Estimating evapotranspiration in a semi-arid catchment: A comparison of hydrological modelling and remote-sensing approaches

Abstract

Reliable spatial data of evapotranspiration (ET) in support of water resources management are limited. ET is a major component of the water balance, in many regions, and therefore it is critical that it be accurately quantified. To identify a product that accurately estimates spatially distributed ET for application in data-scarce regions, an inter-model comparison was conducted between the MOD16 ET dataset and the ET calculated with the calibrated and validated JAMS/J2000 hydrological model in the Sandspruit catchment (South Africa). Annual JAMS-ET and MOD16-ET data were generally consistent. Monthly JAMS-ET and MOD16-ET dynamics are influenced by the response of vegetation to precipitation as well as the atmospheric evaporative demand. The maximum correlation coefficient between JAMS-ET and MOD16-ET was 0.82 and it was evident at Lag 0, showing that both ET estimates are in phase when evaluated at the basin scale. The maximum correlation coefficients between the ET estimators and precipitation were 0.67 and 0.70 for JAMS-ET and MOD16-ET, respectively, and this was evident at Lag 2 (1 lag is 1 month) for both methods. This suggests that there is a 2-month delay in the maximum response of ET to precipitation. The models did not exhibit significant dependence on the seasonal distribution of precipitation. The complementary use of hydrological modelling and satellite-derived data may be greatly advantageous to water resources management, e.g., water allocation studies, ecological reserve determinations and vegetation water use studies. The results of the inter-model comparison also provide motivation for the use of the MOD16 ET dataset to estimate ET in data-scarce regions. Additionally, this study provides evidence for the potential use of validated satellite-based ET data as inputs in hydrological models. This may facilitate a more realistic representation of the catchment hydrological processes.