

Yebra, M; Van Dijk, A; Leuning, R; Huete, A; Guerschman, JP (2013). Evaluation of optical remote sensing to estimate actual evapotranspiration and canopy conductance. REMOTE SENSING OF ENVIRONMENT, 129, 250261. Abstract We compared estimates of actual evapotranspiration (ET) produced with six different vegetation measures derived from the MODerate resolution Imaging Spectroradiometer (MODIS) and three contrasting estimation approaches using measurements from eddy covariance flux towers at 16 FLUXNET sites located over six different land cover types. The aim was to assess optimal approaches in using optical remote sensing to estimate ET. The first two approaches directly regressed various MODIS vegetation indices (VIs) and products such as leaf area index (LAI) and fraction of photosynthetically active radiation (fPAR) with ET and evaporative fraction (EF). In the third approach, the PenmanMonteith (PM) equation was inverted to obtain surface conductance (G(s)), for dry plant canopies. The G(s) values were then regressed against the MODIS data products and used to parameterize the PM equation for retrievals of ET. JackKnife crossvalidation was used to evaluate the various regression models against observed ET. The PMG(s) approach provided the lowest root mean square error (RMSE), and highest determination coefficients (R2) across all sites, with an average RMSE = 38 W m(2) and R2=0.72. Direct regressions of observed ET against the VIs resulted in an average RMSE = 60 W m(2) and R2=0.22, while the EF regressions an average RMSE = 42 W m(2) and R2=0.64. The MODIS LAI and fPAR product produced the poorest estimates of ET (RMSE> 44 W m(2) and R2<0.6); while the VIs each performed best for some of the land cover types. The enhanced vegetation index (EVI) produced the best Er estimates for evergreen needleleaf forest (RMSE = 28.4W m(2), R2=0.66). The normalized difference vegetation index (NDVI) best estimated ET in grassland (RMSE = 23.8 W m(2) and R2=0.68), cropland (RMSE = 29.2 W m(2) and R2=0.86) and woody savannas (RMSE = 25.4 W m(2) and R2=0.82), while the VIbased crop coefficient (Kc) yielded the best estimates for evergreen and deciduous broadleaf forests (RMSE = 27 W m(2) and R2=0.7 in both cases). Using the ensembleaverage of ET as estimated using NDVI, EVI and Kc we computed global grids of dry canopy conductance (G(c)) from which annual statistics were extracted to characterise different functional types. The resulting G(c) values can be used to parameterize land surface models. (C) 2012 Elsevier Inc. All rights reserved. DOI:
ISSN: 00344257 