Estimating evapotranspiration in the Sahel during 2003-2008 using satellite products

The land surface in the Sahel is a strong amplifier of the inter-annual variability of the West African Monsoon (Giannini et al., 2003). Therefore, making regional estimates of evapotranspiration available is critical for improving our understanding of land surface-atmosphere interactions and for a better agricultural and hydrological planning. The objective of this study is to provide time-series of daily evapotranspiration in the Sahel during 2003-2008 using a set of satellite products. Two operational approaches that do not require calibration or prescribed information about plant functional types were evaluated and compared. The first is a two layer model based on the Priestley-Taylor equation for potential evapotranspiration depressed according to multiple stresses (Fisher et al., 2008). The original model version was modified to run at daily time- scale. A soil water constraint for soil evaporation based on thermal inertia from surface temperature replaced the original based on the atmospheric water deficit to reduce dependence from coarse climatic datasets. The second is a single-source approach that estimates the evaporative fraction based on the spatial scatterplots -resembling a triangle- between the morning rise in surface temperature and a vegetation index (Stisen et al., 2008). The evaporative fraction for each pixel is retrieved assuming that the edges of the triangle correspond to hydrological extremes for soil moisture and evapotranspiration, by interpolating between them.

Both models were run using inputs derived from the Meteosat Second Generation SEVIRI sensor (4 km pixel) including Land Surface Temperature (LST), albedo, and daily downward surface longwave and shortwave fluxes (4 km pixel). LAI, fPAR and NDVI were acquired from MODIS. Air temperature from ECWMS (40 km resolution) was used for Fisher's model. Daily model outputs were compared at the site-level with eddy flux measurements in Dahra (Senegal) and in Agoufou (Mali). At a regional level model differences were analyzed spatially using EOF (Empirical Orthogonal Functions) to detect those areas with greatest model divergences and lower confidence in model results.