Assessments of long-term means and trends of eight evapotranspiration products over China based on water balance method

Accurate quantification of evapotranspiration (ET) is essential for a better understanding of hydrological processes and the interactions between hydrological, climatic, and vegetation systems. Existing global ET products have significant differences in describing regional ET and its trends in China. Limited by the short coverage period of observational data such as runoff, previous studies have rarely evaluated the performance of ET products in reproducing long-term ET and its trends. In addition, studies evaluating ET products based on the water balance method often use single input data, which increases the uncertainty of the water balance method to a certain extent.  In order to better understand the applicability of commonly used global land surface ET products in China's watersheds, we evaluated the performance of eight ET products (i.e., 6 widely used global ET products: ERA5L, GLDAS, MODIS, FLUXCOM, GLEAM, PMLV2, and 1 newly released ET product with two different resolution datasets: CAMELE) in simulating monthly and annual ET and their ability to describe long-term trends in ET, using the multi-source input water balance method in 133 small basins in China. The results show that all products overestimate basin ET, whether on monthly or annual scale, with GLEAM and  PMLV2 performing best with an RMSE of less than 50mm/month and the overall deviation less than 50% in most basins, followed by MODIS, and CAMELE's two resolution products having the most overall overestimation. Remote sensing-based evapotranspiration products GLEAM, PMLV2, and MODIS generally have better accuracy, followed by GLDAS. All products have greater uncertainty mainly in the small basins of the Southeastern Rivers and Pearl River Basin with the highest RMSE, while perform better in the upper source areas of the Yangtze River Basin, the Yellow River Basin, the Hai River Basin and the Songliao Basin. The average ET over all basins shows an increasing trend which is 1.49mm/year² from 1980 to 2016 and 1.08mm/year² from 1980 to 2014. All four long-series ET products (i.e., ERA5L, GLDAS, GLEAM, CAMELE-0.25°) capture this trend, but GLEAM and GLDAS overestimate the trend of ET, and the other products underestimate the corresponding trend. ET-WB mainly experiences two stages of change, gradually decreasing from 1980 to 2001, and then begin to rise during 2002-2016. All long-series products capture this change process. All products underestimate the increasing trend of ET in most basins and cannot describe the spatial distribution of ET trend well. The interannual variation of ET-WB has greater fluctuations, and all products underestimate the Cv of ET-WB. In contrast, PMLV2 performs best, followed by GLDAS, GELAM, and MODIS, while FLUXCOM performs worst, followed by CAMELE-0.25°, ERA5L, and CAMELE-0.1°. This latest assessment helps to understand the status and development of current land surface ET datasets and provides guidance for selecting appropriate ET products for use in specific regions within China and its interior.