Upscaling evapotranspiration measurements from multi-site to the satellite pixel scale over heterogeneous land surfaces

Upscaling evapotranspiration measurements from multi-site to the satellite pixel scale over heterogeneous land surfaces

Agricultural and Forest Meteorology

The acquisition of “ground-truth” land surface evapotranspiration (ET) data at the satellite pixel scale over heterogeneous land surfaces is crucial to develop ET estimation models and improve the accuracy of remotely sensed ET values. However, few studies have focused on methods of acquiring ET data at the satellite pixel scale. Based on multi-site eddy covariance (EC) system measurements from the “Multi-Scale Observation Experiment on Evapotranspiration over heterogeneous land surfaces” in the middle reaches of the Heihe River Basin, five upscaling methods were compared and a combined method was developed to acquire “ground-truth” ET data at the satellite pixel scale. First, this study evaluated the performances of three simple upscaling methods (the arithmetic average, area-weighted and footprint-weighted methods). The results showed that the three simple upscaling methods perform well in the relatively homogeneous pixels. For the area-weighted method, the mean absolute percentage error (MAPE) for these pixels was 6.1%. However, the accuracy was worse in the relatively heterogeneous pixels, with a MAPE of 10.8% due to the surface heterogeneity significantly affecting the accuracy of the upscaled results. Second, the upscaling of ET results from heterogeneous land surfaces at the satellite pixel scale can be significantly improved by using two upscaling methods introducing auxiliary variables (the integrated Priestley-Taylor equation method and the area-to-area regression kriging method), that can characterize the heterogeneity of the surface water and heat conditions. Finally, a combined method (applied the area-weighted method for relative homogeneous surfaces, otherwise used the method introducing auxiliary variables) was proposed to acquire both instantaneous and daily “ground-truth” ET data at the satellite pixel scale at the time of a MODIS overpass. The uncertainties of the “ground-truth” ET data were evaluated, taking the large aperture scintillometer (LAS) measurements as the satellite pixel reference. The results show that the proposed upscaling method is reasonable and feasible, and therefore could bridge the gap between in situ ET measurements and remote-sensing estimates of ET.