Intercomparison of four remote-sensing-based energy balance methods to retrieve surface evapotranspiration and water stress of irrigated fields in semi-arid climate

Abstract

Abstract. Instantaneous evapotranspiration rates and surface water stress levels can be deduced from remotely sensed surface temperature data through the surface energy budget. Two families of methods can be defined: the contextual methods, where stress levels are scaled on a given image between hot/dry and cool/wet pixels for a particular vegetation cover, and single-pixel methods, which evaluate latent heat as the residual of the surface energy balance for one pixel independently from the others. Four models, two contextual (S-SEBI and a modified triangle method, named VIT) and two single-pixel (TSEB, SEBS) are applied over one growing season (December–May) for a 4 km × 4 km irrigated agricultural area in the semi-arid northern Mexico. Their performance, both at local and spatial standpoints, are compared relatively to energy balance data acquired at seven locations within the area, as well as an uncalibrated soil–vegetation–atmosphere transfer (SVAT) model forced with local in situ data including observed irrigation and rainfall amounts. Stress levels are not always well retrieved by most models, but S-SEBI as well as TSEB, although slightly biased, show good performance. The drop in model performance is observed for all models when vegetation is senescent, mostly due to a poor partitioning both between turbulent fluxes and between the soil/plant components of the latent heat flux and the available energy. As expected, contextual methods perform well when contrasted soil moisture and vegetation conditions are encountered in the same image (therefore, especially in spring and early summer) while they tend to exaggerate the spread in water status in more homogeneous conditions (especially in winter). Surface energy balance models run with available remotely sensed products prove to be nearly as accurate as the uncalibrated SVAT model forced with in situ data.