Evaluating satellite-derived evapotranspiration estimates against ground-based measurements in arid upper orange catchment, South Africa

Abstract

Evapotranspiration (ET) accounts for ~60% of surface water loss, and accurate estimation is vital for water management in arid regions. This study assessed ET in the Upper Orange catchment, South Africa, using the surface energy balance system (SEBS), Penman–Monteith (PM) and MODIS 16 against eddy covariance (EC) flux tower data. Landsat 8 and MODIS were analyzed for eight days in 2020. Model accuracy was evaluated using RMSE, Pbias and R², while ANOVA was used to test differences across land cover types. SEBS showed the highest accuracy (RMSE = 2.48 mm/day, R² = 0.71), outperforming MODIS (RMSE = 3.21 mm/day, R² = 0.59), while PM performed moderately well (RMSE = 2.85 mm/day, R² = 0.66). Both SEBS and PM overestimated ET by 25%–33%. ANOVA confirmed significant land cover differences (p = 0.04), with grasslands recording the highest ET. Overall, SEBS provided robust and spatially reliable estimates, offering valuable insights for water accounting and policy development