Actual evapotranspiration in drylands derived from in-situ and satellite data: Assessing biophysical constraints

Improving regional estimates of actual evapotranspiration (lambda E) in water-limited regions located at climatic transition zones is critical. This study assesses an lambda E model (PT-JPL model) based on downscaling potential evapotranspiration according to multiple stresses at daily time-scale in two of these regions using MSG-SEVIRI (surface temperature and albedo) and MODIS products (NDVI, LAI and f(PAR)). An open woody savanna in the Sahel (Mali) and a Mediterranean grassland (Spain) were selected as test sites with Eddy Covariance data used for evaluation. The PT-JPL model was modified to run at a daily time step and the outputs from eight algorithms differing in the input variables and also in the formulation of the biophysical constraints (stresses) were compared with the lambda E from the Eddy Covariance. Model outputs were also compared with other modeling studies at similar global chyland ecosystems. The novelty of this paper is the computation of a key model parameter, the soil moisture constraint, relying on the concept of apparent thermal inertia (f(SM-ATI)) computed with surface temperature and albedo observations. Our results showed that f(SM-ATI) from both in-situ and satellite data produced satisfactory results for lambda E at the Sahelian savanna, comparable to parameterizations using field-measured Soil Water Content (SWC) with r(2) greater than 0.80. In the Mediterranean grasslands however, with much lower daily lambda E values, model results were not as good as in the Sahel (r(2)= 0.57-0.31) but still better than reported values from more complex models applied at the site such as the Two Source Model (TSM) or the Penman-Monteith Leuning model (PML). PT-JPL-daily model with a soil moisture constraint based on apparent thermal inertia, f(SM-ATI) offers great potential for regionalization as no field-calibrations are required and water vapor deficit estimates, required in the original version, are not necessary, being air temperature and the available energy (Rn-G) the only input variables required, apart from routinely available satellite products. (c) 2012 Elsevier Inc. All rights reserved.