Divergent performance of multiple satellite-based products for monitoring water use efficiency across diverse agroecosystems

Multiple satellite remote sensing-based gross primary productivity (GPP) and evapotranspiration (ET) products have been developed to simulate the spatiotemporal patterns of carbon and water cycles in terrestrial ecosystems. However, the performance of these products in monitoring cropland water use efficiency (WUE) has rarely been evaluated. In this study, a total of 140 site-years of flux data across the maize, soybean, rice and winter wheat ecosystems were used as a benchmark to assess the performance of such products over 8-day, monthly and annual time scales, including the Breathing Earth System Simulator (BESS), Global Land Surface Satellite (GLASS), Moderate Resolution Imaging Spectroradiometer (MODIS), and the Penman-Monteith-Leuning V2 (PML), respectively. Our site-level evaluation demonstrated that the performance of such satellite-based products for monitoring WUE varied considerably across diverse agroecosystems. BESS WUE outperformed the other three products at all time scales when mixing all crop types together. However, the optimal product varied for specific crop across 8-day, monthly and yearly scales. At the 8-day scale, BESS performed best in simulating WUE for maize, while PML was superior for the other three crops. It indicated that the models that closely coupled carbon and water cycles tend to yield more robust WUE estimates. At the monthly scale, the BESS again provided the most accurate WUE for maize and soybean, whereas the PML product performed best for rice and winter wheat. Nevertheless, all products showed limitations, particularly in capturing the interannual variations of cropland WUE. On an annual scale, BESS exhibited the best accuracy (2.50 vs 2.51 g C kg−1 H2O), followed by PML (2.49 g C kg−1 H2O), MODIS (2.03 g C kg−1 H2O) and GLASS (1.53 g C kg−1 H2O) models. Our evaluation underscores the potential of integrating carbon and water cycles into models to enhance WUE performance, thereby providing a direction for future improvements in model structures.