streamflow simulation in spatially heterogeneous watersheds. Olkeba T. Letaa, Aly El-Kadia,b, Henrietta Dulaib, Kariem Ghazalc a Water Resources Research ...
Assessing the effect of rainfall data scarcity on daily streamflow simulation in spatially heterogeneous watersheds
Olkeba T. Letaa, Aly El-Kadia,b, Henrietta Dulaib, Kariem Ghazalc Water Resources Research Center (WRRC), UH Manoa of Geology and Geophysics (GG), UH Manoa c Dept. of Natural Resources and Environmental Management (NREM), UH Manoa a
b Dept.
Problem/research questions •High rainfall spatial variability in Pacific Islands •Mean annual rainfall ranges 200 to 10,000 mm over the Hawaiian Islands (Giambelluca et al, 2013) •Decrease in number of rain gauging stations •Can we use rainfall data from neighboring watersheds for ungauged watersheds in spatially heterogeneous rainfall patterns? •How do rainfall data scarcity affect watershed model performance compared to relatively well gauged watersheds?
Study area (Oahu island)
Heeia watershed
Nuuanu area watersheds •Small scale watersheds with unique hydrological features, soil types, topography, & highly variable climate conditions.
Study area (American Samoa)
Maloata watershed
Heeia watershed Fagaalu watershed
Nuuanu area watersheds •Small scale watersheds with unique hydrological features, soil types, topography, & highly variable climate conditions.
Objectives
•To develop watershed model for: Heeia watershed: environmental twist Nuuanu watershed: water conservation twist Fagaalu watershed: environmental twist • To assess the effect of rainfall input on watershed model performance in spatially heterogeneous watersheds
Approach Utilize the model Soil and Water Assessment Tool (SWAT)
Soil-water-plant interaction processes and water balance
SWAT modeling approach Meteorological data
DEM, land use, soil maps
Streamflow data
Sub-basins, reach, reservoir, hydrological response units SWAT set-up and run
Parameters sensitivity analysis Sensitive parameters
Calibration and uncertainty Performance evaluation
Validation and uncertainty
Hydro-meteorological (Oahu watersheds)
• Missing values: filled based on contour maps, nearby stations, correlation and interpolation techniques for ungauged sites
Hydro-meteorological (Fagaalu)
• Missing values: filled based on contour maps, nearby stations, correlation and interpolation techniques for ungauged sites
Data and SWAT development Heeia watershed
Data and SWAT development for Nuuanu area
Data and SWAT development for Fagaalu
Data and SWAT development for Maloata
Sensitive parameters for Oahu p-value (α = 0.05)
t-stat
SLSUBBSN CANMX RCHRG_DP LAT_TTIME GW_DELAY GWQMN CH_N2 ESCO
Nuuanu area watersheds
SLSOIL SURLAG GW_REVAP SOL_Z OV_N
Top three significant:
REVAPMN SOL_K HRU_SLP EPCO SOL_AWC ALPHA_BF CH_K2
•Curve number (surface runoff) •Channel hydraulic conduct. (SW-GW interaction) •Baseflow recession constant (groundwater)
CN2 0.0
0.2
0.4
0.6
0.8
1.0
-30
-20
-10
0
10
20
Sensitive parameters for American Samoa p-value (α = 0.05)
t-stat
SOL_AWC CANMX
Fagaalu watershed
SLSUBBSN GW_DELAY CH_N2 RCHRG_DP OV_N REVAPMN SLSOIL EPCO HRU_SLP SURLAG LAT_TTIME GW_REVAP SOL_Z(..) GWQMN ESCO CH_K2 ALPHA_BF CN2
Top three significant: •Curve number (surface runoff) •Baseflow recession constant (groundwater) •Channel hydraulic conductivity (SW-GW interaction) 0.0
0.2
0.4
0.6
0.8
1.0
-10
0
10
20
30
•Top 3 sensitive parameters are the same for Oahu and Am Samoa
Calibrated SWAT parameter values
•Good solutions (NSE ≥ 0.2, Heeia, ≥0.5, Kalihi) parameter values
Observed streamflow well represented with rain records Example of without rain records
Heeia: NSE = 0.47; PBIAS = -13%, r = 0.72; RSR = 0.74
Example of with rain records
Kalihi: NSE = 0.86, PBIAS = 14.7%, r = 0.93, RSR = 0.37
Use of rainfall data within watershed improves SWAT model performance Example of rain records Kalihi
Example of no rain records Heeia
•Rain records: P-factor = 0.83 to 0.93; r-factor = 0.72 to 0.84 •Without rain records: p-factor = 0.41 to 0.76; r-factor = 0.89 to 1.18
Temporal variability of observed streamflow sufficiently captured for watershed with rainfall data With rain records
Fagaalu
NSE = 0.86; PBIAS = -6%, r = 0.94; RSR = 0.37
Without rain records
Maloata
NSE = 0.37, PBIAS = -34%, r = 0.61, RSR = 0.79
•Peak flows missed for Maloata watershed, low NSE
95% prediction uncertainty did not sufficiently bracket observations for ungauged watersheds With rain records
Fagaalu
Maloata Without rain records
•P-factor, is 0.87 for Fagaalu, but only 0.38 for Maloata watershed •Most of the observations brackets when rainfall data within watershed is used, but results reasonably acceptable
Multiple rain-gauges improve model performance
Upstream
Downstream
•Use of multiple rain gauging stations well represents hydrographs.
The Nuuanu area annual water budget
(51% P)
(P)
Streamflow (42% P) (20% P)
• ET and recharge consistent with previous studies
The Fagaalu annual water budget (35% P)
(P)
Streamflow (65% P) (14% P)
•ET inline with Izuka et al, 2007 Tafuna-Leone Plain
Conclusions •SWAT applicability tested for rain gauged and ungauged watersheds; •Use of rain records within watershed improved model performance by capturing local climate spatial variability; •SWAT performed less for watersheds without rain records, but results were reasonably acceptable; •With methods to resolve data scarcity issues and careful statistical evaluation criteria, data from neighboring areas can be used for watersheds without rain records.
Acknowledgements to: