Ability of a soil–vegetation–atmosphere transfer model and a two-source energy balance model to predict evapotranspiration for several crops and climate conditions

Abstract

Abstract. The heterogeneity of Agroecosystems, in terms of hydric conditions, crop types and states, and meteorological forcing, is difficult to characterize precisely at the field scale over an agricultural landscape. This study aims to perform a sensitivity study with respect to the uncertain model inputs of two classical approaches used to map the evapotranspiration of agroecosystems: (1) a surface energy balance (SEB) model, the Two-Source Energy Balance (TSEB) model, forced with thermal infrared (TIR) data as a proxy for the crop hydric conditions, and (2) a soil–vegetation–atmosphere transfer (SVAT) model, the SEtHyS model, where hydric conditions are computed from a soil water budget. To this end, the models' skill was compared using a large and unique in situ database covering different crops and climate conditions, which was acquired over three experimental sites in southern France and Morocco. On average, the models provide 30 min estimations of latent heat flux (LE) with a RMSE of around 55 W m−2 for TSEB and 47 W m−2 for SEtHyS, and estimations of sensible heat flux (H) with a RMSE of around 29 W m−2 for TSEB and 38 W m−2 for SEtHyS. A sensitivity analysis based on realistic errors aimed to estimate the potential decrease in performance induced by the spatialization process. For the SVAT model, the multi-objective calibration iterative procedure (MCIP) is used to determine and test different sets of parameters. TSEB is run with only one set of parameters and provides acceptable performance for all crop stages apart from the early growing season (LAI < 0.2 m2 m−2) and when hydric stress occurs. An in-depth study on the Priestley–Taylor key parameter highlights its marked diurnal cycle and the need to adjust its value to improve flux partitioning between the sensible and latent heat fluxes (1.5 and 1.25 for France and Morocco, respectively). Optimal values of 1.8–2 were highlighted under cloudy conditions, which is of particular interest due to the emergence of low-altitude drone acquisition. Under developed vegetation (LAI > 0.8 m2 m−2) and unstressed conditions, using sets of parameters that only differentiate crop types is a valuable trade-off for SEtHyS. This study provides some scientific elements regarding the joint use of both approaches and TIR imagery, via the development of new data assimilation and calibration strategies.