Résumé
In many tropical areas, evapotranspiration is the most important but least known component of the water cycle. An innovative method, named E3S (for EVASPA S-SEBI Sahel), was developed to provide spatially-distributed estimates of daily actual evapotranspiration (ET d) from remote sensing data in the Sahel. This new method combines the strengths of a contextual approach that is used to estimate the evaporative fraction (EF) from surface temperature vs. albedo scatterograms and of an ensemble approach that derives ET d estimates from a weighted average of evapotranspiration estimated from several EF methods. In this work, the two combined approaches were derived from the simplified surface energy balance index (S-SEBI) model and the EVapotranspiration Assessment from SPAce (EVASPA) tool. Main innovative aspects concern (i) ensemble predictions of ET d through the implementation of a dynamic weighting scheme of several evapotranspiration estimations, (ii) epistemic uncertainty of the estimation of ET d from the analysis of the variability of evapotranspiration estimates, and (iii) a new cloud filtering method that significantly improves the detection of cloud edges that negatively affect EF determination. E3S was applied to MODIS/TERRA and AQUA datasets acquired during the 2005-2008 period over the mesoscale AMMA-CATCH (Analyse Multidisciplinaire de la Mousson Africaine-Couplage de l'Atmosphère Tropicale et du Cycle Hydrologique) observatory in SouthWest Niger. E3S estimates of instantaneous and daily available energy, evaporative fraction, and evapotranspiration were evaluated at a local scale based on two field-monitored plots representing the two main ecosystem types in the area-a millet crop and a fallow savannah bush. In addition to these ground-based observations, the local scale evaluation was performed against continuous simulations by a locally-calibrated soil-vegetation-atmosphere transfer model for the two plots. The RMSE (root mean square error) from this comparison for E3S's ET d estimates from combined AQUA/TERRA sources was 0.5 mm·day −1 , and the determination coefficient was 0.90. E3S significantly improved representation of the evapotranspiration seasonality, compared with a classical implementation of S-SEBI or with the original EVASPA's non-weighted ensemble scheme. At the mesoscale, ET d estimates were obtained with an average epistemic uncertainty of 0.4 mm·day −1. Comparisons with the reference 0.25 •-resolution GLEAM (global land evaporation Amsterdam model) product showed good agreement. These results suggested that E3S could be used Remote Sens. 2020, 12, 380 2 of 34 to produce reliable continuous regional estimations at a kilometric resolution, consistent with land and water management requirements in the Sahel. Moreover, all these innovations could be easily transposed to other contextual approaches.