Satellite retrieval of woody biomass for energetic ...

b i o m a s s a n d b i o e n e r g y 3 6 ( 2 0 1 2 ) 4 3 2 e4 3 8

Available online at www.sciencedirect.com

http://www.elsevier.com/locate/biombioe

Short communication

Satellite retrieval of woody biomass for energetic reuse of riparian vegetation Giovanni Forzieri* Department of Earth Sciences, University of Florence, Via E. Fermi, 2, Arcetri 50125, Firenze, Italy

article info

abstract

Article history:

Streamside vegetation plays a multifunctional role in many interconnected hydraulic,

Received 27 May 2011

ecological and sedimentological processes of great interest for flood risk assessment and

Received in revised form

river restoration practices, which need frequent riparian vegetation maintenance inter-

4 October 2011

ventions. The reuse of riparian vegetation as a potential energy source represents

Accepted 21 October 2011

a sustainable way to support the costs of buffer management. As a consequence, reliable

Available online 10 November 2011

and cost-efficient modelling of woody biomass is crucial to quantify the theoretical energy budget of riparian corridors and to drive decision-making processes. Potential capabilities

Keywords:

of multi-spectral satellite data in retrieving riparian woody biomass (B) are explored in this

Multispectral remote sensing

paper. The method is organized in five sequential steps: 1) riparian vegetation mapping; 2)

Vegetation reuse

Principal Component Analysis of the vegetation spectral signatures; 3) estimation of the

Energy budget

correlation structure between arboreal spectral signatures and ground-observed biome-

Wood chips production

chanical properties e tree high (h) and stem diameter (D); 4) identification/calibration/

Riparian vegetation management

validation of spectral-based predictive models of h and D and 5) use of a standard allometric relationship to calculate the riparian woody biomass. The methodology is tested over a 3 km reach of the forested floodplain of the Avisio river (Trentino Alto Adige, Italy) by using an extended field surveys and a synchronous SPOT-5 multi-spectral image acquired on 28/08/2004. Results showed strong correlation coefficients between spectral signatures and vegetation parameters (rk(h) ¼ 0.783 and rk(D) ¼ 0.695) and valuable satellite capabilities in retrieving biomechanical parameters through tri-parametric power lows (R2(h) ¼ 0.732 and R2(D) ¼ 0.619). The remotely-derived woody biomass map shows comparable estimates than those obtainable thought ground measurements (RMSE(B) ¼ 0.019 m3) and represents a repeatable and accurate device to assess the potential energy budget within riparian corridors. A simple simulated scenario of the riparian management is also provided to assess the biomass-derived net calorific value and the corresponding cost-benefit analysis. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Riparian vegetation plays a crucial role in many interconnected hydraulic-sedimentological-ecological processes

by: (1) providing flow resistance during flooding event and reducing erosion; (2) delivering nutrients to streams from litter fall and large woody debris; (3) stabilizing stream banks through the root mass; (4) controlling water temperature

* Tel.: þ39 055 2055304; fax: þ39 055 2055317. E-mail address: [email protected]. 0961-9534/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.biombioe.2011.10.036


through shade; and (5) adding to the recreational, habitat and aesthetic value of streams [1,2]. Flood risk reduction and river restoration practices need frequent riparian corridor maintenance interventions that can considerably burden with the financial budget of the institutions responsible for land management. The reuse of streamside vegetation as woody biomass represents a sustainable way to support the costs of buffer management (cutting/plantation operations) and to create new economic opportunities in the floodplain areas for energy production [3,4]. Approximated theoretical energy budget derivable from Poplars, Willows and Acacia can reach 160e450 GJ/ha, 178e276 GJ/ha, 178e231 GJ/ha, respectively [5]. Therefore, an accurate assessment of riparian biomass represents an essential tool to drive decision-making processes for a multifunctional riparian vegetation management. Techniques for biomass estimation focus on allometric relationships which use vegetation structural parameters, such as stem diameter and tree height [6], estimated through conventional ground surveys [7,8] or more recent in situ measurement techniques based on terrestrial laser scanning and digital parallel photography [9,10]. However, a systematic monitoring of the vegetation parameters using such field sampling is often infeasible, as these methods are time consuming, expensive and allow characterizing only small portions of riparian buffers. As in many other environmental monitoring problems, remote sensing may provide unprecedented mapping capabilities. In the last years Light Detection and Ranging (LiDAR) data are becoming a widely used tool in river applications thanks to their capacity to capture the 3D structure of monitored vegetated surfaces. Several studies applied airborne laser scanning data for the estimation of stand volume and forest characterization by showing encouraging performances on conifer areas and lower reliability estimates on broad-leaved patters with overlapped crowns and complex plant morphology [11,12]. Furthermore, LiDAR-derived parameterizations of riparian ecosystem are strongly dependent on flight acquisition opportunities and they can only provide limited information about the vegetation dynamics that would require acquisition and processing of long time series of remote sensing data [13]. Satellite images are expected to play a key role in the next years to explain riparian ecosystem dynamics, due to their relative low revisiting time and increasing spatial resolution [14]. Several research studies have been developed to estimate the aboveground biomass using satellite sensors, prevalently focusing on landscape-scale applications [15,16], whereas only few river-scale analyses have been conducted. Dillabaugh and King [17] by using Ikonos imagery demonstrated significant correlations between spectral information and biomass of shrubs in riparian marshlands. Despite the valuable results achieved in multispectral-based methodologies of riverine ecosystems focusing on vegetation spectral indices and image spectral enhancements of forested floodplains [18e20], there is a greater need to progress in satellite remote sensing to develop reliable and time-efficient tools for woody biomass retrieval in forested floodplains. In this paper the spectral-based method described in [21], originally developed for floodplain hydraulic roughness parameterization, is applied to assess the potential of satellite

433

retrieval of riparian woody biomass. The proposed technique is based on the optimization of the vegetation spectral signatures derivable from remote sensing platforms and on the exploration of spectral properties as biomass predictors.

2.

Methods

2.1.

Study area

The mouth of the Avisio river (Trentino Alto Adige, Italy) is used as study area (Fig. 1). It consists in a 3 km reach, which includes the floodplain from the clearway bridge (SS12 e Brennero), in the municipality of Lavis, to the confluence of the Avisio in the Adige River. The total area is w1,041,000 m2, its 54% is made up of vegetated lands, including herbaceous, shrub and arboreal patterns. Riparian forest is mainly made up of high-trunk plant species at the evolutionary stage with homogeneous density and is prevalently located in the confluence zone where the river presents a multi-channel path.

2.2.

Remote sensing and field data

To characterize the spectral properties of riparian vegetation a SPOT-5 image acquired on 28 August, 2004 was used in this study. The image is composed of a 10 10 m resolution shortwave band (wavelength range from 1.58 to 1.75 mm) and three 5 5 m resolution visible/near-infrared bands (wavelength ranges: blue 0.43e0.47 mm, red 0.61e0.68 mm and near infrared 0.78e0.89 mm). The short-wave band was resampled at the same spatial resolution of the finer visible/near-infrared channels. The SPOT-5 image was orthorectified in UTM/ WGS-84 projection using a 1 m orthophoto image and a rational function model and then corrected from the atmospheric effects through the ENVI module FLAASH Model [22]. The vegetation properties of the investigated riverine ecosystem were measured through an extensive field campaign in August 2004 (synchrony between field and remote sensing acquisition). Field surveys indicate three main vegetation classes: 1) mixed arboreal (willows, false acacia, black poplar; 2) shrub (elder, cornel tree, alder) and 3) herbaceous (nettle), see inset pictures in Fig. 1. Over mixed arboreal patterns a set of 17 sample plots with homogeneous vegetation characteristics was selected, over which planimetric extentions, tree height (h) and stem diameter (D) were carried out using GPS and forestry instrumentation (hypsometer with lens, dendrometric tripod). Sample plots are showed in red polygons in Fig. 1.

2.3.

Methodology

In this paper, the main goal is to assess the riparian woody biomass through satellite multi-spectral data. For this purpose vegetation parameters e tree high (h) and stem diameter (D) e that represent input variables in standard allometric relationships for woody biomass calculation, are initially quantified. The investigated relationship that links vegetation parameters and spectral information assumes the following general mathematical formulation:

434


Fig. 1 e Study area located along a 3 km stretch of the Avisio river (Trentino Alto Adige, Italy). Field surveys over mixed arboreal are displayed in red polygons. The performed 5-class riparian vegetation map, including mixed arboreal, shrub, herbaceous patterns, bare soil and water bodies is visualized in colour patterns overlaid on the reference 1 m ortophoto. Pictures show the main vegetated patterns. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Vi ¼ fi ðch1 ; .; ch4 Þ;

(1)

where Vi represents the vegetation parameter (dependent variable h or D), ch1,.,4 are the spectral channels (predictor variables) and fi is the unknown analytical function. In order to isolate the spectral properties of the woody vegetation, a preliminary classification of the riparian ecosystem is performed. The four SPOT multi-spectral bands were classified by a pixel-oriented Maximum Likelihood (ML) classifier with Gaussian class-conditional distributions [23] on a 5-class set of land covers including: mixed arboreal, shrub, herbaceous, bare soil and water. The classification is trained and tested by using sample areas and ground control points, respectively, acquired during the field campaign and by visual inspection of the 1 m ortophoto. The classification accuracies were assessed in terms of Conditional Kappa Statistics and Overall Classification Accuracy (OCA). To prevent problems of multi-collinearity between the predictor variables (chi) in Eq. (1), the Normalized Principal Component Analysis (NPCA) is performed on the multi-spectral bands of the vegetated patterns by aggregating mixed

arboreal, shrub and herbaceous classes. The relationship (1) can be adjusted as: Vi ¼ wi ðNPC1 ; .; NPC4 Þ;

(2)

where wi and NPCj are the new unknown analytical functions and the found principal components, respectively. Correlation structures between predictor/dependent variables were analyzed in terms of Spearman rank coefficient (rk) and significance level ( p-value) by extracting over each sample plot ground measurements (Vi) and the average values of the normalized principal components (NPCj). The model identification is performed through visual interpretation of the obtained scatter plots (Vi versus NPCj) and the Akaike procedure, which finds the most appropriate fitting analytical function (w). To calibrate/validate the identified predictive models we divided the ground-monitored sample plots in independent training and testing sets (11 and 6 plots, respectively). Model accuracies were assessed by means of the Generic Analysis of Variance (ANOVA). Then, the predictive models were spatialized over the whole forested floodplain, by providing


spatial maps of tree height (h) and stem diameter (D). Woody biomass (B) is derived by using the following equation developed on similar riparian forest scenarios [24] and adapted here to express the biomass in terms of m3: 2 D ; B ¼ 0:0012 þ 0:34$h$ ðm$pixelÞ$p$ 2

(3)

where h and D refer to the pixel-scale average values, m is the plant density and pixel is the cell size (5 5 m). Forzieri et al. [21] found that SPOT spectral data acquired on the same study area are not able to capture the spatial variability of plant density. However, ground measurements showed a prevalent spatial homogeneity of plant density for mixed arboreal patterns. As a result, the mean value of the plant density measured over all the sample plots (0.12 #/m2), is retained representative of the investigated riparian vegetation. The corresponding uncertainty analysis is estimated by defining at pixel-scale the confidence intervals of the predicted biomass values (a ¼ 0.05). The resulting simulated woody biomass parameterization is compared with the ground-based biomass over the set of sample areas. The ground-based biomass is calculated by using Eq. (3) with field-observations of h and D. Model reliability is quantified in terms of Root Mean Square Errors (RMSE) between simulated and ground-based woody biomass values.

3.

Discussion

The proposed riparian vegetation classification well distinguishes between mixed arboreal, herbaceous, bare soil and water (colour patterns in Fig. 1) by providing encouraging Conditional Kappa Statistics values (0.51, 0.66, 0.78, 0.77 respectively) and Overall Classification Accuracy (OCA ¼ 69.21%). Modest reliability was assessed for the shrub class (Conditional Kappa Statistics ¼ 0.22) due to the spectral overlapping with the mixed arboreal, as evident in the confusion matrix (not shown here for brevity). Even if more sophisticated approaches for riparian vegetation mapping using remote sensing data fusion techniques could enhance the classification process [17,25], the obtained performances are retained an acceptable compromise between classification accuracy, data availability (remote and field surveys) and computation time. The principal component analysis significantly synthesized the vegetation spectral information. As expected, the first eigenvector of the NPCA exhibited the main data variability (NPC1 ¼ 62.36%, NPC2 ¼ 30.57%, NPC3 ¼ 6.69%, NPC4 ¼ 0.36%). High correlation coefficients (rk(h) ¼ 0.783 and rk(D) ¼ 0.695 with p-value < 0.05) confirm that spectral information represents important ecological indicators of riparian biomass (Fig. 2AeB). The adopted aggregation scheme merging all the vegetation classes allows for the extraction of the spectral signatures whose spectral ranges (and related linear transformation) are able to explain the spatial heterogeneity of vegetation properties of mixed arboreal patterns. Different aggregation schemes of land covers for principal component analysis were also explored (singularly extracting each land cover, by merging all the vegetation classes in addition to bare soil and by merging all the land covers) but

435

showed lower correlation values and were omitted here for brevity purposes. The identification analysis drives the selection of the w analytical model among a set of predictive models. The resulting mathematical relationships are in the following form: Vi ¼ Ci þ Ai $ðNPC1 ÞBi ;

(4)

where NPC1 is the first normalized principal component whereas Ai, Bi and Ci are the unknown coefficients calibrated through non-linear least square method (A(h) ¼ 12.476, B(h) ¼ 4.074, C(h) ¼ 1.784 and A(D) ¼ 0.225, B(D) ¼ 3.880, C(D) ¼ 0.030). Results were accurate, especially in terms of high coefficients of determination (R2(h) ¼ 0.732, R2(D) ¼ 0.619) (Fig. 2C and D) and root mean square error (RMSE(h) ¼ 1.597 m, RMSE(D) ¼ 0.033 m). The adjusted coefficient of determination, given the reduced number of sample plots for the testing set, showed mainly low reliability values (adjR2(h) ¼ 0.465, adjR2(D) ¼ 0.238). Vegetation parameter maps of tree height, stem diameter and woody biomass are shown in Fig. 3A, B and C, respectively. Results showed reasonable floodplain biomass parameterization with higher biomass values closed to the month of the Avisio, due to the more abundant water availability for vegetation growth. Average woody biomass (w100 m3/ha) estimated on the whole study area show comparable values to the reference estimates [24]. Reduced levels of uncertainty (w8%) and low RMSE (RMSE(B) ¼ 0.019 m3) of woody biomass (field data vs. remote sensing data) confirm the suitability of the proposed approach to assess the vegetation properties useful for the assessment of potential energy budget. By the simple conversion coefficients it is possible to estimate the potential energetic budget derivable from the woody biomass. For instance, by assuming that willow is the prevalent plant species in Fig. 3C, the corresponding wood chips net calorific value (E ), related to the whole mixed arboreal patterns of the study area (225,550 m2), can be approximately calculated as follows: CD ¼ a0 $a1 ¼ 520=2:43 ¼ 214kg=m3 W ¼ CD$B ¼ 214$ð2238 179Þ ¼ 478932 38306 kg; E ¼ W$a2 ¼ ð478932 38306Þ$12; 2$103 ¼ 5842 467GJ

(5)

where CD is the willow chips bulk density, B (with corresponding prediction interval) is the remote sensing-derived biomass, W is the total wood weight, E is the net calorific value, a0 is the willow dry mass density (520 kg/m3), a1 is the volumetric index (2.43), a2 is the 30% moisture conversion factor for wood chips (12; 2$103 GJ/kg). The obtained estimation neglects corrections of volumetric and mass density factors. Reference values are taken from [26]. The obtained layer, merged with additional information including mechanization level of yards and transport distances, represents a suitable input for decision-making tools developed to compare costs and environmental benefits of the energy use of riparian vegetation trough multi-criteria/multi-attribute approaches [4]. A simple cost-benefit analysis is provided here based on a set of working assumptions taken from [26] for explicative purposes. The wood-energy supply cost is assumed to be 12.4 V/m3 and include: felling (2 chainsaws), full tree skidding (2 tractors and winch), mechanized processing at the landing site (processor on tractor), chipping

436


Fig. 2 e Upper panels: explorative analysis between the first principal component values (on x-axis) and ground-observed vegetation parameters of the mixed arboreal class (on y-axis). Each black circle represents a sample area. Lower panels: comparison of predicted (on x-axis) and observed parameters (on y-axis). Calibration and validation sample plots are shown in black and grey circles, respectively and the black line indicates the regression line 1:1. Graphics are shown for tree height (h) in panels AeC and stem diameter (D) in panels BeD.

Fig. 3 e Structural parameter maps obtained for the mixed arboreal patterns: tree height (A), stem diameter (B) and woody biomass (C ). Values refer to 5 3 5 m pixel size.


harvesting residues (high power chipper), delivery of chips (truck and trailer). The economical benefit derivable from the wood chips price with 30% moisture content is approximately estimated 90 V/t. Based on the hypnotized scenario the total costs and economical benefits for the considered riparian area amount to 27,751 V and 43,103 V, respectively, with a corresponding gain of 15,352 V.

4.

Conclusion

This paper explores the satellite multi-spectral capability to retrieve riparian vegetation parameters (tree high and stem diameter) used in this study as woody biomass indicators. Results showed strong correlations between spectral-derived values and tree high/stem diameter. Tri-parametric power laws calibrated/validated by using field surveys allowed to remotely-derived spatial maps of vegetation parameters to use in allometric relationships for automatic and reliable spatiotemporal monitoring of riparian woody biomass. Findings indicate that the proposed methodology is useful for a sustainable buffer management and an integrated environmental modelling. The proposed method was tested only over a limited area and additional investigations over different vegetation conditions would be important to fully test the procedure for operational applications. New generation of high spatial resolution satellite hyper-spectral data will furthermore enhance the remote sensing of woody biomass, by discriminating vegetation species with different calorific values (i.e. Poplar vs. Willows vs. Acacia) and by potentially enlarging the spectral information useful for plant density retrieval.

Acknowledgements I thank Maurizio Righetti, Massimo Degetto (Centro Universitario per la Difesa Idrogeologica per la Difesa dell’Ambiente Montano, Univerisity of Trento) and Prof. Federico Preti (Dipartimento di Ingegneria Agraria e Forestale, Univerisity of Florence) for the provision of remote sensing and field data. I particularly acknowledge Prof. Fabio Castelli (Dipartimento di Ingegneria Civile Ambientale, University of Florence) for the collaborative assistance and the two anonymous reviewers who improved the quality of the work.

references

[1] Naiman RJ, DeCamps H, Pollock M. The role of riparian corridors in maintaining regional biodiversity. Ecol Appl 1993;3:209e12. [2] RIPFOR. Hydraulic, sedimentological and ecological problems of multifunctional riparian forest management e RIPFOR e guidelines for end-users. Quality of Life and Management of Living Resources Programme, (Key Action No. 5), 2006. [3] Rockwood DL, Naidu CV, Carter DR, Rahmani M, Spriggs TA, Lin C, et al. Short-rotation woody crops and phytoremediation: opportunities for agroforestry. Agrofor Syst 2004;61e62:51e63.

437

[4] Recchia L, Cini E, Corsi S. Multicriteria analysis to evaluate the energetic reuse of riparian vegetation. Appl Energy 2010; 87:310e9. [5] Candolo G. Energia dalle biomasse vegetali: le opportunita` per le aziende agricole. Agronomica 2006;4:26e35 (in Italian). [6] Castellani C, Scrinzi G, Tabacchi G, Tosi V. Inventario forestale nazionale italiano (in Italian). Trento: Ministero dell’agricoltura e delle foreste; 1984. [7] Winward AH. Monitoring the vegetation resources in riparian areas. USDA Forest Service; 2000. Gen. Tech. Rep. RMRS-GTR-47. [8] Coles-Ritchie MC, Henderson RC, Archer EK, Kennedy C, Kershner JL. Repeatability of riparian vegetation sampling methods: how useful are these techniques for broad-scale, long-term monitoring? USDA Forest Service; 2004. Gen. Tech. Rep. RMRS-GTR-138. [9] Straatsma MW, Warmink JJ, Middelkoop H. Two novel methods for fields measurements of hydrodynamic density of floodplain vegetation using terrestrial laser scanning and digital parallel photography. Int J Remote Sensing 2008;29(5): 1595e617. [10] Ens JA, Farrell FE, Be´langer N. Rapid biomass estimation using optical stem density of willow (Salix spp.) grown in short rotation. Biomass Bioenergy 2009;33:174e9. [11] Straatsma MW, Baptist MJ. Floodplain roughness parameterization using airborne laser scanning and spectral remote sensing. Remote Sensing Environ 2008;112: 1062e80. [12] Forzieri G, Guarnieri L, Vivoni ER, Castelli F, Preti F. Multiple attribute decision-making for individual tree detection using high-resolution laser scanning. For Ecol Manag 2009;258(11): 2501e10. [13] Forzieri G, Guarnieri L, Vivoni ER, Castelli F, Preti F. SpectralALS data fusion for different roughness parameterizations of forested floodplains. River Res Appl 2011;27:826e40. [14] Forzieri G, Castelli F, Preti F. Advances in hydraulic roughness remote sensing. Int J Remote Sensing doi:10.1080/ 01431161.2010.531788, in press. [15] Zheng D, Rademacher J, Chen J, Crow T, Bresee M, Le Moine J, et al. Estimating aboveground biomass using Landsat 7 ETMþ data across a managed landscape in northern Wisconsin, USA. Remote Sensing Environ 2004;93:402e11. [16] Roy PS, Ravan SA. Biomass estimation using satellite remote sensing data - an investigation on possible approaches for natural forest. J Biosciences 2010;21(4):535e61. [17] Dillabaugh KA, King DJ. Riparian marshland composition and biomass mapping using Ikonos imagery. Can J Remote Sensing 2008;34(2):143e58. [18] Colombo R, Bellingeri D, Fasolini D, Marino CM. Retrieval of leaf area index in different vegetation types using high resolution satellite data. Remote Sensing Environ 2003;86: 120e31. [19] Wulder MA, LeDrew EL, Franklin SE, Lavigne MB. Aerial image texture information in the estimation of northern deciduous and mixed wood forest leaf area index (LAI). Remote Sensing Environ 1998;64:64e76. [20] Nagler PL, Glenn EP, Thompson TL, Huete A. Leaf area index and normalized difference vegetation index as predictors of canopy characteristics and light interception by riparian species on the lower Colorado River. Agric For Meteorol 2004; 125:1e17. [21] Forzieri G, Degetto M, Righetti M, Castelli F, Preti F. Satellite multispectral data for improved floodplain roughness modelling. J Hydrology 2011;407(1e4):41e57. [22] ENVI. ENVI reference guide. ITT Visual Information Solutions; 2009. [23] Richards JA, Jia X. Remote sensing digital image analysis. 4th ed. Berlin, Germany: Springer; 2006.

438


[24] Tabacchi G, Maltoni ML. Tavole dendrometriche regionali (in Italian). Bologna: Assessorato agricoltura, ambiente e sviluppo sostenibile; 2000. [25] Forzieri G, Moser G, Vivoni ER, Castelli F. Riparian vegetation mapping for hydraulic roughness estimation using very high

resolution remote sensing and LiDAR data fusion. J Hydraulic Eng 2010;136(11):855e67. [26] Francescato V, Antonini E, Zuccoli Bergomi L. Wood fuels handbook. AIEL e Italian Agriforestry Energy Association Publisher; 2008.