Résumé
Evapotranspiration (ET) plays a crucial role in estimating groundwater recharge, which is critical to ensure water resource management, particularly in countries prone to difficult access to water such as Madagascar. This study estimates ten ET products derived from remote sensing (RS), land surface modeling (LSMs), and reanalysis methods. To evaluate the performance of these datasets, we use a water balance approach, comparing precipitation minus runoff (P−Q) at the catchment scale. This comparison covers nine catchments in Madagascar, focusing on humid and semi-humid regions over monthly to annual timescales from 2000 to 2014. We also take advantage of the GRACE/-FO satellite missions (2002–2023) to estimate large-scale variability in water storage. Analyses were performed at different spatial scales (basin level, bioclimatic zone level, and across the entire island) and various timescales (monthly, annual, and interannual). Results highlight significant differences in ET product performances. ERA5 (the fifth-generation ECMWF atmospheric reanalysis) and GLEAM (Global Land Evaporation Amsterdam Model) show the best performance overall. MERRA-2 (Modern-Era Retrospective Analysis for Research and Applications, Version 2) and GLDAS-CLSM (Global Land Data Assimilation System with the Community Land Surface Model) exhibit significant errors and biases. Understanding these differences requires addressing the uncertainties in the input data and the physical methods employed by each ET product. These results allow us to better understand the impact of extreme weather events (e.g. droughts and cyclones) over water and vegetation dynamics spatialized across Madagascar.