Evaluation of remotely sensed global evapotranspiration data from inland river basins

Remotely sensed (RS) evapotranspiration (ET) data sets have been widely evaluated and applied at regional scale. However, the research is relatively insufficient on inland basins. RS global ET data sets were evaluated in 19 major inland basins around the world in this study. Monthly ET was estimated by using water balance (ET_WB), as well as multi-source precipitation and total water storage anomaly data sets from each basin. The main attributes of the ET changes were identified in each basin. The accuracy of 10 ET products was evaluated by monthly ET_WB series over the entire study region and at the basin scales. Results showed that precipitation changes were the main cause of ET changes in most basins (i.e., 11/19). The impacts of human activities, mainly because of increased irrigation diversion, dominated the ET changes in the five basins located in the western and northern regions of the Eurasian inland basin. Accurate estimated ET series in inland basins with a regular unimodal or U distribution of mean monthly ET were also obtained; this was not the case for basins with irregular seasonal ET variations. In general, ET products exhibited accurate monthly ET estimates in most inland basins, but their accuracy was reduced at the Helmand River, Iranian inland rivers, Qaidam, and Turpan Basins. All products tended to underestimate ET, except for the Penman-Monteith-Leuning product. Overall, the global land surface satellite and global land evaporation Amsterdam model products performed the best in inland basins, while the moderate resolution imaging spectroradiometer (MODIS; MOD16A2) and surface energy balance system (SEBS) products performed relatively poorly. In addition, all products except SEBS and MOD16A2 could capture the spatial ET distribution in inland basins, and the spatial correlation coefficient ranged between 0.83 and 0.93. In general, an ensemble of optimized ET products could improve the accuracy of simulated monthly ET and reduce the uncertainty of the simulated series over the entire study region and at the basin scales.