Upscaling water availability and water use assessments in hydro-social systems: the small reservoirs of the Merguellil catchment (Central Tunisia)
Caractérisation des ressources et usages de multiples hydro-sociosystèmes : les retenues collinaires du bassin du Merguellil (Tunisie Centrale)
Andrew Sean Gellett OGILVIE
A thesis submitted for the joint degree of Doctor of Philosophy to:
King’s College London Department of Geography
& Université de Montpellier Ecole Doctorale SIBAGHE Eaux Continentales et Sociétés
Abstract Small reservoirs have become increasingly widespread across semi-arid regions, due to their ability to reduce transport of eroded soil and harvest scarce and unreliable rainfall for local users. The scale and geographical dispersion of these multiple hydro-social systems restrict their investigation, leading to difficulties in assessing their agricultural potential, their cumulative influence on runoff, and in identifying strategies to support riparian farmers. This research sought to develop a multi-scalar interdisciplinary approach to assess water availability across multiple small reservoirs and understand hydrological and wider drivers’ influence on associated agricultural practices. An Ensemble Kalman Filter approach was developed to combine 30 m Landsat flooded surface area observations with a daily hydrological (GR4J + water balance) model on 7 gauged reservoirs. Data assimilation, providing near-real time corrections, reduced runoff uncertainties generated by highly variable and localised rainfall intensities and lowered daily volume root mean square errors (RMSE) by 50% compared to the initial rainfall-runoff model forecast. Compensating for Landsat’s reduced temporal resolution and correcting outliers, the method correctly reproduced flood dynamics of 5 ha lakes (R =0.9). Validated against extensive field data over 1999-2014, the method notably establishes Landsat imagery’s ability to assess annual water availability on ungauged reservoirs as small as 1 ha (RMSE circa 25%). Applied to 48 small reservoirs and 546 Landsat 5-8 images, the treatment chain identified the significant water scarcity and unreliability that impedes agricultural development on 80% of lakes in the Merguellil upper catchment (Central Tunisia). In parallel, rapid surveys, quantitative questionnaires and semi-directed interviews confirmed minimal withdrawals, yet highlighted the diversification of practices and the peripheral benefits accompanying small reservoir development. Many farmers lack the capabilities to increase their withdrawals and suffer physical and economic water access difficulties, mismanagement, compounded through limited and short-term government assistance. Individual successes resulted from farmers’ economic resilience and means to secure alternate water supplies during dry spells. Faced with limited storage capacities and prolonged droughts, small reservoirs must in this climatic context retain their supplementary irrigation focus and not strive to support widespread intensification of irrigated practices.
3
Main acronyms and abbreviations employed AUC
Area Under Curve
AWEI
Automated Water Extraction Index
CGIAR
Global research partnership for a food secure future
CRDA
Commissariat Régional au Développement Agricole (Tunisia)
CRU
Climatic Research Unit (University of East Anglia)
CV
Coefficient of Variation
DEM
Digital Elevation Model
DG ACTA
Direction Générale de l’Aménagement et de la Conservation des Terres Agricoles (Tunisia)
DGPS
Differential GPS
DGRE
Direction Générale des Ressources en Eau (Tunisia)
E
Evaporation
ENKF
Ensemble Kalman Filter
ETM+
Enhanced Thematic Mapper Plus (Landsat 7)
GIS
Geographic Information System
GPS
Global Positioning System
GR4J
modèle du Génie Rural à 4 paramètres Journalier (France)
GW
Groundwater
GWin
Groundwater inflow
HSV
Height surface volume relationships
IDW
Inverse Distance Weighting
I
Infiltration
IGN
Institut national de l’information géographique et forestière (France)
IRD
Institut de Recherche pour le Développement (France)
4
KF
Kalman Filter
L
Leak
MNDWI
Modified Normalised Difference Water Index
MODIS
Moderate resolution Imaging Spectroradiometer
MSE
Mean Squared Error
NDAI
Normalised Difference Area Index
NDMI
Normalised Difference Moisture Index
NDPI
Normalised Difference Pond Index
NDVI
Normalised Difference Vegetation Index
NDTI
Normalised Difference Turbidity Index
NDWI
Normalised Difference Water Index
NIR
Near Infra Red
NOAA
National Oceanic and Atmospheric Administration
NRMSE
Normalised Root Mean Squared Error
NSE
Nash Sutcliffe Efficiency
O
Overflows
OLI
Operational Land Imager (Landsat 8)
P
Precipitation (rainfall)
PDAI
Percentage Difference Area Index
P ET
Potential Evapotranspiration
Q
Runoff
R
Releases
RMSE
Root Mean Squared Error
RS
Remote Sensing
S
Surface area
s.d.
Standard deviation
SHV
Starting Haze Value
SLC
Scan Line Corrector (Landsat)
SPOT
Satellite Pour l’Observation de la Terre
SR
Small Reservoir
SRTM
Shuttle Radar Topographic Mission
SWIR
Short Wave Infrared
THEIA
Pôle thématique surfaces continentales (France)
TIRS
Thermal Infrared Sensor (Landsat 8) 5
TM
Thematic Mapper (Landsat 4 & 5)
TRMM
Tropical Rainfall Measuring Mission
USGS
U.S. Geological Survey
UTM
Universal Transverse Mercator
V
Volume
VW B+GR4J
Volume from water balance + GR4J model
Vf ield
Volume from field observations
VRS
Volume from remote sensing assessments
VEN KF
Volume from ENKF approach
W
Withdrawal
WB
Water Balance
WGS
World Geodetic System
WSCW
Water and Soil Conservation Works
Z
Stage (water depth)
6
Acknowledgements Difficult to summarise in a few lines all the people who were directly involved, or who helped shape this journey. At the very least, I wish to thank the following: Gilles Belaud, Patrick Le Goulven and Mark Mulligan, for your confidence and support to develop this research, for the rich, stimulating discussions, and your remarkable scientific and human qualities. Sylvain Massuel, for your sound scientific advice, and resourcefulness in helping install hydrological equipment. Jeanne Riaux for introducing me to the joys of anthropology and ethnographic interviews and our entertaining discussions on combining and confronting hydrological and human sciences’ perspectives. Christian Leduc, for your scientific exigence, and your enduring willingness and ingenuity to fund my contracts at IRD and foster this adventure. Roger Calvez, for sharing your intricate knowledge of the Merguellil upper catchment, and for the numerous field visits. Zakia Jenhaoui, for your unmatched translation and people skills. Marina Alazard, for your unique evaporation expertise and Romain Rochette, for your support with field data. Luuk Fleskens, Séverin Pistre, Pierre Ribstein and Rob Wilby, for taking the time to review this work formally and for your valuable perspectives during the viva. Bernard Cappelaere, Roger Moussa, Nick Clifford, Nick Drake for participating in steering committees and providing your broader cross-cutting insights. Jean Charles Clanet and Jacques Lemoalle, for a most appealing foray into IRD G-eau in 2008, which invigorated my interest in academic research. Colleagues and friends within the G-eau joint research group for your support and inspiring influence. Christine Legrand for your tireless administrative assistance as well as the KCL Geography department staff. Vaughan Robinson and Claire Mouchot (Embassy of France) for supporting the development of this first joint PhD between Montpellier University and King’s College London. The Tunisian DG ACTA, DGRE and Kairouan CRDA for their ongoing collaboration and access to historical time series. Mohamed Ayachi for sharing your vast expertise on the catchment. INAT and the IRD representation in Tunis for facilitating the several months I spent in Tunisia. LISAH and CESBIO researchers for your help in accessing wider data. Friends & previous colleagues in Geneva, UK, China and Guinea for sharing parts of this journey and shaping the itinerary. My family or rather the Ogilvie, Rabanit, Henderson, Rouse families for your love and support. Mum, Dad, Karine for feeding (and funding) my curiosity over the years and transmitting your interest in research & international development. Manon for your love, patience, everyday attentions and accepting to live lately with a monomaniac PhD candidate. Gaspard for distracting me and drawing me away from my desk as often as I needed. My second child for motivating me to submit quickly... Merci! 7
To Manon, Gaspard & the bump...
8
Contents I 1 1.1
1.2 1.3
2 2.1 2.2
2.3
2.4
Research overview
22
Research context and objectives Research context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Understanding hydro-social systems . . . . . . . . . . . . . 1.1.1.1 Water uses & drivers . . . . . . . . . . . . . . . . 1.1.1.2 Hydro-social systems . . . . . . . . . . . . . . . . 1.1.2 Assessing water resources across small reservoirs . . . . . . 1.1.3 Potential of remote sensing in hydrology . . . . . . . . . . . 1.1.4 Questions and limitations over remote sensing . . . . . . . . Aims & objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overall methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Upscaling and combining hydro-social assessments . . . . . 1.3.2 Water availability monitoring across small reservoirs . . . . 1.3.3 Water uses & hydro-sociological drivers on small reservoirs
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
Research site and overview of field data Merguellil upper catchment, Central Tunisia . . . . . . . . . . . . . . . . . . . Small reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Small reservoir development . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Inventory of small reservoirs in and around the Merguellil upper catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Digital elevation models and reservoir catchment area delimitation . . 2.2.4 Lake cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hydro-meteorological observation network . . . . . . . . . . . . . . . . . . . . . 2.3.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Rainfall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2.1 Monitoring equipment . . . . . . . . . . . . . . . . . . . . . . 2.3.2.2 Data availability . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Lake pan evaporation . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4 Small reservoirs hydrometric data . . . . . . . . . . . . . . . . . . . . 2.3.4.1 Equipment status . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4.2 Data treatments . . . . . . . . . . . . . . . . . . . . . . . . . Deriving power relations for height-surface-volume . . . . . . . . . . . . . . . . 2.4.1 HSV and power relations . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Power relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2.1 Modelling a common S-V power relation . . . . . . . . . . .
9
23 23 24 24 25 25 26 27 28 29 30 30 31 34 34 35 35 36 38 39 40 40 41 41 44 44 45 45 47 48 48 49 51
2.4.3
2.4.2.2 Validating a common S-V power relation . Silting . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3.1 Erosion rates . . . . . . . . . . . . . . . . . 2.4.3.2 Silting in small reservoirs . . . . . . . . . . 2.4.3.3 Accounting for silting in HSV for ungauged 2.4.3.4 Discussing HSV and silting errors . . . . .
. . . . . . . . . . . . lakes . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
II Upscaling water availability assessments in ungauged reservoirs 3 3.1
3.2
3.3 4 4.1
53 55 55 56 57 59
63
Remote sensing of flooded areas in small reservoirs Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Selecting satellite imagery . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2 Landsat satellites & sensors . . . . . . . . . . . . . . . . . . . . . . . . 3.1.3 Landsat image availability . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.4 Image pretreatments and corrections . . . . . . . . . . . . . . . . . . . 3.1.4.1 SLC-off errors (Pixel loss on Landsat imagery) . . . . . . . . 3.1.4.2 Cloud & shadow errors (Pixel loss on Landsat imagery) . . . 3.1.4.3 Radiometric normalisation . . . . . . . . . . . . . . . . . . . 3.1.5 Water detection by remote sensing . . . . . . . . . . . . . . . . . . . . 3.1.5.1 Water reflectance . . . . . . . . . . . . . . . . . . . . . . . . 3.1.5.2 Multispectral analysis . . . . . . . . . . . . . . . . . . . . . . 3.1.6 Selecting & calibrating water indices . . . . . . . . . . . . . . . . . . . 3.1.6.1 Surface area assessments in the field . . . . . . . . . . . . . . 3.1.6.2 Errors & confusion matrices . . . . . . . . . . . . . . . . . . 3.1.6.3 Validating the threshold . . . . . . . . . . . . . . . . . . . . . 3.1.7 Assessing flood dynamics over time . . . . . . . . . . . . . . . . . . . Results & discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Comparing the performance of spectral water indices . . . . . . . . . . 3.2.2 Remote sensing of flood dynamics with MNDWI . . . . . . . . . . . . 3.2.2.1 Cloud & shadow influence . . . . . . . . . . . . . . . . . . . . 3.2.2.2 SLC-off influence . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2.3 Optimising cloud&shadow and SLC-off thresholds for long term monitoring . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2.4 Applying remote sensing to monitor flood dynamics . . . . . 3.2.2.5 Daily surface area & water availability over time . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
106 110 115 119
Water availability modelling in an Ensemble Kalman Filter Method overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Data assimilation methods . . . . . . . . . . . . . . . . . . 4.1.2 Ensemble Kalman Filter . . . . . . . . . . . . . . . . . . . . 4.1.3 Remote sensing inputs . . . . . . . . . . . . . . . . . . . . . 4.1.4 Water balance modelling . . . . . . . . . . . . . . . . . . . . 4.1.5 Model sensitivity and performance on ungauged reservoirs .
121 121 121 123 125 125 127
10
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
64 64 64 65 66 70 70 73 75 82 82 82 85 85 87 93 93 95 95 105 105 106
4.2
4.5
Runoff modelling with GR4J . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Model structure . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Model calibration . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2.1 Runoff estimation from water balance . . . . . . . . 4.2.2.2 Calibration steps . . . . . . . . . . . . . . . . . . . 4.2.2.3 Transposing parameters to ungauged reservoirs . . . Estimating and interpolating the other water balance fluxes . . . . . . 4.3.1 Interpolating rainfall . . . . . . . . . . . . . . . . . . . . . . . 4.3.1.1 Spatial interpolation methods . . . . . . . . . . . . 4.3.1.2 Inverse distance weighting interpolation of rainfall . 4.3.2 Interpolating lake evaporation . . . . . . . . . . . . . . . . . . 4.3.2.1 Spatial & temporal interpolation . . . . . . . . . . . 4.3.2.2 Transposition coefficients for lake evaporation . . . 4.3.3 Potential evapotranspiration for catchment modelling . . . . 4.3.4 Estimating infiltration rules . . . . . . . . . . . . . . . . . . . 4.3.4.1 Infiltration values for gauged lakes . . . . . . . . . . 4.3.4.2 Modelling infiltration on ungauged SR . . . . . . . . 4.3.5 Modelling overflows . . . . . . . . . . . . . . . . . . . . . . . 4.3.6 Modelling releases . . . . . . . . . . . . . . . . . . . . . . . . 4.3.7 Modelling withdrawals . . . . . . . . . . . . . . . . . . . . . . Results & discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Kalman filter performance on daily volumes . . . . . . . . . . 4.4.2 Kalman Filter performance on annual water availability . . . 4.4.3 Remote sensing uncertainties . . . . . . . . . . . . . . . . . . 4.4.4 Rainfall-runoff modelling difficulties . . . . . . . . . . . . . . 4.4.4.1 Heterogeneous catchment responses . . . . . . . . . 4.4.4.2 Rainfall measurements and interpolation . . . . . . 4.4.4.3 Other data difficulties . . . . . . . . . . . . . . . . . 4.4.5 Monitoring ungauged basins with remote sensing observations Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
III
Understanding small reservoir hydro-social systems
179
5 5.1
Lake benefits and hydro-sociological drivers of water use Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Water availability characterisation . . . . . . . . . . . . . . . . 5.1.1.1 Annual water availability patterns and demand . . . . 5.1.1.2 Water availability index . . . . . . . . . . . . . . . . . 5.1.2 Water use and users characterisation . . . . . . . . . . . . . . . 5.1.2.1 Rapid water use surveys on 56 lakes . . . . . . . . . . 5.1.2.2 Agricultural questionnaires on 22 lakes . . . . . . . . 5.1.2.3 Ethnographic interviews on 4 lakes . . . . . . . . . . . 5.1.3 Characterising lake benefits and extracting hydro-sociological straints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
180 . 180 . 180 . 180 . 181 . 182 . 182 . 183 . 184
4.3
4.4
5.2
11
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
128 128 128 128 129 130 131 131 131 132 133 133 139 141 141 143 145 148 149 149 150 150 155 160 165 165 168 170 171 177
. . . . . . . . . . . . . . . . . . . . . . . . con. . . . 185 . . . . 186
5.2.1
5.3
Supporting and changing agricultural practices around small reservoirs 5.2.1.1 Irrigated practices . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1.2 Cereals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1.3 Livestock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1.4 Fishery and other complementary activities . . . . . . . . . 5.2.1.5 Groundwater recharge and peripheral water use . . . . . . . 5.2.2 Wider benefits of SR . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2.1 Protecting downstream areas . . . . . . . . . . . . . . . . . 5.2.2.2 Other socio-economic benefits . . . . . . . . . . . . . . . . . 5.2.3 Typology of lakes’ agricultural benefits . . . . . . . . . . . . . . . . . . 5.2.4 Water availability drivers of water use . . . . . . . . . . . . . . . . . . 5.2.5 Wider hydro-sociological drivers . . . . . . . . . . . . . . . . . . . . . 5.2.5.1 Water access difficulties . . . . . . . . . . . . . . . . . . . . . 5.2.5.2 Managing hydrological uncertainty . . . . . . . . . . . . . . . 5.2.5.3 User participation and government assistance . . . . . . . . . 5.2.5.4 Local water mismanagement . . . . . . . . . . . . . . . . . . 5.2.5.5 Conflicting livelihood strategies . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
IV
Final conclusions and perspectives
V 6
186 186 191 192 194 195 195 195 196 196 198 209 209 212 214 216 218 219
221
Appendices
226
Supplementary research on rainfall and runoff at the basin scale
227
7 7.1 7.2
Appendix to chapter 2 - Additional content 242 Inventory of lakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Height-surface-volume relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
8 8.1 8.2 8.3
Appendix to chapter 3 - Additional content 253 Background on GPS methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Selecting a spectral water index . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 Calibrating percentage of pixels to remove due to clouds & SLC off . . . . . . . 256
9 9.1 9.2
Appendix to chapter 4 - Additional content Interpolating lake evaporation . . . . . . . . . . . . Further details on estimating water balance fluxes 9.2.1 Infiltration . . . . . . . . . . . . . . . . . 9.2.2 Releases . . . . . . . . . . . . . . . . . . . 9.2.3 Overflows . . . . . . . . . . . . . . . . . . Kalman filter performance . . . . . . . . . . . . . .
9.3 10 10.1
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
262 . 262 . 266 . 266 . 267 . 268 . 268
Appendix to chapter 5 - Forms developed for field surveys and interviews 272 Water survey forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
12
10.2 10.3
Water use questionnaires . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 Semi directed interview topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
VI
Extended summary in French
VII
Bibliography
292 306
List of Figures 1.1.1 1.3.1 1.3.2
Estimated number of small reservoirs for several countries . . . . . . . . 24 Schematic representation of overall methodology . . . . . . . . . . . . . . 30 Location of Merguellil upper catchment and of neighbouring hydrometeorological instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 2.2.2 2.2.3 2.3.1
Merguellil upper catchment and location of instrumented small reservoirs Detecting lakes for inventory using MNDWI on Landsat imagery . . . . . Comparing SRTM vs ASTER catchment delineation for small reservoirs Rainfall and evaporation monitoring equipment on SR in Merguellil catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rainfall data availability for the 7 SR modelled in the following chapters Availability of evaporation data for small reservoirs in and around the Merguellil upper catchment . . . . . . . . . . . . . . . . . . . . . . . . . . Limnimetric monitoring equipment . . . . . . . . . . . . . . . . . . . . . Availability of stage data for small reservoirs in and around the Merguellil upper catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Available rating curves for small reservoirs in and around Merguellil catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Topographic survey of Hoshas small reservoir . . . . . . . . . . . . . . . . Linear regression to derive coefficients of power relation between surface area (S) and volume (V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between volumes estimated from rating curve and through S-V power relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relationship between surface area and volume based on 11 SR rating curves and the interSR relation derived from linear regression . . . . . . B and — parameters for each of the 15 lakes . . . . . . . . . . . . . . . . Lake specific S-V power relations against the interSR power relation . . . Change over time of S-V relationships on two lakes . . . . . . . . . . . . Modelled evolution of B and — parameters over time . . . . . . . . . . . Lake and date specific S-V relation against the interSR power relation adapted for silting over time . . . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 2.3.3 2.3.4 2.3.5 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 2.4.6 2.4.7 2.4.8 2.4.9 2.4.10
13
36 37 39 42 43 45 46 47 48 49 50 50 52 52 54 58 60 61
2.4.11
Lake and date specific S-V relation against the interSR power relation adapted for silting over time . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1.1
Wavelengths and atmospheric transmission of the bands of Landsat 5 TM, Landsat 7 ETM+ and Landsat 8 OLI-TIRS sensors . . . . . . . . . 67 Overlap of Landsat Paths 191 35 and 192 35, SRTMv3 n35e009 digital elevation model and small reservoirs . . . . . . . . . . . . . . . . . . . . . 68 Availability of Landsat images for Merguellil upper catchment . . . . . . 68 Lag between successive Landsat images used here . . . . . . . . . . . . . 69 Chain of treatments applied to the 546 Landsat images . . . . . . . . . . 70 Pixel loss due to SLC failure compared to objects of interest on Landsat 7 29.03.2013 image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Percentage of Landsat images affected by SLC presence over 1 lake cell (Gouazine) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Comparing cloud detection methods on Landsat 8 29.03.2013 image . . . 75 Guettar lake affected by nearby cloud and cloud shadow presence . . . . 75 Percentage of Landsat images affected by clouds when assessed over a whole scene and an individual lake cell . . . . . . . . . . . . . . . . . . . 76 Merguellil catchment small reservoirs visible on full colour composite (bands 453) Landsat 5 2.11.1999 image . . . . . . . . . . . . . . . . . . . 83 DGPS surveys on Morra reservoir . . . . . . . . . . . . . . . . . . . . . . 86 Water detection rate for varying thresholds of MNDWI on Landsat 8 29.03.2013 image for lake Morra . . . . . . . . . . . . . . . . . . . . . . . 91 Incorrectly classified pixels when optimising MNDWI for Water & Non water Producer Accuracies vs Max overall accuracy . . . . . . . . . . . . 92 MNDWI classification output when optimising for Water & Non water Producer accuracies, Max overall accuracy and their overlap . . . . . . . 92 Detection rates for each index and error for lake Morra (cell 30) on Landsat 8 29.03.2013 image . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Compared performance of 7 water indices on 6 lakes during wet period (March 2013) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Compared performance of 7 water indices on 6 lakes during dry period (June 2013) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Optimal threshold (based on overall accuracy) for each lake, date & index100 Errors in validation of calibrated threshold . . . . . . . . . . . . . . . . . 102 Correlation between field measurements and satellite estimation . . . . . 103 Errors in calibration and validation steps with MNDWI and optimal threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Errors from clouds and shadows on scatter plot between field and Landsat observations (Gouazine, lake cell 51) . . . . . . . . . . . . . . . . . . . . 105 PDAI errors as a function of percentage of SLC-off and cloud & shadow pixels over 1 lake cell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Change in R (between field and remotely sensed surface areas) and remaining number of Landsat scenes for increasing percentages of SLC-off and cloud & shadows pixels tolerated (Lake Gouazine) . . . . . . . . . . 108
3.1.2 3.1.3 3.1.4 3.1.5 3.1.6 3.1.7 3.1.8 3.1.9 3.1.10 3.1.11 3.1.12 3.1.13 3.1.14 3.1.15 3.2.1 3.2.2 3.2.3 3.2.4 3.2.5 3.2.6 3.2.7 3.2.8 3.2.9 3.2.10
14
3.2.11 3.2.12 3.2.13 3.2.14 3.2.15 3.2.16 3.2.17 3.2.18 3.2.19
4.1.1 4.1.2 4.3.1 4.3.2 4.3.3 4.3.4 4.3.5 4.3.6 4.3.7 4.3.8 4.3.9 4.3.10 4.3.11 4.3.12 4.4.1 4.4.2 4.4.3 4.4.4 4.4.5 4.4.6 4.4.7
Aggregated surface area error over all images on Gouazine lake according to varying tolerance levels of pixel loss from: . . . . . . . . . . . . . . . . Correlation between remotely sensed surface area and field data over 1999-2014 for Gouazine lake . . . . . . . . . . . . . . . . . . . . . . . . . Surface area time series through remote sensing and field data for Gouazine lake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between remotely sensed surface area and field data . . . . . Comparing remotely sensed flooded surface area and field data over 19992015 for 6 lakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Daily flood dynamics and mean surface area from field data and remote sensing over 1999-2015 for lake Gouazine . . . . . . . . . . . . . . . . . . Surface area time series from field data and remote sensing data over 1999-2015 for 6 lakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mean annual surface area from field data and remote sensing data over 1999-2015 for 6 lakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Correlation between mean daily surface area per year for each lake using field and remote sensing data . . . . . . . . . . . . . . . . . . . . . . . . .
109 111 112 112 114 116 117 118 119
Schematic representation of methodology for hydrological study of small reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Water balance fluxes of small reservoirs . . . . . . . . . . . . . . . . . . 126 Scatterplot of observed and interpolated rainfall at Gouazine . . . . . . . 132 Monthly evaporation over 1995-1999 for 10 lakes . . . . . . . . . . . . . . 134 Mean interannual evaporation over 1995-1999 for 10 lakes against altitude134 Scatterplots of evaporation observations vs IDW and KED interpolated values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 Interannual variations in monthly evaporation . . . . . . . . . . . . . . . 137 Scatterplots of evaporation observations at El Haouareb vs observed and IDW interpolated values for 2 lakes . . . . . . . . . . . . . . . . . . . . . 138 Relationship between pan coefficient and lake size . . . . . . . . . . . . . 140 MODIS-derived interannual 2000-2014 P ET . . . . . . . . . . . . . . . . 142 Monthly P ET variations across 58 lakes in 2000-2003 . . . . . . . . . . . 143 Relationship between pan evaporation and P ET on Gouazine SR . . . . 144 Infiltration values as a function of stage in the lake estimated during depletion periods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 Rate of change of infiltration values with water depth for 14 small reservoirs147 Comparing outputs of Ensemble Kalman Filter with field data and other methods for daily volumes (Gouazine, 1997-2014) . . . . . . . . . . . . . 151 Modelled and observed volume time series for Fidh Ali & Hoshas . . . . 152 Modelled and observed volume time series for Fidh Ben Nasseur & Dekikira153 Modelled and observed volume time series for Morra & Guettar . . . . . 154 Scatterplot between modelled and observed daily volumes . . . . . . . . 156 Comparing outputs of ENKF with field data and other methods for annual water availability (Gouazine, 1997-2014) . . . . . . . . . . . . . . . . 158 Modelled and observed annual water availability for Fidh Ali & Hoshas . 161
15
4.4.8 4.4.9 4.4.10 4.4.11 4.4.12 4.4.13 4.4.14 4.4.15 4.4.16 4.4.17 4.4.18
5.1.1 5.1.2 5.2.1 5.2.2 5.2.3 5.2.4 5.2.5 5.2.6 5.2.7 5.2.8 5.2.9 5.2.10 5.2.11 5.2.12 5.2.13 5.2.14 5.2.15
Modelled and observed annual water availability for Fidh Ben Nasseur & Dekikira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modelled and observed annual water availability for Morra & Guettar . . Error in the number of days water levels fall below a given volume . . . Performance of the WB+GR4J model on Gouazine . . . . . . . . . . . . Observed and modelled daily volumes by WB+GR4J for 6 lakes . . . . . Rainfall interpolation over catchment with and without station in catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modelled daily volumes when degrading knowledge of the HSV, GR4J parameters and infiltration values on Gouazine lake . . . . . . . . . . . . Modelled daily volumes when degrading knowledge of the HSV, GR4J parameters and infiltration values on 3 SR . . . . . . . . . . . . . . . . . Modelled mean annual volumes when degrading the HSV, GR4J parameters and infiltration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Water volumes estimated from Landsat observations across 51 lakes using common scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Water volumes estimated from Landsat observations across 51 lakes using free scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
162 163 164 166 167 169 172 173 174 175 176
Schematic representation of methodology used to determine water uses and associated hydro-sociological drivers . . . . . . . . . . . . . . . . . . 182 Reservoirs selected for ethnographic interviews and ongoing hydrological instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 Observed heterogeneity in the number of fruit trees per farm and per lake based on questionnaires . . . . . . . . . . . . . . . . . . . . . . . . . 187 Percentage of each fruit tree species grown by small holders interviewed . 188 Diversification of agricultural practices following the small reservoir construction based on farmer interviews . . . . . . . . . . . . . . . . . . . . 189 Agricultural water uses around SR in the Merguellil catchment . . . . . . 190 Number of motor pumps in 2005 (on 21 lakes) and 2012 (on 56 lakes) . . 191 Additional uses of SR in the Merguellil catchment . . . . . . . . . . . . . 193 Irrigated agriculture on wells situated directly downstream from the Gouazine lake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Water use typology of 56 lakes studied in (and near) the Merguellil upper catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Mean daily water availability over the dry season per year over 2000-2014 201 Number of days for each lake water volumes fell below a designated volume202 Mean interannual availability over 2007-2014 per lake over the dry season displayed as mean ±1 standard deviation . . . . . . . . . . . . . . . . . . 203 Number of days for each lake water availability fell below 5000 m3 over the whole year and the 6 dry months . . . . . . . . . . . . . . . . . . . . 204 Definition of water availability categories based on number of days during 6 dry months water availability exceeded 5000 m3 . . . . . . . . . . . . . 205 Simulated water deficit on 1 lake based on observed water demand . . . 207 Simulated water availability on 4 lakes based on observed water demand 208
16
5.2.16 5.2.17 5.2.18 5.2.19 5.2.20 7.2.1 7.2.2 7.2.3 7.2.4 7.2.5 7.2.6 8.3.1 8.3.2 8.3.3 8.3.4 8.3.5 8.3.6
9.1.1 9.3.1 9.3.2
9.3.3
Relationship between initial capacity of each lake and the mean daily volume over 2007-2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figurative representation of water availability and water uses for small reservoirs according to the 3x3 categories in table 5.2.2. . . . . . . . . . Map of interannual mean water availability (2007-2014) and number of motor pumps per lake . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Diversity of pumps observed on SR in Merguellil catchment . . . . . . . Location of beneficiary and excluded households around Guettar lake . . Difference in volume estimated through H-V power relation if derived on all or upper values (Gouazine) . . . . . . . . . . . . . . . . . . . . . . . . Comparison of true volume against volume estimated by different power relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of true volume against volume estimated by different power relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relationship between — in S-V power relation and several lake characteristics for 15 lakes studied . . . . . . . . . . . . . . . . . . . . . . . . . Relationship between B in S-V power relation and several lake characteristics for 15 lakes studied . . . . . . . . . . . . . . . . . . . . . . . . . Change of power relation parameters over time for 15 small reservoirs . . Cloud cover (%) across 1 lake cell (Gouazine) for images of different scene cloud cover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Errors from clouds and shadows on remotely sensed surface area time series (lake Gouazine, cell 51) . . . . . . . . . . . . . . . . . . . . . . . . Errors from SLC-off pixels on remotely sensed surface areas time series (lake Gouazine, cell 51) . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimising SLC-off and cloud & shadow thresholds on additional lakes . Optimising SLC-off and cloud & shadow thresholds on additional lakes (cont.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Aggregated surface area error over all images on Morra lake according to varying tolerance levels of pixel loss due to: . . . . . . . . . . . . . . . . .
208 210 211 213 218
247 248 249 250 251 252
257 258 258 259 260 261
Relationships between El Haouareb and lake interpolated monthly evaporation values (mm) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 Performance of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake . . . . 269 Scatterplot between observed daily volume and simulated volumes of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake . . . . . . . . . . 270 RMSE of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake . . . . 271
17
List of Tables 2.4.1 2.4.2 2.4.3
3.1.1 3.1.2 3.1.3 3.1.4 3.1.5 3.2.1 3.2.2
4.2.1 4.3.1 4.3.2 4.3.3 4.3.4 4.4.1 4.4.2 4.4.3 4.4.4
B and — values for S-V power relations for small reservoirs . . . . . . . . 51 Error on volume from interSR power relation against volume from lake specific power relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Silting and associated capacity loss measured on 14 small reservoirs between 1988 and 1998 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Exoatmospheric solar constant per band for ETM+ and TM sensors . . . K1 & K2 constants for correcting thermal radiance to temperature . . . Dates of DGPS surveys and satellite imagery used to calibrate multispectral indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flooded surface area (m2 ) according to three DGPS surveys on 7 lakes . Confusion matrix for remotely sensed flooded area and DGPS contours . Summary of confusion matrix values (%) obtained when validating the optimal MNDWI threshold on 7 lakes on Landsat 8 21.05.2013 image . . R and number of images remaining after removal of scenes with >40% clouds & shadows and >25% SLC-off pixels . . . . . . . . . . . . . . . . Optimal GR4J parameters sets and R values for each catchment . . . . Errors from IDW and KED rainfall interpolation compared to observations at Gouazine rainfall gauge . . . . . . . . . . . . . . . . . . . . . . . Errors from IDW and KED interpolation of evaporation compared to observations on 2 lakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interannual variations in annual evaporation . . . . . . . . . . . . . . . . Infiltration values (mm) for small reservoirs in and around the Merguellil upper catchment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ensemble Kalman Filter performance on daily volumes . . . . . . . . . . Ensemble Kalman Filter performance on annual water availability . . . . Errors when interpolating rainfall events over 20 mm without gauges in the catchments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kalman Filter performance on mean water availability when degrading model inputs and parameters . . . . . . . . . . . . . . . . . . . . . . . . .
79 82 87 87 88 102 110 130 133 135 139 148 157 159 169 171
5.2.1 5.2.2
Water availability and use characteristics for each lake . . . . . . . . . . 199 Categorisation of 48 small reservoirs based on water availability and water uses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
7.1.1 7.1.3
Lakes in the Merguellil upper catchment used in the study . . . . . . . . 242 Lakes around the Merguellil catchment used in the study . . . . . . . . . 245 18
7.2.1
Percentage error on resulting power relation for 1 lake when using different ranges of the true rating curve . . . . . . . . . . . . . . . . . . . . . . 246
8.2.1
Water detection rates on each reservoir based on different calibration methods of MNDWI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 Errors on validation image using threshold derived from calibration (on overall accuracy) for each lake and index . . . . . . . . . . . . . . . . . . 256
8.2.3
9.1.1
Relationship between the interpolated (IDW) monthly evaporation 19952008 values for each lake and for El Haouareb . . . . . . . . . . . . . . . 262
19
20
Oued El Haffar small reservoir (30 000 m3 ) in the Merguellil upper catchment
21
Part I
Research overview
22
Chapter 1
Research context and objectives 1.1
Research context
In Mediterranean and other semi arid areas, the characteristic high spatio-temporal rainfall variability leads to prolonged droughts (Nicault et al. 2008) and intense rainfall events (Cudennec et al. 2007; García-Ruiz et al. 2011) which generate substantial soil erosion. Small reservoirs and other water and soil conservation works (WSCW) have evolved since Neolithic times (Bruins et al. 1986) in an attempt to reduce soil loss and silting of downstream dams and harvest scarce and unreliable water resources for local users (Habi and Morsli 2011). Their simplicity and reduced costs led to significant bottom up development of numerous açudes in Brazil (Burte et al. 2005) and tanks in India (Bouma et al. 2011) (figure 1.1.1). Considering their ability to contribute to food security in rural drought-prone areas, and due to the negative effects of large scale irrigation projects (Wisser et al. 2010; Ngigi et al. 2005), small reservoirs and other rainwater harvesting methods gained growing support from governmental and international projects. Since the 1960s, small reservoirs have become increasingly widespread, notably across parts of Northern and sub-Saharan Africa (Nyssen et al. 2010; Sawunyama et al. 2006; Talineau et al. 1994). By capturing upstream runoff, these structures modify the spatio-temporal distribution of water resources, notably increasing evaporation and in certain cases, returns to groundwater through infiltration (Gay 2004). At a catchment scale, past studies (Dridi 2000; He et al. 2003; Kingumbi et al. 2007; Lacombe 2007; Ogilvie et al. 2016; Gao et al. 2011) have shown reduction around 20% in downstream flows consecutive to the construction of several WSCW within a single catchment. Their influence however is concurrent with other variable and mutually dependent processes (Cudennec et al. 2004), which affect the hydrological behaviour of the catchment including rainfall spatial variability, soil moisture, land use and land cover changes and groundwater interactions (Kingumbi et al. 2007; Lacombe et al. 2008; Ogilvie et al. 2016). Individual influences are difficult to distinguish and the cumulative influence of small reservoirs and other WSCW in catchments over 100 km remains poorly understood (Kongo and Jewitt 2006; Lacombe et al. 2008). Studies in the same catchment over the same period in Tunisia notably identified a reduction in downstream flows from WSCW between 1% and 50%, illustrating the complexity and uncertainties in upscaling and quantifying the influence of individual structures. With the pressures on limited water resources increasing in semi-arid areas as a result of changes in climate, land uses, population, living standards, 23
50
610
120 1000 12000
208000
350 800 100
3000
1700 600
110
15000
1000 425
3000
70000
600 10000
Number of SR per country
0
1 250
2 500
5 000 Kilometres
Figure 1.1.1: Estimated number of small reservoirs for several countries (Adapted from Venot and Krishnan (2011) and based on national, non exhaustive assessments) and tourism (García-Ruiz et al. 2011), greater optimisation of resources (green/blue water) is required and the potential of small reservoirs have come under scrutiny (Mushtaq et al. 2007; Le Goulven et al. 2009).
1.1.1
Understanding hydro-social systems
1.1.1.1
Water uses & drivers
By supplementing water resources and extending the growing season, small reservoirs have the potential to provide supplementary irrigation and support agricultural livelihoods within poorer rural areas (Wisser et al. 2010). Their reduced costs and potential to be implemented at the scale of individual farms or clusters notably supports recent efforts to recognise and promote irrigation and water management practices of small holders (Vincent 2003). Studies revealed multi use systems with diverse benefits including watering livestock, irrigation, fish production as well as recreational and cultural importance. Agricultural production remained sometimes limited (Khlifi et al. 2007; Habi and Morsli 2011; Lacombe 2007; Selmi and Talineau 1994; Faulkner et al. 2008) leading to concerns by donors and investors despite the strong demand and apparent affection for reservoirs by users (Venot and Hirvonen 2013). This contradiction highlighted the difficulty in identifying and quantifying the value of these systems, partly due to the inherently subjective and restrictive frameworks used, centred on issues of agricultural performance, efficiency or productivity (Venot and Krishnan 2011). These may be deceptive and lead to under evaluating their importance. Literature on the reasons which constrain or favour agricultural water use in small reser-
24
voirs remains scarce. Hydrological limitations and user uncertainties over water availability have been highlighted (Mugabe et al. 2003), however other observations pointed to insufficient resources to cover investment & maintenance costs, lack of adequate management structures (Zairi et al. 2005) or the consequence of government strategies and siting criteria (Talineau et al. 1994). The overlapping interactions between drivers of agricultural water use remain poorly understood, often as a result of mono-disciplinary studies considering agronomic, social (i.e. cultural, economic, institutional) and hydrological factors separately. Understanding the dynamics around small reservoirs requires a comprehensive framework to study the reservoir’s hydrology, the riparian community and their interactions. 1.1.1.2
Hydro-social systems
The recent surge in scientific literature in socio-hydrology (Sivapalan et al. 2012; Montanari et al. 2013; Sivakumar 2012; Di Baldassarre et al. 2013; King et al. 2012; Riaux 2013) highlights the growing recognition of the importance and difficulties in studying these mutual interactions between society and its water environment. Despite the growing popularity for the recently coined term, the concept refers to common contemporary approaches which recognise the importance of human interactions within socio-ecological (Ostrom 2009) or socio-hydrological systems. This begins with considering human influences, such as withdrawals, impoundments and other human induced changes on hydrological systems, moving away from theoretically undisturbed natural systems (Thompson et al. 2013). Conversely, it also focusses on the influence of hydrological systems on human interests, notably water variability on water & food security (Bakker 2012) and the consequences of flood dynamics (Di Baldassarre et al. 2013) and droughts. Nevertheless, socio-hydrology by its very definition must also move towards identifying suitable interdisciplinary approaches, capable of providing new insights on the broad, complex water and society interactions and retroactions (Braden et al. 2009; Bakker 2012). These must be transcribed into operational approaches, modus operandi (Sivapalan et al. 2012; Hale et al. 2015) which seek to yield added value from incorporating other disciplinary approaches and not just combining their results. By exploring and borrowing different tools, these may address hydrological questions from a different perspective, (such as the role of institutions on water access and water uses). By broadening our approach and understanding, such approaches can also allow new topics and new questions to be investigated and generated (Riaux and Massuel 2014).
1.1.2
Assessing water resources across small reservoirs
Despite their growing importance worldwide, small reservoirs are often not quantified due to the bottom up development of these or due to incomplete records and crucially little information is available on their water availability. Where inter and intra-annual water variability is high, absence of this information impedes water managers’ capacity to understand the potential of these reservoirs, and can also constrain the use of the resource (Mugabe et al. 2003). Information on water availability volume, duration and timing can help advise users & other stakeholders on what may be grown at different periods of the season (Alsdorf et al. 2007; Liebe et al. 2005). Estimating the spatio temporal variations across several reservoirs and several years, can also help produce spatial assessments (and maps) of which lakes are more prone to droughts and guide the selection of locations where investments 25
(by planners or users) may be more suitable. At the catchment scale, understanding the volumes captured by small reservoirs (and their use) can help understand their influence on downstream flows, feeding into upstream-downstream water allocation debates and decision support tools. Volume changes within reservoir can indeed serves as multiple runoff gauges, to be integrated within larger catchment scale models. Understanding their water availability over time can also help refine water balance assessment, notably depletion fluxes of evaporation and infiltration. Besides initial design capacities, water availability in small reservoirs is determined by site specific hydrological dynamics regulated by both natural (climate, topography, geology, geomorphology, pedology), and human factors (maintenance, leaks, withdrawals, releases). Studies based on hydrological measurements and geochemical analyses identified the water balance of small reservoirs (Gay 2004; Grunberger et al. 2004; Lacombe 2007; Li and Gowing 2005; Nyssen et al. 2010) but highlighted multiple uncertainties (Li and Gowing 2005) in evaporation and groundwater flows as well as the human management which remains poorly understood (Grunberger et al. 2004; Lacombe 2007; Leduc et al. 2007; Kingumbi et al. 2007). Hydrological monitoring of individual lakes allowed the modelling of flood depletion dynamics however the high spatial rainfall variability led to large uncertainties when modelling runoff into the dams. Rainfall runoff modelling of gauged small reservoirs in Tunisia notably failed to exceed R = 0.5 (Lacombe 2007). The heterogeneity in sub-catchments and flood dynamics across reservoirs further increase difficulties when seeking to model water resources in ungauged small reservoirs (Cudennec et al. 2005).
1.1.3
Potential of remote sensing in hydrology
Where numerous small reservoirs are dotted across a single catchment (landscape) and questions about their hydrology arise, their quantity, size and geographical dispersion make the operational monitoring of their water resources both costly and impractical. Instrumenting and maintaining a hydrometric observation network requires significant time, equipment and transport costs (Liebe et al. 2005) which are not compatible with the localised and modest importance of small reservoirs. As a result, contrary to large reservoirs whose size and importance for irrigation and/or hydropower warrants the development of a hydrometric observation network, small reservoirs are rarely instrumented, except for research purposes (Albergel and Rejeb 1997). Remote sensing which began with aerial photography has seen a significant surge in methods and applications following the development of satellite imagery, symbolised by the launch of the Landsat 1 satellite in 1972. Providing coverage of the whole globe at increasing spatial, temporal and spectral resolution and at reducing costs, satellite-borne sensors increased opportunities for spatial analysis of the atmosphere, oceans, continental surfaces and underground resources. Applications notably include meteorology, topography and altimetry (digital elevation models), land use classifications and numerous agricultural applications such as biomass estimations (leaf area index, yields), transpiration, surface temperature, soil moisture, etc. (Bastiaanssen et al. 2000). Remote sensing can be used to study large areas, multiple locations and archive images support historical/diachronic perspectives over several decades. The increasing spatial and temporal resolution and broad range of sensor characteristics enhance the possibilities and allow for global studies as well
26
as finer investigations in hydrology and beyond. Within hydrology, remote sensing (RS) has been used successfully to provide information of interest on flooded areas, hydrological processes and ecosystems. Images were used to assess water surfaces, river depths and widths, as well as investigate water quality issues such as turbidity or salinity, notably through proxies such as chlorosis or increased brightness resulting from high salinity (Bastiaanssen et al. 2000) or visualise geomorphology and map river channels. Remote sensing notably facilitates the assessment and monitoring of large areas such as wetlands including the Niger Inner Delta (Mahé et al. 2011; Seiler et al. 2009; Ogilvie et al. 2015; Bergé-Nguyen and Crétaux 2015), the Okavango (Wolski and MurrayHudson 2008; Gumbricht et al. 2004), the Sudd (Mohamed et al. 2004) in South Sudan, large rivers such as the Amazon (Martinez and Le Toan 2007; Alsdorf et al. 2007) and large lakes notably in East Africa (Swenson and Wahr 2009; Ouma and Tateishi 2006), China (Ma et al. 2007; Qi et al. 2009). Smaller water bodies have recently been included in worldwide inventories, done with MODIS 250 m and Landsat 30 m images (Feng et al. 2015; Verpoorter et al. 2014). A number of localised studies have also highlighted the potential of remote sensing methods to monitor surface area of reservoirs as low as 1 ha. These include small ponds from 1 to 25 ha in West Africa (Niger, Mali, Ghana, Senegal) (Gardelle et al. 2009; Liebe et al. 2005; Annor et al. 2009; Soti et al. 2010), in Southern Africa (Sawunyama et al. 2006), in South India (Mialhe et al. 2008), and in South America (Rodrigues et al. 2011). These works used a variety of sensors (radar, optical: NOAA AVHRR, Landsat, SPOT) and methods but confirmed the potential of RS methods as a cost effective technique to study water bodies. Images were used for inventories but also to monitor flood processes over time and assess water volumes without costly ground surveys. Remote sensing data is also increasingly used in hydrological studies, both hydrological modelling and water balance (Swenson and Wahr 2009; Leauthaud et al. 2013). In ungauged catchments, satellite imagery can provide information on evaporation, rainfall but also runoff through river depth and width or surface area & associated volume changes. It was notably used to validate the outputs of flooding models (Soti et al. 2010), to provide otherwise unmeasurable data such as flooding amplitudes in wetlands (Ogilvie et al. 2015) or in combination with other data, e.g. as inputs of rainfall, evaporation or soil moisture (Soti et al. 2010; Zribi et al. 2011) to hydrological models.
1.1.4
Questions and limitations over remote sensing
Despite highlighting the potential of remote sensing in hydrology, these studies also revealed problems of variable accuracy (Ran and Lu 2012; Annor et al. 2009), notably on smaller lakes (1-10ha) where errors due to the presence of flooded vegetation and shallow waters which affect the reflectance signal are more significant (up to 50%). Numerous spectral indices and classification methods exist to detect water bodies but no consensus exists on their performance and these are often tested and developed empirically on large water bodies (Feyisa et al. 2014). The continued research & development of sensors (new wavelength bands, increased temporal and spatial resolution), as well as innovative pretreatments and classification methods (enhanced corrections, cloud detection, vegetation discrimination, etc.) also open up enhanced possibilities in water detection/hydrology whose accuracy and ability
27
must be validated within different settings (lake size, shape, depth, turbidity; surrounding topography and land use; etc.). These must be continually confronted but also refined and calibrated against accurate ground truth field data. Ground truth in remote studies of small reservoir notably relied on higher resolution satellite imagery or GPS measurements (Annor et al. 2009; Liebe et al. 2005) which carry non negligible uncertainties, especially considering the size of the objects of interest. In several cases, calibration of indices was performed on a single date and field data was not always concomitant with satellite overpass (Liebe et al. 2005). Crucially, though smaller water bodies can be detected through remote sensing, its ability to monitor flood dynamics and provide long term water availability assessments must be investigated (Baup et al. 2014). Such interannual assessment of water availability have been performed with daily 250 m resolution MODIS images across all water bodies or focussing on a single wetland. Baup et al. (2014) using 350 m altimetry data studied water availability in 50 ha lakes. On smaller reservoirs (1-30 ha), 10 m to 30 m resolution studies provided a snapshot of flooding in reservoirs at a given time (Liebe et al. 2005; Feng et al. 2015), or between two dates (Annor et al. 2009). Due to physical and cost constraints, high spatial resolution (< 250 m) images to study medium catchments ( > 1000 km ) with high temporal resolution (< 16 days) remain inaccessible. Until low-cost high spatial and temporal resolution satellite imagery becomes available, the possibilities derived from compromises in terms of spatial resolution, spatial coverage and temporal resolution, must therefore be investigated and validated against hydrological field data. In addition to identifying current limitations, research must also help investigate potential applications of future remote sensing products such as SWOT, Sentinel-2, etc. Sentinel-2 will for instance provide 10m images of the entire globe, every 5 days, over 13 bands from visible to the short wave infrared, prefiguring enhanced opportunities. Suitable automation of water detection over several reservoirs and over several images also impose specific constraints on the methodology (Mialhe et al. 2008; Ogilvie et al. 2015), in terms of detection accuracy, threshold stability, and lengthy computing times when manipulating satellite imagery. Clouds which affect optical images also bear upon image availability and the sensor’s ability to provide long term monitoring. Furthermore, ways to combine this incoming array of increasingly reliable data with traditional hydrological field data must be investigated (Wackernagel 2004; Winsemius 2009). Satellite data may be used as an input variable or as a state variable which can then be used to continually correct model parameters through data assimilation processes such as Kalman filters (Moradkhani et al. 2005; Clark et al. 2008). Studies notably pointed to developing hydrological models of small reservoir catchments but possessed limited time series of field data (Liebe et al. 2005).
1.2
Aims & objectives
The aim of this PhD research was To investigate water availability patterns in small reservoirs and their influence on the practices of small holder farmers in a semi-arid catchment.
28
Specifically, this research sought to determine a suitable approach capable of meeting the following objectives: • To upscale water availability assessments across ungauged small reservoirs • To characterise water use across small reservoirs in the catchment • To identify hydrological and additional (historical, sociological, political, financial) drivers influencing the level of water use by small holders around small reservoirs Research focussed on the Merguellil upper catchment in semi-arid Central Tunisia where national water and soil conservations programmes since the 1960s led to the construction of over 50 small reservoirs. Faced with limited and localised resources as well as competition between upstream and downstream users (Leduc et al. 2007), concerns have been raised over the value of these small reservoirs, due to the evaporation losses and reduced levels of water use. At a local level, this research seeks to contribute to reflections over the value of small reservoirs in terms of their ability to provide sufficient and reliable water resources to small holders, and in terms of the benefits observed. This research notably feeds into ongoing cooperation between IRD and the Direction Générale de l’Aménagement et de la Conservation des Terres Agricoles (DG ACTA) and local representatives at the Kairouan and Siliana Commissariat Régional au Développement Agricole (CRDA). The insights gained should assist stakeholders in optimising water management strategies and agricultural benefits associated with small reservoirs, that may be applied in other semi-arid basins. The focus of several national and international studies over the past 40 years, it benefits from significant hydrological and climatic data, which here allowed the calibration and validation of remote sensing and numerical modelling methods.
1.3
Overall methodology
An outline of the research methodology is provided in the sections below, and detailed, referenced, methodology is provided in the relevant chapters. Research built upon the numerous previous studies (PhD/MSc thesis, Mergusie I and II project reports, Hydromed, Aquastress, CGIAR Comprehensive Assessment) in and around the Merguellil upper catchment, which notably yielded a number of agricultural surveys, substantial historical time series from the vast hydrological and climatic observation network as well as data on a subset of instrumented reservoirs. Details of this data is provided in section 2.3. Research benefited from the logistical support and existing collaboration of the Institut de Recherche pour le Développement (IRD) in Tunis with national authorities including the Direction Générale de l’Aménagement et de la Conservation des Terres Agricoles and the local representatives at the Commissariat Régional au Développement Agricole of Kairouan and Siliana, as well as the Institut National Agronomique de Tunis. Regular field visits and extended periods (Sept-Nov 2011 and March-June 2012) were spent in Tunisia. Regular meeting and visits to Kings College London were organised and the rest of the work was carried out at IRD in Montpellier, France. This PhD research was undertaken with support from the EU FP7 WASSERMed and the ANR Amethyst and Sicmed Mistrals Dyshyme projects.
29
Method Finer scale
Rapid surveys Numerical modelling
Remote sensing Quantitative questionnaires
Hydro climatic monitoring
Historical data Semi -directed interviews
Objectives
Agricultural practices & water uses
Water availability
Hydro-social systems
Socio-economic, institutional drivers
Figure 1.3.1: Schematic representation of overall methodology
1.3.1
Upscaling and combining hydro-social assessments
Part of the challenge resided notably in developing a suitable, original approach, capable of investigating both hydrological and social (here water uses and associated drivers) dynamics across a larger scale (here all reservoirs within a catchment). Research combined hydrological monitoring, numerical modelling, remote sensing and ethnographic methods while developing a nested scale approach, to provide varying depth of analysis, as we proceeded to upscale observations on individual small reservoirs to more than 50 small reservoirs scattered across the 1200 km catchment. Difficulties arise notably from the different boundaries of the objects under consideration (e.g. local withdrawals of lakes, wider benefits of groundwater recharge, wider community benefits or practices) which are compounded when combining approaches from different disciplines, where for instance, institutional drivers may operate at different scale to a reservoir or hydrological boundary (Riaux et al. 2014a).
1.3.2
Water availability monitoring across small reservoirs
To overcome data scarcity and difficulties encountered in modelling water resources in ungauged small reservoirs, the feasibility of combining field data, numerical modelling and remote sensing observations to assess water availability patterns over time in small (including very small) reservoirs was investigated. Chapter 3 sought to develop a suitable method based on Landsat optical imagery and validate it against the extensive stage data available over 7 reservoirs. Chapter 4 then sought to investigate how this remote sensing data could be used to refine water balance modelling of these 7 small reservoirs and associated water availability assessments. These sought to answer the following questions: How can the method be implemented? Are uncertainties due to temporal and spatial resolution acceptable to characterise flood dynamics? Are these uncertainties acceptable when considering 30
long term water availability? What are the benefits of data assimilation of remote sensing and field data through a Kalman filter on gauged and ungauged reservoirs? The resulting method was then applied across 48 reservoirs and results were then analysed in chapter 5 to understand the lake’s hydrological potential to support agricultural practices. Suitable satellite imagery and the necessary corrections to deal with sensor and atmospheric interferences including clouds were reviewed. Spectral water indices were compared and calibrated based on their performance, across a range of lakes of different size (1-30 ha), shape, depth, turbidity, algae, etc. and over several dates using high accuracy DGPS measurements of surface areas. The method’s ability to be automated and thus replicated over numerous reservoirs (space) and images (time) was investigated thanks to exceptional long term stage measurements over several years (up to 15 years) and across several reservoirs. The capacity of remote sensing to provide water availability assessments over 15 years within lakes displaying different flood dynamics (in terms of amplitude and duration) was investigated and discussed. Using significant field data and previous studies, the numerous variables required to model the behaviour of small reservoir were assessed and extrapolated across space (all reservoirs) and time (15 years). A combined rainfall runoff and water balance was then designed and tested against stage field data for the 7 small reservoirs. A data assimilation (Ensemble Kalman filter) method to integrate remotely sensed surface areas with the designed hydrological model was then developed and results compared on the 7 gauged reservoirs and on ungauged reservoirs. The water availability across all small reservoirs, including ungauged reservoirs, over 15 years was then assessed. The sensitivity of modelling outputs to field data notably rainfall, evaporation, levelling and remote sensing (accuracy) was studied. Data treatments were programmed in R which thanks to numerous packages allowed for the integrated numerical modelling, remote sensing treatments and subsequent data analysis within the same environment. The large data sets treated here had implications on the time required to download, treat and analyse the 546 Landsat images.
1.3.3
Water uses & hydro-sociological drivers on small reservoirs
Water uses associated with small reservoirs were examined through a combination of rapid surveys, quantitative questionnaires as well as semi directed interviews borrowed from ethnographic disciplines. A qualitative tool, semi directed interviews are by nature localised and limited in the number of people it can target, hence do not bear the same statistical weight as a large number of replies to traditional quantitative surveys for instance. These were not intended to be multiplied for the sake of increasing representativity (Beaud 1996), and instead sought to illustrate and shed light on specific complex issues which other (faster) methods are unable to. This limitation is addressed by a multi-stage sampling methodology (Mushtaq et al. 2007). Initial rapid surveys enabled preliminary scoping of the type of water uses and number of pumps across all reservoirs, leading to a selection of 22 reservoirs on which quantitative questionnaires were led. These sought to identify in detail agricultural practices, associated withdrawals and identify the reasons behind individual successes and failures, notably water access and management, livelihood strategies, government support, land rights. A further
31
four reservoirs were selected based on physical and social characteristics where in depth semi direct interviews provided an ethnographic description of the reservoirs. Focusing on the history of the site, the origins of the small reservoir and its management and water uses, discussions sought to understand in greater detail the the influence of water user associations, land and water rights, government incentives, economic difficulties and conflicts. The questionnaires repeated across 22 reservoirs highlighted the type of issues and their statistical distribution, while in depth interviews sought to analyse in finer detail the workings and consequences of these. Water availability calculated in the previous section was then compared to agricultural water uses observed on the reservoirs and the multiplicity of hydrological and sociological drivers influencing water uses were extracted. The emphasis was to explore the heterogeneity between reservoirs but also on a single reservoir, where inequalities (financial, political, etc.) were believed to heavily modify the social dynamics around reservoirs and the agricultural practices of individual farmers. Insights into how this interdisciplinary framework was developed and implemented with an anthropologist colleague will be provided in Riaux et al. (in prep.).
32
Tunis
"
Tunis ! (
" Kairouan
Béja ! (
Tunisia '
( !
'
Mediterranean sea
' '
( !
Siliana
Nabeul ( ' !
Zaghouan
' '
# #
# #
'
#
#
#
### # ' # ' # ' # # ## # ' ' '# ' ' #' ' ' # '' ''# ' '' '' ' # ' ' ' '#' ' ' ' '' ' ' '' ' ' # # # Kairouan '#' '# ' # '' # ' ' (# ! # # '' # ' ' ' # # # # '' # ' ''' # # # # ## #' '' ## # # # # ' # # ## #### Kairouan # # # ' plain # #' # # # ' Merguellil #
upper catchment
#
## #
#
#
### # #
#
( !
'
¯
#
Rainfall gauges
'
Instrumented SR
'
Small reservoirs (SR)
( !
Major towns
# '
Sousse
'
Elevation (m.a.s.l.) High : 1713 0
( !
5
10
Sidi Bouzid
20 Kilometres
Low : 0
Figure 1.3.2: Location of Merguellil upper catchment and of neighbouring hydrometeorological instrumentation
33
Chapter 2
Research site and overview of field data 2.1
Merguellil upper catchment, Central Tunisia
The case study area is the Merguellil upper catchment located in Central Tunisia, 60 km West of Kairouan, and whose downstream outlet is defined by the El Haouareb dam built in 1989 (figure 2.2.1). The dam, due to be called Habib in honour of Habib Bourguiba before President Ben Ali came to power, was built to protect the town of Kairouan from important floods, notably after the deadly 1969 floods (Poncet 1970), but also to mobilise water resources. The catchment is of strategic importance for the region as the intermittent Wadi Merguellil contributes to recharging the Kairouan plain aquifer with the Zeroud & Nebhana wadis. This alluvial plain extends over 3 000 km and its groundwater resources supply several thousand hectares (8 800 ha) of irrigated perimeters as well as the municipal and industrial water needs of the town of Kairouan. The Merguellil wadi previously flooded the Kairouan alluvial plain and recharged its phreatic aquifer directly but recharge processes were altered by the El Haouareb dam, where upstream runoff now infiltrates and mix with upstream (Ain Beidha) aquifers (Ben Ammar et al. 2006; Alazard et al. 2011). Representative of many problems found elsewhere in the semi arid Mediterranean basin, notably problems of uneven distribution of water resources and competition between upstream and downstream users (Leduc et al. 2007), groundwater overexploitation, increasing pressure from increased population, tourism and irrigated agriculture to feed national and international markets, and possible climate changes (García-Ruiz et al. 2011), the region has provided an interesting case study for numerous research project over more than 40 years: Mergusie I, II, Hydromed/Wademed, CGIAR Comprehensive Assessment, EU FP6 Aquastress, EU FP7 WASSERMed, Sicmed, Groundwater Arena involving several French (G-eau, LISAH, CESBIO) and Tunisian (ENIT, INAT) teams. These studies centred on biophysical aspects (Baba Sy and Besbes 2001; Cudennec et al. 2005; Kingumbi et al. 2007) throughout the catchment, as well as water uses (Faysse 2001; Feuillette et al. 2003) essentially on the Kairouan plain. Covering 1180 km , the catchment is situated in a semi-arid region, where average rainfall is low (265-515 mm/year increasing with altitude, Lacombe et al. (2008)) and characterised
34
by high annual and interannual variability (329 mm/year ± 131 mm). Intense events (exceeding 50 mm/h), during autumn and spring, cause wadis to overflow and 80% of the annual flow is produced over 12 days (Leduc et al. 2007). The name Merguellil actually refers to these floods, and translates broadly to the river “surging in the night”. Altitude varies between 200 m and 1200 m, with average altitude around 500 m, though 80% of the basin is relatively flat with slopes below 5% (Dridi et al. 2001). Mountain ranges are situated essentially in the North East and North West, though the Djebel Trozza is remarkably located in the downstream part of the basin, and leads to high runoff in the Zitoune and Hammam wadis. Average temperature is 19.2°C (10.7°C in January and 28.6°C in August (Zribi et al. 2011)) and potential evapotranspiration varies between 1.5-8 mm/day for an estimated total of 1 600 mm/year. Land uses are dominated by traditional Mediterranean crops, mostly cereals (30% of catchment surface area) and fruit trees (20%), especially olive groves suited to the extended dry season. Grasslands cover 30%, forest 19% and towns and water courses the remaining 1% (Dridi et al. 2001). 3 500 ha (i.e. 6% of agricultural land area) are irrigated, mostly from unregulated groundwater resources (Le Goulven et al. 2009). There are four interconnected (phreatic and deep) aquifers, covering 600 km , contained within geological formations, spanning the Trias to the Quaternary (Kingumbi 2006). The available piezometric data for the region reveals a lowering of the water table of 30 m over the past 30 years (19752005) in the Bou Hafna Oligocene & Cherichira aquifers (Kingumbi 2006), due in part to intense withdrawals in the region (exceeding 9.5 M m3 /year). Large withdrawals also supply the domestic needs of the coastal, tourist destinations. Upstream, surface and groundwater interactions remain largely unknown, though transfers have been observed in both directions, with floods recharging aquifers and groundwater contributing to river flow. Baseflow is estimated around 18 of annual flow at the outlet (Leduc et al. 2007). Further details on the basin’s climate and hydrology are provided in Ogilvie et al. (2016) in chapter 6.
2.2 2.2.1
Small reservoirs Small reservoir development
Small reservoirs were introduced in northern Tunisia at the beginning of the 20th century (Khlifi et al. 2010), before spreading south into semi-arid central Tunisia. Around Kairouan, these were built with the support of several governmental and international projects, starting notably with a Tunisian American project in the 1960s in the most upper Kairouan region (Selmi et al. 2001). Following a period of investment in larger dams instigated in the 1980s with the construction of large Nebhana dam, Sidi Saad dam on the Zeroud in 1982 and the El Haouareb dam in 1989, the 1st national strategy of water and soil conservation led to the construction of numerous (39) dams in the Kairouan gouvernorate during the 1990s. These investments were part of a nationwide strategy setting ambitious objectives of 1000 small reservoirs, 210 larger reservoirs and 20 large dams. By the late 1990s over 700 small reservoirs had been built nationally for an estimated capacity of 70 M m3 and led to a second phase after 2002, supported locally through a host of additional projects (CNEA 2006). These works were combined with additional water and soil conservation works, including contour benches and protecting river banks. These techniques aim to reduce soil erosion, harvest 35
'
# #
# #
' ' ' ' '
' '
' '
a '
##
' # ' ' ' '
'
#
' '
' ' # MORRA
#
# # # #
#
# '
# ' ###' # #
Mt Trozza
#
'
#
Kairouan plain
##
dam
River network
#
#
# # #
# #
# El Haouareb
e
# '
Kairouan
#
'
# #
#
#
ES SENEGA '
HOSHAS '
'
'
Zebbes
##
! (
#
#
#
BRAHIM ZAHER '
#
''GUETTAR '' # ' BEN NASSEUR ' FIDH # llil gue ' # r e M '' # ' FIDH ALI # #
#
Merguellil upper catchment
#
'
Zit ou n
' '
ABDESSADOK # '
' '
'
m
'
#
'
am
Skh ir
Ha m
#
#
# #
'
''' '# '
JEDELIANE
#
DEKIKIRA # '
JANNET # ' #
SADINE 1 '' SADINE 2
##
GOUAZINE # '
#
HADADA '
'
¯
#
#
#
#
# ## #
#
#
Rainfall gauges
'
Instrumented SR
'
Small reservoirs (SR)
! (
Major towns
# High : 1713
0
5
10
20 Kilometres
'
EL MOUIDHI
Low : 0
Figure 2.2.1: Merguellil upper catchment and location of instrumented small reservoirs (in bold, the lakes modelled in the following chapters) temporary rainfall for upstream users, and in certain cases recharge groundwater. Aggressive precipitation and runoff, combined with cropping and pasture lead to strong erosion, causing soil loss, gullies and the sedimentation of downstream dams. Specific erosion rates in the region vary from 1.8 t/ha/yr to 24.2 t/ha/yr (Albergel et al. 2003; Collinet and Zante 2005). Small reservoirs are built relatively cheaply (Wisser et al. 2010; CNEA 2006), using materials taken locally and composed of an embankment made from compacted soil and a stone cladding. An 80 000 m3 reservoir in 2003 cost around 200 000 DT, near 130 000 Ä at the time. The embankment heights range from 5 to 12 m for a length between 100-300 m. The dam is equipped in certain cases, with a spillway, a water intake tower connected to an outlet valve to lower levels and flush sediment. Initial design capacities range between 17 000 m3 and 1 590 000 m3 though most do not exceed 300 000 m3 (table 7.1.1). Despite these vast differences in water volumes, these lakes were built as part of the same water and soil conservation projects, sharing similar objectives, operation and management (CNEA 2006) and as a result were here considered jointly. For the sake of comparison, the El Haouareb dam has a theoretical 95 M m3 capacity.
2.2.2
Inventory of small reservoirs in and around the Merguellil upper catchment
An inventory of small reservoirs in the catchment was produced based upon cross referencing prior records from local authorities and literature (Kingumbi 2006; Lacombe 2007; CNEA 2006), satellite imagery and field visits. Certain coordinates previously transcribed upon 1: 50 000 IGN maps, led to substantial imprecision (mean 450 m, max 2330 m) and uncertain36
!
!
(a) Basin scale (1180 km )
(b) Zoom on two lakes: Gbatis (0.8 ha) and En Mel (5 ha)
Figure 2.2.2: Detecting lakes for inventory using MNDWI on Landsat imagery ties on the location of certain lakes, compounded by different naming conventions as well as incomplete metadata with certain coordinates provided under the local Carthage 1934 datum (associated with the Clarke 18800 IGN ellipsoid) and others under the international WGS 1984 datum. Available coordinates converted and projected to the WGS 1984 datum were overlaid on a classified Landsat image to allow rapid visualisation of the discrepancies and confirm the integrity of the records. The Modified Normalised Difference Water Index (MNDWI, Xu (2006)) which exploits the reduced reflectance in the Short Wave Infra Red spectral bands was applied to discriminate water bodies on 30 m Landsat images (cf. section 3.1.5.2). Where no field data is available to calibrate indices, unsupervised classifications, such as K-means (Jain 2010), may also be used to detect flooded areas. A single cloud free image, chosen when water levels were high across several lakes (autumn 2011) to increase the likelihood of identifying all lakes. Lakes as small as 0.5 ha were detected but the process required manual intervention (i.e. was not automated) to exclude additional pixels, essentially river stretches and temporary flooded areas. Additional filters based on shape, size, turbidity could automate the process when working over large areas. Similar assessment of water bodies have notably been carried out but focussed on larger reservoirs or used 2000 Landsat (GeoCover) images, meaning recent and small non permanent water bodies would be omitted (Verpoorter et al. 2014), and therefore could not be used here. High spatial resolution images made available freely through Microsoft Bing as part of the Bing Global Ortho project and dating for our region between April 2011 and June 2012 were then used in a second phase to cross check the inventory and confirm coordinates. The MDNWI spectral analysis step was essential to quickly distinguish water bodies which may not apparent on true colour composite imagery, especially if taken during the dry season. Where high spatial resolution multispectral imagery is acquired, these steps may be combined. Field surveys undertaken over several months over 2011-2014 then allowed to confirm the location and coordinates of each lake. Only 1 false positive was recorded, due to highly reflective rock formations, and helped identify 56 lakes in the Merguellil upper catchment (table 7.1.1), 10 more than the previous inventory. Three were built recently, but the others had been overlooked or possibly removed due to the heavy silting observed on some of these (cf. section 5.2.4). These small reservoirs, based on their catchment surface areas calculated 37
below, retain river flows and sediment over 21% of the basin and have a combined theoretical capacity of 10.7M m3 . In practice this is affected by silting, but the design capacity, remains considerable compared to the average annual runoff of 17 M m3 . Details are provided in table 5.2.1. In parallel, data from 13 reservoirs situated outside the catchment was also used in this study. These were part of the Hydromed project which sought to study the behaviour of small reservoirs across the Tunisian NW-SE mountain range during the late 1990s. Considering the significant hydrometric observations acquired on these sites and the similarities in geology, topography, rainfall, and corresponding behaviour and dynamics, data from a subset of these reservoirs was notably used here as in previous works (Gay 2004; Lacombe 2007). Observations notably continued on El Gouazine (hereafter Gouazine) after this initial project, providing reliable and prolonged observations. The updated inventory of lakes was then converted to .SHP and .KML files for use in ArcMap, Envi and Google Earth and notably mobile devices for use in the field.
2.2.3
Digital elevation models and reservoir catchment area delimitation
Catchment area delimitation is easily achieved using topographic information from maps, field surveys (when the scale is compatible) or increasingly based on remotely sensed digital elevation models (DEM) acquired from satellite, Lidar, aerial photography. The catchment of each reservoir’s catchment were here delineated based on the Aster, Shuttle Radar Topography Mission (SRTM, the successor to GTOPO30) publicly available digital elevation models. Catchments for certain reservoirs had previously been delineated using the Aster DEM (Proaño 2012) but revealed incoherences. Despite yielding a 30 m horizontal resolution, the “detail of topographic expression” is often nearer 100 m and 120 m as stated within the actual Aster readme file and other studies (Vanonckelen et al. 2013). The 2nd “finished” or “edited” version of SRTM with a 3 arc second horizontal resolution, i.e. 90 m along the equator, was therefore used and compared with the Aster outputs. Vertical error on SRTM is reputedly inferior to 16 m and even below 7 m in parts of the globe (Gorokhovich and Voustianiouk 2006). Provided through USGS, treated and void filled versions available via mirror sites were used (Ambiotek, CGIAR CSI). The Hydrosheds DEM (Lehner et al. 2013) based on SRTM but incorporating certain improvement was also used and compared. A later version 3 of SRTM was recently released providing data at 30 m resolution. Data, initially collected using C-band radar signals of wavelength (5.6 cm) at 1-arc second resolution, were resampled outside of the USA to 90 m for public distribution until the 23.09.2014. 30 m data has therefore been progressively released for part of Africa and Europe. The late release of the 30 m version was therefore not used for catchment delineation but was used to topographically correct Landsat images in chapter 3 and when interpolating rainfall in chapter 4. Resulting catchments defined using Geographic Information System tools, were then compared amongst each other and with high resolution satellite imagery (Google Earth and Bing) which display natural boundaries such as mountain ridges, but also river channels remarkable from the obvious erosion in the area. Contour lines and river networks from 38
Figure 2.2.3: Comparing SRTM (in grey) vs ASTER (in colour) catchment delineation for small reservoirs (green squares) digitised 1: 50 000 topographic maps (Lacombe 2007) were used to confirm their coherence. The beneficial complementarity of the Aster and SRTMv2 products was highlighted notably for delineating small catchments which here ranged from 0.4 to 22 km . Aster performed well on 85% of catchments while Hydrosheds and SRTM on 72% and 69% respectively. The higher planimetric resolution of Aster was beneficial here as the DEM was able to more accurately define areas which drain into another reservoir or inversely must be included (cf. top right of figure 2.2.3). Conversely, its lower topographic accuracy led it to exclude parts of certain catchments which were captured by the SRTM products (Hydrosheds and V2). Appropriate manual corrections were performed to combine as necessary the catchments delineated. The coherence and accuracy of the final delimited catchments was also confirmed during field visits by visualizing in real time on handheld GPS the catchment boundaries defined over ridges. The catchments were here defined to the dam wall. In theory when modelling runoff into the catchment we should exclude the surface area flooded as this input is already modelled as direct rainfall (which is equivalent to having a runoff coefficient of 1 for this area). Considering the size of the lake compared to the catchment size, the influence was negligible. Indeed counting a 2 ha lake even on the smallest catchment (0.4 km ) corresponds to overestimating runoff by around 5%.
2.2.4
Lake cells
In order to automate the remote sensing process (3) and allow the calculation of clouds, and water pixels, cells around each lake were also delineated. A similar approach was notably used to study flood dynamics within designated areas of a large wetland (Ogilvie 39
et al. 2015). Cells were manually created based on the high resolution images provided through Microsoft Bing interface (more recent and spatially defined for our catchment than on Google Earth). Automated drafting of the cell polygons using a classified Landsat image was not satisfactory considering the steps required to remove unwanted polygons (river sections notably) and adding extra polygons (as not all lakes were flooded) and remodelling to account for very high floods. Likewise automatic delineation of small reservoirs using SRTM based on topography was not attempted considering the very flat topography of certain lakes, the low spatial resolution and the low precision in altitude. Delineation sought to account for large floods by defining a 10% buffer around geomorphological boundaries of the lake, but meanders were excluded to conform with the surface-volume laws used (cf. section 2.4).
2.3 2.3.1
Hydro-meteorological observation network Data sources
Data collected through numerous research projects implemented over the past 40 years in the Kairouan region were reviewed and updated thanks to cooperation with the Direction Générale de l’Aménagement et de la Conservation des Terres Agricoles (DG ACTA) and Direction Générale des Ressources en Eau (DGRE) in Tunis and local representatives at the Kairouan and Siliana Commissariat Régional au Développement Agricole (CRDA). After 2011, data availability reduced drastically though following the Tunisian Revolution. Budget restrictions, staff turnover and policy uncertainties notably led to a freeze on transport and equipment costs which led to a decline in the quality of observations, as a result of insufficient maintenance and repairs and irregular remuneration of local observers. Changes were also introduced in the management of the observation network whose contract was discontinued due to performance issues. Remotely transmitted pluviometric and limnimetric equipment (operating through phone networks) was also introduced but combined with insufficient field visits to maintain and clean equipment, these time series could not be used here as suffered from significant uncertainties. The land on which certain stations are located was also the subject of tensions around its ownership with local populations. Cooperation agreements between the DGRE and IRD were also lately not renewed, preventing access to the limited ongoing data collection. Databases managed by the DGRE and the CRDA focussing on catchment hydrometeorology and a handful of reservoirs, as well as databases managed by the DG ACTA with support from IRD G-eau and LISAH centred on monitoring small reservoirs, notably those under the Hydromed project (Albergel and Rejeb 1997) were used. Data was collected as Hydraccess databases, excel files, observer sheets and unread cartridges from limnimetric and pluviometric instruments. The water availability assessment method was developed over 7 reservoirs. Selection of lakes was guided by data availability (stage, rating curves...) necessary to calibrate and validate the modelling and remote sensing methodology, differences in size, topography (shape, depth), as well as flood behaviour in terms of amplitude and duration to confront and assess the method’s performance in different settings. The characteristics of these lakes (n° 10 Hoshas, n°15 Fidh Ali, n°16 Fidh Ben Nasseur, n°19 Guettar, n°30 Morra, n°51 40
Gouazine and n°72 Dekikira) are highlighted in tables 7.1.1 and 7.1.3. Data availability (presented and discussed below) varied widely as a result of initial design differences in the observation network (implemented by different parties (government, research...) and for different purposes) as well as data acquisition difficulties (malfunctions, discontinuity) over the past 25 years. To extend stage and rainfall time series and the analysis & modelling possibilities, additional instrumentation on three reservoirs (Morra, Hoshas and Guettar) and support to ongoing observations on another reservoir (Gouazine) was implemented as part of this PhD. Data treatments to assess and extrapolate the multiple inputs required for hydrological modelling are discussed in chapter 4.
2.3.2
Rainfall
2.3.2.1
Monitoring equipment
A vast meteorological network composed of over 50 stations situated in and around the catchment was available (figure 2.2.1). Rainfall monitoring equipment consisted of manual rainfall gauges (pluviometres) read daily by paid observers and 13 automatic tipping bucket rainfall gauges (pluviographs). The latter were 400 cm (SPIEA type) rainfall gauges funnelling water onto a tipping bucket mechanisms wired up to electronic equipment capable of recording the number of oscillations of the buckets, and therefore instantaneous intensities of rainfall and distribution during the day. These are installed away from obstacles, horizontal surface at 1.5 m above ground level. Buckets are designed to tip under the weight of a known quantity of water. When connected to a surface area of 400 cm this corresponds to a 0.5 mm rainfall impulse. Calibrated within the factory, these must then be tested again on site, to identify any malfunction and adjust as necessary using a known water volume the capacity of each bucket. The tipping of the buckets are counted either mechanically and recorded on paper or electrically by counting impulsions. These are (generally) also connected to a totalising rainfall bucket to cross check the measurement. These stations are located as part of larger meteorological stations, or placed individually throughout the catchment, near gauging sites, on lakes and in upstream areas. Time series began as early as 1888 in Makhtar but are subject to frequent data gaps due to dysfunctions, interruptions and errors introducing reliability and representativity errors which must be accounted for. Of 5 reservoirs instrumented at some point in the catchment, rainfall gauges on only two small reservoirs continued to be partly maintained by the DG ACTA (Gouazine & Guettar) after 2011. On Gouazine, IRD provided assistance to download and extract the data from the old Elsyde Oedipe electronic equipment and pay regular visits to the local observers. At El Guettar, the rainfall logger was not accessible until 2012 when malfunctions were identified. In parallel, a pluviograph had been provided for the local observer to record daily rainfall, however the records provided were inadequate, due to inadequate training and knowledge. As highlighted in chapter 5, this remunerated role was historically attributed to the dam operator, and was associated with political influences and can therefore difficultly be changed. On Hoshas and Morra existing equipment was refurbished, using new mercury circuits for the tipping bucket mechanisms and Hobo devices to record the electric pulses created with every tilt of the buckets. Manual rainfall gauges read by local observers were also installed to provide daily measurements and cross reference the automatic measurements (figure 2.3.1). 41
(a) Pluviograph (left) and pluviometre (right) and evaporation pan (front) located on the Gouazine dam wall
(b) Installed (pluviometre)
manual
rainfall
gauge (c) Tipping bucket rainfall gauge (pluviograph) with replaced mercury contactors
Figure 2.3.1: Rainfall and evaporation monitoring equipment on SR in Merguellil catchment
42
Dekikira
Guettar
Small reservoir
Fidh Ben Nasseur
Fidh Ali
Hoshas
Gouazine
Morra2
Morra1
1995
2000
2005
2010
2015
Figure 2.3.2: Rainfall data availability for the 7 SR modelled in the following chapters
43
2.3.2.2
Data availability
Data from our installed Hobo devices (31 files from 2 stations) were downloaded during fieldwork and ongoing field visits by IRD Tunis staff. Data was then exported from the proprietary Hoboware software into R to extract, compile and convert instantaneous 0.5 mm counts to daily intensities. Calibration and malfunctions noted during field visits were removed accordingly and to maintain homogeneity with previous records and values provided by local observers, rainfall for a given day was summed over the period 07.00 AM until 07:00 AM of the following morning. As in chapter 6, time series were subjected to homogeneity checks using double mass curves (Brunet-Moret 1971). Rainfall data availability is shown in figure 2.3.2 for the 7 small reservoirs we focussed on. Data availability was notably impacted by continuity issues due to research projects ending, as well as problems with monitoring equipment which introduced gaps within time series. These included clogging from silt, leaves, insects, tipping bucket malfunction over time, and mercury circuit damage and led to complete loss of data or underestimation. Rainfall interpolation using the whole database of rainfall gauges sought to minimise the consequences while improving where possible estimation in the upper catchment of each reservoir. Data availability however declined significantly after 2011, as a result of the difficulties mentioned above, leading to rainfall data being available only at the four reservoirs we monitored, the El Haouareb dam and the Skhira station upstream. Alternate sources of data, including national statistics provided through the NOAA observation network were not considered here, due to the reduced number of stations available (no stations situated within the basin). Remote sensing data such as TRMM (Tropical Rainfall Monitoring Mission) was considered at the basin scale, but results confirmed the difficulties for such products to provide accurate rainfall estimates at this spatial scale and daily or event time scale (Soti et al. 2010; Ogilvie et al. 2016).
2.3.3
Lake pan evaporation
Evaporation observations from Colorado type sunken pans installed on the shores of several lakes in and around the catchment were collated from the multiple data sources and complementary data for Gouazine and El Mouidhi provided by the DG ACTA. Pans are 1 m square Colorado pans, and often 50-60 cm deep but under Hydromed these were 1m deep, the first 60 cm of which were buried. Paid observers take daily measurements of the volume withdrawn using an incorporated millimetric stage ladder or based on the volume of water added to reach the level (e.g. 1 litre equals 0.9 mm). Daily rainfall values are subtracted, leading to negative values in the observer readings due to errors, which were therefore set to 0. For El Haouareb, values provided by the authorities already incorporate a 0.76 transposition coefficient to convert pan values to lake evaporation, and were therefore converted back to pan evaporation values for comparison with other sources (Alazard et al. 2015). Transposition coefficients are required due to the difference in conditions notably the greater heating experienced by the pan sides. Floating pans, situated on lakes can help reduce this problem but present greater operational difficulties and errors due to wind & waves can occur. Out of 25 sites scattered along the Tunisian Dorsale mountain range, 4 are situated in the catchment and 9 within 25 km of its boundaries. Data availability illustrated in figure 44
Dekikira
El.Mouidhi
Brahim.Zaher
Abdessadok
Sadine.1
Jedeliane
Fidh.Ali
Fidh.Ben.Nasseur
Gouazine
Morra
Jannet
Hadada
El.Haouareb 1990
1995
2000
2005
2010
2015
Figure 2.3.3: Availability of evaporation data for small reservoirs in and around the Merguellil upper catchment (in bold, the lakes modelled in the following chapters) 2.3.3 shows the high availability in El Haouareb (1989-2015) and Gouazine and El Mouidhi, as well as data availability for 10 small reservoirs during the late 1990s under the Hydromed project. Data was available for 3 of the 7 small reservoirs over which the water balance modelling was developed (Gouazine, Dekikira, Fidh Ben Nasseur). On other lakes, due to absent or incomplete data, this was interpolated as described in section 4.3.2.
2.3.4
Small reservoirs hydrometric data
2.3.4.1
Equipment status
As part of previous research projects, 18 reservoirs in and around the catchment were equipped with limnimetric stage ladders read daily by paid observers and automatic stage pressure transducers over the past 20 years. Considering the decline in the number (figure 2.3.5) and quality of observations partly due to political and financial difficulties after the 2011 Tunisian revolution, complementary instrumentation was implemented on three small reservoirs: Guettar, Morra and Hoshas. At Gouazine, the existing equipment (SPI-3 type transducers connected to a Chloé data station) after inspection with the DG ACTA was retained. On Morra and Hoshas, where automatic stage monitoring had ceased, Schlumberger Water Services (SWS) Mini-Diver pressure transducers were installed for 15-minute logging of water levels (figure 2.3.4). On Guettar complementary SWS Mini-Diver pressure trans45
(a) Limnimetric ladders
(b) Pressure transducer on Morra SR. Here, the SWS mini diver was not installed in a tubewell as the lake doesn’t dry out and data must be downloaded regularly
(c) Specific casing created to prevent theft on SWS mini divers
Figure 2.3.4: Limnimetric monitoring equipment
46
Jede ane Es Senega Sad ne 2 Sad ne 1 M che E Anse E Mou dh Dek k a B ah m Zahe Baoue e Abdessadok Hadada Hoshas Janne Mo a Gue a Gouaz ne F dh A F dh Ben Nasseu 1995
2000
2005
2010
2015
F gure 2 3 5 Ava ab ty of stage data for sma reservo rs n and around the Mergue upper catchment ( n bo d the akes mode ed n the fo ow ng chapters) ducers were nsta ed by IRD staff once the data qua ty ssues of the ex st ng transducer were d scovered These are compact dev ces wh ch can be used n var ous sett ngs (we s akes etc ) to record var at ons n the water co umn above them These prov de cont nuous measurements at user defined (15 m n) nterva or when the stage var at on exceeds a certa n va ue or percentage They a so record temperature wh ch prov des an easy way to detect whether the transducer s exposed to the a r or not due to the arger amp tude of temperature var at ons of a r than water They must a so be compensated for atmospher c pressure us ng comp ementary Baro-Divers p aced n contact w th atmosphere Three were nsta ed throughout the upper catchment one at E Haouareb (for other research works done on the dam) one at Hoshas (for Guettar and for Hoshas Mini-Divers) and one at Morra Equ pment was nsta ed read and ma nta ned dur ng my fie dwork and fo owed up thanks to regu ar fie d v s ts by IRD Tun s staff 2.3.4.2
Data treatments
Stage data for sma reservo rs was comp ed from ex st ng databases and comp ementary fie d data acqu red from IRD and partners (DG ACTA and CRDA) up to 2014 Based on the qua ty of the observat ons and n order to be comb ned w th the h gh Landsat ava ab ty after 1999 data from the 7 sma reservo rs were used to mode water ava ab ty
47
Sadine 2 Sadine 1 Morra M'richet El Anse Jedeliane Jannet Hoshas Hadada Gouazine Fidh Ben Nasseur Fidh Ali Es Senega El Mouidhi Guettar Dekikira Brahim Zaher Baouejer Abdessadok 1990
1995
2000
2005
2010
2015
Figure 2.4.1: Available rating curves for small reservoirs in and around Merguellil catchment (in bold, the lakes modelled in the following chapters) (Dekikira, Gouazine, Morra, Hoshas, Guettar, Fidh Ali, Fidh Ben Nasseur) and to develop the Ensemble Kalman filter approach. Complementary stage data obtained on two lakes with SWS Mini-Divers was atmospherically corrected and regular field observations at the stage ladder were used to calibrate the water column. The drift over time of the equipment remained low, generally under 5 cm over 3 months. This can be explained firstly by the errors in ladder readings, which are of the order of 1 cm when waters are calm, 2 cm when the lake is choppy. Additionally the rope (initially fishing line and then kevlar) used on Hoshas and Guettar can stretch a little. The divers are subject to minimal error: Mini-Divers used have 10 m range with 0.2 cm resolution and theoretical precision ±0.25%. Instantaneous (15 min) values from pressure transducers provided valuable information on the flood peaks and short term flood dynamics (infiltration, releases, etc.). For daily modelling of the small reservoirs, the value at 7 AM was used in order to allow time series to be combined and corroborated with the field observations. Daily observer readings (precision 0.5 cm) where available, were collected, input manually and used to cross-check automatic readings.
2.4 2.4.1
Deriving power relations for height-surface-volume HSV and power relations
Height-surface-volume (HSV) rating curves available for the 18 lakes instrumented in and around the Merguellil catchment are shown in figure 2.4.1. These are notably required to compare surface areas of small reservoirs with remotely sensed areas (chapter 3) and for water balance modelling (chapter 4). Four of the 7 lakes used in this study to develop the method in chapters 3 and 4 were regularly surveyed (every 2 years) under the Hydromed 48
Figure 2.4.2: Topographic survey of Hoshas small reservoir project to account for silting. On Morra the original rating curve was not updated after 1.1.1992 (Pabiot 1999) and some uncertainty remains over the origin of the rating curve, possibly derived from the design curve. Due to the lake remaining flooded throughout the year and the difficulties in bringing a boat to this remote location, no updates to this rating curve have yet been undertaken. Lake Guettar was surveyed in 2002 and Hoshas was surveyed in 2014 by IRD Tunis staff following the acquisition of a Differential GPS (DGPS). The provided DGPS measurements were then interpolated based on Delaunay Triangular Interpolation Networks (TIN) well suited to bathymetric applications to derive height-surface-volume relationships (figure 2.4.2).
2.4.2
Power relations
In the absence of HSV curves for ungauged reservoirs, power relations can be developed. These have been previously used in numerous studies and can be derived based on surveys over a subset of the lakes studied (table 2.4.1). The relationship must where possible be locally determined to account for geomorphological differences which are therefore assumed to be relatively regular across the region of interest. Power relations are based on the principle that when representing the lake as a half pyramid (i.e. triangular shape lake), volumes (V ) and surface area (S) are related as per equation 2.4.1) where B is a constant and — = 1.5. In practice, the lakes do not follow a true pyramid shape, and if slopes are convex — < 1.5 and — > 1.5 if slopes are concave (Liebe et al. 2005). V = B ú S—
(2.4.1)
Similar relations (equation 2.4.2 and 2.4.3) can be defined between height and volume 49
12.5
y = 3 + 0.68 ⋅ x, r 2 = 0.99
ln(V)
10.0
7.5
5.0
2.5 6
8
10
ln(S)
Figure 2.4.3: Linear regression to derive coefficients of power relation between surface area (S) and volume (V )
y = 939 + 1.1 ⋅ x, r 2 = 0.995
V from S−V power relation (m3)
2e+05
1e+05
0e+00 0e+00
1e+05
2e+05
V from S−V rating curve (m3)
Figure 2.4.4: Correlation between volumes estimated from rating curve and through S-V power relation
50
Table 2.4.1: B and — values for S-V power relations for small reservoirs Study
Area
—
B
Gourdin et al. (2003)
Cote d’Ivoire
1.4953
0.00405
Cadier (1996)
Brazil
1.5882
0.003549
Lacombe (2007)
Tunisia
1.7299
0.001413
Liebe et al. (2005)
Ghana
1.4367
0.00857
Sawunyama et al. (2006)
Limpopo
1.3272
0.002308
Ogilvie (this study)
Tunisia
1.6282
0.003935
based on a triangular representation (Lacombe 2007). V = A ú H–
(2.4.2)
S = – ú A ú H –≠1
(2.4.3)
The parameters for individual lakes can then derived by simple linear regression of equation 2.4.4 using the surface and volume data from instrumented lakes which yields the values of — as the slope and ln(B) as the intercept as illustrated in figure 2.4.3. In certain cases (Cadier 1996; Lacombe 2007) lower values (corresponding to under 15% maximum volume) are removed and points are weighted by their volume. These seek to remove the points where the lake profiles are less reliable. Here this was shown to improve estimation at the upper range and therefore R (as similar % errors on large values affect R more strongly) but at the lower range leads to larger errors. On Gouazine for instance using this method leads to errors below 1% above 4 m (4.6 ha) but over 40% at 1 m (1.05 ha). Using all values leads to errors of 8% and 19% respectively (figure 7.2.1 and table 7.2.1). Considering the fact that water levels on several lakes never reach the theoretical upper range of the rating curve (7 ha at Guettar, 15 ha at Morra) and are typically within the lower half (when not completely dry), we chose to favour the representation of the lower volumes in the power relation. The resulting relationships for Gouazine are illustrated in figure 2.4.4. R levels of fit for power relations on all lakes were excellent (R =0.99) in line with high values found by previous studies (Liebe et al. 2005; Sawunyama et al. 2006) and confirmed the possibility of using power relationships across a variety of lake morphologies. ln(V ) = ln(B) + — ú ln(S) 2.4.2.1
(2.4.4)
Modelling a common S-V power relation
A common inter lake (“interSR”) power relation based upon 11 available rating curves for reservoirs in and around the catchment was developed here and tested against four further rating curves. Initial rating curves were used as silting introduces additional complexity and irregularities in the shape of lakes and therefore errors on rating curves. The progressive modification of the parameters due to silting is discussed in section 2.4.3. Rating curves acquired after the construction were therefore not used here (Guettar, Hoshas), nor for those too far away (Zayet, Mrira, etc.) or of significantly different sizes (Jedeliane, El 51
y = −5.6 + 1.6 ⋅ x, r 2 = 0.97
SR
10
Abdessadok Brahim Zaher El Mouidhi
LN_V
Fidh Ali Fidh Ben Nasseur 5
Gouazine Hadada Jannet Morra Mrichet Sadine 2
0
2.5
5.0
7.5
10.0
LN_S
Figure 2.4.5: Relationship between surface area and volume based on 11 SR rating curves and the interSR relation (blue line) derived from linear regression
3.0 Sadine 1
●
beta
2.5
Mouidhi
2.0
Baouejer
●
Fidh Ben Nasseur Saadine 2 Abdessadok Brahim Zaher Morra Dekikira ● ●
Fidh Ali Inter−SR Janet
Hadada
1.5
Es Senega El Gouazine M'Richet
0.00
0.01
0.02
0.03
B
Figure 2.4.6: B and — parameters for each of the 15 lakes 52
Table 2.4.2: Error on volume from interSR power relation against volume from lake specific power relations 1 ha
2 ha
3 ha
4 ha
5 ha
6 ha
7 ha
mean
Es Senega
2.73
7.18
13.44
18.10
Baouejer
43.14
16.91
3.85
Sadine 1
35.55
73.09
Dekikira
104.54
101.87
100.33
99.24
Abdessadok
12.57
0.62
7.61
12.27
8.27
Brahim Zaher
46.38
33.87
27.06
22.43
32.43
Gouazine
53.03
73.29
86.35
96.22
104.23
Fidh Ali
11.61
12.23
12.58
12.84
13.03
Fidh Ben Nasseur
32.23
41.73
Hadada
32.64
29.89
28.23
Janet
9.96
6.81
4.91
3.55
Morra
7.80
9.49
10.46
11.14
M’Richet
5.65
10.02
20.36
12.01
Mouidhi
10.04
15.41
27.47
17.64
Saadine 2
28.47
37.88
42.79
36.38
10.36 21.30 54.32 98.40
97.72
111.01
97.14
116.93
99.12
102.14 12.46 36.98 30.26 6.31
11.66
12.09
12.45
11.78
Haouareb). The full range of values provided in the rating curves were exploited and not just those for larger volumes as stated above. Here rather than averaging the parameters from each lakes (Lacombe 2007) which gives greater importance to the parameters of lakes with significantly different values, the parameters of the interSR power relation were assessed as above through a linear model which seeks to minimise the errors (i.e. here optimise the Nash Sutcliffe Efficiency) between the interSR predicted volume and individual rating curves volume for a given surface area (figure 2.4.5). V = 0.0039352 ú S 1.6282
(2.4.5)
The resulting parameters (equation 2.4.5 ) and figure 2.4.6 reveals that on the whole lakes here have predominantly concave slopes (— >1.5). Across all lakes, the optimal — varied moderately (coefficient of variation of 19%) while values for B varied significantly (CV = 137%). The variability in the values was similar to that of a previous study in the area (Lacombe 2007), though absolute values were different. 2.4.2.2
Validating a common S-V power relation
Tested on 4 small reservoirs (Es Senega, Sadine 1, Baouejer, Dekikira) of different shape and sizes not used to derive the interSR power relation, it worked well on two lakes (Es Senega and Baouejer) where mean errors were 10% and 21% respectively (table 2.4.2) over the range of floods experienced by these lakes (1 ha - 4 ha). On two other lakes (Sadine 1 and Dekikira) errors were much more significant reaching 54% and 99% respectively. Over all 15 lakes, the mean error within the range of flood they experienced was 33%, but results 53
Es.Senega
Baouejer
Sadine.1
5e+04
5e+04
5e+04
Dekikira
Abdessadok
6e+05 4e+05 2e+05 0e+00 1e+05
Brahim.Zaher
1e+05
Gouazine
1e+05
Fidh.Ali
5e+04
1e+05
Fidh.Ben.Nasseur
5e+04
1e+05
Hadada
Volume (m3)
6e+05 4e+05 2e+05 0e+00 5e+04
1e+05
Janet
5e+04
1e+05
Morra
5e+04
1e+05
M.Richet
5e+04
1e+05
Mouidhi
5e+04
1e+05
Saadine.2
6e+05 4e+05 2e+05 0e+00 5e+04
1e+05
5e+04
1e+05
5e+04
1e+05
5e+04
1e+05
5e+04
1e+05
Surface area (m2)
Figure 2.4.7: Lake specific S-V power relations against the interSR power relation (in grey)
54
showed the significant differences in the ability of a single power relation to model the S-V of individual lakes. The inter SR performed well on 8 out of 15 lakes (errors < 25%) of varying sizes and depth. On 5 lakes, however, errors reached between 30% and 55% and on two lakes 100%. The relationship notably performed poorly on small lakes where the slopes were concave (— > 1.5) (Sadine 1, Sadine 2, Fidh Ben Nasseur). Higher — values correspond to lakes with steep banks, where errors in the power relation are greater as minor shifts in S or the parameters lead to greater volume errors (Liebe et al. 2005). Using a different relationship for the lakes where — congregates around the lower pole of 1.4 - 1.7 (figure 2.4.6) and for lakes where — values congregate above 1.7 did not improve performance (figure 7.2.2). Performance was also lower for two larger, shallower lakes but no correlation was identified between — and B parameters and lake characteristics to further refine the parameter values for each lake (figure 7.2.4 and 7.2.5). At a local scale, surface area, volume and notably mean depth which is related to the shape of the lake floor bed and the bank slopes can be expected to be correlated with —. Lakes include typically shallow oval shaped lakes situated in floodplain (Gouazine) and deeper triangular lakes situated within stream channels (Morra). Here the absence of correlation can be partly explained by the fact that mean depth does not sufficiently account for differences in the slope of lake banks. The presence of multiple sets of optimal parameters being available as a result of the significant influence of B is also expected to have influenced the result (i.e. here for a shallow lake we can use — >1.5 as B corrects for this). In the absence of field data on all lakes or variables to differentiate parameters for types of lakes, a generic but site specific power relation must be used. Errors though significant, remain comparable to those from other methods such as Fowler and Nelson relations (Sawunyama et al. 2006). The influence of silting is discussed below. The poor fit of power relations derived for other areas are shown in figures 7.2.2 and confirm the need to derive a local relationship which reflects the local geomorphology of lakes. Relationships from Limpopo performed remarkably poorly, indicating that lakes there were much shallower than in the Tunisian context. The relationship derived for açudes in Brazil showed the greatest similarity with the Tunisian context.
2.4.3
Silting
2.4.3.1
Erosion rates
Intense rains, high runoff combined with degraded and uncovered soils lead to significant erosion in Mediterranean areas. Exceptional storms (over 100 mm/day) can scar the landscape but erosion in Mediterranean areas, unlike tropical areas, is generated predominantly by runoff which occurs when soils are saturated (Roose 1991). A detailed study on a single lake in the catchment (Fidh Ali) showed that sheet runoff mobilised 10 to 50 times more sediment than erosion from rainfall (Collinet and Zante 2005). In practice, erosion is very heterogeneous as a result of localised intense rainfall and runoff, as well as differences in the local lithology & pedology, catchment size, land cover (cropland, forest...) as well as efforts to reduce soil erosion notably through contour benches. Large events as shown in chapter 6 occur around 5 times a year but the spatial and temporal variability means certain parts of the catchment may be unaffected for several years. Studies in the region based on successive 55
Table 2.4.3: Silting and associated capacity loss measured on 14 small reservoirs between 1988 and 1998 (compiled from Albergel et al. 2004) Lake
Silt per year (m3 )
Capacity loss per year (%)
Es Senega
3108
3.87
Baouejer
930
1.41
Sadine 1
4415
9.92
Dekikira
4260
1.94
Abdessadok
3245
3.51
Brahim Zaher
4153
4.82
Gouazine
2004
0.85
Fidh Ali
6605
4.90
Fidh Ben Nasseur
1524
3.23
Hadada
3515
4.14
Janet
9318
9.88
M’Richet
1239
2.92
El Mouidhi
3855
2.70
Saadine 2
10285
12.54
Mean
4176
4.76
levelling when lakes are dry and bathymetric measurements of many lakes showed that specific erosion rates varied from 1.8 t/ha/yr to 24.2 t/ha/yr (Albergel et al. 2003; Collinet and Zante 2005) on Gouazine and Fidh Ali respectively over the 1990-1999 period. Clayey soils protected by a layer of stones are less affected than marly soils found at Fidh Ali. Lower crop development (88% vs 55% Fidh Ali and Gouazine respectively) and greater forest cover play a role, as well as the greater contour bench development on Gouazine (30%) vs 5% on Fidh Ali. Silting accordingly is not proportional to the catchment area and local studies (Hentati et al. 2010; Collinet and Zante 2005) pointed to the difficulty in identifying the factors which determine the vulnerability to erosion and rate of silting though the frequency of events and the land use (cropping) during strong rainfall events is determinant. Their influence is also not constant, as WSCW are shown to degrade and breaches potentially increasing erosion processes (Baccari et al. 2008), while farmland abandon & forest expansion as seen in other Mediterranean areas can also in time modify erosion patterns and amplitude (García-Ruiz et al. 2011). 2.4.3.2
Silting in small reservoirs
Silting has a detrimental effect on reservoirs as it reduces their capacity, eventually leading to a finite lifespan when they will cease to contain water, as discussed in section 5.2.4. Here, based on successive observations during the 1990s, Albergel et al. (2003) showed that the decline in capacity despite its heterogeneity could be modelled through linear regression. Over several years, mean silting in the catchment led to an average capacity loss of 4.6% year and 22% over 7.7 years based on 19 small dams and 5 larger dams (Albergel et al. 2004).
56
Certain lakes such as Gouazine benefited from low levels of silting (0.9%/year capacity loss,CNEA (2006)), while others had more significant capacity loss (4.6%/year for Fidh Ali). Studies similarly showed that 71% of lakes would last beyond 30 years, but some such as Fidh Ali would not exceed a 10 year lifespan (Collinet and Zante 2005). In fact, by 1999 after 9 years, the lake had lost 54 000 m3 (33%) of its maximum capacity, while another lake (Sadine 2) lost 61% within 3 years and 100% by 6 years. Lakes can fill slower but also faster than expected due to a single large event (Ben Mammou and Louati 2007), notably when these are under dimensioned (Habi and Morsli 2011). Silting was notably strong on Sadine, Fidh Ali and Jannet where mean silting reaches 8700 m3 /year compared to only 2930 m3 /year on the other 11 SR (table 2.4.3). 2.4.3.3
Accounting for silting in HSV for ungauged lakes
Rating curves presented in section 2.4 were updated over time to account for silting. Figure 2.4.8 shows how the S-V relations evolve as a result of silting for Gouazine and Fidh Ali where silting was minimal (11% loss in maximum capacity over 17 years) and significant (33% over 9 years) respectively. Considering the impossibility of levelling and constantly updating all lakes within a catchment, power relation rating curves can be adapted for silting through a progressive shift in the parameters (Cadier 1996; Lacombe 2007), though in practice, silting occurs from sudden, discrete events, not a linear incremental process. Furthermore difficulties occur as silting is heterogeneous within lakes. In Fidh Ali for instance the alluvial cone created at the mouth of the river rose by 4.5m between January 1993 and September 1998 but on 75% of the lakebed surface area, the layer of sediment does not exceed 1m depth (Albergel et al. 2004). The — and B parameters for 70 S-V power relations on all 15 lakes over the 1990-2007 period were calculated. The coefficients for these 70 relations are illustrated in figures 7.2.6a) & b). Apart from Mouidhi and Sadine, — is shown to increase gradually over time. The slope of — values against time for these (12) lakes follows a linear trend, increasing by 0.03119/year. This is coherent with the result found by Lacombe (2007) on – and the relationship between – and —. Lacombe (2007) had also noted the particularity of Mouidhi and Sadine whose rapid silting and small size are understood to lead to these discrepancies. The evolution over time of the power relation was here calculated for all lakes in the catchment by setting a gradual annual increase of — by 0.03119 and an associated decrease in maximum capacity Vmax of 2930 m3 . Initial Vmax was here known from the inventories and used to calculate Smax calculated based on the initial interSR B and — parameters. By supposing that the maximum surface area at the spillway does not evolve over time, which is acceptable based on the true rating curves, the resulting B was calculated over time based on the annual increase of — and an updated Vmax . Vmax was updated by reducing it by a conservative average silting value of 2930 m3 . The power relation parameters and their evolution over time were then tested on 4 small reservoirs (figures 2.4.10 and 2.4.11). Despite the heterogeneity in the lakes morphology and the rate of silting, the interSR power relation adapted over 25 years (i.e. up to 2014 for lakes built in 1990) for silting succeeded in containing errors around 50%. On Fidh Ali and Gouazine the parameters used were able to shift the power relations and maintain acceptable
57
100000
1996 1995 2001 2004 ●
1990
2007 2006 2005
1993
75000
Volume (m3)
●
50000 ●
●
25000 ●
●
●
0
● ● ●
0
20000
40000
60000
Surface area (m2)
(a) Gouazine
100000 ●
1990 1994 1993 1997 1999 1995
●
75000
Volume (m3)
●
●
50000 ●
●
●
25000 ●
●
● ● ● ●
0
● ● ● ●
0
●
10000
20000
30000
40000
Surface area (m2)
(b) Fidh Ali
Figure 2.4.8: Change over time of S-V relationships on two lakes 58
levels of errors (under 45%) relative to the site specific updated surveys. On Sadine 2, errors after 5 years increased as the extreme silting was less well modelled leading to mean relative errors of 50%. On Dekikira after 11 years, the estimated power relation modelled the decline remarkably well with relative errors remaining around 55%. On Gouazine, even after 17 years relative errors remained below 80%. 2.4.3.4
Discussing HSV and silting errors
Absolute errors are significant and increase over time and at the higher range of surfacevolumes, as can be expected from seeking to model morphology of different lakes and account for their heterogeneous silting. Nevertheless, for water balance modelling, the absolute volume is not used, but rather the daily change in volume which except at the lowest levels, will not be affected by silting. Figure 2.4.8 indeed shows how the slope of the relationship remains relatively constant over time at the higher range as indeed silting does not modify the morphology of the upper parts of the lake. Silting however affects absolute volumes and therefore the water availability assessments performed in chapters 4 and 5. The sensitivity of results to these uncertainties was therefore assessed and showed that for a majority of lakes in the basin, these were not significant when considering annual or seasonal water availability due to the very small and sparse volumes available (section 5.2.4).
59
3.5
Lake ●
Abdessadok Baouejer
3.0
Brahim Zaher Dekikira
beta
El Gouazine Es Senega 2.5
Fidh Ali
●
Fidh Ben Nasseur
● ●
●
Hadada ●
●
2.0
Janet M'richet
● ●
1.5
0
5
10
15
Years
(a) —
0.015
Lake ●
Abdessadok
0.010 Baouejer Brahim Zaher El Gouazine
B
Es Senega Fidh Ali Fidh Ben Nasseur Hadada
0.005
Janet ●
M'richet
●
● ●
0.000
●
0
●
●●
5
●
●
10
15
Years
(b) B
Figure 2.4.9: Modelled evolution (black line) of B and — parameters over time
60
1990
1993
4e+05
1995
●
●
● ●
●
● ●
● ● ●
1e+05 ●●
●
●
● ● ● ●● ●● ●●
● ● ● ●● ●● ● ●● ●● ●
1996
●
●
● ● ●
● ● ● ●● ●● ●● ●● ●●
2001
3e+05 ●
2004
●
●
●
● ● ● ●● ●● ●● ●● ●●
● ●
Power relation
●
● ●
● ● ●
●
● ● ● ● ●
1e+05
●
●
●
● ●
●
●
● ● ● ● ●
●
●
2e+05
●
●
●
● ●
Volume (m3)
●
●
● ●
●
● ● ●
4e+05
0e+00
●
●
●
●
2e+05
0e+00
●
● ●
3e+05
● ●● ●●
● ● ● ● ●●
2005
● ●
● ●
● ●
● ●
● ●
● ●● ●● ●●
● ● ● ● ●●
2006
●
Modelled
●
Lake and date specific
● ●
● ●
2007
4e+05 ●
●
●
3e+05
●
●
●
● ●
● ●
0e+00
● ●● ●●
0
●●
● ●
●
● ●
● ● ● ●● ●●
80000
0
● ●
● ●
●
●
●
●
●
40000
●
●
●
●
1e+05
● ●
● ●
2e+05
●
●
● ●
● ● ● ● ● ●
● ● ● ● ●●
40000
80000
●●
0
●
40000
80000
Surface area (m2)
(a) Gouazine
1990
1993
1994
●
●
●
2e+05 ● ● ● ●
●
●
●
● ●
1e+05
●
●
●
●
Volume (m3)
0e+00
● ●
●
●
●
●
●
●
●
●
● ● ● ● ● ●● ● ● ● ●●●● ●●● ●●●● ● ●●
● ● ● ● ● ● ●● ● ●●● ● ●● ●● ●●●● ● ●●
1995
●
●
●
● ● ● ● ● ● ● ●●● ●● ● ● ●● ● ●●●● ● ●●
1997
●
Power relation
1999
●
●
●
Modelled
●
Lake and date specific
●
2e+05
●
●
● ●
1e+05
●
● ●
● ●● ●● ●● ●●●●● ● ●●
0
●
●
● ●
●
● ●
0
●
●
● ● ● ● ●● ● ●● ●●● ●●●●● ● ●●
40000
●
●
● ● ● ●
●
●
● ● ●
0e+00
●
●
●
● ● ●● ●● ●● ● ●●● ●● ●●●● ● ●●
40000
● ●
0
40000
Surface area (m2)
(b) Fidh Ali
Figure 2.4.10: Lake and date specific S-V relation against the interSR power relation adapted for silting over time
61
1992
1995
1997
100000
75000
Volume (m3)
●
Power relation 50000
●
Modelled
●
Lake and date specific
●
●
● ● ● ●
●
25000
●
● ● ●
●
●
● ●
●
● ●
● ●
● ●●
●
●● ● ● ● ●
0
●
0
● ●
●
10000
20000
● ●
●
● ●
● ●● ●● ●● ●● ●● ●
●
●
0
10000
20000
●
0
● ●
●
●
●
● ● ● ●● ● ●● ●● ● ● ●
10000
20000
Surface area (m2)
(a) Sadine 2
1990
1993
1996
●
4e+05
●
●
● ●
●
●
● ●
3e+05
● ●
●
2e+05 ● ●
1e+05
Volume (m3)
● ●
● ● ●
●
●
● ●
●
●
●
●
● ●
●
● ●
● ●
●
●
● ●
● ● ● ● ●● ●● ●● ● ● ● ●● ●
●● ● ●
● ● ● ● ●● ● ●●
● ● ● ● ● ● ● ● ●● ● ●● ●
●
●
●
● ● ●
● ●
●
●
● ●
●
●
●
40000
●
●
● ●
Lake and date specific
●
●
1e+05
Modelled
●
●
●
2e+05
●
●
●
0
Power relation
●
●
3e+05
● ● ● ● ● ●● ●● ● ● ● ●● ●
●
● ●
2005
4e+05
●
● ●
●
2001
0e+00
●
●
●
0e+00
●
● ●
● ●
● ●● ●● ●● ● ●● ●
80000
0
●
● ●
● ●
40000
●
80000
Surface area (m2)
(b) Dekikira
Figure 2.4.11: Lake and date specific S-V relation against the interSR power relation adapted for silting over time
62
Part II
Upscaling water availability assessments in ungauged reservoirs
63
Chapter 3
Remote sensing of flooded areas in small reservoirs 3.1
Materials and methods
3.1.1
Selecting satellite imagery
Remote sensing exploits differences in the electromagnetic waves emitted and reflected by objects under investigation to distinguish and classify land surface areas. Passive sensors exploit the energy provided by natural emitters, i.e. solar energy reflected or re-emitted as thermal energy by the earth’s surface. Active sensors send their own energy signal and measure the fraction reflected or backscattered by the target. Passive sensors are found aboard many commercial satellites and the same principle is also exploited by cameras, radiometers, spectroradiometers, etc. Active sensors include Radar (Radio Detection and Ranging), Lidar (Light Detection and Ranging) and Laser (Light Amplification by Simulated Emitted Radiation). Both types of sensors may be used in water detection studies and criteria to select suitable satellite data include physical specifications of the sensor and satellite, as well as image availability and cost considerations. Technical characteristics of the sensors shape their spectral coverage and resolution (i.e. the scope and width of the individual wavelength bands). The altitude of the satellite’s orbit determines the spatial coverage or swath as well as the spatial/geometric resolution of each image which varies from several centimetres to kilometres. Many earth observation satellites follow a heliosynchronous near-polar orbit circling the earth many times a day. Combined with the earth’s rotation, they orbit the globe along latitudinal strips. The revisit time of a target then depends on altitude and orbit of the satellite, though steerable sensors can increase temporal resolution of the sensor by focussing on the area of interest from different angles. Radiometric resolution defined by the number of bits (binary numbers) on the spectrometer determines its sensitivity and ability to capture greater detail in energy differences. Sensors and satellites each possess strengths and weaknesses, and compromises based upon specific objectives must be identified. Passive systems dependent on sunlight can only operate during daylight hours and reflectance values are affected by cloud presence, thus reducing the number and frequency of suitable images (Liebe et al. 2005). Active systems such as Radar can penetrate clouds, but their ability to detect water is restrained by 64
changes to the surface’s roughness due to ripples (wind, waves) or vegetation (Ji et al. 2009; Alsdorf et al. 2007). Passive sensors also remain more accessible in terms of costs and their treatment and analysis is more straightforward (Ji et al. 2009). From geometric principles, high spatial resolution is notably incompatible with large spatial coverage and with high spectral resolution. MODIS (Moderate Resolution Imaging Spectroradiometer) products provide daily images, however the 250 m spatial resolution is inadequate for monitoring small reservoirs, as each pixels corresponds to an area of 6.25 ha. AVHRR (Advanced Very High Resolution Radiometer) also provides daily coverage of the globe since 1982 (freely since 1992) but at 1.1 km resolution. Both have been used successfully to monitor floods over large areas such as wetlands (Ogilvie et al. 2015; Sakamoto et al. 2007; Mahé et al. 2011). Conversely, high spatial resolution imagery exist, notably SPOT (Satellite Pour l’Observation de la Terre) as well as Quickbird and Ikonos which offer images under 5 m resolution. Their reduced spatial coverage (10-20 km wide for the highest resolution) make them less suitable for medium sized catchments. Several images may be mosaicked, however for long term monitoring, the very high costs are currently prohibitive. Research discounts are available, however a single SPOT 10m colour image still cost a non negligible 450Ä through the CNES ISIS programme in 2013. Prices will continue to drop making higher resolution products more accessible. Radar sensors providing long term time series include ERS-1 (1991-1996) and ERS-2 (1995-present) SAR (synthetic aperture radar) and Envisat ASAR (advanced synthetic aperture radar, since 2002) with 30 m spatial resolution but with a 35-day cycle. To study long term flood dynamics and water availability in small reservoirs (1-10ha) across a medium size catchment (1000 km ), products with sufficient temporal resolution and coverage, geometric resolution, at affordable costs were required. Differences in spectral characteristics though important to optimise the water detection, can be partly addressed using the variety of classification methods available. Landsat was chosen amongst the numerous available satellite products, as it provides free multispectral images every 16 days since the 1970s at sufficient geometric resolution (30 m since Landsat 4). Landsat images previously cost $4400 per scene (Bastiaanssen et al. 2000) but prices reduced down to $700 by the year 2000 and are free since 2009. Exploited for numerous land use studies, it has notably been used on small water bodies (Qi et al. 2009; Liebe et al. 2005; Mialhe et al. 2008; Sawunyama et al. 2006).
3.1.2
Landsat satellites & sensors
The Landsat programme funded and managed by NASA launched the Landsat 1 satellite in 1972 and launched in 2013 Landsat 8 providing unparalleled cover of the globe at medium resolution over more than four decades. Considering our interest to monitor lakes since the 1990s, images from Landsat 5, 7 and 8 were used. Landsat 6 crashed into the sea on its launch day in 1993. With a heliosynchronous near polar orbit at an altitude of 705 km, Landsat 5 and its successors revisit the same region of the globe every 16 days. Covering a swath of approximately 170 km N-S by 183 km W-E, they provide multispectral images in 30 m spatial resolution in 6 or more spectral bands from the visible to the infrared part of the spectrum. Landsat TM (thematic mapper) sensor aboard Landsat 5 possessed radiometers
65
for three bands in the visible, two in the near infrared and one in the short wave infrared. A thermal infrared (TIR) band acquired data at 120 m spatial resolution, resampled to 30 m for images processed since 25.02.2010 and 60 m before this. ETM+ (enhanced thematic mapper plus) aboard Landsat 7 acquires thermal infrared at 60 m (again resampled to 30 m since 25.02.2010) and a 15 m panchromatic band. To optimise the radiometric resolution, bands can also collect in low gain settings when surface brightness is high (i.e. low resolution when large range of values) or conversely in high gain (high resolution when low range of values). Thermal infrared on ETM+ acquires in both gain settings. Landsat 8 OLI-TIRS sensors adds a 30 m cirrus and 30 m coastal aerosol band, and two 100 m TIR replace the previous 60 m thermal IR band. As all sensors, these exploit wavelengths situated within suitable atmospheric windows, where molecular absorption by water, vapours, gazes (carbon dioxide, ozone) and suspended particles are minimal (cf. figure 3.1.1). Landsat 5 TM sensor started acquiring on 1.03.1984 and exceeded its design lifetime by 24 years before inevitably failing in November 2011 (Bédard et al. 2008). The ETM+ sensor was launched on 15.04.1999 but on 31.05.2003 the scan line corrector (SLC) failed permanently as a result of mechanical failure (Goslee 2011). The SLC serves to compensate for the forward motion of the satellite and align the scans. As a result, certain area are scanned twice while others are not, leading to a loss of around 22% of pixels distributed along oblique lines (Zeng et al. 2013). Methods to deal with this issue exist but the continued operation of Landsat 5 TM sensor provided a valuable complement to ETM+, especially considering the similarity in spectral and spatial resolutions. Landsat 8 was launched on 11.02.2013 ensuring continuity of the Landsat programme. Though the temporal resolution of Landsat satellites is 16 days, their acquisition cycle are staggered by 8 days. Landsat 7 revisits 8 days later the same area as Landsat 5 and Landsat 8 which effectively replaced Landsat 5 maintained the same 8 day lag with Landsat 7. The combination of two satellites can therefore be exploited to provide images every 8 days rather than 16 days.
3.1.3
Landsat image availability
Landsat images managed by USGS can be selected and ordered using USGS GloVIS, USGS Earth Explorer or NASA’s Reverb tools. Based on location and dates, suitable tiles and their metadata can be previewed, before being downloaded directly or ordered. In some cases, only LandsatLook JPEG versions are immediately available and full data (Level 1) products must be ordered via the GloVIS interface. Once processed, these are made available freely on a file transfer protocol (FTP) server, typically within 1-3 days and can be downloaded using their proprietary Bulk Download Application programme. The Worldwide Reference System (WRS) indexes Landsat images using path numbers relating to the orbit cycles and rows relating to scene centres. 233 paths and 248 rows compose the WRS-2 system used with the Landsat 4-8 orbits. Swaths from successive orbit cycles overlap by 7-80% as one moves from the equator to the poles due to the constant swath width. The Merguellil upper catchment is located at the overlap of two paths: path 191 row 35 and path 192 row 35 (figure 3.1.2). Path 192 is visited 7 days after path 191 leading to increased image availability. Figure3.1.3 below shows all available Landsat images for the Upper Merguellil catchment and figure 3.1.4a the average delay per month between
66
Figure 3.1.1: Wavelengths and atmospheric transmission of the bands of Landsat 5 TM, Landsat 7 ETM+ and Landsat 8 OLI-TIRS sensors (adapted from http://landsat.usgs.gov/ldcm_vs_previous.php) two images. The number of images per month does not reveal whether these are spread out within the month or overlap as with Landsat 5 and Landsat 8 sensors. Actual image availability depends on image quality, notably the presence of clouds and the influence of the scan line corrector failure as investigated in section 3.2.2.3 (figure 3.1.4b). Considering data availability of both Landsat images and stage data for several lakes, the period 1999-2014 was used. The 2013-2014 years following the launch of Landsat 8 where satellite imagery, stage data and differential GPS (DGPS) contours could be acquired were used to develop the method, notably selecting and calibrating water indices. High image availability following the commissioning of Landsat 7 in 1999 and before the scan line corrector failed also coincided with significant stage data from instrumented lakes which could be exploited to test the method’s suitability. No Landsat images were available for this catchment over 1992 and 1999. Over 1999-2014 we therefore acquired and treated a total of 546 Landsat images: • 133 Landsat 5 TM images from 21.12.2000 to 02.11.2011 • 56 Landsat 7 SLC-on images from 14.08.1999 to 05.05.2003 • 287 Landsat 7 SLC-off images from 16.08.2003 to 05.01.2015 • 70 Landsat 8 images from 29.03.2013 to 05.12.2014. The first of these is pre-WRS, i.e. taken before the satellite was at its final orbit. These images have the same geometric precision however cover a different swath, which must be accounted for in their treatment, notably defining cropping coordinates. A summary of the chain of treatments applied to the 546 Landsat images is provided in figure 3.1.5. 67
Small reservoirs Merguellil catchment
Landsa t 192 3 5 path
Landsa t 191 3 5 path
SRTM v3 n35e009 tile
Figure 3.1.2: Overlap of Landsat Paths 191 35 and 192 35, SRTMv3 n35e009 digital elevation model and small reservoirs
● ●● ● ● ● ● ● ●● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ●● ●
●
● ● ● ●
● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
Landsat 5
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
Landsat 7 SLC−on
●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●●● ● ●● ● ● ● ● ●
●● ● ● ●● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ●●●● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
1985
1990
1995
2000
2005
2010
Landsat 7 SLC−off
Landsat 8
2015
Figure 3.1.3: Availability of Landsat images for Merguellil upper catchment
68
Days between images (mean per month)
60
40
20
0 2000
2005
2010
2015
(a) All images for Merguellil upper catchment
Days between images (mean per month)
60
40
20
0 2000
2005
2010
(b) Images after optimisation for clouds & SLC failure over one lake cell (Gouazine)
Figure 3.1.4: Lag between successive Landsat images used here 69
2015
Colour key
Landsat 5-8 (30m, 16 day)
RS data
Surface area data (GPS, stage)
High resolution imagery (G. Earth)
Cloud, shadow & SLC-off assessment (& optimisation) Field instrumentation
Pretreatments (radiometric, atmospheric, topographic corrections) MNDWI water detection (& index calibration)
Outputs
Grid cell definition around SR
Surface-vol relations
Flooded surface area Si+X
Remote sensing of flooded volumes
Flooded volumes Vi+X
Figure 3.1.5: Chain of treatments applied to the 546 Landsat images
3.1.4
Image pretreatments and corrections
Landsat images were distributed as Level 1 Terrain Corrected (L1T) products. This orthorectification seeks to improve geometric accuracy by using digital elevation models (including SRTM and GTOPO30) and ground control points (GCP) from the Global Land Survey 2000 (GLS2000). Accurate georeferencing is required in order to overlay auxiliary data on images, such as our DGPS contours, and perform spatial assessment of their concordance (confusion matrices). Landsat 8 reportedly benefits from very high geodetic accuracy, around 18 m before corrections and around 10 m following GCP corrections (Storey et al. 2014). Landsat 7 also achieved around 15 m after terrain corrections, but these vary with location and depend on the quality of the ground control points. Bands are supplied as individual GeoTIFF files, using the Universal Transverse Mercator (UTM) projection associated with the World Geodetic System (WGS) 84 datum. Landsat products are provided in a single compressed .tar.gz file, along with a metadata (.MTL) file providing parameter values to radiometrically correct the images. Our area of interest being situated within a single Landsat tile (figure 3.1.2), no mosaicking of images was required. Images were cropped to the region of interest during treatments to reduce file size and optimise speed. Landsat 7 images after the SLC failure are provided with a ’gap mask’ indicating the location of the no data pixels. No cloud mask is provided but can be generated as detailed below. 3.1.4.1
SLC-off errors (Pixel loss on Landsat imagery)
Following the permanent failure of the scan line corrector on TM sensor on 31.05.2003, resulting Landsat 7 images were affected by the loss of around 22% of pixels (Zeng et al. 2013). These are distributed along oblique lines ranging from 14 pixels wide at the edges to 2 at the centre which is less affected (Maxwell 2004). The lines created by the SLC failure are different to the subtle stripping phenomenon which can be observed on Landsat images as a result of the fact that the sensor uses 16 photodetectors which screen the area, and can produce minor differences due to differing sensitivity. SLC-off pixels suffer from a complete loss of information, i.e. no treatment or correction may be developed to restitute the information, unlike other interferences which could in theory be modelled and corrected for. Methods designed to address the loss of pixels from SLC failure must instead rely on nearby spectral information either spatially or temporally. 70
Accounting for SLC pixel loss Spatial interpolation exploits the spectral data from nearby pixels to derive clusters and classify land uses. These notably use resampling algorithms as well as maximum likelihood and decision tree classification methods (Maxwell 2004; Bédard et al. 2008). Geostatistical tools such as ordinary kriging (Zhang et al. 2007) can also be used. These however assume that nearby pixels share similar reflectance profile which is not compatible with small scale analysis developed here (Zeng et al. 2013). Our objects of interest vary between 1 ha and 30 ha but as figure 3.1.6a illustrates these are easily contained within a single SLC-off band which varies between 60 m and 420 m wide. Where SLC bands are situated over small reservoirs, spectral interpolation would in many cases fail to detect a lake, and be unable to estimate its size with reasonable accuracy. This method is however well suited for larger lakes and land use studies (figure 3.1.6b). Temporal methods rely on additional spectral data for the same location from another image close in time. Other Landsat 7 SLC-off images may be used as the gaps do not fully overlap on successive images (Zhu et al. 2012), alternatively Landsat 5 and Landsat 8 images or images from other sensors may be combined (Chen et al. 2011; Roy et al. 2008). The local linear histogram matching (LLHM) method employed by the USGS (Scaramuzza et al. 2004) relied on this to previously provide SLC corrected images. Users must now choose amongst methods including the Neighborhood Similar Pixel Interpolator (NSPI), the Geostatistical NSPI (GNSPI), maximum a posteriori and Markov random field theory and the weighted linear regression method which have respective advantages and limitations (Zeng et al. 2013; Zhu and Woodcock 2014, 2012), notably with respect to accuracy in heterogeneous regions. These methods were however developed for large scale (land use) studies where the rate of change of nearby pixels is low and/or homogeneous. Methods rely on the relative stability of pixel reflectance between nearby dates or at least assume “similar patterns of spectral differences between dates” (Chen et al. 2011; Zeng et al. 2013). This is not the case when monitoring small scale objects and very localised changes, i.e. a few water pixels varying at a very different rate to surrounding land use/vegetation between successive images. Methods such as the NSPI are also designed to use several images to correct a single image before land use classification, making these also less suited in terms of computing time to correct numerous Landsat images as required here. Both spatial and temporal correction methods were therefore not suitable here. Instead the pixel loss due to SLC failure over each lake and for each image were identified in order to exclude in post treatment images with excessive percentage of SLC related pixel loss as well as cloud&shadow interference as detailed below. USGS previously provided a mask indicating the location of pixels loss from SLC failure. This can be generated from any Landsat band and identifying pixels where the digital number is 0, instead of the 1 to 255 range. Using the lake cells created (cf. section 2.2.4) we then extract the number of no data pixels over each lake and compile using R the results for each lake cell and date. As SLC off pixels are effectively no data pixels, these can not be classed as flooded areas unlike cloudy pixels which can be lead to commission errors (false positives) (figure 3.2.8). SLC off pixels on the contrary can only lead to an underestimate of the flooded area (figure 8.3.3). In parallel, this assessment provided insights into the range of pixel loss over time from SLC failure over small lakes, as a result of the heterogeneous size and location of SLC-off bands within images and between consecutive images.
71
(a) 20 ha small reservoir in the Merguellil
(b) 1,000 ha lake
Figure 3.1.6: Pixel loss due to SLC failure compared to objects of interest on Landsat 7 29.03.2013 image
% of L7 SLC−off images
30
20
10
0 0
50
% of SLC−off pixels in lake cell
100
Figure 3.1.7: Percentage of Landsat images affected by SLC presence over 1 lake cell (Gouazine)
72
3.1.4.2
Cloud & shadow errors (Pixel loss on Landsat imagery)
Electromagnetic waves within the visible to infrared spectrum used by passive sensors such as those aboard the Landsat satellites are heavily influenced by clouds. In the visible wavelengths, clouds prevent much of the solar energy from reaching the land surface and is reflected, hence their brightness (whiteness). In the infrared bands, much of the electromagnetic energy is absorbed by the droplets in clouds. No systematic corrections on pixel values can be performed as the varied nature and properties (temperature, thickness, altitude, water content) of clouds disturb reflectance values differently. Accordingly, their influence on land use classification is uneven and not systematic and cloud presence does not always interfere with water detection (Ogilvie et al. 2015). In addition to masking pixels by being impenetrable by visible wavelengths, clouds also affect pixels by the shadow they cast on the earth’s surface. Cloud shadows can in certain cases have the same low reflectance values as freshly burned areas or as water bodies (Liebe et al. 2005). Cloud & shadow detection Similar gap-fill corrections as developed for SLC pixel loss can be used for clouds and shadows but again the small scale of our objects of interest makes these unsuitable (Feng et al. 2015). Cloud ratings across the whole scene and within each quarter are indicated within the metadata files and can be used to discard images, especially when purchased. Due to the scale of small reservoirs (i.e. a 10 ha (0.1 km ) small reservoir within a 33 300 km Landsat 8 image), not all lakes are affected the same way by a cloudy tile (figure 3.1.9). Considering images with less than 20% clouds would lead to excluding 60% of images (i.e. 327 out of 546 images) if assessed at the whole scene level against only 31% (i.e. 168 out of 546 images) if assessed over a single 10 ha lake (figure 3.1.10). The 73% increase in images deemed suitable highlights the need for spatial analysis of clouds over individual lakes, in order to retain as many images as possible to derive flooding time series. Though manual detection of clouds may appear straightforward, the wide variations in reflectance and temperature of the earth’s surface and of clouds make automated detection difficult (Irish et al. 2006). The high ratio between high visible reflectance and strong infrared absorption, as well the fact that clouds are colder than the earth’s surface can be used to detect clouds (Huang et al. 2010; Martinuzzi et al. 2007; Irish et al. 2006. Clouds are however not always bright, white and cold while shadowed pixels can be brighter than nearby surface reflectances (Zhu and Woodcock 2012). The latter is notably an issue when shadows are cast on snow, ice, bright rocks and other very bright areas areas. These are sparse in semi-arid areas, but images are notably affected by subtle shadows from “semi transparent clouds”, notably cirrus clouds which are harder to detect (Zhu and Woodcock 2012). Landsat 8 seeks to address this partly by adding a 1380 nm band 9 designed to detect high cirrus clouds. This wavelength present on other lower spatial resolution sensors is strongly absorbed by the water vapour in the (lower) atmosphere therefore only high cirrus clouds where water vapour is low will reflect solar radiation in this wavelength. Information on cirrus cloud cover is now provided by USGS in the Quality Assurance file of Landsat 8, but these are not available for previous sensors. Attempts to model cloud & cirrus shadows have been made based upon estimating cloud height based on temperature, view angle and solar illumination geometry reported in the metadata files (Huang et al. 2010; Zhu and Woodcock 2012).
73
The Automated Cloud Cover Assessment (ACCA) algorithms (Irish et al. 2006) used originally by the USGS exploit the visible, infrared and thermal infrared bands to provide cloud masks for Landsat 7 images. These identify up to 90% of clouds but only 75% of time (Irish et al. 2006) and don’t account for cirrus clouds or shadows, just like more recent methods such as Goslee (2011). The Fmask algorithm (3.2.1 beta version) developed by Zhu and Woodcock (2012) specifically designed to deal with clouds and shadows on Landsat imagery was used here. Fmask uses top of atmosphere reflectances for all visible and IR bands, and brightness temperature for band 6 (equation 3.1.19) are used as inputs. The method then combines probability masks with a series of rule based tests to identify potential cloud pixels based on the cloud spectral behaviour described above. The upgraded version (3.2.1) used here also exploits the Landsat 8 cirrus band. It reports a 96.41% overall accuracy vs 84.8% for ACCA. Other methods seek to combine the information from successive images to correct the reflectance values (Zhu and Woodcock 2014; Goodwin et al. 2013). An improved version of Fmask called Tmask notably adds this temporal dimension in the analysis of clouds to improve cloud shadow detection (Zhu and Woodcock 2014). It combines the outputs from the Fmask algorithm with a model attempting to anticipate the spectral change in pixels based on a stack of several successive images, at least 15 clear observations. Changes which are not deemed to be due to land use change are then identified as cloud or cloud shadow. As discussed with SLC off corrections methods such as NSPI, these are not suited to the detection of ephemeral change (Zhu and Woodcock 2014), such as small scale, rapid changes in water surface area, which do not follow the same gradual pattern as nearby land use. It should also be noted that the method is highly consuming in computer processing time (20 hours CPU for 33 Landsat image). Implementing Fmask cloud & shadow detection Fmask 3.2.1 was used here via an executable based on Matlab code, which completed calculations much faster than a version adapted for R (under 2 minutes per image vs 60 mins). The output GeoTIFF mask file is then coded as 0=clear land pixel, 1=clear water pixel, 2=cloud shadow, 3=snow, 4=cloud, 255=no observation. Files were cropped to the area of interest only after cloud detection to ensure the shadows from clouds on the edge of our area of interest were not cropped and undetected. Figure 3.1.8a shows how on one image, cloud covers only 1.7% of the image according to the USGS metadata and other methods (e.g. Goslee 2011). With the addition of the 1380 nm wavelength band allows the detection of cirrus clouds leading to a total of 8.2% of clouds (figure 3.1.8b) and with Fmask an extra 3.3% of pixels are covered by shadows (figure 3.1.8c). On a single cell, the error from incomplete detection can be significant, with for example no thick clouds detected by conventional methods but 87% of the lake cell covered by clouds and shadows according to Fmask. The grossly overestimated flooded area detected as a result in this cell, confirm the pertinence of this additional cloud detection. The presence of clouds and shadows over each lake cell was then extracted and compiled over each 546 images in R to exclude in post treatment images with excessive percentage of cloud & shadow interference, as well as SLC related pixel loss as detailed in section 3.1.7. In parallel, the relative influence of clouds and shadows on results was studied. This was done by post-treating within R our water detection rasters and coding as NA pixels which contained clouds and/or shadows based on the georeferenced Fmask raster file. The flooded 74
(a) Detection of clouds only (as per (b) Detection of clouds incl. cirrus (c) Detection of clouds incl. cirACCA(Irish et al. 2006), Goslee clouds (as on Landsat 8) rus clouds & cloud shadow (as per (2011) methods) Fmask method)
Figure 3.1.8: Comparing cloud detection methods on Landsat 8 29.03.2013 image
Figure 3.1.9: Guettar lake (in green) affected by nearby cloud (white) and cloud shadow (grey) presence areas when keeping all cloud and shadows pixels and when removing clouds, shadows and then both were compiled (figure 3.2.8). This also enables us to assess in addition to cloud and shadow presence over the lake cell, the cloud and shadow presence (and therefore error) over the pixels classed as flooded. (This could not be done for SLC pixels as these no data pixels can not overlap pixels classed as flooded.) The total error however is also composed of areas not detected as flooded but still obstructed by cloud or shadows. Though cloud and shadows share similar spectral properties with water and generally result in commission errors (false positives) and therefore overestimation, in some cases, clouds and shadows can in some cases not be classed as water as shown by the underestimation errors due to cloud and shadows in figure 3.2.8. 3.1.4.3
Radiometric normalisation
The radiance values measured by a sensor differ from the energy reflected by the earth’s surface, due to atmospheric interferences (aerosols, clouds...) as well as noise & distortions due to the sensor irregularities, its orbit, and the solar elevation angle which changes through the year. A series of corrections are required to estimate the reflectance values at 75
60
% images
40
20
0 0
25
50
% of cloud pixels over whole scene
75
100
(a) Across whole scene
60
% images
40
20
0 0
25
50
% of cloud & shadow pixels over lake cell
75
100
(b) Across Gouazine lake cell
Figure 3.1.10: Percentage of Landsat images affected by clouds when assessed over a whole scene and an individual lake cell
76
the top of atmosphere and those at the earth’s surface (top of canopy). For certain applications such as single image classifications, radiance values can be used directly. Where several images are compared in multi temporal studies, or where a method requires true reflectance values, corrections are required. Level 2A products which incorporate necessary radiometric and atmospheric corrections have recently been introduced by USGS. Images are corrected on request (http://landsat.usgs.gov/CDR_LSR.php) to land surface (“top of canopy”) reflectance using the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS). Similarly, the THEIA land data centre provides corrected Landsat images over France based on a treatment chain they developed (MACCS). At the time of research, level 2A corrected images were not available for Landsat 8 or before 2002 for our catchment. The following radiometric (including atmospheric & topographic) corrections required were therefore derived for each sensor and band combination and coded in R. On the three images used to compare water detection band ratio indices, all bands except panchromatic were corrected. On the other 543 images, only the bands required for MNDWI (green & SWIR) were corrected using one R programme for each sensor which looped through each band and image of interest. The raster files for each band were cropped to the area of interest to reduce calculation time. Packages such as teamLUCC in R can automate a number of these corrections steps, however they lack flexibility and were not yet adapted to treat Landsat 8 images which contain different files and metadata. Functions belonging to the R rgdal, raster, sp, and landsat packages were used. Necessary coefficients to perform corrections were extracted automatically for each band and image from the metadata files as these are sensor dependent and can change over time (Goslee 2011). Adjustments were made to deal with the changes in the band names and metadata formats over the period 2012-2015. Digital Numbers to Top of Atmosphere (TOA) radiance The signal detected by the Landsat sensors in each pixel is provided in quantized calibrated (QCAL) digital numbers (DN) as a result of a linear transformation of the actual at sensor (top of atmosphere, TOA) spectral radiance. Digital numbers range from 0 to 255. This calibration process (equation 3.1.1) can be reversed to obtain the TOA radiance values, L (equation 3.1.2) (Chander et al. 2007).
(DN ≠ DNmin ) ú (Lmax ≠ Lmin ) = (DNmax ≠ DNmin ) ú (L ≠ Lmin ) L=
Lmax ≠ Lmin (DN ≠ DNmin ) + Lmin DNmax ≠ DNmin
(3.1.1)
(3.1.2)
This can be rewritten successively as L=
Lmax ≠ Lmin Lmax ≠ Lmin ú DN + Lmin ≠ DNmin DNmax ≠ DNmin DNmax ≠ DNmin L = Grescale ú DN + Brescale
where
77
(3.1.3) (3.1.4)
Lmax ≠ Lmin Lmax ≠ Lmin = DNmax ≠ DNmin 255
(3.1.5)
Lmax ≠ Lmin DNmin DNmax ≠ DNmin
(3.1.6)
Brescale = Lmin ≠ Grescale ú DNmin = Lmin
(3.1.7)
Grescale =
Brescale = Lmin ≠
Grescale and Brescale are the “radiance multiplicative” and “radiance additive” rescaling factors, sometimes called gain and bias (or offset). These are sensor and band specific and can be found within the metadata files as “RADIANCE_MULT_BAND_x”, and “RADIANCE_ADD_BAND_x”. These both simplify as shown above, as DNmin is 0 here. TOA radiance to TOA reflectance At sensor radiance must first be converted to at sensor reflectance to correct for intra-annual differences in earth to sun distance and solar illumination. The solar elevation angle notably changes due to changes to the earth-sun geometry over the months and can be assessed based on date and time of image acquisition. flT OA = T OAref lectance =
L ú fi ú Esun Exsun ú cos(90 ≠ sunelev )
(3.1.8)
Esun is the earth to sun distance in astronomical units (AU) provided in the metadata files (EARTH_SUN_DISTANCE) for OLI/TIRS. For other sensors (ETM+, TM), this can be calculated based on the date of acquisition indicated in the metadata (DATE_ACQUIRED) using functions such as ESDIST in the R Landsat (Goslee 2011) package. sunelev is the solar elevation angle in degrees for the scene centre provided by the metadata file (SUN_ELEVATION). The complementary angle (90-sunelev ) provides the local solar zenith angle. Exsun is the band specific exoatmospheric solar irradiance constant in W m≠2 ú µm≠1 calculated based on equation 3.1.9. For OLI/TIRS, Lmax and REFmax are provided for each band. For ETM+ and TM sensors, Exsun for each band must be taken from the literature (table 3.1.1, Chander et al. (2009)). Exsun =
Lmax ú (fi ú Esun ) REFmax
(3.1.9)
TOA reflectance to TOC reflectance Atmospheric corrections are then necessary to convert at satellite (top of atmosphere) reflectance to surface (top of canopy, i.e. ground) reflectance. These seek to account for the distorting effect of atmospheric disturbances (air molecule, aerosol particles, water vapour, carbon dioxide, methane, etc.) which absorb and scatter electromagnetic waves and estimate the reflectance if it was measured at ground level (Hagolle et al. 2015). These effects are more complex to remove than satellite to top of atmosphere corrections and rely on absolute or relative correction methods. The pseudo invariant features (PIF) methods exploit the principle that the reflectance of certain pixels (buildings, etc.) within images are expected to remain stable over time, as their physical properties do not evolve. Variations are then due to atmospheric interferences and a regression equation between reflectance value for this invariant object in successive
78
Table 3.1.1: Exoatmospheric solar constant per band for ETM+ and TM sensors Landsat 7 ETM+
Landsat 5 TM
1
1997
2
1812
3
1533
4
1039
5
230.8
220
6
-
-
7
84.9
8
1362
1796
images is used to perform a relative normalisation of each image (Liang et al. 2001). Histogram matching is another relative method which seeks to shift the histograms from hazy portions of an image to match the histogram of non hazy portions. This assumes that each portions will contain similar variety in objects and reflectances. They also both assume relatively homogeneous aerosol distributions across the image. Both these methods can require visual inspection of the images and reduce the dynamic range of the bands, leading to loss of detail (Goslee 2011). Furthermore these do not account for the fact that atmospheric disturbances are band (wavelength) specific and therefore not recommended when using band ratio approaches as here (Lu et al. 2002). Absolute corrections include two approaches, either modelling of the atmospheric processes and thus disturbances or analysing the data in the bands to identify and remove the error due to atmospheric effects. Physically based models of (part of the) atmospheric effects rely notably on radiative transfer theory and require estimating atmospheric parameters (Tauz , Tauv , Edown and Lhaze ) and/or exploiting ground data to calibrate the corrections (Lu et al. 2002). Key inputs into models such as SMAC include the aerosol optical thickness which can be obtained for selected sites with aerosol and water vapour data from the Aerosol Robotic Network (AERONET). An AERONET site is operational in the vicinity of the Merguellil upper catchment but only since 2013, and could therefore not be used in this study using historical Landsat 5, 7 and 8 data. Due to the widespread difficulty in acquiring in situ data for remote locations and for all periods of interest, image based calibration models were developed. These include COSTZ, DOS and improved DOS (Chavez 1996). DOS methods provide a simple, widely used method for atmospheric corrections, proven to perform well and as reliably as other algorithms and methods (Paolini et al. 2006; Lu et al. 2002; Song et al. 2001). Dark Object Subtraction (DOS) The Dark Object Subtraction method performs corrections directly from TOA radiance to TOC (surface) reflectance, i.e. encompasses corrections for both at sensor radiance to reflectance due to illumination geometry (Paolini et al. 2006) and for atmospheric disturbances (equation 3.1.13). The DOS method supposes that certain objects are dark (in complete shadow) and that radiance values measured for certain dark objects within an image are due to atmospheric scattering (path radiance, Lhaze ) (Chavez 1996). These dark objects are selected from the lowest DN value across at least 79
1000 pixels. This defines the starting haze value (SHV) which is converted from DN to TOA radiance according to equation 3.1.4. The path radiance Lhaze is then assessed according to equation 3.1.12, as an allowance is made considering that pixels are rarely completely dark and 1% natural reflectance is subtracted. In its simplest form, SHV is calculated for each band but the improved DOS calculates Lhaze for all bands using only one band (blue band) to “correlate haze values and maintain the spectral relationship between bands” (Goslee 2011; Chavez 1996). In the SWIR bands, where minimal scattering occurs, SHV is defined as 0 (Goslee 2011). Lhaze converted to satellite reflectance is then subtracted to reflectance values across the image (equation 3.1.13). Tauv is the atmospheric transmittance from the target to the sensor, Tauz is the atmospheric transmittance from the sun to the target and Edown is the downwelling diffuse spectral irradiance at the surface. When we don’t consider atmospheric disturbances, (i.e. Lhaze =0, Edown =0, Tauv & Tauz =1), the model used is termed apparent reflectance and equation simplifies to equation 3.1.8 for at sensor reflectance. Lhaze = SHV ≠ Lsat1% Lsat1% = 0.01 ú
Esun ú cos(90 ≠ sunelev ) fi ú Esun
Lhaze = Grescale ú DNmin + Brescale ≠ 0.01 ú flT OC.DOS =
Esun ú cos(90 ≠ sunelev ) fi ú Esun
(Lsat ≠ Lhaze ) ú fi ú Esun Tauv ú (Exsun ú cos(90 ≠ sunelev ) ú Tauz + Edown )
(3.1.10) (3.1.11)
(3.1.12)
(3.1.13)
Tauv = 1
(3.1.14)
Tauz = 1
(3.1.15)
Edown = 0
(3.1.16)
Several variants of the DOS method exist which include more or less simplifying assumptions on Tauv , Tauz & Edown (Song et al. 2001). The modified DOS (termed DOS4) notably seeks to account not only for haze but also for aerosols. Song et al. (2001) highlight that overall accuracy is however not improved and Goslee (2011)confirm that corrections are similar between DOS. Furthermore DOS3 & DOS1 were more adapted in normalising successive images which is here more important than absolute reflectance values for the purposes of classifying and comparing changes over successive images (Song et al. 2001). DOS3 being slower as requires solving through iteration (Goslee 2011), DOS1 level of correction which assumes no atmospheric transmission losses in either direction (Tauv & Tauz =1) and no diffuse downward radiation at the ground (Edown =0), i.e. corrects only for additive scattering was used here. The Landsat R package provides functions to perform the DOS corrections but these were also not adapted for differences in the Landsat 8 bands and metadata file
80
and were therefore coded manually. Topographic normalisation Where topographic differences are significant, differences in illumination on inclined surfaces can alter surface reflectance values. These are easily observed on either side of hills for instance and this shading effect must be accounted for when measuring surface reflectance. This is different to orthorectification discussed in section 3.1.4 where the topography is used to ensure the geometry & georeferencing is correct, here the objective is to account for terrain in the reflectance value. Corrections done so far assume the surface is flat and these must therefore be adapted to account for spatialised topographic differences. Methods rely on a digital elevation model to derive slope and aspect (direction faced) information which along with information on the satellite and solar geometry can model illumination values, as in equation 3.1.17. “i is the incident angle, slope angle is ◊p , solar zenith angle is ◊z, solar azimuth angle „a, and aspect angle „o. IL = cos“i = cos◊p cos◊z + sin◊p sin◊z cos(„a ≠ „o )
(3.1.17)
Slope and aspect can be derived using GIS software or in R with the Landsat package, while solar azimuth and zenith angle is provided in the metadata files. Topographic corrections are then performed using one of several statistical and/or empirical algorithms (Riano et al. 2003; Lu et al. 2002). The Minnaert algorithm (Minnaert 1941) was shown to perform as well or better than C correction, Cosine correction and Gamma correction (Vanonckelen et al. 2013; Hantson and Chuvieco 2011). Minnaert adds a band specific constant K to the cosine correction, a trigonometric correction defined in equation 3.1.18 where flt is the reflectance value generated by the inclined surface, and flh is the value on a theoretical flat surface. flh = flt ú (
cos◊z K ) IL
(3.1.18)
A function is included within the Landsat R package to perform this topographic correction. Gao and Zhang (2009) showed that SRTM 90 m DEM was suitable to perform topographic corrections on Landsat imagery and here the improved SRTM v3 30m DEM tiles mosaicked, projected to UTM and bilinear resampled to 30 m were used. Brightness temperature For thermal bands (Band 10 for Landsat 8 and Band B6_VCID_1 for Landsat 7, B60.TIF for Landsat 5) which are required to detect clouds, DN values can be converted to TOA radiance using equation 3.1.4 and then to (brightness) temperature (Kelvin) using equation 3.1.19. brightness.temperature =
K2 K1 úÁ log( L
+ 1)
(3.1.19)
K1 and K2 can be found in the Landsat 8 metadata as K1_CONSTANT_BAND_10 and K2_CONSTANT_BAND_10 respectively. For Landsat 5 and Landsat 7 these are not available in the metadata files but can be found in the literature. Á is emissivity, and can be estimated as 0.95 to account for atmospheric interferences.
81
Table 3.1.2: K1 & K2 constants for correcting thermal radiance to temperature Landsat 5 TM
Landsat 7 ETM+
K1
607.76
666.09
K2
1260.56
1282.71
3.1.5
Water detection by remote sensing
3.1.5.1
Water reflectance
The different level of absorption and reflectance in several wavelengths can be used to discriminate water over other land uses. The reflectance of water consists of three factors: surface reflectance, volume reflectance & bottom reflectance (Mialhe et al. 2008). Surface reflectance occurs mostly in the visible shorter wavelengths notably blue hence its colour, while longer wavelengths in the infrared spectrum are absorbed by water. Conversely in these longer NIR & SWIR bands, spectral reflectance of soil & healthy vegetation is high, and this high contrast can be exploited to distinguish between water and land areas (Toya et al. 2002; Mather 1999). Reflectance values of vegetation peak in the infrared spectrum while remaining low in the blue and red bands, due to absorption by chlorophyll. Reflectance of vegetation is also high in the green wavelengths, leading to its green colour. Soil reflectance depends on soil composition but has a gradual increase in the spectral response from visible to IR. When vegetation or soil is humid, the reflectance in SWIR & NIR bands decreases somewhat due to the water content. This is notably used to help distinguish types of vegetation and water stress. When water is more turbid, the suspended solids lead to more reflectance as one moves towards the red wavelengths. Water which contains submerged vegetation will also not absorb as strongly in the infrared bands as the vegetation present will reflect these wavelengths more than pure water. Turbidity, dissolved and suspended organic & inorganic matter, vegetation and algal content affect essentially volume reflectance and the level of influence of volume reflectance depends on how deeply light penetrates water (Mialhe et al. 2008; Mather 1999). Longer wavelengths are less able to penetrate water, with for instance green wavelengths reaching depths of 10 m but only 10 cm at 0.8-1.1 µm. The NIR and more so the SWIR bands are therefore less affected by the presence of vegetation & suspended solids in the water column, allowing these to more accurately classify turbid water as water than shorter wavelengths. Longer wavelengths will also be less affected by bottom reflectance which is composed of soil and vegetation. Its contribution depends on the depth and composition of the water column and that of the lake floor. 3.1.5.2
Multispectral analysis
The information from these various wavelengths can be combined in multispectral analysis to identify and classify land use features. A true colour composite image where the red, green and blue reflectance bands are displayed in red, green and blue colours performs the same tasks as our eyes and allows us to identify water, soil, vegetated and urban areas. Specific combinations of reflectance bands can be used to discriminate various features of interest. False colour compositions exploiting the low and high infrared reflectance values of
82
!
Figure 3.1.11: Merguellil catchment small reservoirs visible on full colour composite (bands 453) Landsat 5 2.11.1999 image water bodies (453) is illustrated in figure 3.1.11. To discriminate automatically the different features present in the landscape, methods exploit these difference in spectral profiles, mathematically. Geometry and texture can also assist classification processes but are essentially used with high resolution imagery and for other land use features (forests, gridded irrigation plots, etc.) Classification methods for water detection employ algorithms to group pixels into clusters with similar reflectance profile. In supervised classifications, learning zones based on external knowledge of the field or spectral information are incorporated from the onset and help calibrate the method. In unsupervised classifications, the classes produced by the algorithm are analysed a posteriori according to their spectral profiles or by exploiting secondary informations such as land use maps or field knowledge. Methods include k-means classification based on fuzzy logic which remains highly effective in distinguishing land uses (Jain 2010; Ogilvie et al. 2015). Maximum Likelihood Classifications (MLC) based on Bayesian probability function has been used over different scales but for automatic delineation this method was reportedly affected by residual noise and is not calibrated with ground truth data to improve the method (Annor et al. 2009; Liebe et al. 2005). Furthermore they are resource intensive (computational and operator time to define training zones) and can difficultly be applied and automated to series of images. Reflectance values on a single band notably the SWIR and NIR or thermal bands may be thresholded to distinguish water bodies (Mialhe et al. 2008). Considering the difficulties observed when water mixes with vegetation, several spectral water indices were developed to make use of the contrasting reflectance of objects in several bands (wavelengths). Using two or more spectral bands, these indices have been widely used in water studies and have been shown to perform as well (Mialhe et al. 2008) or better than other methods notably MLC (Feyisa et al. 2014). Working pixel by pixel unlike classifications such as MLC, they are also less likely to overclassify areas with large variations in their covariance matrix. Often based on the normalised difference of two spectral
83
bands, the ratio helps removes some of the residual noise after radiometric corrections and constant thresholds may be defined which facilitate the automation of water detection over successive images. Spectral water indices Numerous spectral water indices exist in the literature. One of the most well known band ratio index is the NDVI (Normalised Difference Vegetation Index) (Rouse et al. 1973) which exploits the contrast between the peak reflectance in the infrared band and the low reflectance in the red to monitor vegetation. It has also been used to detect water bodies (Ma et al. 2007; Mohamed et al. 2004) as the index becomes positive in presence of vegetation and negative for water bodies. It remains however limited in its ability to distinguish terrestrial from aquatic vegetation. McFeeters (1996) developed the NDWI (Normalised Difference Water Index) exploiting the difference in reflectance of water in the green and NIR bands. Xu (2006) however showed that this index failed to correctly distinguish water from built up features where NIR reflectance was also lower than green reflectance which led to similar positive NDWI values. In remote, rural areas, this is not expected to be an issue. Xu (2006) proposed a modified NDWI (MNDWI) where the NIR band was substituted by the SWIR band. In parallel, Gao (1996) developed a NDWI but using NIR and SWIR bands. Xu (2006) later defined this as NDMI (Normalised Difference Moisture Index). Rogers and Kearney (2004) also developed an index they called NDWI but using red and SWIR bands. Lacaux et al. (2007) developed the NDPI (Normalised Difference Pond Index) which also exploits the low reflectance of water in SWIR and contrast with green. NDPI is effectively the opposite of MNDWI developed by Xu (2006). NDTI (Normalised Difference Turbidity Index) was also developed using the red and green bands, and exploits the principle that turbid water reacts like bare soils, i.e. low reflectance in green and high in red. For SWIR, two bands exist on Landsat TM/ETM+/OLI-TIRS sensors (figure 3.1.1). Band 5 for TM/ETM+ and band 7 for OLI-TIRS was used in accordance with other studies (Ji et al. 2009). Ouma and Tateishi (2006) tested using ETM+ band 7 for SWIR in MNDWI/NDPI but showed that this did not perform as well as with band 5. N DV I =
N IR ≠ R N IR + R
(3.1.20)
N DW I =
G ≠ N IR G + N IR
(3.1.21)
G ≠ SW IR G + SW IR
(3.1.22)
N IR ≠ SW IR N IR + SW IR
(3.1.23)
SW IR ≠ G SW IR + G
(3.1.24)
R≠G R+G
(3.1.25)
M N DW I = N DM I =
N DP I =
N DT I =
Another spectral water index is AWEI (Feyisa et al. 2014). Developed for Landsat images, it exploits wavelengths within 5 bands to specifically detect water pixels and distinguish 84
over dark surfaces. Two variants exist, AWEInsh and AWEIsh, the latter being optimised to remove (urban and topographic) shadow pixels. Both can be used in succession in the presence of highly reflective areas (snow, roofs, etc.) Unlike band ratio indices, the AWEI equations are not normalised. The equation and factors were empirically determined over several large lakes to optimise the difference between water and other land surfaces.
3.1.6
AW EInsh = 4 ú (G ≠ SW IR1 ) ≠ (0.25 ú N IR4 + 2.75 ú SW IR2 )
(3.1.26)
AW EIsh = B + 2.5 ú G ≠ 1.5 ú (N IR4 + SW IR1 ) ≠ 0.25 ú SW IR2
(3.1.27)
Selecting & calibrating water indices
The performance of individual indices has been discussed and compared in studies (Ji et al. 2009, 2015; Feyisa et al. 2014) which focussed on large water bodies. Exploiting different wavelength combinations, indices are more or less suited to detecting certain types of water and their behaviour in varying situations must be assessed. Furthermore the thresholds for each index must be calibrated. McFeeters (1996) when developing NDWI identified positive values of the index as water but in practice a more precision adjustment of threshold can improve accuracy of the methods (Ji et al. 2009; Xu 2006). Furthermore, the stability of threshold must be assessed in order to allow automation over successive images (Ogilvie et al. 2015). The suitability of all 7 spectral indices described above were therefore tested on small water bodies. NDPI being the opposite to MDNWI, its performance was not assessed separately. The AWEIsh version of AWEI optimised to deal with potential shadows was used here. For TM, ETM+, OLI-TIRS sensors, all bands required for the 7 indices presented above are available, but on certain sensors bands forbids the use of certain indices. The absence of SWIR bands on early Landsat (pre TM) sensors and SPOT 4-5 10m images notably precludes the use of the MNDWI or AWEI indices. In those cases, NDWI, NDVI and NDTI can be considered. 3.1.6.1
Surface area assessments in the field
DGPS (Differential Global Positioning System) contours of the flooded area over three dates and 7 lakes were used to calibrate the thresholds of each index and assess their accuracy. In large scale studies, where direct measurement of the flooded area is not feasible due to scale or access, notably wetlands and large lakes, classifications using higher resolution images may be used. In the Niger Inner Delta for instance, k-means classification on concomitant Landsat (30 m) images were used to calibrate water detection on MODIS (250 m) images (Ogilvie et al. 2015). Annor et al. (2009) used MLC as ground truth. These classifications however incorporate errors due to pretreatments and the classification algorithm. Stage data may also be used but requires the use of Height-Surface-Volume relationships which when available remain associated with significant uncertainties, notably at low stage values which are exacerbated over time through silting (cf. section 2.4). Furthermore HSV relations will only provide a surface area value making a spatialised assessment of errors impossible, except in the rare cases where a 3D georeferenced model of the lake is available. 85
Figure 3.1.12: DGPS surveys on Morra reservoir 21 contours around the shores of 7 lakes were taken over three field dates (table 3.1.3). Lakes were selected to study how well the 7 calibrated indices could assess flooded areas on small reservoirs presenting specificities in terms of size, depth, shape and presence of vegetation. Lakes notably included 2 reservoirs under 1 ha, two between 1-8 ha and 2 above 8 ha. With the exception of Morra, these reservoirs were not the same as the 7 lakes used subsequently in section 3.1.7 and chapter 4 to study flood dynamics over 1999-2014. Selection of the lakes was partly constrained by low water availability due to relatively poor rainfall in 2012-2013 which led to only 14 lakes being flooded according to a 10 m 19.03.2013 SPOT image. Using a different subset of lakes also allowed to further validate the method on lakes not used in the calibration of the water index thresholds. Accessibility of small reservoirs was considered, both in terms of relative ease of access by 4x4 and in the ability to circle the lake on foot. Where necessary, boats can be used to access locations unavailable on foot. Minor portions of the lake perimeter (less than 10%) which could not be surveyed on foot, due to cliff faces or treacherous silty deposits were transcribed and waypoints were corrected in post treatment. Field trips were timed to coincide with the 16-day Landsat 8 acquisition cycle and programmed during the spring, before the small reservoirs have dried out and when cloud cover is reduced. Where logistics allow, the contours should be taken only after the image quality (clouds, haze) has been checked online or contours should be postponed in doubt. The lag between satellite overpass and field observation was inferior to 1 week, similar or inferior to previous studies (Liebe et al. 2005; Annor et al. 2009; Sawunyama et al. 2006). The three campaigns were staggered between March and June 2013 in order to calibrate multi-spectral
86
Landsat 8
26.03.13
21.05.13
11-14.06.13
29.03.13 (2013_088 preWRS2)
24.05.13 (2013_144)
09.06.13 (2013_160)
Table 3.1.3: Dates of DGPS surveys and satellite imagery used to calibrate multi-spectral indices Cell n°
26.03.13
21.05.13
11-14.06.13
Morra
30
76669
73311
70746
El Kraroub
25
77820
59193
52891
Gbatis
18
9870
6692
3653
Garia S / El Kesra
28
25112
22491
20697
En Mel
21
56020
48527
44168
O. Bouchaha A
35
1826
2091
N/A
Mdinia
50
N/A
61965
63028
Table 3.1.4: Flooded surface area (m2 ) according to three DGPS surveys on 7 lakes indices when lake water levels were high and low and study how greater proportions of shallow waters, vegetation growth, algae, turbidity, affect their performance. Multiple dates were also required to assess the stability of thresholds, for the process to be semi-automated on different images. Two series of contours on 6 lakes (as 1 was dry and 1 was inaccessible) in March and June were used to assess their performance in calibration mode. A third series of contours on the same 7 lakes in May when vegetation and algae growth was evident but in reduced proportions was used in validation mode. A fourth campaign was organised in July but only two lakes remained flooded, therefore these contours were used in the next validation phase which also looked at the suitability of the chosen index and threshold on a different set of lakes (section 3.1.7). Converted waypoints were checked against freely available high resolution imagery (Microsoft Bing in ArcMap) to check for any obvious errors. Files were then cleaned up based on field notes and drawings of the bushes and shrubs bypassed and inaccessible portions (cliff faces...). The uncertainty from these corrections is estimated to be in the order of ±0.5 m. On other portions, observer error during DGPS acquisition is estimated around ±0.2m due to unclear shorelines on flat beaches and the fact that a waypoint was taken every 2 seconds while circling the shore. The surface area is then calculated from the geometry of the polygons in GIS software. 3.1.6.2
Errors & confusion matrices
Confusion matrices Confusion matrices provide an assessment of how many pixels are correctly and incorrectly classified in each class as defined in table 3.1.5. These are widely used in remote sensing studies and perform a spatial assessment of pixels omitted and committed by the spectral water index compared to ground truth, here the georeferenced contours. They provide an assessment of the accuracy of the classification and can thus be used to calibrate thresholds based on minimising error or maximising accuracy rates defined 87
Table 3.1.5: Confusion matrix for remotely sensed flooded area and DGPS contours DGPS contour = Water
DGPS contour = Non water
Remote sensing water index = Water
T Pw
F Pw
Remote sensing water index = Non water
F Ps
T Ps
below. Producer accuracy (P rod.accu) is used to define how well a certain area is classified. This is computed based on the ratio of pixels correctly detected in each class over the true number of pixels in that class. Equation 3.1.28 below defines water producer accuracy, i.e. the percentage of water pixels detected correctly by remote sensing as water. Errors of omission are defined according to equation 3.1.29 and assess the percentage of pixels omitted from each class. P rod.accuW ater =
T Pw T Pw + F Ps F Ps T Pw + F Ps
(3.1.29)
T Ps F Pw = T Ps + F Pw T Ps + F Pw
(3.1.30)
OmissionW ater = 1 ≠ P rod.accuW ater = OmissionN on.water = 1 ≠
(3.1.28)
User accuracy (U ser.accu) is defined as the ratio of pixels correctly detected in each class over the total number of pixels classified in that class (equation 3.1.31). Though omission errors estimates the number of pixels wrongly omitted and therefore the underestimation, commission error (equation 3.1.32) does not assess pixels wrongly committed in terms of surface area overestimation. The denominator is not pixels committed in terms of true surface area but in terms of the total number of pixels classified by the water index. It provides instead an assessment of the reliability of the classification. Omission errors indeed refer to the product (producer), i.e. focussing on the lake surface area and assess what proportion the method has missed. Commission errors focus on the user and assess what proportion of the pixels it classified as water are indeed water. (In other terms, on one hand omission errors assess how much of a shape the child didn’t paint relative to the size of the shape, while commission errors assess how much of a child’s painting is outside the shape boundaries but relative to the surface area (s)he painted...). U ser.accuW ater = CommissionW ater = 1 ≠
T Pw T Pw + F Pw
T Pw F Pw = T Pw + F Pw T Pw + F Pw
CommissionN on.water = 1 ≠
T Ps F Ps + T Ps
(3.1.31) (3.1.32) (3.1.33)
Overall accuracy (Overall.accu) is defined in equation 3.1.34, as the number of pixels
88
correctly classified over the total number of pixels. Other pixel based accuracy measurements between classifications and reference data exist, notably the Kappa coefficient and McNemar’s test which focus on the frequency of pixels correctly and incorrectly classified. Feyisa et al. (2014) however showed that additional statistics did not provided added benefit when optimising water detection thresholds. Overall.accu =
T Pw + T Ps T Pw + F Pw + F Ps + T Ps
(3.1.34)
To derive confusion matrices, the shapefiles of our DGPS contours for each date were converted to 1 m resolution Raster GeoTIFF files where flooded areas were coded as 1 and non-flooded areas as 0. The three relevant Landsat 8 images pretreated for clouds and radiometric corrections as described above were used to generate the 6 band ratio indices. These were upscaled to 1 m resolution to be compatible with our contours. Considering the influence of a second decimal in the thresholds (Ogilvie et al. 2015), these were then varied between -1 and 1 in increments of 0.01. As the AWEI is not normalised, iteration was used to identify a suitable range to study the thresholds. Pixels where the index value was superior to the threshold were classified as water, except for NDVI and NDTI where it was the opposite. Lake cells (defined in section 2.2.4) were then used to extract on both the contour GeoTIFF file and each of the 6 Landsat classified images the relevant pixels for each lake. In addition to allowing automatic extraction of flooded areas for each lake, these cells help remove certain outliers, i.e. pixels which do not belong to the lake but share similar reflectance profile, such as surrounding flooded vegetation, river stretches and temporary ponds or waterlogging. Four new spatial data frames in R were then created for each image to define the pixels where both the DGPS and the classified index agreed and disagreed on the presence of water and non water (i.e. T Pw , etc.). Confusion matrix errors were then calculated using the number of pixels in each spatial data frame. The process was then repeated for each threshold value, for each index and over the three Landsat images. Spatial data frames have similar properties to raster files but allow for much faster processing in R (4 min vs 18 min). Area error indices The performance of spectral indices in detecting flooded area can be assessed based on errors in the actual surface area, notably through simple calculations of relative errors, including the Normalised Difference Area Index (NDAI) (Liebe et al. 2005; Annor et al. 2009) or Percentage Deviation Area Index (PDAI) (Sawunyama et al. 2006). This is however not a pixel based assessment and it therefore assumes that all pixels identified as flooded are in the right area, and may disguise spatial errors of overclassification and underclassification. The lake cells help reduce this problem but within a lake cell, wet vegetation may still wrongly be classified as seen on figure 3.1.14. Surface area is calculated from the sum of correctly located areas (T Pw ) but also areas wrongly classified water (F Pw ) areas. Here though 21% of the true water pixels are incorrectly classified (omission error), surface area error is only 0.3% as the 2098 m incorrectly classified in effect compensate for the 2100 m not detected in the south of the lake. The reduced PDAI can be described as an artefact, the result of beneficial uncertainty, i.e. where errors of underestimation on certain parts of the lake are compensated by errors of overestimation in other parts. PDAI provides a valuable assessment of how well remote sensing fulfils our requirement of estimating surface
89
area, optimising over PDAI does not ensure the most spatially accurate performance and may instead favour an optimal combination of incorrectly detected pixels. SRS ≠ SDGP S SRS + SDGP S
(3.1.35)
|SRS ≠ SDGP S | ú 100 SDGP S
(3.1.36)
N DAI = P DAI =
Selecting threshold calibration method Thresholds were therefore optimised based on more systematic pixel based accuracy assessments. The rate of change of the multiple water detection rates as a function of thresholds are shown in figure 3.1.13 for a single lake and index (MNDWI). As thresholds increase, less pixels are classified as water resulting in an increase in water omission errors and a decrease in commission errors. Optimising water omission or water commission errors alone is not sufficient, as this would lead to a threshold which respectively grossly overclassifies or underclassifies pixels as water. Optimising for both water omission and water commission errors can be done (e.g. Feyisa et al. 2014) by looking at their intersection (figure 3.1.13b), i.e. where both errors are comparable in size or where their sum is minimal which is not always identical (table 8.2.1). With only two complementary classes when water omission errors increase, non water omission errors decrease (figure 3.1.13a) as pixels shift out of one class into the other. Their intersection could then also be used to optimise thresholds. Overall accuracy gives equal weight to correctly identifying all classes, making it unsuitable when we wish to optimise for a single class amongst many. With two classes, overall accuracy rates remain high for a broad range of thresholds (> 90% for MNDWI thresholds between -0.13 and 0.02) as it effectively seeks to minimising omission errors for both water & non water pixels. As a result even when water classification accuracy declines, the correctly classified non water pixels maintain a high overall rate. The marginal peak may nevertheless be used to optimise thresholds (Fisher and Danaher 2013). The thresholds where these various errors converge or are minimal are however marginally different (figure 3.1.13b). The influence from choosing to optimise thresholds based on the following combinations of the four different detection rates was investigated: 1. maximising producer accuracy of water & non water pixels 2. maximising producer and user accuracy individually of water pixels (i.e. intersection of these rates) 3. maximising the sum of producer and user accuracy of water pixels 4. maximising overall accuracy Table 8.2.1 shows the optimal threshold according to the optimisation method used and the associated errors. The coefficient of variations in the optimal thresholds obtained for each lake varied between 6% and 70% indicating potentially significant differences. Though overall accuracy rates remained fair (> 70%) these disguised non negligible differences in the actual surface area error, reducing to 0% for certain thresholds but reaching 55% (and 393% on the smallest lakes). PDAI index is indeed highly sensitive to differences in thresholds (figure 3.1.13a). Figures 3.1.14 and 3.1.15 illustrate on two lakes the difference in pixels classified 90
1.0 0.8 0.6 0.4
Water detection rate
0.2
● ● ●
0.0
● ●
−0.4
−0.2
0.0
Water omission error Water commission error Non water omission error Overall accuracy PDAI
0.2
0.4
MNDWI threshold
0.5
(a) Amplitude of the errors
● ● ●
0.2
0.3
●
0.0
0.1
Water detection rate
0.4
●
Water omission error Water commission error Non water omission error Overall accuracy PDAI
−0.2
−0.1
0.0
0.1
0.2
MNDWI threshold
(b) Zoom on intersection of errors
Figure 3.1.13: Water detection rate for varying thresholds of MNDWI on Landsat 8 29.03.2013 image for lake Morra
91
Source: Esri, DigitalGlobe, GeoEye, i-cubed, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community
Figure 3.1.14: Incorrectly classified pixels when optimising MNDWI for Water & Non water Producer Accuracies (dark blue) vs Max overall accuracy (lighter blue) (Landsat 8 29.03.2013 image, lake Gbatis, cell 18)
Figure 3.1.15: MNDWI classification output when optimising for Water & Non water Producer accuracies (left), Max overall accuracy (middle) and their overlap (right) (Landsat 8 29.03.2013 image, lake Morra, cell 30)
92
using different optimisation methods. Optimising along method 2 provided consistently good results but failed to contain PDAI errors on the smallest reservoirs. Method 4 in addition to inherently producing superior overall accuracy rates, also provided for all six reservoirs acceptable PDAI errors and the smallest PDAI error for the two smallest reservoirs (lake 18 and 35). Thresholds for each index and lake were therefore optimised based on overall accuracy. We must however that sensitivity to the calibration method remains limited as here when optimising on overall accuracy over all 7 lakes and 3 images, MNDWI threshold was -0.09 (-0.1 for March and -0.08 for June) and when optimising on PDAI it was -0.11 (-0.09 March, -0.12 for June). 3.1.6.3
Validating the threshold
The performance of each calibrated spectral water index was then assessed and compared across the 7 lakes and over the March and June Landsat 8 images. The overall accuracy and PDAI achieved as well as the stability of the optimal thresholds were discussed. The median threshold from these 12 calibration events (6 lakes in March, 6 in June) was then taken as the optimal threshold to be used over each lake & image. The median was preferred over mean threshold to reduce the influence of single lakes (and dates) where the optimal threshold is significantly different due to mixed reflectance issues. Similarly, an optimal inter-lake and inter-date threshold was not determined by minimising errors (optimising overall accuracy) over aggregated pixels from all lakes, as we sought to give equal weighting to each lake rather than allowing large lakes to steer threshold values. This also ensured mixed pixels found across small reservoirs were better accounted for in the calibration (Jain 2010). The associated errors when using the threshold from this calibration was then assessed using a third May 2013 Landsat 8 image and the concomitant DGPS contours over the same 7 lakes. The spectral water index which minimised water detection errors over lakes and dates, and which is therefore most suited to being extended to other lakes and images was then selected.
3.1.7
Assessing flood dynamics over time
After selecting and calibrating the optimal water detection index, the method was applied to the successive 546 Landsat images to monitor flood dynamics across 7 reservoirs over 15 years (1999-2014). DGPS contours allowed the calibration and validation of the index, however the method’s ability to be applied to successive images and derive flood dynamics was then tested on 7 small reservoirs benefiting from significant field data available. With the exception of Morra, these lakes were not used in the development of the thresholds, giving further weight to the validation of the method. Originating from previous projects (Albergel and Rejeb 1997) as well as specific instrumentation for this research, the available stage data and associated stage-surface area rating curves derived from topographic levelling are detailed in section 2.4 for each reservoir. The Gouazine reservoir benefited from both the longest (over 15 years) and most accurate time series, based on automatic transducers data subject to ongoing corroboration by observers, and regular (6) updates to the stage-surface area rating curves and was therefore favoured when developing the method. Optimising cloud+shadow & SLC thresholds In the absence of methods capable of detecting and correcting for interferences from clouds, shadows and SLC pixel loss, the 93
thresholds (% of lake cell affected) over which their presence detrimentally affected the water detection were investigated by comparing with stage derived surface areas. Optimisation of the thresholds was done based on three variables: R , the number of images and Mean Square Error of the surface area integrated over time. The R calculates the quality of the fit between the satellite and field data, however this will tend towards removing all cloud & SLC interferences, and therefore reduce the number of available observations. Considering the objective of maintaining sufficient repetitivity of the satellite observations in order to derive flood dynamics, retaining images with non negligible quantity of clouds is preferable and errors introduced by these interferences can be smoothed where necessary (Ogilvie et al. 2015). To optimise for the quality of the relationship over time, thresholds were calibrated in order to minimise mean squared error (MSE) between remotely sensed surface area and field surface area integrated over time. This indeed depends on sufficient and reliable observations of surface area over time to provide the best fit with field derived dynamics. This information is then used to assess mean water availability per year to ascertain whether the remote sensing observations can be used to provide flood statistics to stakeholders. These calculations were programmed in R. The 546 Landsat images downloaded were treated for cloud and SLC off detection and radiometrically corrected as detailed above. Using the chosen (MNDWI) index and optimal threshold (of -0.9), flooded surface areas were extracted for each lake using the predefined lake cells. A first R programme was written to loop through all the folders and perform these pretreatments and calculations. A second R programme then looped through all the output files to collate the number of flooded pixels, the number of clouds, shadows and SLC off pixels within each lake cell and for each image. The process took 7 minutes in R for each image. Available stage data was converted to surface area for each lake based on spline interpolation of the H-S relationship. The successive updates of the H-S relationship (up to 6 on Gouazine) were used to interpolate the relevant sections of the stage time series. Values of clouds & shadows and SLC off pixels above which to exclude the satellite observations of flooded area were then varied for each interference respectively in 5% increments. For R and the number of images the influence of clouds & shadows and SLC errors are independent, meaning clouds and shadow thresholds could be assessed on non SLC affected images only and conversely SLC errors studied on images with no clouds & shadow issues. When considering surface area over time, their influence can no longer be studied individually. Surface area over time was calculated by linearly interpolating discrete observations into daily time series and assessing the area under the curves (AUC) using the R pracma package. Cloud and shadow cover detected reached up to 140% due to assessing their number in 30 m resolution pixels contained (even partly) within the polygon-shaped lake cell (cf. section 2.2.4). Gouazine (Lake 51) which possessed both the longest time series and the most accurate data and rating curves (updated 6 times) was used to optimise these thresholds. These were then assessed on other lakes to confirm their pertinence/validity. The results in terms of accuracy of the remotely sensed flood dynamics and water availability assessments over the seven reservoirs were then discussed.
94
3.2 3.2.1
Results & discussion Comparing the performance of spectral water indices
Comparing overall accuracies When individually optimised for overall accuracy, all spectral water indices performed well against the March DGPS contours, achieving consistently high overall accuracy rates (figure 3.2.2), and low water omission and commission rates (figure 3.2.1). NDMI achieves marginally lower overall accuracy than the other 6 indices reaching 82% on average rather than 90%. PDAI shows indices were all able to achieve a reasonable level of water detection accuracy, below 20% on 4 out 6 lakes. On lake 18 PDAI varied between 9% and 27% depending on the index, showing remarkable precision considering its size of 1 ha and the coarse 30 m image resolution. In June, overall accuracy rates declined marginally for MNDWI, NDVI, NDTI and NDWI, while improving marginally for NDMI (figure 3.2.3). PDAI values with NDVI increased beyond 25% for small and large lakes. Lake 35 had dried out in June but on lake 18 PDAI values degraded for NDVI, NDWI and NDMI and all exceeded 50%. Only MNDWI and NDTI maintained PDAI rates below 25% for all lakes. Problems were reported with AWEIsh. These results confirm the range of precision found in previous studies, under 25% for small reservoirs around 1-2 ha, and 10-50% for lakes between 1-5ha (Sawunyama et al. 2006; Annor et al. 2009). On the smallest lake (lake 35), only 0.2 ha, the error reaches 50% with all indices which is to be expected considering it corresponds to just over 2 pixels. These indices can therefore be applied on Landsat images of 30m with reasonable accuracy to detect lakes of 1ha and above. Residual errors observed are first and foremost due to the coarse resolution, leading to more pronounced difficulties on smaller lakes, considering the greater influence of a single misclassified pixel on the result. 1 ha corresponds to a little more than 9 pixels. Smaller lakes also intrinsically have a greater proportion of mixed pixels, i.e. partly flooded and partly shoreline (figure 3.1.14) where the reflectance is a mix of water, soil, vegetation (Ji et al. 2009). The relative proportions of each land class in the mixed pixel will determine the overall reflectance which will therefore be different than pure pixels. For these to be considered and included in the classification, calibration must therefore ensure mixed pixels are included in the training (calibration) zones (Ji et al. 2009; Jain 2010). Triangular shaped lakes which possess a narrow inlet will also be affected by this issue. These errors reduce as resolution of the sensors increase, and therefore can be lower with SPOT, Sentinel-2 products. Smaller lakes also tend to have shallower depth, as shown through the stage-surface power relations, and therefore a higher proportion of shallow waters, where the boundaries between water, soil & vegetation are blurred (Mialhe et al. 2008). In heterogeneous water environments, DGPS delineation errors are greater (nearer ±1m) and the spectral reflectance of flooded areas becomes a combination of water, soil and vegetation making accurate discrimination more complex. These problems become relevant on all lakes on the June image due to the lower water levels and greater algae and vegetation growth. NDVI which suffers from inability to correctly distinguish vegetation from flooded vegetation sees its performance diminish accordingly. NDMI on the other hand performed well, coherent with its objective to detect moisture and all types of water. In previous studies it was also found to perform well during the dry season in the presence of shallow, turbid, vegetated waters (Ogilvie et al. 2015) thanks to the middle infrared band which penetrates water less and is 95
0.4
−0.4
0.2
0.4
1.0 0.8 0.6
0.6
Water detection rate
0.8
1.0 0.4
0.0
NDWI threshold
0.4
0.0
Water detection rate 0.2
0.6
0.4
0.0 0.0
NDMI threshold
0.8
1.0 0.2
0.2
1.0 0.8 0.6 0.4 0.2
Water detection rate
0.0
MNDWI threshold
0.0 −0.4
0.4
Water detection rate
0.0 −0.4
0.4
0.2
AWEIsh threshold
0.2
0.0
0.2
1.0 0.8 0.6 0.4
Water detection rate
0.0
0.2
1.0 0.8 0.6 0.4
Water detection rate
0.2 0.0 −0.4
−0.4
0.0
0.2
NDTI threshold
0.4
−0.4
0.0
0.2
0.4
NDVI threshold
Figure 3.2.1: Detection rates for each index and error for lake Morra (cell 30) on Landsat 8 29.03.2013 image. Legend as in figure 3.1.13a. less affected by turbidity & vegetation in the water. On lake 18 however its performance was inferior to NDTI and MNDWI which also shine through their ability to detect turbid waters, and distinguish flooded vegetation over wet vegetation on land. The MNDWI was shown in previous studies (Feyisa et al. 2014; Ji et al. 2015; Jiang et al. 2014; Ogilvie et al. 2015) to perform better than other band ratio indices, and here its suitability on small lakes was confirmed. Feyisa et al. (2014) found that AWEIsh performed better than MNDWI on 4 out 5 sites improving accuracies by up to 50%. Here it also exceeded MNDWI in terms of overall accuracy on the 5 lakes however only marginally as mean overall accuracies were 89.4%, 89.4% and 89.9% for MNDWI, AWEInsh, and AWEI respectively. Though proficient on the first image, the index could not be studied or used on the other images as on June and May images, the AWEI index was unable to detect water. Problems occurred as a result of inherent discrepancies in the radiometric corrections which the nature of the AWEI relation seem to have amplified, rather than reduce as with other normalised band ratio indices. This problem was indeed not observed using the uncorrected raw bands nor with other corrected band ratio indices. Developed on pure pixels, the ability of AWEI to accurately detect flooded areas in small reservoirs combining issues of mixed pixels and shallow waters remains still to be assessed.
96
1.00
0.95
● ●
●
●
●
●
●
Overall accuracy rate
● ●
● ●
● ●
● ●
● ● ●
●
● ●
0.90 ● ●
●
● ● ●
●
● ●
Small reservoirs ● ●
●
18
●
21
●
25
●
28
●
30
●
35
●
0.85
●
●
●
●
●
●
●
0.80
●
●
0.75 AWEInsh
AWEIsh
MNDWI
NDMI
NDTI
NDVI
NDWI
(a) Overall accuracy rate
100
75
PDAI
Small reservoirs
50
●
●
●
●
●
●
●
18
●
21
●
25
●
28
●
30
●
35
●
●
●
25
●
● ●
● ●
●
● ●
● ● ●
●
●
●
●
●
●
●
● ●
●
● ●
●
●
0 AWEInsh
AWEIsh
MNDWI
NDMI
NDTI
●
● ●
●
NDVI
NDWI
(b) PDAI
Figure 3.2.2: Compared performance of 7 water indices on 6 lakes during wet period (March 2013)
97
1.00
●
0.95
●
● ● ●
●
●
●
●
0.90
● ● ●
Small reservoirs
● ●
●
Overall accuracy rate
● ●
●
●
●
●
●
18
●
21
●
25
●
28
●
30
●
50
● ●
● ● ● ● ●
●
●
0.85
●
●
0.80
●
0.75 AWEInsh
AWEIsh
MNDWI
NDMI
NDTI
NDVI
NDWI
(a) Overall accuracy rate
100
●
●
75
PDAI
Small reservoirs
●
50
●
●
●
●
18
●
21
●
25
●
28
●
30
●
50
●
25
● ● ●
●
●
●
●
●
●
●
●
●
● ●
● ●
● ●
● ●
0
● ●
●
AWEInsh
AWEIsh
MNDWI
NDMI
NDTI
NDVI
NDWI
(b) PDAI
Figure 3.2.3: Compared performance of 7 water indices on 6 lakes during dry period (June 2013)
98
Threshold stability Considering the objective of automating water detection over successive Landsat images, the index used must both be able to provide high accuracies and maintain relative threshold stability. Though the different indices all peak around the same thresholds, the rate of change of accuracies as a function of thresholds are different (figure 3.2.1). The steeper slopes of omission and commission errors notably for NDTI and NDMI indices indicate greater sensitivity to incremental changes in thresholds. Accordingly overall accuracy rates peak over a reduced range of thresholds, compared to AWEI, MNDWI and NDWI. A marginal shift in threshold will then introduce larger errors and this greater instability has implications on the amplitude of errors when using a fixed threshold across lakes & images. The significant rate of change observed confirm that thresholds must be assessed and optimised up to 2 decimals. AWEIsh appears highly stable to changes in threshold which is coherent with the absence of normalisation and therefore thresholds spread wider than -1 to 1 and change more gradually. Thresholds for each index, lake and date optimised against overall accuracy are shown in figure 3.2.4. Thresholds are markedly different on the March image for lake 35, which due to its very small size is composed exclusively of mixed pixels. As a result indices such as MNDWI, NDMI or NDVI which are more likely to classify these pixels as water, are in effect calibrated to a higher threshold to reduce the number of pixels detected as water. This is also observed though in lower proportion on the small 1 ha lake 18. In the June image, thresholds for each lake also appear somewhat more scattered than in March, again as a result of the presence of the mixed reflectance signals from shallow waters and greater algae which the threshold calibration seeks to account for. The marginal delay (below 1 week) between the satellite overpass (acquisition) and the ground truthing is not seen to influence thresholds as if water levels had declined significantly between the image overpass and the DGPS contour, a systematic shift would be observed in the thresholds to effectively reduce accordingly the water area detected. The variance in the optimal thresholds calculated for each index are shown in table 8.2.3. Variation coefficients was not used here due to its inherent instability when the average is near 0 (leading to exaggerated CV variations) as with NDTI & NDMI. NDTI stands out for the stability of its thresholds followed by NDVI. The variance of NDWI, NDMI and MNDWI resulting from these 12 calibration points are all relatively close. AWEI due to the problems mentioned earlier could not be studied for threshold stability in June. Considering only the optimal thresholds for the 6 lakes on the March image, the AWEIsh variance (2.24 ú 10≠4 ) was second lowest after NDTI, confirming the “fairly stable optimal threshold value” (Feyisa et al. 2014) declared by its inventor. Validating indices and thresholds Stability in the optimal threshold shown in figure 3.2.4 combine however with the sensitivity to threshold change shown in figure 3.2.1 to determine the amplitude of errors from using a fixed threshold over several images & lakes. Water detection errors compared to DGPS contours when using the optimal threshold based on the 12 calibration events on the May 2013 validation image are shown for each lake and index in table 8.2.3 and figure 3.2.5. Overall accuracy rates remain very high for all 5 indices ranging from 80% to 93% on all 6 lakes, except again for NDMI which ranges from 70% to 89%. Large clouds and shadows covered 46% and 50% respectively of the Kraroub lake cell on this date, making validation over this lake infeasible. Small and large lakes were nevertheless represented. On 99
march
june
● ●
●
● ●
0.8
●●●●●●●●
0.9
● ● ●
●● ●● ●● ●●● ●
●
●● ● ● ●
●
●
MNDWI
●
0.9
(all)
●●●●● ●●● ●
●
●●●●●●●●
●●
●●
●
●
●●●
●●●●●
0.8
●
● ●
●
● ● ● ● ● ●● ● ●
● ● ●● ●
●●
0.8
●●●●●
●●
●
●
● ●
0.8
●
●
●
●
●
●
●
● ●
●
●
●
●●
−0.1
25
●
28
●
30
●
35
●
50
● ● ● ●● ● ●● ●●
NDWI
●● ● ●● ●
−0.2
21
●
●
●
0.8
18
●
NDVI
● ●● ●●● ● ●● ● ● ●●
●● ●●●
● ●● ● ●
0.9
0.9
●
● ●
● ● ●
0.9
Small reservoirs
● ●
NDTI
Overall accuracy rate
●
NDMI
● ● ●●●
●●
0.0
0.1
0.2−0.2
−0.1
0.0
0.1
0.2−0.2
−0.1
0.0
0.1
0.2
Figure 3.2.4: Optimal threshold (based on overall accuracy) for each lake, date & index
100
the smaller lakes (18 and 35), NDVI and NDWI performed poorly, failing completely to detect the smallest lake. On lake 18 their omission rates were 76% and 67%, leading to PDAI errors of 73% and 60% respectively. These errors are partly due to index stability issues but largely to their inability to correctly distinguish flooded vegetation as even with an image specific threshold NDVI can only minimise both omission and commission errors to 50% each resulting in PDAI of 46%. In May water levels had reduced compared to March, leading to greater issues of shallow waters & vegetation. NDMI was able to detect the two smallest lakes but low user accuracy led to gross PDAI errors, as it overclassified many vegetated pixels. MNDWI and NDTI therefore fared the best with these smaller reservoirs, though MNDWI achieved better accuracies on these small lakes and achieved consistently the lowest PDAI errors, as a result of the lower rate of change of the MNDWI index over NDTI observed in figure 3.2.1. Though not calibrated against PDAI, this remains our objective and MNDWI achieved surface area errors ranging from 1 to 28%, mean 10.5% with standard deviation of 8.7%. On the smallest lakes (35 and 18), PDAI errors were 14 and 8%. These results provide confirmation of the suitability of the MNDWI index to be used with a constant threshold (-0.09) over several lakes and images (figure 3.2.7, table 3.2.1). MNDWI performed well in different situations and the threshold remains sufficiently stable to comply with our prerequisite to automate and extend the procedure throughout a catchment and over several years. Previous studies on small reservoirs had found NDAI errors around 10% but reaching 50% or more, notably on the smaller lakes in the range 1.5 - 5ha. Here errors remained small even on lakes of this size despite the same spatial resolution of the images, pointing to the improved ability of indices in detecting mixed pixels caused by shallow waters and to the benefits of including small reservoirs in the calibration as these require a slightly greater threshold. Annor et al. (2009) had found that remote sensing overestimated surface areas systematically by around 7%, due to the maximum likelihood classifier used which relies on spectral profiles and has a tendency to overestimate. Here the calibration of the threshold effectively provides for this and no additional corrective factor was required to convert to field measurements as shown by the 1:1: linear relationship in figure 3.2.6. Liebe et al. (2005) had also required a corrective factor to account for depletion fluxes (evaporation, infiltration, withdrawals) caused by the lag between the field and satellite acquisitions. With ground measures taken sufficiently close this also proved unnecessary. R Pearson correlation coefficient reaches 0.99 over these 19 values, improving marginally over Liebe et al. (2005) (0.9). Band ratio indices where necessary may be combined with other indices and methods to improve detection of more complex areas, such as shallow waters during the dry season. NDMI may be combined with MNDWI (Ogilvie et al. 2015) or alternative combinations could be researched to deal with the greater vegetation problems in water bodies that do not dry out in summer. Considering the satisfactory levels of accuracy obtained, this was not required here. Optimising thresholds for each lake and image individually could also reduce PDAI errors below 5%, highlighting the potential for further improvements by developing (unsupervised) classification methods capable of being used over numerous locations and images reliably and efficiently (in terms of CPU and user input). Optimising thresholds for each lake but not each image would not improve performance on many lakes (where the optimal threshold was also -0.09), except the smallest lakes. 101
95
● ●
90
● ● ●
● ●
●
●
● ● ●
● ● ●
●
●
● ●
● ●
● ●
● ●
●
●
●
● ●
85
●
80
●
OV accu
●
● ●
75 70
●
75
● ●
●
● ● ●
●
Prod accu
● ● ● ●
●
50
● ●
Accuracy rates
●
25
●
0 100
●
50
●
18
● ● ●
●
21
●
●
28
● ● ● ●
●
30
●
35
●
50
User accu
75
● ● ●
● ● ●
●
●
●
●
● ● ●
Small reservoirs
●
● ●
25 0 250
●
●
●
●
●
200 150
PDAI
●
100
● ●
50 ● ● ● ● ●
● ● ●
● ● ● ●
MNDWI
NDMI
NDTI
0
●
●
● ●
● ● ●
NDVI
NDWI
Figure 3.2.5: Errors in validation of calibrated threshold
Table 3.2.1: Summary of confusion matrix values (%) obtained when validating the optimal MNDWI threshold on 7 lakes on Landsat 8 21.05.2013 image (N.B. clouds and shadows were present over lake 25) Cell n°
18
21
25
28
30
35
50
Lake name
Gbatis
En Mel
Kraroub
Garia S.
Morra
Bouchaha A
Mdinia
Surface area (ha)
0.67
4.85
5.92
2.25
7.33
0.18
6.20
Overall accu
92.6
87.9
64.8
92.7
89.6
79.6
93.9
Prod accu
77
81
86
67
88
42
90
User accu
71
88
38
93
93
49
91
PDAI
8
7
244
28
5
14
1
102
80000 ●●
y = −1845 + 0.99 ⋅ x, r 2 = 0.99
● ●
●●
60000
Remotely sensed surface area (m²)
●
●
Date
●
●
40000
20000
●
June
●
March
●
May
● ● ●
● ● ● ● ●
0 0
20000
40000
60000
80000
Surface area by DGPS (m²)
Figure 3.2.6: Correlation between field measurements and satellite estimation
103
0.95
● ● ● ● ● ●
0.90
● ●
● ●
● ●
Calibration
0.85
Step
Overall accuracy
0.80
● ●
Calibration
● ●
Validation
Date 0.95
● ●
june march may
0.90 Validation
0.85
0.80
18
21
25
28
30
35
50
Small reservoir
● ●
30
● ●
20 Calibration
● ●
10
Step
Surface area error (%)
● ● ● ● ● ●
● ●
Calibration
● ●
Validation
0 Date 30
● ●
june march may
20 Validation
10
0 18
21
25
28
30
35
50
Small reservoir
Figure 3.2.7: Errors in calibration and validation steps with MNDWI and optimal threshold
104
with clouds and shadows ● ●
●●● ● ● ● ● ●●● ●● ● ● ● ● ●
● ●
Remotely sensed surface area (m²)
1e+05
5e+04
0e+00
●●● ●●
●
●
with clouds only
● ●● ●●● ●
● ●
● ●
●
● ● ●●
●
●●● ●
● ● ●● ● ● ●● ● ●● ● ●● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ● ● ●● ● ● ● ● ●●●●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●● ●● ● ●● ● ● ●● ●● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ● ●● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ●● ● ●● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●● ●●● ● ●● ●
●
●
● ●●● ● ●
● ●
●● ●● ● ● ● ●● ● ●● ● ●● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ●● ● ● ● ● ● ● ●● ● ●● ●●● ● ● ● ● ● ● ●● ● ● ● ●●●●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●●● ● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ●●● ●● ●●●● ● ● ● ● ●● ●● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ●● ●●● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ●●●● ●●● ● ● ● ●● ●
Cloud & shadow %
100
without clouds and shadows
●
●
●
50
●
0
● ●
0e+00
●
●
●
1e+05
5e+04
●
● ●●
●
●
● ●
●
with shadows only ●
●●● ● ● ● ● ● ● ●● ● ● ● ●
● ●●●●●● ● ●● ●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●●●●● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ●●● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ●●●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ●●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●●● ● ●● ● ●●●●● ●● ● ● ●●●●●●●●●●●● ●
● ●●● ●● ● ●● ●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ●●●●● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ●●● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ● ●●●●● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●●●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ●●●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ●●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●● ● ● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●●●●●●● ●● ● ●●●●● ●● ● ● ●●●●● ●●●●●●● ● ●
0
0
●
30000
60000
90000
30000
60000
90000
Field surface area (m²)
Figure 3.2.8: Errors from clouds and shadows on scatter plot between field and Landsat observations (Gouazine, lake cell 51)
3.2.2
Remote sensing of flood dynamics with MNDWI
3.2.2.1
Cloud & shadow influence
The influence of clouds and shadows on the correlation between field observations and satellite observations over 1999-2014 for 1 lake are illustrated and differentiated in figure 3.2.8. Clouds clearly lead to an overestimation of the surface area with errors up to 100% when clouds cover the whole surface area of a (dry) lake. However the effect was not systematic, as when clouds were detected across the whole cell, MNDWI also classified only some of these clouds as water, as shown by the underestimation. The proportion of clouds over the lake is not determinant (figure 3.2.9) as the resulting error depended on cloud properties. The diversity in the nature and properties/characteristics (thickness, aerosol content...) of clouds are indeed firstly responsible for the heterogeneous reflectance observed and the resulting classification difficulties as discussed in section 3.1.4.2. On a handful of images (5 out of 84) with clouds over 75%, errors are relatively close to the flooded area, which may be due to cloud transparency or the actual overlap between clouds & flooded areas. When considering a single lake of the order of magnitude of small reservoirs (10 ha) 28% of the 546 Landsat images were affected by clouds, in varying proportions shown on figure 3.1.10b. In other catchments, proximity to coastal areas and more pronounced orography may increase the proportion of images affected by clouds. 105
The influence of shadows is more moderate and less frequent (5% of images) but also leads to significant overestimations. In 30% of these cases, the clouds were detected across the whole lake cell but again MNDWI classified between 0 and 100% of cloud shadows as water, as a result of difficulties in both detecting shadows and accounting for these properly.The lower frequency is partly due to the inherent difficulties in discriminating shadows and the overlap of clouds and shadows where only clouds are detected. 3.2.2.2
SLC-off influence
Unlike clouds which interfere more or less with the ground reflectance, pixels affected by SLC are inoperative and therefore can only lead to underestimation. As discussed in section 3.1.4.1, the influence of SLC is variable within the image and across images, leading to heterogeneous influence on water detection. When considering a single 10 ha lake cell, 35% of its surface area was affected by pixel loss on average over the 287 Landsat 7 SLC off images. Figure 3.1.7 illustrates however that over 30% of Landsat 7 images post SLC failure led to minor pixel loss (40% clouds & shadows and >25% SLC-off pixels Lake
Number of images remaining
R
Fidh Ali
308
0.70
Fidh Ben Nasseur
298
0.91
Hoshas
310
0.19
Guettar
302
0.66
Gouazine
269
0.92
Morra
250
0.43
Dekikira
240
0.73
lakes studied here. The calibration of clouds and SLC-affected images removed a significant 51% of images, leaving on average 282 ±27 images as the actual number of Landsat observations suitable here for observations across the 7 small reservoirs studied here (table 3.2.2). This equates to 1.5 images/month over the 1999-2014 period studied though these are unevenly distributed as seen in figure 3.1.4b due to the recent overlapping acquisitions of Landsat 5, 7 & 8 as well as the low overall cloud cover at this period (March 2013 - December 2014) over our area of interest. Significantly when studying one lake, 59% of the Landsat 7 post-2003 images were discarded due to SLC off pixels, and 31% of all Landsat images were removed due to excessive cloud presence. Variations between lakes in the number of images available were high when using the same thresholds, due to greater cloud presence on higher altitude lakes (Morra) as well as the irregular SLC interference detailed earlier. 3.2.2.4
Applying remote sensing to monitor flood dynamics
The results from applying the method to the Gouazine reservoir and 6 other lakes (Dekikira, Hoshas, Fidh Ali, Fidh Ben Nasseur, Guettar, Morra) for which stage and rating curves were available are presented here. Discrete observations & flood dynamics over time On Gouazine lake, and to a lesser extent on Dekikira, which benefit from the longest and most accurate time series, the method performed well (R =0.91 and 0.73), yielding average PDAI error of 14% and 20% and median PDAI error of only 3% and 13% error. These results provide further validation of the water detection method applied and the calibration process, and confirm its applicability to detect flooded areas accurately over another reservoir (not used in calibration) and over several years. They also confirm that multiple sensors notably Landsat 8 OLI-TIRS, with Landsat 7 ETM+ and Landsat 5 TM, can be used with same method, in line with what Vogelmann et al. (2001) showed on TM and ETM+. Though the 1:1 linear relation in figure 3.2.12 confirms that surface areas are directly provided by the remotely sensed calibration and no additional corrective factor is allowed, the figure also illustrates the greater difficulties when water levels are low and the resulting overestimation observed. When applied to other lakes, where time series are less complete, the method displayed varying levels of fit (table 3.2.2) between remotely sensed surface areas and field data and 110
100000 ●
2
y = 2453 + 1 ⋅ x, r = 0.922
●
● ●
● ● ● ● ● ● ●
● ●
● ●
●
●
● ●● ● ●
Remotely sensed surface area (m²
75000
●
●
●
●● ● ●
●
● ●
●
● ● ●
● ●
● ●
●
●
50000
● ● ● ● ● ● ● ●
25000 ●
●
● ● ● ●
● ●
0
● ● ● ● ● ● ●
0
●
●
●
● ●
● ●● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ●●● ● ● ● ●● ● ● ●● ● ●● ●● ● ● ●●●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●
●
●
● ●
● ● ● ● ● ●
●
●
●
●
●
● ● ●●
●●
● ●● ● ● ● ●
●●●
●
● ●
●
●
● ● ●
●
25000
50000
75000
Surface area − field data (m²)
Figure 3.2.12: Correlation between remotely sensed surface area and field data over 19992014 for Gouazine lake
111
150000 ●
Surface area − field data
●
Remotely sensed surface area ● ●
● ●
100000
●
● ●
●
●
● ●
Surface area (m²)
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●
50000
0
2000
● ●
2005
2010
2015
Date
Figure 3.2.13: Surface area time series through remote sensing and field data for Gouazine lake
100000 60000 ●
●
y = 2906 + 1 ⋅ x, r 2 = 0.665
● ●
●
●
● ●
● ●
75000
●
● ●
y = 10850 + 0.72 ⋅ x, r = 0.725
●
●
●
● ● ●
●
●
●
●● ● ● ● ● ●
●
● ● ●● ● ● ● ●
●
●
● ● ●
●
●
●
●
● ●●
● ●
●
●● ● ● ●
● ● ●
● ●
● ● ●
● ●
●
●
25000
● ●
●
●
● ● ●
●
●
●
●
● ●
●
●
●
●●
50000
●
●
● ●
●●
●
●
Remotely sensed surface area (m²)
Remotely sensed surface area (m²)
●
● ●
●
2
● ●
40000
●
●
●
●
●
● ●
●
●
● ●
●
●●
● ●
● ● ●● ● ● ●● ● ● ●
●
● ●
● ●●
●
●
●
●
●●
●
●
●
●● ●● ●
● ● ●
●
● ●
● ● ●
●● ●●
●
● ●
20000
●
● ●
●
● ●● ●●
● ● ●
●
● ●
●
● ● ●
● ●
● ●
●
●●
●● ● ● ●
●
●
● ● ●
●
●
0
●
25000
50000
0
75000
● ●
0
Surface area − field data (m²)
●
●
10000
20000
30000
40000
Surface area − field data (m²)
(a) Dekikira lake
(b) Guettar lake
Figure 3.2.14: Correlation between remotely sensed surface area and field data
112
illustrated a variety of issues. The reduced amplitude of the flood on the smallest lakes (Fidh Ben Nasseur and Hoshas) and crucially the rapid decline leads to difficulties. Out of 7 events on these two lakes, 2 minor events (0.5 ha flood) were completely undetected and two were detected after surface area had declined by 50% (figures 3.2.15a and 3.2.15b ). The available observations were however accurate (R =0.9), showing the potential of the method even on small lakes, subject to sufficient temporal resolution. On these lakes, 1 ha floods were completely infiltrated and evaporated within 21 days. Newer optical sensors providing images every 5 days (e.g. Sentinel-2) will improve monitoring of smallest reservoirs but remains dependent on low cloud cover. Here, the complementarity of remote sensing with rainfall-runoff modelling appears therefore necessary (cf. chapter 4) and relevant as the temporal frequency of images remains low to capture all rapid events. On other small lakes (Guettar, Fidh Ali), the method no longer fails to detect individual events, however these longer time series reveal scattering in the results. Figure 3.2.14b shows the relationship between field and remotely sensed surface areas (R =0.66) and illustrates the significant spread in the data, which also impacts the ability to derive coherent flood dynamics (figure 3.2.15c). Reducing cloud&shadows and SLC thresholds were shown to not improve the results, and instead scattering here is due to the inherent uncertainties in the method, partly incomplete detection of clouds and classification difficulties identified in calibration. These classification errors are more significant on small lakes due partly to pixel size and the greater bearing individual incorrectly detected pixels have, but also the greater proportion of mixed pixels and presence of shallow waters where reflectance of soil, vegetation and water are amalgamated. When investigating single outliers, we see that overestimation can also occur as a result of image acquisition directly after an event. The downstream parts of the wadi inlet included in the lake cell will be detected by the MNDWI and counted , whereas the stage-surface relationship will only estimate for a given cell the lake surface area. Overestimation can also occur as a result of sheet runoff surrounding the lake. In another, the gross overestimation can be attributed to incomplete detection of clouds, considering the significant clouds presence around the reservoir but not directly above it (figure 3.1.9). The addition the Landsat 8 band for cirrus clouds as well as continued research & development of new tools such as Tmask reveals the existing attempts to reduce such errors. On the largest lake studied here (Morra), the results reveal interesting insights. Firstly the method highlighted the obsolescence of the stage-surface rating curve, which was confirmed through DGPS contours of the surface area (figure 3.2.15e). Due to the lake remaining flooded through the year and the access difficulties to perform bathymetric surveys, the initial (design) HSV rating curve has not been updated over the past 20 years. Though R can not be meaningfully be assessed considering the errors in the rating curves, the method highlights its capability in reproducing flood declines, though residual difficulties due to outliers are still apparent (figure 3.2.15e). Despite the lake being larger and therefore benefiting from a greater number of pixels, we see that the method’s accuracy also depends on the amplitude in the variations in the surface area over time. On lakes such as Gouazine where surface area fluctuates between 0 m and 90,000 m within a year, whereas on Morra the amplitude of this deep lake is often contained between 65,000 m and 100,000 m . The amplitude of the surface area variations on Morra are then contained within the uncertainties of the remote sensing methods (estimated within 20% in calibration and validation) making 113
20000
50000 ●
Surface area − field data
●
Remotely sensed surface area
●
●
Surface area − field data
●
Remotely sensed surface area
● ●
●
●
●
Surface area (m²)
●
Surface area (m²)
40000
●
15000
●
10000
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
5000 ●
● ● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
●
● ● ● ●● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●
10000 ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●
2000
● ● ●
20000
●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
30000
● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
2001
2002
2003
●●● ●●●●●●●●● ●● ●● ●●● ● ●●●●●●●●● ●● ●
2011
●
●
●
●●●●●
2012
●●●
2014
2015
2016
Date
(a) Fidh Ben Nasseur
(b) Hoshas
50000
● ● ●
●
Surface area − field data
●
Remotely sensed surface area
●
●
● ● ● ● ● ● ●
● ●
●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
30000
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●
20000
10000
Surface area − field data
● ●
Remotely sensed surface area
● ●
DGPS contours
●
● ● ●● ● ●
40000
●
●
● ● ● ●
● ●
● ● ●
●
● ●●
●
● ●
● ● ●●●
● ● ● ● ● ● ● ● ●
●●
●●●
●●
● ● ●● ●● ●
●●
● ● ●
●
Surface area (m²)
40000
● ● ● ● ●
Surface area (m²)
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●●●●●●●●●●● ● ●● ●●● ●● ●
2013
Date
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●● ●● ●
● ● ● ● ●● ●● ● ●●
● ●
● ●
●
●● ● ●
● ● ●
● ●
● ● ●
● ●
●
●
● ●●
●
●
● ●
● ●
●
●
● ● ● ●
●● ●
● ●
●
20000
● ●
●
● ● ● ●
● ● ● ● ● ● ● ●●●
●
● ●
●
● ●
●
● ● ● ● ● ●
● ●●
● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ●
0
●
●
● ● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ●
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
2000
●
●
●
● ● ●
● ●
●
●
●
● ● ● ● ● ● ● ● ●● ● ● ●● ● ●
●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ●
2005
●
●
●
●●●●●●●●
●
● ● ●
● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ●● ● ● ●
● ●
●
● ● ●
●
●
● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●
● ● ● ● ● ● ● ●
● ● ● ● ●
●
2010
0
●
●●● ●●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
2015
2012
2013
2014
Date
2015
2016
Date
(c) Fidh Ali
(d) Guettar 125000
150000 ● ●
● Surface area − field data
● ●
Remotely sensed surface area
●
Surface area − field data
●
Remotely sensed surface area ●
● ● ●
100000
DGPS contours
● ● ● ● ● ● ●
125000
● ● ●
●● ●
● ● ● ●
● ● ● ●● ●
● ●●
●
●
●
● ●
●
● ●
● ● ●
100000
● ● ●● ●
●
● ●
● ● ●
● ●●
●
● ● ● ●
●
● ●
● ● ●
● ●
● ● ● ● ●
75000
●
Surface area (m²)
Surface area (m²)
●
●
● ● ●
●
75000
● ● ● ●
● ●
● ● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●● ● ● ●
●
●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●
● ●
50000
● ● ● ●● ● ● ● ●
●
● ● ● ● ●
● ●
● ● ●● ●
●
25000
●
● ● ● ● ● ● ●● ●
●
● ●
0
50000 2010
2012
2014
● ● ● ● ● ●● ● ● ● ● ● ●● ●
● ● ● ● ●
● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
2000
2016
●● ● ●● ●
●
● ● ● ● ●
● ●
●
● ● ●
● ● ● ● ●
● ●● ● ● ●
2005
● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●
●
●●
●●
● ●
●
● ● ● ● ● ● ●● ●
● ● ● ● ●
● ● ● ●
● ●
● ● ● ● ●
● ● ● ●
●
●
● ● ● ● ●
● ● ● ●
● ●
● ●
●
● ● ● ●● ● ● ● ● ●
● ●
●
●● ●● ● ● ● ●● ●● ● ● ● ● ●
●
● ●
● ●
●
● ● ● ●
2010
● ● ●
2015
Date
Date
(e) Morra
(f) Dekikira
Figure 3.2.15: Comparing remotely sensed flooded surface area and field data over 1999-2015 for 6 lakes
114
trends more difficult to extract from this scattering in the results. Accordingly, it is not only the size of the lake but also the morphology of the lake, whereby a deep, lake which has minimal surface area expansion during floods will be more affected by the uncertainties and scattering. Errors were not due to calibrating the MNDWI threshold on a range of smaller lakes as the optimal threshold when calibrating Morra alone was identical. 3.2.2.5
Daily surface area & water availability over time
Considering the end-objective of studying water availability, these individual points were then interpolated to derive a daily time series of the surface area, necessary precursor to assessing water availability over time from discrete observations. Surface areas can be converted to volumes using lake specific surface-volume rating curves or more generic power relations (cf. section 2.4) and converted into statistical information for stakeholders on the volume, timing and duration of the flood (cf. chapter 5). Here comparisons were performed using surface areas in order to avoid (at this stage) introducing additional uncertainties by applying surface-volume relationships. Figures 3.2.16a and 3.2.17 illustrate the difficulty in interpolating points with variable and non negligible time lag into a hydrologically coherent daily time series. The hydrological dynamic of steep rises associated with intense rapid floods in opposition to the gradual slow declines are notably not fully represented. Alternate interpolation methods (to the simple linear interpolation used here for illustration purposes) can be applied but will similarly struggle to accurately represent flood dynamics across different lakes. Daily time series of surface area aggregated and converted into mean annual surface area are shown in figures 3.2.16b and 3.2.18. The fit with the mean surface area assessed from stage data is excellent for Gouazine (R =0.9) and Dekikira (R =0.89). On other lakes, the absence of stage data over several periods (years) limited the number of years which could be compared, but for available years figure 3.2.19 indicates a good overall fit (R =0.88). This highlights how scattering in the outputs may prevent precise modelling of flood dynamics but the method (as such) remains capable of providing a valuable tool with which to assess water resources. Nevertheless errors certain years were not negligible (>50%) and highlighted how individual points do not carry the same importance when modelling daily surface areas. Individual outliers can naturally have a significant influence, but the associated error will depend on the lag with successive observations, as when constrained by nearby points its influence will be moderated. Furthermore, Landsat observations close to the peak of an event are more important than other observations which are situated along the gradual decline of water surface areas and which can therefore be more easily interpolated. The absence of observations for extended periods can also severely impact Landsat images’ ability to assess with reasonable accuracy mean surface areas (and thus mean water availability). In some cases, flood events will be missed whilst in other cases, in the absence of data, the surface area and thus volume is by default interpolated linearly between successive observations though water levels may have declined and risen in between (e.g. Hoshas in figure 3.2.17). Sensors benefiting from increased temporal resolution will inherently allow a greater representation of the flood dynamic (e.g. MODIS in Ogilvie et al. (2015)). Alternate optimisation of clouds & SLC could also in time be developed to maintain more observations
115
Remotely sensed surface area (m²)
1e+05
5e+04
0e+00 2000
2005
2010
2015
2010
2015
Surface area − field data (m²)
Date
1e+05
5e+04
0e+00 2000
2005
Date
(a) Interpolated daily flood time series
Mean daily surface area (m²) per year − RS data
60000
40000
20000
0 2000
2005
2010
2015
2010
2015
Year
Mean daily surface area (m²) per year − field data
60000
40000
20000
0 2000
2005
Year
(b) Annual mean surface area
Figure 3.2.16: Daily flood dynamics and mean surface area from field data and remote sensing over 1999-2015 for lake Gouazine
116
Remotely sensed surface area (m²)
Remotely sensed surface area (m²)
50000
40000
30000
20000
10000
30000
20000
10000
0
0 2000
2005
2010
2015
2000
2005
Date
2010
2015
2010
2015
2010
2015
2010
2015
2010
2015
2010
2015
Date
Surface area − field data (m²)
Surface area − field data (m²)
50000
40000
30000
20000
10000
30000
20000
10000
0
0 2000
2005
2010
2015
2000
2005
Date
Date
(b) Hoshas
Remotely sensed surface area (m²)
Remotely sensed surface area (m²)
(a) Fidh Ali
50000
40000
30000
20000
10000
100000
75000
50000
25000
0
0 2000
2005
2010
2000
2015
2005
Date
Date 100000
Surface area − field data (m²)
Surface area − field data (m²)
50000
40000
30000
20000
10000
75000
50000
25000
0
0 2000
2005
2010
2000
2015
2005
Date
Date
(d) Dekikira
150000
Remotely sensed surface area (m²)
Remotely sensed surface area (m²)
(c) Fidh Ben Nasseur
100000
50000
50000
40000
30000
20000
10000
0
0 2012
2013
2014
2015
2016
2000
2005
Date
Date 50000
Surface area − field data (m²)
Surface area − field data (m²)
150000
100000
50000
40000
30000
20000
10000
0
0 2012
2013
2014
2015
2016
2000
Date
2005
Date
(e) Morra
(f) Guettar
Figure 3.2.17: Surface area time series from field data and remote sensing data over 19992015 for 6 lakes
117
Mean daily surface area (m²) per year − RS data
Mean daily surface area (m²) per year − RS data
15000 30000
20000
10000
10000
5000
0
0 2000
2005
2010
2015
2000
2005
Year
2010
2015
2010
2015
2010
2015
2010
2015
Year
Mean daily surface area (m²) per year − field data
Mean daily surface area (m²) per year − field data
15000 30000
20000
10000
10000
5000
0
0 2000
2005
2010
2015
2000
2005
Year
Year
(a) Fidh Ali
(b) Hoshas 80000
Mean daily surface area (m²) per year − RS data
Mean daily surface area (m²) per year − RS data
10000
7500
5000
2500
60000
40000
20000
0
0 2000
2005
2010
2000
2015
2005
Year
Year 80000
Mean daily surface area (m²) per year − field data
Mean daily surface area (m²) per year − field data
10000
7500
5000
2500
60000
40000
20000
0
0 2000
2005
2010
2000
2015
2005
Year
Year
(c) Fidh Ben Nasseur
(d) Dekikira 10000
Mean daily surface area (m²) per year − RS data
Mean daily surface area (m²) per year − RS data
100000
75000
50000
25000
7500
5000
2500
0
0 2000
2005
2010
2015
2000
2005
Year
2015
2010
2015
10000
Mean daily surface area (m²) per year − field data
100000
Mean daily surface area (m²) per year − field data
2010
Year
75000
50000
25000
0
7500
5000
2500
0 2000
2005
2010
2015
2000
Year
2005
Year
(e) Morra
(f) Guettar
Figure 3.2.18: Mean annual surface area from field data and remote sensing data over 19992015 for 6 lakes
118
y = 787 + 0.84 ⋅ x, r 2 = 0.888
Mean daily surface area (m²) per year − RS data
●
60000
●
●
● ●
●
40000
SR ●
Dekikira
●
Fidh Ali
●
Fidh Ben Nasseur
●
Gouazine
●
Guettar
●
Hoshas
● ●
● ● ● ●
●
● ●
●
20000 ●
● ●
●
●
● ●
●
● ●
●
●
●
●
●
0
●
●
● ● ●
● ●● ●● ●
0
●
●
● ●
●
●
20000
40000
60000
Mean daily surface area (m²) per year − field data
Figure 3.2.19: Correlation between mean daily surface area per year for each lake using field and remote sensing data at these critical stages (flood peaks or every x days) and reduce omission and overestimation errors on these critical points through post treatment smoothing, spatial and temporal interpolation or additional imagery sources. To address the absence of observations over extended periods and problems of incoherent hydrological dynamics, we seek in the next chapter to combine remote sensing observations with a conjunct rainfall-runoff water balance model. A Kalman filter is also employed to reduce the influence of certain outliers based on the respective confidence given to the hydrological model outputs and the remote sensing observations.
3.3
Conclusions
This chapter sought to investigate the possibility of using remote sensing to assess surface area of small reservoirs over time and develop a suitable methodology. In addition to necessary radiometric corrections, appropriate detection and subsequent calibration of the number of cloud and SLC-off pixels were defined and implemented to treat series of Landsat 5 to Landsat 8 images. The performance of 6 spectral indices used in water detection were then calibrated against field data and compared to assess their performance on very small lakes where the greater presence of mixed pixels from shallow waters and vegetation leads 119
to different reflectance properties than open water. Results on 7 lakes of varying sizes and over three dates, confirmed the ability of MNDWI to detect a range of small reservoirs with high overall accuracy rates and stable thresholds. Surface area errors remained remarkably low, under 15% on 5 lakes including the smallest ones, considering the demanding objective of detecting areas under 1 ha with 30 m spatial imagery. Applied over the 546 images, the method showed its ability to monitor flooded surface areas over larger lakes. Residual errors from cloud uncertainties, due to undetected cirrus clouds and shadows, as well as greater vegetation, shallow water and associated mixed pixels led to a number of outliers which can be detrimental when studying individual flood rises and declines notably on smaller lakes. Difficulties were also encountered where the errors on individual observations were contained within the range of amplitude of the flood. Furthermore, filtering images with excessive cloud and SLC presence reduced temporal resolution to 1.5 images/month which can lead to minor floods or those on high infiltration lakes to be overlooked by Landsat time series. Nevertheless when considering water availability patterns over time as is our objective here, overall correlation between estimated mean surface area and field mean surfaces reached were high, reaching R = 0.88. These outputs were then applied and combined with field data in the following chapters (4 and 5) to assess water availability across all small reservoirs. The benefit of integrating remotely sensed surfaces areas with a water balance model through an Ensemble Kalman filter, notably sought to improve the modelling of the flood dynamics and compensate for Landsat’s reduced temporal resolution.
120
Chapter 4
Water availability modelling in an Ensemble Kalman Filter 4.1
Method overview
After developing in chapter 3 a method capable of detecting and monitoring with sufficient accuracy the surface area of lakes over time, the aptitude of remote sensing to contribute to improving water availability assessments on small reservoirs was specifically explored. Methods to combine remotely sensed information with available field data on reservoirs’ hydrological processes were researched. A daily hydrological (rainfall-runoff + water balance) model to simulate volumetric changes in small reservoirs combined with a Kalman filter to reevaluate in real time model outputs based on Landsat observations was developed here (figure 4.1.1). The benefits of this combined model were assessed on 7 gauged reservoirs and compared with results obtained using only Landsat observations or water balance modelling. Finally, the sensitivity of the model to downgrading input values and moving towards conditions found on the ungauged reservoirs was assessed.
4.1.1
Data assimilation methods
Data assimilation seeks to combine external sources of data or observations to beneficially correct or calibrate in real time (i.e. as observations become available) model outputs. Widely relied on in meteorology, it has become increasingly used in other scientific fields, including hydrology (Beven and Freer 2001; Clark et al. 2008; Moradkhani et al. 2005; Xie and Zhang 2010; Boulet et al. 2002) notably to combine the benefits of increasingly available and valuable (precise, accurate, higher temporal and spatial resolution) remote sensing data. Here, Landsat observations are used to constantly reevaluate the outputs of a reservoir’s daily water balance model based on field observations on instrumented sites. Data assimilation techniques can be grouped in two categories: variational data assimilation (e.g. 3D-Var, 4D-Var) which is notably used in the modelling of large systems such as atmospheric circulation models, oceanography etc. and stochastic data assimilation which include Best Linear Unbiased Estimator (BLUE) and Kalman filters, suited to smaller scale systems and less demanding in calculation times. 121
Colour key
Surface area data (GPS, stage)
Landsat 5-8 (30m, 16 day)
RS data
High resolution imagery (G. Earth)
Cloud, shadow & SLC-off assessment (& optimisation) Field instrumentation
Pretreatments (radiometric, atmospheric, topographic corrections)
User interviews
MNDWI water detection (& index calibration) Grid cell definition around SR
Outputs
Surface-vol relations
Flooded surface area Si+x
Remote sensing of flooded volumes Rainfall Digital elevation model
Flooded volumes Vi+x
PET
SR catchment delineation
Interpolated P
Interpolated PET
Evaporation
GR4J modelling (daily, conceptual)
Infiltration
Rainfall
H-S-V relations Releases Withdrawals
Water balance modelling (daily)
Inflow
Hydrological modelling (rainfall-runoff + water balance) Repeat x days
Vforecasti+1 Vforecasti+x Repeat over 1999-2015 546 Landsat images 51 small reservoirs
Ensemble Kalman Filter
Vobsi+x
Vupdate = V forecast + Kk*[Yobs - H(Vforecast)] Vupdatei+x
Ensemble Kalman Filter data assimilation
Figure 4.1.1: Schematic representation of methodology for hydrological study of small reservoirs
122
Classical Kalman filters (KF) widely used for linear systems can not be used here, due to the non linearity of the Surface-Volume relation used in equations 4.1.8 and 4.1.9. Ensemble (Evensen 2003) and Extended Kalman Filters are suitable for non linear systems, but the Extended KF, just like the variational methods, notably require a tangent model and deriving the adjoint. Vermeulen and Heemink (2006) described the possibility of using a reduced model to implement the adjoint and certain interfaces such as Tapenade (http://www-sop.inria.fr/tropics/tapenade.html) can help implement this, after converting initial code into Fortran. Ensemble KF is a stochastic method which removes the complexity associated with requiring an adjoint and which which has been used with success on a number of research works to treat non linear systems, including where initial states are highly uncertain (Gillijns et al. 2006) as may be the case with rainfall runoff modelling during certain events and more so on ungauged reservoirs.
4.1.2
Ensemble Kalman Filter
The Kalman filter operates in a two-step process: initial forecast and update steps. The forecast step (equation 4.1.3) seeks to model Vt+1 from Vt and is here the result of our daily hydrological (rainfall runoff + water balance) model i.e. Vf orecast (VW B+GR4J ). This new value has an associated error wk . When an external observation Vobs (here VRS from Landsat) is available, the forecast value is updated (equation 4.1.4) based on this additional data using the Kalman gain equation (equation 4.1.7). A sequential type of data assimilation, it only updates model states based on prior observations and does not integrate knowledge from observations occurring after the modelled date (Moradkhani et al. 2005). This new value, Vupdate (here VEN KF ) is then fed into our water balance daily model and updated again when a new remote sensing (RS) observation (filtered for cloud & SLC off issues) is available. In the ensemble version of the Kalman filter, we generate n values of an initial state and run each ensemble member through the forecast and update step. The n values of the initial state are generated based upon a random synthetic error y so that values have mean value initial state and predefined covariance Cy. Here where initial states are the same as Vt and not an intermediary variable or other, y can be taken to be wk . The n ensemble of external observations are also generated randomly in order to obtain a normal (Gaussian) distribution with error vk , i.e. centred on the observation value and with predefined covariance Cv. The method is indeed based upon a probabilistic interpretation of uncertainty, where yk , vk and wk errors are assumed random variables with zero mean and predefined covariances Cy, Cw andCv (Reichle et al. (2002)). They must also be assumed statistically independent, i.e. mutually uncorrelated and uncorrelated in time, i.e. white noise. In real life, noise is not Gaussian but this is a reasonable approximation and the KF algorithm seeks to correct estimation event if Gaussian noise poorly estimated (Moradkhani et al. 2005). The key is that errors are not defined as zero, as the method is designed to deal with and optimise noise as it travels through the system. Each ensemble member initial state is run through the model (forecast step) and then updated with an n observation. The new updated state is then obtained from the mean of all ensemble update states. This process is then repeated over all the time steps used in our hydrological model (cf. figure 4.1.1). n is typically comprised between 40 and 100, as below
123
40 studies show greater errors but marginal benefits in numbers greater than 100 (Gillijns et al. 2006). Considering that computer CPU (central processing unit) calculation times for the KF part was not an issue here, n = 100 was programmed in R. The Kalman equations are as follows: Vupdate = Vf orecast + Kk ú [Yobs ≠ H(Vf orecast )]
(4.1.1)
Yobs = H ú Vobs + vk
(4.1.2)
Vf orecast = f (V ) + wk
(4.1.3)
Vupdate = Vf orecast + Kk ú [Vobs + vk ≠ Vf orecast ]
(4.1.4)
Vupdate = Kk ú (Vobs + vk ) + (1 ≠ Kk ) ú Vf orecast
(4.1.5)
where H is the model to convert observed state values to the observations used here. In many cases including this study, observations are equivalent to state values and H therefore simplifies to 1. In practice, in R it is a matrix where days where acceptable Landsat observations (i.e. acceptable cloud & SLC cover) are coded as 1 and defaulting to 0 when no acceptable observations are available to update the forecast. Kk is the Kalman gain defined as: Kk = Cy ú H T ú (H ú Cw ú H T + Cv)≠1
(4.1.6)
Kk = Cy ú (Cw + Cv)≠1
(4.1.7)
where Cv (also referred to as R) is the observation error covariance matrix, Cw (also referred to as P b) is the forecast error covariance matrix and Cy is the cross covariance matrix between the state variable and the forecast. As described in equation 4.1.4, the amount the Kalman filter modifies the initial forecast based on additional observation is driven by the Kalman gain and thus the covariance values Cw and Cv. The Kalman filter can increase or decrease a forecast, but the extent is driven by the Kalman gain, which in turn translates the confidence given to either sources of data, i.e. the model’s forecast or the additional observation (of a state variable). The nature of the Kalman filtering only allows the Kalman gain to operate here within the range between the forecast and the observation i.e. drawing Vf orecast more or less close to the remote sensing observation. It can therefore increase or decrease an under or overestimated prediction but only as far as the Landsat observation value, therefore when both are overestimated, the Kalman gain can not reduce the error beyond the individual errors of using only remote sensing or hydrological model values. The method’s performance therefore improves where these covariance values can be estimated with greater precision or reliability. Inherently unmeasurable, the associated errors yk , vk and wk may be estimated but the covariances should also be tuned or constrained as
124
necessary to obtain the desired level of correction, notably varying the model error covariance Cw accordingly. In certain cases, Cv can be estimated based upon the known errors in instruments, and here would include errors from both sensors, but also radiometric correction and detection errors. Here we benefit from field observations against which we can assess the errors of remote sensing observations and our hydrological model. For Cv these can be estimated by comparing Vf ield and VRS (after removing outliers). Stage related volumes include their own element of error (ladder readings, HSV imprecisions and evolving flood bed topography) but here these are neglected compared to the errors from remote sensing and hydrological modelling. Cw can likewise be estimated from the errors observed between the forecast volumes (i.e. VW B+GR4J ) and observed Vf ield . Cw remained constant as recommended by Clark et al. (2008), though other studies also updated it after every run based on the new ensemble modelled values. This would however prevent the use of the Kalman filter in prediction mode, i.e. on other periods or lakes, as this would require continuous ground truth data to refine the covariances. For Cy, as the state variable used is directly the volume (and not an intermediary state variable), Cy is equivalent to Cw. The initial Cv and Cw values estimated using the covariances between stage observations and model outputs and stage and remote sensing observations were used. Alternate combinations were tested but these did not lead to performance improvements. These display a ratio around 1:20, i.e. Cw variance is 20 times greater than Cy variance and contribute to attributing greater confidence in the Landsat values.
4.1.3
Remote sensing inputs
The remote sensing inputs were the results developed in chapter 3, i.e. MNDWI assessments on Landsat 5-8 images of surface areas for individual lake cells, filtered to images with less than 40% clouds and 25% SLC off pixels. These were converted using the available rating curves detailed in chapter 2, and the developed inter small reservoirs rating curves for ungauged reservoirs described in section 2.4. In order to allow comparison over time with available field data (V f ield ) and the Ensemble Kalman Filter (VEN KF ), the remote sensing data (VRS ) was linearly interpolated to provide a continuous time series. Spline interpolation, Lowess smoothing, and the TSGF spline function (Forkel et al. 2013) approach to gap fill and smooth daily time series were tested but failed here to provide significant benefit (cf. figures 9.3.1 in chapter 9). The latter function was notably developed to smooth remote sensing outputs such as NDVI values. Attempts at smoothing the ENKF outputs using smoothers notably Loess also failed to improve significantly the results (e.g. figures 9.3.1). Smoothing difficulties occur as time series do not follow a seasonal pattern, like large floods in wetlands (Ogilvie et al. 2015), making it harder for the algorithm to ascertain which values must be dropped and on the contrary which peaks must be kept.
4.1.4
Water balance modelling
The remote sensing observations are here combined with a daily water balance (WB) model of small reservoirs. These have been used in many locations around the world including Sub-Saharan Africa (Desconnets et al. 1997; Martin-Rosales and Leduc 2003), Brazil (Molle 125
Figure 4.1.2: Water balance fluxes of small reservoirs 1991), Mexico (Avalos 2004), India (Massuel et al. 2014b) and Tunisia (Grunberger et al. 2004; Zammouri and Feki 2005) and seek to account on a daily basis for volumetric differences in water in the reservoir based on observed or modelled assessments of all input and output fluxes. The multiple fluxes which compose the water balance of small reservoirs are illustrated in figure 4.1.2 and can be assessed through a combination of direct field measurements, rainfall-runoff modelling as well as geochemical methods (Montoroi et al. 2002). Due to the numerous fluxes involved (equation 4.1.8), accounting for daily changes in volume even on gauged reservoirs requires estimating, extrapolating and/or neglecting certain fluxes, based on reasonable assumptions. Considering the objective of testing the feasibility of transferring the method to ungauged lakes, the water balance model was designed to be fed by widely available data, and/or estimates or rules based on instrumented reservoirs. Daily runoff into the reservoir is assessed using a simple GR4J (Perrin et al. 2003) rainfall runoff model to account for limited data availability precluding the use of a distributed physically based model. Inputs are rainfall, and potential evapotranspiration (assessed as in sections 4.3.1 and 4.3.3) and the model is calibrated on gauged reservoirs. The required starting volume (V0 ) was set to coincide on ungauged lakes with a Landsat observation. Though rainfall, reservoir stage and pan evaporation can be measured directly, difficulties occur due to the uncertainties in estimating the other fluxes, groundwater fluxes (infiltration and groundwater inflow), withdrawals and evaporation (Li and Gowing 2005). These are estimated through local understanding of hydrological processes based on ongoing field instrumentation on four reservoirs and previous studies in the catchment, focussing notably on depletion phases, when the water balance simplifies, reducing the number of unknowns. The values, rules or mathematical relationships derived for these fluxes are presented in sections below. A daily time step is used in the water balance, as higher temporal resolution data is not available, except over certain periods for rainfall and stage measurements but this was not systematic. —V = P + GWin + Q ≠ E ≠ I ≠ W ≠ R ≠ O ≠ L
(4.1.8)
A volumetric balance, P and E values are notably multiplied by the surface area, which 126
as shown below leads to a non linear system, based on equation 2.4.1. Sn≠1 = f (Vn≠1 ) = (
4.1.5
Vn≠1 —1 ) B
(4.1.9)
Model sensitivity and performance on ungauged reservoirs
The performance of the Kalman Filter (VEN KF ) were assessed against available field data (Vf ield ) and compared with the performance of using only hydrological model (VW B+GR4J ) and only remote sensing (VRS ) data. Nash Sutcliffe Efficiency (NSE) values were calculated but considering their sensitivity to timing of outputs and ability to disguise certain errors (Moussa 2010), Root Mean Squared Error (RMSE) and Normalised RMSE (over maxmin values) were also used. The performance in terms of individual daily volumes was investigated as well as on annual water availability, considering the end objective of exploring interannual water availability patterns. After combining remote sensing observations with specific models developed for 7 gauged reservoirs (i.e. relevant I, withdrawals, calibrated GR4J, updated HSV) the method’s performance as these inputs are degraded was tested. This sought on one hand to study the model’s response to greater uncertainties in the input variables and parameters, to identify the sensitivity of the model to these inputs and to identify where additional field data can provide most benefits. On the other hand, this led to identifying the level of uncertainty if applying an ENKF model based on generic rules on all ungauged SR in the catchment. Evaporation, transposition coefficients, overflows, releases, withdrawals were interpolated based on basin wide data or assumed to follow similar rules across all lakes. The ways these variables are assessed are described in section 4.3 below. The same KF parameters were also used across all gauged reservoirs. These inputs can therefore be derived the same way for gauged and ungauged reservoirs. Ungauged reservoirs distinguish themselves from gauged reservoirs, by the fact that there is no rainfall gauge in the catchment, no estimated runoff (Qobs ) against which to calibrate the GR4J model, no prior knowledge of infiltration, and no specific HSV updated over time. These must be extrapolated, estimated or transposed from the information gathered across the 14 gauged and 7 modelled reservoirs as also described in sections below. The influence of uncertainties from limited P observation networks, from using standard GR4J parameters, from using a standard I rule, and from using the interSR relationship on the WB model, as well as the ability of RS observations to correct for these uncertainties was therefore assessed. As before, errors were assessed in terms of RMSE and NSE based on both individual values and annual water availability statistics. Two lakes (Gouazine and Fidh Ali) were chosen considering the relative reliability of the P , E, stage (Z) and updated HSV relations and the length of their observations, against which modelled annual availability could be compared. Two further lakes were chosen as they displayed significantly different conditions, in terms of I, HSV and GR4J parameters, (i.e. their behaviour complied less with our generic rules) and allowed for testing the model in significantly degraded conditions. These four lakes were not subject to additional fluxes such as major withdrawals & leaks, reducing potential additional uncertainties. The results guided the procedure used to generalise the approach and assess water availability on all lakes.
127
4.2 4.2.1
Runoff modelling with GR4J Model structure
To estimate runoff (Q) into small reservoirs and feed into the daily WB model of lakes, the GR4J (Perrin et al. 2003) rainfall runoff model was used. Its choice was driven by several factors. A lumped conceptual model with four parameters, it is well suited to the relative scarcity of data in these catchment and to the objective of using the model of several ungauged catchments. Developed and successfully employed in numerous semi arid catchments of comparable size, it requires only rainfall (P ), potential evapotranspiration (P ET ) and catchment size as variables and relies on a simple two reservoir structure and setting four parameters, which notably seek to account for antecedent soil humidity, important in this context (Ogilvie et al. 2016). The structure of the GR4J model and full details of the model structure are described in Perrin et al. (2003). The four parameters are: • X1 production store capacity [mm] • X2 groundwater exchange coefficient [mm/d] • X3 routing store capacity [mm] • X4 unit hydrograph time constant [d] Despite this conceptual approach, the model is in fact empirical, as was developed on a great number of data sets which progressively led to defining and optimising the model structure and operation. This reduced number of parameters also ensures greatest robustness and facilitates the transposition of model parameters to neighbouring catchments (Perrin et al. 2003). As with all hydrological modelling, moving from empirical and black box type models down to physically based models allows increasing opportunities to represent and modulate the range of physical processes active in different catchments, and adapt these as they evolve over time. Increasing model complexity however multiplies the constraints on the resolution, availability and reliability of data, as well as potentially increasing computing and operator time. Model selection is then a compromise between sufficient detail to represent the processes necessary to assess the objectives and composing with data availability. A daily time step was used to capture the short intense rainfall events in the catchment and to concord with both the available data (P , P ET , Qobs to calibrate) and the water balance time step. The airGR code developed for R by Laurent Coron and Charles Perrin was used in beta testing. Its behaviour and results were initially compared with the Excel version of the model which confirmed its reliability. The R version notably allowed for additional optimisation functions, including the HBAN detailed below, which yielded superior results (R increase from 0.69 to 0.72 over short calibration periods).
4.2.2
Model calibration
4.2.2.1
Runoff estimation from water balance
To calibrate the GR4J model, runoff (Qobs ) needed to be assessed on these 7 gauged reservoirs. In many cases including in the Merguellil catchment, runoff can not be measured 128
directly due to the unstable riverbed which causes the stage discharge relationship to evolve over time and due the diffuse sheet runoff and subsurface runoff bypassing the hydrometric monitoring equipment. Its estimation required to calibrate and validate non physical rainfall runoff models must instead be estimated though the water balance method. The high rainfall intensity characteristic of Mediterranean climates combined with limited vegetation, low soil water holding capacities and prominent topography can lead to violent Hortonian runoff events (Lacombe et al. 2008). During floods, the water balance equation can then be simplified as several fluxes can be neglected (GWin , E, I) or assumed null (withdrawals) over 24 hours. The latter are typically absent during floods, due to reduced irrigation needs (greater rainfall, reduced potential evapotranspiration) and pumps being displaced to prevent damage from flooding. Here GWin and withdrawals were neglected, and the water balance simplified to equation 4.2.1. Releases though rare, occur on the most significant events (1% of all events, Lacombe 2007) and must be accounted for to estimate the actual runoff from these extreme events. In 95% of cases, releases occurred after the flood and could be identified based on significant decreases (around 10 000 m3 /day) on instantaneous (15 min) time series, as described in section 9.2.2. Overflows must also be accounted and were estimated on several lakes using a stage ladder in the spillway, a rating curve and observations from the dam operator, as discussed in section 9.2.3. Other fluxes (P , E, I) were assessed as detailed below on a daily basis. The change in volume was assessed using available stage measurements and relevant rating curves detailed in section 2.4. Q in m3 /day values were converted to mm based on the catchment sizes assessed as detailed in section 2.2.3. On days without stage values, Q was set to NA which are recognised by the model and not accounted for in the calibration of the model. These nevertheless lead to certain difficulties as discussed in the results. Q = —V ≠ P + E + I + O + R 4.2.2.2
(4.2.1)
Calibration steps
Default initial states (i.e. 0) and reservoir levels which correspond to 50% of the production and routing stores were used. By default a 1 year warm up period is used by the model, as it corresponds to an acceptable duration to initialise the model and notably for the initial states, here the stores (reservoir levels), to evolve to realistic values (Seibert and Vis 2012). The stage time series available for each lake are shown in section 2.3.4 and overlapped in all cases with our Landsat observations which began in 1999 were used. Rainfall and P ET data (cf. section 4.3.3 below) were interpolated for each subcatchment over 1997-2014 in order to provide gap filled continuous time series as inputs. The objective function, i.e. error criterion to minimise, was NSE calculated on initial flows which gives large importance to difference between observed and modelled volumes. This ensured large events were well accounted for, as required here considering their determining influence on water availability. NSE, close to R , is widely used in hydrology and can be calculated with flow (Q) square root (sqrtQ) or log (lnQ) transformed flows. sqrtQ gives more importance to both high and low flows, while lnQ gives most importance to the lowest flows. NSE on sqrtQ was considered but found to be disproportionately affected by the numerous low flow days and therefore failed to reproduce the larger floods as well.
129
A boolean criterion to define events over 10 mm rainfall in the time series against which to optimise parameters was also specified to further reinforce the importance given to larger events considering the limited number of events which generate floods. Using larger rainfall thresholds did not lead to improved model performance. This criteria also helped reduce the influence of the water balance uncertainties which on days of low rainfall lead to greater errors on Qobs , and would therefore be more difficult to model accordingly. Qobs can even as a result become marginally negative and was therefore artificially set to 0 for modelling purposes. A specific function (CreateInputsCrit) also allowed specifying instances when Qobs is null and an user defined epsilon value (0.0001) was added to these time steps to remove computation problems (null values). q (Qobs ≠ Qsim ) N ash = 1 ≠ q (Qobs ≠ Qobs.mean )
(4.2.2)
The R airGR package provides the HBAN optimisation function which used the steepest descent method or a local optimisation quasi-newton method (Limited-memory BroydenFletcher-Goldfarb-Shanno B, i.e. L BFGS B method). Both were tested here but the HBAN optimisation algorithm provided superior results. Both are iterative algorithms, and starting values can be provided for the parameters but using the recommended parameters (5.9, 0, 4.5, 0.2) were not shown here to improve model performance. Restrictions such as range or fixed values for certain parameters (e.g. time of transfer through catchment) or Bayesian parameter sets can also be defined. Considering the limited number of parameters in the model, no prior sensitivity tests of the model to parameters changes was carried out and all were sought to be optimised. The optimal parameters for each catchment are shown in table 4.2.1 with the Nash values obtained over the whole periods. X1 and X4 values notably appeared coherent but X2 and X3 were not necessarily physically relevant. The suitability of parameters sets over time was tested and corrected as necessary as certain parameters led to unstable behaviour over time, e.g. base flow of 5-10 mm/day being incorrectly generated after several years on Fidh Ali. Table 4.2.1: Optimal GR4J parameters sets and R values for each catchment X1
X2
X3
X4
R
Gouazine
20.680
-10687.743
1040.510
0.5
0.62
Fidh Ali
20.372
-10795.156
1050.967
0.5
0.28
Fidh Ben Nasseur
3.929
-4702.422
5889.227
0.685
0.34
Guettar
0
-4374.832
1045.836
0.989
0.23
Hoshas
61.576
-7794.297
215.855
1.272
0.56
Dekikira
0.141
-16.362
38.775
1.813
0.58
Morra
0
5.392
4090.426
0.5
0.33
4.2.2.3
Transposing parameters to ungauged reservoirs
Transferring parameter sets typically relies on identifying similarities between catchments based on their location or physical properties (Parajka et al. 2005) or employing statistical
130
(e.g. Bayesian) methods, but remains complex and prone to uncertainties. Considering the limited data availability of each catchment, no individually physically based comparison could be performed between catchments, which varied in sizes, slope, lithology, land cover (forest vs cropped), and varying levels of water and soil conservation works. Here, considering the objective of combining GR4J model outputs with remote sensing observations, the influence of using a standard set of parameters across the 7 small reservoirs was assessed to see how significantly it affected the runoff modelling and how the Kalman filter may compensate or not for this. Parameters from four lakes showing modest similarities (Fidh Ali, Fidh Ben Nasseur, Gouazine, Guettar) were therefore used. R values remained close when interchanging parameter sets between these 4 lakes, confirming their relatively similar rainfall-runoff response (runoff coefficients).
4.3
Estimating and interpolating the other water balance fluxes
4.3.1
Interpolating rainfall
4.3.1.1
Spatial interpolation methods
Daily rainfall values at the lake and across the lake’s subcatchment were required here as inputs to the water balance and GR4J models respectively. These were spatially interpolated from the network (described in 2.3.2) of pluviometric stations within and around the reservoir catchments to provide all lakes including ungauged ones, with complete daily time series between 1999-2014. On gauged lakes, interpolation is required both to fill data gaps at the gauging station (figure 2.3.2) and for catchment rainfall estimation. Rainfall interpolation relies on positive spatial autocorrelation principle whereby nearby points which tend to have similar values are combined and weighted appropriately to assess rainfall. Deterministic methods seek to identify suitable weights of the nearby observations/stations based notably upon their respective locations. Inverse distance weighting (IDW) as the name suggests attributes a weight inversely proportional to the distance between the station and the point of interest. Thiessen polygons (also known as Voronoi polygons and Dirichlet cells) determines polygons which represent the surface area of influence of a given station and determines accordingly the weight to use in the interpolation. This was shown at the catchment scale (chapter 6) and by other studies (Goovaerts 2000; Van Der Heijden and Haberlandt 2010) to not perform as well as IDW. Geostatistical methods such as Kriging seek to model statistically the way the variable varies over space, i.e. in addition to distance between observations, it can therefore seek to account for another variable (often altitude for rainfall), using methods such as ordinary, simple, cokriging and kriging with an external drift (KED). These methods have been widely used across the world and are fully documented in several papers including Hengl et al. (2007). Studies disagree on the benefits of incorporating a digital elevation model over simple IDW, which depends on the strength of the relationship between altitude and rainfall (Wackernagel 2004; Van Der Heijden and Haberlandt 2010; Feki et al. 2012), which may be influenced by orographic and coastal effects. Comparative benefits also depend on the extent to which this altitudinal bias is inherently accounted for in the distribution of rain gauges. 131
KED
IDW
80
●
P interpolated at station
60
●
40
● ●
●
●
● ●
●
●
● ● ●
●
●
●
● ●
●
20
●
●
0 0
●
●
● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●●
40
●
●
●
●
●
●
●
● ●
● ●
●
●
●
● ●
● ● ●
●
●
● ●
● ●
● ●
● ●● ● ●● ● ● ●● ● ● ●● ● ● ●●
80
120
0
●
●
● ●
40
●
80
120
P at gauging station
Figure 4.3.1: Scatterplot of observed and interpolated rainfall at Gouazine 4.3.1.2
Inverse distance weighting interpolation of rainfall
IDW was used after comparing known rainfall values at instrumented sites with the value obtained using IDW and KED interpolation of nearby stations (i.e. with the catchment stations removed) showed marginal benefit of KED over IDW. On two reservoirs (Gouazine and Morra), both methods led to high errors, over 50% (25 mm) as illustrated in figure 4.3.1 and table 4.3.1. 53 and 35 events over 20 mm were considered here based on their importance in generating runoff and flooding reservoirs (cf. chapter 6). Interpolations were performed in R with the HydroTSM package automated to interpolate values at each lake cell (blocks) over successive dates. A cropped SRTM V3 30m digital elevation model (DEM) tile was used (cf. section 2.2.3). Altitudes of each stations were extracted from the DEM, and their coordinates collated and corrected using available inventories and high resolution imagery available freely through Microsoft Bing Maps and Google Earth). The calculation unit used for IDW was 30 m in order to be contained within single lake cells and coherent with that used with KED. Lake cells were delineated as detailed in section 2.2.4. Resulting rainfall values were then multiplied by the lake surface area, obtained by converting the modelled Vn≠1 values using site specific rating curves for gauged reservoirs and the interSR power relation for ungauged reservoirs. IDW as a deterministic approach also ensures that where a value at the station is available the interpolated value will match this value unlike kriging and other geostatistical approaches. Kriging is also substantially slower (prohibitively so, as 1 day required over 10 minutes, meaning 1 year could take over 2 days and the 15 years studied over 30 days).
132
Table 4.3.1: Errors from IDW and KED rainfall interpolation compared to observations at Gouazine rainfall gauge KED
IDW
Mean error (mm)
25
26
Mean error (%)
66
70
Error standard deviation
23%
29%
Error range
11 - 101%
0 - 100%
Here interpolation was not sensitive to the actual method but suffered from rainfall gauge densities too low to deal with the strong spatial variability in rainfall events. This is characteristic of rainfall patterns within Mediterranean climates (Neppel et al. 1998) where very localised storm cells can be undetected by nearby rainfall gauges and which no interpolation method can restitute without external data. At the event scale, altitude is therefore not sufficient to correctly modulate over space the amplitude of events. Suitable variograms to improve geostatistical interpolation are more complex and advanced methods combining daily meteorological satellite observations (Meteosat, TRMM, GPM, JAXA Global Rainfall Watch) or ground radar or even phone signal networks (Overeem et al. 2013; Doumounia et al. 2014) could help improve interpolation. Hydrological studies must meanwhile cope with these large errors which are an important source of uncertainty in rainfall runoff modelling. Their influence on model outputs were further discussed in section 4.4.4.2. Catchment rainfall for GR4J was interpolated using IDW over the catchments delineated as described in section 2.2.3.
4.3.2
Interpolating lake evaporation
4.3.2.1
Spatial & temporal interpolation
Evaporation values (cf. section 2.3.3 ) were here interpolated over time (to fill gaps) and space (to ungauged lakes) using a combination of spatial interpolation and linear regression between each lake and the El Haouareb dam which benefits from continuous observations. Interpolations were carried out with the R HydroTSM package as for rainfall. Monthly values were used considering the relatively minor intra-monthly differences, except on rainfall days. Localised rainfall can also generate large uncertainties as reduced evaporation on nearby stations may not be representative of rainfall and evaporation at the interpolated station. This was also coherent with the final objective of assessing flood trends and annual water availability, not daily fluctuations. Dealing with spatial variations Variations in monthly evaporation values based on data from 10 reservoirs over 1995-1999 period confirmed high variability between lakes (figure 4.3.2), where monthly mean values had variation coefficients ranging between 8% and 51%. Greater in absolute terms in summer months (reaching s.d. of 67 mm), they were proportionally greater during winter month. Differences were more moderate than rainfall variations as can be expected considering the more uniform temperature and air humidity over large areas. Less sensitive to spatial variations, evaporation can reportedly be assessed
133
400 ● ● ● ●
● ● ●
●
300
●
● ●
●
Monthly evaporation (mm)
● ● ●
● ●
●
● ●
●
● ● ● ●
● ●
●
● ● ● ● ●
● ●
●
●
●
●
100
● ● ● ●
●
● ● ● ● ● ● ●
● ● ● ●
● ●
● ● ●
● ● ●
● ● ● ●
●
● ●
● ● ● ● ●
●
● ● ● ● ●
● ● ● ● ● ● ●
● ● ● ● ● ●
●
●
●
● ● ● ● ● ●
●
●
● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ●
● ●
● ● ●
● ●
●
● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ●
●
● ●
●
● ● ●
●
●
●
● ● ●
● ●
● ● ● ●
● ●
● ● ●
● ● ●
● ● ●
●
● ●
● ● ● ● ● ●
●
● ●
●
● ●
● ● ● ● ● ●
● ● ● ● ● ●
● ●
●
●
●
● ●
● ●
● ●
●
● ●
●
●
● ● ●
●
● ● ● ● ●
● ● ● ● ● ● ●
● ●
●
● ●
● ● ● ●
●
●
● ● ● ●
●
●
● ● ● ● ●
● ● ● ● ●
● ●
● ● ●
● ● ● ● ● ●
● ●
● ● ● ●
● ●
● ●
● ●
● ● ●
● ● ●
● ● ●
●
● ● ● ● ●
● ●
●
● ● ● ●
●
● ● ●
● ● ● ●
●
● ● ● ●
● ●
● ● ●
● ●
●
● ● ● ●
● ● ● ● ● ● ●
● ● ● ●
●
●
●
●
●
●
●
●
● ● ●
●
● ●
●
●
● ●
●
● ●
●
● ● ●
●
● ●
● ● ● ●
● ● ●
●
●
●
●
● ●
●
●
●
● ●
200
●
● ●
● ● ●
●
● ●
●
●
● ● ●
● ●
●
●
●
●
●
●
● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●
0 1996
1997
1998
1999
Figure 4.3.2: Monthly evaporation over 1995-1999 for 10 lakes (with mean ±1 s.d.) 2500
● ●
2000
●
● ●
●
●
Annual evaporation (mm)
●
1500
●
●
1000
500
0 200
400
600
800
Altitude (m)
Figure 4.3.3: Mean interannual evaporation over 1995-1999 for 10 lakes against altitude (with linear fit and 95% confidence interval) 134
Table 4.3.2: Errors from IDW and KED interpolation of evaporation compared to observations on 2 lakes Lake
Interpolation method
Mean error
Error standard deviation
Fidh Ben Nasseur
KED
15%
14%
IDW
14%
13%
KED
40%
35%
IDW
30%
21%
Hadada
using data every 50-60 km (Baccour et al. 2012). Evaporation pans in the catchment conform with this distance but are not distributed in every direction. IDW spatial interpolation was used to estimate E values at all lakes over 1999-2008 when sufficient stations (5) were available. Split samples done using E data from 6 stations and assessing the errors on interpolation values over 1995-1999 using KED and IDW at 4 stations showed greater performance of IDW. R reached 0.95 and 0.85 for IDW and KED respectively but both tended to underestimate evaporation in the higher months. The linear regression between altitude and mean interannual evaporation over 1995-1999 for 10 lakes tends marginally towards a reduction in evaporation as altitude increases, but the poor fit at the extremities indicates low confidence (figure 4.3.3), with two lakes (Sadine and Abdessadok) notably experiencing very high evaporation despite their high altitude. Evaporation at the scale of water bodies depends essentially on temperature and the deficit of air saturation; heavily influenced by wind which fuels evaporation by shifting saturated lower layers and non saturated higher layers over the lake or pan (Musy and Higy 2004). Altitude is expected to influence evaporation partly as it reduces temperature and thus the evaporative power of air, but other factors here counteract notably greater wind exposure at higher altitude. Greater uncertainties on Hadada than Fidh Ben Nasseur, despite being closer to evaporation gauges (6 km vs 20 km) also highlight the localised variations in solar and wind exposure. Where full meteorological data (wind, solar radiance, relative humidity) are available these can usefully improve geostatistical interpolation. KED also improves its performance when a large number (at least 5, ideally 10 and more) stations are available to derive a suitable and stable variogram (Baccour et al. 2012). Alternate sources of data such as daily 1 km MODIS solar radiance data could also be investigated. Dealing with temporal variability After 2008, evaporation values were not reliably available in the northern and higher altitude part of the catchment (figure 2.3.3). In order to complete time series for the later years, 2x2 relations between each interpolated lake and El Haouareb were developed to modulate monthly values according to the general tendency up to 2014. Interannual variations though minimal at an annual scale hide significant variability over certain months (figure 4.3.5 and table 4.3.3). Disparities were proportionally more significant during the winter and spring months when evaporation is lower but also from greater climatic variability, where the earlier onset of warmer weather can increase E drastically from 70 mm in April 1996 to 192 mm in April 2000. Linear regressions between stations rely on the same positive correlation principle that
135
●
●
●
300 ●
●
●
●
Fidh.Ben.Nasseur_IDW
●
●
● ● ● ●
200
●
● ●
● ●
● ● ● ● ●
Interpolated value (mm)
●
100
● ●●● ● ●
● ●● ●
●
● ●
●●● ●
● ●
● ●
● ●
●
●
●
300
●
● ●
●
●
●
Fidh.Ben.Nasseur_KED
●
●
● ● ●
200
● ●
● ● ● ● ● ●
● ● ● ●●
● ● ● ●
●
● ●
●● ● ●
● ●
●
● ● ●
100
● ●
●
●
●
100
200
300
Station value (mm)
(a) Fidh Ben Nasseur
400
●
●
●
●
300
● ● ●
Hadada_IDW
●
● ●
●
●
200
● ●
●
● ●
Interpolated value (mm)
●
100 ● ● ● ●●
●● ● ●●● ●●
● ● ●● ●
●
●
● ●
●
● ●●
●
● ● ●
●
400
●
300
● ●
●
●
200 ●
●
●
100
●● ● ●
● ●
● ● ●● ● ● ●●
●
● ● ●
●
● ●
●
● ●
●
●
● ●
●
●
●
●
●
●
Hadada_KED
●
●
●●
●
100
200
300
Station value (mm)
(b) Hadada
Figure 4.3.4: Scatterplots of evaporation observations vs IDW and KED interpolated values (with linear fit and 95% confidence interval) 136
400
● ●
●
● ● ● ● ● ● ● ● ● ●
300
●
Monthly evaporation (mm)
● ● ●
● ● ●
● ●
●
● ●
●
● ●
● ●
●
● ● ● ●
● ● ● ● ●
200
● ●
●
● ● ● ●
● ●
●
● ●
● ● ●
●
●
● ●
● ● ● ●
●
100
●
● ● ● ● ● ● ●
● ● ● ●
● ● ●
● ●
● ● ● ● ● ● ●
● ● ●
● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ●
● ●
● ● ● ●
● ● ● ● ● ● ● ● ● ●
● ● ● ●
●
● ● ● ●
● ● ●
● ●
● ●
●
● ● ●
● ●
●
0 October
January
April
July
(a) Gouazine (1995-2014)
400 ● ● ● ● ●
Monthly evaporation (mm)
300
●
● ● ● ● ● ● ● ● ● ● ●
●
200
● ● ● ● ● ● ● ● ● ● ● ● ●
●
100
●
●
● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ●
● ● ● ●
● ●
●
●
●
● ● ● ● ● ●
●
● ●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ●
● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ●
●
● ● ● ● ●
● ●
● ● ● ● ●
July
(b) El Haouareb (1993-2014)
Figure 4.3.5: Interannual variations in monthly evaporation 137
● ● ● ● ● ●
● ●
● ● ●
April
●
●
● ●
January
● ●
●
● ● ● ●
0 October
● ● ●
● ●
● ●
● ● ● ● ●
● ●
●
300
y = −0.978 + 0.974 ⋅ x, r 2 = 0.962
●
●
● ●
Fidh.Ben.Nasseur_IDW
●
●
● ●
● ● ●
200
● ●
●
●
●
●
● ● ●
100 ●●
● ● ●●●● ● ● ●
●
●
●
●
●
● ●
●
● ● ● ●
●
●
●
● ●
●
2
y = −22 + 1.16 ⋅ x, r = 0.94
● ●
300
●
●
● ● ●
●
●
●
●
200
Fidh.Ben.Nasseur_data
Fidh Ben Nasseur monthly E (mm)
●
●
● ●
●
● ● ●
● ●
●
●
●
●
●
●
●
100 ● ● ●
●
●● ●● ●●
●
● ● ●
●
●
● ●
100
200
300
El Haouareb monthly E (mm)
(a) Fidh Ben Nasseur
400
●
2
y = −22.4 + 1.05 ⋅ x, r = 0.949 ●
300
● ●
● ●
200 ●
●
●
● ● ●
100 ● ●
● ●
● ●● ● ● ● ● ●
●
●
● ●
●●
● ●
●
Hadada_IDW
●
Hadada monthly E (mm)
●
●
●
●
●
●
● ● ●
●
● ● ● ●
0 400
300
●
y = −28.2 + 0.944 ⋅ x, r 2 = 0.929
●
● ●
●
● ●
●
200
● ●
●
●
●
●
● ● ●
100
●
●
●●
● ● ●●● ●● ● ● ●
● ●
● ● ●
●
● ●
●
Hadada_data
●
● ●
●
●
● ●
●
0 100
200
300
El Haouareb monthly E (mm)
(b) Hadada
Figure 4.3.6: Scatterplots of evaporation observations at El Haouareb vs observed and IDW interpolated values for 2 lakes (with linear fit and 95% confidence interval) 138
Table 4.3.3: Interannual variations in annual evaporation Data period
Mean E
s.d.
Annual CV
Max CV
Gouazine
16 years
1814 mm
154 mm
8.5%
30% (49 mm)
El Haouareb
22 years
2074 mm
152 mm
7.3%
22% (34 mm)
IDW interpolation also uses. In effect, IDW interpolation allows a different E variogram (E variations) across the catchment to be derived for each month. After 2008, the mean variogram over 1999-2008 is used across the catchment, and modulated over time based on observations on 1 station downstream. This assumes that evaporation variations over time are homogeneous across all areas of the basin, which though not the case, is an acceptable compromise considering the absence of relevant data. Lacombe (2007) notably showed good correlation between stations within the catchment and figure 4.3.6 illustrates on Hadada and Fidh Ben Nasseur the high quality of the fits (R > 0.92) obtained here between monthly values at El Haouareb and both stations when using station data and data resulting from IDW interpolation. The strong relationship (intercept, slope, R ) between each lake and El Haouareb are detailed in table 9.1.1, confirming linear relationships may be used. The associated errors of using linear regressions on Fidh Ali reach 12.6% s.d. 11.2% and when combined with IDW interpolation reach 13.4% s.d. 11.2%. On Hadada these are much greater, reaching 29.7% s.d. 24.4% and 53% s.d. 39.6%, respectively, when both sources of uncertainties are combined. This corresponds to an error below 2 mm most months but of 4.5 mm during the July and August on a daily E of 11.5 mm. In the absence of large withdrawals, a difference/error of 4.5 mm will affect the rate of the flood decline, as evaporation and infiltration are both typically in the range 1-10 mm/day. Values were estimated using the linear relation only after 2008 to reduce errors on previous years and retain as much as possible the site specific available data. The implications of using IDW and linear interpolated values vs direct observations on the modelling outputs were investigated in section 4.4.4.2. Where several stations contain data over the whole period, more accurate linear relations between subsets of stations could be derived, accounting for proximity and altitude. Using Gouazine (for example only, as available data does not allow linear interpolation over 2008-2014) situated at higher altitude than El Haouareb improved the linear relationship and errors reduced marginally to 40.4 % s.d. 24.4% at Hadada but increased errors on Fidh Ben Nasseur to 24.6% s.d. 18.0%. Seasonal variograms or interpolation combining additional meteorological data could also be developed. 4.3.2.2
Transposition coefficients for lake evaporation
Pan coefficients are required to convert pan evaporation values to lake evaporation and account notably for the fact that the sides of the pan heat up more quickly. Values in the literature vary between 0.5 and 1 (e.g. 0.6 for ET0 from a Colorado pan in Allen et al. (1998)) but these must be assessed locally in order to account for climatic differences and lake characteristics. Greater depth, lower water temperature and greater salinity can indeed reduce evaporation (McMahon et al. 2013). At higher temperatures when E reaches 12 mm/day, the pan heating effect is more noticeable leading to lower pan coefficients. This reduction was observed empirically during summer months (Alazard et al. 2015; Linacre 139
1.2
●
1.0
●
● ●
Pan coefficient
●●
●
Study
● ● ● ●
0.8
●
Lacombe
●
Molle
●
●
●
0.6
0
10
20
30
Lake size(ha)
Figure 4.3.7: Relationship between pan coefficient and lake size (with linear fit and 95% confidence interval) 1994). On the El Haouareb dam in the catchment, a detailed water balance study (Alazard et al. 2015) showed that the pan coefficient reached 1 over March to May and reduced to 0.76 the rest of the year, leading to a value around 0.8 on the whole. Elake = Ct ú Epan
(4.3.1)
On small water bodies, studies showed an inverse relationship with surface area (figure 4.3.7), explained by the fact that large bodies accumulate greater humidity above the reservoir which reduces the evaporation process, fuelled by the necessary relative difference in air humidity (Cadier 1996; Molle 1991; Riou 1972). Molle (1991) showed a marginal increase in evaporation on smaller lakes, and Ct estimated around 0.9 for lakes under 1 ha, 0.85 up to 5ha and 0.8 between 5-10 ha. On 12 Tunisian lakes, Lacombe (2007) used a regression model to estimate both the evaporation pan coefficient and infiltration (based on a leakage coefficient) and showed a similar trend (figure 4.3.7). The method was however found to be inoperative on the lakes where infiltration is constant, and the obtained Ct values varied between 0.68 and 1.56 and were declared to not necessarily be true transferable values, but rather parameters suited to this particular model. Others (Grunberger et al. 2004) on the same lakes had used 1 for simplicity, assimilating the values in the pan to E on the lakes. Considering the body of work available on evaporation and that many studies converge over a Ct of 0.8 in semi arid areas, and for water bodies of this order of magnitude, this value was used here. Though Ct may vary a little between seasons, lake size, salinity and as water surface areas decline, the sensitivity of evaporation values to incremental changes in Ct is limited. Uncertainties on Ct of 25% (i.e. Ct = 0.8 ±25%) result in a change in daily evaporation by 1-2 mm which is clearly minor compared to the uncertainties on other 140
depletion fluxes, notably infiltration. Where the focus is on flood dynamics and single flood declines, the importance will be greater, but here in the context of annual water availability assessments, this uncertainty can be neglected.
4.3.3
Potential evapotranspiration for catchment modelling
Inherently difficult to measure over a catchment, potential evapotranspiration, which represents the evaporation and transpiration if water was not a limiting factor, is generally calculated using energy balance equations such as those developed by Penman and Monteith who also coined these terms (McMahon et al. 2013). Inputs must be assessed using a network of air temperature, relative air humidity, radiation and wind observations, which are only gathered at weather stations on the outskirts of the basin. CRU proposes P ET values at a coarse resolution of 0.5°, i.e. 55 km at the equator but here P ET inputs to the GR4J model were obtained from MODIS-derived 1 km datasets. These MOD16 products exploit global weather data set but are combined with MODIS derived land cover types, leaf area index and albedo (Mu et al. 2011) to provide monthly estimates of P ET over 2000-2014. The 180 tiles were downloaded and treated in R to extract the relevant values per pixel and catchment (figure 4.3.8). Values range between 1600 to 2000 mm and decline with altitude as expected and peak over the El Haouareb dam. Values are comparable but up to 15% greater than previous estimates, based on Penman Monteith for Kairouan or for evaporation from the lake (Kingumbi et al. 2007; Alazard et al. 2015). Monthly variations between lakes have a standard deviation of s.d. 5 mm though the use of meteorological inputs (temperature, vapour pressure and solar radiation) at 1° resolution will also restrict the amplitude of spatial resolution. Interannual variations shown in figure 4.3.9 are moderate reaching s.d. 10-15 mm during the warmer months and monthly interannual means for each catchments were used to extend GR4J modelling to 1999. Previous sensitivity studies concluded to reduced influence of P ET variations over time on several models, including the GR4J model (Oudin et al. 2005). In certain instances, reference or potential evapotranspiration can be estimated using pan evaporation values and a P ET pan coefficient. This can be estimated based on relative humidity, air temperature and wind data, but should ideally be empirically derived. Previous studies estimated the pan coefficient around 0.6 including in semi arid environments (Allen et al. 1998; Riou and Chartier 1985; Lacombe 2007). Using Gouazine pan data over 20002008 and the MODIS P ET data (figure 4.3.10) suggests a P ET pan coefficient value around 0.7 but potentially greater during colder months, though large scattering in the results confirm the inherent difficulties in using pan evaporation to assess P ET , notably in drier climates. During warmer months, the pan coefficient reduces progressively as the pan is more affected by heating of its sides, but the correlation with P ET improves as a result of the more homogeneous climatic and vegetation conditions across the catchment.
4.3.4
Estimating infiltration rules
Infiltration on small reservoirs can be studied through in situ measurements, hydrodynamic methods, mass balance methods with tracers and geochemical (isotopes notably) (Grunberger et al. 2004) and water balance methods. Discrete hydrodynamic studies using observations on piezometers downstream of lakes 141
36.0
35.8
latitude
Annual PET (mm) 2000 1900 35.6 1800 1700 1600
35.4
35.2 9.00
9.25
9.50
9.75
10.00
longitude
(a) 1 km P ET values over Merguellil upper catchment
UTM 32N Northing
3960000
Annual PET (mm) 1850 3940000
1800
1750
1700
3920000
3900000
520000
540000
560000
580000
UTM 32N Easting
(b) P ET interpolated per small reservoir sub catchment
Figure 4.3.8: MODIS-derived interannual 2000-2014 P ET 142
Figure 4.3.9: Monthly P ET variations across 58 lakes in 2000-2003 (with mean ±1 s.d.) were available but for limited periods and single reservoirs. On Gouazine 1997-1998 studies notably focussed on groundwater infiltration and propagation, notably on three wells situated downstream of the lake and used for irrigation (Gay 2004). A piezometer situated beside the dam at Hoshas lake was instrumented by the CRDA and again in 2011 as part of this research but vandalism (allegedly due to belief it could be used as a borehole) led to the installation being broken by a sledgehammer and loss of the pressure transducer within a few weeks. Similar problems were reported by other IRD staff on installations in northern Tunisia. Punctual measurements taken by the CRDA and ourselves could inform of the groundwater trends at Hoshas but could not here be used for infiltration studies. 4.3.4.1
Infiltration values for gauged lakes
Infiltration values or relations for 11 lakes in and around the catchment had been derived in previous works by Lacombe (2007) based on water balance methods. Additional data acquired was used to confirm infiltration values on Gouazine and derive new relationships for Guettar, Dekikira and Hoshas. For these three lakes, where infiltration had not been previously estimated, 1262, 651 and 1546 days of depletion were used respectively. Infiltration was assessed during lake depletion phases where certain fluxes can be neglected as detailed in the appendix section 9.2.1. The required inputs for the water balance (stage values, HSV rating curves, evaporation and rainfall values) were assessed as described previously. The resulting relationships between I and stage (in terms of head of water, not stage values as the lake silts up) are illustrated in figure 4.3.12 and table 4.3.4. Infiltration values found for Hoshas were very high, reaching up and over 50 mm/day similarly to values at El Mouidhi (Lacombe 2007). This is coherent with the gravel substrata/soil (which gave the name to the lake) and the rapid decline observed on stage time 143
● ●
● ●
●
● ● ●
● ● ●
● ● ●
Monthly PET (mm)
●
●
●
● ●
● ●
100
● ●
● ●
●
●
●
●
●
●
●
● ● ●
●
●
●
● ●
●
●
●
●
●
●
●
●● ●
●
●
●●
●
●
●
●
●
●●
●
●
●
●
200
●
●
● ●
●
● ● ●
●
● ●●
●
● ●
●
●●
●● ●
●●
●
● ●
● ●
● ●
●
●
●
●
●
●●
0 100
200
300
Monthly Pan evaporation (mm)
(a) P ET = f (Epan )
2.0
●
● ● ●
1.5
● ●
●●
●
●
● ●
● ● ●
● ●
Ct
●
1.0
●
●
●●
● ● ●
●
● ●
● ●
●
● ●
● ● ●
●
●●
●
● ●
● ●
● ● ● ●
● ●
●●
● ●
●
●●
● ●
●
● ●●
●
●
●
●
●
● ● ●
●
●
●
●
●
●
●
● ● ●
●
●
●
●
●
●
●
● ●
●● ● ●
●
●
●● ●
●
●
0.5
0.0 100
200
300
Monthly Pan evaporation (mm)
(b) Transposition coefficient Ct = f (Epan)
Figure 4.3.10: Relationship between pan evaporation and P ET on Gouazine SR 144
series reaching 50-100 mm/day. Infiltration increases with the surface area (in contact with water) and the volume in line with Darcy’s law. Outliers in figure 4.3.11 also highlight the complexity in deriving simple rules for infiltration, where antecedent soil humidity will naturally play a role and infiltration rates at the beginning of a depletion phase will be greater due to subsoil/pores not being saturated. As a result even on lower floods when the head of water is low (e.g. maximum amplitude 1 ha, stage value 400 cm) infiltration rates exceeded 50 mm/day when this was the first flood after over 18 months of drought. Other outliers are notably due to uncertainties at the beginning of an event, where the actual volume increase can not be accurately assessed due to the stage level of the day before being below the ladder but not necessarily 0. Releases on this lake were initially rare and are no longer possible due to the valve being jammed and therefore could be neglected here. On Morra, Fidh Ben Nasseur and the other 7 lakes studied, infiltration was slower, due in part to clayey soils and in line with Darcy’s law could be modelled based on water head in the lake. Relationships can be derived based on table 4.3.4, as in the example equations 4.3.2 below where Zwater is the absolute depth of water not the stage value to account for the silting of the bottom of the lake. IF idhBenN asseur = 3.06 + 0.00314 ú Zwater
(4.3.2)
On 4 lakes infiltration did not follow a linear decline with stage due in part to the more impermeable and sometimes heterogeneous substrata. On Gouazine for instance, infiltration values developed in Lacombe (2007) and completed here with recent observation over 429 depletion days (figure 4.3.11) show that infiltration rate is lower for larger surface areas. Though surprising, this can be explained by the presence of a sandy layer on part of the lake floor which has high infiltration rates (Grunberger et al. 2004). As the flooded area increases, the proportion of the lake in contact with this sandy layer decreases, reducing the overall infiltration rate. At 4.5 m stage which corresponded then to 2.2 ha, their infiltration rate of 250 m3 /day corresponds to a rate of 11 mm/day close to the 13 mm value here. On Guettar and Fidh Ali (figure 4.3.11 and table 4.3.4), the water balance showed that infiltration values were constant, around 10 mm/day and 2.7 mm/day respectively due to differences in the soil substrata. Progressive silting of the lake bed can lead to changes in the infiltration properties, notably making it less impermeable. The recent data on Gouazine and Hoshas do not however indicate a noticeable change in infiltration properties over time and confirm what had been observed by Lacombe (2007). This can be partly explained by the fact that though the bottom of the lake becomes silted and slows infiltration down, the added silt means that the same volume of water covers a greater part of the sides of the lake. These higher parts of the banks which become in contact with water have not been affected by silting and therefore have faster infiltration rates (Lacombe 2007). The silting is also concentrated over small areas of the lake, at the alluvial cone where the wadi flows into the lake. 4.3.4.2
Modelling infiltration on ungauged SR
Mean infiltration values for 14 reservoirs across the range of water depths experienced by each reservoir, ranged from 2 mm to 28 mm/day; mean 8.2 mm, s.d. 7.5mm. Figure 4.3.12 highlights the stark difference between two lakes with very high infiltration due to gravely 145
30
150
●
●
●
●
●
●
●
●
y = −45 + 0.25 ⋅ x, r 2 = 0.418
●
● ●
100
●
Daily infiltration (mm)
20 ●
● ●
●
Daily infiltration (mm)
● ●
● ●
● ●
●
●
● ● ● ●
●
●
●
● ●
10
● ●
50
● ●
●
●
●
●
● ●
● ● ● ●
●
0
0 200
300
400
400
500
500
600
700
800
Stage value (cm)
Stage value (cm)
(a) Hoshas
(b) Gouazine
15 ● ● ● ●
● ● ●
30
●
●
●
●
●
●
●
● ●
●
10
●
● ●
Daily infiltration (mm)
● ●
●
●
●
●
●
●
● ●
●
● ● ●
●
●
● ●
●
● ●
●
● ●
20
●
●
●
●
●
●
● ●
● ●
● ● ● ● ●
●
●
●
●
● ●
●
●
●
● ● ● ●
●
● ●
●
●
●
●
●
●
●
● ● ● ●
●
5
Daily infiltration (mm)
●
● ● ● ●
● ● ●
10
●
● ● ● ● ● ●
●
● ● ● ●
● ●
●
●
● ●
●
0
● ●
0 400
500
600
700
200
Stage value (cm)
300
400
Stage value (cm)
(c) Dekikira
(d) Guettar
Figure 4.3.11: Infiltration values as a function of stage in the lake estimated during depletion periods
146
80
High I
60
Lake ● ●
Abdessadok Brahim Zaher Dekikira
Infiltration (mm)
El Guettar Fidh Ali Fidh Ben Nasseur 40
Gouazine Hadada Hoshas ●
Janet M'richet Morra
●
20
Mouidhi Saadine 2
Low I ● ● ●
●
0 0
2000
4000
Water depth (mm)
Figure 4.3.12: Rate of change of infiltration values with water depth for 14 small reservoirs soils (Hoshas and El Mouidhi), and the other 12 small reservoirs. Infiltration values varied within a much closer range of 2 mm to 13.6 mm, mean 5.6 mm, s.d. 3.6 mm within these 12 lakes. Furthermore the rise in infiltration as the head of water increases is minimal as shown from the slope of the curves below (mean increase in infiltration of 0.7 mm per additional metre of water). Considering the low depth of these reservoirs (1 - 4 m), this equates to a shift in daily infiltration of 2.1 mm. Equation 4.3.3 can then be used to model infiltration on most reservoirs based on water depth which can be determined from remotely sensed surface areas and H-V power relations. On El Mouidhi and Hoshas, mean infiltration values across the range of water depths rose to 24 mm and a steep increase as a function of increasing water stage was observed. On lakes with known sandy soils and rapid infiltration a different infiltration equation 4.3.4, corresponding to an increase in I by 17 mm per m of additional water can be used. In the absence of field knowledge, lithological and geological maps of sufficient resolution may be used to determine the relationship to use. The sensitivity of the model and water availability assessments to these infiltration uncertainties are explored in section 4.4.5. Ilow = 0.00072 ú Zwater + 4.5
(4.3.3)
Ihigh = 0.017 ú Zwater + 2.8
(4.3.4)
147
Table 4.3.4: Infiltration values (mm) for small reservoirs in and around the Merguellil upper catchment. Values except for Dekikira, Gouazine, Hoshas and Guettar were adapted from Lacombe (2007) Lake
Mean infiltration
Infiltration for Zmin (intercept)
Dekikira
2.7
2.70
2.7
0
Abdessadok
3.7
2.55
5.5
0.98
Brahim Zaher
13.6
13.60
13.6
0
Gouazine
9
13.00
7.5
-1.38
Fidh Ali
3.6
3.60
3.6
0
Fidh Ben Nasseur
7.8
3.06
12.5
3.14
Hadada
5.3
2.96
7.9
1.23
Janet
2.1
1.92
2.3
0.3
Morra
2
1.48
2.5
0.53
M’richet
2.4
1.24
3.0
0.58
Mouidhi
20.5
2.01
55.5
9.72
Saadine 2
4.6
3.02
6.8
1.87
Hoshas
28
3.62
77.1
24.50
Guettar
10
10.00
10.0
0
4.3.5
Infiltration Infiltration rise per m for Zmax (slope*1000)
Modelling overflows
During large floods, overflows can occur through the spillway. In extreme cases, when the speed of the flood is too significant to be released through the spillway or by opening the valve, water can also pass over the dam wall as reported in isolated instances by dam guards. This remains exceptional, as the spillway and releases are designed to avoid this occurrence considering the threat it constitutes to the dam wall, to populations and to interests downstream. Exceptional overflows were notably reported during interviews with farmers around Fidh Ben Nasseur and Guettar and over Jannet and Sadine 1 & 2 (Albergel and Rejeb 1997) but these volumes have never been estimated. Here overflows were accounted for in the water balance model by defining maximum capacity of the lakes and a simple ifelse condition preventing the modelled volume to exceed the Vmax . On small reservoirs where additional levelling was undertaken, the maximum capacities were accordingly updated over time. On the other small reservoirs, despite the initial rating curves not being available, the maximum (design) capacity of the lakes was available in the inventories (cf. table7.1.1). This maximum capacity was then updated for each subsequent year based upon the silting laws discussed in section 2.4.3 (and subject to the same uncertainties). The implications of the errors resulting from heterogeneous and non continuous silting are discussed in light of the model outputs in section 4.4.5. The use of an overflow condition notably improved model performance by capping where necessary (on Dekikira for instance) the error from GR4J overestimation, as well as those originating from RS outliers (e.g. Fidh Ben Nasseur). 148
4.3.6
Modelling releases
Releases occur by the dam operator to lower lake levels when a flood may threaten to reach maximum capacity and to protect the infrastructure (dam wall). These releases can be as a result of the amplitude of the current flood or in prevision of another forthcoming event and occur just after the flood in 95% of cases and 5% during the flood (Lacombe 2007). Interviews with the dam operator sought to confirm release strategies and eventual guidelines, which are likely to vary over the seasons to maintain resources for the dry season and preserve sufficient capacity during the flood seasons (García-Ruiz et al. 2011). Minor releases can also occur at other periods to flush out sediments and vegetation from the conduit, though these are increasingly rare. Releases therefore depend on numerous factors whose combination could not be modelled, such as the forecast of other storms, dam operator strategies, CRDA guidelines and instructions, presence of lakes downstream which should or shouldn’t be subject to upstream releases, local demand for water as well as physical problems with the valve which in many cases became either temporarily or irreparably stuck. Interviews conducted highlighted that in many cases, releases had also ceased over time, either due to such technical problems or through pressure from other users who did not understand and appreciate the need for this practice and refused to waste the water in such a way. In the absence of adequate data for all lakes and over time, broad rules based on releases estimated through a water balance on gauged lakes were defined here. On Gouazine releases reached between 50 000 m3 and 80 000 m3 after the significant 2005 and 2006 events. Considering the peak capacity of 237 000 m3 , a conservative rule whereby if modelled volume exceeded 80% of Vmax , it was reduced progressively to reach 80% of Vmax . Data on releases at Fidh Ali confirmed similar behaviour with releases reaching a maximum of 50 000 m3 in 1995. Other releases identified, remained of the order of 1000-5000 m3 /year i.e. the equivalent of 1 to 5 days of heavy pumping (Lacombe 2007). The influence of this rule can notably be observed on the modelling of events of January 2003, January 2006 and February 2005 on Gouazine.
4.3.7
Modelling withdrawals
Withdrawals for water used were estimated based upon rapid field surveys and detailed interviews (both semi directed interviews and quantitative questionnaires) as described in chapter 5. Iboutons (Massuel et al. 2009) temperature sensors placed on water pipes to monitor the decrease in temperature occurring during pumping were tested on two lakes by IRD Tunis staff but results were not conclusive as the difference between air and water temperatures was too minor to be reliably detected. These are better suited to studies on groundwater abstraction where the amplitude in temperature is more marked. Reticence from users was also expressed, which must be understood in light of the previous totalitarian state under President Ben Ali. Interviews revealed that unlike irrigation from boreholes or larger lakes and canals, irrigation around small reservoirs, is excessively heterogeneous in terms of its amplitude and frequency, making extrapolation over time and other reservoirs difficult. Withdrawals are predominantly for watering fruit trees and some market gardening/vegetables but pumping rates vary widely between users, lakes, seasons & years, due to rainfall and water availability, but largely due to numerous socio-economic reasons including access to pumps and pipes, 149
supporting running and maintenance costs as well as alternative income strategies. Based on observations, discussions and theoretical pumping possibilities (i.e. number of pumps), lakes were classified within three broad water use categories, supporting either A) no withdrawals, B) occasional watering for fruit trees and isolated market gardening attempts, C) intensive water use for market gardening and/or fruit trees. Lakes in group B) reported minimal withdrawals for olive trees twice a month during dry months and once a month in the winter though many ceased all watering between November and March if rainfall was sufficient. Water requirements are higher when trees are young (first 2-3 years) and most vulnerable to temporary droughts (CNEA (2006), interviews), and decline as trees become older and hardier. Pumps are thermal/petrol motored pumps of 7-11 horse power with flow rates around 3-5 litres per second (Lacombe 2007). Based on watering 8 hours/day, once per month in winter, twice per month in summer and a maximum value of 5 pumps in operation, this leads to 1200 m3 /month in summer and 600 m3 /month in winter, or 40 m3 /day in summer and 20 m3 /day in winter. This is line with the average value of 10 000 m3 /year identified through water balance of several lakes in the area by Lacombe (2007). Considering infiltration (of 7 mm) and evaporation (10 mm) on a small (2 ha) surface area during the summer months represents over 300 m3 /day this withdrawal scenario is not significant and was neglected in light of the uncertainties in the other fluxes and HSV. Water use was here essentially limited by socio-economic factors, and as result, during years of high water availability, no expansion or modification of strategies was recorded. On numerous lakes, withdrawals had in fact reduced over time (compared to previous interviews in 2005 by Lacombe (2007)), due to the costs of repairs as well as crop failures, shifting either temporarily or permanently into group A). Accordingly, only for group C) (Guettar, O. Daoued or Fadden Boras) lakes did withdrawals represent a significant flux. Greater number of pumps/users were not responsible for the increase but rather intense market gardening or intense arboriculture which can lead to withdrawals during 8 hours, for 3 and more days/week. Two heavy such users on a lake can increase withdrawals to a non negligible 4100 m3 /month on the lake or 130 m3 /day, close to typical daily I and E values. The water balance performed here on Guettar despite several uncertainties notably showed days where an additional 200 m3 were withdrawn and over the summer months 4500 m3 /month were withdrawn. Lacombe (2007) had identified lakes where up to 24 000 m3 /year were withdrawn which if based on intensive irrigation over april-september represents around 4000 m3 /month . O.Daoued and Fadden Boras which had similar number of users and patterns could therefore be modelled based on the same assumption. No constraint to reduce irrigation when levels dropped was modelled as users enjoyed unregulated access to the water resources and displayed opportunistic behaviour, preferring to withdraw frequently while water was available, rather than seeking to manage resources to prolong watering periods.
4.4 4.4.1
Results & discussion Kalman filter performance on daily volumes
Results from applying a Kalman filter based on remotely sensed values to the outputs of a WB+GR4J model are shown in figures 4.4.1, 4.4.2, 4.4.3 and 4.4.4 for the 7 gauged 150
200000 VWB+GR4J
150000 100000 50000 0 200000
VENKF
150000
Volume (m3)
100000 50000 0 200000
VRS
150000 100000 50000 0 200000 150000
Vfield
100000 50000 0 2000
2005
2010
2015
Date
(a) Modelled and observed volume time series for Gouazine 1997-2014
75000 RMSEWB+GR4J
50000
25000
0
75000 RMSEENKF
50000
25000
0
75000 RMSERS
50000
25000
0 2000
2005
2010
2015
(b) RMSE values on daily volumes (m3 )
Figure 4.4.1: Comparing outputs of Ensemble Kalman Filter with field data and other methods for daily volumes (Gouazine, 1997-2014)
151
1e+05 VWB+GR4J
5e+04
0e+00
1e+05 VENKF
Volume (m3)
5e+04
0e+00
1e+05 VRS
5e+04
0e+00
1e+05 Vfield
5e+04
0e+00 2000
2005
2010
2015
Date
(a) Fidh Ali
1e+05 VWB+GR4J
5e+04
0e+00
1e+05 VENKF
Volume (m3)
5e+04
0e+00
1e+05 VRS
5e+04
0e+00
1e+05 Vfield
5e+04
0e+00 2002
2004
2006
2008
2010
2012
2014
Date
(b) Hoshas
Figure 4.4.2: Modelled and observed volume time series for Fidh Ali & Hoshas
152
50000 40000 VWB+GR4J
30000 20000 10000 0 50000 40000
VENKF
30000
Volume (m3)
20000 10000 0 50000 40000
VRS
30000 20000 10000 0 50000 40000
Vfield
30000 20000 10000 0 2000
2005
2010
2015
Date
(a) Fidh Ben Nasseur
200000 VWB+GR4J
150000 100000 50000 0 200000
VENKF
150000
Volume (m3)
100000 50000 0 200000
VRS
150000 100000 50000 0 200000
Vfield
150000 100000 50000 0 2000
2005
2010
2015
Date
(b) Dekikira
Figure 4.4.3: Modelled and observed volume time series for Fidh Ben Nasseur & Dekikira 153
6e+05 VWB+GR4J
4e+05 2e+05 0e+00 6e+05
VENKF
4e+05
Volume (m3)
2e+05 0e+00 6e+05
VRS
4e+05 2e+05 0e+00 6e+05
Vfield
4e+05 2e+05 0e+00 2000
2005
2010
2015
Date
(a) Morra 3e+05
VWB+GR4J
2e+05
1e+05
0e+00 3e+05
2e+05 VENKF
Volume (m3)
1e+05
0e+00 3e+05
2e+05 VRS
1e+05
0e+00 3e+05
2e+05 Vfield
1e+05
0e+00 2004
2006
2008
2010
2012
2014
Date
(b) Guettar
Figure 4.4.4: Modelled and observed volume time series for Morra & Guettar
154
reservoirs where site specific field observations were available. These highlight on Gouazine and Dekikira the ability of the method to represent the amplitude, frequency and timing of several major floods on certain lakes, but in detail the model performance was variable as shown by the R and RMSE values calculated on individual dates for each lake (table 4.4.1). R levels were high on lakes such as Gouazine and Dekikira, but greater errors were observed on other lakes due notably to more preponderant remote sensing uncertainties (Hoshas, Morra) and less reliable hydrometric field data (HSV on Morra and Guettar). These issues are discussed in detail below. Significantly, the combination with Landsat observations led to improved performance over results obtained using a WB+GR4J model developed with specific data (P , I, E, HSV etc.), raising R values except on the two smallest lakes, Hoshas and Fidh Ben Nasseur). The Landsat observations notably modulated the initial forecast, usefully correcting the modelled flood peaks which were both under and overestimated, e.g. flood underestimated in 2003 and overestimated in 2007 on figure 4.4.1a. The decline rates and overall hydrological dynamics of our WB model appeared coherent and well represented across the reservoirs, despite the difference in infiltration at Hoshas and the flood amplitudes (cf. figure 4.4.2). The initial errors on flood peaks have a significant knock on effect on the modelled decline and figure 4.4.5a clearly illustrates the correction from RS which draws WB+GR4J values closer to the 1:1 line, raising the R value from 0.57 to 0.82 on Gouazine and reducing RMSE. RMSE values for each year between field data and simulated data (table 4.4.1), over the whole period reduced by 50% compared to the initial forecast (i.e without the Landsat corrections) on 4 of the lakes (Dekikira, Gouazine, Fidh Ali and Guettar). Excluding the much larger dam (Morra), RMSE values varied between 10 000 m3 and 30 000 m3 (mean 20 900 m3 ). RMSE errors were lower on the two smaller reservoirs but remained very high proportionally to the range of flood values, explaining the low R . On Morra, NRMSE were also significant, despite its large size, again due to remote sensing issues discussed below. The KF nevertheless improved noticeably over the initial forecast as seen in figure 4.4.2. As the lake rarely dries out, GR4J errors from overestimation of a previous event were carried through into the second event, leading to a significant drift, which the KF usefully corrected. On lakes which dry out, the model is in effect reset and previous errors are removed. As shown in figure 4.4.5a and 4.4.5, the fit (and R ) between individual RS observations and field observations can be greater than from the combined method. This is essentially due to being calculated on only 24 (on average) Landsat observations per year, rather than the 365 simulated days per year as with the other two outputs (i.e. VW B+GR4J and VEN KF ), where the effect of individual errors are exacerbated. The KF allowed for a more coherent and accurate flood dynamic than simply interpolating VRS observations as seen for instance on the flood rise in 2003 and the decline in 2005 on figure 4.4.1a. Nevertheless, on lakes where the WB+GR4J performance was low, it also contributed to degrading the remote sensing estimations.
4.4.2
Kalman Filter performance on annual water availability
The method’s performance in assessing annual water availability rather than fine flood dynamics (i.e. individual observations) is shown in figures 4.4.6, 4.4.7, 4.4.8, 4.4.9 and summarised in tables 4.4.2. The ENKF method displayed superior results than on individual values,
155
VWB+GR4J
VENKF
VRS
Daily volume (m3 − sim la e
aa
3e+05
2e+05
1e+05
0e+00 0e+00
1e+05
2e+05
0e+00
1e+05
2e+05
Daily volume (m3 − fiel
0e+00
1e+05
2e+05
aa
(a) Gouazine VWB+GR4J
VENKF
VRS
Daily volume (m3 − sim la e
aa
1e+05
5e+04
0e+00
0
25000
50000
75000
0
25000
50000
Daily volume (m3 − fiel
75000
0
25000
50000
75000
aa
(b) Fidh Ali VWB+GR4J
VENKF
VRS
1000000
Daily volume (m3 − sim la e
aa
750000
500000
250000
0
0e+00
2e+05
4e+05
6e+05 0e+00
2e+05
4e+05
Daily volume (m3 − fiel
6e+05 0e+00
2e+05
4e+05
6e+05
aa
(c) Morra
Figure 4.4.5: Scatterplot between modelled and observed daily volumes 156
Table 4.4.1: Ensemble Kalman Filter performance on daily volumes (a) NSE
NSE Lakes
VW B+GR4J
VEN KF
VRS
Gouazine
0.57
0.82
0.92
Dekikira
0.69
0.76
0.73
Fidh Ali
0.18
0.55
0.70
Fidh Ben Nasseur
0.45
0.45
0.91
Morra
0.11
0.44
0.43
Hoshas
0.38
0.02
0.19
Guettar
0.18
0.48
0.66
(b) RMSE and NRMSE
RMSE (m3 )
NRMSE
Lakes
VW B+GR4J
VEN KF
VRS
VW B+GR4J
VEN KF
VRS
Gouazine
45200
25100
23600
0.44
0.25
0.23
Dekikira
44000
22100
25600
0.39
0.20
0.23
Fidh Ali
39200
20900
19800
0.83
0.44
0.42
Fidh Ben Nasseur
6500
11500
1800
9.43
16.68
2.61
Morra
274300
258800
92800
0.91
0.85
0.31
Hoshas
2700
16600
18900
0.88
5.40
6.14
Guettar
62500
29000
42300
1.14
0.53
0.77
157
150000
Vfield
100000 50000 0
VWB+GR4J
100000 50000 0 150000 100000
VENKF
Mean daily volume (m3) per year
150000
50000 0 150000 VRS
100000 50000 0 2000
2005
2010
2015
(a) Modelled and observed mean daily water volumes per year VWB+GR4J
VENKF
VRS
150000
RMSE on annual mean daily water volume
Mean daily volume (m3) per yea − sim la e
aa
30000
100000
50000
0
20000
10000
0
0
30000
60000
90000
0
30000
60000
90000
Mean daily volume (m3) per yea − fiel
0
30000
60000
aa
90000
VWB+GR4J
VENKF
VRS
(b) Scatterplot between modelled and observed mean daily water(c) RMSE on mean daily water volumes volumes per year per year (m3 )
Figure 4.4.6: Comparing outputs of ENKF with field data and other methods for annual water availability (Gouazine, 1997-2014)
158
Table 4.4.2: Ensemble Kalman Filter performance on annual water availability (a) NSE
NSE Lakes
VW B+GR4J
VEN KF
VRS
Gouazine
0.42
0.92
0.84
Dekikira
0.74
0.89
0.54
Fidh Ali
0.13
0.94
0.82
Fidh Ben Nasseur
0.75
0.86
0.01
Morra
0.81
0.99
0.89
Hoshas
0.40
0.06
0.42
Guettar
0.01
0.65
0.81
(b) RMSE and NRMSE
RMSE (m3 )
Mean daily water volume (m3 )
VW B+GR4J
VEN KF
VRS
Gouazine
42800
36600
11200
18900
Dekikira
59000
46100
13900
35500
Fidh Ali
32200
35100
6900
13100
Fidh Ben Nasseur
1000
2400
700
900
Morra
448900
166700
133900
87900
Hoshas
800
800
6700
5000
Guettar
29000
60601
18539
9293
Lakes
(c) NRMSE
NRMSE
Lakes VW B+GR4J
VEN KF
VRS
Gouazine
0.36
0.11
0.13
Dekikira
0.41
0.12
0.32
Fidh Ali
0.74
0.15
0.27
Fidh Ben Nasseur
3.48
0.94
1.33
Morra
0.55
0.44
0.29
Hoshas
0.26
2.06
1.54
Guettar
1.11
0.34
0.17
159
leading to very high levels of R and improving on the initial WB+GR4J results, except on Hoshas. R notably reaches above 0.86 on 5 of the 7 lakes. Mean RMSE (excluding the larger Morra dam) reduces here to 9700 m3 , due to the annual smoothing of observations. On Hoshas, VW B+GR4J continued to perform better than VEN KF despite underestimating all events, due to the small and short floods experienced which leads to a drastic, incorrect increase in water availability from single remote sensing outliers. Nevertheless, if we consider the order of magnitude of the RMSE and mean water availability on Hoshas we see that the method correctly modelled a mean water availability under 5000 m3 all years (except for 2007). This was the case on all lakes, indicating that these allow for valuable comparisons of availability between lakes. Again, by modelling the decline between two Landsat observation, the KF also improved upon VRS for water availability assessments, except on lakes where the poor initial forecast degraded the KF performance (e.g. Guettar and Morra). Considering the relative ease to assess the absence of water, as clouds only lead to an overestimation of water areas (cf. chapter 3), remote sensing observations also allowed for accurate estimates of periods when the lake is dry or below a given level as shown in figures 4.4.10, including on Hoshas where the number of dry days was less affected by outliers as the lake dries out rapidly. Direct remote sensing observations were more proficient than KF in detecting periods when the lake was dry as the KF suffered from small events artificially modelled by the GR4J, but for higher volumes KF provided a more accurate description of the number of days.
4.4.3
Remote sensing uncertainties
Across all lakes, the presence of remote sensing outliers (overestimations) can notably degrade model outputs but their influence as discussed in chapter 3 is not homogeneous. On smaller lakes, issues of greater errors due to clouds and mixed pixels lead to proportionally greater errors, reducing the KF performance. Though less prevalent on Fidh Ali and Fidh Ben Nasseur, despite their similar sizes, these were observed on Hoshas. Many outliers (15 over 2013-2014) were effectively removed through the filtering of clouds and SLC off but two overestimations were not identified. Stage values were considered reliable as not subject to gaps and nearby observations all confirmed the absence of storm and floods at that period. These were therefore due to undetected cloud presence, notably cirrus clouds or shadows, which on a lake of this surface area has a significant effect. The use of the Vmax condition defined for overflows based on the known maximum capacity helped reduce these errors (by over 100%) by capping them to below 130 000 m3 . Grid cells were here intentionally widely defined (e.g. 5 ha where lakes typically reach 3 ha) as the precise location of areas flooded at very high waters were not known. These errors could notably be improved using high resolution DEM or levelling further lakes. Improvements in the way clouds are detected as well as increased temporal and spatial accuracy will further reduce these errors. Higher spatial resolution will increase precision, while more frequent images will allow outliers to be corrected faster, reducing water availability errors which depend on the lag between subsequent correct observations. Here, cloudy observations above a certain threshold were removed but improvements in the method may be possible by defining specific Kalman gain values for each RS observation to reflect for the presence of clouds at the image level and the associated greater uncertainty over specific observations.
160
60000
Vfield
40000 20000 0
40000
VWB+GR4J
20000 0 60000 40000
VENKF
Mean daily volume (m3) per year
60000
20000 0 60000
VRS
40000 20000 0 2000
2002
2004
(a) Fidh Ali
20000 Vfield
15000 10000 5000 0
VWB+GR4J
15000 10000 5000 0 20000 15000
VENKF
Mean daily volume (m3) per year
20000
10000 5000 0 20000 15000
VRS
10000 5000 0 2000
2005
2010
2015
(b) Hoshas
Figure 4.4.7: Modelled and observed annual water availability for Fidh Ali & Hoshas
161
15000
Vfield
10000 5000 0
VGR4J+WB
10000 5000 0 15000 10000
VENKF
Mean daily volume (m3) per year
15000
5000 0 15000
VRS
10000 5000 0 2000
2005
2010
2015
2010
2015
(a) Fidh Ben Nasseur
150000 100000 Vfield
50000 0
VWB+GR4J
100000 50000 0 150000 100000
VENKF
Mean daily volume (m3) per year
150000
50000 0 150000 100000
VRS
50000 0 2000
2005
(b) Dekikira
Figure 4.4.8: Modelled and observed annual water availability for Fidh Ben Nasseur & Dekikira
162
6e+05
4e+05 Vfield
2e+05
4e+05
VWB+GR4J
2e+05
0e+00 6e+05
4e+05
VENKF
Mean daily volume (m3) per year
0e+00 6e+05
2e+05
0e+00 6e+05
4e+05 VRS
2e+05
0e+00 2000
2005
2010
2015
(a) Morra
1e+05 Vfield
5e+04
0e+00
5e+04
0e+00
1e+05 VENKF
Mean daily volume (m3) per year
VWB+GR4J
1e+05
5e+04
0e+00
1e+05 VRS
5e+04
0e+00 2002.5
2005.0
2007.5
2010.0
2012.5
2015.0
(b) Guettar
Figure 4.4.9: Modelled and observed annual water availability for Morra & Guettar
163
Absolute error on nb of days water volume is inferior
1000
WB+GR4J ENKF RS
500
0 0
50000
Water volume (m3)
100000
150000
(a) Gouazine (over 1999-2014)
Absolute error on nb of days water volume is inferior
750
500
WB+GR4J ENKF RS
250
0 0
25000
50000
75000
Water volume (m3)
100000
(b) Fidh Ali (over 1999-2004)
Figure 4.4.10: Error in the number of days water levels fall below a given volume 164
The numerous outliers observed over 2013-2014 on several lakes highlight that though increased availability (due to the concomitant operation of Landsat 7 and Landsat 8 sensors) may offer greater potential to follow floods, it also increases the likelihoods of errors from undetected clouds. On Morra, despite its large size, KF performance remained low, due partly to remote sensing issues highlighted in chapter 3. The range of amplitudes on this very large lake are minor and contained with the % error of surface area estimates from our MNDWI method. As a result, the R for individual observations is heavily affected, but R on mean annual availability performed very well.
4.4.4
Rainfall-runoff modelling difficulties
Numerous difficulties also occurred as a result of low performance of the WB and specifically the GR4J model. As shown in figures 4.4.11 and 4.4.12, the water balance model reproduced the flood decline dynamics correctly in all lakes, despite their highly different flood amplitudes and decline processes (e.g. high infiltration at Hoshas, withdrawals at Guettar, etc.). Major difficulties occurred due to the GR4J rainfall-runoff modelling which incorrectly reproduced the amplitude of events. R on GR4J indeed only reached values around 0.5-0.6 on Gouazine and Dekikira, but nearer 0.2-0.3 on other lakes, notably on lakes with less extensive and reliable field data (rainfall, stage and rating curves) (table 4.2.1). In semi arid basins, a value superior to 0.5 can be considered acceptable considering the extreme conditions (i.e extended dry or very minimal events and very significant events (100 mm/day). On three out of 13 lakes in the region, Lacombe (2007) had also only succeeded in obtaining a value of 0.5 or more, despite focussing on shorter periods where each lake was followed with its own evaporation pan, rainfall gauge and regularly surveyed. The reasons of these low R values are explored below. 4.4.4.1
Heterogeneous catchment responses
The low performance translated partly difficulties to model the catchment’s response. As discussed in Ogilvie et al. (2016) (cf. chapter 6) at the catchment scale, the intensity but also land cover, antecedent soil humidity or conservation works such as contour benches can significantly influence runoff coefficients. Model parameters such as X1 notably seek to account for the soil humidity and the threshold effect, leading to greater runoff once X1 is saturated. The lumped (i.e. not spatialised) nature of the GR4J model makes accounting for localised changes in catchment behaviour (WSCW, land cover and cropping) difficult however. Model choice guided by limited data availability precluded the selection of a more data intensive semi-distributed and/or physical model capable of accounting for discrete changes over time in land cover and land use. Changing model parameters over time can alternatively indirectly account for this but only at the catchment scale. On Gouazine, where numerous studies discuss the possible reduction in runoff from the development of contour benches on 43% of its catchment area (Nasri 2007), calibrating over 1997-2003 led to a routing store capacity (X3 parameter) 5 times greater than over the whole period, possibly pointing to the greater retention capacity from WSCW. Model performance improved (R rose from 0.62 to 0.67) but only marginally as it remained affected by the other difficulties discussed below. Furthermore constant parameters were required here in order to extend the model to periods and small reservoirs without stage data for calibration. Similar anthropic 165
3e+05 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
Field data
●
Modelled data (airGR+WB)
●
●
●
● ●
●
●
●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●
Volume (m3)
2e+05
1e+05
0e+00
2000
● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●
2005
2010
2015
Date
flow [mm/d]
●
●
20 60 120
01/2002
01/2004
01/2006
01/2008
01/2010
01/2012
01/2014 observed simulated
4 ●
●
0
2
4
01/1998
2
● ● ●
● ● ●● ● ●● ● ● ●● ● ● ● ● ●
0
● ● ●
● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ●● ●● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●●● ●● ●
0
●
● ●
●
●
●
●● ●
● ●
● ● ● ●● ●●
●
2
●
4
6
01/2000
0.005 0.025
●
01/2002
01/2004
01/2006
01/2008
01/2010
Ptot Qobs Qsim
0.002
●
0
●
0.005
●
precip. & flow regime [mm/d]
Q.sim
●
01/2000
10 12 14
6
01/1998
8
8
6
precip. [mm/d]
(a) Observed and modelled daily volumes by WB+GR4J
01/01
01/04
01/07
01/10
31/12
31−days rolling mean
8
Q.obs
(b) Observed and modelled daily flow values by GR4J
Figure 4.4.11: Performance of the WB+GR4J model on Gouazine
166
01/2012
01/2014
Volume (m3)
75000
50000
25000
0
●
40000
● ●
Field data
● ●
Modelled data (airGR+WB)
●
Field data
●
Modelled data (airGR+WB) ●
● ●
30000
● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
2000
2005
2010
● ●
20000
● ● ● ● ● ● ● ● ● ●
●
● ● ●
● ● ● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ●● ● ●● ●● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
●
2000
●
● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ●● ●● ●● ● ●● ● ●● ●● ●● ●● ●● ●● ●● ●● ● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ●● ●
● ● ● ●
● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ●● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●●●●●●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●● ● ● ●●●●●● ●
2005
2010
150000
Volume (m3)
Volume (m3)
2015
100000
50000
Modelled data (airGR+WB)
● ● ●● ● ● ● ● ●
2015
● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
0
2000
2005
2010
(d) Dekikira
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●
Field data
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
3e+05
2e+05
●
m
4e+05
Modelled data (airGR+WB)
m
Volume (m3)
5e+05
2015
Date
●
6e+05
●
●
● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ●● ● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
(c) Fidh Ben Nasseur
●
●
● ●
2010
Date
●
●
● ●
●
Field data
200000
Modelled data (airGR+WB)
2000
● ● ● ●
●
(b) Hoshas
Field data
0
●
●
Date
●
10000
●
●
2005
(a) Fidh Ali
20000
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
10000
2015
●
●
Date
30000
●
●
Volume (m3)
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●
●
●●
● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
1e+05 2000
2005
2010
2015
Date
(e) Morra
( ) Guettar
F gure 4 4 12 Observed and mode ed da y vo umes by WB+GR4J for 6 akes
167
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●
changes in catchments were noted, including a sand extraction quarry reported to have modified and banked the river channel upstream of Guettar, and a small but non negligible (around 5 000 m3 ) reservoir built upstream of Hoshas which could theoretically reduce runoff on the smaller events (2.5 mm) by around a third (though the operation of the low outlet valve is unknown). 4.4.4.2
Rainfall measurements and interpolation
Part of the errors arise from the rainfall observation network which as discussed in sections 2.3.2 and 4.3.1 can fail to capture an event or anyway the highest intensities, due to the significant spatial variability and very localised nature (spatial extent) of rainfall events across Mediterranean catchments. When artificially removing the upstream station at Gouazine, mean error on interpolated rainfall reached 6.2 mm (20.3%), underestimating rainfall on 44 out of 53 events. Errors are calculated against rainfall interpolated with all stations in the catchment, which is naturally also subject to uncertainties, considering the difficulties in assessing rainfall over a large area. Despite the gently undulating topography in this catchment, these results highlight the importance of an upstream station to correctly detect the greater orographic rainfall intensities upstream and provide an indication of possible sources of errors on the other lakes which do not benefit from upstream stations. When removing all stations in the catchment, as is the case in ungauged catchments, errors on interpolated rainfall at lakes reached 60-70% on events over 20 mm rainfall (section 4.3.1). On catchment rainfall, the same approach revealed lower but significant uncertainties reaching 20-50% on average (table 4.4.3 and figure 4.4.13). On individual events, the errors can be greater with events being totally undetected (4 out of 35 at Morra, 1 out of 21 at Guettar, 0 out of 53 at Gouazine). The amplitude of errors will largely depend on the presence, proximity and reliability of gauging stations situated around the upstream reaches. The presence of nearby rainfall stations at 1 km and 3 km and at similar altitudes contributed to reducing the error on Gouazine and Guettar. On the contrary at Morra the absence of a rainfall station for over 6 km and outside this region of marked topography (dam is at 720 m altitude), led to significant underestimation of rainfall, with 32 out of 35 events being underestimated. KED interpolation at Morra did not help reduce errors. At this scale of analysis, the required density is therefore significantly lower than the 30 km suggested by Gandin (1970) in Baccour et al. (2012). When applying rainfall data interpolated for Gouazine without the upstream station, the performance of the GR4J model decreased only marginally (R =0.55), however the WB+GR4J R declined from 0.58 to 0.24. The rainfall underestimation from the absence of an upstream station forces the model to increase runoff coefficients in order to restitute the measured runoff. This leads to overestimation on other events which were more accurately measured due to their larger spatial extent. The knock on effect of these overestimations on decline values explains the lower WB+GR4J performance. The KF continues to improve performance and correct for these errors but NSE on annual availability reduces from 0.92 to 0.84, and RMSE values for the daily observations increase by over 20%. These result highlight the order of magnitude of uncertainties from limited rainfall station density which are further exacerbated by inherent measurement errors in rainfall. These notably stem from capture (up to 80% error due to winds, inclined rainfall) (Musy and
168
Table 4.4.3: Errors when interpolating rainfall events over 20 mm without gauges in the catchments Mean error (mm)
Mean error (%)
s.d. error
Range
Gouazine IDW
6.0
19.6
4.1 mm
0.3 - 17 mm
Guettar IDW
7.6
25.7
6.2 mm
1 - 21.6 mm
Morra IDW
15.8
49.4
10.4 mm
1.4 - 41.1 mm
Morra KED
16.2
50.7
10.2 mm
0.4 - 41.1 mm
Guettar
Morra
Gouazine
75 ●
IDW without station in catchment
●
●
●
50 ● ●
● ● ● ●
●
●
●
●
●
●
●
● ●
●
●
● ● ●
●
●●
●
● ●
●●
●
● ● ●● ● ● ●●
●
●
25
●● ●
●
● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ●
●
● ●● ●● ●
● ● ● ●
● ● ●
● ●● ●●● ● ●
● ● ●
0
● ● ●
●
0
25
50
75
0
25
●
50
75
0
25
50
75
IDW with station in catchment
Figure 4.4.13: Rainfall interpolation over catchment with and without station in catchment
169
Higy 2004) but also equipment malfunction, observed on numerous occasions, where dirt causes pluviographs to slow down, or empty before its tips sufficiently to close the circuit and create a recorded impulse. Vegetation (pollen, leaves) or insects also clogged the funnel of the pluviograph despite the presence of a filter, leading to significant errors (100% order) which could not systematically be corrected with observer readings. During intense rainfall, the tipping of the buckets may also be too slow and losses can occur. Observer readings are also subject to human error, in addition to their systematic bias (reading times & accuracy, rounding up or down). Homogeneity tests performed in chapter 6 only identify significant long term shifts as caused by a change in equipment or location but not punctual errors. Interpolation helped tamper some of these errors. 4.4.4.3
Other data difficulties
Part of the GR4J modelling errors also arose due to an artificial lag generated from disparities in the way data was acquired and reported. Rainfall was indeed sometimes assigned to the morning of the observation rather than the day before. Corrections were done on individual lakes but could not be identified over 15 years of data and over the 50 stations. Errors were also harder to distinguish when events occurred over 2-3 successive days. Likewise, stage readings taken at 7 AM did not systematically identify the flood peak as flood levels may continue to rise or on the contrary may have already declined through high infiltration (e.g. Hoshas decline of 50 cm within 7 hours) or manual releases. In the absence of instantaneous time series, these could not be identified leading to calibration errors. Likewise, events occurring over successive days suffered from the arbitrary daily division of P and Z data. Lacombe (2007) showed that 93% of events (over 10 mm) were separated by 24 hours and that only 0.3 events were still active before and after 7 AM. These statistics however disguise some of the difficulties as in semi arid contexts the exceptional events (i.e. the other 7%) are determinant on local hydrology. On Fidh Ali (over 29.09.2001 to 1.10.2001) the very large event was therefore poorly modelled as the substantial rains scattered over 3 successive days, led to very high runoff on the third day only, due to saturated soils and delayed subsurface flows, causing calibration difficulties. Calibration errors also arise from absent stage observations before the first storms, which may not always be 0 as ladders and transducers are installed near the edge for logistical reasons of installation and regular access to instruments and therefore do not always stretch to the lowest levels. These should more reliably be identified and distinguished from NA and 0 observations. Where possible, short gaps were interpolated based upon previous values and estimated depletion fluxes (evaporation, infiltration, etc.). Alternatively these were assumed to be 0 as excluding these events, led to a reduction of NSE from 0.62 to 0.46. Remote sensing with frequent and precise enough images, could in time help fill such gaps in time series. Insufficient levelling may also explain certain difficulties, as for instance on Fidh Ali, rating curves were no longer updated after 1999, meaning the estimated volume in figure 4.4.12 may be overestimated as a result of greater silting which occurred after the large 2001 event. The lake had already lost much of its capacity (37 000 m3 over 9 years) pointing to high erosivity of its catchment, but despite recent silting studies on this lake (e.g. Hajji et al. (2015)), no additional survey could confirm this, forcing us to use average silting laws.
170
Table 4.4.4: Kalman Filter performance on mean water availability when degrading model inputs and parameters RMSE (m3 /day)
Lakes
NRMSE
VW B+GR4J
VEN KF
VRS
Gouazine
87600
12100
Dekikira
49800
Fidh Ali Hoshas
4.4.5
VW B+GR4J
VEN KF
VRS
15800
0.79
0.11
0.14
17000
28100
0.44
0.15
0.25
23000
16900
17900
0.45
0.33
0.35
28000
1600
1100
8.59
0.49
0.34
Monitoring ungauged basins with remote sensing observations
The performance of the KF after inputs (I, HSV & silting) and parameters (GR4J) were degraded and estimated as they would be over ungauged lakes is shown in figures 4.4.14 and 4.4.15 for four lakes. On Gouazine, where the change in GR4J parameters and HSV were marginal (cf. section 2.4) was severely impacted by the reduced infiltration from the average rule (section 4.3.4) which prevented the model from reproducing the emptying of the lakes between successive events, leading to a rising drift in volumes and errors. The Vmax condition helped avoid errors increasing further but not reduce it. On Hoshas the model was sensitive to both the change in the GR4J parameters and infiltration rule, leading to significant overestimation of inflow and insufficiently rapid infiltration. Using the high infiltration rule developed led to a strong improvement in the forecast and an associated 50% decrease in RMSE confirming the sensitivity of the model to this. I nevertheless remained too low and could not correct for the overestimated inflow volumes. On Dekikira, the significant differences in the HSV model were overshadowed by the change in GR4J parameters which led to vast underestimation in the inflow. Using estimations on certain inputs and parameters clearly degraded the performance of the WB+GR4J inital forecast with RMSE errors rising drastically (over 100%). The Kalman filter using regular Landsat observations continued to provide valuable corrections and RMSE errors on both individual observations and annual availability remained remarkably similar to those with the site specific model on Gouazine and Hoshas. On Fidh Ali and Dekikira, their rise was contained to around 50%. As the quality of the water balance model degrades, the benefit of assimilation with remote sensing observations naturally increases. However the benefit of VEN KF over simply exploiting interpolated VRS values also declines, due to the initial forecast becoming so uncertain. As before, remote sensing observations alone provide superior performance on individual observations but significantly on mean water availability assessment, their RMSE reduce from being around 50% greater than VEN KF to merely 10% greater (table 4.4.4). Where the initial prediction is particularly inadequate as on Hoshas, simple interpolation of the VRS observations even performed better than VEN KF . This is coherent and confirms that the use of KF is valuable when there is sufficient confidence in the model, i.e. for gauged lakes, but as uncertainties (on I, HSV, silting and GR4J parameters) rise, KF errors can become greater than the VRS errors. Remarkably VRS performed well when using the
171
200000 VWB+GR4J
150000 100000 50000 0 200000
VENKF
150000
Volume (m3)
100000 50000 0 200000
VRS
150000 100000 50000 0 200000 150000
Vfield
100000 50000 0 2000
2005
2010
2015
Date VWB+GR4J
VENKF
VRS
Daily volume (m3 − sim la e
aa
200000
150000
100000
50000
0 0e+00
1e+05
2e+05
0e+00
1e+05
2e+05
Daily volume (m3 − fiel
0e+00
1e+05
2e+05
aa
Figure 4.4.14: Modelled daily volumes when degrading knowledge of the HSV, GR4J parameters and infiltration values on Gouazine lake
172
200000
VWB+GR4J
VENKF
VRS
150000 VWB+GR4J
100000 50000 0
150000
200000
Volume (m3)
50000 0 200000 150000 VRS
100000 50000
Daily volume (m3 − sim la e
VENKF
100000
aa
150000
100000
50000 0 200000 150000 Vfield
100000 50000
0
0 2000
2005
2010
0
2015
50000
100000
150000
0
50000
100000
150000
Daily volume (m3 − fiel
Date
0
50000
100000
150000
aa
(a) Dekikira 125000
VWB+GR4J
100000
VENKF
VRS
100000 VWB+GR4J
75000 50000 25000 0 125000
75000
Volume (m3)
50000 25000 0 125000 100000
VRS
75000 50000 25000
Daily volume (m3 − sim la e
VENKF
75000
aa
100000
50000
25000
0 125000 100000 Vfield
75000 50000 25000
0
0 2000
2005
2010
0
2015
25000
50000
75000
0
25000
50000
75000
Daily volume (m3 − fiel
Date
0
25000
50000
75000
aa
(b) Fidh Ali VWB+GR4J
VENKF
VRS
VWB+GR4J
1e+05
5e+04
0e+00 1e+05
Volume (m3)
0e+00
1e+05 VRS
5e+04
Daily volume (m3 − sim la e
VENKF
5e+04
aa
1e+05
5e+04
0e+00
1e+05 Vfield
5e+04
0e+00 0e+00 2000
2005
2010
0
2015
10000
20000
30000
40000
0
10000
20000
30000
Daily volume (m3 − fiel
Date
40000
0
10000
20000
30000
40000
aa
(c) Hoshas
Figure 4.4.15: Modelled daily volumes when degrading knowledge of the HSV, GR4J parameters and infiltration values on 3 SR
173
60000 150000 40000 Vfield
Vfield
100000
20000
50000 0
0 60000
50000 0
150000
50000
20000
0 60000
40000
VENKF
VENKF
100000
Mean daily volume (m3) per year
100000
40000
VWB+GR4J
VWB+GR4J
Mean daily volume (m3) per year
150000
20000
0
0 60000
150000 40000 VRS
VRS
100000
20000
50000 0
0 2000
2005
2010
2015
2000
(a) Gouazine
2010
2015
(b) Fidh Ali 60000
120000 90000
40000 Vfield
Vfield
60000
20000
30000 0 120000
0 60000
30000 0 120000 90000
30000 0 120000
20000
0 60000
40000 VENKF
VENKF
60000
40000
VWB+GR4J
VWB+GR4J
60000
Mean daily volume (m3) per year
90000
Mean daily volume (m3) per year
2005
20000
0 60000
90000
40000 VRS
VRS
60000
20000
30000 0
0 2000
2005
2010
2015
2000
(c) Dekikira
2005
2010
2015
(d) Hoshas
Figure 4.4.16: Modelled mean annual volumes when degrading the HSV, GR4J parameters and infiltration standard HSV relation and silting laws developed in 2.4, confirming their suitability. The need to account for silting was confirmed as when removing this, mean water availability RMSE rose on both Fidh Ali and Gouazine by a further 50%. As with the site specific models, when accounting for the range of amplitudes of each lake, NRMSE errors remained low and below 0.5 in all four lakes when using the KF method or simply interpolating VRS (table 4.4.4). These can therefore be used to provide valuable insights in the water availability patterns of lakes (figure 4.4.16) and allow comparisons between lakes. Considering the absence of tangible benefit from using the ENKF on ungauged basins where uncertainties become significant, simple interpolation of Landsat observations converted to volumes using our standard HSV and accounting for silting were used to estimate water availability across all reservoirs. Figure 4.4.17 illustrates the first insights provided by the approach into hydrological variations across all ungauged small reservoirs over 19992014. This information is analysed in chapter 5 to understand water availability patterns and hydrological constraints across SR in the Merguellil catchment.
174
1000000 750000 500000 250000 0
Trozza_sud
Hoshas
Ain_Faouar
Fidh_Mbarek
Maiz
Mazil
Fidh_Ali
Fidh_BenNasseur
Bouksab
Gbatis
Guettar
Marrouki
Mahbes
En Mel
Sidi_Sofiane
Smili_1
Smili_2
Kraroub
Garia_3
Garia_2
Garia_S
Raouess
Marrouki_2
Morra
Habsa_B
Habsa_A
Fidh_Zitoune
El_Haffar
Bouchaha_A
Bouchaha_B
Daoued_1
Fadden_Boras
Dhahbi
Skhira_5
Skhira_1
Skhira_2
Skhira_3
Skhira_4
Skhira_C
Skhira_B
Skhira_A
Salem_Thabet
Mdinia
Gouazine
Ben_Houria
Abdessadok
Brahim_Zaher
Abda
Hammam
Chauoba_Hamra
Dekikira
1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0
Volume (m3)
1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 1000000 750000 500000 250000 0 2000
2005
2010
2015 2000
2005
2010
2015 2000
2005
2010
2015
Year
Figure 4.4.17: Water volumes estimated from Landsat observations across 51 lakes - using common scale
175
Trozza_sud
Hoshas
2000
2005
2010
2015
0 2000
2005
2010
2015
2000
Mazil
2000
2005
2010
2015
2005
2010
2015
2005
2010
2015
2000
2010
2015
2000
2005
2010
2015
Smili_2
0 2000
2005
2000
2010
2015
2005
2010
2015
Garia_S
2000
2005
2010
2015
2005
2010
2015
2005
2010
2015
2005
2010
2015
2000
Bouchaha_A
2005
2010
2015
2000
2005
2010
2015
2000
Dhahbi
2005
2010
2015
2000
2005
2010
2015
2005
2010
2015
Skhira_4
2010
2015
2000
Skhira_A
2005
2010
2015
2005
2010
2015
2000
Ben_Houria
2005
2010
2000
2015
2000
2005
2010
2015
Hammam
2010
2015
2005
2010
2015
2005
2010
2015
2010
2015
2005
2010
2000
2015
2005
2010
2015
El_Haffar
30000 20000 10000 0 2000
2005
2010
2015
Fadden_Boras
2005
2010
2015
2000
2005
2010
2015
Skhira_2
2005
2010
2015
2000
2005
2010
2015
Skhira_B
2005
2010
2015
2000
2005
2010
2015
Gouazine
2005
2010
2015
2000
2005
2010
2015
2010
2015
Abda 20000 0
2000
2005
2010
2015
2000
2005
Dekikira 200000 100000 0
100000 50000 0 2000
2000
Brahim_Zaher
Chauoba_Hamra
4e+05 2e+05 0e+00
2005
0 2005
2015
Morra
50000 2000
2010
200000 100000 0 2000
50000 0
0
2005
100000 50000 0
Abdessadok
10000
2015
Mdinia
0 2000
2010
1000000 500000 0
40000
0
2000
Skhira_C
Salem_Thabet
40000
2005
40000 20000 0
0 2005
2015
20000 0 2000
100000 2000
2010
Garia_2
0 2000
Skhira_3
80000 40000 0
2015
100000
0 2000
2010
Skhira_1
40000
2005
80000 40000 0
Skhira_5
20000 10000 0
2005
40000 20000 0
0 2000
2000
Daoued_1
20000
0
2015
0e+00
Bouchaha_B
10000
2010
Smili_1
Fidh_Zitoune
40000 20000 0
40000 20000 0 2000
2005
4e+05 2000
Habsa_A
0
2015
40000 20000 0 2000
Habsa_B 20000
2010
Marrouki_2
0 2000
2015
10000 0 2000
10000
0
2010
Marrouki
Garia_3
Raouess
100000
2005
2000 150000 100000 50000 0
20000 10000 0 2000
2005
1e+05 5e+04 0e+00
Kraroub
1000000 500000 0
20000
2015
Sidi_Sofiane
40000 20000 0
8e+05 4e+05 0e+00 2000
2010
Guettar
En Mel
0
2005
0 2005
2000
Fidh_BenNasseur
100000 2000
Mahbes 100000
2015 40000 20000 0
Gbatis 100000 50000 0
2000
2010
1e+05 5e+04 0e+00 2000
Bouksab 40000 20000 0
2005
Fidh_Ali
100000 50000 0
4e+05 2e+05 0e+00
Fidh_Mbarek 40000 20000 0
20000
Maiz
Volume (m3)
Ain_Faouar
100000 50000 0
40000 20000 0
2000
2005
2010
2015
2000
2005
2010
2015
Year
Figure 4.4.18: Water volumes estimated from Landsat observations across 51 lakes - using free scale
176
4.5
Conclusions
After assessing the feasibility of using Landsat observations to monitor surface areas across small reservoirs and developing an appropriate methodology, this chapter sought to identify how this additional information could improve the monitoring of their water resources. On gauged lakes, this information coupled with an Ensemble Kalman Filter to a water balance model showed its potential to correct and improve upon previous modelling attempts across small reservoirs. Remote sensing observations provided vital corrections to the flood amplitudes incorrectly estimated by the GR4J model which suffered notably from rainfall accuracy issues, while the site specific rules on depletion fluxes (infiltration, withdrawals, etc.) led to an accurate modelling of the flood decline. Overall performance reached high levels of R and reduced RMSE errors (down to 10 000 m3 ) notably when considering annual water availability as required here. Uncertainties due to remote sensing evoked in chapter 3, naturally have knock on effects, especially on smaller reservoirs or where the range of fluctuations is smaller and potentially contained within the precision of the method. These were expected considering the demanding objective of using medium resolution images (30 m) on small objects with limited field data, and highlight the potential which will grow as new sources of higher temporal and spatial resolution satellite imagery, as well as improved cloud corrections and detection methods, become available. On ungauged lakes, developing a suitable water balance model was constrained by numerous uncertainties, generated by inaccurate rainfall observations and the difficulty of extrapolating ratings curves and rules on silting and infiltration, considering the heterogeneity in the catchment. Greater characterisation of the catchment may allow for more detailed modelling attempts, but on lakes where the reservoir’s behaviour falls outside the commonly observed behaviour the rise in uncertainties degrades significantly the water balance model. As a result, these uncertainties exceed those from the remote sensing and it becomes preferable to rely exclusively on interpolated Landsat surface area observations to assess volumes over time. These maintained acceptable RMSE values which will reduce as sensors provide more frequent and accurate observations with which to monitor the flood. In parallel, uncertainties on the rating curve and silting remain high and should be further reduced by additional levelling to extend the range of lakes and account for silting over time. New methods including drone based sensors open up increased possibilities to acquire at lower costs (time, equipment) sufficient topographic detail to render surface volumes rating curves (Massuel et al. 2014a). These also help reduce remote sensing uncertainties by appropriately capping (defining maximum) surface area estimates. These results are then used in the following chapter to explore water availability patterns across all lakes over 15 years, and shed light on hydrological constraints. Alternate modelling approaches may also be investigated to characterise land cover of individual catchments over time and better account for the physical properties affecting runoff processes. Models with a longer time scale (e.g. 8 days to coincide with Landsat observations) in order to reduce the difficulties in modelling events which spread over successive days or alternatively an hourly model where field data allows may also help improve performance. The Kalman filter approach may also be varied to seek to correct not individual observations, but rather model variables or parameters. This could indeed be used to seek to rectify flood inflow values, changing input rainfall values or model parameters
177
as necessary, notably on ungauged lakes and when more frequent satellite observations (i.e. closer to flood peaks) are available. Similarly, over decline phases, where sufficient confidence in infiltration and evaporation exists, the remote sensing observations could be used to identify withdrawal rates. The KF method may also be enhanced by fine tuning (Moradkhani et al. 2005) the covariances to compose with both sources of uncertainty and provide greater confidence to remote sensing observations (on larger lakes, clear days, etc.) than field data based on additional criteria (lake size, cloud presence across image, etc.).
178
Part III
Understanding small reservoir hydro-social systems
179
Chapter 5
Lake benefits and hydro-sociological drivers of water use 5.1 5.1.1
Methods Water availability characterisation
The method developed using remotely sensed water surface areas in chapters 3 and 4 was applied to investigate water availability over 15 years across 48 lakes (shown in figure 5.2.18). Results sought to explore where water availability is sufficient (i.e. their potential) to support agricultural practices and whether the amplitude and reliability of water resources had a determining influence over the development of irrigated water uses. Lakes cells defined in 2.2.4 were used to extract and compile in R the surface area for each lake in the catchment over all 546 treated Landsat MNDWI images spanning August 1999 to January 2015. Surface areas were then converted to volumes using the relationships detailed in 2.4, where the B and — model parameters were incremented over time to account for silting, based on known maximum capacity and construction date. Time series were spline interpolated and gap filled using the TSGF extension (Forkel et al. 2013) in order to allow water availability statistics to be calculated over time. A secondary cap based on maximum capacity of lakes was used to reduce remote sensing outliers. Lakes built recently in the Merguellil catchment (e.g. Mdinia 2 in 2012) as well as as minor lakes (e.g. Bouksab 2) for which maximum capacity data was not available were excluded here from the water availability analysis. 5.1.1.1
Annual water availability patterns and demand
Annual water availability (here defined as the mean water availability per day) was first calculated to identify the order of magnitude of available resources and their variability over 1999-2014. To facilitate comparison between lakes, and account for the recent construction of 5 lakes (El Hammam, El Marrouki 2, Mazil, Garia S, Mdinia), the variability in interannual water availability was also assessed over the common period 2007-2014, as averaging over different period would bias results considering the significant interannual rainfall variability 180
(cf. Ogilvie et al. (2016) and chapter 6). Though construction was completed in 2006 for Mdinia, data from 2007 was used to allow time for the lake to fill after a major flood, as due to its size and unlike many smaller lakes, the initial months were not representative of its subsequent water levels. Daoued 3 & Mdinia 2 built in 2012 were not studied considering the much shorter time series available. Flooding in these cells were clearly detected after 2012, pointing to the method’s ability to identify the year lakes when lakes were commissioned. Based upon the information provided during interviews, withdrawal patterns including average pumping duration and frequencies during the dry and rainy seasons for lakes were calculated (cf. section 4.3.7). These theoretical watering needs were then used to derive monthly pumping rate scenarios for lakes, classed in three categories according to the water use typology derived here. Higher resolution in the withdrawals assessment per lake was not relevant considering the high uncertainties observed in the number of pumps actually in operation each year and on each lake. These scenarios were then compared with the estimated water availability. Availability and demand were assessed over the 6 months period between April and September where reduced rainfall and increased temperature and potential evapotranspiration lead to greater water demand. Assessing availability over 6 months also helped contain the influence of remote sensing uncertainties and clouds in individual observations. 5.1.1.2
Water availability index
Considering water uses depend on the amplitude but also the timing and duration of the flood, the number of days that water levels exceed a given volume were calculated for each lake over the dry season and the whole year. This index defined the number of days water levels remained under 5000 m3 was then used to account for the strong disparity in the number of days lakes were flooded and highlight the irrigation potential. The 5000 m3 value reflected sufficient water availability to develop intensive irrigated activities by two users (estimated around 4500 m3 /month in section 4.3.7) or supplementary irrigation by up to 20 farmers. Significantly it also complies based on the height-surface-volume (HSV) relations used with a minimal water depth of 1 m, required according to interviews by users to operate pumps without silt from the lake floor bed clogging the pumps. In terms of surface area, this represents a minimum surface area of 0.5 ha, and therefore complied with technical requirements of being inferior to the size of the smallest lakes (e.g. Bouchaha A). Aggregating a minimum of 6 Landsat pixels also reduced the presence of single outliers. These statistics were calculated over the whole year and over the 6 driest months (AprilSeptember) when watering needs are greatest. The 48 lakes were then categorised based on the length of time they succeeded or failed to meet the required water volume. Considering the uncertainties in modelling silting over time, which led certain lakes to be considered completely silted with time, results were compared with calculations run using the initial volume without accounting for silting. This represents a best case scenario and seeks to understand water availability conditions experienced by new lakes or if dredging or raising dam walls were carried out to maintain initial capacity. In the same way as the surface area assessments, it also provides in effect a measure of the flooding duration, which can provide insights into their potential, irrespective of flooding volumes and their uncertainties.
181
Method Initial Finer scale
Basin scale
typology
Rapid surveys (56 SR)
Water availability assessments (48 SR)
Representative sample Quantitative questionnaires (22 SR) Micro scale Semi directed interviews (4 SR)
Water availability typology
Agricultural water use typology
Water availability driver
Outputs
Socio-economic, i ti ti al i
Figure 5.1.1: Schematic representation of methodology used to determine water uses and associated hydro-sociological drivers
5.1.2
Water use and users characterisation
As described in section 1.3, a multi stage approach was used to identify broad water use characteristics at the catchment scale, and define with greater detail water uses and profile water users on a selection of lakes (figure 5.1.1). These sought to exploit the strengths and weaknesses of different approaches, composing notably with greater speed to investigate uses at the basin scale, and greater detail on specific illustrative sites. 5.1.2.1
Rapid water use surveys on 56 lakes
An inventory of pumps and water uses around each reservoir was established during field visits over 2011-2013. Surveys were undertaken between April and June, at the beginning of the dry season, when lakes are expected to have been filled by rainfall and when irrigated agricultural activities begin due to reduced precipitation and increased potential evapotranspiration. Pumps are also often removed during winter periods, due to the reduced watering needs and the risk of floods which could submerge and damage the pumps. 55 reservoirs in the Merguellil catchment and the Gouazine lake in the immediate vicinity (and studied extensively in part II) were surveyed. Complementing the inventory on reservoir characteristics (cf. section 2.2.2), these observations focussed on the agricultural water uses supported by the lake as well as general characteristics affecting its use (silting, damages). These noted the presence and number of pumps, watering pipes (bergater) and cisterns which allow watering from the lake. They also qualitatively listed the presence of nearby habitations, fruit trees, market gardening, rainfed crops, grazing and watering livestock nearby. Flood recession cropping on the banks of the lake to benefit from increased and prolonged soil moisture was also recorded, as were 182
nearby wells under the influence of lakes, considering their potential benefit for groundwater recharge (Selmi and Zekri 1995). Furthermore all water supply sources (spring, boreholes, wadi) were inventoried to highlight the role of these lakes in supplementary irrigation. Forms are appended in section 10.1. Information collected through these water use surveys on 56 reservoirs was compared with previous inventories undertaken in 1999 and 2005, at similar periods of the year. In 1999, interviews focussed on 43 reservoirs while in 2005 extensive agricultural use surveys were carried out on 25 lakes in the region. Additional local studies (Selmi et al. 2001; Selmi and Zekri 1995) and reports including original project documents and evaluation reports (e.g. CNEA 2006) were used, notably to identify the original objectives of lakes and the wider assistance provided to riparian farmers. 5.1.2.2
Agricultural questionnaires on 22 lakes
22 lakes were selected out of the 56 surveyed for further analysis of agricultural practices and water uses. The sample was chosen to include a cross section of lake size, geographic situation, water availability and water use characteristics. Incorporating physical and social parameters throughout the method is essential to capture the diversity of uses, users and hydro-sociological constraints, considering the consequences inappropriate analysis and evaluations frameworks can yield (Venot and Cecchi 2011; Selmi and Talineau 1994). Water availability assessments, defined from the Landsat data, identified lakes with negligible, unreliable and good hydrological potential. Rapid surveys distinguished lakes with and without motor pumps, and lakes potentially supporting the recharge of nearby wells. The latter were distinguished considering the different dynamics, hydrologically and in terms of water access, government support, involvement in water use association, prior irrigation experience, etc. (Selmi and Talineau 1994). Lakes with no identified exploitation were included here to specifically explore the reasons behind these. These initial categories helped guide the selection of 22 reservoirs on which to develop quantitative agricultural interviews. In order to compare over time agricultural uses and water management with data collected in 2005-2006, where possible, priority to lakes already studied in 2005 was given. 48 farmers were interviewed with the assistance of a translator, with previous work experience in the area and with a good understanding of the lakes, farmers and sociopolitical context. Quantitative questionnaires covered the following topics: identification of the household and livelihood strategies, farm characteristics, agricultural practices across each plot (crop types, surface areas, yields, watering regimes, livestock), water access and water uses, lake withdrawals, water management, and future perspectives. Questionnaires were developed based upon previous agronomic questionnaires (Feuillette et al. 2003) adapted for the small exploitations around small reservoirs. Additional questions on land rights and government assistance to draw out potential constraints influencing water use were also included based upon the insights gained in the semi directed interviews. The replies to the questionnaires were treated statistically to describe the variety of agricultural practices and their evolution over time based upon the 2005 questionnaires. Forms are provided in appendix section 10.2. All interviewees were informed of the objectives of the study, of the use of their answers, of the absence of any obligation to take part and of confidentiality issues, as agreed to by the King’s College Research Ethics Committee. Both men and women were
183
approached, only children, unsound and very elderly people were not targeted. Considering the limited number of people living around the reservoirs no formal selection of interviewees was necessary. 5.1.2.3
Ethnographic interviews on 4 lakes
In parallel, 25 semi-directed interviews were undertaken (over 2011-2013) on four reservoirs (selected out of the 22 lakes as presented in section 1.3. These sought to complement the information gained through questionnaires, which can in certain cases create a formal atmosphere, and lead to shortened replies and constrained discussion (Beaud 1996). The technique, borrowed from ethnography, seeks to build greater trust with the interviewee and allow for more personal, confidential or sensitive information to be disclosed, relating here to conflicts or political and economic constraints. The trust gained and position as an external observer can then lead to interviewees revealing information, in a similar way to a confession (Beaud 1996). In parallel, the unbridled discussion encourages and enables the interviewee to delve into additional details and raise complementary issues. This method therefore reduces the risk of analysing small reservoirs using a predetermined, constrained and subjective framework (Venot and Cecchi 2011) unable to capture the wider, sometimes unsuspected uses, benefits and knowledge of riparian farmers. Discussions focussed around predetermined topics designed on the basis of literature review and initial visits to the reservoirs. These sought to paint a portrait of how lives had evolved around the reservoir, following its construction (récit de vie) and centred on the history of the site and small reservoir project, its management and associated water uses. For dam operators, additional questions focussed on their duties and their hydrological understanding of the resource and its variations. The framework is appended in section 10.3. Printed high resolution satellite images (provided through Microsoft Bing and Google Earth) of the lakes were used as a visual aid for interviewees and interviewers to represent the location of plots and pumps, and stimulate discussions (carte parlée). The open questions aim to provide a thread to the discussion, and the challenge resides in allowing the interviewee enough room to talk freely and stray off course into potentially interesting revelations, whilst identifying when it is necessary to reign in the discussion. Likewise, the interviewer must recognise opportune moments and issues potentially raised by the interviewee on which to expand. Successful interviews rely on appropriate use of the informal framework, but depend largely on the ability to gain their confidence, through the approach (initial contact, flexibility in interview times, etc.), shared informal moments (meeting the family, sharing tea, coffee or food, visiting their farm, etc.), and repeat visits. Interviews were carried out with the dam operator and riparian farmers on all four lakes (figure 5.1.2). A second phase of complementary interviews based and building upon issues raised through the analysis of the first interviews focussed exclusively on Guettar reservoir, chosen due to the greater number of users and variety of issues revealed. These semi directed interviews were carried out in close collaboration with an anthropologist and with an interpretor. The lessons learnt from developing and implementing this interdisciplinary research are discussed in Riaux et al. (in prep.). Interviews were not recorded, but transcribed and typed up collectively the same day with the interpreter present to iron out the inherently common (Beaud 1996) uncertainties and incoherences. The circumstances and
184
(a) Morra
(b) Guettar
(c) Gouazine
(d) Hoshas
Figure 5.1.2: Reservoirs selected for ethnographic interviews and ongoing hydrological instrumentation progress of the interview were also described (where, when, other activities going on, other people present, silences, etc.) as these can assist and alter the interpretation given to certain replies.
5.1.3
Characterising lake benefits and extracting hydro-sociological constraints
The information extracted through quantitative agricultural interviews and qualitative semi directed interviews were combined to characterise the users and their agricultural practices on 22 small reservoirs. Based on the information gathered through rapid surveys across all reservoirs, these identified key water use characteristics of each lake, and led to a typology of the lake’s agricultural benefits. The water use and water availability typologies were then crossed (on the common subset of 48 lakes) to identify where water availability is a driver constraining or supporting agricultural water use. The incoherences highlighted as a result, associated with the indepth knowledge gained through interviews, then helped shed light on the multiple social, financial and institutional drivers (water access, land rights, role of management structures, participation) which must be accounted for to understand agricultural development around small reservoirs. Interpretation and analysis of the dynamics around the reservoir was performed through an iterative process, whereby initial hypotheses were gradually confirmed or invalidated by
185
the replies of the interviewees and the hydrological assessments. By repeating the same questions with each interviewee, the understanding of the dynamics around the lake gradually increase, as does the confidence in the replies obtained. Interviews trigger new questions and hypotheses which progressively reorient, refine and nuance our understanding of this hydro-sociological system. Second interviews with the same farmer notably helped confront our evolving theory and clarify new interrogations. The multiple replies obtained allow us to triangulate the information (e.g. who shares the pumps, did the government assist, what happened to the water user associations, conflicts, etc.) until we reach a saturation point where the additional information only reinforces our understanding of the system but ceases to provide new insights (Olivier de Sardan 1995). The observations seek to identify and define the relationships and interactions between our variables, leading to a representation of the inner workings of the system, sharing some (relative) similarity with hydrological approaches. Insights gained from interviews were also fed into the design of the questionnaires (e.g. questions on land rights, water user associations) in order to identify the spatial distribution of these drivers and help upscale local observations.
5.2 5.2.1
Results Supporting and changing agricultural practices around small reservoirs
Agriculture around the lakes is composed of cereals, fruit trees and market gardening, combined with livestock. The introduction of small reservoirs was first and foremost accompanied by a change in agricultural practices, notably the development and expansion of irrigated activities, sometimes at the expense of other livelihood strategies. 5.2.1.1
Irrigated practices
Fruit trees Fruit trees were observed on the banks of over 80% of all lakes in the catchment and 58% of these fruit farmers declared taking up arboriculture activities following the construction of the lake. Though the number of trees varied between farmers, all farmers on the lakes surveyed had planted fruit trees, indicating the homogeneity of the practice. On some lakes, up to 4 400 fruit trees over 20 ha were planted following the construction of the reservoir. On others however, very small gardens were developed (13 trees on 0.5 ha at Skhira 3), and disparities were high between beneficiaries on the same lakes (figure 5.2.1). On average 720 trees over 6.3 ha were planted, contrasting with other results on the catchment (CNEA 2006) where 2000-3000 fruit trees were planted over more than 20 ha following the development of small reservoirs. These can be explained partly by the greater proportion of lakes with large exploitations used in their study, and Selmi et al. (2001) on 57 lakes in the wider Kairouan governorate found an average of 8 ha planted with fruit trees. In northern Tunisia around Jendouba, Khlifi et al. (2010) related an increase in the average surface area for trees from 3 ha to 23 ha over 8 years following the introduction of small reservoirs, however the Jendouba region benefits from significantly higher rainfall (450-950 mm from South to North), not accounting for additional socio-economic differences which may explain this dynamic. In the Merguellil upper catchment, only a few farmers strive to expand the number of trees on their land (El Maiz, Sidi Sofiane 2, Mahbes, Morra), with 186
Number of fruit trees per farm on each lake
4000
3000
2000
1000
Smili_2
Skhira_5
Skhira_4
Skhira_3
Sidi_Sofiane_2
Sidi_Sofiane
Morra
Maiz
Mahbes
Kraroub
Guettar
Gbatis
Garia_S
Fidh_Zitoune
Fidh_Mbarek
Fidh_BenNasseur
Fadden_Boras
En.Mel
Daoued_1
Bouksab
Bouchaha_A
0
Figure 5.2.1: Observed heterogeneity in the number of fruit trees per farm and per lake based on questionnaires most others having already exploited all their land and only planting new trees to replace dead ones. On smaller farms, fruit trees and cereals were grown within the same plots. The heterogeneity observed confirms the inadequacy in using national statistics at this scale (CNEA 2006) and where high resolution multi-spectral satellite imagery is available, a diachronic quantification of the small agricultural plots around each lake could provide valuable insights. Fruit trees observed are olive, almond, grenade, fig, apricot, peach, apple and cherry (figure 5.2.2). Except on several lakes in the north of the catchment near Kesra (O. Daoued and Skhira lakes) where cherry trees dominate, olive trees are the preferred choice for farmers. These are favoured by the local populations for their resistance to droughts, low labour, prior experience in growing, as well as historical and socio-cultural reasons (medical benefits, sign of wealth, etc.) (Selmi et al. 2001). Yields vary widely over the years, rendering precise assessments of the amount or percentage of the production sold difficult. Olive harvests are very low the first 5 years and then provide increasing yields every second year, subject to suitable hydro-climatic conditions. Olive yields in the catchment vary between 14 kg (1 galba) and 70 kg (5 galba) for certain very old trees. Lachaal et al. (1997) indicate that yields in Southern Tunisia can reach 29 kg/tree on average but remain around 13 kg in Central Tunisia. 14 kg of olives produce a galba of oil (4-5 litres) and if farmers have the means they transform the olives and sell the oil directly. Olives sell at around 0.8 Tunisian Dinar (DT) per kg while 1 litre of oil fetches 5 DT, i.e twice as much per kg. Here, trees provided olive oil for the extended family consumption (between 100 and 300 litres of oil) 187
2.4%
14.1%
0.3% Olive 5.5%
Almond Pomegranate Fig Apricot
9.4% 61.2%
Peach Cherry Apple
0.7%
6.3%
Figure 5.2.2: Percentage of each fruit tree species grown by small holders interviewed and on good years, 65% of people surveyed declared being able to sell excess produce from their fruit trees. These were both farmers with large exploitations, and smaller farmers with only 100 trees who used their olive production to purchase other goods. There are 1441 oil mills (huileries) spread throughout the country and production can be easily commercialised through the Office National de l’Huile. Reported almond yields varied between 5 and 30 kg/tree/year, apricots 10 and 50 kg/tree, figs 10 and 60 kg/tree, and cherry 20 to 50 kg/tree. Market gardening Market gardening had been attempted at some stage on 80% of lakes surveyed and by 65% of farmers interviewed. Only 20% of all farmers continue this activity today. These grow a range of vegetables mostly tomato, chilli and watermelon but also fennel, courgettes (zucchini), green peas as well as winter vegetables such as onions and broad beans. Surface areas were typically small (under 0.5 ha) and mostly for personal consumption except for two farmers surveyed for whom market gardening became and remains today an income generating activity (Ain Smili, Guettar). These values are close to those observed by Selmi et al. (2001) who had previously observed an average of 0.25 ha for market gardening on lakes in the region, and similarly observed a rapid decline of activity within three years. Market gardening where successful, is said to increase mean revenue by over 400% (from 800 DT to 3800 DT), according to regional data (CNEA 2006). Associated withdrawals These practices were supported where and when water availability allowed by withdrawals from the lakes using motor pumps, cisterns or smaller tanks. Rapid surveys confirmed that 77% of all reservoirs supported withdrawals at some point and 66% still do today, highlighting an interest for this additional resource across most lakes.
188
100
% of farmers interviewed around SR
90 80 70 60 50 40 30 20
Cereals & livestock
10
Fruit trees Market gardening
0
Time after SR construction
Figure 5.2.3: Diversification of agricultural practices following the small reservoir construction based on farmer interviews These figures however disguise the relatively low water use actually witnessed around reservoirs with irrigated practices in the vast majority of cases limited to occasional watering of small plots of fruit trees as more intensive fruit farming or market gardening activities remain exceptional. Motor pumps were indeed only available on 43% of lakes and in limited number, while on the other 23% withdrawals consisted of occasional cisterns or smaller tanks. Nevertheless, 85% of farmers interviewed relied, albeit minimally, on the lake. The 7 farmers who did not rely on the lake had access to a well nearby (2 farmers), or only relied on rainfall once trees became hardier (2), or purchased cisterns (3), in many cases taken from the Merguellil wadi, near Ain Bou Khris (En Mel, Guettar lakes) and directly downstream of Skhira gauging station for Fidh Zitoune. Where motor pumps were available, their numbers ranged between 1 and 7 (mean 2.5) but across all lakes the mean value fell to barely 1 motor pump per lake (figure 5.2.5). Comparing 2005 data on the same reservoirs showed that the average number remained stable. No pumps had been installed on lakes which had no pumps in 2005 despite some of these holding water resource potential (Garia S, Gouazine) and on 19% of lakes pump numbers reduced due to damages or the reduction in irrigated activities (En Mel). On a third of lakes however, additional pumps had been acquired and replaced over time, signalling a continued or increased interest in the lake’s resources. Interviews revealed that pumps are shared between 2.9 users on average but this is subject to changes over time, due notably to economic constraints and conflicts described below. Individually purchased pumps are often shared between family members with their own exploitation or working on an (as yet) undivided family exploitation. Pumps provided by the government to a cluster of independent farmers, were over time used by only 1 or 2 main users, as a result of conflicts and abandon of agricultural practices by other users. The number of pumps and beneficiaries
189
(a) Gravity fed irrigation of fruit trees
(b) Market gardening plots
Figure 5.2.4: Agricultural water uses around SR in the Merguellil catchment
190
10
8 Pumps.2012
6
Number of pumps per lake
4
2
0 10
8 Pumps.2005
6
4
2
Smili_2
Smili_1
Trozza_sud
Skhira_D
Skhira_C
Skhira_5
Skhira_B
Skhira_4
Skhira_A
Skhira_2
Skhira_3
Skhira_1
Skhira_27
Sidi_Sofiane
Salem_Thabet
Sidi_Sofiane_2
Morra
Raouess
Mdinia_2
Mazil
Mdinia
Maiz
Marrouki_2
Mahbes
Marrouki
Hoshas
Kraroub
Hoshas_amont
Habsa_B
Habsa_A
Hammam
Guettar
Gbatis
Gouazine
Garia_3
Garia_2
Garia_S
Fidh_Zitoune
Fidh_Ali
Fidh_Mbarek
Fidh_BenNasseur
En Mel
El_Haffar
Fadden_Boras
− Dhahbi
ao e
Daoued_1
Bouksab_3
Chauoba_Hamra
Bouksab
Bouksab_2
Bouchaha_B
Ain_Faouar
Ben_Houria
Bouchaha_A
Abda
0
Figure 5.2.5: Number of motor pumps in 2005 (on 21 lakes) and 2012 (on 56 lakes) were however notably difficult to assess, as revealed by the combination of user surveys, questionnaires and interviews. Pumps can be placed in storage to protect them from floods, but also for extended periods when these need repairs or when farmers leave to work as labourers. Other pumps may be on site but not used due to financial difficulties but also conflicts notably within the water use associations. The number of pumps nevertheless provides a useful metric of the potential for agricultural exploitation of the lake, and a proxy indicator of the interest of farmers and of the government’s support for irrigation. Data provided by farmers on the number of users sharing the pump was also subject to significant discrepancies as was shown through triangulation of their answers. Other fruit farmers relied essentially on cisterns or small tanks, individually owned or purchased from local entrepreneurs, due to their distance from the lake or in the absence of access to motor pump and pipes. These were witnessed on 11 lakes including lakes where water uses were not suspected from the rapid surveys, due to their portable and mobile nature. From our interviews, these were only allowed on 45% of lakes surveyed. On half of these, their use was restricted to riparian farmers, with as many as 20 families reported to use cisterns around lakes (Garia S, Skhira 3, O. El Haffar or El Mahbes) while on the other half, cisterns were also provided to farmers from further afield, for free (50 cisterns/year at Morra, 100 at Skhira 4) or sold as on O.Daoued 3 where 4 cisterns of 3000 l at a time operate, and up to 50 cisterns per day are withdrawn during summer months. The withdrawals associated with these heterogeneous practices are estimated and compared with water availability in section 5.2.4. Water access issues are discussed further in 5.2.5.1. 5.2.1.2
Cereals
Cereal cropping continues to be practised by 70% of farmers around small reservoirs, but 20% declared ceasing rainfed cereals after the lake’s construction to concentrate on fruit trees. The other 10% had never practised this traditional livelihood activity. Farmers explained their conversion from cereals to arboriculture by the reduced space and means to continue 191
cereals. Land area has notably reduced through inheritances and the inability to leave plots as fallow led to reduced land fertility, increasing the need for costly inputs. Costs to plough and harvest small plots around 1 ha are also proportionally greater. This conversion remains much less significant than around small reservoirs around Jendouba where in 2 years, people had allegedly all converted to fruit farming and market gardening (Khlifi et al. 2010). Cereals are essentially durum wheat, barley and occasionally oats. The latter two are used as fodder. Farmers owned on average 10 ha, ranging between 1 – 60 ha. In certain (five) instances, farmers also rented land for cereals. Profits are either shared 50:50, or the land is rented at a rate of 200-400 DT/ha/year. Rented surface areas varied between 5 and 10 ha except for one farmer (El Mahbes) who rented 200 ha. Reported yields reach 1.3 tonnes per hectare, but varied between 0.5 t/ha and 3 t/ha. The production provides for the families’ personal consumption and seeds for the following year. 73% of farmers also sold excess production during the wetter years. Though traditionally rainfed, in very rare instances (Morra and Sidi Sofiane), farmers had installed sprinklers (up to 9 over 3 ha) to exploit water from the lake during the early growth stages. In both cases, the practice ceased rapidly as unreliable rainfall during the rest of the growth period during which they chose not to continue watering reduced any benefit. Similar practices in the Jendouba region had also been observed on a handful (3 out of 18) of reservoirs but again only over the first two years (Khlifi et al. 2010). In certain cases (Skhira), soil humidity on the contours of lake were used to advantage by farmers to grow cereals (flood recession cropping), however benefits can be outweighed by potential flooding and several farmers preferred to refrain from cropping near the reservoirs to reduce risk of damage to their crops. Accordingly, farmers also complained about the loss of land as a result of the flooded area of the reservoir. 5.2.1.3
Livestock
Livestock is a major traditional activity throughout the gouvernorat which continues to be practised by 70% of farmers surveyed and represents a significant source of capital and income notably from the sale of lambs around the traditional Muslim festivities. 27% of farmers declared abandoning this activity following the construction of the lake, as farmers reported having less time to water and feed their stock, but also due to increasing pressure on land leading certain years to conflicts over fodder. Several people (Guettar, Mahbes, Skhira) also reported increased theft since the revolution and the generalised insecurity that followed. This led families to sell off part of their flocks of sheep, or keep them in their compound at night in order to reduce the risk of losing this valuable financial resource. Others also admitted to changing their strategy and opting for bovine or even in one case dromedaries (Fidh M’Barek) partly due to these being harder to steal and calves reaching up to 1400 DT. Farmers surveyed owned between 1 and 4 cows, 2 to 60 sheep (mean 10), and several (17%) also owned up to 15 goats. Farmers which continued livestock herding also declared reducing their herds & flocks by up to two thirds. 70% of lakes surveyed were reportedly used to water livestock. In certain remote areas with limited alternate water supply, they reduce significantly the distance livestock must travel, thereby increasing productivity of the herd. On Skhira, a farmer with more than 400 fruit trees and no other water source felt that the lake’s greatest value was facilitating
192
(a) Livestock grazing on the lakebed
(b) Cereal cropping in the lakebed
Figure 5.2.6: Additional uses of SR in the Merguellil catchment
193
Downstream well Irrigated agriculture
Downstream well
Gouazine lake
0
50 100
200 Metres
Figure 5.2.7: Irrigated agriculture on wells situated directly downstream from the Gouazine lake watering livestock. Likewise on Gouazine, the local population had allegedly requested the lake for livestock purposes. On the other 30%, existing alternatives such as the nearby wadi (e.g. at Gbatis), or water fountains and concerns over lake water quality and salinity (Guettar, En Mel, Fidh M’Barek) explained why herders did not use the lake for watering livestock. Lakes also support nomadic herders passing through (e.g. Morra) or settling in northern lakes for a few months to find fodder and water (Skhira, En Mel). 5.2.1.4
Fishery and other complementary activities
No fishery activities were identified over the small reservoirs in the region. The absence of perennial water stocks to support the development of fish larvae are clearly a major constraint, however at Morra reservoir, which only dried out once due to a temporary leak from a faulty valve, this activity could have been developed successfully. Apiculture is well developed in these rural areas, and bee keepers are allowed to put their hives around the lakes (Guettar, Mahbes) to take advantage of the lake and the surrounding vegetation. Other complementary agricultural activities of farmers around the lakes include collecting fodder (17 DT per truck around Guettar, Fidh Zitoune), picking pine nuts in the forested mountainous regions around Kesra, and gathering rosemary (e.g. Fidh Zitoune) to be converted into oil (9-10 DT for 100kg). These activities were however not dependent on the small reservoirs. Furthermore only 17 out of 48 farmers focus exclusively on agriculture, the other 31 continuing to have an additional source of income.
194
5.2.1.5
Groundwater recharge and peripheral water use
A distinction was made here between the activities developed directly on the lake and those developed on wells which may partly benefit from the influence of the lake. These wells are influenced by different water availability patterns and were often initiated long before the development of the lake upstream. In certain areas, small reservoirs display very rapid infiltration, notably Hoshas (“gravel”, a reference to its subsoil) or Mouidhi (cf. figure 4.3.12). According to the available literature 8 reservoirs were specifically designed to favour recharge (cf. table 5.2.1) of the local (Bou Hafna, Cherichira) aquifers (CNEA 2006). Geochemical studies in the region (Gay 2004; Montoroi et al. 2002) showed that infiltration from Gouazine reservoir reached 300 m3 /day and that this signal was observed on wells exploiting shallow aquifers 350 m downstream within 100 days (figure 5.2.7). Previous reports (CNEA 2006) had assessed that 4 to 10 shallow wells benefited from this recharge on 4 lakes (Salem Thabet, Hoshas, Dhahbi and Abda). Here, this was reported by farmers on 12 lakes, who witnessed the positive influence of the lake on the level and recovery times of their wells. Infiltration rates will depend on the local geology, lithology, successive deposits on the lake bed and stage levels, while the infiltrated volumes will depend on the amplitude and duration of floods as well as evaporation losses. By contributing to recharging groundwater locally, small reservoirs indirectly benefit the users of these local wells which support fruit tree and market gardening activities. The positive effect was notably documented by Selmi and Talineau (1994); Selmi and Zekri (1995) on two wells downstream of Gouazine where levels rose from 18m depth to 7m and from 6m to 2m respectively after the dam was built, as well as in other parts of Tunisia (Zairi et al. 2005). Around the Kesra plateau, where no deep boreholes for irrigation are reported and where drinking water is piped over long distances, the wells situated on the banks of the Skhira 3-5 lakes were widely used. On Sidi Sofiane, users also favoured the well directly downstream of the lake due to its greater availability. In other cases (Guettar), farmers with a well downstream alternated between the lake and the well, as the lake didn’t suffer from drawdown issues. In addition to formal wells, farmers also seek to tap part of the groundwater or subsurface flows supported by the lake, by digging in the river bed downstream of the embankment. This was notably observed after the lake ran dry to harness residual soil humidity (Guettar) but also as a more long term strategy either because of greater availability (Bouksab 3) or because of conflicts with landowners bordering the lake shore who won’t allow bergater pipes through their land (El Maiz).
5.2.2
Wider benefits of SR
5.2.2.1
Protecting downstream areas
Reservoirs as part of the national water and soil conservation strategy were also designed to collect and reduce silting of downstream areas and lakes. Sedimentation rates are highly variable between subcatchments, and affected partly by size, rainfall intensities, lithology and land use. Based on mean sedimentation rates across 14 reservoirs in the region, over 200 000 m3 sediment may be collected annually by reservoirs in the catchment (cf. section
195
2.4.3). This corresponds to a reduction in silting of the downstream El Haouareb dam of the order of 15% estimated around 1.33 to 1.48 M m3 /year. These results are of the same order of magnitude as estimates in CNEA (2006) which identified a reduction in silting of 2.6 M m3 over 15 years. Other lakes, were designed to protect roads (Bouksab 2 & 3) or plots (Fidh Ben Nasseur) situated directly downstream from floods. For users, the captured silt causes difficulties and discontent, notably reducing available capacities over time as discussed in the water availability assessments (section 5.2.4). Furthermore, problems were reported due to the detrimental flushing of accumulated silt and debris onto the land of farmers situated below (Bouchaha, Fidh Ben Nasseur). On another lake, damage to the wall combined with heavy silting, has led to regressive erosion flushing the accumulated sediment, effectively shifting the reservoir from being a retainer of sediment to a producer. 5.2.2.2
Other socio-economic benefits
People interviewed also mentioned the wider value of lakes, with certain highlighting the contributing to landscape, shaded ecosystems (Mahbes) as well as recreational benefits, with children swimming in some lakes and their large duck populations attracting hunters. Miscellaneous benefits reported also include using water for house construction and washing wool in the lake (Morra, Guettar). Household use was only mentioned on 8 out of 22 lakes surveyed, as public fountains are relatively widespread, though can be used during common water shortages. Biodiversity benefits have yet to be investigated. On the whole farmers, despite limited agricultural benefits and discontent, expressed attachment to their lakes, which were also for some a source of prestige (e.g. Ain Smili). Significantly, beyond being a material resource capable of increasing local production, lakes appeared to play a key role in “structuring and maintaining local social groups” (Riaux et al. 2014b).
5.2.3
Typology of lakes’ agricultural benefits
Based upon the qualification of water uses and users across 22 lakes and the rapid surveys carried out across the catchment, the benefits across all 56 lakes were characterised (figure 5.2.8). Existing typologies focus notably on physical characteristics (age of structure, state, slope) (Lacombe 2007; Albergel and Rejeb 1997), or on detailed agricultural practices (sources of income, trajectories) (Selmi et al. 2001; Khlifi et al. 2010), but here the objective was to focus on the benefits across all lakes from the farmers’ perspective. The withdrawals associated with these lakes are detailed in section 4.3.7. Lakes with negligible benefits to users These (16) lakes currently support negligible agricultural withdrawals, and do not visibly recharge any nearby wells. These include (9) lakes which never did support agriculture and have essentially a protective value, reducing erosion and flooding of downstream areas. This was in certain cases their primary objective (table 5.2.1) but certain larger lakes failed to develop irrigated practices, largely due to their isolated location (Chaouba El Hamra and Marouki 2). Other lakes ceased to support agriculture after becoming completely silted (Ben Houria, Bouchaha B) or sufficiently to be abandoned (Skhira A) or now only allow for the occasional watering of a few trees, up to 100. 196
Negligible
Residual
Isolated
bene ts
bene ts
high bene ts
16 lakes
27 lakes
13 lakes
Limited withdrawals
Isolated intense withdrawals
No direct withdrawals
Essentially protection
Up to 1200 m3/month/lake
300 fruit trees/farm 0.5 ha market gardening
Up to 4100 m3/month/lake
900 fruit trees/farm Up to 2.5 ha market gardening
Figure 5.2.8: Water use typology of 56 lakes studied in (and near) the Merguellil upper catchment
197
Lakes with residual benefits to users These (27) lakes support occasional withdrawals, and provide small, irregular but noticeable benefits to nearby farmers. These support agriculture directly through motor pumps (13) or through cisterns (8) while a total of 10 (also) support water levels of nearby wells exploited for irrigation. The number of pumps remains low (1 per lake on average) and withdrawals essentially support the occasional watering of fruit trees. Sampled exploitations possess on average 300 (up to 1000) fruit trees, 7 ha of cereals, between 10 and 50 heads of sheep, and when availability is greater, occasional small scale market gardening (0.5 ha) for personal consumption continues to be observed. This largest category includes lakes of all sizes, including very small lakes (under 40 000 m3 capacity) where single riparian farmers continue to occasionally pump on the lake, to large lakes (over 800 000 m3 capacity) where the low number of pumps installed (2-3) reflect the low exploitation of these lakes. Several of these lakes were built to support irrigation, while others (Skhira 4&5) built in the 1960s to protect from soil erosion, have supported the development of several orchards through direct pumping and nearby wells. Lakes with isolated, high benefits These (13) lakes support modest regular withdrawals and have assisted the development of isolated but significant exploitations. Two of these are currently exploited using only cisterns while the others withdraw using motor pumps (2.5 pumps per lake on average), and two lakes also contribute to recharging nearby wells. Activities focus essentially around fruit trees which reached 900 on average across our sample (and up to 4400), 15 ha of cereals and between 10 and 50 heads of livestock essentially sheep. Commercial market gardening of 2-5 ha were attempted but farmers preferred to focus on fruit trees, even using drip irrigation on three lakes. Lakes were often large (table 5.2.1) but also include two smaller lakes (40 000 m3 ). In most cases though, the benefits are the results of individual entrepreneurs and the lakes and pumps appear to have contributed in reinforcing existing capabilities. These lakes suffer from low equity, where benefits are confined to single users, who possessed existing capital or reliable sources of income (e.g. Fadden Boras).
5.2.4
Water availability drivers of water use
The Landsat observations were used to derive quantitative assessments of the periods the lake was flooded and could potentially support irrigation. Figure 5.2.9 illustrates the vast differences in interannual water availability across 48 lakes which reached over 200 000 m3 /day on large lakes and remained close to 0 for many of the small reservoirs over 20072014. As seen from figure 5.2.11, lake volumes displayed stark interannual variability over the summer months, expectedly high on small lakes but also remarkably high on the large Kraroub lake (variation coefficient = 0.76), contrasting with other large lakes such as Morra and Mdinia (variation coefficient = 0.23). Amongst smaller lakes, vast differences are also observed, notably between lakes such as Gouazine and Mahbes. This variability has a substantial influence on the ability of lakes to provide both substantial and consistent water resources, which is clearly summarised in figure 5.2.10b for all lakes. As described in section 5.1.1.2 this information was then used to identify the number of days lake levels reduced below 5000 m3 and assess the suitability of each lake to provide a consistent resource, over the right period and in sufficient quantity. Figure 5.2.12a highlights the vast disparities
198
Residual benefits
Negligible benefits
Table 5.2.1: Water availability and use characteristics for each lake Lake
Initial capacity (m3 )
Mean availability (m3 /day, 2007-2014)
Ben_Houria
17 000
341
Bouchaha_B
34 000
Bouksab_2 Bouksab_3
Nb of pumps
Nearby wells
Initial objective
Y
0
N
Protection (Kairouan)
620
Y
0
N
Irrigation
N/A
N/A
Y
0
N
Protection (road)
N/A
N/A
Y
0
N
Protection (road)
120 000
15 876
N
0
N
Irrigation
Garia_2
19 000
403
Y
0
N
Recharge
Garia_3
25 000
1 239
Y
0
N
Recharge
Hoshas_amont
N/A
N/A
N
0
N
N/A
Marrouki_2
56 000
3 691
N
0
N
Irrigation
Raouess
18 000
1 285
N
0
Y
N/A
Skhira_1
181 000
11 494
N
0
N
Protection
Skhira_27
N/A
N/A
N
0
N
Protection
Skhira_A
72 000
16 277
Y
0
N
Protection
Skhira_B
120 000
11 269
N
0
N
Protection
Skhira_C
52 000
16 896
N
0
N
Protection
Skhira_D
N/A
N/A
N
0
N
Protection
Abda
37 000
4 972
N
0
Y
Recharge
Ain_Faouar
66 000
3 309
Y
0
Y
Recharge
Bouchaha_A
18 000
1 560
Y
1
N
Irrigation
Bouksab
55 000
8 152
Y
0
Y
Irrigation
Dhahbi
26 000
2 799
N
0
Y
Recharge
El_Haffar
30 000
1 720
Y
0
N
N/A
Fidh_Ali
134 000
30 092
Y
2
N
Irrigation
Fidh_BenNasseur
47 000
3 751
Y
1
Y
Protection (plots)
Fidh_Mbarek
53 000
4 287
Y
1
N
Irrigation
Gbatis
106 000
17 598
Y
2
N
Irrigation
Gouazine
237 000
65 435
Y
0
Y
Irrigation
Habsa_A
50 000
3 964
Y
0
N
Irrigation
Habsa_B
35 000
912
Y
0
N
Irrigation
Chaouba_Hamra
199
Withdrawals
Residual benefits Isolated, high benefits
Lake
Initial capacity (m3 )
Mean availability (m3 /day, 2007-2014)
Hammam
850 000
107 826
Hoshas
130 000
Maiz
Nb of pumps
Nearby wells
Initial objective
Y
2
N
Irrigation
7 365
Y
0
Y
Recharge
500 000
132 408
Y
5
N
Irrigation
Marrouki
153 000
21 179
Y
0
Y
Recharge
Mazil
104 000
30 438
Y
4
N
N/A
Mdinia
1 200 000
229 921
Y
3
N
N/A
Mdinia_2
N/A
N/A
Y
0
N
N/A
Salem_Thabet
63 000
6 426
N
0
Y
Recharge
Skhira_2
38 000
2 748
Y
0
N
Protection
Skhira_3
79 000
3 238
Y
0
N
Protection
Skhira_4
160 000
28 526
Y
3
N
Protection
Skhira_5
60 000
23 085
Y
3
Y
Protection
Smili_2
35 000
6 333
Y
1
N
Irrigation
Trozza_sud
50 000
13 760
Y
1
N
Irrigation
Daoued_1
95 000
7 350
Y
4
N
N/A
Daoued_2-3
350 000
N/A
Y
1
N
N/A
En Mel
1 000 000
172 073
Y
7
N
Recharge
Fadden_Boras
94 000
21 742
Y
2
N
N/A
Fidh_Zitoune
40 000
5 244
Y
1
N
N/A
Garia_S
1 500 000
53 483
Y
0
N
N/A
Guettar
150 000
28 126
Y
6
Y
Irrigation
Kraroub
1 590 000
229 970
Y
2
N
Irrigation
Mahbes
180 000
42 863
Y
4
N
Irrigation
Morra
705 000
302 106
Y
2
N
Irrigation
Sidi_Sofiane
40 000
6 696
Y
2
Y
Irrigation
Sidi_Sofiane_2
N/A
N/A
Y
0
N
Irrigation
Smili_1
130 000
19 677
Y
2
N
Irrigation
44
23
13
Number of lakes where present
200
Withdrawals
Mdinia 200000 100000 0
100000 50000 0
2000 2004 2008 2012
40000 20000 0
Mean daily volume (m3) ove ap −sep
0 2000 2004 2008 2012 Trozza_sud
2000 2004 2008 2012 20000 10000 0
2000 2004 2008 2012 0.50 0.00
10000
0
Ben_Houria
Smili_1
2000 2004 2008 2012 0.50 0.00
Abda
0 2000 2004 2008 2012
20000 10000 0
Hoshas
0 2000 2004 2008 2012
Ain_Faouar
Fidh_Mbarek
0 2000 2004 2008 2012
2000 2004 2008 2012
Sidi_Sofiane
Smili_2
1000 0 2000 2004 2008 2012 Raouess 2 1 0
2000 2004 2008 2012 0.50 0.00
2000 2004 2008 2012
Marrouki_2
400 200 0
El_Haffar
2000 2004 2008 2012 Dhahbi
2000 2004 2008 2012
Skhira_5
0.50 0.00
Skhira_B
Garia_2
2000 2004 2008 2012
Habsa_B
Habsa_A 4000 2000 0 2000 2004 2008 2012
Bouchaha_A
Bouchaha_B 1.0
Skhira_2
2000 2004 2008 2012 0.50 0.00
2000 2004 2008 2012 0.50 0.00
2000 1000 0 2000 2004 2008 2012
0.50 0.00
2000 2004 2008 2012
2000 2004 2008 2012
Skhira_C
2000 2004 2008 2012
0.0 2000 2004 2008 2012
0.50 0.00
Garia_3
2000 2004 2008 2012 0.50 0.00
400 200 0
500 0
Bouksab 200 100 0
2000 2004 2008 2012
2000 2004 2008 2012
Fidh_Zitoune
2000 2004 2008 2012
Fidh_BenNasseur
2000 2004 2008 2012 0.50 0.00
2000 1000 0
0 2000 2004 2008 2012
1500 1000 500 0
2000
Daoued_1 4000
2000
0 2000 2004 2008 2012
2000 2004 2008 2012
Skhira_1
4000
1000 500 0
2000 2004 2008 2012
2000 2004 2008 2012
Skhira_4
0.50 0.00
20000
0
Fadden_Boras
2000 2004 2008 2012 Fidh_Ali
20000
8000
2000 2004 2008 2012 10000 5000 0
0 Marrouki
Gbatis
40000 20000 0
20000 2000 2004 2008 2012
2000 2004 2008 2012
Mahbes
Chauoba_Hamra
Guettar
100000 0 Gouazine
2000 2004 2008 2012
2000 2004 2008 2012
Maiz
2000 2004 2008 2012
0.50 0.00
1000 500 0 2000 2004 2008 2012
Mazil
0 2000 2004 2008 2012
Kraroub
2000 2004 2008 2012
Hammam 100000
0e+00
2e+05 0e+00
0e+00
120000 60000 0
Morra 2e+05
2000 2004 2008 2012
En.Mel 2e+05
Garia_S
2000 2004 2008 2012
Skhira_A
2000 2004 2008 2012
Skhira_3
2000 2004 2008 2012 0.50 0.00
Salem_Thabet
2000 2004 2008 2012
Year
Figure 5.2.9: Mean daily water availability over the dry season per year over 2000-2014
201
Ben_Houria Ain_Faouar Abdessadok Abda ● ●●● ● ●●● ●●●●● ● ●● ● ● ● ● ●●●● ●● ●● ●●● ● ● ● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●● ●●●●●●●●●●●●● Marrouki ●● ●● ●● ●● ●● ●● ●● ●●● ●● ●● ●● ●● ●● ●● ●● ● ●● ●● ●● ●●●●●●● ●● ●● ●● ●● ●● ● ● ●● ●● ●● ● ●● ●● ● ● ●● ●● ●● ● ● ●● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●● ●●● ●●●●●●● ● Mahbes ●●● ● ●● ● ● ●●●● ●● ● ● ●●●●● ●● ● ● ● ● ●● ● ●● ● ● ●●●●●● ● ● ●● ● ● ●● ● ●●●●●●● ●● ● ● ● ●● ●●● ●● ●● ●● ●● ● ●● ●● ● ●● ●● ●● ● ●● ●● ●●●●●● ●● ●● ●● ●● ●● ●●● ● ●● ●● ●● ●● ● ●● ●● ● ●● ●●● ●●●●●● ●● ●● ●● ●●●●● ●● ● ●● ● ●●●●● ● ●● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●● ●●●●●● ● ●●● ●●●●●●●●●● ●●● ● ●●●●●●● ●● ●● ●● ●● ● ●●●●●● ●●● ●● ● ●● ● ●● ● ●● ● ●●●●●●●●● ●●●●● ●●● ●● ●● ●● ●● ●● ●●●●● ●●●● Garia_S ●● ● ●● ● ●● ● ●● ●● ●●● ●● ●●●●●● ●● ● ●● ●● ●●● ●● ●● ●●●● ●●●●● ●●●●●● ●● ●● ●● ●● ●● ●● ●●● ●● ●● ●●●●●●●●●●● ●● ●● ●●●●●●●●●●●●●●●● ● ●● ● ●● ●● ●● ●● ● ●● ●
Number of days per year water volume is inferior
300
200
100
0
●●● ● ●● ●● ●●●● ●●● ●● ●●●● ●●● ●●● ●●● ●● ●●● ●●● ●● ●● ●●●● ●● ●●● ●●● ●●● ●● ●●●●● ●● ●●● ●● ●● ●● ●●●●●● ●● ● ●● ●●● ●● ●●● ● ●● ●● ● ●●● ● ● ●●● ●● ● ●● ● ●●●● ●●●●● ●● ● ● ●●● ●●● ●●● ● ●●●●●●●●●●●●●●● ●●● ● ● ● ●● ●● ● ●● ● ●● ●●● ● ●● ●● ●● ●●●● ●● ●● ●● ●● ●●●● ● ●●●●●● ●●●●●● ● ● ●● ● ●● ●●●●● ●● ●● ● ● ●● ●● ●●●●●● ●● ● ●●●●● ●●●● ●●●●● ● ● ●●●●●● ●● ● ● ● ●●●●●● ●●●●● ●●● ●●●●●● ● ●● ●●●● ● ●●●●●● ●●● ● ●● ●● ●● ●●●●●● ●●●●● ● ●● ● ●●●●●● ●● ●●● ● ●●●● ●● ●● ●●●● ● ● ●●●●● ●● ●●●●● ● ● ●● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ●● ●● ●●●● ● ●● ●● ●● ● ●● ●●●●● ●● ●● ●●● ●●● ● ● ●●●●● ● ●●● ● ●● ●● ●● ● ●●● ●●●●● ● ●● ●●● ● ●●● ●● ● ● ●● ● ● ●● ●● ●● ●●●● ●●● ●● ●● ●●● ●●● ●●●●●● ●● ●●●●● ●● ●●● ●●● ● ●● ●● ●● ● ● ●●● ●● ● ●● ●● ● ●● ●● ● ● ●●●●●● ● ● ●● ● ●●●●● ●● ●● ●● ●●● ●●● ●●● ●●● ●●● ●●● ● ● ●● ● ●●●●●● ●● ●● ●● ●● ● ●● ●●●● ●●●●●● ●● ●● ● ●●●●● ● ● ●● ●● ●● ● ● ●● ●●● ●● ● ●● ● ●●● ● ●●●● ●● ● ●● ● ● ●● ● ● ● ●●● ●●●●●●● ●● ● ●● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●●● ● ● ● ● ● ● ●●●● ●●●● ●● ● ●●● ●●● ●● ●● Hammam ● ● ●● ● ●●● ●●●●●●● ●● ●● ●●● ● ●●●● ●● ● ●●● ●● ● ● ● ● ● ●●● ● ●● ●● ●● ●● ● ●●● ●● ●● ● ● ● ● ●● ● ●● ●● ●● ●● ● ●● ● ●●● ●● ●● ● ●●● ● ● ●●● ● ●● ● ● ●● ●● ● ●● ● ●●● ●●● ●●● ●● ● ●●● ● ●● ●● ● ● ●● ●● ●● Gouazine ● ●● ●● ●●● ●●●● ●● ●●● ● ●● ●● ●●●●●●● ● ● ●● ●●● ● ●● ●●●●●● ●● ●●● ● ●● ●● ●●● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ●● ●● ● ● ● ●●●●●● ●● ●● ●● ● ●● ●● ●● ● ● ●● ● ● ●● ●● ●●●● ●●● ●● ●●●●●● ●● ● ● ● ●● ●●●● ● ●●● ●●● ●● ● ● ●●●●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●● ●● ● ●● ●● Dekikira ●●● ● ● ●● ●● ●● ●●●●●●● ● ●●● ●●● ● ● ●● ● ● ●●● ● ●●●●●●● ● ●●●●● ●● ● ● ●● ● ● ● ●● ●● ●● ●● ●● ● ●●●●●●● ● ● ●● ● ● ● ● ●● ●●●●●● ●● ● ● ●●● ●●● ● ● ●●● ●● ● ● ●●● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ●● ●●●● ● ● ● ●● ● ● ●● ● ● ● ●●● ●● ● ● ●● ● ●●●● ●●● ●● ● ●● ● ● ● ● ● ●● ●●● ●● ● ●● ●● ● ● ● ● ●●●● ● ●● ● ● ●● ●● ● ●● ●●●●● ●● ●● ● ● ●● ●● ●● En.Mel ●● ●● ●●●●● ●● ●● ● ● ● ● ● ●● ●●●●●● ●●●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● Maiz ●● ● ●● ● ● ●● ●● ●● ● ●● ●● ● ●●●● ● ●●●●●● ●●●●● ●● ● ●● ● ●● ●● ●●●●● ●●● ● ●● ●● ●●●● ● ●● ●● ● ●● ● ●● ● ● ●●●● ●●●●●● ●●● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ●●● ●● ● ●● ●● ●● ● ● ● ● ●●●● ●● ● ●●● ● ● ●● ● ●● ● ●●●●●●●● ●● ●● ●●● ●●● ●● ● ●● ●●●● ●●●● ● ●●● ●● ●●● ●● ●● ●●●● ●● ●● ●●● ● ●●●●●● ●● ●● ● ●● ●●●● ●●● ● ● ● ●● ● ●●● ●●●● ● ● ●●●● ● ● ● ● ● ● ●●●●● ●● ●●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ●● ● ● ●● ● ● ● ● ●●●●●●● ● ● ●● ●● ● ● ●● ●● ●● ● ● ●●●●●●●● ●● ● ●● ●●●●●●● ●● ● ●●● ● ●● ●●●●● ● ● ● ●●● ● ●● ● ●● ●● ●●● ●●●●● ●● ●● ● ●● ●● ●● ● ●●●●● ●●● ● ● ●● ●● ●● ● ●●● ● ● ● ●●●●●●●●● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●●●● ●● ●●●● Kraroub ● ● ●● ●● ●●● ●●● ●● ●● ● ●●●●●●● ● ●● ●● ●● ● ●●● ●●●●●●● ● ●● ● ●● ● ●● ●●● ●● ● ●● ●●●●●● ● ● ● ● ● ●● ● ●● ●●●●●● ●● ● ● ●● ● ● ●● ●●●●●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●●● ●● ● ●●●● ● ● ● ●● ●● ●● ●●●● ● ●● ●● ● ● ●● ● ●●● ●● ● ●●●● ● ● ●●● ●●●● ● ●●●●●●● ● ●● ●● ● ●● ● ●● ●●●● ●●● ●● ● ●●● ●●●● ●● ●●● ● ● ● ●● ●●● ● ●● ●●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●●●● ● ●●●● ●●● ● ●●●●● ●● ●●●●● ● ●● ● ● ●● ●● ●● ●●● ● ● ●● ●●●● ●● ●● ● ●● ● ●● ●● ●●●●●● ●● ●●● ● ●● ●● ● ●●●●●● ● ●● ●● ● ●●● ● ●● ●●● ●●● ●● ● ●●●● ● ● ●●● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ●● ● ●●● ●●●●● ●● ●● ●● ●●●●●● ●●●● ● ●●● ●● ●●●● ●● ●●●● ●● ● ● ● ● ●●●●●● ●● ●●● ●●● ●●●●● ●●● ●●●● ●● ●● ●● ●●●● ●●● ●●●●● ●● ●● ● ● ● ●●●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●●●●●●●● ●●●●● ●●●●●● ● ● ●●●●●●●●●● ● ●● ●● ●●● ●●●●●● ●●●●● ●● ●● ●●●●● ● ●● ● ● ● ●●●● ●●● ●● ● ●●● ●● ●●●● ● ●● ●●● ●●● ● ● ● ● ●● ● ●●● Mdinia ●●●● ●● ●●● ● ●●● ●● ●● ●● ●●● ●● ● ●●●● ●● Morra ● ●●● ●● ●●●●●● ●●●●●● ●●● ●●●●● ● ●●●●●●●●●●●●●●●●●●●●● ● ●●● ●●● ●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●● ●●●●●●●●●●●●● ●● ●●●●● ● ●● ●● ●● ●● ●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●●●●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●●● ●
0
50000
100000
150000
Water volume (m3)
(a) Volumes up to 150 000 m3 Ben_Houria Habsa_B Bouchaha_A Garia_2 Bouchaha_B Garia_3 El_Haffar Raouess Skhira_3 Dhahbi Skhira_2 Habsa_A Marrouki_2 Ain_Faouar Hoshas Fidh_BenNasseur Salem_Thabet Abda Fidh_Zitoune Fidh_Mbarek Sidi_Sofiane Smili_2 Brahim_Zaher Skhira_1 Daoued_1 Bouksab Skhira_B Skhira_A Abdessadok Marrouki Gbatis Smili_1 Trozza_sud Gouazine Skhira_C Chauoba_Hamra Guettar Fidh_Ali Skhira_5 Maiz Skhira_4 Mahbes Mazil Fadden_Boras Kraroub
Number of days per year water volume is inferior
300
200
100
En.Mel Dekikira Hammam Garia_S Mdinia Morra
0 0
2500
5000
Water volume (m3)
7500
10000
12500
(b) Zoom on volumes up to 10 000 m3
Figure 5.2.10: Number of days for each lake water volumes fell below a designated volume
202
Interannual mean availability (m3/day) ove ap −sep
between lakes and led to three groups of lakes to be defined based on their water availability patterns (figure 5.2.13).
3e+05
2e+05
1e+05
Mdinia Garia_S Morra Hammam En.Mel Kraroub Dekikira Mahbes Mazil Maiz Guettar Fadden_Boras Gouazine Marrouki Fidh_Ali Smili_1 Gbatis Trozza_sud Abdessadok Brahim_Zaher Skhira_4 Hoshas Skhira_1 Daoued_1 Ain_Faouar Fidh_Mbarek Fidh_BenNasseur Bouksab Sidi_Sofiane Smili_2 Garia_3 Garia_2 Raouess Marrouki_2 Habsa_B Habsa_A Fidh_Zitoune El_Haffar Bouchaha_A Bouchaha_B Dhahbi Skhira_5 Skhira_2 Skhira_3 Skhira_C Skhira_B Skhira_A Salem_Thabet Ben_Houria Abda Chauoba_Hamra
0e+00
Figure 5.2.11: Mean interannual availability over 2007-2014 per lake over the dry season displayed as mean ±1 standard deviation Negligible water availability The first category consists of lakes with negligible water potential. 27 lakes never reached the required water availability and a further 6 met this requirement for less than 45 days during the 6 dry months (i.e. less than 25% of dry season days). A minimum of 45 days on average during the dry season was considered necessary to support any market gardening activities based on minimal crop cycles lengths. These lakes are typically small and therefore unable to hold water for any length of time, due to lower depths and associated greater exposure to evaporation. Some of these problems originate from their design, with four lakes (Ben Houria, Garia 2, Bouchaha A and Raouess) under 20 000 m3 . Several of these were designed for protection or recharge and appear under-dimensioned for water uses, despite farmers having planted trees and installed a pump besides these. Other somewhat larger lakes, suffered from progressive silting (Bouchaha B, Garia 3, Haffar) or like Hoshas (130 000 m3 ) experience very short floods due to high infiltration. For 20 of these, figure 5.2.12bconfirmed that uncertainties over silting rates were not significant as even under the best case scenario of maintaining initial capacity through dredging or building new dams at these locations and dimensions, these continued to fail to meet the defined availability requirements. On the other 13 lakes, sensitivity to the silting model was much greater as seen in the differences between figures 5.2.12a and 5.2.12b for 203
Mdinia Garia_S Morra Hammam En.Mel Kraroub Mahbes Mazil Maiz Guettar Fadden_Boras Gouazine Marrouki Fidh_Ali Smili_1 Gbatis Trozza_sud Skhira_4 Hoshas Skhira_1 Daoued_1 Ain_Faouar Fidh_Mbarek Fidh_BenNasseur Bouksab Sidi_Sofiane Smili_2 Garia_3 Garia_2 Raouess Marrouki_2 Habsa_B Habsa_A Fidh_Zitoune El_Haffar Bouchaha_A Bouchaha_B Dhahbi Skhira_5 Skhira_2 Skhira_3 Skhira_C Skhira_B Skhira_A Salem_Thabet Ben_Houria Abda Chauoba_Hamra
Number of days per year water volume < 5 000 (m3)
Mdinia Garia_S Morra Hammam En.Mel Kraroub Mahbes Mazil Maiz Guettar Fadden_Boras Gouazine Marrouki Fidh_Ali Smili_1 Gbatis Trozza_sud Skhira_4 Hoshas Skhira_1 Daoued_1 Ain_Faouar Fidh_Mbarek Fidh_BenNasseur Bouksab Sidi_Sofiane Smili_2 Garia_3 Garia_2 Raouess Marrouki_2 Habsa_B Habsa_A Fidh_Zitoune El_Haffar Bouchaha_A Bouchaha_B Dhahbi Skhira_5 Skhira_2 Skhira_3 Skhira_C Skhira_B Skhira_A Salem_Thabet Ben_Houria Abda Chauoba_Hamra
Number of days per year water volume < 5 000 (m3) 300
200
100
0
(a) With silting modelled
300
200
100
0
(b) Best case scenario, i.e. with no silting
Figure 5.2.12: Number of days for each lake water availability fell below 5000 m3 over the whole year (light red) and the 6 dry months (dark red)
204
Mdinia Garia_S Morra Hammam En.Mel Kraroub Mahbes Mazil Maiz Guettar Fadden_Boras Gouazine Marrouki Fidh_Ali Smili_1 Gbatis Trozza_sud Skhira_4 Hoshas Skhira_1 Daoued_1 Ain_Faouar Fidh_Mbarek Fidh_BenNasseur Bouksab Sidi_Sofiane Smili_2 Garia_3 Garia_2 Raouess Marrouki_2 Habsa_B Habsa_A Fidh_Zitoune El_Haffar Bouchaha_A Bouchaha_B Dhahbi Skhira_5 Skhira_2 Skhira_3 Skhira_C Skhira_B Skhira_A Salem_Thabet Ben_Houria Abda Chauoba_Hamra
Number of days per year water volume < 5 000 (m3)
Mdinia Garia_S Morra Hammam En.Mel Kraroub Mahbes Mazil Maiz Guettar Fadden_Boras Gouazine Marrouki Fidh_Ali Smili_1 Gbatis Trozza_sud Skhira_4 Hoshas Skhira_1 Daoued_1 Ain_Faouar Fidh_Mbarek Fidh_BenNasseur Bouksab Sidi_Sofiane Smili_2 Garia_3 Garia_2 Raouess Marrouki_2 Habsa_B Habsa_A Fidh_Zitoune El_Haffar Bouchaha_A Bouchaha_B Dhahbi Skhira_5 Skhira_2 Skhira_3 Skhira_C Skhira_B Skhira_A Salem_Thabet Ben_Houria Abda Chauoba_Hamra
Number of days per year water volume < 5 000 (m3)
Good availability
Good availability
Unreliable availability Negligible availability
300
200
45
100
135
0
(a) With silting modelled
Unreliable availability Negligible availability
300
200
45
100
135
0
(b) Best case scenario, i.e. with no silting
Figure 5.2.13: Definition of water availability categories based on number of days during 6 dry months water availability exceeded 5000 m3
205
instance on Chauoba El Hamra lake or several of the Skhira lakes. Field visits confirmed that these still supported (minimal) storage capacity and occasional water uses, confirming the difficulties expressed in section 2.4 to model silting across all lakes sometimes over more than 30 years. In the upstream area of the Skhira plateau, lakes were built in the 1960s. The lower topography and heavy cereal cropping will have reduced soil loss and silting rates, allowing these to still provide water to users after 50 years. The Skhira 5, Daoued 1 and Smili 2 lakes were notably exploited by pumps and wells downstream, while on others (Chauoba El Hamra) no pumps were observed due its isolated location but availability was present. In the absence of updated levelling to identify the actual available volumes and considering their ability to support albeit minimal watering, these 13 lakes were therefore placed in the next class of lakes. Interestingly, Landsat flood dynamics observed on these lakes also pointed to their ability to provide insights into the silting patterns of lakes. Increased silting leads to lower depths, meaning the same surface area over time will evaporate or infiltrate faster, which can be observed from the surface area dynamics. This was clearly observed on 4 lakes which field visits confirmed where silted (Ben Houria, Garia 2, Bouchaha B and Raouess). The 13 lakes where silting uncertainties were greater also continued to display large and extended flood periods based on Landsat surface assessments (as seen in the previous chapter, figure 4.4.18). On all these lakes, water availability can be limited and unreliable, providing inopportune conditions to develop more intense irrigated activities but when not silted, allow for occasional watering. Figure 5.2.14 illustrates how the cumulated water volumes present on lakes such as Ain Faouar over the 6 month growing period were insufficient most years to support the water demand for minimal watering of trees, even without accounting for the silting over time. In most cases, as seen in figure 5.2.10b, water availability patterns are similar when considering lower volumes, however for very small lakes, here Habsa A and Fidh Zitoune, their availability patterns around 2500 m3 is significantly greater pointing to reliable though minimal water volumes. On Fidh Zitoune, one pump was notably installed to water some trees. Unreliable water availability This second category gathers the lakes where uncertainties over water availability are greatest, and where lakes are dry on average between 1.5 and 4.5 months out of the 6 dry months. It includes 9 small (40 000 m3 ) and large lakes (Gouazine and Maiz 230 000 m3 and 500 000 m3 respectively) where the significant variability leads to numerous droughts despite the interannual water availability over the dry months exceeding 5000 m3 /day over 2007-2014 (figure 5.2.11). As seen in the previous chapter (figure 4.4.1a) and in figure 5.2.9, certain years water availability may fall to zero for extended periods. Conversely, years of good availability, more intense localised irrigation of vegetables is possible but farmers must compose with the uncertainty through other resources, be they physical (other watering points) or economic (means to purchase cisterns). Figure 5.2.15 which accounts for the estimated water demand shows how available volumes during summer months on Gouazine barely sufficed to meet the basic watering demand of 1200 m3 /month during summer months several years. Smaller lakes, such as Daoued 1 succeeded, though sometimes barely, in meeting the high withdrawals observed of 4 100 m3 /month during summer months, though dry spells at some point during 206
Ain_Faouar
1200
Mean daily volume (m3) over apr−sep
800
400
0
2000
2004
2008
2012
Year
Figure 5.2.14: Simulated water deficit on 1 lake based on observed water demand those 6 months were likely. For 8 of these 9 lakes, results were insensitive to the effect of silting uncertainties, i.e. even after construction or if these were dredged annually, these suffer from significant uncertainty. Only Fadden Boras in the absence of silting showed reduced variability in the number of days above 5000 m3 /day (figure 5.2.12b). Field visits confirmed it was used for irrigation, and good water availability was reported by the farmer. Landsat imagery also showed prolonged flood patterns and despite its modest size (maximum capacity of 84 000 m3 ) may in fact benefit from good water availability, subject to confirmation through levelling. As mentioned above, 13 additional lakes shift from the negligible to unreliable category due to silting uncertainties. Good water availability The 6 lakes in the third category displayed water availability superior to 5000 m3 for more than 4.5 months during the 6 dry months (i.e. over 75% of dry season days). These were the 6 largest lakes (over 500 000 m3 ) which is coherent with the fact that storage is determinant in semi arid areas where floods are rare. However El Maiz was excluded from this category, suffering despite its similar capacity from greater water insecurity. Conversely, as mentioned above Fadden Boras may also despite its smaller size benefit from reliable water availability. The mean availability of lakes as shown in figure 5.2.16 is indeed only partly correlated (R = 0.6) with the maximum capacity of lakes, as can be expected considering the spatial rainfall and associated runoff variability as well as differences in catchment runoff coefficients and lake characteristics (notably high infiltration). Part of these will have been accounted for during the dimensioning of dams, however Garia S, though reliable, shows relatively low mean water volumes compared to 207
Guettar
Garia_2
Mean daily volume (m3) ove ap −sep
Mean daily volume (m3) over apr−sep
600
400
200
40000
20000
0
0 2000
2004
2008
2000
2012
2004
2008
2012
Year
Year Daoued_1
Gouazine
Mean daily volume (m3) over apr−sep
Mean daily volume (m3) over apr−sep
150000 10000
5000
0
100000
50000
0 2000
2004
2008
2012
2000
2004
Year
2008
2012
Year
Figure 5.2.15: Simulated water availability on 4 lakes based on observed water demand
●
3e+05
y = 5119 + 0.141 ⋅ x, r 2 = 0.636
●
Mean daily volume (m3) over 2007−2014
●
2e+05
●
●
●
1e+05
●
● ●
● ●
●
● ● ● ●● ● ● ●● ● ● ●● ● ●●●● ●
0e+00
0
●
● ●
● ● ●
●
● ●
● ●
●●
● ●
500000
1000000
1500000
Initial maximum capacity (m3)
Figure 5.2.16: Relationship between initial capacity of each lake and the mean daily volume over 2007-2014
208
Table 5.2.2: Categorisation of 48 small reservoirs based on water availability and water uses Water availability
Agricultural benefits
Negligible
Unreliable
Good
Negligible
Ben Houria, Bouchaha B, Garia 2, Garia 3, Marrouki 2, Raouess
Chauoba El Hamra, Skhira 1, Skhira A, Skhira B, Skhira C
Residual
Abda, Ain Faouar, Bouchaha A, Dhahbi, El Haffar, Fidh Ben Nasseur, Fidh M’Barek, Habsa A, Habsa B, Hoshas, Salem Thabet, Skhira 2, Skhira 3
Bouksab, Fidh Ali, Gbatis, Gouazine, Maiz, Marrouki, Mazil, Skhira 4, Skhira 5, Smili 2, Trozza Sud
Hammam, Mdinia
High (isolated)
Fidh Zitoune
Daoued 1, Fadden Boras, Guettar, Mahbes, Sidi Sofiane, Smili 1
En Mel, Garia S, Kraroub, Morra
other lakes of its capacity and may have been over-dimensioned considering its reduced catchment size (5 km compared to 20 km for other large lakes). These larger lakes remain affected by interannual variability, as seen in the high standard deviation values (figure 5.2.11), but would not here affect irrigation practices. Morra for instance only dried out once and due to a major leak on a broken outlet valve. Significantly, these lakes are the ones which can realistically provide year round irrigation, i.e. support farmers through the critical dry periods in line with their objective in many semi arid regions. Many other lakes, only offer temporary storage after rains when needs are lower and are not reliably capable of supporting irrigated practices. Water use on large lakes remains limited though and estimated withdrawals of 4100 m3 /month on Morra are negligible considering the 300 000 m3 average water availability. No correlation existed between the availability of lakes and the water use levels, assessed as the number of fruit trees grown or number of pumps around lakes. The discrepancies between the water uses and water availability characteristics of each lake (table 5.2.2 and illustrated in figure 5.2.17) confirmed the need to account for other wider factors to explain the agricultural development around lakes.
5.2.5
Wider hydro-sociological drivers
5.2.5.1
Water access difficulties
As discussed in section 5.2.1.1, the limited number of pumps available on lakes constitutes a first obstacle to water use. 56% of lakes are currently not equipped with any pumps, while the others are equipped with only 2.5 pumps per lake on average. The number of 209
High (isolated)
Daoued_1 Fadden_Boras Sidi_Sofiane Mahbes
Fidh_Zitoune
Kraroub
Morra
Guettar
En Mel Garia_S
Maiz
Skhira_3 Habsa_B
Residual
Increasing water use
Smili_1
El_Haffar Fidh_BenNasseur Fidh_Mbarek
Marrouki Bouksab Gbatis Skhira_5
Skhira_2 Ain_Faouar Bouchaha_A Hoshas Salem_Thabet
Abda
Dhahbi Habsa_A
Mdinia
Trozza_sud
Fidh_Ali Gouazine
Skhira_4 Smili_2
Hammam
Mazil
Skhira_A
Negligible
Marrouki_2 Raouess Bouchaha_B Garia_3
Skhira_1 Chauoba_Hamra Skhira_C
Garia_2 Ben_Houria
Skhira_B
Negligible
Unreliable
Good
Increasing water availability
Figure 5.2.17: Figurative representation of water availability and water uses for small reservoirs according to the 3x3 categories in table 5.2.2.
210
¯
Gouazine
Daoued_1 Skhira_1 Skhira_C Skhira_4 Skhira_5 Skhira_A
Bouchaha_A Garia_S Mdinia
Fadden_Boras
Fidh_Zitoune
Smili_1 Sidi_Sofiane
Kraroub
Mahbes En Mel Bouksab
Mazil Fidh_Ali
Maiz
Morra
Hammam
Number of pumps (2012) Water availability (m3/day, 2007-2014) 1
100
2
1 000
3
5 000
4-5
10 000
6-7
Marrouki
Kairouan plain El Haouareb dam
Trozza_sud
0
100 000
2.5
5
10 Kilometers
Figure 5.2.18: Map of interannual mean water availability (2007-2014) and number of motor pumps per lake pumps per lake is not related to mean water availability (figure 5.2.18) indicating the wider socio-economic and institutional factors which influence water access around lakes. Based on interviews over 22 lakes, over half (54%) of these pumps were provided by the government following the construction of the lakes. The others were purchased by farmers and their costs and use were typically shared between 2-3 brothers or cousins. Prices for a typical (Hatts 10 CV) motor pump reportedly cost between 1000-2000 DT, which farmers without pumps stated being unable to afford. Credits are notably difficult to obtain due to the absence of land titles (titre bleu) on 73% of exploitations surveyed. Farmers also highlighted the severe difficulties in paying credits back, with one being bailed out by the government and another having severe interests to pay. Farmers who had purchased their own pump were typically wealthier exploitations having inherited large plots of land, or farmers who had additional external resources (regular employment, or family cash flows), however a third of the private pumps had been purchased without such assistance. Furthermore, several users reported problems with their pumps, from lack of regular maintenance or direct damage essentially from floods. Due to economic difficulties to maintain and repair pumps, as well as difficulties to obtain spare parts for certain pump brands, a noteworthy 16 out of 59 pumps were reported out of order at the time on 22 lakes. Government provided pumps were sometimes taken back for repairs but farmers were still required to cover repair costs. Farmers stated needing to pay 200 DT or more for repairs or a new battery and extended periods before the pumps were (sometimes) repaired where reported. Likewise, due to running costs of gasoline and oil, certain users chose not to exploit their pumps every year. Access to shared pumps is also limited due to institutional issues
211
discussed in section 5.2.5.4. Water distribution costs also create a financial hurdle for farmers and several requesting bergater pipes from local authorities. Users rely on motor driven pumps to draw water from the lake or sometimes from wells situated directly downstream, and/or fill cisterns and tanks, sometimes from the lake’s outlet valve. Drip irrigation was only recorded on three exploitations (Sidi Sofiane, Fadden Boras and El Kraroub), while in all other cases, water was conveyed to the plots using bergater pipes and a more or less complex network of trenches and impoundments to channel water to each tree. Some farmers had built a basin and practised gravity-fed irrigation (Mahbes, Daoued). In Jendouba, water irrigation from lakes was done using sprinklers, with each farm having 3 to 6 sprinklers each (Khlifi et al. 2010), highlighting the very different nature of exploitations from small reservoirs. Physical factors, here topography, also constrain water extraction with the pumps being unable to water the plots of certain farmers situated uphill (Guettar) or further away. Where possible these fill cisterns (Sidi Sofiane 2) or use two pumps in sequence but the latter was abandoned due to prohibitive costs (En Mel). At El Kraroub, one farmer succeeded in watering his land situated over 1 kilometre away, while on other lakes (Morra, Ain Smili), remote beneficiaries were allowed to draw cisterns at a very low price on the lake (3DT/cistern, i.e. to cover the price of gasoline). The marly, silty soils also prevent farmers from using their pumps when water levels fall below a metre due to silt clogging the pumps. 5.2.5.2
Managing hydrological uncertainty
Farmers faced with high variability and associated uncertainty over water resources managed in certain instances to develop successful enterprises due partly to their ability to deal with water shortages. To save their fruit trees, farmers reported needing 1 cistern of 3000-4000 litres for 10-12 trees, 2-5 times per year, depending on rainfall and the age of the trees. Droughts during the summer months representing a minimum costs of 500 DT to water 1 ha of trees twice at 25-30 DT/cistern. These costs of complementary water supply were difficult to bear by many farmers and several lost part of their younger fruit trees and market gardening attempts during the initial droughts. Over 60% of farmers interviewed had attempted market gardening but more than half of these ceased after just one unsuccessful attempt, due to the high costs involved. Their limited capabilities (Sen 1999) affects their willingness to take on risks and most users around these lakes continue to focus exclusively on olive and other fruit trees. Some farmers continue to attempt limited market gardening if water levels are high in the spring but only resort to watering their fruit trees should the lake run dry. These uncertainties can prevent them from planting or lead them to reduce investments, e.g. applying lower than optimal doses reducing their production levels (Khlifi et al. 2010; Mugabe et al. 2003) Only those with existing other sources of livelihoods appear willing and capable to bear such risks, due to their increased financial resilience and their ability to access alternate sources of water. 32 farmers interviewed with parallel access to wells and cisterns developed greater irrigated activity partly through their capacity to overcome these temporary shortages. Those with wells (15 farmers) notably exploited both sources of water, in certain cases displacing their pump between the two, while others resorted to water from the wadi or buying cisterns when required.
212
(a) Government provided shared pump
(b) Individually owned and repaired pump
Figure 5.2.19: Diversity of pumps observed on SR in Merguellil catchment
213
Uncertainties were also reported due to the uncontrolled withdrawals of other users which led to farmers being unable to predict how long water supplies would last. Their understanding of flood dynamics remained low, with significant discrepancies reported through interviews, e.g. that when full, the lake could be used for three to four years, when in fact a single flood would dry up within a year. This contrasts with experiences in other subSaharan areas where farmers had a relatively accurate understanding of how often and how long the lakes filled up which may be related to the extreme importance placed by users in very arid areas over the water supply. 5.2.5.3
User participation and government assistance
Objectives and siting of reservoirs Part of the low valorisation of the reservoirs’ resources also stem from the siting of the reservoirs and their selection criteria (Selmi and Talineau 1994; Talineau et al. 1994; Wisser et al. 2010). Criteria found in related project documents referred to obvious physical (hydrological, geological) parameters coherent with the water and soil conservation objectives however apparently failed to include a social component relating to water use potential despite the irrigation objectives. Studies here (Talineau et al. 1994; Selmi and Zekri 1995) and elsewhere (Habi and Morsli 2011) already pointed to the importance of including socio-economic analysis to identify water availability potential, and identify the structure, practices, potential and initial results of the first production units around small reservoirs, if these are to form a “central part of a rural development policy” (Talineau et al. 1994). This initial consultation can help understand the local challenges to harness the resource and reduce problems of appropriation, and entrust greater responsibility over ownership and management of structures to end users, which lacked here. The interviewed farmers confirmed they were not consulted on the construction of the dam, and no pre-assessment of their interest, objectives and capabilities was carried out. In certain cases (Chaouba El Hamra, El Marrouki 2) the clear absence of beneficiaries around the lake appears to contradict the stated objectives of developing irrigation. Government objectives for these reservoirs alternate between controlling sediment, groundwater recharge and upstream use, however the associated strategies appear somewhat fuzzy (Riaux et al. 2014b). On lakes built to support irrigation, no pumps or trees were provided, or sometimes only at a much later stage therefore losing out on the impetus created by the construction of the lake. On others, people reported (Gbatis, En Mel and Mahbes) that the government forbid lake users to sell market gardening products produced around the lakes. The development of water use around lakes in many cases remained merely an “implicit” objective (Selmi and Zekri 1995), behind the priority protection of downstream infrastructure. This may originate from opportunism towards funding opportunities or progressive rewriting of objectives over time but 20 years after the conclusions were published, the means developed to encourage upstream use remain limited. This may partly explain some of the incoherences observed where beneficiaries were provided with horse drawn cisterns despite having no livestock (Sidi Sofiane 2) and thus sold off, or why large irrigation projects such as those at Morra and Kraroub never materialised. At Morra, the electrification strategy was never finalised, leading to a built, yet inoperative pumping station.
214
Participation and appropriation On certain lakes, users reported agreeing (Gouazine) or even asking (Fidh M’Barek, Guettar and Fidh Zitoune) for the lake, however in most cases the lake was apparently imposed, sometimes at the expense of their land and livelihoods. Compensation for taken land was allegedly offered but out of 29 exploitations with land under the dam, and 12 of these with land titles to prove it, only 2 reported obtaining some compensation (Garia S, El Kraroub). At Kraroub this amounted to 8000 DT for 11 ha. Certain users benefited instead from positions of influence and (temporary) remuneration as the dam operator (Fidh Zitoune, Guettar), though where possible these were local government employees living or working on site (e.g. Hoshas, Fidh Zitoune). After 10 to 30 years of existence for most lakes, many users still complain about their lake and the fact it has taken their land without compensation, or that it infiltrates and silts up too quickly, despite these both being part of their objectives. These were not always clearly presented and on one lake, intended for recharge (Hoshas), farmers were only advised after the construction that water use would be forbidden, placing undue expectations on the lake. Initial discussions could have helped clarify the real potential of these lakes and reduce the associated bitterness experienced by certain farmers. On Fidh Zitoune, the lady interviewed revealed having bought the land before the lake construction knowing a lake was planned, emphasising the level of hopes placed on this resource. Limited ongoing government assistance In the remote Merguellil upper catchment, public action for development is limited compared to the downstream Kairouan plain, and is compensated for through subventions and individualised small grants (Riaux et al. 2014b). The government has a tradition of supporting local farmers, notably providing subsidies to shift livelihood strategies from livestock to agriculture and encouraged water use around reservoirs. Following their construction, 15 out of the 22 lakes interviewed had benefited from a pump, provided by the government or the funding body (e.g. EU project) and over 1 in 2 of the farmers interviewed depended at least partly on a public pump. In addition, nearly half of farmers (45%) spoken to benefited from fruit trees saplings and one in three from bergater piping. The number of trees was allegedly conditioned to the number of holes dug however large differences were noted between lakes rather than users, with around 50 trees on El Maiz, Fidh M’Barek, Fidh Zitoune and between 300 and 1000 on Guettar and Mahbes. These disparities may stem from the different sources of funding at the time, but also from uneven distribution observed on certain lakes, where trees were shared by 3 out of 8 families instead of all riparian families. This concentration of public assets by powerful individuals is discussed further in the next section. Trees provided were a variety of fruit trees (peach, almond, olive, etc.) however the greater watering needs of peach and almond fruit trees led to several dying during the first years (Bouksab, Sidi Sofiane 2). The gifting of fruit trees served to help beneficiaries develop income but trees were also intended to reduce soil erosion and silting on reservoirs (Selmi et al. 2001; Khlifi et al. 2010). Government assistance appears to have been too limited and punctual for many users to make the transition from traditional rainfed activities to irrigated activities, which demand additional knowledge about technology (Oweis and Hachum 2006), water supply, fertiliser, access to markets, credit etc. Pumps and saplings were provided, but there was very little continued support given to farmers. “The EU project provided aid, and then disappears. . . people can’t cope”. Farmers expected further support and emphasised their reliance on 215
the government, possibly as a result of the historically interventionist government. Many farmers notably placed requests with the local governor in Kairouan or Ousseltia for help or subsidies for pumps as well as bergater and farmers felt that the government should pay to repair pumps, as these had been provided by them. Several dams suffered damage over the years but maintenance or repairs had not been carried out on 9 of the 22 reservoirs surveyed, which led farmers to complain in front of the regional delegation offices. Damages ranged from jammed outlet valves to serious breaches to the dam wall (El Hammam, El Maiz, Trozza Sud, Ben Houria) or an unfinished spillway as on Mdinia. The need for greater involvement by the government was previously pointed out but appears to have been insufficient and is not confined to this region (Selmi and Talineau 1994; Zairi et al. 2005). The management and maintenance of these smaller dams is indeed poorly budgeted (Selmi and Zekri 1995). This differs widely with experiences in China (Mushtaq et al. 2007) where desilting is planned every 2 years and the equivalent of 343 man days/year are planned for repair and maintenance. Likewise, the price of repairs on pumps is accounted for, and funds to actively maintain and invest in reservoirs. Local pond management groups are responsible and involved directly in the work. Despite the different context, continued support from the government and greater involvement by end users from the onset may have helped develop more successful water use. Greater monitoring may also have helped solve certain incongruities, e.g. why the 3 pumps connected to canals providing water to an irrigated perimeter (Kraroub) never functioned beyond the inauguration or why the pumps were never recovered. To be complete however, mediation efforts by the local government in water use conflicts on other lakes (Maiz) failed due to the power of certain local chiefs. Furthermore the absence of wider investments, such as roads hinder intensification of agricultural activities. The remote, isolated nature of certain lakes and exploitations constrains the intensification of agricultural practices as farmers reported being too far away from main roads and urban centres for people to come buy their production on site or for them to sell it on markets. At Ain Smili, the farmer lost 2 tonnes of chilli when the dirt track road became impracticable after heavy rains. Likewise at Morra, the absence of a good road, reportedly limits the interest in intensifying production and farmers throw out excess milk production. 5.2.5.4
Local water mismanagement
Ineffective water user associations Due to the limited number of pumps, pumps were shared between 3 users on average though this reached up to 8 users initially. The government when provided pumps insisted on the creation of Associations d’Intérêt Collectif (AIC) (similar to water user associations) to organize the management of the pumps. This was initiated across Tunisia and in 1997 22 AIC including 5 around Kairouan had been identified (Selmi and Talineau 1994). On the 22 reservoirs studied, 11 had formally created but in 2014 all had effectively dissolved. On 5 of these, they ceased due to conflicts between members and chiefs (El Kraroub, El Maiz, Guettar, Mahbes, O. Fadden Boras), on 3 they ceased due to the low number of users (1 user on Ain Smili, 4 on Sidi Sofiane and 2 pumps on Morra), while on 3 (Gbatis, Fidh Zitoune and O.Daoued) they ceased when the shared pump broke. AIC were created based upon the identification of numerous beneficiaries,
216
whose interest and activities on the lake declined over time with for instance 24 farmers identified in 2005 on Sidi Sofiane for only 1 pump and 4 farmers today. On the other lakes though requested by CRDA no AIC were created, due to the low number of users (2 to 3 on Bouksab, Fidh M’Barek, Bouchaha, Sidi Sofiane 2) and their existing family ties (En Mel, Sidi Sofiane, etc.). Conflicts within the AIC often originated over the use of the pump with several chiefs keeping it under lock and key (Kraroub, Maiz). One of these chiefs declared that the pump was his, but as water was unreliable, he uses cisterns. He “can’t sell it as it’s a state pump, could give it, but is keeping it in case he has a well one day”. Furthermore AIC chiefs prevented farmers using the lake with their own pumps, notably refusing that pipes run through their land bordering the lake shore. This led to the absurd situation of farmers placing the pump on a dug out, 50 metres downstream, which had been used during the construction of the lake for water. On two other lakes, governance of the AIC and possession of the shared pumps was taken over by the same person. In many cases, the roles of dam operator or AIC chief were given to local government employees present on site, or others with close links with the former ruling party, the Rassemblement constitutionnel démocratique (RCD). The importance of these ties with government officials and/or bribes were highlighted by several interviewees, as well as the importance of the local government representative who in certain areas was more proactive in reaping benefits for his constituency. Conflicts also occurred over the absence of provisions made to maintain and repair pumps, which eventually broke down. Management by default In most cases, people don’t feel a need for an AIC management, largely due to the low number of users on the lake and the family ties between them. Failed initial experiences may tint their perception and people negatively associate these with the previous totalitarian government. They also “already have enough problems with the drinking water user associations” (Mahbes, Skhira 4). In the absence of formalised or active AIC, management of the pumps and access to the lake is done through informal gatherings and discussions, though in some cases, established influences continue to play a role, with the dam operator (often the chief of AIC previously) having the final say. These on most lakes, but not all (Bouksab, Fidh M’Barek), continue to open valves, prevent children from swimming or restrict livestock to prevent pollution. The absence of clear management led to unregulated withdrawals on lakes. Farmers withdraw what they can afford while water is available, which on lakes such as Guettar and Mahbes where several pumps operate leads to a rapid decline in resources and a shortening of the flood period. This tragedy of the commons (Hardin 1968) was already observed on certain lakes where pumps and water resources are shared by Selmi and Talineau (1994). On Guettar this situation led to certain farmers calling for instating control and fees over water access, but coordinated by an independent government engineer. Water access on the lake remains restricted to riparian families of the lake, though such delineation can appear arbitrary as in figure 5.2.20. Farmers buying land around the lake are entitled to access water, reflecting the traditional bind between land and water rights, and “if the CRDA allows it”. External cisterns were only exceptionally allowed (Ain Smili, O.Daoued, Morra) and where required, farmers prevented external people from using the lake, sometimes with the CRDA’s help (Fidh Zitoune). The continued reference to the 217
Guettar lake
Excluded household
Beneficiary households
0
100
200
Metres
Figure 5.2.20: Location of beneficiary and excluded households around Guettar lake CRDA reflects how engrained the government’s presence is and translates a posture, where farmers seem to be merely managing a government lake where they gained rights, rather than actually appropriating themselves the lake. People have gradually resorted to individual pumps and government pumps have become private, due to users ceasing irrigated activities or members acquiring these after paying for their repair. Pumps often remain shared, but amongst family members which can lead to additional family members, previously outside the AIC, gaining access. Conflicts nevertheless endure, over water turns or covering repairs and gasoline costs. No ongoing maintenance is foreseen and the person who breaks it must repair it, leading to sometimes long delays before an agreement is found. In other parts of the catchment (Skhira 3-5), where no formal structures were created, lakes are managed at the community level (“each cluster of houses has its lake”). The age of the lakes, lower density of users and greater dependency on lakes for household water supply may partly explain the different management which on many accounts appeared successful. Village level management of ponds is also reported in China and Western Africa and may help reduce problems of mismanagement and overexploitation. 5.2.5.5
Conflicting livelihood strategies
Out of those who abandoned market gardening activities, 19 stated that water was the dominant constraint however 15 identified other reasons, which included the costs and low profits obtained, the intense labour and presence required. Small farmers who attempted commercial market gardening on Morra notably struggled with the need to have sufficient capital to cope until the harvest, and eventually preferred to continue working as labourers. With fruit trees the return on investment is minimal the first years as these only start to produce after 4-5 years, with good yields reported after 10 years. As a result, fruit 218
farming requires a long term investment, which for many farmers led to them pursuing complementary livelihoods on the side. Certain farmers (7) with large herds of sheep (between 50 and 60) and significant surface area (10-60 ha) of cereals showed no real interest in intensifying their fruit farming activities, preferring to rely on traditional agricultural livelihoods. Other farmers had already started to diversify their income due to land surface area reducing through inheritance and soil fertility declining over time, and work as labourers (Selmi et al. 2001; CNEA 2006). Many, when they have interesting opportunities (including abroad in Libya or Cameroon) pursue these activities and when contracts are scarce and/or water levels were high, choose to focus on agriculture and seek to intensify their production (market gardening, fertilisers, etc.). Out of 28 people who previously had a secondary source of income, 25 continue to have a secondary activity, implying that only three people were able to move towards agriculture exclusively. These sell higher value fruit (cherry, fig, apricots) as well as some market gardening, started fruit farming after the lake was built, and only one benefited from substantial family wealth. Conversely out of 20 farmers who previously only focussed on agriculture, 6 have now had to seek complementary activities, highlighting again the difficulties they faced in fruit farming.
5.3
Conclusions
The combination of qualitative and quantitative interviews over a selection of lakes highlighted the vast disparities in the benefits and practices between lakes, farmers and over time. Adopting a wide framework allowed to delve beyond initial conclusions of low water use based on limited pump numbers and identify benefits from cisterns and tanks, for livestock and for nearby wells. Historical insights also highlight the subtle changes in livelihood strategies that accompanied the introduction of the lakes and the numerous attempts to exploit their resources. These attempts to develop market gardening and fruit farming on 56% of lakes bear witness to the level of interest for local users in this additional resource, but in the vast majority of cases, agricultural water uses were limited to occasional watering of olive groves and other fruit trees. Numerous farmers reported attempting market gardening or intensifying trees but were deceived due to the unfavourable hydro meteorological conditions. Exploiting the Landsat derived water availability assessment provided unprecedented insight into the water availability patterns and constraints present on lakes in the catchment. On some of the smallest lakes, due to their size or rapid infiltration, their initial potential was already negligible and clearer communication with the riparian communities would have avoided disappointments and associated disaffection. On lakes where volumes can be significant and irrigated activities can be developed, farmers must be equipped with strategies to cope with the variability and associated uncertainty. For most this was limiting, and led to the loss of production, while those with the sufficient economic resilience or other water supplies survived the droughts. Though a limiting condition, water availability was indeed not a sufficient condition to understand the disparities in practices. On lakes where water availability is high, individual successes are reported and again occurred as a result of individual capabilities. The absence of ongoing government support to develop and structure agricultural activities and water management limited the number of beneficiaries, due largely to problems over 219
access to pumps, pipes, repairs and AIC conflicts. Farmers were in certain cases excluded from the start (Selmi and Talineau 1994) and others more subtly over time, through conflicts, mismanagement and appropriation of the pumps. Where the local government was proactive as on Guettar and Mahbes this led to wider levels of exploitation compared to other similar lakes, but benefits remained confined to a select few. Improvements can still be made to support practices on such lakes, where young entrepreneurs are investing their efforts. To this end, the information gathered here on the lake’s potential in terms of water availability may help make informed choices over future investments and risk management strategies. The government which provided in many cases the impetus for farmers to convert to fruit farming should where possible accompany the development of irrigation by providing continued financial support, as well as cognitive support. This requires a comprehensive approach and integrating the needs of end users throughout: from site selection and beneficiary selection, to maintaining funds to provide ongoing maintenance, support, repairs or even crop insurance policies. Here the incoherent government strategy led on the contrary to significant expectations and demand from the beneficiaries on the government which were not met. The means put forward only allowed reservoir water use to be developed as a complementary activity and most riparian farmers thus continue to earn their income as before either through cereal and livestock or through contracts as labourers elsewhere.
220
Part IV
Final conclusions and perspectives
221
Water resources Water availability assessments across ungauged reservoirs The research demonstrated the potential of 30 m Landsat imagery to monitor long term water availability (1999-2014) in ungauged small reservoirs of 1 ha to 12 ha surface area. Landsat surface area assessments provide valuable information on the amplitude and durations of the flood and when combined with surface-volume relations inform users on water availability, here with volumetric errors of the order of 10 000 m3 in mean annual availability. The method presents operational potential and can be transposed to other catchments to assess and monitor long term water availability patterns in small reservoirs which for economic and logistical reasons can generally not be instrumented. Relying on freely available satellite imagery, the method requires minimal field data to calibrate the water index thresholds and to derive surface-volume relationships reflecting local geomorphological conditions. Considering the increased development of small reservoirs across sub-Saharan Africa and worldwide (Wisser et al. 2010), due to their lower costs and support to small holders, this information may be used to inform farmers and stakeholders of the water resource potential contained by individual reservoirs and target investments appropriately. Here, the method provided the first insights into water availability across 51 reservoirs scattered across a large catchment (1200 km ), and highlighted the reduced availability and significant variability which affects over 80% of lakes.
Mutual benefits of remote sensing data assimilation in hydrological modelling Where field data is available to develop hydrological modelling of small reservoirs (i.e. rainfall, evaporation, infiltration data), data assimilation methods exploiting Landsat observations are shown here to significantly reduce uncertainties associated with modelling reservoirs in populated semi-arid catchments. An Ensemble Kalman Filter approach was developed to combine Landsat observations with a daily hydrological model (GR4J + water balance) on 7 reservoirs, providing valuable near-real time corrections and reducing runoff uncertainties due to highly variable and localised rainfall intensities. The high spatial and temporal rainfall requirements, as well as significant heterogeneity observed in the hydrological processes (infiltration notably) across reservoirs, further reinforce the value of remote sensing data assimilation in data sparse environments. The approach also revealed mutual benefits by improving the Landsat-derived flood dynamics, notably reproducing a more accurate flood decline and correcting remote sensing outliers.
Further hydrological applications Continued research is required to highlight the increasing precision and accuracy resulting from the higher temporal and spatial resolution imagery, notably forthcoming Sentinel-2 imagery and SWOT imagery, which is become more accessible to a wider scientific (research and operational) community, due to reduced costs and growing competencies. Greater resolution in time in altimetric measurements from sensors, aboard satellites or drones, could also reduce uncertainties from silting leading to reduced volumetric uncertainties (Massuel
222
et al. 2014a). Overall, the existing level of performance and complementarity confirm the growing value of remote sensing in contemporary hydrology which must be seized and opens up numerous increased opportunities (Winsemius 2009; Wackernagel 2004). These will however continue to depend on extensive field instrumentation to correctly calibrate, validate and correct results from remote sensing. These results also provide opportunities for further research in hydrological modelling, as the water volumes collected in each reservoir may be exploited as individual runoff gauges (Liebe et al. 2009). This spatialised information provides runoff data for 51 sub catchments and can be incorporated to refine semi-distributed models of the Merguellil upper catchment. Specifically, this will help in assessing the cumulative influence of small reservoirs, currently estimated between 1% and 50% (Ogilvie et al. 2016; Lacombe et al. 2008; Kingumbi et al. 2007) due to ongoing difficulties in distinguishing their role in the decline in downstream flows over other climatic and human pressures. Likewise, evaporation across all reservoirs may be assessed based on the surface area time series, refining their water balance and assessments of groundwater recharge. Combined with a greater understanding of water use in upstream reservoirs, these results will provide much needed data to feed into catchment-wide water management discussions (Lacombe 2007). The high demand for groundwater in the downstream Kairouan plain indeed introduced demand for upstream-downstream optimisation (Le Goulven et al. 2009) and could lead to defining alternate resource management scenarios, seeking to optimise groundwater recharge for downstream but also upstream well users.
Water uses and associated drivers Wider value of small reservoir investments Water withdrawals identified here through a combination of in-depth interviews, questionnaires and wider surveys remain minimal, supporting in most cases the occasional watering of fruit trees. Their value can however not be assessed simply against performance or efficiency measures (Venot and Cecchi 2011) and depend on the perspective and yardstick used. Despite minimal market gardening and high income crops, the majority of reservoirs led to crop diversification and/or an increase in the surface area of fruit trees, supported at least in part by the lake’s resources. The olive harvest provide a non negligible revenue to the families, providing olive production for the whole extended family. During good years, farmers can develop limited lucrative activities (selling excess) and fruit farms in the long term guarantee income growth (Selmi and Zekri 1995). Associated ethnographic research (Riaux et al. 2014b) points to the wider social and political dimension of this resource and users appear attached to these lakes, especially in remote areas where livelihoods are difficult. Heterogeneity of practices inter and intra lakes was high and isolated successful enterprises based around intensive fruit farming were also observed. In the same way that government subsidies maintain deficient or inefficient agricultural sectors in numerous countries, these investments have supported remote rural areas, affected by erosion, poor soils and low and irregular rainfall, albeit with low equity and often incoherently. They probably contributed to slow the inevitable fate of rural exodus, “without the lake, this area would die”. Considering some of these lakes are over 40 years old and con223
tinue to provide residual, non negligible benefits beyond their life expectancy, these clearly have a positive impact. People in the area have a saying “Wadi Merguellil belongs to the mountain people but the profits are for the Kairouan people”. These projects which allowed some development around small reservoirs may be a step in the right direction in helping local upstream small holders (Vincent 2003) exploiting the wadi Merguellil and should be encouraged.
Longer term strategies and integrated approaches The majority of small reservoirs which depend on erratic rainfall volumes are however limited in their ability to develop irrigated agriculture on any significant scale. Lakes can support dry spells during the rainy season, and on good rainfall years intensification during dry months if supported adequately to optimise production and minimise risks could be performed. However fundamentally small reservoirs, despite gradual shifts in the discourse (Venot and Krishnan 2011) are designed for supplementary irrigation and their limited storage does not address the years when rainfall is low, preventing a long term transition to irrigated agriculture. Already, this reduced storage capacity, prevents support to farmers when they need it the most: people through their traditional livelihoods “know how to live when there is rain, it’s when there is none that it’s a problem”. If the risks associated with uncertainty can be lifted, the lake ceases to be both “a help and a threat” and people can serenely envisage developing agricultural activities. In China’s Zhanghe Irrigation System, the uncertainty over water resources was addressed by inserting small reservoirs into an integrated irrigation network, whereby ponds are interconnected with canals (Mushtaq et al. 2007). As a result ponds are fed by rainfall for 58% and the rest is completed with water from canals and other ponds, waving fears over insecurity. Users also typically have access to 2 ponds, further reducing drought risks. Clearly this example of alternate irrigation strategy can not be transposed to many arid areas however it highlights that ponds may only be adapted as a suitable irrigation strategy when it supports or is supported by another water supply strategy (groundwater, springs, cisterns...). In the Merguellil, those who have succeeded in intensifying their irrigated activity have done so because they were also able to provide this alternate water supply, notably by purchasing cisterns or using their wells. For farmers the solution is a borehole which epitomises in their eyes a secure, reliable, constant water supply “it’s OK for a garden, for agriculture we need a borehole”. In any case, reservoirs by capturing silt have a dedicated lifespan and must not be perceived as a perennial solution. 30 years on, future strategies are required to support the seeds sown by these water and soil conservation programmes.
224
The Merguellil upper catchment: view towards the South East (with the Djebel Trozza in the distance)
225
Part V
Appendices
226
Chapter 6
Supplementary research on rainfall and runoff at the basin scale In parallel to the work presented in this PhD, research focussed at the whole basin scale on investigating trends in annual rainfall and annual runoff in the Merguellil upper catchment. It notably sought to understand the high variability in runoff coefficients influenced by rainfall intensities and surface area, antecedent soil moisture and land cover, including small reservoirs. For reasons of overall coherence, considering the different scope of this work, this research was not included within the main body of the final PhD manuscript. Nevertheless, as it provides some useful overview on rainfall and runoff processes in this catchment, relevant at the local subcatchment scale, this was included here for convenience. The work is published in the Hydrological Sciences Journal as detailed in the reference Ogilvie et al. (2016).
227
Hydrological Sciences Journal
ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: http://www.tandfonline.com/loi/thsj20
Réponse hydrologique d’un bassin semi-aride aux événements pluviométriques et aménagements de versant (bassin du Merguellil, Tunisie centrale) Andrew Ogilvie, Patrick Le Goulven, Christian Leduc, Roger Calvez & Mark Mulligan To cite this article: Andrew Ogilvie, Patrick Le Goulven, Christian Leduc, Roger Calvez & Mark Mulligan (2016): Réponse hydrologique d’un bassin semi-aride aux événements pluviométriques et aménagements de versant (bassin du Merguellil, Tunisie centrale), Hydrological Sciences Journal, DOI: 10.1080/02626667.2014.934249 To link to this article: http://dx.doi.org/10.1080/02626667.2014.934249
Accepted author version posted online: 12 Jun 2014. Published online: 25 Jan 2016. Submit your article to this journal
Article views: 44
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=thsj20 Download by: [IRD Documentation]
Date: 27 January 2016, At: 01:35
228
HYDROLOGICAL SCIENCES JOURNAL – JOURNAL DES SCIENCES HYDROLOGIQUES, 2016 http://dx.doi.org/10.1080/02626667.2014.934249
Réponse hydrologique d’un bassin semi-aride aux événements pluviométriques et aménagements de versant (bassin du Merguellil, Tunisie centrale) Andrew Ogilviea,b, Patrick Le Goulvena, Christian Leduca, Roger Calveza and Mark Mulliganb Institut de Recherche pour le Développement, UMR G-eau, Montpellier Cedex 5, France; bDepartment of Geography, King’s College London,, Londres, WC2R 2LS, Royaume Uni
Downloaded by [IRD Documentation] at 01:35 27 January 2016
a
RÉSUMÉ
ARTICLE HISTORY
L’influence cumulée des aménagements de versants sur les écoulements en zones semi-arides demeure peu étudiée et comprise à l’échelle de grands bassins versants. En Tunisie centrale, nous étudions les variations de la réponse hydrologique du bassin du Merguellil à l’échelle annuelle et pour 114 événements entre 1989 et 2010. Sur cette période, les aménagements de conservation des eaux et des sols ont progressivement drainé de 5% à 30% des surfaces du bassin. L’analyse statistique révèle la forte variabilité des pluies et des débits annuels, mais ne distingue aucune tendance déficitaire. L’approche événementielle démontre que les variations des écoulements annuels sont liées aux apports générés par un nombre très limité (cinq à six par an en moyenne) d’épisodes pluvieux supérieurs à 15 mm. Le nombre très variable de ces évènements ainsi que les écarts importants observés sur leur coefficient de ruissellement (KR) respectif sont déterminants sur les apports annuels. Pour les événements de plus de 60 mm, les différences d’intensités pluviométriques, de couvert végétal et d’humidité du sol réduisent le ruissellement jusqu’à 80%. Toutefois, les KR d’averses de moins de 40 mm, sur des sols au couvert végétal et d’humidité semblables, ont également diminué de 45% après l’extension des aménagements de CES à la fin des années 1990.
Reçu le 22 novembre 2013 accepté le 29 mai 2014
Hydrological response of a semi-arid catchment to rainfall events and water and soil conservation works (Merguellil catchment, central Tunisia)
KEYWORDS
ABSTRACT
The cumulative influence of water and soil conservation works (WSCW) on runoff in large semi-arid catchments remains poorly understood. In the Merguellil basin, in central Tunisia, where WSCW now drain over 30% of the surface area, changes in the hydrological response were studied annually and over 114 rainfall events during the 1989–2010 period. At the annual scale, statistical analyses highlighted strong rainfall and runoff variability but did not indicate a downward trend. Annual variations in downstream flows were closely linked to the runoff generated by a small number of rainfall events over 15 mm (five to six per year). Changes in the number of these events but also large variations in their runoff coefficient (KR) had a determining influence over annual runoff. Differences in rainfall intensity, land cover and antecedent soil moisture reduced runoff from rainfall events over 60 mm by up to 80%. Nevertheless, the runoff coefficients of rainfall events below 40 mm, occurring in similar land cover and soil moisture conditions, also decreased by 45% following the increase in WSCW in the late 1990s.
Introduction Les aménagements de conservation des eaux et des sols (CES) connaissent depuis plusieurs décennies une forte expansion, favorisée par le soutien de projets gouvernementaux et internationaux en zones semi-arides, notamment au Brésil (Burte 2008), au Mexique (Avalos 2004), en Inde (Bouma et al. 2011), et en Afrique du Nord et subsaharienne (Talineau et al. 1994, Sawunyama et al. 2006, Nyssen et al. 2010). La construction de banquettes doit freiner et retenir l’érosion des versants tandis que celle de retenues collinaires doit stocker les sédiments et réduire l’envasement d’ouvrages stratégiques en aval. En parallèle, ces aménagements permettent de mobiliser des ressources en eau pour les usagers à l’amont, et CONTACT Andrew Ogilvie
EDITEUR
Z.W. Kundzewicz EDITEUR ASSOCIÉ
C. Cudennec MOTS CLEFS
conservation des eaux et des sols; coefficient de ruissellement; variabilité pluviométrique; influence anthropique; bassin semiaride; approche événementielle water and soil conservation works; runoff coefficients; rainfall variability; human influence; semi-arid basin; event scale
peuvent parfois favoriser la recharge de nappes. Ils viennent en complément des techniques traditionnelles de gestion et de conservation des eaux et des sols connues dès l’antiquité (Roose 1991). A l’échelle locale, de nombreuses études réalisées à partir de mesures hydrologiques et d’analyses géochimiques ont permis d’évaluer le bilan hydrique et l’influence des retenues collinaires et des banquettes, notamment en Tunisie (Gay 2004, Grunberger et al. 2004, Al Ali et al. 2008), Inde (Li et Gowing 2005), Ethiopie (Nyssen et al. 2010) et Palestine (AlSeekh et Mohammad 2009). Ces ouvrages ont dans leur environnement immédiat des conséquences sur la redistribution des ressources hydriques, réduisant jusque 80% les écoulements sur des bassins de quelques dizaines d’hectares
[email protected]
© 2016 IAHS
229
Downloaded by [IRD Documentation] at 01:35 27 January 2016
2
A. OGILVIE ET AL.
(Al-Seekh and Mohammad 2009, Nyssen et al. 2010). Dans des bassins versants de plus de 100 km2 leur influence cumulée demeure cependant peu étudiée et comprise (Kongo et Jewitt 2006, Lacombe et al. 2008). La complexité de l’analyse réside dans la difficulté à différencier l’effet de ces ouvrages des autres processus simultanés, interférents, eux mêmes variables dans le temps et l’espace (pluie, occupation du sol, etc.) (Cudennec et al. 2004, 2007) et qui influencent aussi la réponse hydrologique d’un bassin versant. Parmi ces facteurs on distingue notamment la pluviométrie, l’humidité initiale du sol, l’occupation du sol, les états de sols et du couvert végétal et les aménagements (prélèvements, barrages, etc.) (Lacombe et al. 2008). L’étude de leur influence repose principalement sur des approches de modélisation du bassin entier (Cudennec et al. 2004, Kingumbi et al. 2007, Lacombe et al. 2008) ou l’étude des variations des coefficients de ruissellement (He et al. 2003, Lacombe 2007, Gao et al. 2011), et s’apparente notamment à l’étude de l’influence à cette échelle de changements d’occupation du sols (PeñaArancibia et al. 2012). Sur la partie centrale du bassin du fleuve Jaune, des études montrent une réduction des écoulements variant de 6% (Gao et al. 2011) à près de 50% (He et al. 2003) suite aux aménagements de CES. De la même manière, dans le bassin du Merguellil, en Tunisie Centrale, des études basées sur différentes méthodes ont évalué l’influence cumulée des aménagements de CES entre 1% et 50% sur les même périodes (Dridi 2000, Kingumbi 2006, Lacombe 2007). Ces résultats témoignent de la complexité à représenter et évaluer l’influence des ouvrages de CES à cette échelle. Cette étude sur le bassin du Merguellil vise d’abord au travers d’une analyse statistique à identifier la présence de tendances dans les chroniques de débits et de pluies sur plus de 45 ans. Les facteurs climatiques permettant d’expliquer les
variations des écoulements sont étudiés aux échelles annuelle puis événementielle, afin de mieux tenir compte du régime pluviométrique en zone semi-aride. L’approche événementielle se concentre sur 114 épisodes pluvieux sur la période 1989–2010, concomitante à l’aménagement des CES. L’analyse multicritères des coefficients de ruissellement des événements traités permet d’étudier l’influence du cumul et de l’intensité pluviométrique et des états de sol (couverture du sol et humidité antérieure) dans l’objectif de distinguer leurs effets sur la fonction de production d’un bassin versant. L’influence des CES est recherchée dans les éventuelles anomalies de la réponse hydrologique du bassin versant.
Matériels et méthodes Site d’étude et aménagements Le site d’étude est le bassin amont du Merguellil en Tunisie Centrale, d’une superficie de 1183 km2 et délimité à l’aval par le barrage El Haouareb construit en 1989 (Fig. 1). L’altitude varie entre 200 et 1200 m, pour une altitude moyenne de 500 m. Situé en zone semi-aride, les précipitations annuelles sont faibles (265 mm dans la plaine et 515 mm à l’amont) (Leduc et al. 2007) et caractérisées par des évènements intenses, surtout au printemps et à l’automne, qui entraînent la crue des oueds. Les écoulements intermittents de l’Oued Merguellil approvisionnent le barrage El Haouareb qui, lui-même, alimente la nappe de la plaine de Kairouan à l’aval fortement exploitée pour l’agriculture irriguée et l’alimentation en eau potable. La température moyenne est de 19,2°C (10,7°C en janvier et 28,6°C en août) (Zribi et al. 2011) et l’évapotranspiration potentielle atteint 1600 mm/an. Les ouvrages de CES ont été réalisés dès les années 1960, dans le cadre d’un grand programme d’aménagements de
Figure 1. Bassin amont du Merguellil et localisation des stations de mesure et aménagements de CES.
230
Downloaded by [IRD Documentation] at 01:35 27 January 2016
HYDROLOGICAL SCIENCES JOURNAL – JOURNAL DES SCIENCES HYDROLOGIQUES
CES le long de la Dorsale tunisienne (chaîne montagneuse NE −SW) prévoyant la construction de plus de 850 retenues collinaires et l’aménagement de 1 M d’ha par des banquettes anti-érosives (Nasri 2007). Dans le bassin du Merguellil, ces structures contrôlent actuellement plus de 30% de la superficie du bassin versant. On dénombre 50 retenues qui drainent 20% de la superficie du bassin, pour une capacité cumulée de 5,8 hm3, dont 30 construites après 1989 lorsque le développement des CES a connu une rapide expansion (Lacombe et al. 2008). Les banquettes, d’une capacité linéaire moyenne de 85 mm par mètre linéaire couvrent 23% du bassin (Dridi 2000). Ces aménagements avaient été recensés à partir d’images satellites SPOT 2,5 m et 10 m de 2003 et des données de la Direction Générale des Ressources en Eau (Ben Mansour 2000, Lacombe 2007). Des visites de terrain ont permis la mise à jour de ces inventaires, et la prise en compte de l’envasement actuel de retenues. Ces ouvrages perdent chaque année une part de leur capacité par envasement; en moyenne 4,6% par an pour les réservoirs (Ben Mammou et Louati 2007) et 3% par an pour les banquettes (Baccari et al. 2008), mais ces valeurs varient fortement dans le temps et l’espace en fonction du nombre et de l’importance des épisodes pluvieux subis. La céréaliculture et l’arboriculture dominent les usages agricoles dans le bassin (Dridi 2000): 50% du bassin est cultivé en céréales pérennes (blé), oliviers et amandiers. Le reste des superficies est composé de terres de parcours (30%), forêt (19% surtout à l’amont) et 1% de bâtis (Lacombe et al. 2008). Les terres irriguées sont estimées à 3500 ha, et exploitent quasi exclusivement les ressources souterraines de plusieurs nappes en accès libre (Le Goulven et al. 2009). On recense quatre entités hydrogéologiques principales interconnectées, couvrant environ 600 km2 et contenues dans des formations géologiques qui vont du Trias au Quaternaire (Kingumbi 2006). Le suivi piézométrique depuis plus de 40 ans a permis de mettre en évidence une baisse piézométrique de 0,25 m à 1 m par an sur la période 1978–1989 dans la nappe oligocène de Bou Hafna, conséquence des nombreux prélèvements (Kingumbi 2006). Les relations eaux de surface-eaux souterraines dans le bassin amont sont peu connues mais des échanges dans les deux directions ont été observés: recharge des nappes par les crues et soutien des nappes aux débits de base. L’écoulement de base est estimé à environ 1/8 des écoulements annuels au barrage (Leduc et al. 2007). Données et analyses Ce site bénéficie d’un vaste réseau de mesures hydrologiques et climatiques depuis plus de 45 ans, géré et entretenu par la Direction Générale des Ressources en Eau et les Commissariats Régionaux au Développement Agricole (Ministère de l’Agriculture tunisien). Donnés pluviométriques Le réseau de suivi pluviométrique est composé de plus de 80 pluviomètres dont 13 pluviographes à augets basculeurs situés à l’intérieur et aux abords du bassin. Les chroniques sont
3
lacunaires et certaines stations ne sont plus en activité, mais elles remontent jusque 1888 pour la plus ancienne (Makhtar). Pour notre étude pluviométrique, nous avons exploité les données journalières sur la période 1902–2010 durant laquelle entre trois et 80 stations disposent de données. Ces données ont été nettoyées des erreurs systématiques, notamment de calibrages, codages et retranscription, puis contrôlées par la méthode des doubles masses (double cumul) (Brunet-Moret 1971). Cette méthode se base sur le principe de pseudo-proportionnalité entre les données de deux pluviomètres voisins pour vérifier leur homogénéité (Kingumbi 2006). Nous procédons ensuite à l’interpolation spatiale des précipitations sur la surface du bassin, par la méthode de pondération inverse à la distance. Cette méthode robuste et répandue possède l’avantage sur le krigeage que les facteurs de pondération sont invariants au cours du temps, ce qui réduit le temps de calcul lors de l’estimation de précipitations journalières sur 50 années. Le module SPATIAL du logiciel HydrAccess a été utilisé ici. Malgré le caractère anisotrope du milieu, la répartition homogène des postes pluviométriques intègre implicitement l’influence du gradient altitudinal sur les précipitations (Feki et al. 2012) pour la période 1960–2010. Sur la première partie du 20e siècle, le faible nombre de stations pluviométriques en activité et leur distribution hétérogène pourrait introduire un biais dans l’interpolation. L’étude des chroniques avant 1960 a donc été réalisée sur les 18 stations en activité et non sur la pluie interpolée sur le bassin entier. Ceci permet également de mettre en avant les différences et extrêmes locaux qui sont lissés par l’interpolation. Les chroniques de données ont été étudiées à partir du calcul de statistiques (moyennes, écart type, données centrées réduites) puis soumises à des tests de détection de rupture dont les tests de Pettitt et Buishand (Pettitt 1979, Lubes-Niel et al. 1998) et la segmentation de Hubert (Hubert et al. 1989). Les tests non paramétriques de Mann-Kendall et de la pente de Sen (Yue et al. 2002) sont utilisés en complément pour déceler et quantifier la pente d’une éventuelle tendance linéaire sur les données climatiques. Ces tests statistiques ne sont pas utilisés pour détecter les erreurs de mesures car ils amalgament erreurs systématiques et variations interannuelles de la pluie. La méthode du vecteur régional pourrait être appliquée ici pour réduire les problèmes d’homogénéisation des données. Les tests de tendance sont réalisés sur le cumul de pluie annuel, sur le nombre de jours de pluie (>2 mm), et sur le nombre d’événements intenses >20 mm/jour et >30 mm/jour. Les logiciels Khronostat (IRD 1998) et Makesens (Salmi et al. 2002) ont été exploités pour effectuer ces tests. Donnés hydrologiques Pour des raisons logistiques, notamment la largeur du cours d’eau et les apports souterrains, aucune station hydrométrique ne mesure les débits à l’exutoire du bassin amont, qui correspond à l’entrée du barrage El Haouareb. Les stations les plus proches en activité sont celles de Haffouz qui collecte les écoulements sur 675 km2 et qui peut être combinée avec les données de la station de Zebbes (sous-
231
Downloaded by [IRD Documentation] at 01:35 27 January 2016
A. OGILVIE ET AL.
bassin de l’Oued Zebbes, 180 km2). Cependant ces deux stations ne permettent pas de prendre en compte les apports des oueds Ben Zitoune et Hammam, petits bassins versants de 90 km2 situés à l’aval captant les écoulements sur le Mont Trozza. De plus, ces stations souffrent de problèmes de fiabilité et de représentativité, à cause de l’absence de seuils stables et des faibles écoulements qui passent à coté des stations de mesure. Les écoulements à l’exutoire ont donc été calculés par un bilan hydrique au barrage El Haouareb à partir des cotes journalières, des mesures de pluie, d’évaporation et des estimations sur les éventuels prélèvements, vidanges et lâchers. Les cotes sont converties en surface et volume à l’aide de courbes d’étalonnage qui ont été revues suite aux nivellements successifs de 1989, 1994, 1997 et 2006. L’infiltration journalière varie entre 5 et 12 mm, en fonction de la cote du barrage mais également du colmatage progressif du fond de la retenue (Alazard et al. 2011). Malgré des améliorations successives dans l’estimation des flux du bilan du barrage (Kingumbi 2006, Alazard et al. 2011), l’estimation des apports par clôture du bilan introduit plusieurs sources d’erreur, notamment sur la courbe d’étalonnage et les incertitudes sur les autres flux (évaporation et échanges souterrains). Les apports ont été calculés depuis la mise en eau du barrage en 1989 jusqu’à 2010. En l’absence de mesures au site d’El Haouareb avant 1989, les chroniques de la station d’Haffouz sur la période 1966 à 2010 (Bouzaiane et Laforgue 1986) ont été comparées pour étudier l’évolution des apports sur le plus long terme. Les données de Haffouz sont en partie corrélées avec les données d’El Haouareb (R2 = 0,6). En complément, les quelques années d’observations sur la période 1974–1982 à Sidi Boujdaria, 4,5 km à l’amont du barrage, ont été comparées avec les valeurs estimées par Nazoumou (2002) à la même station sur 1989–1998. L’année hydrologique qui s’étend de septembre à août a été utilisée pour l’étude annuelle. Les données hydrométriques ont été soumises aux mêmes tests statistiques que les données de pluie. Coefficients de ruissellement (KR) A partir des données d’écoulement à l’exutoire et des pluies interpolées sur le bassin, nous avons calculé la lame d’eau ruisselée (mm) et la hauteur d’eau précipitée (mm) sur ce bassin de 1183 km2. Les coefficients de ruissellement sont calculés par le rapport entre la lame d’eau ruisselée et la hauteur d’eau précipitée et exprimés ici en pourcentage. Après une analyse des corrélations entre apports et pluies aux pas de temps annuel et mensuel, l’étude s’est centrée sur 114 épisodes pluvieux de plus de 15 mm recensés sur la période 1989–2010. Les événements de faible importance ont été écartés car, d’une part, les erreurs sur la courbe d’étalonnage, l’évaporation et l’infiltration sont proportionnellement plus fortes sur ces petits apports et, d’autre part, les événements plus importants sont déterminants pour le cumul annuel des écoulements. Le seuil de 15 mm a été choisi après analyse de tous les événements sur la période 1989–2010 et correspond en moyenne à la pluviométrie minimale pour observer un écoulement notable (supérieur à 100 000 m3). La bonne corrélation (R2 = 0,95) obtenue entre les apports
35 Apports annuels par événement > 15 mm (hm3)
4
30 y = 0,84x – 1.90 R² = 0,95 25
20
15
10
5
0 0
10
20
30
40
Somme des apports annuels (hm3)
Figure 2. Relation entre les apports annuels au barrage El Haouareb et les apports par les événements de plus de 15 mm sur la période 1989–2010.
générés par les épisodes pluvieux de plus de 15 mm et les apports annuels (Fig. 2) indique que ces événements contribuent de façon déterminante aux écoulements annuels et justifie que l’étude se concentre sur ces 114 événements. Les pluies et débits pour les épisodes pluvieux de plus de 15 mm ont été cumulés sur les jours successifs où l’écoulement était supérieur aux débits de base, qui rejoignent ici rapidement des valeurs de l’ordre de 1 m3/s. Ceci permet de tenir compte du délai à cette échelle entre une pluie à l’amont et l’apport au barrage, et des difficultés à différencier par jour de pluie l’apport pour des événements qui s’étendent sur deux et trois jours consécutifs. Ces 114 événements ont été analysés et regroupés en plusieurs classes sur la base de quatre critères, sélectionnés pour leur impact (de 1er ou de 2e ordre) sur la réponse hydrologique d’un bassin versant semi-aride. Ces paramètres sont le cumul pluviométrique, l’intensité en 24 h, la couverture du sol et le niveau de saturation hydrique. D’autres variables influencent la genèse des écoulements, notamment l’étendue spatiale des pluies (superficie et localisation) et l’intensité horaire, mais peu de stations pluviométriques sur le bassin disposent de pluviographes pour la mesure des intensités. La couverture du sol a été évaluée ici sur la base de valeurs d’indice de végétation par différence normalisée, NDVI (Rouse et al. 1973) afin de différencier les événements de mai à octobre aux sols peu couverts et les événements de novembre à avril qui surviennent lorsque le couvert végétal est plus important. Les moyennes mensuelles de NDVI pour le bassin sur la période 1989–2006 ont été calculées à partir des données bimensuelles pour des pixels de 8 km (Tucker et al. 2005) fournies par le Global Inventory Modeling and Mapping Studies. Une indication du degré de saturation hydrique est fournie par la différence entre le cumul de pluie sur les 20 jours précédant l’événement et l’évapotranspiration potentielle sur la même période. Des mesures journalières d’évaporation sur un bac Colorado situé au barrage El Haouareb ont été utilisées pour le calcul
232
HYDROLOGICAL SCIENCES JOURNAL – JOURNAL DES SCIENCES HYDROLOGIQUES
de l’évapotranspiration potentielle. L’étude a ensuite porté sur l’influence de ces variables sur les variations des KR au travers de statistiques descriptives (moyenne, écart type, médiane, etc), tests de corrélations et régressions pas à pas pour différentes classes d’événements. Les corrélations sont évaluées à l’aide du coefficient de Spearman (rs), qui permet de détecter tout type de relation monotone (linéaire, puissance, exponentielle, etc.). La régression pas à pas est une méthode itérative permettant d’identifier les variables explicatives pour modeler les variations de KR. Les disparités interannuelles des apports au barrage ont ensuite été interprétées sur la base du nombre d’événements et de ces paramètres. Enfin, l’influence des ouvrages de CES a été recherchée parmi les anomalies de KR pour des classes d’événements comparables.
Evolution des écoulements annuels Sur la période 1989–2010, les apports annuels au barrage El Haouareb se caractérisent par une alternance d’années excédentaires et déficitaires (Fig. 3) par rapport à la moyenne des apports annuels de 16,5 hm3. L’écart type de 9,4 hm3/an et le coefficient de variation de 0,6 illustrent cette forte variabilité interannuelle. Cinq années déficitaires dont 4 consécutives sont recensées entre 1996–1997 et 2001–2002, mais les tests statistiques de Pettitt, Buishand, Hubert, MannKendall et Sen ne détectent aucune rupture ou tendance dans l’évolution des écoulements annuels. Sur la période 1989–1998, la moyenne interannuelle calculée ici est 10% inférieure aux résultats précédents (Kingumbi 2006, Lacombe 2007) due à une différence de méthode, notamment l’estimation des prélèvements et de l’infiltration. Les années déficitaires entre 1998 et 2002 ont également réduit de 15% supplémentaires la moyenne interannuelle sur 1989–2010 par rapport aux calculs précédents sur 1989–1998. Sur la période 1966–2010, les chroniques de la station de Haffouz confirment la forte variabilité interannuelle, avec une moyenne de 21,5 hm3 et un écart type de 22,5 hm3. L’ampleur de cet écart-type peut en partie s’expliquer par les problèmes significatifs de fiabilité à cette station hydrométrique et des
changements d’instrumentation. Sur la période 1989–2010 la moyenne interannuelle atteint 19,5 hm3, et est supérieure ou inférieure à la moyenne des apports sur la période antérieure à 1989 selon que l’on prenne en compte l’année exceptionnelle 1969–1970 et les années fortement déficitaires de 1966 à 1969. Ainsi sur 1969–1989 la moyenne atteint 26,4 hm3 tandis que sur 1966–1989 sans prendre en compte l’année 1969–1970 la moyenne est de 18,8 hm3. Ces résultats démontrent la difficulté à comparer les tendances entre deux périodes dans des conditions d’extrême variabilité interannuelle. Les tests statistiques appliqués ne démontrent toutefois aucune tendance sur les écoulements annuels sur le long terme. Des études précédentes (Dridi 2000, Kingumbi 2006) avaient identifié une baisse de 40% sur le module annuel 1989–1998 au barrage El Haouareb par rapport aux chroniques antérieures (32 hm3/an pour 1974–1982) mesurées à Sidi Boujdaria, 4,5 km à l’amont (Bouzaiane and Laforgue 1986). L’absence de mesures aux deux stations sur les mêmes périodes ainsi que la forte infiltration entre ces deux points de mesure complique cependant toute comparaison. L’évaluation par Nazoumou (2002) des écoulements à Sidi Boujdaria sur 1989–1998 (35 hm3/an) à partir d’estimation d’infiltration tendrait même à montrer une augmentation des écoulements annuels de 8% après 1989.
Influence climatique sur les apports annuels Evolution des précipitations annuelles L’analyse pluviométrique a révélé une forte variabilité naturelle (Fig. 4), avec une moyenne de pluie annuelle sur le bassin amont de 329 mm pour la période de 1960–2010, et un écart type important de 131 mm, soit un coefficient de variation proche de 0,4. La pluviométrie de 1969–1970 fut exceptionnelle et atteignit 1000 mm, suite à de violents orages en septembre et octobre 1969 (Camus 1985). Une succession d’années déficitaires est recensée entre 1976–1977 et 1988–1989, notamment liée à la réduction des événements de plus de 30 mm (Kingumbi et al. 2005). Cependant les tests statistiques de Pettitt, Buishand, Hubert, Mann Kendall et Sen indiquent avec 99% de confiance qu’aucune tendance déficitaire sur la pluviométrie moyenne n’est observée. Les
125
Apports annuels à Haffouz
Apports annuels au barrage El Haouareb
100
Moyenne interannuelle au barrage El Haouareb Apports annuels (hm3)
Downloaded by [IRD Documentation] at 01:35 27 January 2016
Résultats et discussion
75
50
25
0 1969–1970
1979–1980
5
1989–1990
Figure 3. Apports annuels au barrage El Haouareb (1989–2010) et à la station d’Haffouz (1966–2010).
233
1999–2000
2009–2010
6
A. OGILVIE ET AL. 6 5 4
écart type
3 2 1 0 –1 –2 1960–1961
1965–1966
1970–1971
1975–1976
1980–1981
1985–1986
1990–1991
1995–1996
2000–2001
2005–2006
indiquent une baisse du cumul de pluie de 25% sur 1915–2010 et 1929–2010 respectivement. Aucune tendance ne fut observée sur les nombres de jours de pluie et d’événements >20 mm/jour et >30 mm/jour.
données récentes indiquent une période excédentaire sur 5 années consécutives de 2002–2003 à 2006–2007 et confortent les résultats antérieurs qui concluaient sur une absence de baisse des précipitations sur le bassin et à l’échelle de la Tunisie (Sakiss et al. 1994, Zahar 1997, Benzarti et Habaieb 2001, Kingumbi et al. 2005, Lacombe 2007, Slimani et al. 2007, Proaño 2012). L’analyse par les mêmes tests statistiques de l’évolution du nombre d’épisodes pluvieux de plus de 20 et 30 mm par jour sur les pluies interpolées dans le bassin a aussi révélé l’absence de tendance statistiquement significative, en accord avec les études précédentes (Kingumbi et al. 2005, Proaño 2012). Sur une période plus longue (1902–2010) la moyenne interannuelle atteint 367 mm et l’écart type 140 mm, confirmant que la période étudiée (1960–2010) est représentative du régime pluviométrique observé sur le long terme. La moyenne est supérieure de 10% à celle sur 1960–2010 mais ceci résulte du biais introduit par le faible nombre de stations en activité au début du 20e siècle et leur surreprésentation dans les parties hautes et plus pluvieuses du bassin. L’étude des tendances sur les 18 stations aux chroniques les plus longues n’indique en effet pour 16 stations aucune rupture sur la moyenne pluviométrique. Toutefois, on note que deux stations (Ben Nessim et Garrat) situées au Sud du bassin
Relation entre précipitations annuelles et écoulements annuels La comparaison de l’évolution des écoulements à l’exutoire du bassin amont du Merguellil et des valeurs de précipitations annuelles interpolées sur le bassin a révélé une faible relation entre ces deux variables (Fig. 5). La légère corrélation obtenue (R2 = 0,66) indique une tendance à l’augmentation des écoulements les années excédentaires, mais l’écoulement n’est pas directement lié au volume annuel de pluie. Une corrélation faible entre ces deux variables à cette échelle de temps et d’espace n’est pas surprenante dans une zone semiaride où les écoulements sont événementiels et est cohérente avec les résultats antérieurs, qui concluent que les variations du cumul pluviométrique ne peuvent à elles seules expliquer la diminution des ressources hydriques (Kingumbi 2006, Leduc et al. 2007). A l’échelle mensuelle, la corrélation bien que supérieure, témoigne également des fortes disparités
40
25 R² = 0,66
R² = 0,67
35 20 Apports mensuels (hm3)
30 Apports annuels (hm3)
Downloaded by [IRD Documentation] at 01:35 27 January 2016
Figure 4. Valeurs centrées réduites des pluies annuelles sur le bassin amont du Merguellil sur 1960–2010 (nombre d’écart-types par rapport à la moyenne pluviométrique interannuelle sur la période 1960–2010).
25 20 15
15
10
10 5 5 0
0 0
100
200
300
400
500
600
0
Pluies annuelles (mm)
50
100
150
200
Pluies mensuelles (mm)
Figure 5. Relations entre les pluies et les apports au barrage El Haouareb pour la période 1989–2010 aux pas de temps annuel et mensuel.
234
HYDROLOGICAL SCIENCES JOURNAL – JOURNAL DES SCIENCES HYDROLOGIQUES
La corrélation entre la pluviométrie et les apports au barrage au pas de temps annuel mais également mensuel ne permettant pas d’expliquer les variations interannuelles des écoulements, l’analyse s’est focalisée sur les événements pluvieux de plus de 15 mm. Ceux-ci contribuent à plus de deux tiers en moyenne des débits annuels (Fig. 2) et la bonne corrélation (R2 = 0,95) entre les apports générés par les épisodes pluvieux de plus de 15 mm et les apports annuels indique que ces événements sont déterminants sur le module annuel. L’étude des 114 événements a révélé une forte hétérogénéité de leur cumul pluviométrique et de leur nombre qui oscille entre 1 (2007 et 2009) et 10 (1989) par an. Sur la période 1989–2010, on dénombre en moyenne seulement 5,4 épisodes par an de plus de 15 mm. Sur les années excédentaires, un seul événement extrême peut même générer un apport dépassant 10 hm3, soit plus de 50% des apports moyens interannuels (Fig. 6). Ce caractère événementiel est cohérent avec les observations précédentes en zone semi-aride (Cudennec et al. 2005) et confirme la nécessité de raisonner sur les événements.
Influence des KR sur les apports annuels Les coefficients de ruissellement peuvent varier d’un facteur de 2 voire 3, selon qu’un événement au cumul semblable se produise sur un sol nu ou sur un sol couvert et peu saturé (Tableau 1). La lame de pluie, l’intensité, la couverture du sol et le cumul de pluie antérieur lors de l’épisode pluvieux ont tous un effet significatif sur la réponse hydrologique d’un bassin et
12
Apports (hm3)
12
Nombre d'épisodes pluvieux (> 15 mm) par an Apports (hm 3)
10
10
8
8
6
6
4
4
2
2
0
0
01/09/1989 01/09/1991 01/09/1993 01/09/1995 01/09/1997 01/09/1999 01/09/2001 01/09/2003 01/09/2005 01/09/2007 01/09/2009
Figure 6. Nombre d’épisodes pluvieux de plus de 15 mm sur la période 1989–2010 et leur apport au barrage El Haouareb.
2 Apports annuels
1.5 Nombre événements > 15 mm
1 0.5 Ecart type
Downloaded by [IRD Documentation] at 01:35 27 January 2016
Influence des événements
A l’inverse du cumul pluviométrique annuel, le nombre d’événements est un facteur déterminant du module annuel à El Haouareb, puisqu’une année d’apport excédentaire comportera en moyenne 7,1 événements de plus de 15 mm contre seulement 3,6 pour les années déficitaires. La quantité de ces événements annuels permet d’expliquer de nombreuses variations observées et notamment la baisse des apports durant les années 1997–2002 (Fig. 7). Néanmoins, la corrélation n’est pas satisfaisante sur certaines années, notamment 1994–1995 et 2006–2007. La prise en compte de la lame précipitée par les événements de plus de 15 mm ou par les événements plus violents (plus de 40 mm, 60 mm et 80 mm) n’a pas permis d’améliorer la corrélation. Il est nécessaire de prendre en compte les variations des coefficients de ruissellement de ces 5,4 événements annuels pour expliquer les variations des apports annuels.
Nombre d'épisodes pluvieux
observées sur les apports à El Haouareb pour un volume de pluie mensuel semblable (Fig. 5).
7
0 –0.5 –1 –1.5 –2 1989–1990
1994–1995
1999–2000
2004–2005
2009–2010
Figure 7. Valeurs centrées réduites des apports annuels au barrage El Haouareb et du nombre d’épisodes pluvieux de plus de 15 mm par an (1989–2010) (nombre d’écarts types par rapport à la moyenne interannuelle sur la période 1989–2010).
235
Downloaded by [IRD Documentation] at 01:35 27 January 2016
8
A. OGILVIE ET AL.
leurs influences peuvent se combiner au point qu’un épisode comparable puisse générer un apport extrêmement variable en fonction des circonstances dans lesquels il se produit. Les variations des apports sur toute la période 1989–2010 peuvent ainsi s’expliquer par le nombre d’événements annuels et les conditions de chaque événement. Des années avec uniquement deux à trois événements ont typiquement généré un apport insuffisant, à moins qu’un ou plusieurs événements ne coïncident avec des conditions favorisant un fort ruissellement. Ainsi en 1994–1995, les trois épisodes de plus de 15 mm sont intervenus sur des sols apparemment peu couverts et après d’autres averses, permettant à ces trois événements relativement faibles d’apporter plus de 16 hm3. De même en février 2005 un événement de 100 mm amena 11 hm3 tandis qu’en décembre 2003 un événement de 85 mm n’amena que 3,8 hm3 à cause d’une plus faible intensité journalière et des sols moins saturés. Sur la période fortement déficitaire de 1996–2002, en plus d’un nombre restreint d’événements, les épisodes importants ont eu lieu sur des sols couverts, peu humides et donc avec des KR naturellement bas, tandis que les événements de fin d’été potentiellement ruisselants ont été relativement faibles entrainant peu d’écoulement.
Variabilité des coefficients de ruissellement Paramètres influant le KR des événements Les KR calculés pour 114 événements ont été regroupés en plusieurs catégories sur la base des quatre paramètres sélectionnés pour étudier leur influence sur la réponse hydrologique du bassin versant (Tableau 1). Les KR moyens augmentent de 4,6% à 7,15% lorsque le cumul de pluie croît de moins de 40 mm à plus de 80 mm par épisode. La réponse hydrologique du bassin varie de façon semblable selon que l’intensité journalière est supérieure ou inférieure à 30 mm. Les KR sont également 50% plus forts pour des événements d’été que d’hiver lorsque les sols agricoles sont nus, n’ont pas encore été labourés ou semés et donc favorisent le ruissellement. Les sols d’hiver ont eux une meilleure couverture végétale qui ralentit et réduit le ruissellement. L’intensité horaire des événements d’été est souvent aussi plus forte, ce qui contribue à augmenter le ruissellement (Chargui et al. 2013). Finalement les KR moyens augmentent
Tableau 2. Coefficients de corrélations de Spearman entre les valeurs de KR et quatre paramètres étudiés pour des événements au cumul de pluie entre 15 et 100 mm.
KR tous événements cumul >80 mm cumul 60–80 mm cumul 40–60 mm cumul 15–40 mm
Intensité des pluies 0,153 0,657 −0,048 −0,042 −0,036
Cumul de Couverture pluie du sol 0,180 −0,318 −0,086 0,091 −0,048 −0,638 0,034 0,028 −0,043 −0,444
Saturation hydrique 0,249 0,886 0,357 −0,081 0,251
également de 3,6% à 6,4% selon le degré de saturation du système hydrologique, confirmant que la capacité à ruisseler est plus importante lorsque la pluie antérieure est élevée. La régression pas à pas effectuée avec les quatre paramètres sur les variations des KR indique que la couverture du sol, le niveau de saturation hydrique, et l’intensité pluviométrique journalière sont des variables explicatives du modèle, mais pas le volume total de l’événement. Leur corrélation avec KR demeure toutefois faible (Tableau 2). Les tendances observées sur les valeurs de KR en fonction de ces quatre critères sont cohérentes avec la réponse hydrologique attendue, mais l’on observe effectivement des écarts types importants et de grandes variations entre les KR minimum et maximum au sein de ces classes (Tableau 1). Ces écarts, présents dans d’autres bassins semi-arides (Martinez-Mena et al. 1998), indiquent que la combinaison de ces paramètres importants, ne sont pas les seuls à influencer fortement le KR pour un événement donné. Interaction des facteurs sur les KR Sur 6 épisodes pluvieux au cumul supérieur à 80 mm, les KR s’échelonnent entre 2,9% et 12,5%. Les plus forts KR sont observés pour les événements aux plus fortes intensités en 24 h, qui varient de 30 à 87 mm/jour selon les épisodes. La régression pas à pas effectuée sur cette classe d’événement a confirmé la pluie journalière comme étant la variable explicative dominante du modèle KR et le coefficient de Spearman de cette relation linéaire atteint 0,66 (Fig. 8). La visualisation des isohyètes journalières (Fig. 9) témoigne de l’importance de la présence et de l’étendue spatiale des fortes intensités pluviométriques sur les KR élevés. Tous ces épisodes de plus de 80 mm étant des phénomènes d’hiver sur des sols couverts,
Tableau 1. Variations des KR (%) pour différentes catégories d’épisodes pluvieux. Type d’événement Cumul de pluie
KR moyen
KR médian
KR min
KR max
Nombre d’événements
Ecart type
15–40 mm 40–80 mm >80 mm
4,60 5,65 7,15
3,24 5,71 6,63
0,22 0,18 2,85
26,35 10,9 12,49
83 25 6
4,5 2,9 3,93
30 mm/j
4,53 7,62
3,38 5,92
0,18 2,09
20,45 26,34
98 16
3,77 5,43
Elevée (NDVI >0,2) Faible (NDVI 15 mm aux KR élevés. L’influence des ouvrages de CES a donc été recherchée sur des petites variations de KR pour des événements comparables. Pour les événements inférieurs à 40 mm, la moyenne des KR a diminué de 40% après 1996, en accord avec les résultats obtenus par Lacombe (2007). Les KR moyens sont demeurés faibles, autour de 3,6%, sur la période 2003–2010 (Fig. 12). Pour affiner ce résultat et prendre en compte certains paramètres qui peuvent en plus des ouvrages de CES influencer le KR, nous avons comparé l’évolution des KR pour différentes sous-classes d’événements. Les KR ont diminué après 1996 de plus de 50% pour des épisodes sur des sols au couvert végétal faible et pour ceux sur des sols couverts, peu humides. Sur des sols couverts et humides, les KR ont baissé de 10% entre les périodes étudiées. Sur les épisodes pluvieux de plus de 40 mm, les KR moyens ont baissé de 20% après 2003. Néanmoins, le nombre limité d’épisodes dans cette catégorie (31 sur 21 ans) ne permet pas de disposer d’un nombre significatif d’événements dans des conditions comparables (sur sols nus, couverts, saturés, etc.), et d’estimer dans quelles proportions l’influence de ces ouvrages, qui se confond avec l’influence des sols, réduit de façon supplémentaire le ruissellement.
La baisse des KR observée sur des événements inférieurs à 40 mm confirme qu’il y a une modification de la réponse hydrologique du bassin versant sur la période qui suit les aménagements de CES. Cette diminution étant constatée lorsque le cumul pluviométrique, le couvert végétal et le degré de saturation hydrique des événements étaient similaires, tend à démontrer que des variations sur ces facteurs ne peuvent avoir causé cette réduction. Inversement, l’absence de baisse des KR observée pour des épisodes où les sols sont couverts et saturés semble conforter l’hypothèse d’une influence des aménagements de CES. En effet lorsque le cumul de pluie antérieur est important, les ouvrages de CES seraient en partie remplis et leur influence sur le ruissellement serait logiquement moindre. La continuité des faibles KR observés après 2003 tendraient cependant à réfuter l’hypothèse d’une réduction au cours du temps de l’influence des aménagements de CES due à la dégradation (brèches et envasement) des ces ouvrages (Lacombe 2007). Comme les événements de moins de 40 mm contribuent à 60% en moyenne des apports annuels sur la période 1989–1996, la baisse de 40% observée après 1996 sur leur KR moyen entrainerait une diminution des écoulements annuels due aux ouvrages de CES de l’ordre de 25%. Ceci correspondrait à un ruissellement capté autour de 4 hm3/an, ce qui est cohérent avec la capacité totale des retenues collinaires estimée à près de 6 hm3. Lacombe (2007) avait identifié une baisse des écoulements de 28–32% sur le bassin suite à
Tous événements
Sols peu couverts
Sols couverts, secs ou peu humides
1989–1996 1996–2003 2003–2010
Sols couverts et humides
0
2
4
6
KR moyens (%)
Figure 12. Evolution des KR moyens pour des événements pluvieux de 15–40 mm.
239
8
10
Downloaded by [IRD Documentation] at 01:35 27 January 2016
12
A. OGILVIE ET AL.
l’augmentation des superficies contrôlées par les aménagements de CES de 5% à 26%. Considérant les éventuels autres facteurs (pluies, occupation du sol, etc.) stationnaires entre les périodes, il attribua cette réduction à la seule influence des aménagements de CES. Dridi (2000) en s’appuyant sur des statistiques de pluie, de coefficients de ruissellement et de la capacité moyenne des banquettes (85 mm) et des retenues, avait estimé que leur impact cumulé réduisait les débits annuels de 30%. Kingumbi (2006) en simulant l’absence de CES sur ce bassin, à l’aide d’un modèle couplé surface−souterrain distribué à base conceptuelle développé sur MODCOU, avait évalué la réduction des débits à uniquement 1%. Il émettait l’hypothèse que la baisse globale des écoulements serait due à une réduction des écoulements de base suite au rabattement du niveau des nappes environnantes. Son travail s’était heurté aux incertitudes résiduelles sur l’infiltration et l’évaporation ainsi qu’à la difficulté à extrapoler le fonctionnement des CES étudiés à d’autres CES implantés dans des sous-bassins aux profils pédologiques, géologiques et géomorphologiques variés. Un autre modèle développé sous SWAT (Abouabdillah 2010) n’avait pas permis d’affiner l’estimation de l’influence des aménagements de CES.
Conclusion L’analyse des fortes variations pluviométriques et hydrométriques dans ce bassin semi-aride par différents tests statistiques n’indique pas de rupture au cours des 45 dernières années. Cette étude a montré que les écoulements annuels ne sont pas corrélés avec le cumul pluviométrique mais sont liés aux apports générés par un nombre très limité d’épisodes pluvieux de plus de 15 mm (5,4 par an en moyenne). Le nombre très variable de ces épisodes, mais également les circonstances dans lesquelles ils se produisent, sont déterminants sur les apports annuels, illustrant l’importance d’une approche événementielle en région semi-aride. Les influences du cumul et de l’intensité pluviométrique, de l’état du couvert végétal du sol ainsi que du niveau de saturation du système hydrologique sont clairement démontrées sur les variations des coefficients de ruissellement de 114 événements étudiés. Les années d’apports déficitaires entre 1996–2002 sur le bassin du Merguellil ne sont pas liées à l’aménagement concomitant des ouvrages de CES comme pourrait sembler l’indiquer la baisse constatée sur les KR annuels. Ce déficit s’explique avant tout par un nombre réduit d’épisodes de plus de 15 mm par an et des circonstances défavorables, notamment sur des sols d’hiver, couverts, peu saturés et peu ruisselants. Cependant, l’analyse détaillée des 114 événements révèle une baisse de plus de 40% sur les KR pour des événements comparables suite à l’aménagement des ouvrages de CES. Ces ouvrages contrôlant près de 30% des superficies du bassin auraient donc également contribué à réduire les apports annuels d’environ 25% depuis 1996. Les difficultés à interpréter les variations des KR sur les épisodes pluvieux les plus faibles témoignent de la difficulté résiduelle à représenter à cette échelle spatiale le fonctionnement hydrologique d’un bassin semi-aride hétérogène. Une discrétisation dans un
modèle distribué ou semi distribué demeure nécessaire pour prendre en compte les disparités spatiales et temporelles des intensités pluviométriques et des états de sols qui influent fortement sur les fonctions de production.
Finance Les auteurs remercient le Commissariat Régional au Développement Agricole de Kairouan, Tunisie et la Direction Générale des Ressources en Eau de Tunis. Ces travaux ont été partiellement financés par les projets FP7 WASSERMed, ANR AMETHYST et SICMED Dyshyme.
Déclaration de divulgation Aucun conflit d'intérêts potentiel n’a été rapporté par le(s) auteur(s).
References Abouabdillah, A., 2010. Hydrological modeling in a data-poor Mediterranean catchment (Merguellil, Tunisia). Assessing scenarios of land management and climate change. Thesis (PhD). University of Tuscia, Italy. Al Ali, Y., et al., 2008. Water and sediment balances of a contour bench terracing system in a semi-arid cultivated zone (El Gouazine, central Tunisia). Hydrological Sciences Journal, 53 (4), 883–892. Alazard, M., et al., 2011. Estimating groundwater fluxes by hydrodynamic and geochemical approaches in a heterogeneous Mediterranean system (central Tunisia). IAHS Publication, 345, 253–258. Al-Seekh, S.H. and Mohammad, A.G., 2009. The effect of water harvesting techniques on runoff, sedimentation, and soil properties. Environmental Management, 44 (1), 37–45. Avalos, J.E., 2004. Modélisation hydrologique globale conceptuelle appliquée aux petits bassins versants en zone semi-aride du nordMexique. Revue des sciences de l’eau, 17(2), 195–212. Baccari, N., et al., 2008. Efficiency of contour benches, filling-in and silting-up of a hillside reservoir in a semi-arid climate in Tunisia. Comptes Rendus Geosciences, 340 (1), 38–48. Baccour, H., Slimani, M., and Cudennec, C., 2012. Structures spatiales de l’évapotranspiration de référence et des variables climatiques corrélées en Tunisie. Hydrological Sciences Journal, 57 (4), 818–829. Ben Mammou, A. and Louati, M., 2007. Évolution temporelle de l’envasement des retenues de barrages de Tunisie. Revue des sciences de l’eau, 20 (2), 201–210. Ben Mansour, H., 2000. Apport de la télédétection pour l’étude de la dynamique des aménagements de conservation des eaux et du sol (banquettes) et impact sur les apports du Merguellil (Tunisie Centrale). Thesis (DAA). Ecole Nationale Supérieure Agronomique de Rennes. Benzarti, Z. and Habaieb, H., 2001. Étude de la persistance de la sécheresse en Tunisie par utilisation des chaînes de Markov (1909– 1996). Sécheresse, 12 (4), 215–220. Bouma, J.A., Biggs, T.W., and Bouwer, L.M., 2011. The downstream externalities of harvesting rainwater in semi-arid watersheds: an Indian case study. Agricultural Water Management, 98 (7), 1162– 1170. Bouzaiane, S. and Laforgue, A., 1986. Monographie hydrologique des oueds Zeroud et Merguellil. Tunis: Direction Générale des Ressources en Eau, Orstom. Brunet-Moret, Y., 1971. Etude de l’homogénéité de séries chronologiques de précipitations annuelles par la méthode des doubles masses. Cahiers Orstom, Série Hydrologie, VIII, 4, 3–31. Burte, J., 2008. Les petits aquifères alluviaux dans les zones cristallines semi-arides: fonctionnement et stratégies de gestion de l’eau. Thesis (PhD). Université de Montpellier 2, France and Universidade Federal do Ceara, Brazil.
240
Downloaded by [IRD Documentation] at 01:35 27 January 2016
HYDROLOGICAL SCIENCES JOURNAL – JOURNAL DES SCIENCES HYDROLOGIQUES
Camus, H., 1985. Etude pluviométrique des bassins versants des oueds Zeroud et Merguellil. Tunis: Direction des Ressources en Eau, Orstom. Chargui, S., Slimani, M., and Cudennec, C., 2013. Statistical distribution of rainy events characteristics and instantaneous hyetographs generation (Merguellil watershed in central Tunisia). Arabian Journal of Geosciences, 6 (5), 1581–1590. Cudennec, C., Leduc, C., and Koutsoyiannis, D., 2007. Dryland hydrology in Mediterranean regions–a review. Hydrological Sciences Journal, 52 (6), 1077–1087. doi:10.1623/hysj.52.6.1077 Cudennec, C., Sarraza, M., and Nasri, S., 2004. Modelisation robuste de l’impact agrégé de retenues collinaires sur l’hydrologie de surface. Revue des sciences de l’eau, 17 (2), 181–194. Cudennec, C., Slimani, M., and Le Goulven, P., 2005. Accounting for sparsely observed rainfall space-time variability in a rainfall-runoff model of a semiarid Tunisian basin. Hydrological Sciences Journal, 50 (4), 617–630. Dridi, B., 2000. Impact des aménagements sur la disponibilité des eaux de surface dans le bassin versant du Merguellil. Thesis (PhD). Université Louis Pasteur (Strasbourg 1). Feki, H., Slimani, M., and Cudennec, C., 2012. Incorporating elevation in rainfall interpolation in Tunisia using geostatistical methods. Hydrological Sciences Journal, 57 (7), 1294–1314. Gao, P., et al., 2011. Changes in streamflow and sediment discharge and the response to human activities in the middle reaches of the Yellow River. Hydrology and Earth System Sciences, 15 (1), 1–10. Gay, D., 2004. Fonctionnement et bilan de retenues artificielles en Tunisie: approche hydrochimique et isotopique. Thesis (PhD), Université Paris XI, France. Grunberger, O., Montoroi, J., and Nasri, S., 2004. Quantification of water exchange between a hill reservoir and groundwater using hydrological and isotopic modelling (El Gouazine, Tunisia). Comptes Rendus Geosciences, 336 (16), 1453–1462. He, X., et al., 2003. Down-scale analysis for water scarcity in response to soil–water conservation on Loess Plateau of China. Agriculture, Ecosystems & Environment, 94 (3), 355–361. Hubert, P., Carbonnel, J.P. et Chaouche, A.,1989, Segmentation des séries hydrométéorologiques. Application à des séries de précipitations et de débits de l'Afrique de l'Ouest, Journal of Hydrology, 110, 349–367. IRD, 1998. Khronostat, version 1.01. Montpellier: IRD. Kingumbi, A., Bargaoui, Z., and Hubert, P., 2005. Investigation of the rainfall variability in central Tunisia. Hydrological Sciences Journal, 50 (3), 493–508. http://www.tandfonline.com/doi/abs/10.1623/hysj.50.3. 493.65027 Kingumbi, A., 2006. Modélisation hydrologique d’un bassin affecté par des changements d’occupation. Cas du Merguellil en Tunisie Centrale. Thesis (PhD). Université de Tunis El Manar, Ecole Nationale d’Ingénieurs de Tunis. Kingumbi, A., et al., 2007. Modélisation hydrologique stochastique d’un bassin affecté par des changements d’occupation: cas du Merguellil en Tunisie centrale. Hydrological Sciences Journal, 52 (6), 1232–1252. http://www.tandfonline.com/doi/abs/10.1623/hysj.52.6.1232 Kongo, V. and Jewitt, G., 2006. Preliminary investigation of catchment hydrology in response to agricultural water use innovations: A case study of the Potshini catchment – South Africa. Physics and Chemistry of the Earth, Parts A/B/C, 31 (15–16), 976–987. Lacombe, G., 2007. Evolution et usages de la ressource en eau dans un bassin versant aménagé semi-aride. Le cas du Merguellil en Tunisie centrale. Thesis (PhD). Université Montpellier II. Lacombe, G., Cappelaere, B., and Leduc, C., 2008. Hydrological impact of water and soil conservation works in the Merguellil catchment of central Tunisia. Journal of Hydrology, 359 (3–4), 210–224. Le Goulven, P., et al., 2009. Sharing scarce resources in Mediterranean river basin: wadi Merguellil in Central Tunisia. In: F. Molle and F. Wester, eds. River basin trajectories: societies, environments and development. Wallingford: CAB International, 147–170. Leduc, C., et al., 2007. Impacts of hydrological changes in the Mediterranean zone: environmental modifications and rural development in the Merguellil catchment, central Tunisia. Hydrological
13
Sciences Journal, 52 (6), 1162–1178. http://www.tandfonline.com/ doi/abs/10.1623/hysj.52.6.1162 Li, Q. and Gowing, J., 2005. A daily water balance modelling approach for simulating performance of tank-based irrigation systems. Water Resources Management, 19 (3), 211–231. Lubes-Niel, H., et al., 1998. Variabilité climatique et statistiques. Etude par simulation de la puissance et de la robustesse de quelques tests utilisés pour vérifier l’homogénéité de chroniques. Revue des sciences de l’eau, 11(3), 383–408. Martinez-Mena, M., Albaladejo, J., and Castillo, V.M., 1998. Factors influencing surface runo generation in a Mediterranean semi-arid environment: chicamo watershed, SE Spain. Hydrological Processes, 12, 741–754. Nasri, S., 2007. Caracteristiques et impacts hydrologiques de banquettes en cascade sur un versant semi-aride en Tunisie centrale. Hydrological Sciences Journal, 52 (6), 1134–1145. http://www.tandfonline.com/doi/ abs/10.1623/hysj.52.6.1134 Nazoumou, Y., 2002. Impact des barrages sur la recharge des nappes en zone aride: étude par modélisation numérique sur le cas de Kairouan (Tunisie centrale). Thesis (PhD). Université de Tunis El Manar. Nyssen, J., et al., 2010. Impact of soil and water conservation measures on catchment hydrological response-a case in north Ethiopia. Hydrological Processes, 24, 1880–1895. Peña-Arancibia, J.L., et al., 2012. Detecting changes in streamflow after partial woodland clearing in two large catchments in the seasonal tropics. Journal of Hydrology, 416–417, 60–71. Pettitt, A.N., 1979. A non-parametric approach to the change-point problem. Journal of the Royal Statistical Society. Series C (Applied Statistics), 28, 126–135. Proaño, D., 2012. Bilan offres-demandes sur le bassin versant du Merguellil à l’aide de la plateforme WEAP. Thesis (MSc). Université de Montpellier 2, France. Roose, E., 1991. Conservation des sols en zones méditerranéennes. Synthèse et proposition d’une nouvelle stategie de lutte antierosive: la GCES. Cah. ORSTOM Pédol, 26 (2), 145–181. Rouse, J.W., et al., 1973. Monitoring vegetation systems in the great plains with ERTS. In: Proceedings of the 3rd ERTS symposium, Washington, DC: Goddard Space Flight Center, 48–62. Sakiss, N., et al., 1994. La pluviométrie en Tunisie a-t-elle changé depuis 2000 ans? Tunis: Institut National de la Météorologie & Institut National Agronomique de Tunisie. Salmi, T., et al., 2002. Trends of annual values of atmospheric pollutants by the Mann-Kendall test and Sen’s slope estimates: the excel template application MAKESENS. Helsinki: Finnish Meteorological Institute. Sawunyama, T., Senzanje, A., and Mhizha, A., 2006. Estimation of small reservoir storage capacities in Limpopo River Basin using geographical information systems (GIS) and remotely sensed surface areas: case of Mzingwane catchment. Physics and Chemistry of the Earth, Parts A/B/C, 31 (15–16), 935–943. Slimani, M., Cudennec, C., and Feki, H., 2007. Structure du gradient pluviométrique de la transition Méditerranée—Sahara en Tunisie: déterminants géographiques et saisonnalité. Hydrological Sciences Journal, 52 (6), 1088–1102. http://www.tandfonline.com/doi/abs/10. 1623/hysj.52.6.1088 Talineau, J.C., Selmi, S., and Alaya, K., 1994. Lacs collinaires en Tunisie semi-aride. Sécheresse, 5 (4), 251–256. Tucker, C., et al., 2005. An extended AVHRR 8-km NDVI dataset compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing, 26 (20), 4485–4498. Yue, S., Pilon, P., and Cavadias, G., 2002. Power of the Mann–Kendall and Spearman’s rho tests for detecting monotonic trends in hydrological series. Journal of hydrology, 259 (1), 254–271. Zahar, Y., 1997. Eléments d’hydrologie pour l’aménagement: modélisation spatiale et temporelle des précipitations extrêmes et érosives en Tunisie centrale. Thesis (PhD). Université des lettres, des arts et sciences humaines, Tunis I Zribi, M., et al., 2011. Soil surface moisture estimation over a semi-arid region using ENVISAT ASAR radar data for soil evaporation evaluation. Hydrology and Earth System Sciences, 15 (1), 345–358.
241
Chapter 7
Appendix to chapter 2 Additional content 7.1
Inventory of lakes
Table 7.1.1: Lakes in the Merguellil upper catchment used in the study (coordinates in WGS84 UTM 32N)
Lake
Easting (m)
Northing (m)
Altitude Catchment Initial (m) area capacity (km ) (103 m3 )
Abda
561709
3945521
323
4.04
37
1969
Ain Faouar
558342
3947480
313
2.95
66
1994
Ain Smili 1
555944
3959320
436
5.00
130
1994
Ain Smili 2 (Ben Klikia)
555677
3959267
436
1.72
35
1994
Ben Houria (Tabal Jebbes)
554563
3946468
371
2.30
17
1970
Bouksab
556404
3953563
349
2.77
55
1985
Bouksab 2
556104
3953488
350
-
-
-
Bouksab 3
556214
3953475
350
-
-
-
Chauoba El Hamra
553234
3946533
413
2.03
120
1996
Dhahbi
562130
3942471
265
17.89
26
1973
El Guettar
553826
3956371
393
4.98
150
1992
El Gbatis
552765
3954945
365
2.02
106
1995
242
Construction year
Lake
Easting (m)
Northing (m)
Altitude Catchment Initial (m) area capacity (km ) (103 m3 )
El Habsa A
531972
3948970
678
3.96
50
1992
El Habsa B
532734
3950132
633
0.48
35
1994
El Hammam
554992
3940551
395
17.39
850
2002
El Kraroub
547666
3958564
492
20.51
1590
1995
El Mahbes
554563
3956593
387
8.80
180
1993
El Maiz
554340
3951316
356
12.35
500
1986
El Marrouki
566276
3942522
297
4.67
153
2002
El Marrouki 2 (Ghouil)
565774
3941200
278
0.63
56
1972
El Mazil (Hannidet or El Gassaa)
550815
3953046
395
5.38
104
2002
El Morra
535963
3949070
588
11.69
705
1992
En Mel
553276
3954941
373
17.02
1000
1997
Fidh Ali
553435
3951696
350
2.74
134
1991
Fidh Ben Nasseur
553364
3953479
368
1.82
47
1990
Fidh Mbarek
552147
3950472
390
1.01
53
1992
Garia S (Oued Thal)
544941
3959185
540
5.36
1500
2005
Garia 2
542797
3957287
525
0.85
19
1968
Garia 3
542906
3955901
514
0.99
25
1968
Hoshas
560139
3946798
306
7.90
130
1986
Hoshas amont (road dam)
561368
3947528
341
6.75
-
-
Mdinia
533437
3957444
632
22.83
1200
2006
Mdinia 2
534564
3960723
799
1.19
-
2012
O.Bouchaha A
533165
3961899
835
0.43
18
1990
O.Bouchaha B
533276
3961941
823
0.36
34
1990
O.Daoued 1
530248
3964088
838
2.94
95
1994
O.Daoued 2
530220
3963976
830
2.98
-
2012
O.Daoued 3
530399
3963846
832
8.28
350
2012
O.El Haffar
540830
3959756
700
0.42
30
1993
243
Construction year
Lake
Easting (m)
Northing (m)
Altitude Catchment Initial (m) area capacity (km ) (103 m3 )
O.Fadden Boras
538148
3958137
690
3.83
94
1992
O.Fidh Zitoune
533470
3952801
679
2.53
40
1993
O.Raouess
539105
3956551
578
1.75
18
1994
Salem Thabet
564980
3944050
307
3.09
63
1970
Sidi sofiane
560550
3959107
481
0.71
40
1994
Sidi sofiane 2
559920
3958030
458
-
-
-
Skhira 1 (Fidh Serwi S4)
526628
3962701
836
2.01
181
1967
Skhira 2 (El Hinchir S3)
525590
3962575
830
1.88
38
1967
Skhira 27
518419
3961101
884
2.90
-
1968
Skhira 3 (Oule M’rabet S2 or Gnaech)
525391
3961927
820
2.38
79
1967
Skhira 4 (El Mkebrta S18)
525044
3961168
811
4.76
160
1967
Skhira 5 (El Hfaya S12)
524808
3960621
807
4.21
60
1967
Skhira A (22, Fidh Grawa S20)
520051
3959737
864
2.96
72
1968
Skhira B (20, Fidh Smara S5)
519674
3960940
875
4.06
120
1968
Skhira C (18, Bou Haasine S1)
520983
3961335
820
2.06
52
1967
Skhira D
521930
3959775
818
0.44
-
-
Trozza sud (Ben Zitoune)
555193
3931410
368
3.71
50
1993
244
Construction year
Table 7.1.3: Lakes around the Merguellil catchment used in the study (coordinates in 32N UTM WGS84) Lake
Easting (m)
Northing (m)
Altitude Catchment Initial (m) area capacity (km ) (103 m3 )
Abdessadok
522600
3948499
795
3.15
-
1990
Baouejer
485385
3937536
852
1.61
-
1991
Bou Haya
461317
3873837
823
5.26
-
1994
Brahim Zaher
521147
3934407
528
6.64
-
1992
Dekikira
561489
3971357
406
3.31
219.1
1991
El Gouazine
563230
3973777
397
16.64
237
1990
El Mouidhi
576959
3900003
240
3.01
142.77
1991
Es Senega
509498
3927160
629
3.27
80.4
1991
Hadada
511742
3966004
1010
3.66
84.97
1992
Jannet
517357
3969846
823
5.70
94.28
1992
Jedeliane
500335
3938320
782
-
1550
1992
M’Richet
553478
3994395
606
-
42.4
1992
Sadine 2
507262
3961662
868
-
82
1990
245
Construction year
7.2
Height-surface-volume relations
Table 7.2.1: Percentage error on resulting power relation for 1 lake when using different ranges of the true rating curve from 0.13
from 0.63
from 1.13
from 1.63
from 2.13
from 2.63
from 3.13
0.13
24.66
74.87
75.84
52.16
24.47
0.91
18.49
0.63
16.27
0.37
0.67
7.21
17.26
26.70
35.26
1.13
18.62
7.97
7.77
12.98
19.90
26.66
33.04
1.63
12.68
4.79
4.64
8.70
14.26
19.86
25.30
2.13
5.36
0.48
0.59
2.66
7.24
11.98
16.71
2.63
0.32
4.30
4.38
1.87
1.80
5.71
9.71
3.13
5.29
7.58
7.64
5.78
2.94
0.18
3.46
3.63
7.72
8.45
8.49
7.25
5.22
2.88
0.34
4.13
7.95
7.30
7.31
6.63
5.35
3.76
1.94
4.63
6.23
4.39
4.38
4.20
3.59
2.69
1.54
5.13
5.29
2.42
2.39
2.63
2.61
2.30
1.74
5.63
4.47
0.69
0.65
1.26
1.75
1.98
1.96
6.13
2.32
2.22
2.27
1.34
0.40
0.30
0.77
6.43
0.06
4.83
4.89
3.80
2.63
1.68
0.96
6.63
0.33
5.50
5.55
4.36
3.03
1.92
1.02
7.13
2.73
8.44
8.51
7.08
5.42
3.94
2.66
mean error
7.52
9.04
9.12
8.12
7.37
7.09
9.68
s.d.
6.87
17.21
17.43
11.76
7.14
8.84
11.79
0
246
3e+05
V from power relation (m3)
from.0.13 2e+05
from.0.63 from.1.13 from.1.63 from.2.13 from.2.63 from.3.13 from.3.63
1e+05
from.4.13
0e+00 0e+00
1e+05
2e+05
3e+05
V from rating curve (m3)
(a) Zoom on upper values
30000
V from power relation (m3)
from.0.13 from.0.63 20000
from.1.13 from.1.63 from.2.13 from.2.63 from.3.13 from.3.63
10000
from.4.13
0 0
10000
20000
30000
V from rating curve (m3)
(b) Zoom on lower values
Figure 7.2.1: Difference in volume estimated through H-V power relation if derived on all or upper values (Gouazine)
247
Site specific power relation
1e+05
Guordin (Cote d'Ivoire)
Cadier (Brazil)
● ●● ●
●
5e+04
●●
● ●
● ●
0e+00
● ●● ●●
●
●
● ● ● ●●●● ●
●
V from power relation (m3)
Lacombe (Tunisia)
●
●
●
●●
●
●
●
● ● ● ●● ●●●
Liebe (Ghana)
●
●
Sawunyama (Limpopo)
●
1e+05
●● ● ●
5e+04 ● ●
0e+00
●● ●●●
●
●
● ● ● ● ●●●●
Interlake relation (Tunisia)
●
●
●
●
●●
●
●
●●●● ● ● ●
Lower range lakes (Tunisia)
●
●
●
●
●●
●
●
Upper range lakes (Tunisia) ●
● ●
1e+05 ●● ●
● ●
●
0e+00
●
0
●
●
●
● ● ●● ●●
25000
●
●
●
● ●● ●●
●
●
●
5e+04
●●
●●
50000
75000
●
100000 0
●
● ●● ●●●
25000
50000
75000
●
100000 0
●
●
25000
50000
75000
100000
V from rating curve (m3)
(a) Es Senega
Site specific power relation
Guordin (Cote d'Ivoire)
Cadier (Brazil)
●
60000 ●
40000
● ●
20000
●
●
●
0
● ● ●●
●
●
● ●● ● ●
●
Lacombe (Tunisia)
V from power relation (m3)
●
●
●
●
●
●
●
● ● ●● ●
●
Liebe (Ghana)
●
●
●
●
●
●
●
Sawunyama (Limpopo)
60000
40000 ●
20000 ●
0
● ● ●●
●
●
●
●
●
●
● ●● ● ●
Interlake relation (Tunisia)
●
●
●
●
●
●
●
●
● ●● ● ●
Lower range lakes (Tunisia)
●
●
●
●
●
●
●
●
●
Upper range lakes (Tunisia)
60000
40000 ●
20000 ●
0
●
● ● ●●
●
0
10000
●
●
●
● ● ● ●●
20000
30000
●
●
●
●
40000 0
●
●
10000
●
●
●
● ●
● ● ● ●●
20000
30000
40000 0
●
●
10000
●
●
●
●
●
20000
30000
40000
V from rating curve (m3)
(b) Sadine 1
Figure 7.2.2: Comparison of true volume against volume estimated by different power relations 248
Site specific power relation 60000
Guordin (Cote d'Ivoire)
Cadier (Brazil)
●
●
40000
●
● ● ●
20000
●
● ●
● ●
0
●
●
● ●● ●
●
●
●
●
●
●
●
●
●
● ●
Lacombe (Tunisia)
V from power relation (m3)
●
●
●
●
Liebe (Ghana)
●
Sawunyama (Limpopo)
●
60000 ●
40000
● ● ●
20000
● ● ● ● ●
0
●
●● ●
●
●
●
●
●
●
●
●
●● ● ●
Interlake relation (Tunisia)
Lower range lakes (Tunisia)
●
●
●
●
●
●
●
Upper range lakes (Tunisia) ●
●
60000
● ●
●
40000
● ●
●
●
● ●
20000
●
● ●
●
●
●
●
●
●
●
●
●
●
● ● ●
● ● ●
0
●
●
0
20000
40000
60000
● ●
0
20000
40000
60000
●
0
20000
40000
60000
V from rating curve (m3)
(a) Baouejer
Site specific power relation
Guordin (Cote d'Ivoire)
Cadier (Brazil)
●
200000
● ● ●
150000
●
●
●
100000 ●
● ●
●
● ● ●
●
●
●
●
● ● ●● ●
● ●
●
●
● ● ● ●
Lacombe (Tunisia)
V from power relation (m3)
●
●
●
50000 0
●
●
●
●
Liebe (Ghana)
Sawunyama (Limpopo)
200000 ●
150000 ● ●
100000
●
● ● ●
●
50000
●
● ●
0
● ● ● ●● ●
● ● ●
Interlake relation (Tunisia)
●
● ●● ● ● ● ●
Lower range lakes (Tunisia)
●
●
●
●
●
●
Upper range lakes (Tunisia)
●
200000
●
●
● ●
150000
● ●
100000
● ●
●
●
●
50000
●
● ● ● ●
● ● ●
0
●
●
●
0
●
●
●
50000 100000 150000 200000
0
● ● ●
50000 100000 150000 200000
0
50000 100000 150000 200000
V from rating curve (m3)
(b) Dekikira
Figure 7.2.3: Comparison of true volume against volume estimated by different power relations 249
Alti
Height_wall
2.5
2.5
2.0
2.0
●
●
● ●
1.5 200
400
600
SR
● ●
1.5
800
●
1000
8
9
Length_wall
10
11
12
Abdessadok Baouejer
Length_SR Brahim Zaher
2.5
Dekikira
2.5
2.0
2.0
●
El Gouazine
●
● ●
1.5
120
160
200
Es Senega
●
1.5
80
beta
●
250
500
Mean_depth
750
1000
Fidh Ali
1250
Fidh Ben Nasseur
S
Hadada 2.5
2.5
2.0
●
2.0
●
2
1.5 3
●
4
5
6
7
●
30000
V
Janet M'Richet
●
●
1.5
●
Morra 60000
90000
●
Mouidhi Saadine 2 Sadine 1
2.5 2.0
● ●
1.5
●
2e+05
4e+05
6e+05
value
Figure 7.2.4: Relationship between — in S-V power relation and several lake characteristics for 15 lakes studied
250
Alti
Height_wall
0.03
0.03
0.02
0.02
0.01
0.01
● ●
0.00 ● 200
400
600
800
SR
● ●●
0.00 1000
8
9
Length_wall
●
10
11
12
Abdessadok Baouejer
Length_SR
0.03
0.03
0.02
0.02
Brahim Zaher Dekikira El Gouazine
0.01
0.01
● ●●
0.00
B
80
●
0.00
120
160
200
Es Senega
●
●
250
500
Mean_depth
750
1000
Fidh Ali
1250
Fidh Ben Nasseur
S
0.03
0.03
0.02
Hadada ●
0.02
0.01
●
0.00
M'Richet
0.01
●
2
3
●
4
5
6
0.00
7
Janet
● ●
Morra
●
30000
V
60000
90000
●
Mouidhi Saadine 2
0.03
Sadine 1
0.02 0.01 0.00
● ●
●
2e+05
4e+05
6e+05
value
Figure 7.2.5: Relationship between B in S-V power relation and several lake characteristics for 15 lakes studied
251
Abdessadok
Baouejer
Brahim Zaher
Dekikira
El Gouazine
Es Senega
Fidh Ali
Fidh Ben Nasseur
Hadada
Janet
M'richet
Mouidhi
3 ● ●
2
●
1
3
beta
2 1
3 ●
●
●
2
●
1 Sadine 1
Sadine 2
3 ● ●
●
2
●
●
●
1
●
●
0
5
10
15
0
5
10
15
Years
(a) — coefficient
Abdessadok
Baouejer
Brahim Zaher
●
8e−04
6e−04
0.00015
4e−04 2e−04 0e+00
●
4e−04
0.00005
2e−04
Es Senega
0.010 0.005
B
0.0000 Fidh Ben Nasseur
0.015
0.004
8e−04
0.010
0.003
6e−04
0.002
4e−04
0.001
2e−04
0.000
0.000
Hadada
Janet
0.0100
0.0025
Fidh Ali
0.005
0.000
0.0050
0e+00
0.00000
●
El Gouazine
3e−04
0e+00 M'richet
●
Dekikira
0.0075
6e−04
0.00010
0.0100
Mouidhi 0.00012
0.003
0.0075
0.00008
2e−04
0.002
1e−04
0.001
0.00004
0.000
0.00000
0.0050 0.0025
●
0e+00
0.0000 Sadine 1
●
●
Sadine 2 0.015
●
2.0 0.010
1.5
0.005
1.0 0.5 0.0
0.000 ●●
● ●● ● ●
0
5
−0.005 10
15
0
5
10
15
Years
(b) B coefficient
Figure 7.2.6: Change of power relation parameters over time for 15 small reservoirs
252
Chapter 8
Appendix to chapter 3 Additional content 8.1
Background on GPS methods
The Global Positioning System (GPS) is one of the global geolocation systems, funded & controlled by the US Department of Defence (DOD). Others included Glonass (by Russia), and Galileo the European project. It relies on 24 satellites orbiting at 20,200 km from earth with a revolving period of 12 hours, allowing a minimum of 4, often near 8 satellites to be seen from all points of the earth. Locations are determined through triangulation using at least four satellites, and the distance between satellites and the receiver is measured through the travel time of radio signals. GPS therefore require accurate synchronisation of emitters, made possible by cesium atomic clocks. Signals are sent in two wavelengths of 19.04 cm and 24.44 cm but only the coarse acquisition code on 1 band was previously accessible to civilians the other being restricted by the US DOD. In 2000, selective availability which previously limited precision to 100 m was lifted. The Standard Positioning Service of GPS however has a theoretical precision of 10-15 m in all three dimensions due to errors based on the position of the satellite (taken from ephemeris data), synchronisation of the receiver clocks and ionospheric interferences. Differential GPS relies on correcting coordinates based on errors observed on nearby fixed receivers (within 100 km) whose coordinates are precisely known. These error are transmitted through radio beacons, satellites or internet and allow corrections in real time or post treatment. New systems such as WAAS (Wide Area Augmentation System) (and similarly European Geostationary Navigation Overlay Service (EGNOS) in Europe) provide corrections to many basic GPS receivers capable of improving latitude and longitude accuracies to 3-5m. More sophisticated corrections such as the RTK based on the phase of the received signal allow precision of the order of the centimetre in all three dimensions. A TOPCON DGPS with RTK enabled was used here.
253
8.2
Selecting a spectral water index
Table 8.2.1: Water detection rates on each reservoir based on different calibration methods of MNDWI
18
30
35
28
Optimal threshold
Overall accu
Prod accu (water)
Prod accu (non water)
User accu (water)
PDAI
Omissions (water & non water)
-0.17
85.8
85
86
60
55
Omissionscommission
-0.09
91.4
79
95
79
0.3
Om+com
-0.09
91.4
79
95
79
0.3
Overall accu
-0.09
91.4
79
95
79
0.3
PDAI
-0.09
91.4
79
95
79
0.3
Omissions
-0.08
87.3
87
87
90
-3.7
Omissionscommission
-0.09
87.7
90
85
89
1.0
Om+com
-0.12
89.0
95
81
87
9.0
Overall accu
-0.12
89.0
95
81
87
9.0
PDAI
-0.09
87.7
90
85
89
1.0
Omissions
-0.18
71.5
79
69
42
294.3
Omissioncommission
-0.17
76.0
40
86
45
97.2
Om+com
-0.2
70.4
100
62
43
392.9
Overall accu
-0.17 -0.03
80.5
28
95
64
-50.7
PDAI
-0.14 -0.03
76.0
40
86
45
-1.5
Omissions
-0.19
90.2
91
90
71
29
Omissionscommission
-0.13
91.2
79
96
79
0.4
Om+com
-0.19
90.2
91
90
71
29
Overall accu
-0.1
91.5
73
97
85
-14.0
254
21
25
Optimal threshold
Overall accu
Prod accu (water)
Prod accu (non water)
User accu (water)
PDAI
PDAI
-0.13
91.2
79
96
79
0.4
Omissions
-0.09
89.3
89
90
86
2.8
Omissionscommission
-0.08
88.6
87
90
87
-0.4
Om+com
-0.09
89.3
89
90
86
2.8
Overall accu
-0.09
89.3
89
90
86
2.8
PDAI
-0.08
88.6
87
90
87
-0.4
Omissions
-0.14
93.5
93
93
83
13.3
Omissionscommission
-0.09
94.0
88
96
88
-0.5
Om+com
-0.1
94.4
90
96
88
2.9
Overall accu
-0.1
94.4
90
96
88
2.9
PDAI
-0.09
94.0
88
96
88
-0.5
255
Table 8.2.3: Errors on validation image using threshold derived from calibration (on overall accuracy) for each lake and index Calibration threshold variance NDWI
1.41 x 10-3
NDVI
7.17 x 10-4
NDTI
8.18 x 10-5
MNDWI
NDMI
8.3
1.10 x 10-3
1.56 x 10-3
18
21
25
28
30
35
50
Ov accu
89.7
87.8
65.3
90.9
89.6
80
93.4
Prod accu
32.8
79
89.8
57.9
88.8
0
91.4
User accu
81.3
88.7
39
90.4
92.2
NA
88.7
PDAI
60
15
240
40
7
100
0
Ov accu
89.2
88.3
67.7
89.9
89
80
93.4
Prod accu
24
77
85
51
87
0
91
User accu
90
92
40
99
93
NA
89
PDAI
73
17
228
44
5
100
3
Ov accu
88.1
88
69
92
91.2
81.1
93.4
Prod accu
74
77
91
65
89
26
86
User accu
55
91
42
90
95
56
93
PDAI
34
15
234
28
7
14
7
Ov accu
92.6
87.9
64.8
92.7
89.6
79.6
93.9
Prod accu
77
81
86
67
88
42
90
User accu
71
88
38
93
93
49
91
PDAI
8
7
244
28
5
14
1
Ov accu
77.2
78.6
62.9
89.3
86.6
69.8
84.9
Prod accu
88
62
81
68
81
56
68
User accu
36
79
36
73
94
34
83
PDAI
142
4
244
8
13
244
11
Calibrating percentage of pixels to remove due to clouds & SLC off
256
% of images with scene cloud cover 50%) scene cloud cover
Figure 8.3.1: Cloud cover (%) across 1 lake cell (Gouazine) for images of different scene cloud cover
257
● ●
●
●●
●●● ●
●
● ● ● ● ●● ●
●
●
●
●● ●● ● ● ●●
●●
●● ● ●
● ●●● ● ●●● ●●
● ●
●
●
●
● ● ●
●
● ●●
●
●
●
● ●
●
● ●●
● ●
● ●
Remotely sensed surface area (m²)
1e+05
●
●
●
5e+04
0e+00
● ●
●
● ●●
●
● ●
●
● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ●
● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●
2000
2005
2010
Cloud & shadow %
100
50
0
2015
Date
Figure 8.3.2: Errors from clouds and shadows on remotely sensed surface area time series (lake Gouazine, cell 51)
●
● ●
1e+05
●
Remotely sensed surface area (m²)
●
5e+04
0e+00
●
● ●
●
● ●
●
●
● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ●● ● ●● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ●
2000
●
2005
2010
SLC−off %
100
50
0
2015
Date
Figure 8.3.3: Errors from SLC-off pixels on remotely sensed surface areas time series (lake Gouazine, cell 51)
258
1.00
1.00
SLC−off %
Cloud & shadow %
0.75
0.75 ●●●●●●●●● ●●●●●●●●●
●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●● ●●●●●
100
●●●●●●● ●●● ●● ● ● ●● ●●●●●●● ●● ●● ●● ●● ●● ●● ● ● ● ● ● ● ● ●●●●●●●●●●●● ●●●●●●●
0.50
50
R² value
R² value
●●●●●●●●●●●●
0.25
●●●●●●●●●●●●
0.50
100
●●●● ●●●●●●●●●●●● ●●●●●●●● ●●●●●● ●●●●●●●●●● ●●●●●●●● ●●●●●●●●●●●●
50
0.25 ● ●● ●● ● ● ●● ●● ●● ●● ●● ●
0
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
0.00
0
0.00 0
50
100
150
0
Cloud & shadow % 550
500
400
350
300
0
50
100
100
150
● ● ● ● ●●●●●●●● ●●●●●●● ● ●●●●●●● ●●● ● ●●●●●●● ● ● ●●●●●●● ● ● ●●●●●●● ● ●●● ● ●●●●●●● ● ● ●●●●●●● ● ● ● ● ●●●●●●● ● ● ● ● ● ● ● ●●●●●●●● ● ● ● ● ● ●●●●●●● ● ● ● ● ● ● ● ● ● ●●● ●●●●●●●● ●● ●● ●●●●●●● ●● ● ● ● ●● ●●●●●●●● ●●●●●●● ● ● ● ● ●● ● ● ●● ●●●●●●● ●● ●● ●●●●●●●● ●●●●●●● ●● ●● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ●● ●●●●●●●●● ●● ●● ● ● ● ● ● ● ● ● ● ● ●●●●● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●●● ●● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●●● ●● ●● ● ●● ●● ● ●● ● ● ● ●● ●● ●● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ●● ●● ● ● ●● ●●● ●● ●● ●● ● ●● ●● ●● ● ●●● ●● ●● ●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ●●● ● ● ● ●● ●● ●● ●● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●
SLC−off %
100
50
Nb of images
●●●●●● ●●●●●● ●●●●●● ●●●●● ●● ● ●●●●●● ● ●●● ●●●●●●● ●● ●● ● ● ●●●●●● ● ● ● ● ● ● ●●●●●●●● ●● ●● ●● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ●●●●● ●●● ●● ●●● ●●●●● ●● ● ●● ● ● ●● ●● ●● ●●●●●●● ● ●● ●● ●● ●● ●● ●● ●● ●● ●●●●●● ●● ●● ● ● ●● ●●●●●●●●● ●● ●● ● ● ●● ●● ●● ●● ●● ●●● ●● ●● ● ●● ●● ●●●● ●●● ●● ● ●● ● ● ● ● ● ● ● ● ●●●●● ● ●● ● ●●● ●●● ● ●●●● ●● ● ●●● ●●●● ● ● ● ●●●●●●●●●●● ●● ●● ●● ● ● ●● ●●●●●●●●●●●● ● ●●● ● ● ●●●● ● ●● ●●●●●●● ●● ●● ●●● ●●● ●● ●●●●●●●● ●●●●●● ● ● ● ●●●●●● ●●●● ●● ●●● ● ● ●●●●●●●●●●●●●● ●●● ●●● ● ●●● ●●●●●●● ●●●●●●●●● ●●●●●● ●●●●●● ●● ●●●● ●● ●●●●●●● ●● ●● ●● ●● ●● ● ●● ●● ●●● ●● ●● ●● ●● ●● ●● ●●●●●●●● ●● ●● ●● ●●● ●● ● ● ●●●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●● ●●● ●●● ●●● ●●●●●●●●● ● ●● ●●● ●● ● ● ●● ●● ●● ●● ●● ●● ●●● ●● ●●● ●● ●● ●● ●●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●●●●●●●● ● ● ● ● ●● ●● ●● ●●●●● ●● ●● ●● ●● ●● ●● ●●●● ● ●●●●●●●●●●●● ● ● ● ● ● ●●●●● ●●●● ●●●● ●●●●● ●●●● ●●● ●●●● ●●●● ●●●●●●● ●●●●●●●●
450
50
SLC−off %
●● ●● ●● ●● ●● ● ● ●● ●● ●● ●● ●● ● ● ●● ●● ●● ●● ●● ● ●
500
Nb of images
●●●● ●●●●●●●●
●●●●●●●●●●●● ●●●●●●●●●●●● ●●●●●● ●●●●●●
400
300
0
Cloud & shadow %
100
50
0
200
150
0
50
Cloud & shadow %
100
150
SLC−off %
(a) Fidh Ali
1.00
1.00
SLC−off %
Cloud & shadow % 0.75
●● ●●● ●●● ●● ●●● ●●● ●●● ●●● ● ● ●● ●● ●● ●● ●● ● ●● ●●● ●●● ●●● ● ● ●● ●●● ●●● ●●● ●●● ●● ● ● ● ●●●●●● ● ●●●●●●●●●● ●● ●● ●● ●● ● ●● ●● ●● ● ●● ●● ●● ● ● ●● ●●●●●● ● ●● ● ● ● ● ● ●● ●● ●● ●● ●● ●● ●● ●●● ●● ●● ●● ●● ●● ●● ●● ● ● ●● ●● ●● ●● ●● ●● ●● ●● ●●● ● ● ●● ●● ●● ●● ●● ●● ●● ●● ● ● ● ● ● ● ●●●●●●●●●● ●● ●● ●● ●● ●● ●● ●● ● ● ● ●●●●●●●● ● ● ● ● ●● ● ●●●● ●●●●●●●● ●●●●●●●●●● ●●●●● ● ● ● ● ●●●●●●●● ●● ●● ●● ●●●●● ●● ●●●●● ●●● ●●●●●●●● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●●●●●● ●●●●●●●●●● ●● ●●● ●●● ● ● ●●● ● ● ●●●● ●●●●● ● ●
0.50
100
50 ● ● ●● ●● ● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ●● ●● ●● ● ● ●● ●● ● ●● ●● ●● ●● ●
0.25
R² value
R² value
0.75
● ●●● ●●● ●●●● ●● ● ●●●●● ●●● ●●●● ● ●●●● ●●● ●● ● ●● ●●●● ●● ●●●● ●●●● ●●● ●● ●●●● ●●●● ● ●● ●●● ●●● ●● ● ● ● ●●●●●● ●● ●● ●● ●● ●●● ●●●● ●●● ●● ●●● ●●● ●● ●●●● ● ● ● ●● ●●● ●●●● ● ●● ● ● ●●● ● ●● ● ● ●● ●●● ●● ●●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ●●● ●● ● ●● ● ●● ●●●●● ●● ●●●●●● ● ●● ● ● ● ●●●●●● ●●●●● ●●● ●●● ●●●●● ●● ● ●● ●●●●● ● ● ●●●●●
0.50
●●●●●●●●●●●●●● ●●●●● ● ●●●● ●●●●●●
0.25 0
0.00
50
0
0.00 0
50
100
150
0
Cloud & shadow % 550
500
400
350
300 0
50
100
100
150
150
● ● ● ● ● ●●●●● ● ●●●●●● ● ●●●●●● ●● ●●●●● ●● ●●●●● ●● ●● ● ●●●●●● ● ●● ●●●●● ● ●● ●●●●● ●● ●● ●●●●● ●● ● ●●●●● ● ● ●●●●● ● ●● ● ● ● ● ● ● ● ● ● ●●●●●●● ●●●●● ●● ●●● ●● ●● ●●●●● ● ●●● ●● ●●●●● ● ● ●●●●● ● ●● ●● ●● ●● ● ● ● ●● ●● ● ●●●●●● ●●● ● ● ● ● ● ● ●● ●● ● ● ● ●●● ●● ●●● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●●● ●● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ●● ●● ●● ●● ● ●● ● ●●● ● ●● ●● ●● ●● ● ● ● ●● ●● ●●● ● ●● ● ●● ●●●● ●● ●● ●● ●● ● ●● ● ●●● ● ● ●● ●●●● ● ●●● ●● ●● ●● ●● ●●● ●● ● ● ● ●● ●● ● ●●● ●●● ●● ●● ●● ● ●● ●● ●● ●● ●● ● ● ●● ●● ●● ● ● ●● ●● ● ● ● ●● ●● ● ●
SLC−off %
100
50
Nb of images
●●●●● ●●●●● ●●●●● ●●●●● ●●●●● ●●●●● ●●●●●● ●●●●●●● ●●●● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ● ● ● ●●● ●● ●● ●●●● ● ●●●●●● ● ● ● ● ● ●●● ●●●●●●●●● ●● ●●●●●●● ● ●●●● ●●●●●●●●●●●● ● ● ●●●●● ●●●● ●●● ●● ●● ●● ●● ●● ● ●● ●● ●●● ● ●●●●●● ● ● ● ● ● ●● ● ●●●● ●●● ●●●● ●● ●● ●●●● ●●●●● ●●●●●●●● ● ● ● ● ● ● ●●●●●●●●●● ●●●●● ●●●●●●●●● ●●● ● ●● ● ● ● ●●● ●●●●● ●●●●● ●●● ● ●●●●●●●●●●●●●● ●●● ●●● ●● ●● ●●● ●●●●●●● ● ● ● ● ●●● ●●● ●●●●●●●●● ●● ●●●● ●●●● ● ●●●●● ●●●●● ●●● ●●●●●●●●●●●●●●● ●●●●●● ●●● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●● ●● ●● ●● ●●●●●●● ●● ●●●●●● ●●●●● ●●●●●●●●● ●●● ●● ● ●●●● ●● ●● ●●●● ●●●●●●●●●●● ●● ●● ●● ● ● ● ● ● ● ● ●●●●●●● ●●●●●●●●●●● ●● ● ● ● ●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ● ●● ●● ● ●● ● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ● ●●●●● ●●●●● ●●● ●● ●●●●●●● ●● ●●●●●●●● ●● ● ●●●●●● ● ● ● ● ● ● ●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●● ●●●●●●●● ●●●●●●●●●●●●●●●
450
50
SLC−off % ●●●●● ●●●●● ●●●●● ●●●●● ●●●●● ●●●●●
500
Nb of images
100
400
300
0
Cloud & shadow %
100
50
0
200 0
Cloud & shadow %
50
100
150
SLC−off %
(b) Guettar
Figure 8.3.4: Optimising SLC-off and cloud & shadow thresholds on additional lakes
259
1.00
1.00
SLC−off %
Cloud & shadow % 0.75
100 ●● ●● ● ● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●●● ● ●● ●● ●● ●● ●● ●● ●● ●● ●● ●●●● ● ●●●●●●●●●●●● ●●●●●●●●●●●●●
0.50
50
●●●●●●●●●●●●●
●●●●●●● ● ●● ● ● ●●●●●●● ● ●● ●●● ●● ●● ●● ● ● ● ● ● ●●●●●●●●●●●●●●●● ●●
0.25
0
50
100 ● ●● ●●● ●● ● ●● ●
0.50
100
50
● ● ● ●● ● ●● ●●●●● ●● ●● ●●● ●● ● ●●● ●● ●● ●● ●● ●●●●
0.25 0
● ●●●● ●●● ●●●●●●●●● ●●●●●●●●●●●●● ●●●●●● ● ●● ●● ●● ●● ●● ●● ● ●●●●●●●●●●●●●● ● ● ● ● ● ●●●●●● ● ● ●●● ●● ●● ●● ● ●●● ●● ●● ●● ●●●● ●● ●● ●● ●● ●● ●● ● ●●●●●●●●●●●●● ●● ●● ●● ● ● ●●●●●●● ●● ●● ●● ●● ●● ●● ●● ●● ●● ● ●● ●●●●●● ●●●●●●●●●●●●●●
0.00
R² value
R² value
0.75
0
●●● ● ● ●● ● ●●●●●●●● ● ● ● ● ●● ●●● ● ● ●●● ● ●● ●●● ● ● ●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●● ●● ●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●
0.00
150
0
50
Cloud & shadow %
100
150
SLC−off % 500
●●●●●●● ●●●●●●●
●
●●●●●●● ● ● ●●●●●●●●●● ●●●●●●● ●●● ●● ● ●●●●●●●● ●●●●● ● ● ● ● ●●● ●●●●●●● ●●● ●● ●●●●●●●● ●● ● ●● ● ●● ● ●●●●●●● ● ● ●●● ● ●●●●●●●● ● ●● ●● ● ●● ● ● ●● ●●●●●●● ●●●●●● ●● ●● ● ●●●●●●● ● ● ●●● ●● ●●● ● ● ●●●● ●●●●●●●●● ● ● ● ● ● ● ● ● ●●●●● ● ● ● ● ● ● ●●●●●●● ●● ● ● ●●● ● ●●● ●● ●●●●●●● ● ● ●●●●●●● ● ● ● ● ● ●● ● ●● ●●●● ●●●●● ● ●●●●●● ● ● ● ● ●● ●● ●●●● ● ●●●●●●● ●●●●●● ●●●●●●●●● ●●●●●●● ●●● ●● ● ● ● ● ● ● ● ●●● ●● ●● ● ●●●●●●● ●●● ●●●●● ● ●● ●●●●●●●●●● ●● ●●●●●●● ● ●●●●●●●●●●●●●●●●●● ● ● ●●●●● ● ● ● ●●●●●●●●●●● ●● ● ●●● ●●●● ● ● ●●●● ● ● ●●●●● ● ●●●●●●● ●● ●● ● ●●●●●●●●●●●● ●●●●●● ●●●●●●●● ●●●● ● ●●●●●●●●
400
300
● ●
SLC−off %
100
50
Nb of images
Nb of images
●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●● ●● ● ● ●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●● ● ● ● ●● ●●●●●●●●●●●●●●● ●●● ●●●●●●●●●●●●●●●●●● ● ●●● ●●●●●●●●●●●●●●● ● ●●● ●●●●●●●●●●●●●●●● ●● ● ● ●● ●● ●● ●●● ● ●● ● ● ● ● ● ●●●● ●●● ●● ● ● ●●● ●● ● ●●● ●●● ● ● ● ● ● ● ● ● ● ● ●●● ● ●●● ●● ●●● ● ●●●● ● ●● ●● ●● ●● ● ● ●● ●● ● ●●● ● ●● ●● ●● ● ● ●● ● ●●● ● ●● ●● ●● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ●●● ● ● ●●● ● ● ● ● ●● ●● ●● ●● ●● ● ●● ● ● ● ● ●● ●● ●● ●● ● ● ● ●● ●● ● ●● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●
●●●●●●●
500
50
100
300
0
Cloud & shadow %
100
50
0
200
200 0
400
150
0
50
Cloud & shadow %
100
150
SLC−off %
(a) Morra
1.00
1.00
SLC−off %
Cloud & shadow % 0.75
● ●● ●● ●● ●● ●● ● ● ● ●● ●● ● ●●●●●●●●●●●●●● ●●●●●●● ●●● ●● ●●●●●●●● ●● ●● ●● ●● ●●●●●●●●●●●●●● ●●●●● ●●●●●●● ●●●●●●● ●●●●● ● ●●●●●●●● ●● ●●●●●● ●● ●●●●● ●● ●● ● ●●●● ● ●●●●●●● ●● ●●●●●● ● ● ● ● ● ●● ●●●● ●● ●●●● ● ●●●●● ● ●●●●●●●●●● ● ●●●● ●● ●● ●● ●●●●● ● ● ●●● ●● ●● ●● ●● ●● ● ●● ●●● ● ●●● ●●●
0.50
0.25
100
50
R² value
R² value
0.75
●●● ●●●●●● ●● ●●● ●● ●●● ●●● ●●●●●● ●● ●● ● ●● ●● ●● ●●●● ●● ● ●● ● ● ● ●●● ● ● ● ● ● ●● ●● ● ●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●●● ●● ●● ●●●●●●●●●●●●●●●●● ● ●● ● ●●● ● ● ● ● ●●●●●●●●●●●●●●●●● ●●●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●
0.50
0.25 0 ● ● ● ● ● ● ● ●● ●● ●● ●● ●● ●● ● ● ● ●● ●● ●● ●● ●● ●● ●
50
0
●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●
0.00
0.00 0
50
100
150
0
Cloud & shadow % 500
300
200 0
50
100
100
150
150
● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●● ●● ●● ●● ●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●● ●● ● ●● ●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●● ● ●●●●●●●●●●●●●●●●●●● ●● ● ●● ● ● ●●●●●●●●●●●●●●●●●● ●● ● ● ●●●●●●●●●●●●●●●●● ● ●●●●● ●●●●●●●●●●●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●●● ●●● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ●●●●● ● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●● ● ● ●● ● ● ● ● ● ● ●● ●●●●●● ● ● ● ●● ●●● ●● ● ●● ● ● ●● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ●● ●● ● ● ●● ●● ● ● ● ● ●● ● ●● ● ●● ●● ● ● ● ●● ● ●● ●● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ●● ●●● ● ●● ● ● ● ●● ●● ● ● ●● ● ● ●●● ● ●●● ●
●
SLC−off %
100
50
Nb of images
● ●● ●●● ●●●●●●● ●●●●●●●●●●●●● ● ● ●●●●●● ● ●●●● ●●●●●●●●●●● ●●● ●●●●● ●●●●●●● ● ●● ● ● ● ●● ●● ●●● ●● ●●●●●● ●● ●● ● ●●●●●●● ●● ●● ●● ●●●●● ●●● ● ●● ● ● ●●●● ●●●●● ● ● ●● ● ●● ● ●● ● ●●●● ● ● ● ● ● ●● ●●●●●●● ●● ● ●●● ● ● ● ● ● ● ●●●●●●● ●●● ●● ● ●●●● ●●●●●●●● ● ● ● ● ●●●● ●● ●●●●●● ●● ●●●●●●● ●● ●●●●● ●● ● ● ● ●●●●● ● ●●●●●●● ● ●●● ●●●●●●●● ●●● ● ● ● ● ●●●●●●● ● ●●●●● ●● ● ● ●●● ●● ●●●● ● ● ● ●●●● ● ●● ●●●●●●● ● ●●●●● ●●● ●● ●●●●●● ●● ● ● ● ● ● ● ● ● ●●● ● ●●●● ●●● ●●● ●●●●●●● ●● ●●● ● ●●●●●●●● ●●● ●●●●● ●●●●● ●●●●●●
400
50
SLC−off %
●●●●●●● ●●●●●●● ●●●●●●● ●●●●●●● ●●●●●●●
500
Nb of images
100
400
300
0
200
0
Cloud & shadow %
50
100
Cloud & shadow %
100
50
0
150
SLC−off %
(b) Dekikira
Figure 8.3.5: Optimising SLC-off and cloud & shadow thresholds on additional lakes (cont.)
260
●
●
●
●
●
●
●
8e+15
AUC MSE
6e+15
4e+15
2e+15
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
0e+00 0
50
100
150
Cloud & shadow %
(a) Clouds & shadows
●
5e+15
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
4e+15
●
AUC MSE
3e+15
●
2e+15
● ●
1e+15 ●
●
●
● ● ●
● ●
0e+00 0
●
●
50
100
150
SLC−off %
(b) SLC-off
Figure 8.3.6: Aggregated surface area error over all images on Morra lake according to varying tolerance levels of pixel loss due to:
261
Chapter 9
Appendix to chapter 4 Additional content 9.1
Interpolating lake evaporation
Table 9.1.1: Relationship between the interpolated (IDW) monthly evaporation 1995-2008 values for each lake and for El Haouareb
Lake
R
Intercept
Slope
Trozza_sud
0.99
1.195
0.972
Marrouki
1.00
-0.088
0.994
Marrouki_2
1.00
-0.132
0.992
Dhahbi
0.99
-0.932
0.993
Salem_Thabet
0.99
-0.561
0.990
Abda
0.97
-2.030
0.994
Hammam
0.97
-1.786
0.990
Chauoba_Hamra
0.92
-5.361
1.012
Hoshas_amont
0.96
-2.588
0.994
Hoshas
0.95
-3.150
0.999
Ain_Faouar
0.93
-4.493
1.008
Fidh_Mbarek
0.89
-7.033
1.024
Maiz
0.89
-7.266
1.027
Mazil
0.89
-6.346
1.016
Fidh_Ali
0.88
-7.365
1.027
Fidh_BenNasseur
0.88
-6.973
1.023 262
Lake
R
Intercept
Slope
Bouksab
0.89
-5.687
1.012
Gbatis
0.89
-5.898
1.012
Guettar
0.89
-4.332
0.997
Mahbes
0.90
-3.762
0.992
En Mel
0.89
-5.722
1.011
Sidi_sofiane
0.93
3.017
0.924
Smili_1
0.92
0.572
0.949
Smili_2
0.92
0.172
0.953
Kraroub
0.92
-0.780
0.955
Garia_3
0.93
0.517
0.941
Garia_2
0.93
0.750
0.935
Garia_S
0.93
0.592
0.936
Raouess
0.93
2.426
0.913
Morra
0.90
6.889
0.901
Habsa_B
0.90
8.662
0.877
Habsa_A
0.90
9.676
0.872
Fidh_Zitoune
0.91
6.703
0.881
El_Haffar
0.94
1.652
0.914
Bouchaha_A
0.93
-0.177
0.880
Bouchaha_B
0.93
-0.149
0.881
Daoued_1
0.93
-3.733
0.873
Daoued_3
0.93
-3.476
0.873
Fadden_Boras
0.93
2.260
0.905
Skhira_5
0.91
-4.849
0.861
Skhira_1
0.92
-6.025
0.867
Skhira_2
0.92
-6.744
0.866
Skhira_3
0.91
-6.107
0.865
Skhira_4
0.91
-5.457
0.863
Skhira_27
0.89
-11.880
0.863
Skhira_D
0.90
-6.151
0.859
Skhira_C
0.90
-9.830
0.863
Skhira_B
0.90
-10.397
0.862
263
Lake
R
Intercept
Slope
Skhira_A
0.90
-8.513
0.860
Mdinia
0.93
2.817
0.881
Gouazine
0.85
8.919
0.821
Ben_Houria
0.92
-5.283
1.013
Mdinia_2
0.94
0.975
0.885
Abdessadok
0.82
19.554
0.797
Arara
0.95
-1.336
0.891
Baouejer
0.94
-7.821
0.894
Bou_Haya
0.95
1.684
0.892
Brahim_Zaher
0.92
-1.765
0.919
Marrouki_3
0.99
-0.069
0.987
Marrouki_4
0.99
0.020
0.987
Marrouki_5
1.00
0.080
0.987
Mouidhi
0.84
32.377
0.807
Ogla
0.95
5.598
0.877
Senega
0.94
-3.056
0.901
Hadada
0.87
-19.575
0.848
Jannet
0.90
-17.959
0.870
Jedeliane
0.93
-13.466
0.904
M_Richet
0.94
5.601
0.861
Mrira
0.94
-3.515
0.889
Sadine_2
0.77
-12.440
0.902
Sbaihia
0.95
4.360
0.882
Zayet
0.96
7.129
0.897
Skhira_12
0.88
4.267
0.838
Dekikira
0.89
11.159
0.838
264
●
●
● ●
● ● ●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
● ●
● ●
200
●
●
●
●
●
●●
● ● ●
●
● ●
●
● ● ● ● ● ●
100
●
●●
●
●
● ● ● ●
● ● ● ●
● ●
●
●
●
● ● ●
●
●
●
●
● ●●
● ●
●
●
●
●
●
●
●
100
●
● ●
●
● ● ●
● ●
●● ●
● ● ●●
●
●
● ●● ● ●●● ●● ● ● ● ●● ● ● ● ● ● ● ●●●● ● ● ●● ● ● ● ● ● ● ●
● ●●
●
●
400
● ● ●
● ● ● ● ●
● ●
●
● ●
●
●
●
●
●
●
●
● ● ● ●
●
●
● ●
●
●
●
●
●
●
●
300
●
●
● ●
200
●●
●
●
●
●
100
●
●
●
●
●
●
● ●
●
200
●
●
●
● ● ●
●
●
●
●
●
●
●
●
● ● ●
● ● ●
●
●
● ● ●
● ●
●
●
●
●
● ● ●
● ●
●●
●
● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ●
●
●
● ●
●●
●● ● ●
● ● ●
●
●● ●
●
●
●
●
●
●
El.Gouazine monthly E (mm)
Dekikira monthly E (mm)
●
●
●
● ●
●
300
● ● ●
●
●
●
●
●
● ●
● ●
●
●
300
●●
●
●
●
100
El Haouareb monthly E (mm)
●
200
300
400
200
300
400
200
300
400
El Haouareb monthly E (mm)
●
400
● ● ● ● ● ● ● ● ●
●
●
●
300
●
● ●
●
●
● ●
●
●●
● ●
● ●
●
●
300
●
●
●
●
●
Guettar monthly E (mm)
El.Morra monthly E (mm)
● ●
●
● ●
●
● ● ● ●
●
● ●
●
● ● ●
●● ●
● ●
●
●
● ●
●● ● ●
●
●
●
● ●
● ●
●● ●●
●
●
● ● ●
●
●
● ●
●
● ●
●
200
200
●
●
●
● ● ●
●
● ●
100
●
● ● ●● ●
●● ●● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●
●
● ● ●
●
●
● ●●
●
●
100
●
● ● ● ● ●
●
●
100
200
300
100
400
El Haouareb monthly E (mm)
El Haouareb monthly E (mm)
●
●
400
400
● ● ●
●
● ●
● ●
● ●
● ●
●
●
● ● ●
●
Fidh.Ali monthly E (mm)
●
●
Fidh.Ben.Nasseur monthly E (mm)
300
●
●
● ●
● ●
●
●●
●
●
●●
●
●
● ●
●
● ●
● ●
●
● ● ●
200
● ●
● ●
100
●
● ● ● ●● ● ●●● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ●
● ●
●
● ●
● ●
●
●
● ● ●
●
●
●
● ●
●
●
● ●
● ●
200
●
●
●●
● ●
●
●
●
100
● ●
● ●
●
● ● ●
●
●
● ● ●● ●
●
● ● ● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
300
● ●
● ● ● ●
●●
● ●
100
200
300
100
400
El Haouareb monthly E (mm)
El Haouareb monthly E (mm)
400 ●
●
●
●
●
●
● ●
●
● ●
● ●●
●
● ●
300
● ●
●
Hoshas monthly E (mm)
● ● ●
●
●
●
●● ●
●
100
●● ● ● ● ●● ●
● ●
●● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●
●
● ● ●●
●● ●
● ●
●
●
● ●● ● ● ● ● ●● ● ● ●
●
● ●
● ● ●
● ● ● ● ● ●
●
● ●
●
●
200
●
●
●● ●
● ●
● ●
●
● ●
●
●
● ●
●
● ● ●● ● ● ● ● ● ● ● ● ● ● ●
● ●
●
●
● ●
●
●
●
100
200
300
400
El Haouareb monthly E (mm)
Figure 9.1.1: Relationships between El Haouareb and lake interpolated monthly evaporation values (mm)
265
9.2 9.2.1
Further details on estimating water balance fluxes Infiltration
Infiltration can be assessed by exploiting the fact that during depletion phases, P and Qin fall to 0 and closing the water budget then simplifies to equation 9.2.1. Infiltration is then possible to calculate based on daily water levels and estimated evaporation. This supposes an absence of any withdrawals or releases as well as negligible GWin and/or leaks. In practice, releases as shown below can be identified from the instantaneous curves (where available), while withdrawals depending on their amplitude and available volumes may be harder to identify. As discussed in section 4.3.7, local knowledge of the irrigation practices (number of pumps, irrigation regimes) can ease the determination of withdrawals. Here several lakes were not exploited (Hoshas, Gouazine) and on others the occasional nature of withdrawals compared to the duration of the depletion phases over which infiltration was calculated allowed us to identify such outliers. On periods without W or R or by assuming that the outliers are due to W or R considering the amount of days available we can obtain a value or equation for I as shown in section 4.3.4. Leaks (L) can occur from seepage through the dam wall, through faulty valves or in exceptional cases, as observed on El Hammam for instance, as a result of a partial rupture of the dam wall from large floods. Except in the later case, their amplitude remain minimal and constant. Though groundwater inflow is neglected in certain water budgets (Li and Gowing 2005; Lacombe 2007) other studies in the Merguellil (Selmi and Talineau 1994; Gay 2004; Montoroi et al. 2002) and elsewhere (Matsuno et al. 2003; Massuel et al. 2014b) revealed the existence of non negligible groundwater flows. Despite their shallow depth, small reservoirs can be fed by delayed subsurface flows as well as local resurgences or springs. On Gouazine, this was calculated using isotopic methods and showed to be of the order of 50 m3 /day (Gay 2004; Grunberger et al. 2004). This corresponds to 2.5 mm/day when the lake is at a low 2 ha and less when surface area rises. Furthermore, in the days after an event (retarded) subsurface flows can also contribute. Gay (1999) on Kamech small reservoir in Northern Tunisia notably showed that there was delayed flow from rains up to 1 month before. Lacombe (2007) despite neglecting GWin highlights that up to 10% of days had an additional inflow (from delayed subsurface or GWin ) which were therefore removed. Certain lakes such as Aïn Smili whose names (Aïn) denotes the presence of a spring are expected to have non negligible GWin . Selmi and Talineau (1994) notably stated that in 1994 despite drought, several reservoirs filled up due to groundwater flows in the Merguellil catchment. Where groundwater inflow is not negligible, groundwater inflow which is inherently difficult to measure directly could be estimated using a combination of hydrodynamic and geochemical methods. Several tracers can be used to refine the water balance and notably estimate groundwater fluxes. These include artificially introduced tracers, or naturally occurring chemical and isotopic tracers. Chemical methods include monitoring differences in electrical conductivity, as a result of increased salt concentration from evaporation or groundwater inflow. pH, cations (calcium, magnesium etc.) or anions (chloride, silica, etc.) monitoring are also widely used, especially before the advent of affordable isotopic methods (Fontes 1976; Risi 2009). This was beyond the scope of this study, considering the intense monitoring required during depletion periods with no additional rainfall, and which could here not be assumed by the observers. Though interviews questioned the presence of springs and/or baseflow, the presence and amplitude of 266
subsurface and groundwater flows on other small reservoirs remains to be investigated fully. Their role in supporting water availability during low flow periods is of notable interest. Infiltration estimates as mentioned above may indeed correspond to I + L ≠ GWin ≠ Qsubsurf ace . This may lead to an underestimate of absolute I values (from GWin ) but is not detrimental for the water balance modelling if these interferences remain minimal and (proportionally) constant over time. Where leaks occur only over certain periods through parts of the dam wall or are repaired, these are much harder to account for properly. As in Lacombe (2007), the influence of Qsubsurf ace or temporary leaks can however be detected in the daily infiltration estimates which are assumed to remain either relatively constant over time or vary as the water levels decline according to Darcy’s law. I = —V ≠ E + GWin ≠ W ≠ L
9.2.2
(9.2.1)
Releases
In some cases the occurrence and duration of releases are mentioned in the daily reports by the paid observers. Readings are nevertheless unreliable and conduits have not been gauged, and on instrumented reservoirs, releases are assessed from the stage readings and water balance analysis (i.e. —V ≠ E ≠ I). Releases occuring during the flood can therefore not be estimated but releases after or before the flood could be determined based on instantaneous (15 min interval) stage readings. These combine with other depletion factors including infiltration and evaporation which can be estimated and their presence can be seen in the outliers of Guettar for instance in figure 4.3.11. Minor anomalies on the contrary are essentially due to uncertainties over each flux (E, SHV). These releases can however be disguised/amalgamated with other fluxes, unvoluntary leaks and voluntary withdrawals. Leaks where observed may difficultly be estimated but their constant nature means these are effectively included in the infiltration terms. (On Morra for instance a minor leak from the valve is reported by the guard, effectively leading to greater losses from the reservoir). In the Tunisian context, these can be differentiated based on two parameters: the time of year and volume. In general, releases occur during the winter months when water levels are high and storms prevail while withdrawals are concentrated on the drier summer months. In practice, interviews showed that market gardening and olive grove were also watered during the winter months, but their amplitude is indeed greater in spring and summer months. Differentiation on the volumes can however be done considering the generally low number of pumps and withdrawals on these small reservoirs, compared to the released volumes which are significant and exceed any reasonable rate of withdrawals. Thermally powered these pumps have withdrawals rates below 4.2 litres/s (Lacombe 2007). Considering the watering needs and significantly the diesel costs, pumping rates reported during interviews did not exceed 8 hours per day (cf. section 4.3.7). which corresponds to a maximum of 121 m3 /day per pump in use. Pump numbers were shown to reach 7 on all lakes and a maximum of 10 on En Mel in 2005. Theoretical maximum withdrawals rates can then be assessed for each lake or a maximum withdrawal rate of 1200 m3 /day can be used to discriminate withdrawals from releases. Lacombe (2007) had similarly obtained on a different cross section of lakes in and around the Merguellil catchment a value of 1078 m3 /day, based on three pumps functioning 24 hours/day.
267
9.2.3
Overflows
The volumes passing through the spillway can be estimated based on instantaneous 15 min data over the duration the lake levels exceed the spillway and knowledge of the cross section’s geometry. The spillways of small reservoirs were in certain instances instrumented (9 lakes used in Lacombe (2007)) with a stage ladder allowing for an estimation of the volumes which flow through this channel based on the known geometry/cross section of the spillway and time series of the flow. On the 3 SR studied by Lacombe (2007) overflow corresponded to on average 18% of inflow. These vary widely between reservoirs and years with lakes including Fidh Ali & Morra reported to never having overflowed since their construction while Hadada and Brahim Zaher overflowed over several years and up to 40% of all inflow. Recent testimonies during the interviews revealed that this since occurred on Morra but has not been measured. On Fidh Ali they notably occur as a result of progressive silting. On Saadine 2 and Jannet, these began to occur nearly systematically from 3 years after the construction as a result of very strong silting. For our WB+GR4J model, data on overflows was available for selected small reservoirs (Dekikira, Gouazine, Fidh Ali, Fidh Ben Nasseur). Contradictions between the data and information provided in the interviews (e.g. no overflows on Guettar in the data but reported overflows & damage in which year in interviews) led to doubts over the reliability of these reports. They indeed require the presence of the dam operator during the storm, who during these larger storms may be more inclined to operate the valve or put pumps and livestock under shelter than monitor the stage ladder, as confirmed in interviews. This information was nevertheless used in the calibration and validation of the GR4J model on the instrumented small reservoirs as during large events the volume which overflowed must be added to the volume collected in the dam. Further instrumentation of these could provide interesting insights into the actual volumes overflowing, notably on Morra where the configuration of the spillway directly opposite the river inflow means it may overflow significantly.
9.3
Kalman filter performance
268
VWB+GR4J
200000 150000 100000 50000 0
VENKF
200000 150000 100000 50000 0
200000 150000 100000 50000 0
VRS
Volume (m3)
VENSKF
200000 150000 100000 50000 0
Vfield
200000 150000 100000 50000 0
VENSKF_lowess
200000 150000 100000 50000 0
VRS_smooth
200000 150000 100000 50000 0 2000
2005
2010
2015
Date
(a) Modelled and observed volume time series
150000 Vfield
100000 50000 0
VWB+GR4J
150000 100000 50000 0 150000
VENKF
100000 50000 0 150000
VENSKF
100000 50000 0
VENSKF_lowess
150000 100000 50000 0
VRS_smooth
150000 100000 50000 0 150000
VRS
100000 50000 0 2000
2005
2010
2015
(b) Modelled and observed mean daily water availability per year (m3 )
Figure 9.3.1: Performance of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake for 1997-2014. These alternate outputs were later discarded.
269
VWB+GR4J
VENKF
VENSKF
VRS
VENSKF_lowess
VRS_smooth
3e+05 2e+05 1e+05
Daily volume (m3 − sim la e
aa
0e+00
3e+05 2e+05 1e+05 0e+00
3e+05 2e+05 1e+05 0e+00 0e+00
1e+05
2e+05
0e+00
Daily volume (m3 − fiel
1e+05
2e+05
aa
Figure 9.3.2: Scatterplot between observed daily volume and simulated volumes of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake for 1997-2014. These alternate outputs were later discarded.
270
125000 RMSEWB+GR4J
100000 75000 50000 25000 0 125000
RMSEENKF
100000 75000 50000 25000 0 125000
RMSEENSKF
100000 75000 50000 25000
RMSEENSKF_lowess
0 125000 100000 75000 50000 25000 0 125000
RMSERS_smooth
100000 75000 50000 25000 0 125000
RMSERS
100000 75000 50000 25000 0 2000
2005
2010
2015
(a) RMSE on daily water volumes (m3 )
RMSE on annual mean daily water volume
30000
20000
10000
0 VWB+GR4J
VENKF
VENSKF
VENSKF_llowess
VRS
VRS_smooth
(b) RMSE on mean daily water availability per year (m3 )
Figure 9.3.3: RMSE of additional outputs including smoothed VEN KF and smoothed VRS and Ensemble Square Root KF (VEN SKF ) on Gouazine lake. These alternate outputs were later discarded. 271
Chapter 10
Appendix to chapter 5 - Forms developed for field surveys and interviews 10.1
Water survey forms
272
Inventaire rapide des usages sur les retenues de l’amont du Merguellil
Nom de la retenue : _________________________ Date de la visite :________________________ Coords GPS à l’extrémité rive droite ____________________
gauche
de la digue :
___________________
Datum : WGS 84
Altitude (sur digue) :___________m
Carthage
Prendre 1 photo du lac :
OK
Description Présence d’eau : Oui
Très faible
Non
Arbustes/forte végétation dans le fond de la retenue : Oui Déversoir :
OK
Tour de prise : OK
Non
Dommages sérieux
Inexistant
Colmatée
Inexistante
Autres dégats (digue, etc)_____________________________________________________________ __________________________________________________________________________________ Estimation hauteur digue : 10 m
Mesure_____________________________________________________________________
Usages
Observés
Renseignés
Commentaires
Pompes (nombre) Bargater (nombre) Citernes (nombre) Bidons (eau domestique)
A proximité (ca. 500 m) du lac : Habitations :
Oui
Non
Fruitiers (oliviers, amandiers,…) : Oui
Non
Cultures sous pluie :
Oui
Non
Maraichage :
Oui
Non
Culture de décrue :
Oui
Non
Bétail :
Oui
Non
Autres usages : ____________________________________________________________________
273
Inventaire rapide des usages sur les retenues de l’amont du Merguellil
Autres points d’eau à proximité (ca. 500 m) Puits
Profondeur nappe :____________ m selon riverain
mesure ce jour
Source
Si oui, à l’amont
Aval
Autre bassin
Lac
Si oui, à l’amont
Aval
Autre bassin
Prélèvements dans l’oued
Si oui, à l’amont
Aval
Autre bassin
Pompe n°
Nombre d’exploitants
1
2
3
Origine
Pompe n°
Don projet Subvention Achat Don projet Subvention Achat Don projet Subvention Achat
4
Nombre d’exploitants
Origine Don projet Subvention Achat Don projet Subvention Achat Don projet Subvention Achat
5
6
Nb total pers. exploitant lac :____________
Y a-t-il une AIC sur le lac ?
Active
Inactive
Inexistante
Si active, nb de personnes :___________ Si inactive, depuis quand ?____________ Pourquoi ?_______________________________________ __________________________________________________________________________________
Ses cultures
ha ou nb arbres
Irrigation Provenance
Technique
Fréquence : saison sèche
Oliviers Autres fruitiers Céréales Maraichage Autres Provenance = Lac : L ; Puits : P ; Source : S ; Citerne : C ; Oued : O Technique= Micro bassins : B ; Raie : R ; Inondation : I ; Goutte à goutte : G ; Aspersion : A
274
Fréquence : saison humide
10.2
Water use questionnaires
275
Questionnaires pour enquêtes détaillées sur les usages agricoles autour de 4 petits réservoirs de l’amont du Merguellil : El Guettar, El Gouazine, El Morra et Hoshas. 1. Identification et structure familiale Date de l’enquête :
Nom de la retenue :
Nom de l’enquêté :
Sexe : M
Origine géographique : Ici Nbre d’années vivant ici :
Membres du ménage Lien de parenté
Ailleurs
F
préciser Présent avant aménagement : Oui
Age
Activité au sein de l’exploitation
Non
Acivité extérieure à l’exploitation (type d’activité et nbre de mois/an)
Enquêté
Membres de la famille proche (parents, enfants, frère, sœur) vivant hors du ménage Lien de parenté Activité extérieure à l’exploitation (type d’activité et nbre de mois/an)
Bénéficiez vous d’aide familiale sur l’exploitation en dehors du ménage ? Oui
Non
1
276
2. Activités agricoles par parcelle Parcelle n° Près du lac
Ailleurs
préciser :
Indiquer la localisation de la parcelle sur la carte satellite fournie Cette parcelle est-elle :
A vous
Louée
Avez vous un titre ?
Partagée Oui
Non
Quelle année avez-vous commencé cette culture ? Pourquoi ?
Avez vous reçu des aides du gouvernement ou projet ? Oui Lesquelles :
Cultiviez vous avant ?
Oui
Non
Non
Si oui, quoi ?
Si non, que faisiez vous ?
2
277
Parcelle n° …… - Cultures en cours (hiver 2013) Culture(s) sur cette parcelle
Culture principale Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Culture 2 Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Préciser espèce/variété Superficie ou nombre d’arbres Production par arbre
parcelle
ha
par arbre
parcelle
ha
Dates de semis Dates de récolte Irrigation Provenance de l’eau
Technique d’irrigation
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
Fréquence et durée d’irrigation Période sèche Période humide Itinéraire technique Main d’œuvre
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
personnes
personnes
Quels intrants ? Labour ? Proportion pour la famille Valeur obtenue à la revente
%
% TND
TND 3
278
Parcelle n° …… - Cultures récoltées (hiver 2012 - été 2013) Culture(s) sur cette parcelle
Culture principale Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Culture 2 Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Préciser espèce/variété Superficie ou nombre d’arbres Production par arbre
parcelle
ha
par arbre
parcelle
ha
Dates de semis Dates de récolte Irrigation Provenance de l’eau
Technique d’irrigation
Fréquence et durée d’irrigation Période sèche Période humide Itinéraire technique Main d’œuvre
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
personnes
personnes
%
%
Quels intrants ? Labour ? Proportion pour la famille Valeur obtenue à la revente
TND
TND 4
279
Parcelle n° …… - Cultures projetées (été 2014) Culture(s) sur cette parcelle
Culture principale Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Culture 2 Olives (huile) Olives (table) Autres fruitiers Céréales Maraichage Autres
Préciser espèce/variété Superficie ou nombre d’arbres Production par arbre
parcelle
ha
par arbre
parcelle
ha
Dates de semis Dates de récolte Irrigation Provenance de l’eau
Technique d’irrigation
Fréquence et durée d’irrigation Période sèche Période humide Itinéraire technique Main d’œuvre
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
Lac Source Puits/forage Citerne Oued Raie Inondation Goutte à goutte Aspersion Micro bassins
personnes
personnes
%
%
Quels intrants ? Labour ? Proportion pour la famille Valeur obtenue à la revente
TND
TND 5
280
3. Activités agricoles de l’exploitation Cultivez vous d’autres parcelles ?
Oui
Non
Si oui, remplir fiche(s) n°2 supplémentaire pour chaque parcelle Superficie totale :
ha
Superficie exploitée en 2012-2013:
Superifice irriguée en 2012-2013:
ha
Si pas tout, pourquoi :
ha
Si pas d’irrigation, avez vous essayé ? Oui
Si pas tout, pourquoi :
Non
Combien de fois, années ? Que s'est il passé?
Avez-vous essayé le maraîchage ? Oui
Non
Combien de fois, années ? Que s'est il passé?
Avez-vous bénéficié de crédits ou d’aides financières sur l’exploitation ? Oui
Non
Lesquels ?
Source financement campagne agricole revenu agric
Nombre d’employés ?
revenu extra agric
permanents
aide d’un parent
crédit/emprunt
journaliers (au maximum)
6
281
Matériel utilisé dans l’exploitation A vous
En location
En association
Nombre de tracteurs Nombre de moissonneuses Nombre de camionnettes
Cheptel (combien de tetes de bêtes femelles) actuellement Bovins (vaches)
Ovins (moutons)
Caprins (chèvres)
Vendez-vous quelques animaux : régulièrement
Autres (spécifier)
exceptionnellement
Avant le lac, quelles étaient vos activités (agricoles)
7
282
4. Vos accès et usages de l’eau sur l’exploitation Votre accès à l’eau Nombre Depuis quelle année ? Pompe sur le lac
Puits/forage Profondeur : m Baisse de la nappe Oui Si oui, m sur Source
Non ans
Citerne du lac de l’oued inconnu Prélèvement direct dans l’oued
par an
Financement (plusieurs choix possibles) Don projet Subvention Achat Crédit/emprunt Don projet Subvention Achat Crédit/emprunt Don projet Subvention Achat Crédit/emprunt Don projet Subvention Achat Crédit/emprunt Don projet Subvention Achat Crédit/emprunt
Propriété
A tous A vous A vous et d’autres A quelqu’un d’autre A tous A vous A vous et d’autres A quelqu’un d’autre A tous A vous A vous et d’autres A quelqu’un d’autre A tous A vous A vous et d’autres A quelqu’un d’autre A tous A vous A vous et d’autres A quelqu’un d’autre
Accès partagé ? Oui Non Oui Non Oui Non Oui Non Oui Non
Etat actuel
Bon Hors service Quels problèmes : Bon Hors service Quels problèmes : Bon Hors service Quels problèmes : Bon Hors service Quels problèmes : Bon Hors service Quels problèmes :
Localiser sur la carte leurs accès à l’eau 8
283
Régime d’irrigation/arrosage Moyen de prélèvement
Pompe sur lac
Puits/forage
Pompe thermique(gazoil) Pompe électrique Marque Puissance Coût Pompe thermique(gazoil) Pompe électrique Marque Puissance Coût
Période sèche Durée des prélèvements
Fréquence des prélèvements
heures ou litres gazoil
Période humide Durée des Fréquence prélèvements des prélèvements
heures ou litres gazoil
heures ou litres gazoil
heures ou litres gazoil
Source heures, litres ou bidons
heures, litres ou bidons
heures, litres ou bidons
heures, litres ou bidons
Citerne
Prélèvement direct dans l’oued
Pompe thermique(gazoil) Pompe électrique Marque Puissance Coût
Moyen de distribution à la parcelle
Quelles cultures / parcelles ?
Bargater Réseau Arrosoir Autre Bargater Réseau Arrosoir Autre Bargater Réseau Arrosoir Autre Bargater Réseau Arrosoir Autre Bargater Réseau Arrosoir Autre 9
284
5. Evolution des prélèvements en eau sur le lac Prenez vous moins d’eau du lac que par le passé ? Oui Si oui, pourquoi ?
Avez-vous :
Non
diminué l’apport par culture changé de culture
Avez vous essayé le goutte à goutte ? Si non, pourquoi ?
diminué la surface irriguée changé de système d’irrigation
Oui
Non
Prenez vous plus d’eau du lac que par le passé ? Oui Si oui, pourquoi ?
Non
Avez-vous
augmenté l’apport par culture changé de culture
augmenté la surface irriguée changé de système d’irrigation
Durant la campagne 2012-2013 combien de mois ne pouviez vous pas arroser avec le lac?
Quand le lac est à sec comment faites vous pour arroser ?
Moyen d’exhaure sur la retenue (en 2013, tous les utilsateurs) Vous Autres Commentaires Pompes (nombre) Bargater (nombre) Citernes (nombre)
10
285
Usages de l’eau sur la retenue (en 2013, tous les utilsateurs) Vous Autres Commentaires Cultures Oui Non Oui Non maraîchères Ne sait pas Fruitiers
Oui
Non
Oui Non Ne sait pas
Culture de décrue
Oui
Non
Oui Non Ne sait pas
Pour cheptel
Oui
Non
Oui Non Ne sait pas
Eau domestique
Oui Non Nombre bidons/jour
Oui Non Ne sait pas
Puits à proximité
Oui Non Nombre
Autres usages
Oui Non Préciser
Oui Non Ne sait pas Nombre Oui Non Ne sait pas Préciser
6. Gestion de l’eau dans la retenue A qui appartient l’eau du lac ? Village Association
Etat
Aviez vous des terres sous le lac ? Oui Non Si oui, avez-vous reçu compensation ? Oui
Tous
Non
A combien d’autres personnes appartenaient les terres ? Qui peut exploiter l’eau du lac ? Village Qui décide de l’accès aux eaux du lac? Chef du village Association
personnes
Tous
personnes personnes
Autre
préciser
11
286
Comment répartissez vous l’accès aux eaux du lac ?
Y a-t-il ou y a-t-il eu une association (AIC/GDA) ? Oui Si oui détails, si non pourquoi ?
Non
Quelqu’un peut il augmenter ses cultures et prélèvements ? Oui Doit il demander l’autorisation ? Oui Non A qui ? Un nouvel arrivant peut il prélever ? Oui Sous quelles conditions ?
Non
Non
Combien de personnes partagent votre pompe ?
personnes
Comment se répartit l’accès sur votre pompe ?
Qui paie les entretiens/réparations ?
7. Perceptions et perspectives La retenue a-t-elle été utile ? Oui Quels sont ces défauts ?
Non
Que faudrait il faire pour améliorer le lac?
Faut il construire de nouveaux lacs ? Oui Si non, que faudrait il construire ?
Non
12
287
La disponibilité en eau a t –elle changé depuis la construction du barrage ? Augmenté Diminué Sans changement Combien d’années pourrez vous continuer à exploiter retenue?
années
Si l’eau diminue, comment vous ou vos enfants vont-ils continuer l’agriculture ? diminuer l’apport par culture diminuer la surface irriguée changer de culture préciser changer de système d’irrigation
préciser
arrêter l’agriculture Si vous en aviez la possibilité, quel investissement feriez vous ?
Entretien de la retenue Quel entretien de la retenue est fait?
Quand ? Par qui ? Qui devrait le faire ?
Connaissez vous d’autres lacs à proximité ? Oui Lesquels?
Non
Sont ils souvent en eau ? Oui Les avez-vous déjà exploités ? Oui Si non, pourquoi ?
Pourriez vous en profiter ? Oui
Non Non
Non
Si non, pourquoi ?
13
288
10.3
Semi directed interview topics
289
Framework for interviews – Small holders around small reservoirs, Merguellil upper catchment, Tunisia Identification : Location, date
Researchers & translator present Impressions
Topic Origins of the reservoir
Sub topic Before the reservoir
Questions Since when farmer lives here Prior presence of ponds of lakes To whom belonged the land? Cropped land? How the reservoir came Date of construction about Project developer (government, international aid, etc) Financing ? Surveying before construction (infiltration, etc ?) Discussion/consultation by project ? Compensation for land? Construction of the Construction Who did the works reservoir Local, external company ? Duration and construction method Local participation Did he assist with the works ? Voluntary/paid work? Other farmers? Functioning of the Water availability Frequent/regular flood reservoir Storage capacity & evolution over time Duration of the flood (when, how many months) Speed flood recedes (when, how long) Change over time ? According to which markers/observations ? Satisfaction level ? Reservoir operation Need to open the floodgate (before, during the storm) ? Who ? Spillway operated ? Training and guidelines on the behaviour of the dam/floodgate ? Management of the water Management structure Type of management ? resource Set up by whom ? Organisation Management rules Rights over the water Limits on withdrawals ? (who, when, how much) Water uses Agricultural withdrawals Period of the year, duration, frequency ? Type of pump. Origin (private purchase, supplied by govt, project etc) ? Management of shared pumps Solutions when reservoir dry Agricultural practices Irrigated crops (crop type, surface area)
290
Other users & water uses
Other agricultural/livelihood activities Size of the property ? Land rights Future agricultural/irrigation projects ? Other properties nearby Withdrawals by these users ? Number of pumps and famillies using water ? Cisterns, tanks and containers? Change over time in withdrawals & uses
Other users
Other uses
Drinking water, recreation/swimming, livestock, fishing, hunting, etc ? Other uses? Conflicts ? Problems ? Anecdotes? Where ? Names ? Know if closest ones are used, flood often ? Name (if willing) Sex Approximate age Location of house Location of irrigated cropland Location of non irrigated land Other economic activities
Miscellaneous
Other reservoirs
Identification participant
Knowledge of other reservoirs in area ? of
the
291
Part VI
Extended summary in French
292
Résumé Les retenues collinaires connaissent un essor dans les zones semi-arides pour leur capacité à réduire le transport des sédiments, et à capter les ressources pluviométriques aléatoires pour la petite agriculture. L’ampleur et l’éparpillement de ces multiples hydro-sociosystèmes limitent leur étude, entrainant de fortes incertitudes sur leur potentiel hydrique et leur influence cumulée sur les écoulements, ainsi que sur les stratégies permettant de soutenir les agriculteurs. Cette recherche vise à développer une approche multi-échelle, interdisciplinaire pour quantifier les disponibilités en eau de multiples retenues et éclairer les facteurs hydrologiques et sociaux influençant les pratiques agricoles. Un Filtre de Kalman Ensemble couplant des observations 30 m Landsat de surfaces inondées avec un modèle hydrologique journalier (GR4J + bilan hydrique) est développé sur 7 retenues jaugées. L’assimilation de données réduit l’erreur quadratique moyenne (RMSE) sur les volumes de 50% par rapport à l’ébauche du modèle pluie-débit, diminuant notamment les incertitudes générées par les précipitations intenses et localisées peu ou mal détectées. Compensant la faible résolution temporelle et des valeurs aberrantes issues des images Landsat, la méthode permet le suivi de dynamiques de crue de retenues de l’ordre de 5 ha (R =0.9). Validée sur des données de terrain de longue durée (1999-2014), la méthode démontre notamment le potentiel des images Landsat à quantifier la disponibilité en eau annuelle de retenues non jaugées dès 1 ha (RMSE autour de 25%). Appliquée à 48 retenues et 546 images Landsat 5, 7 et 8, la chaîne de traitement confirme les fortes incertitudes et pénuries en eau, qui contraignent le développement agricole sur 80% des lacs du bassin amont du Merguellil en Tunisie Centrale. La combinaison d’inventaires, d’enquêtes agricoles et d’entretiens semi-directifs atteste des prélèvements minimes mais relève la diversification des pratiques agricoles et les bénéfices périphériques accompagnant ces aménagements. De nombreux agriculteurs ne disposent pas des capabilités nécessaires pour augmenter leurs prélèvements et souffrent de problèmes physiques et économiques d’accès à l’eau, exacerbés par une gestion inefficace et un appui limité et de courte durée. Les réussites individuelles recensées témoignent de la résilience économique de certains agriculteurs, disposant de ressources complémentaires pour faire face aux pénuries. Au vu des capacités limitées et des sécheresses durables, les retenues collinaires dans ce contexte climatique doivent maintenir leur objectif initial d’irrigation de complément et non chercher à soutenir une intensification à plus grande échelle de l’agriculture irriguée.
293
Résumé étendu Contexte Les retenues collinaires connaissent depuis plusieurs décennies une forte expansion, favorisée par le soutien de projets gouvernementaux et internationaux en zones semi-arides, notamment au Brésil (Burte et al. 2005), au Mexique (Avalos 2004), en Inde (Bouma et al. 2011), et en Afrique du Nord et subsaharienne (Talineau et al. 1994; Sawunyama et al. 2006). Combinés avec d’autres techniques de conservation des eaux et des sols (CES), ces aménagements permettent de réduire l’érosion et l’envasement des barrages situés en aval, mais également de mobiliser des ressources en eau pour les utilisateurs à l’amont, voire de favoriser la recharge de nappes. Dans les zones où les précipitations sont limitées et irrégulières, ces ouvrages de faible coût ont le potentiel de soutenir les pratiques de petits agriculteurs (Vincent 2003) et ne souffrent pas des mêmes effets négatifs que les projets d’irrigation de grande envergure (Wisser et al. 2010). Modifiant la distribution spatio-temporelle des ressources, ces retenues ont toutefois une influence sur les écoulements à l’échelle locale et en grand nombre peuvent réduire les écoulement à l’échelle d’un bassin versant de 1% à 50% (He et al. 2003; Gao et al. 2011; Lacombe 2007; Kingumbi et al. 2007; Ogilvie et al. 2016) (cf. chapitre 6). Face aux pressions climatiques et anthropiques sur les ressources en eau en zone semi aride et le besoin d’optimiser la gestion de ressources, les bénéfices de ces retenues sont parfois remises en question (Mushtaq et al. 2007; Le Goulven et al. 2009). Des études ont montré de multiples bénéfices à ces systèmes multi-usages (agriculture, élevage, pêche...) mais les productions agricoles associées demeurent parfois faibles (Khlifi et al. 2007; Habi and Morsli 2011; Selmi and Talineau 1994; Faulkner et al. 2008), soulignant l’importance d’approches adaptées pour étudier ces systèmes (Venot and Cecchi 2011). La littérature sur les facteurs qui favorisent ou contraignent le développement de l’agriculture autour des retenues demeure limitée mais souligne l’importance de facteurs socio-économiques, des structures de gestion et des politiques publiques (Mugabe et al. 2003; Zairi et al. 2005; Talineau et al. 1994) en plus des contraintes hydrologiques. Les interactions entre ces facteurs demeurent peu identifiées, étant souvent étudiées individuellement au travers d’études monodisciplinaires en agronomie, sociologie ou hydrologie. La multiplication récente de travaux en socio-hydrologie (Sivapalan et al. 2012; Montanari et al. 2013; Sivakumar 2012) témoigne de l’importance et des difficultés à étudier les relations complexes entre l’eau et les sociétés. Ce concept, inhérent aux recherches existantes sur l’étude des influences anthropiques dans les systèmes hydrologiques et sur les interactions entre eau et sociétés, souligne notamment le besoin d’incorporer les outils d’autres disciplines afin de fournir de nouveaux éclairages, voire de soulever de nouvelles questions
294
de recherche (Riaux and Massuel 2014). A l’exception de quelques retenues instrumentées à des fins de recherche (Albergel and Rejeb 1997), les disponibilités en eau de retenues collinaires demeurent souvent inconnues pour des raisons logistiques et économiques, compte tenu de leur nombre, importance et éparpillement. Ce manque de données empêche notamment les gestionnaires de comprendre le potentiel de chaque retenue et de sélectionner les sites où privilégier leurs investissements. De même, des informations sur les volumes, périodes et durées de crues peuvent aider les usagers à optimiser leurs stratégies agricoles (Liebe et al. 2005; Mugabe et al. 2003). Des études hydrologiques et géochimiques ont permis de simuler le bilan hydrique de retenues (Grunberger et al. 2004; Li and Gowing 2005) mais identifient des incertitudes fortes sur l’infiltration et les flux anthropiques (prélèvements, lâchers). La forte variabilité spatiale des pluies génère également des difficultés à modéliser les apports aux retenues et l’hétérogénéité des sous-bassins et des dynamiques de chaque retenue rend la modélisation de retenues non jaugées délicate (Lacombe 2007; Cudennec et al. 2005). Le développement continu de nouveaux capteurs satellites procurant des images à faible coût et aux résolutions spatiale, temporelle et spectrale grandissantes permet de nombreuses applications en hydrologie. Utilisés dans l’étude hydrologique de zone humides ou de grands lacs (Ogilvie et al. 2015; Swenson and Wahr 2009; Wolski and Murray-Hudson 2008), ils ont également été appliqués sur des petits plans d’eau de plus de 1 ha (Feng et al. 2015; Gardelle et al. 2009; Liebe et al. 2005). Ces travaux témoignent toutefois d’incertitudes et les possibilités offertes par de nouveaux capteurs, indices et traitements, doivent être étudiées dans différents contextes et adaptées aux particularités de ces petits objets. De plus le potentiel de l’imagerie satellite pour suivre les dynamiques de crues et quantifier les disponibilités en eau de multiples petites retenues n’a pas été identifié (Baup et al. 2014). Ceci impose notamment des contraintes spécifiques en termes de précision, de stabilité des indices, et de fréquence d’images qui doivent être compatibles avec la résolution temporelle et spatiale des images disponibles, et requiert la présence de suffisamment de données de terrain pour valider la méthode. Enfin, des approches permettant de combiner ces nouvelles sources de données avec des approches hydrologiques traditionnelles doivent être étudiées (Wackernagel 2004; Winsemius 2009). Les observations satellites peuvent notamment être exploitées comme variable d’entrée ou d’état de modèles qui permettent de corriger en continu le fonctionnement du modèle au travers de techniques d’assimilation de données, tels que les filtres de Kalman (Moradkhani et al. 2005; Clark et al. 2008).
Objectifs Dans ce contexte, ces travaux de recherche visent à: Étudier la disponibilité en eau des retenues collinaires et identifier leur influence sur les pratiques des petits agriculteurs en zone semi-aride. Plus précisément, les objectifs étaient de développer une approche permettant de: • Quantifier la disponibilité en eau dans de multiples retenues collinaires (non jaugées) • Caractériser les usages de l’eau de multiples retenues collinaires
295
• Identifier les autres facteurs (historiques, socio-économiques, institutionnels) influençant le développement des usages autour des retenues collinaires Ces travaux se sont focalisés sur le bassin amont du Merguellil (1200 km ), en Tunisie Centrale, où les stratégies nationales de conservation des eaux et des sols depuis plus de 50 ans ont entrainé la construction d’une cinquantaine de retenues collinaires. Au niveau local, ces études doivent notamment contribuer aux réflexions menées depuis plusieurs années sur la pertinence de ces aménagements, compte tenu des fortes pertes par évaporation et faibles usages observés, et de la forte demande en eau des utilisateurs exploitant les ressources (souterraines) à l’aval du bassin. Cette recherche s’inscrit dans la coopération entre l’IRD et la Direction Générale de l’Aménagement et de la Conservation des Terres Agricoles (DG ACTA) et les partenaires locaux du Commissariat Régional au Développement Agricole (CRDA) de Kairouan et Siliana. Faisant l’objet de travaux de recherches depuis plus de 40 ans, ce bassin bénéficie d’un réseau significatif de données hydro-climatiques à l’échelle des retenues et du bassin, permettant la calibration et validation de méthodes de télédétection et de modélisation numérique.
Méthode Régionaliser les dynamiques hydro-sociologiques L’un des enjeux de ces travaux fut de développer une approche originale permettant d’étudier à la fois les dynamiques hydrologiques et sociales (ici usages de l’eau et facteurs d’influence) sur une large échelle (ici toutes les retenues d’un bassin versant). L’approche proposée combine l’instrumentation et les observations hydrologiques, la modélisation numérique, la télédétection et les enquêtes agricoles/agronomiques et ethnographiques, et développe une approche multi-échelle afin de fournir différents degrés d’analyses, pour passer d’observations sur une retenue à une cinquantaine de retenues dans le bassin versant. La complexité réside également dans la multiplicité des frontières des objets étudiés (par ex. prélèvements sur les lacs, recharge de nappes, bénéfices ou pratiques agricoles d’une communauté), exacerbée par la confrontation de plusieurs disciplines, où les influences institutionnelles peuvent opérer à une échelle différente de celle d’un ouvrage hydraulique ou d’un bassin versant (Riaux et al. 2014a).
Quantifier les disponibilités en eau Après un 1er chapitre dévolu à la présentation du contexte et des objectifs de la thèse, le 2ème chapitre fournit un aperçu du site d’étude, et des principales données hydro-climatiques exploitées. Le chapitre 3 développe ensuite une méthode pour délimiter les surfaces en eau dans toutes les retenues du bassin à partir d’images Landsat 5, 7 et 8. Celle ci repose sur la définition d’une chaine de traitements radiométriques adaptée aux bandes spécifiques à chaque capteur et sur la quantification et optimisation du nombre d’images exclus à cause de nuages, de leur ombre et de la défaillance du Scan Line Corrector (SLC) sur le capteur ETM+ de Landsat 7. Sept indices couramment utilisés pour la télédétection de l’eau sont comparés afin d’étudier leur performance sur des retenues collinaires qui présentent des difficultés particulières liées 296
à leur petite superficie et à la présence de nombreux pixels où les signatures spectrales de sols humides, végétation plus ou moins inondée, et eau libre se mélangent. Les indices sont calibrés et validés à l’aide des mesures de terrain au DGPS des surfaces en eau de 7 lacs sur trois dates. Ils sont ensuite évalués sur leur précision en hautes eaux et basses eaux ainsi que sur la stabilité des valeurs d’indice, condition nécessaire pour automatiser la méthode sur tous les lacs et toutes les images. La chaine de traitement appliquée aux 546 images entre 1999 et 2014 permet enfin d’évaluer le potentiel de la méthode pour suivre les dynamiques de crue et pour caractériser les variations annuelles des surfaces en eau (prélude à la disponibilité en eau annuelle puis interannuelle), à l’aide de chroniques d’observations de terrain disponibles pour 7 lacs sur plusieurs années (entre 2 et 15 ans). Le chapitre 4 développe ensuite une approche pour coupler ces informations satellitaires avec un modèle hydrologique des retenues (pluie-débit+bilan hydrique), afin d’estimer au plus près la disponibilité en eau de tous les réservoirs. A partir de l’instrumentation de 4 retenues et d’une base de données acquise sur 13 autres à proximité au cours des années 1990 et 2000, les multiples flux nécessaires au bilan hydrique des retenues sont interpolés (pluie, évaporation) ou estimés (infiltration, prélèvements, lâchers) à partir des observations. Les écoulements entrants sont estimés à partir d’un modèle GR4J fonctionnant au pas de temps journalier. De type conceptuel et global, le modèle est choisi pour sa robustesse et le faible nombre de variables et de paramètres, facilitant son application dans un contexte de données limitées et la transposition aux bassins non jaugés. Un Filtre de Kalman d’ensemble est ensuite développé afin de corriger l’ébauche du modèle GR4J+bilan hydrique à l’aide des observations Landsat. La performance de l’approche à restituer les dynamiques hydriques ainsi que la disponibilité en eau annuelle est évaluée sur les 7 mêmes retenues collinaires. Les résultats sont également comparés à ceux obtenus à partir uniquement des données spatiales et des modèles hydrologiques. Enfin la sensibilité des résultats à l’augmentation des incertitudes sur la pluviométrie, l’infiltration, le transfert de paramètres et l’envasement des retenues (modification des relations hauteursurface-volume) est évaluée afin d’appréhender les incertitudes liées à l’application de la méthode sur des bassins non jaugés. Les résultats sont ensuite appliqués dans le chapitre 5 à chacune des retenues du bassin afin d’estimer les disponibilités en eau annuelles et sur la saison sèche, ainsi que leurs variations sur la période 1999-2014. Un indice évaluant le nombre de jours par an et sur la saison sèche où le volume disponible est inférieur aux besoins agricoles est ensuite défini afin de fournir une typologie du potentiel hydrique de chaque retenue à soutenir l’agriculture irriguée.
Usages et contraintes hydro-sociologiques En parallèle, les usages de l’eau sont étudiés dans le chapitre 5 au travers d’inventaires rapides, d’enquêtes de pratiques et d’entretiens semi-directifs sur des échantillons de taille variable de retenues. Les inventaires rapides permettent d’identifier le type d’usages ainsi que le nombre de pompes menant à une première caractérisation des retenues. Des enquêtes sont ensuite menées sur 22 retenues afin de caractériser les pratiques et prélèvements des agriculteurs et comprendre l’influence des problèmes d’accès à l’eau, des conflits, des stratégies agraires ou de l’appui institutionnel. Des entretiens semi-directifs sont également
297
menés sur 4 retenues, sélectionnées sur la base de critères physiques et sociaux. Ceux-ci explorent notamment l’histoire du site, les origines de l’aménagement, sa gestion et ses usages, et permettent de décrire avec plus de finesse l’influence des associations d’intérêt collectif (groupe d’utilisateurs), des droits fonciers, des difficultés économiques, des aides publiques et des conflits. Empruntés à l’ethnographie, ces entretiens permettent une étude détaillée mais par nature localisée et ils ne sont pas destinés à être multipliés à volonté (Beaud 1996). Les enquêtes permettent ainsi d’appréhender la variabilité et la distribution spatiale des pratiques et contraintes, tandis que les entretiens fournissent un éclairage détaillé sur les origines et conséquences de celles-ci. Cette approche doit expliciter l’hétérogénéité des pratiques entre les retenues mais également au sein d’une même retenue, compte tenu des inégalités (économiques et politiques) qui vont modifier profondément les dynamiques d’usages de chaque agriculteur. Suite à la description et caractérisation des bénéfices fournis par les retenues, les incohérences entre les usages et le potentiel hydrique calculé à l’aide des résultats des chapitres 3 et 4 sont mises en exergue. Les nombreuses influences socio-économiques et institutionnelles qui promeuvent ou freinent l’usage de l’eau autour des retenues sont enfin discutées.
Principaux résultats et discussion Chapitre 3 - Télédétection des surfaces en eau dans les retenues collinaires Ce chapitre visait à caractériser le potentiel d’images Landsat gratuites de moyenne résolution spatiale (30 m) et temporelle (16 jours) pour quantifier au cours du temps les variations des surfaces en eau dans tous les retenues et développer une approche adaptée. Comparé à 6 autres indices de détection de l’eau, l’indice MNDWI permet de quantifier avec la même valeur d’indice les surfaces en eau de 7 retenues de différentes tailles et dans différentes conditions (3 images en hautes eaux et basses eaux) avec la plus grande précision (figure 3.2.5). Les erreurs en terme de surface varient de 1% à 28% (moyenne 10,5%, écart type 8,7%) et même sur les plus petites retenues (autour de 0.5 ha) les erreurs ne dépassent pas 14% et 8%. Les taux de précision obtenus par les matrices de confusion sont supérieurs à 87% sur 6 des 7 retenues. D’autres indices tel que le NDTI ont été très performant (en calibration) pour détecter les petites retenues mais étant plus sensibles aux variations de l’indice ils entrainent des erreurs plus significatives (en validation) lors de leur transposition sur d’autres images. Lors de l’application de la méthode à 546 images Landsat, les erreurs moyennes en terme de surface demeurent en deçà de 20% mais la présence de valeurs aberrantes réduit notamment les valeurs de R sur certaines retenues. Ces erreurs proviennent d’une part des difficultés à détecter les nuages de type cirrus et leur ombre. L’ajout d’une nouvelle bande pour leur détection sur Landsat 8 et les recherches continues pour le développement de meilleurs outils (Zhu and Woodcock 2014), témoignent de la reconnaissance de ces sources d’erreurs. Sur les petites retenues, la proportion plus importante de pixels mélangeant eau, végétation, et sols ainsi que l’importance proportionnellement plus importante de chaque pixel mal classé réduit d’autre part les valeurs de R (autour de 0.7). Sur certains lacs profonds et de grande taille, des difficultés à suivre les dynamiques apparaissent également à cause de vari298
ations plus restreintes des surfaces en eau mais restent contenues dans l’intervalle d’erreur (10-20%), contrairement à d’autres lacs où l’amplitude des surfaces est plus importante (65 000 m à 100 000 m et 0 m à 90 000 m respectivement). La calibration du pourcentage de pixels affectés par des nuages (et leur ombre) et par la défaillance du Scan Line Corrector (SLC-off) sur Landsat 7 mène à retenir en moyenne 282 ±27 images sur les 546 traitées. Les images avec plus de 25% de SLC-off pixels ou plus de 40% de nuages au dessus de la retenue sont retirées. Cette exclusion de 51% des images, équivaut à conserver en moyenne 1,5 images/mois sur 1999-2014 bien qu’en pratique les écarts soient plus faibles ces dernières années à cause des acquisitions concomitantes de Landsat 5 et 7 puis Landsat 7 et 8. Sur un lac donné, 59% des images Landsat 7 SLC-off étaient inutilisés, et 31% de toutes les images à cause de nuages et de leur ombre. Ces valeurs varient en fonction de l’altitude de certain lacs et de la présence irrégulière des interférences SLC qui sont plus importantes sur les bords de l’image. Ici, les méthodes de correction qui s’appuient sur l’interpolation spatiale et/ou temporelle ne sont pas adaptées compte tenu de la faible dimension des retenues et de la forte variabilité des surfaces en eau entre les images, contrairement aux autres occupations du sol pour lesquelles ces méthodes sont développées (Chen et al. 2011; Zeng et al. 2013). L’absence d’observations fréquentes entraine notamment la perte d’information sur les dynamiques de crues rapides, sur les petites retenues et celles à forte infiltration. Deux crues sur 7 ne sont pas détectées sur les plus petits lacs et deux autres crues sont identifiées après une décrue de 50%, confirmant l’intérêt des images à plus haute résolution temporelle pour un suivi de crue. Malgré ces difficultés sur certaines dynamiques de crue, la méthode se montre efficace pour estimer et comparer les variations interannuelles des surfaces en eau de tous les lacs, avec une forte correspondance entre les valeurs estimées par télédétection et par les données de terrain (R =0.88, figure 3.2.19). Des images à plus haute résolution spatiale mais également temporelle permettront de réduire par la suite ces sources d’incertitudes. L’influence des valeurs aberrantes dépend en effet également du laps de temps avant la correction apportée par la prochaine observation. Confirmant le potentiel d’images Landsat pour l’étude des surfaces en eau de retenues collinaires (dès 0.5 ha), le chapitre suivant vise à combiner l’information spatiale avec un modèle hydrologique afin de pallier leurs limitations respectives et améliorer l’estimation des disponibilités en eau.
Chapitre 4 - Modélisation des disponibilités hydriques à l’aide d’un Filtre de Kalman d’ensemble (ENKF) Sur des bassin jaugés, l’intégration des observations Landsat par un filtre de Kalman d’ensemble dans un modèle hydrologique (GR4J+bilan hydrique) fournit des corrections déterminantes sur l’amplitude des crues mal estimées par le modèle pluie débit (GR4J). Celui ci souffre notamment des précipitations intenses et localisées, caractéristiques de cette zone semi aride, et peu ou mal détectées par le réseau pluviométrique (cf. section 4.4.4.2). Les erreurs de l’ébauche (du modèle) sur les amplitudes des crues ont des conséquences sur toutes les valeurs de décrue, que le ENKF permet de corriger, augmentant notamment le R (par ex. de 0.57 à 0.82). Les corrections sont d’autant plus importantes sur des lacs qui se vidangent lentement, où les surestimations du modèle GR4J ne permettaient pas au lac de se vidanger entre deux évènements, entrainant une dérive progressive des résultats. Inversement, la con299
naissance fine du bilan hydrique (notamment l’infiltration, les prélèvements, etc.) permet d’améliorer la représentation des dynamiques de décrue mal restituées par les observations Landsat trop peu fréquentes. Sur les valeurs de disponibilité annuelle, les erreurs quadratiques moyennes (RMSE) diminuent en moyenne de 30% par rapport aux estimations faites à partir exclusivement d’observations Landsat et de 68% par rapport aux estimations du modèle hydrologique seul (table4.4.2). Comme le montre l’étude de sensibilité, lorsque la connaissance du milieu se dégrade et que les incertitudes sur le modèle deviennent importantes, les corrections du ENKF sont encore plus utiles et parviennent à contenir les erreurs de NRMSE en dessous de 50% sur tous les lacs (table 4.4.4). En l’absence de stations dans le bassin, les précipitations peuvent être sous-estimées jusqu’à 60-70% et se combinent avec les difficultés du modèle à prendre en compte les états de surface, le cumul pluviométrique d’évènements sur plusieurs jours ainsi que les aménagements anthropiques mal représentés par un modèle global (cf. section 4.4.4). Ces difficultés confirment la complexe modélisation hydrologique de ces petits objets et les disparités entre les besoins de données à haute résolution spatiale et temporelle et leur disponibilité. La plus-value apportée par le modèle diminue également et lors de la transposition vers les bassins non jaugés, les RSME sur les volumes annuels ne sont que 10% inférieurs à ceux obtenus par simple interpolation des observations Landsat. En l’absence de données complémentaires permettant d’adapter les variables et paramètres du modèle (notamment précipitation, infiltration, paramètres GR4J) il demeure ainsi plus adapté d’estimer les variations interannuelles de volumes sur les retenues non jaugées à l’aide exclusivement des données Landsat. Les erreurs sont de l’ordre de 10 000 m3 sur des volumes annuels de l’ordre de 30 000 m3 à 60 000 m3 (table 4.4.2). Sur les plus petites retenues ou celles qui s’assèchent rapidement par une forte infiltration, les erreurs sont proportionnellement plus importantes mais permettent toutefois de fournir un ordre de grandeur suffisant pour comparer leur potentiel hydrique, indiquant une disponibilité moyenne de 1000 m3 comparé à plus de 30 000 m3 pour les autres. Malgré les nombreuses sources d’incertitudes, cette méthode permet ainsi d’estimer et de comparer les disponibilités interannuelles sur tous les lacs du bassin (figures 4.4.17 et 4.4.18). Par ailleurs, la méthode montre une forte fiabilité à identifier le nombre de jours où le lac est à sec (figure 4.4.10). Les erreurs de télédétection (nuages, végétation, etc.) discutées dans le chapitre 3 sont ici partiellement modérées par la connaissance du volume maximal mais des ajustements du ENKF sont également possibles afin de paramétrer les covariances (Moradkhani et al. 2005) et attribuer une confiance supérieure aux observations Landsat sur des images plus nettes (moins de nuages, aérosols) ou sur des plus grands lacs. L’approche du filtre de Kalman peut également chercher à modifier les variables ou paramètres d’entrée du modèle plutôt que les sorties, permettant ainsi de corriger les valeurs de pluie ou des paramètres du modèle GR4J. La traduction des surfaces en volumes est effectuée à l’aide d’une relation surface-volume développée à partir d’observations sur 13 lacs. Les incertitudes associées dues à l’envasement progressif mais hétérogène des retenues sont discutées en section 2.4. Celles-ci pourraient être réduites à l’aide de mesures de terrain complémentaires, et notamment des méthodes aéroportées (drone, cerf volant) (Massuel et al. 2014a) qui ouvrent de nouvelles possibilités pour restituer à moindre frais (temps, équipement) les différences géomorphologiques et l’envasement au cours du temps. 300
Chapitre 5 - Bénéfices et influences hydro-sociologiques sur les usages des retenues Usages et bénéfices L’analyse des trajectoires des pratiques identifie les modifications subtiles qui suivirent la construction du barrage avec 58% des agriculteurs sondés qui commencèrent la culture d’arbres fruitiers, sachant que 20% abandonnèrent également la céréaliculture et l’élevage, en partie à cause de réduction de leurs terres par héritage et une fertilité des sols en déclin. Le nombre d’arbres fruitiers varie d’une douzaine à plus de 4400 par exploitation (moyenne 720 par exploitation), témoignant de la forte hétérogénéité entre agriculteurs sur les même retenues (figure 5.2.1). Ces arbres, essentiellement des oliviers (61%), fournissent suffisamment d’huile pour la consommation familiale (de 100 à 300 litres) et les bonnes années 65% des enquêtés affirment pouvoir vendre un excédent. Le maraichage, tenté par 65% des agriculteurs enquêtés, n’est actuellement pratiqué que par 20% des agriculteurs et essentiellement pour une consommation familiale sur 0,5 ha en moyenne. 60% des agriculteurs complètent leurs revenus avec des activités extra-agricoles (ouvriers, commerçant, fonctionnaire). Les ressources en eau de 66% des lacs sont exploitées par les riverains mais ces chiffres masquent les faibles prélèvements observés, essentiellement limités à l’arrosage occasionnel des arbres, l’irrigation plus intensive demeurant exceptionnelle. De fait, seulement 43% des retenues sont équipées avec des pompes (et seulement 1 pompe par lac en moyenne) tandis que les autres sont exploités à l’aide de citernes ou simples bidons. Ces points d’eau profitent également aux 70% de riverains pratiquant l’élevage. Une contribution de ces lacs sur les puits exploitant la nappe phréatique est recensée sur 12 lacs, témoignant du soutien indirect des ces ouvrages sur l’agriculture irriguée menée à proximité (section 5.2.1.5). Malgré quelques (4) lacs qui restent en eau en continu, aucune pisciculture n’est pratiquée. Enfin, ces lacs qui s’envasent progressivement, complètent leur rôle de réduire l’érosion des terres et grands barrages à l’aval, capturant quelque 200 000 m3 par an. Des usages récréatifs sont également recensés et les entretiens relèvent l’importance des retenues pour la construction et le maintien du groupe social (Riaux et al. 2014b). La combinaison des approches qualitatives et quantitatives a permis d’identifier les fortes disparités entre retenues et entre agriculteurs au cours du temps et débouche sur une typologie de 56 retenues en termes de bénéfices pour les riverains: • Retenues aux bénéfices négligeables : 16 lacs qui en l’absence de prélèvements ou de recharge de puits à proximité, conservent essentiellement une fonction de protection en retenant les eaux et sédiments lors de crues. • Retenues aux bénéfices résiduels : 27 lacs où des prélèvements occasionnels (ou des puits à proximité pour 6 lacs) fournissent des bénéfices marginaux, irréguliers mais non négligeables aux riverains. Ceux ci cultivent en moyenne 300 arbres fruitiers et parfois 0,5 ha de maraichage pour leur consommation personnelle. • Retenues favorisant un ou plusieurs entrepreneurs : 13 lacs où des prélèvements modestes mais réguliers soutiennent le développement d’exploitations isolées mais significatives. Ceux-ci cultivent en moyenne 900 arbres fruitiers (jusque 4400) et disposent de 2,5 pompes/lac. Certains ont favorisé des essais plus significatifs de maraichage (2-5
301
ha) mais l’accent demeure sur les arbres fruitiers, qui sont même irrigués par goutte à goutte sur 3 exploitations. Influences hydro-sociologiques L’approche développée dans les chapitres 3 et 4 permet l’estimation des disponibilités interannuelles de 48 lacs sur l’année et la saison sèche. 8 retenues, construites récemment (comme Mdinia 2 en 2012) ou pour lesquelles les données de capacité maximale n’étaient pas disponibles (lacs mineurs comme Bouksab 2) étaient exclues de cette analyse. A l’aide d’un indicateur du nombre de jours durant la saison sèche où le volume est supérieur à 5000 m3 , trois classes de retenues sont identifiées (cf. section 5.1.1.2 et 5.2.4) : • Disponibilité en eau négligeable : 20 lacs qui par leur taille et faible profondeur ne permettent pas de conserver suffisamment de ressources durant la période sèche pour permettre l’arrosage de cultures. • Disponibilité en eau aléatoire : 22 lacs, souvent de petite capacité (< 40 000 m3 ) mais pas exclusivement (230 000 m3 et 500 000 m3 ), qui peuvent soutenir l’agriculture mais souffrent de forte variabilité inter et intra annuelle (étant à sec entre 1,5 et 4,5 mois de la saison sèche en moyenne). Les incertitudes sur l’envasement des retenues sont fortes pour 13 lacs. • Disponibilité en eau bonne : 6 lacs, de capacité supérieure à 500 000 m3 , où la disponibilité devient fiable (plus de 4,5 mois sur 6 en moyenne) et qui confirment l’importance du stockage dans ces zones semi arides aux crues limitées. Les incohérences fortes entre les disponibilités hydriques de chaque lac et l’intensité des pratiques (cf. figure 5.2.17) confirment le besoin de prendre en compte d’autres facteurs discutés en section 5.2.5. Bien que facteur limitant, la disponibilité hydrique n’est pas une condition suffisante pour éclairer les disparités de pratiques observées sur les retenues. Sur les lacs aux ressources aléatoires, les agriculteurs doivent élaborer des stratégies pour composer avec la forte variabilité et les risques associés. Pour la plupart cela entraine la perte de production et l’abandon de pratiques, tandis que ceux avec d’autres ressources économiques (pour acheter des citernes) ou physiques (accès à d’autres points d’eau) ont survécu aux pénuries d’eau et ont pu développer leurs activités agricoles. Où la disponibilité est bonne, des succès isolés sont recensés, mais ces lacs souffrent de faible équité d’accès, ayant surtout renforcé les individus possédant du capital ou une autre source fiable de revenus. L’absence d’appui sur le long terme de l’état pour développer et structurer les activités agricoles et la gestion des ressources limite le nombre de bénéficiaires, à cause des problèmes d’accès aux pompes, tuyaux, réparations et des conflits. Certains agriculteurs ont été exclus d’emblée tandis que d’autres ont été écartés progressivement, à la suite de conflits, de mauvaise gestion ou d’appropriation des pompes. Le développement de l’irrigation suppose une approche intégrée qui prend en compte les besoins des utilisateurs depuis la sélection des sites et des bénéficiaires, jusqu’au maintien de ressources financières pour aider à l’entretien, aux réparations voire même pour développer des assurances en cas de perte de production. Les incohérences observées ici ont développé des attentes fortes des agriculteurs mais les moyens mis en œuvre permettent seulement le
302
développement d’une activité subsidiaire et la majorité des agriculteurs continuent à vivre des céréales, de l’élevage et de contrats saisonniers comme ouvriers.
Conclusions finales et perspectives Ressources en eau Régionalisation des disponibilités en eau des retenues Exploitant des images Landsat gratuites de 30 m de résolution, cette approche a permis le suivi de la disponibilité en eau entre 1999 et 2014 sur des lacs de l’ordre de 1 ha à 12 ha de superficie. Les surfaces en eau estimées par télédétection fournissent des informations sur l’amplitude et la durée de la crue et combinées avec des données de terrain pour développer des relations surface-volume locales, quantifient les variations interannuelles des volumes moyens journaliers avec une précision de l’ordre de 10 000 m3 . En présence de données de terrain supplémentaires, un filtre de Kalman d’ensemble permet de combiner les images Landsat avec un modèle hydrologique, améliorant la modélisation de la décrue et corrigeant en partie les erreurs issues de la télédétection. L’hétérogénéité observée ici sur les flux du bilan hydrique ainsi que la forte variabilité spatiale et temporelle des pluies explique certaines des difficultés à modéliser ces bassins anthropisés, et soulignent d’autant plus la valeur ajoutée des observations satellites. Cette méthode présente un potentiel d’application opérationnel et peut être transposée à d’autres bassins pour étudier et suivre la disponibilité sur le long terme de retenues qui ne peuvent pas être instrumentées pour des raisons économiques et logistiques. Compte tenu de l’intérêt porté à ces retenues au niveau mondial (Wisser et al. 2010) pour leur faible coût et soutien aux petits agriculteurs, cette méthode peut permettre d’informer les usagers et décideurs du potentiel hydrique de ces retenues et cibler les investissements en conséquence. Ici la méthode permet notamment de mettre en exergue la disponibilité limitée et forte variabilité qui affecte plus de 80% des retenues. Perspectives de recherche Compte tenu de l’objectif exigeant d’exploiter des images de moyenne résolution spatiale sur des objets de petite taille, ces résultats sont encourageants et des recherches complémentaires permettront d’éclairer le potentiel des produits de plus haute résolution temporelles et spatiales (Sentinel-2, Swot). Des améliorations dans la détection des nuages et en altimétrie aérienne permettront également de réduire respectivement certaines aberrations et les incertitudes liées à l’envasement. De manière générale, le niveau actuel des performances et des complémentarités confirme la valeur grandissante de l’imagerie satellitaire en hydrologie et son exploitation ouvre de nouvelles perspectives (Winsemius 2009; Wackernagel 2004). Ces approches continueront toutefois à dépendre d’instrumentations et d’observations de terrain pour calibrer, valider et corriger les valeurs obtenues par télédétection. Ces résultats ouvrent également des perspectives en modélisation hydrologique, exploitant les volumes estimés dans chaque retenue pour estimer les débits en de multiples lieux du bassin (Liebe et al. 2009). Dans ce bassin, ils permettront à l’aide d’un modèle semi-distribué d’éclairer le rôle des retenues collinaires dans la
303
réduction des débits observés (Lacombe et al. 2008; Ogilvie et al. 2016; Kingumbi et al. 2007). Les chroniques de surfaces en eau peuvent également permettre d’affiner l’évaporation de toutes les retenues et le bilan hydrique à l’échelle du bassin. Ces résultats alimenteront en partie les discussions sur l’optimisation des ressources entre l’amont et l’aval, plébiscitée par la forte demande agricole de la plaine de Kairouan et les inquiétudes sur la forte évaporation et la faible valorisation de retenues à l’amont (Lacombe 2007; Le Goulven et al. 2009). Celles ci pourraient in fine mener à la définition de nouveaux scénarios de gestion, visant à optimiser la recharge pour favoriser les utilisateur de l’aval voire également pour favoriser le développement de puits à l’amont comme dans d’autres régions (Zammouri and Feki 2005; Massuel et al. 2014b).
Usages de l’eau et contraintes Bénéfices plus large des retenues Les prélèvements en eau demeurent minimes et permettent essentiellement l’arrosage de quelques arbres fruitiers. La valeur de ces aménagements ne peut cependant se réduire à de simples mesures d’efficience ou de performances, toujours subjectives (Venot and Cecchi 2011). L’utilisation d’un cadre plus large permet notamment de dépasser les conclusions initiales de sous exploitation, identifiant les modifications des pratiques, les bénéfices plus larges apportés aux riverains (citernes, élevage, recharge de puits), ainsi que l’intérêt porté par les riverains pour ces ressources. Malgré peu de maraichage et de cultures de rente, les ressources en eau des retenues ont favorisé une diversification des pratiques et une augmentation de l’arboriculture, qui sur le long terme augmentera leurs revenus (Selmi and Zekri 1995). Les récoltes d’olives fournissent de l’huile d’olive pour la famille étendue voire un revenu les années d’excédent. Les recherches ethnographiques associées montrent la dimension sociale et politique de cette ressource et les riverains sont attachés à leur retenue, notamment dans les zones reculées où les conditions de vie sont difficiles. Enfin sur quelques lacs, quelques entreprises individuelles d’arboriculture intensive se sont greffées. De la même manière que certaines aides de l’état soutiennent des secteurs agricoles déficitaires ou non compétitifs, ces investissements sont ici venus en aide à des zones rurales reculées aux sols érodés peu fertiles, et aux pluies faibles et éparses. Malgré une approche parfois peu cohérente et souffrant de faible équité, ils ont contribué à ralentir quelque peu l’exode rural, «sans ce lac, la région meure». Plus qu’un accès à l’eau, c’est l’investissement et l’attention portés sur ces territoires qui a peut être permis de réduire l’exode et généré l’engouement des riverains pour ces objets aux ressources parfois peu exploitées. Les gens de la région disent que «l’oued Merguellil appartient aux montagnards de l’amont mais que les bénéfices sont pour les gens de Kairouan»; ces projets qui ont soutenu des développements autour des retenues sont un pas en avant pour aider ces petits agriculteurs (Vincent 2003) et doivent être encouragés. Stratégies sur le long terme La majorité des retenues qui dépendent d’une pluviométrie faible et aléatoire sont toutefois limitées dans leur potentiel à développer une agriculture intensive de grande échelle. Les retenues peuvent combler des pénuries durant la saison sèche et les bonnes années soutenir une intensification à condition d’optimiser la production et gérer le risque. Cependant, et ce 304
malgré l’évolution des discours (Venot and Krishnan 2011), ces ouvrages demeurent destinés à une irrigation de complément, et leur faible capacité qui les rend inexploitables les années de faible pluviométrie, empêche toute transition durable vers l’agriculture irriguée. Or «les gens savent vivre les années de pluie, c’est lorsqu’il n’y en a pas que c’est un problème». Ces retenues doivent alors être intégrées dans une stratégie d’accès à l’eau plus vaste, à l’instar de ceux qui ont pu intensifier leurs activités à l’aide de citerne, nappe ou source. Pour les agriculteurs rencontrés, la solution demeure un forage, qui symbolise à leurs yeux, une source d’eau fiable et continue («le lac c’est ok pour un jardin, pour l’agriculture, il faut un forage»). En capturant les sédiments, les retenues ont une durée de vie limitée et ne doivent pas être considérée comme une solution pérenne. Après 30 ans, de nouvelles stratégies sont nécessaires afin de soutenir les acquis des premiers programmes de conservations des eaux et des sols.
305
Part VII
Bibliography
306
Bibliography Alazard, M., C. Leduc, R. Virrion, S. Guidon, A. Ben Salem, and Y. Travi (2011), Estimating groundwater fluxes by hydrodynamic and geochemical approaches in a heterogeneous Mediterranean system (central Tunisia), IAHS Publication, 345, 253–258. Alazard, M., C. Leduc, Y. Travi, G. Boulet, and A. Ben Salem (2015), Estimating evaporation in semi-arid areas facing data scarcity: Example of the El Haouareb dam (Merguellil catchment, Central Tunisia), Journal of Hydrology: Regional Studies, 3, 265–284, doi: 10.1016/j.ejrh.2014.11.007. Albergel, J., and N. Rejeb (1997), Les lacs collinaires en Tunisie : Enjeux, Contraintes et Perspectives, Comptes rendus de l’Académie d’agriculture de France, 83 (2), 77–88. Albergel, J., Y. Pepin, S. Nasri, and M. Boufaroua (2003), Erosion et transport solide dans des petits bassins versants méditerranéens, IAHS Publication, 278, 373–379. Albergel, J., S. Nasri, M. Boufaroua, A. Droubi, and A. A. Merzouk (2004), Petits barrages et lacs collinaires, aménagements originaux de conservation des eaux et de protection des infrastructures aval: exemples des petits barrages en Afrique du Nord et au Proche Orient, Sécheresse, 15 (1), 78–86. Allen, R., L. Pereira, D. Raes, and M. Smith (1998), Crop evapotranspiration-Guidelines for computing crop water requirements-FAO Irrigation and drainage paper 56, FAO, Rome. Alsdorf, D., P. Bates, and J. Melack (2007), Spatial and temporal complexity of the Amazon flood measured from space, Geophysical Research Letters, 34 (8), 6, doi:10.1029/ 2007GL029447. Annor, F. O., N. van de Giesen, J. Liebe, P. van de Zaag, A. Tilmant, and S. Odai (2009), Delineation of small reservoirs using radar imagery in a semi-arid environment: A case study in the upper east region of Ghana, Physics and Chemistry of the Earth, Parts A/B/C, 34 (4-5), 309–315, doi:10.1016/j.pce.2008.08.005. Avalos, J. E. (2004), Modélisation hydrologique globale conceptuelle appliquée aux petits bassins versants en zone semi-aride du nord-Mexique, Revue des sciences de l’eau, 17 (2), 195–212, doi:10.7202/705530ar. Baba Sy, M. O., and M. Besbes (2001), Dam-aquifer multisystem modelling for the Wadi Merguellil basin, Tunisia, IAHS Publication, 269 (269), 135–138.
307
Baccari, N., M. Boussema, J. Lamachere, and S. Nasri (2008), Efficiency of contour benches, filling-in and silting-up of a hillside reservoir in a semi-arid climate in Tunisia, Comptes Rendus Geosciences, 340 (1), 38–48, doi:10.1016/j.crte.2007.09.020. Baccour, H., M. Slimani, and C. Cudennec (2012), Structures spatiales de l’évapotranspiration de référence et des variables climatiques corrélées en Tunisie, Hydrological Sciences Journal, 57 (4), 818–829, doi:10.1080/02626667.2012.672986. Bakker, K. (2012), Water security: research challenges and opportunities, Science, 337, 914–915, doi:10.1126/science.1226337. Bastiaanssen, W. G. M., D. J. Molden, and I. W. Makin (2000), Remote sensing for irrigated agriculture: examples from research and possible applications, Agricultural Water Management, 46 (2), 137–155, doi:10.1016/S0378-3774(00)00080-9. Baup, F., F. Frappart, and J. Maubant (2014), Combining high-resolution satellite images and altimetry to estimate the volume of small lakes, Hydrology and Earth System Sciences, 18 (5), 2007–2020, doi:10.5194/hess-18-2007-2014. Beaud, S. (1996), L’usage de l’entretien en sciences sociales. Plaidoyer pour l’"entretien ethnographique", Politix, 9 (35), 226–257, doi:10.3406/polix.1996.1966. Bédard, F., G. Reichert, R. Dobbins, and I. Trépanier (2008), Evaluation of segment-based gap-filled Landsat ETM+ SLC-off satellite data for land cover classification in southern Saskatchewan, Canada, International Journal of Remote Sensing, 29 (7), 2041–2054, doi: 10.1080/01431160701281064. Ben Ammar, S., K. Zouari, C. Leduc, and J. M’Barek (2006), Caractérisation isotopique de la relation barrage - nappe dans le bassin du Merguellil (Plaine de Kairouan, Tunisie centrale), Hydrological Sciences Journal, 51 (2), 272–284, doi:10.1623/hysj.51.2.272. Ben Mammou, A., and M. Louati (2007), Évolution temporelle de l’envasement des retenues de barrages de Tunisie, Revue des sciences de l’eau, 20 (2), 201–210. Bergé-Nguyen, M., and J.-F. Crétaux (2015), Inundations in the Inner Niger Delta: Monitoring and Analysis Using MODIS and Global Precipitation Datasets, Remote Sensing, 7 (2), 2127–2151, doi:10.3390/rs70202127. Beven, K., and J. Freer (2001), Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems using the GLUE methodology, Journal of Hydrology, 249 (1-4), 11–29, doi:10.1016/S0022-1694(01)00421-8. Boulet, G., Y. Kerr, A. Chehbouni, and J. D. Kalma (2002), Deriving catchment-scale water and energy balance parameters using data assimilation based on extended Kalman filtering, Hydrological Sciences Journal, 47 (2015), 449–467, doi:10.1080/02626660209492946. Bouma, J. A., T. W. Biggs, and L. M. Bouwer (2011), The downstream externalities of harvesting rainwater in semi-arid watersheds: An Indian case study, Agricultural Water Management, 98 (7), 1162–1170, doi:10.1016/j.agwat.2011.02.010. Braden, J. B., et al. (2009), Social science in a water observing system, Water Resources Research, 45 (W11301), doi:10.1029/2009WR008216. 308
Bruins, H., M. Evenari, and U. Nessler (1986), Rainwater-harvesting agriculture for food production in arid zones: the challenge of the African famine, Applied Geography, 6, 13–32, doi:10.1016/0143-6228(86)90026-3. Brunet-Moret, Y. (1971), Etude de l’homogénéité de séries chronologiques de précipitations annuelles par la méthode des doubles masses, Cahiers Orstom, Série Hydrologie, VIII (4), 3–31. Burte, J., A. Coudrain, H. Frischkorn, I. Chaffaut, and P. Kosuth (2005), Impacts anthropiques sur les termes du bilan hydrologique d’un aquifère alluvial dans le Nordeste semi-aride, Brésil, Hydrological Sciences Journal, 50 (1), 95–110, doi:10.1623/hysj.50.1.95. 56337. Cadier, E. (1996), Hydrologie des petits bassins du Nordeste Brésilien semi-aride: typologie des bassins et transposition écoulements annuels Small watershed hydrology in semi-arid north-eastern Brazil: basin typology and transposition of annual runoff data, Journal of Hydrology, 182 (1-4), 117–141, doi:10.1016/0022-1694(95)02933-8. Chander, G., B. L. Markham, and J. a. Barsi (2007), Revised landsat-5 thematic mapper radiometric calibration, IEEE Geoscience and Remote Sensing Letters, 4 (3), 490–494, doi:10.1109/LGRS.2007.898285. Chander, G., B. L. Markham, and D. L. Helder (2009), Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors, Remote Sensing of Environment, 113 (5), 893–903, doi:10.1016/j.rse.2009.01.007. Chavez, P. S. J. (1996), Image-based atmospheric corrections- revisited and improved, Photogrammetric Engineering and Remote Sensing, 62 (9), 1025–1035, doi:0099-1112/96/ 6209-1025. Chen, J., X. Zhu, J. E. Vogelmann, F. Gao, and S. Jin (2011), A simple and effective method for filling gaps in Landsat ETM+ SLC-off images, Remote Sensing of Environment, 115 (4), 1053–1064, doi:10.1016/j.rse.2010.12.010. Clark, M. P., D. E. Rupp, R. A. Woods, X. Zheng, R. P. Ibbitt, A. G. Slater, J. Schmidt, and M. J. Uddstrom (2008), Hydrological data assimilation with the ensemble Kalman filter: Use of streamflow observations to update states in a distributed hydrological model, Advances in Water Resources, 31 (10), 1309–1324, doi:10.1016/j.advwatres.2008.06.005. CNEA (2006), Etude d’impact des travaux de conservation des eaux et du sol dans le gouvernorat de Kairouan, Tech. rep., Centre National des Etudes Agricoles (CNEA), Tunisie. Collinet, J., and P. Zante (2005), Analyse du ravinement de bassins versants à retenues collinaires sur sols à fortes dynamiques structurales (Tunisie), Géomorphologie: relief, processus, environnement, 1, 61–74, doi:10.4000/geomorphologie.257. Cudennec, C., M. Sarraza, and S. Nasri (2004), Modelisation robuste de l’impact agrégé de retenues collinaires sur l’hydrologie de surface, Revue des sciences de l’eau, 17 (2), 181–194, doi:10.7202/705529ar. 309
Cudennec, C., M. Slimani, and P. Le Goulven (2005), Accounting for sparsely observed rainfall space-time variability in a rainfall-runoff model of a semiarid Tunisian basin, Hydrological Sciences Journal, 50 (4), 617–630, doi:10.1623/hysj.2005.50.4.617. Cudennec, C., C. Leduc, and D. Koutsoyiannis (2007), Dryland hydrology in Mediterranean regions - a review, Hydrological Sciences Journal, 52 (6), 1077–1087, doi:10.1623/hysj.52. 6.1077. Desconnets, J., J. Taupin, T. Lebel, and C. Leduc (1997), Hydrology of the HAPEX-Sahel Central Super-Site: surface water drainage and aquifer recharge through the pool systems, Journal of Hydrology, 188-189, 155–178, doi:10.1016/S0022-1694(96)03158-7. Di Baldassarre, G., A. Viglione, G. Carr, L. Kuil, J. L. Salinas, and G. Blöschl (2013), Socio-hydrology: conceptualising human-flood interactions, Hydrology and Earth System Sciences, 17 (8), 3295–3303, doi:10.5194/hess-17-3295-2013. Doumounia, A., M. Gosset, F. Cazenave, M. Kacou, and F. Zougmore (2014), Rainfall monitoring based on microwave links from cellular telecommunication networks: First results from a West African test bed, Geophysical Research Letters, 41 (16), 6016–6022, doi:10.1002/2014GL060724. Dridi, B. (2000), Impact des aménagements sur la disponibilité des eaux de surface dans le bassin versant du Merguellil, Ph.D. thesis, Université Louis Pasteur (Strasbourg 1). Dridi, B., J. Bourges, J. Collinet, A. V. Auzet, and P. Garreta (2001), Impact des aménagements sur la ressource en eaux dans le bassin du Merguellil (Tunisie centrale), in Hydrologie des régions méditerranéennes, vol. 59, edited by E. Servat and J. Albergel, pp. 192–203, UNESCO, IRD, Montpellier. Evensen, G. (2003), The Ensemble Kalman Filter: Theoretical formulation and practical implementation, Ocean Dynamics, 53 (4), 343–367, doi:10.1007/s10236-003-0036-9. Faulkner, J. W., T. Steenhuis, N. van de Giesen, M. Andreini, and J. R. Liebe (2008), Water use and productivity of two small reservoir irrigation schemes in Ghana’s upper east region, Irrigation and Drainage, 57 (2), 151–163, doi:10.1002/ird.384. Faysse, N. (2001), L’influence des règles collectives d’allocation de l’eau sur les choix stratégiques des agriculteurs : des petits périmètres irrigués tunisiens aux prélèvements en rivière dans le bassin de l’Adour, Ph.D. thesis, Université de Paris X Nanterre. Feki, H., M. Slimani, and C. Cudennec (2012), Incorporating elevation in rainfall interpolation in Tunisia using geostatistical methods, Hydrological Sciences Journal, 57 (7), 1294–1314, doi:10.1080/02626667.2012.710334. Feng, M., J. O. Sexton, S. Channan, and J. R. Townshend (2015), A global, highresolution (30-m) inland water body dataset for 2000: first results of a topographicspectral classification algorithm, International Journal of Digital Earth, pp. 1–21, doi: 10.1080/17538947.2015.1026420. Feuillette, S., F. Bousquet, and P. Le Goulven (2003), SINUSE: a multi-agent model to negotiate water demand management on a free access water table, Environmental Modelling and Software, 18 (5), 413–427, doi:10.1016/S1364-8152(03)00006-9. 310
Feyisa, G. L., H. Meilby, R. Fensholt, and S. R. Proud (2014), Automated Water Extraction Index: A new technique for surface water mapping using Landsat imagery, Remote Sensing of Environment, 140, 23–35, doi:10.1016/j.rse.2013.08.029. Fisher, A., and T. Danaher (2013), A water index for SPOT5 HRG satellite imagery, New South Wales, Australia, determined by linear discriminant analysis, Remote Sensing, 5 (11), 5907–5925, doi:10.3390/rs5115907. Fontes, J.-C. (1976), Isotopes du milieu et cycles des eaux naturelles: quelques aspects, Ph.D. thesis, Université Paris VI, France. Forkel, M., N. Carvalhais, J. Verbesselt, M. Mahecha, C. Neigh, and M. Reichstein (2013), Trend Change Detection in NDVI Time Series: Effects of Inter-Annual Variability and Methodology, Remote Sensing, 5 (5), 2113–2144, doi:10.3390/rs5052113. Gao, B.-C. (1996), NDWI - A normalized difference water index for remote sensing of vegetation liquid water from space, Remote Sensing of Environment, 58 (3), 257–266, doi: 10.1016/S0034-4257(96)00067-3. Gao, P., X.-M. Mu, F. Wang, and R. Li (2011), Changes in streamflow and sediment discharge and the response to human activities in the middle reaches of the Yellow River, Hydrology and Earth System Sciences, 15 (1), 1–10, doi:10.5194/hess-15-1-2011. Gao, Y., and W. Zhang (2009), LULC classification and topographic correction of Landsat7 ETM+ Imagery in the Yangjia river Watershed: The influence of DEM resolution, Sensors, 9 (3), 1980–1995, doi:10.3390/s90301980. García-Ruiz, J. M., J. I. López-Moreno, S. M. Vicente-Serrano, T. Lasanta-Martínez, and S. Beguería (2011), Mediterranean water resources in a global change scenario, EarthScience Reviews, 105 (3-4), 121–139, doi:10.1016/j.earscirev.2011.01.006. Gardelle, J., P. Hiernaux, L. Kergoat, and M. Grippa (2009), Less rain, more water in ponds: a remote sensing study of the dynamics of surface waters from 1950 to present in pastoral Sahel (Gourma region, Mali), Hydrology and Earth System Sciences, 14 (2), 309–324, doi:10.5194/hess-14-309-2010. Gay, D. (1999), Géochimie isotopique des relations hydrologiques entre lac de retenue et aquifère. Petit barrage collinaire de Kamech, Tunisie, Master’s thesis, Université Paris XI, France. Gay, D. (2004), Fonctionnement et bilan de retenues artificielles en Tunisie: approche hydrochimique et isotopique, Ph.D. thesis, Université Paris XI, France. Gillijns, S., O. Mendoza, J. Chandrasekar, B. De Moor, D. Bernstein, and A. Ridley (2006), What is the ensemble Kalman filter and how well does it work?, in American Control Conference 2006, p. 6, IEEE, doi:10.1109/ACC.2006.1657419. Goodwin, N. R., L. J. Collett, R. J. Denham, N. Flood, and D. Tindall (2013), Cloud and cloud shadow screening across Queensland, Australia: An automated method for Landsat TM/ETM+ time series, Remote Sensing of Environment, 134, 50–65, doi:10.1016/j.rse. 2013.02.019. 311
Goovaerts, P. (2000), Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall, Journal of Hydrology, 228 (1-2), 113–129, doi:10.1016/ S0022-1694(00)00144-X. Gorokhovich, Y., and a. Voustianiouk (2006), Accuracy assessment of the processed SRTMbased elevation data by CGIAR using field data from USA and Thailand and its relation to the terrain characteristics, Remote Sensing of Environment, 104 (4), 409–415, doi: 10.1016/j.rse.2006.05.012. Goslee, S. C. (2011), Analyzing Remote Sensing Data in R : The landsat Package, Journal of Statistical Software, 43 (4), 1–25. Gourdin, F., P. Cecchi, D. Corbin, and A. Casenave (2003), Caractérisation hydrologique des petits barrages, in L’eau en partage: les petits barrages de Côte d’Ivoire, edited by P. Cecchi, pp. 75–95, IRD Editions, Paris. Grunberger, O., J. Montoroi, and S. Nasri (2004), Quantification of water exchange between a hill reservoir and groundwater using hydrological and isotopic modelling (El Gouazine, Tunisia), Comptes Rendus Geosciences, 336 (16), 1453–1462, doi:10.1016/j.crte.2004.08. 006. Gumbricht, T., P. Wolski, P. Frost, and T. McCarthy (2004), Forecasting the spatial extent of the annual flood in the Okavango Delta, Botswana, Journal of Hydrology, 290 (3-4), 178–191, doi:10.1016/j.jhydrol.2003.11.010. Habi, M., and B. Morsli (2011), Contraintes et perspectives des retenues collinaires dans le Nord-Ouest algérien, Sécheresse, 22 (1), 49–56, doi:10.1684/sec.2011.0293. Hagolle, O., M. Huc, D. Pascual, and G. Dedieu (2015), A Multi-Temporal and MultiSpectral Method to Estimate Aerosol Optical Thickness over Land, for the Atmospheric Correction of FormoSat-2, LandSat, VENS and Sentinel-2 Images, Remote Sensing, 7 (3), 2668–2691, doi:10.3390/rs70302668. Hajji, O., S. Abidi, T. Hermassi, H. Habaieb, and M. M. Raouf (2015), Mapping multifactor vulnerability to the siltation of small lakes of central Tunisia, International Journal of Innovation and Applied Studies, 10 (4), 1251–1266. Hale, R. L., et al. (2015), iSAW : Integrating Structure , Actors , and Water to study socio-hydro-ecological systems, Earth’s Future, 3, 110–132, doi:10.1002/2014EF000295. Hantson, S., and E. Chuvieco (2011), Evaluation of different topographic correction methods for Landsat imagery, International Journal of Applied Earth Observation and Geoinformation, 13 (5), 691–700, doi:10.1016/j.jag.2011.05.001. Hardin, G. (1968), The tragedy of the commons., Science, 162 (3859), 1243–8, doi:10.1126/ science.162.3859.1243. He, X., Z. LI, M. Hao, K. Tang, and F. Zheng (2003), Down-scale analysis for water scarcity in response to soil -water conservation on Loess Plateau of China, Agriculture, Ecosystems and Environment, 94 (3), 355–361, doi:10.1016/S0167-8809(02)00039-7.
312
Hengl, T., G. B. Heuvelink, and D. G. Rossiter (2007), About regression-kriging: From equations to case studies, Computers and Geosciences, 33 (10), 1301–1315, doi:10.1016/j. cageo.2007.05.001. Hentati, A., A. Kawamura, H. Amaguchi, and Y. Iseri (2010), Evaluation of sedimentation vulnerability at small hillside reservoirs in the semi-arid region of Tunisia using the SelfOrganizing Map, Geomorphology, 122 (1-2), 56–64, doi:10.1016/j.geomorph.2010.05.013. Huang, C., N. Thomas, S. N. Goward, J. G. Masek, Z. Zhu, J. R. G. Townshend, and J. E. Vogelmann (2010), Automated masking of cloud and cloud shadow for forest change analysis using Landsat images, International Journal of Remote Sensing, 31 (20), 5449– 5464, doi:10.1080/01431160903369642. Irish, R. R., J. L. Barker, S. N. Goward, and T. Arvidson (2006), Characterization of the Landsat-7 ETM+ Automated Cloud-Cover Assessment (ACCA) Algorithm, Photogrammetric Engineering and Remote Sensing, 72 (10), 1179–1188, doi:10.14358/PERS.72.10. 1179. Jain, A. K. (2010), Data clustering: 50 years beyond K-means, Pattern Recognition Letters, 31 (8), 651–666, doi:10.1016/j.patrec.2009.09.011. Ji, L., L. Zhang, and B. Wylie (2009), Analysis of Dynamic Thresholds for the Normalized Difference Water Index, Photogrammetric Engineering and Remote Sensing, 75 (11), 1307– 1317, doi:10.14358/PERS.75.11.1307. Ji, L., X. Geng, K. Sun, Y. Zhao, and P. Gong (2015), Target Detection Method for Water Mapping Using Landsat 8 OLI/TIRS Imagery, Water, 7 (2), 794–817, doi: 10.3390/w7020794. Jiang, H., M. Feng, Y. Zhu, N. Lu, J. Huang, and T. Xiao (2014), An Automated Method for Extracting Rivers and Lakes from Landsat Imagery, Remote Sensing, 6 (6), 5067–5089, doi:10.3390/rs6065067. Khlifi, S., M. N. Ben Haha, A. Ben Rhouma, M. Bennour, S. Baccouche, and M. Hammami (2007), Évaluation de l’irrigation à partir de retenues collinaires: cas des lacs Kharrouba, Sned et Jinna dans la région de Jendouba (Tunisie), Sécheresse, 18 (2), 81–87, doi:10. 1684/sec.2007.0073. Khlifi, S., M. Ameur, N. Mtimet, N. Ghazouani, and N. Belhadj (2010), Impacts of small hill dams on agricultural development of hilly land in the Jendouba region of northwestern Tunisia, Agricultural Water Management, 97 (1), 50–56, doi:10.1016/j.agwat.2009.08.010. King, E. G., F. C. O’Donnell, and K. K. Caylor (2012), Reframing hydrology education to solve coupled human and environmental problems, Hydrology and Earth System Sciences, 16 (11), 4023–4031, doi:10.5194/hess-16-4023-2012. Kingumbi, A. (2006), Modélisation hydrologique d’un bassin affecté par des changements d’occupation. Cas du Merguellil en Tunisie Centrale, Ph.D. thesis, Université de Tunis El Manar, Ecole Nationale d’Ingénieurs de Tunis.
313
Kingumbi, A., Z. Bargaoui, E. Ledoux, M. Besbes, and P. Hubert (2007), Modélisation hydrologique stochastique d’un bassin affecté par des changements d’occupation: cas du Merguellil en Tunisie centrale, Hydrological Sciences Journal, 52 (6), 1232–1252, doi:10. 1623/hysj.52.6.1232. Kongo, V., and G. Jewitt (2006), Preliminary investigation of catchment hydrology in response to agricultural water use innovations: A case study of the Potshini catchment South Africa, Physics and Chemistry of the Earth, Parts A/B/C, 31 (15-16), 976–987, doi:10.1016/j.pce.2006.08.014. Lacaux, J., Y. Tourre, C. Vignolles, J. Ndione, and M. Lafaye (2007), Classification of ponds from high-spatial resolution remote sensing: Application to Rift Valley Fever epidemics in Senegal, Remote Sensing of Environment, 106 (1), 66–74, doi:10.1016/j.rse.2006.07.012. Lachaal, L., B. Thabet, and L. Mahfoudi (1997), Multi-market modelling analysis of agricultural policies in Tunisia, in Agricultural growth, sustainable resource management and poverty alleviation in the low rainfall area of West Asia and North Africa, edited by N. Chaherli, P. Hazell, T. Ngaido, T. Nordblom, and P. Oram, pp. 193–198, Amman, Jordan. Lacombe, G. (2007), Evolution et usages de la ressource en eau dans un bassin versant amenagé semi-aride. Le cas du Merguellil en Tunisie Centrale, Ph.D. thesis, Université Montpellier II. Lacombe, G., B. Cappelaere, and C. Leduc (2008), Hydrological impact of water and soil conservation works in the Merguellil catchment of central Tunisia, Journal of Hydrology, 359 (3-4), 210–224, doi:10.1016/j.jhydrol.2008.07.001. Le Goulven, P., C. Leduc, M. S. Bachta, and J.-C. Poussin (2009), Sharing scarce resources in Mediterranean river rasin: Wadi Merguellil in Central Tunisia, in River Basin Trajectories: Societies, Environments and Development, edited by F. Molle and F. Wester, chap. 7, pp. 147–170, CAB International. Leauthaud, C., G. Belaud, S. Duvail, R. Moussa, O. Grünberger, and J. Albergel (2013), Characterizing floods in the poorly gauged wetlands of the Tana River Delta, Kenya, using a water balance model and satellite data, Hydrology and Earth System Sciences, 17 (8), 3059–3075, doi:10.5194/hess-17-3059-2013. Leduc, C., et al. (2007), Impacts of hydrological changes in the Mediterranean zone: environmental modifications and rural development in the Merguellil catchment, central Tunisia, Hydrological Sciences Journal, 52 (6), 1162–1178, doi:10.1623/hysj.52.6.1162. Lehner, B., K. Verdin, and A. Jarvis (2013), HydroSHEDS Technical Documentation Version 1.2, Tech. rep., World Wildlife Fund US, Washington DC, USA. Li, Q., and J. Gowing (2005), A Daily Water Balance Modelling Approach for Simulating Performance of Tank-Based Irrigation Systems, Water Resources Management, 19 (3), 211–231, doi:10.1007/s11269-005-2702-9.
314
Liang, S., H. Fang, and M. Chen (2001), Atmospheric correction of Landsat ETM+ land surface imagery. I. Methods, IEEE Transactions on Geoscience and Remote Sensing, 39 (11), 2490–2498, doi:10.1109/36.964986. Liebe, J., N. van de Giesen, and M. Andreini (2005), Estimation of small reservoir storage capacities in a semi-arid environment, Physics and Chemistry of the Earth, Parts A/B/C, 30 (6-7), 448–454, doi:10.1016/j.pce.2005.06.011. Liebe, J. R., N. van de Giesen, M. Andreini, M. T. Walter, and T. S. Steenhuis (2009), Determining watershed response in data poor environments with remotely sensed small reservoirs as runoff gauges, Water Resources Research, 45 (7), W07,410, doi:10.1029/ 2008WR007369. Linacre, E. (1994), Estimating U.S. Class A Pan Evaporation from Few Climate Data, Water International, 19 (1), 5–14, doi:10.1080/02508069408686189. Lu, D., P. Mausel, E. Brondizio, and E. Moran (2002), Assessment of atmospheric correction methods for Landsat TM data applicable to Amazon basin LBA research, International Journal of Remote Sensing, 23 (13), 2651–2671, doi:10.1080/01431160110109642. Ma, M., X. Wang, F. Veroustraete, and L. Dong (2007), Change in area of Ebinur Lake during the 1998-2005 period, International Journal of Remote Sensing, 28 (24), 5523– 5533, doi:10.1080/01431160601009698. Mahé, G., D. Orange, A. Mariko, and J. P. Bricquet (2011), Estimation of the flooded area of the Inner Delta of the River Niger in Mali by hydrological balance and satellite data, IAHS Publication, 344, 138–143. Martin-Rosales, W., and C. Leduc (2003), Dynamiques de vidange d’une mare temporaire au Sahel : l’exemple de Banizoumbou (Sud-Ouest du Niger), Comptes Rendus Geoscience, 335 (5), 461–468, doi:10.1016/S1631-0713(03)00059-2. Martinez, J.-M., and T. Le Toan (2007), Mapping of flood dynamics and spatial distribution of vegetation in the Amazon floodplain using multitemporal SAR data, Remote Sensing of Environment, 108 (3), 209–223, doi:10.1016/j.rse.2006.11.012. Martinuzzi, S., W. Gould, and O. González (2007), Creating cloud-free Landsat ETM+ data sets in tropical landscapes: cloud and cloud-shadow removal, General Technical Report IITF-GTR-32, pp. 1–18. Massuel, S., J. Perrin, M. Wajid, C. Mascre, and B. Dewandel (2009), A simple, low-cost method to monitor duration of ground water pumping, Ground Water, 47 (1), 141–145, doi:10.1111/j.1745-6584.2008.00511.x. Massuel, S., D. Feurer, A. Ogilvie, R. Calvez, and R. Rochette (2014a), Vers l’amélioration du bilan hydrologique des retenues collinaires par la prise de vue aéroportée légère, in Drones et moyens légers aéroportés d’observation, p. 1, Montpellier, France. Massuel, S., J. Perrin, C. Mascre, W. Mohamed, A. Boisson, and S. Ahmed (2014b), Managed aquifer recharge in South India: What to expect from small percolation tanks in hard rock?, Journal of Hydrology, 512, 157–167, doi:10.1016/j.jhydrol.2014.02.062. 315
Mather, P. (1999), Computer processing of remote-sensed images: An introduction, John Wiley and Sons Inc, Chichester, New York. Matsuno, Y., M. Tasumi, W. van der Hoek, R. Sakthivadivel, and K. Otsuki (2003), Analysis of return flows in a tank cascade system in Sri Lanka, Paddy and Water Environment, 1 (4), 173–181, doi:10.1007/s10333-003-0029-9. Maxwell, S. (2004), Filling Landsat ETM+ SLC-off Gaps Using a Segmentation Model Approach, Photogrammetric Engineering and Remote Sensing, 70 (10), 1109–1111. McFeeters, S. K. (1996), The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features, International Journal of Remote Sensing, 17 (7), 1425–1432, doi:10.1080/01431169608948714. McMahon, T. a., M. C. Peel, L. Lowe, R. Srikanthan, and T. R. McVicar (2013), Estimating actual, potential, reference crop and pan evaporation using standard meteorological data: a pragmatic synthesis, Hydrology and Earth System Sciences, 17 (4), 1331–1363, doi:10. 5194/hess-17-1331-2013. Mialhe, F., Y. Gunnell, and C. Mering (2008), Synoptic assessment of water resource variability in reservoirs by remote sensing: General approach and application to the runoff harvesting systems of south India, Water Resources Research, 44 (5), 14, doi: 10.1029/2007WR006065. Minnaert, M. (1941), The reciprocity principle in lunar photometry, The Astrophysical Journal, 93, 403–410, doi:10.1086/144279. Mohamed, Y., W. G. M. Bastiaanssen, and H. H. G. Savenije (2004), Spatial variability of evaporation and moisture storage in the swamps of the upper Nile studied by remote sensing techniques, Journal of Hydrology, 289 (1-4), 145–164, doi:10.1016/j.jhydrol.2003. 11.038. Molle, F. (1991), Caractéristiques et potentialités des Açudes du nord-est brésilien / Caracteristics and potentialities of the brazilian northeast Acudes, Ph.D. thesis, Université de Montpellier 2, France. Montanari, A., et al. (2013), Panta Rhei - Everything Flows: Change in hydrology and society - The IAHS Scientific Decade 2013 - 2022, Hydrological Sciences Journal, 58 (6), 1256–1275, doi:10.1080/02626667.2013.809088. Montoroi, J.-p., O. Grunberger, and S. Nasri (2002), Groundwater geochemistry of a small reservoir catchment in Central Tunisia, Applied Geochemistry, 17 (8), 1047–1060, doi: 10.1016/S0883-2927(02)00076-8. Moradkhani, H., S. Sorooshian, H. V. Gupta, and P. R. Houser (2005), Dual state - parameter estimation of hydrological models using ensemble Kalman filter, Advances in Water Resources, 28 (2), 135–147, doi:10.1016/j.advwatres.2004.09.002. Moussa, R. (2010), When monstrosity can be beautiful while normality can be ugly: assessing the performance of event-based flood models, Hydrological Sciences Journal, 55 (6), 1074–1084, doi:10.1080/02626667.2010.505893. 316
Mu, Q., M. Zhao, and S. W. Running (2011), Improvements to a MODIS global terrestrial evapotranspiration algorithm, Remote Sensing of Environment, 115 (8), 1781–1800, doi: 10.1016/j.rse.2011.02.019. Mugabe, F., M. Hodnett, and A. Senzanje (2003), Opportunities for increasing productive water use from dam water: a case study from semi-arid Zimbabwe, Agricultural Water Management, 62 (2), 149–163, doi:10.1016/S0378-3774(03)00077-5. Mushtaq, S., D. Dawe, and M. Hafeez (2007), Economic evaluation of small multi-purpose ponds in the Zhanghe irrigation system, China, Agricultural Water Management, 91 (1-3), 61–70, doi:10.1016/j.agwat.2007.04.006. Musy, A., and C. Higy (2004), Hydrologie: Une science de la nature, 314 pp., Presses polytechniques et universitaires romandes. Nasri, S. (2007), Caracteristiques et impacts hydrologiques de banquettes en cascade sur un versant semi-aride en Tunisie centrale, Hydrological Sciences Journal, 52 (6), 1134–1145, doi:10.1623/hysj.52.6.1134. Neppel, L., M. Desbordes, and J. M. Masson (1998), Influence de l’évolution dans l’espace et le temps d’un réseau de pluviomètres sur l’observation des surfaces de pluie en fonction de leur aire, Revue des sciences de l’eau, 11 (1), 43–60, doi:10.7202/705296ar. Ngigi, S. N., H. H. G. Savenije, J. N. Thome, J. Rockström, and F. W. T. P. De Vries (2005), Agro-hydrological evaluation of on-farm rainwater storage systems for supplemental irrigation in Laikipia district, Kenya, Agricultural Water Management, 73 (1), 21–41, doi:10.1016/j.agwat.2004.09.021. Nicault, A., S. Alleaume, S. Brewer, M. Carrer, P. Nola, and J. Guiot (2008), Mediterranean drought fluctuation during the last 500 years based on tree-ring data, Climate Dynamics, 31 (2-3), 227–245, doi:10.1007/s00382-007-0349-3. Nyssen, J., et al. (2010), Impact of soil and water conservation measures on catchment hydrological response-a case in north Ethiopia, Hydrological Processes, 24, 1880–1895, doi:10.1002/hyp.7628. Ogilvie, A., G. Belaud, C. Delenne, J.-S. Bailly, J.-C. Bader, A. Oleksiak, L. Ferry, and D. Martin (2015), Decadal monitoring of the Niger Inner Delta flood dynamics using MODIS optical data, Journal of Hydrology, 523, 368–383, doi:10.1016/j.jhydrol.2015.01. 036. Ogilvie, A., P. Le Goulven, C. Leduc, R. Calvez, and M. Mulligan (2016), Réponse hydrologique d’un bassin semi-aride aux événements pluviométriques et aménagements de versant (bassin du Merguellil, Tunisie centrale), Hydrological Sciences Journal, doi: 10.1080/02626667.2014.934249. Olivier de Sardan, J.-P. (1995), La politique du terrain, Enquête, 1, 71–109. Ostrom, E. (2009), A general framework for analyzing sustainability of social-ecological systems., Science, 325, 419–422, doi:10.1126/science.1172133.
317
Oudin, L., F. Hervieu, C. Michel, C. Perrin, V. Andréassian, F. Anctil, and C. Loumagne (2005), Which potential evapotranspiration input for a lumped rainfall-runoff model? Part 2 - Towards a simple and efficient potential evapotranspiration model for rainfall-runoff modelling, Journal of Hydrology, 303 (1-4), 290–306, doi:10.1016/j.jhydrol.2004.08.026. Ouma, Y. O., and R. Tateishi (2006), A water index for rapid mapping of shoreline changes of five East African Rift Valley lakes: an empirical analysis using Landsat TM and ETM+ data, International Journal of Remote Sensing, 27 (15), 3153–3181, doi: 10.1080/01431160500309934. Overeem, A., H. Leijnse, and R. Uijlenhoet (2013), Country-wide rainfall maps from cellular communication networks., Proceedings of the National Academy of Sciences of the United States of America, 110 (8), 2741–5, doi:10.1073/pnas.1217961110. Oweis, T., and A. Hachum (2006), Water harvesting and supplemental irrigation for improved water productivity of dry farming systems in West Asia and North Africa, Agricultural Water Management, 80 (1), 57–73, doi:10.1016/j.agwat.2005.07.004. Pabiot, F. (1999), Optimisation de la gestion d’un barrage collinaire en zone semi-aride, Master’s thesis, Ecole Nationale Supérieure Agronomique de Rennes, France. Paolini, L., F. Grings, J. a. Sobrino, J. C. Jiménez Muñoz, and H. Karszenbaum (2006), Radiometric correction effects in Landsat multi-date/multi-sensor change detection studies, International Journal of Remote Sensing, 27 (4), 685–704, doi:10.1080/ 01431160500183057. Parajka, J., R. Merz, and G. Blöschl (2005), A comparison of regionalisation methods for catchment model parameters, Hydrology and Earth System Sciences Discussions, 2 (2), 509–542, doi:10.5194/hessd-2-509-2005. Perrin, C., C. Michel, and V. Andréassian (2003), Improvement of a parsimonious model for streamflow simulation, Journal of Hydrology, 279 (1-4), 275–289, doi:10.1016/ S0022-1694(03)00225-7. Poncet, J. (1970), La "catastrophe" climatique de l’automne 1969 en Tunisie, Annales de Géographie, 79 (435), 581–595, doi:10.3406/geo.1970.15175. Proaño, D. (2012), Bilan offres-demandes sur le bassin versant du Merguellil à l’aide de la plateforme WEAP, Master’s thesis, Université de Montpellier 2, France. Qi, S., D. G. Brown, Q. Tian, L. Jiang, T. Zhao, and K. M. Bergen (2009), Inundation Extent and Flood Frequency Mapping Using LANDSAT Imagery and Digital Elevation Models, GIScience and Remote Sensing, 46 (1), 101–127, doi:10.2747/1548-1603.46.1.101. Ran, L., and X. X. Lu (2012), Delineation of reservoirs using remote sensing and their storage estimate: an example of the Yellow River basin, China, Hydrological Processes, 26 (8), 1215–1229, doi:10.1002/hyp.8224. Reichle, R. H., D. B. McLaughlin, and D. Entekhabi (2002), Hydrologic Data Assimilation with the Ensemble Kalman Filter, Monthly Weather Review, 130 (1), 103–114, doi:10. 1175/1520-0493(2002)1302.0.CO;2. 318
Riano, D., E. Chuvieco, J. Salas, and I. Aguado (2003), Assessment of different topographic corrections in landsat-TM data for mapping vegetation types (2003), IEEE Transactions on Geoscience and Remote Sensing, 41 (5), 1056–1061, doi:10.1109/TGRS.2003.811693. Riaux, J. (2013), Engager la construction d’un regard sociohydrologique : des archives catalyseurs de l’interdisciplinarité, Natures Sciences Sociétés, 21 (1), 15–23, doi:10.1051/ nss/2013061. Riaux, J., and S. Massuel (2014), Construire un regard sociohydrologique (2). Le terrain en commun, générateur de convergences scientifiques, Natures Sciences Sociétés, 22 (4), 329–339, doi:10.1051/nss/2014046. Riaux, J., C. Leduc, N. Ben Aissa, J. Burte, R. Calvez, H. Habaieb, A. Ogilvie, S. Massuel, and R. Rochette (2014a), Understanding and Defining sociohydrological spaces and their boundaries: an interdisciplinary perspective from collective fieldwork, in EGU General Assembly, Vienna, Austria. Riaux, J., A. Ogilvie, and Z. Jenhaoui (2014b), Les retenues collinaires font-elles ressource? Réflexions à partir de la Tunisie Centrale, in 6e colloque de la T.M.A for H.S.E.S, Monastir, Tunisia. Riou, C. (1972), Etude de l’évaporation en Afrique centrale, Cahiers Orstom, Série Hydrologie, pp. 39–52. Riou, C., and R. Chartier (1985), Evapotranspiration en zone semi-aride de deux couverts végétaux (gazon, blé) obtenue par plusieurs méthodes. I. - Evaluation de l’ETP (conditions hydriques non limitantes), Agronomie, 5 (3), 261–266. Risi, C. (2009), Les isotopes stables de l’eau : applications à l’étude du cycle de l’eau et des variations du climat, Ph.D. thesis, Université Paris VI, France. Rodrigues, L. N., E. E. Sano, T. S. Steenhuis, and D. P. Passo (2011), Estimation of Small Reservoir Storage Capacities with Remote Sensing in the Brazilian Savannah Region, Water Resources Management, 26 (4), 873–882, doi:10.1007/s11269-011-9941-8. Rogers, a. S., and M. S. Kearney (2004), Reducing signature variability in unmixing coastal marsh Thematic Mapper scenes using spectral indices, International Journal of Remote Sensing, 25 (12), 2317–2335, doi:10.1080/01431160310001618103. Roose, E. (1991), Conservation des sols en zones méditerranéennes - Synthèse et proposition d’une nouvelle stratégie de lutte antiérosive: la CGES, Cahiers Orstom, Série Pédologie, XXVI (2), 145–181. Rouse, J., J. Haas, J. Schell, and D. Deering (1973), Monitoring vegetation systems in the Great Plains with ERTS, in Proceedings 3rd ERTS Symposium, pp. 309–317, NASA SP353, Washington DC, USA. Roy, D. P., J. Ju, P. Lewis, C. Schaaf, F. Gao, M. Hansen, and E. Lindquist (2008), Multitemporal MODIS-Landsat data fusion for relative radiometric normalization, gap filling, and prediction of Landsat data, Remote Sensing of Environment, 112 (6), 3112–3130, doi:10.1016/j.rse.2008.03.009. 319
Sakamoto, T., N. Van Nguyen, A. Kotera, H. Ohno, N. Ishitsuka, and M. Yokozawa (2007), Detecting temporal changes in the extent of annual flooding within the Cambodia and the Vietnamese Mekong Delta from MODIS time-series imagery, Remote Sensing of Environment, 109 (3), 295–313, doi:10.1016/j.rse.2007.01.011. Sawunyama, T., A. Senzanje, and A. Mhizha (2006), Estimation of small reservoir storage capacities in Limpopo River Basin using geographical information systems (GIS) and remotely sensed surface areas: Case of Mzingwane catchment, Physics and Chemistry of the Earth, Parts A/B/C, 31 (15-16), 935–943, doi:10.1016/j.pce.2006.08.008. Scaramuzza, P., E. Micijevic, and G. Chander (2004), SLC gap-filled products phase one methodology, Landsat Technical Notes. Seibert, J., and M. J. P. Vis (2012), Teaching hydrological modeling with a user-friendly catchment-runoff-model software package, Hydrology and Earth System Sciences, 16 (9), 3315–3325, doi:10.5194/hess-16-3315-2012. Seiler, R., J. Schmidt, O. Diallo, and E. Csaplovics (2009), Flood monitoring in a semi-arid environment using spatially high resolution radar and optical data, Journal of Environmental Management, 90 (7), 2121–9, doi:10.1016/j.jenvman.2007.07.035. Selmi, S., and J.-C. Talineau (1994), Des lacs collinaires pour un développement durable en Tunisie semi-aride : Systèmes irrigués, Les Cahiers de la recherche-développement, 37, 33–46. Selmi, S., and S. Zekri (1995), Evaluation économique et environnementale des lacs collinaires en Tunisie. Le cas d’El Gouazine (Ousslatia - Kairouan), in L’homme peut il refaire ce qu’il a défait, edited by R. Pontanier, chap. 27, pp. 439–448, John Libbey Eurotext. Selmi, S., M. B. Sai, and M. Hammami (2001), La valorisation des ressources en eau aléatoires et non pérennes par le développement de l’olivier autour des lacs collinaires en Tunisie, Sécheresse, 12 (1), 45–50. Sen, A. (1999), Commodities and capabilities, OUP Catalogue. Sivakumar, B. (2012), Socio-hydrology: not a new science, but a recycled and re-worded hydrosociology, Hydrological Processes, 26 (24), 3788–3790, doi:10.1002/hyp.9511. Sivapalan, M., H. H. G. Savenije, and G. Blöschl (2012), Socio-hydrology: A new science of people and water, Hydrological Processes, 26 (8), 1270–1276, doi:10.1002/hyp.8426. Song, C., C. E. Woodcock, K. C. Seto, M. P. Lenney, and S. A. Macomber (2001), Classification and Change Detection Using Landsat TM Data, Remote Sensing of Environment, 75 (2), 230–244, doi:10.1016/S0034-4257(00)00169-3. Soti, V., C. Puech, D. Lo Seen, A. Bertran, C. Vignolles, B. Mondet, N. Dessay, and A. Tran (2010), The potential for remote sensing and hydrologic modelling to assess the spatiotemporal dynamics of ponds in the Ferlo Region (Senegal), Hydrology and Earth System Sciences, 14 (8), 1449–1464, doi:10.5194/hess-14-1449-2010.
320
Storey, J., M. Choate, and K. Lee (2014), Landsat 8 Operational Land Imager OnOrbit Geometric Calibration and Performance, Remote Sensing, (3), 11,127–11,152, doi: 10.3390/rs61111127. Swenson, S., and J. Wahr (2009), Monitoring the water balance of Lake Victoria, East Africa, from space, Journal of Hydrology, 370 (1), 163–176, doi:10.1016/j.jhydrol.2009.03.008. Talineau, J., S. Selmi, and K. Alaya (1994), Lacs collinaires en Tunisie semi-aride, Sécheresse, 5 (4), 251–256. Thompson, S. E., M. Sivapalan, C. J. Harman, V. Srinivasan, M. R. Hipsey, P. Reed, A. Montanari, and G. Blöschl (2013), Developing predictive insight into changing water systems: Use-inspired hydrologic science for the anthropocene, Hydrology and Earth System Sciences, 17 (12), 5013–5039, doi:10.5194/hess-17-5013-2013. Toya, J., A. Pietroniro, L. W. Martz, and T. D. Prowse (2002), A multi-sensor approach to wetland flood monitoring, Hydrological Processes, 16 (8), 1569–1581, doi:10.1002/hyp. 1021. Van Der Heijden, S., and U. Haberlandt (2010), Influence of spatial interpolation methods for climate variables on the simulation of discharge and nitrate fate with SWAT, Advances in Geosciences, 27, 91–98, doi:10.5194/adgeo-27-91-2010. Vanonckelen, S., S. Lhermitte, and A. Van Rompaey (2013), The effect of atmospheric and topographic correction methods on land cover classification accuracy, International Journal of Applied Earth Observation and Geoinformation, 24 (1), 9–21, doi:10.1016/j.jag. 2013.02.003. Venot, J.-P., and P. Cecchi (2011), Valeurs d’usage ou performances techniques : comment apprécier le rôle des petits barrages en Afrique subsaharienne ?, Cahiers Agricultures, 20 (1), 112–117, doi:10.1684/agr.2010.0457. Venot, J.-P., and M. Hirvonen (2013), Enduring Controversy: Small Reservoirs in SubSaharan Africa, Society and Natural Resources, 26 (8), 883–897, doi:10.1080/08941920. 2012.723306. Venot, J. P., and J. Krishnan (2011), Discursive framing: Debates over small reservoirs in the Rural South, Water Alternatives, 4 (3), 316–324. Vermeulen, P. T. M., and a. W. Heemink (2006), Model-Reduced Variational Data Assimilation, Monthly Weather Review, 134 (10), 2888–2899, doi:10.1175/MWR3209.1. Verpoorter, C., T. Kutser, D. a. Seekell, and L. J. Tranvik (2014), A Global Inventory of Lakes Based on High-Resolution Satellite Imagery, Geophysical Research Letters, 41, 6396–6402, doi:10.1002/2014GL060641. Vincent, L. F. (2003), Towards a smallholder hydrology for equitable and sustainable water management, Natural Resources Forum, 27 (2), 108–116, doi:10.1111/1477-8947.00046. Vogelmann, J. E., D. Helder, R. Morfitt, M. J. Choate, J. W. Merchant, and H. Bulley (2001), Effects of Landsat 5 Thematic Mapper and Landsat 7 Enhanced Thematic Mapper 321
plus radiometric and geometric calibrations and corrections on landscape characterization, Remote Sensing of Environment, 78 (1-2), 55–70, doi:10.1016/S0034-4257(01)00249-8. Wackernagel, H. (2004), Géostatistique et assimilation séquentielle de données, H.d.r., Université Pierre et Marie Curie, France. Winsemius, H. C. (2009), Satellite data as complementary information for hydrological modelling, Ph.D. thesis, Technische Universiteit Delft, NL. Wisser, D., S. Frolking, E. M. Douglas, B. M. Fekete, A. H. Schumann, and C. J. Vörösmarty (2010), The significance of local water resources captured in small reservoirs for crop production - A global-scale analysis, Journal of Hydrology, 384 (3-4), 264–275, doi:10. 1016/j.jhydrol.2009.07.032. Wolski, P., and M. Murray-Hudson (2008), An investigation of permanent and transient changes in flood distribution and outflows in the Okavango Delta, Botswana, Physics and Chemistry of the Earth, Parts A/B/C, 33 (1-2), 157–164, doi:10.1016/j.pce.2007.04.008. Xie, X., and D. Zhang (2010), Data assimilation for distributed hydrological catchment modeling via ensemble Kalman filter, Advances in Water Resources, 33 (6), 678–690, doi: 10.1016/j.advwatres.2010.03.012. Xu, H. (2006), Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery, International Journal of Remote Sensing, 27 (14), 3025–3033, doi:10.1080/01431160600589179. Zairi, M., I. Karray, and H. Ben Dhia (2005), Evaluation participative de l’impact des travaux de conservation des eaux et des sols (CES) dans le région de Sidi M’hadheb (Sud-Est tunisien), Sécheresse, 16 (1), 53–60. Zammouri, M., and H. Feki (2005), Managing releases from small upland reservoirs for downstream recharge in semi-arid basins (Northeast of Tunisia), Journal of Hydrology, 314 (1), 125–138, doi:10.1016/j.jhydrol.2005.03.011. Zeng, C., H. Shen, and L. Zhang (2013), Recovering missing pixels for Landsat ETM+ SLCoff imagery using multi-temporal regression analysis and a regularization method, Remote Sensing of Environment, 131, 182–194, doi:10.1016/j.rse.2012.12.012. Zhang, C., W. Li, and D. Travis (2007), Gaps-fill of SLC-off Landsat ETM+ satellite image using a geostatistical approach, International Journal of Remote Sensing, 28 (22), 5103– 5122, doi:10.1080/01431160701250416. Zhu, X., D. Liu, and J. Chen (2012), A new geostatistical approach for filling gaps in Landsat ETM+ SLC-off images, Remote Sensing of Environment, 124, 49–60, doi:10.1016/j.rse. 2012.04.019. Zhu, Z., and C. E. Woodcock (2012), Object-based cloud and cloud shadow detection in Landsat imagery, Remote Sensing of Environment, 118, 83–94, doi:10.1016/j.rse.2011.10. 028.
322
Zhu, Z., and C. E. Woodcock (2014), Automated cloud, cloud shadow, and snow detection in multitemporal Landsat data: An algorithm designed specifically for monitoring land cover change, Remote Sensing of Environment, 152, 217–234, doi:10.1016/j.rse.2014.06.012. Zribi, M., A. Chahbi, M. Shabou, Z. Lili-Chabaane, B. Duchemin, N. Baghdadi, R. Amri, and A. Chehbouni (2011), Soil surface moisture estimation over a semi-arid region using ENVISAT ASAR radar data for soil evaporation evaluation, Hydrology and Earth System Sciences, 15 (1), 345–358, doi:10.5194/hess-15-345-2011.
323
324
Résumé Les retenues collinaires connaissent un essor dans les zones semi-arides pour leur capacité à réduire le transport des sédiments, et à capter les ressources pluviométriques aléatoires pour la petite agriculture. L’ampleur et l’éparpillement de ces multiples hydro-sociosystèmes limitent leur étude, entrainant de fortes incertitudes sur leur potentiel hydrique et leur influence cumulée sur les écoulements, ainsi que sur les stratégies permettant de soutenir les agriculteurs. Cette recherche vise à développer une approche multi-échelle, interdisciplinaire pour quantifier les disponibilités en eau de multiples retenues et éclairer les facteurs hydrologiques et sociaux influençant les pratiques agricoles. Un Filtre de Kalman Ensemble couplant des observations 30 m Landsat de surfaces inondées avec un modèle hydrologique journalier (GR4J + bilan hydrique) est développé sur 7 retenues jaugées. L’assimilation de données réduit l’erreur quadratique moyenne (RMSE) sur les volumes de 50% par rapport à l’ébauche du modèle pluie-débit, diminuant notamment les incertitudes générées par les précipitations intenses et localisées peu ou mal détectées. Compensant la faible résolution temporelle et des valeurs aberrantes issues des images Landsat, la méthode permet le suivi de dynamiques de crue de retenues de l’ordre de 5 ha (R =0.9). Validée sur des données de terrain de longue durée (1999-2014), la méthode démontre notamment le potentiel des images Landsat à quantifier la disponibilité en eau annuelle de retenues non jaugées dès 1 ha (RMSE autour de 25%). Appliquée à 48 retenues et 546 images Landsat 5, 7 et 8, la chaîne de traitement confirme les fortes incertitudes et pénuries en eau, qui contraignent le développement agricole sur 80% des lacs du bassin amont du Merguellil en Tunisie Centrale. La combinaison d’inventaires, d’enquêtes agricoles et d’entretiens semi-directifs atteste des prélèvements minimes mais relève la diversification des pratiques agricoles et les bénéfices périphériques accompagnant ces aménagements. De nombreux agriculteurs ne disposent pas des capabilités nécessaires pour augmenter leurs prélèvements et souffrent de problèmes physiques et économiques d’accès à l’eau, exacerbés par une gestion inefficace et un appui limité et de courte durée. Les réussites individuelles recensées témoignent de la résilience économique de certains agriculteurs, disposant de ressources complémentaires pour faire face aux pénuries. Au vu des capacités limitées et des sécheresses durables, les retenues collinaires dans ce contexte climatique doivent maintenir leur objectif initial d’irrigation de complément et non chercher à soutenir une intensification à plus grande échelle de l’agriculture irriguée. Mots-clés: conservation eaux et sols, modélisation hydrologique, socio-hydrologie, bilan hydrique, usages agricoles, télédétection Abstract Small reservoirs have become increasingly widespread across semi-arid regions, due to their ability to reduce transport of eroded soil and harvest scarce and unreliable rainfall for local users. The scale and geographical dispersion of these multiple hydro-social systems restrict their investigation, leading to difficulties in assessing their agricultural potential, their cumulative influence on runoff, and in identifying strategies to support riparian farmers. This research sought to develop a multi-scalar interdisciplinary approach to assess water availability across multiple small reservoirs and understand hydrological and wider drivers’ influence on associated agricultural practices. An Ensemble Kalman Filter approach was developed to combine 30 m Landsat flooded surface area observations with a daily hydrological (GR4J + water balance) model on 7 gauged reservoirs. Data assimilation, providing nearreal time corrections, reduced runoff uncertainties generated by highly variable and localised rainfall intensities and lowered daily volume root mean square errors (RMSE) by 50% compared to the initial rainfall-runoff model forecast. Compensating for Landsat’s reduced temporal resolution and correcting outliers, the method correctly reproduced flood dynamics of 5 ha lakes (R =0.9). Validated against extensive field data over 1999-2014, the method notably establishes Landsat imagery’s ability to assess annual water availability on ungauged reservoirs as small as 1 ha (RMSE circa 25%). Applied to 48 small reservoirs and 546 Landsat 5-8 images, the treatment chain identified the significant water scarcity and unreliability that impedes agricultural development on 80% of lakes in the Merguellil upper catchment (Central Tunisia). In parallel, rapid surveys, quantitative questionnaires and semi-directed interviews confirmed minimal withdrawals, yet highlighted the diversification of practices and the peripheral benefits accompanying small reservoir development. Many farmers lack the capabilities to increase their withdrawals and suffer physical and economic water access difficulties, mismanagement, compounded through limited and short-term government assistance. Individual successes resulted from farmers’ economic resilience and means to secure alternate water supplies during dry spells. Faced with limited storage capacities and prolonged droughts, small reservoirs must in this climatic context retain their supplementary irrigation focus and not strive to support widespread intensification of irrigated practices. Keywords: water and soil conservation, hydrological modelling, socio-hydrology, water balance, water uses, remote sensing
