Evaluation of ERA5, ERA-Interim, JRA55 and MERRA2 reanalysis precipitation datasets over the Poyang Lake Basin in China

Increasing variability of precipitation induced by climate change raises uncertainties for ecosystem services and agricultural production in the Poyang Lake Basin (PLB), China. Reanalysis precipitation datasets (RPDs) are attractive alternatives for monitoring the hydro-climate cycles. However, their applicability in China has mainly been evaluated at the national scale, while less attention has been paid to the PLB or southeastern China. Here, we apply various metrics to evaluate the magnitude and spatial similarity of annual, monthly, different intensities and the spatial structure of precipitation from four reanalysis-the fifth global atmospheric analysis dataset of the European Center for Medium-Range Weather Forecasts (ECMWF) (ERA5), Interim ECMWF Reanalysis (ERA-Interim), Japanese 55-year Reanalysis project (JRA55) and Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2), against the rain-gauges over the PLB during 1980-2018. The results showed that the detection capacity of ERA5 and MERRA2 shifted around 2002, whereas that of ERA-Interim and JRA55 did not substantially change. Before 2002, ERA-Interim had the lowest biases for the magnitude of annual, monthly and different intensities of precipitation (except for heavy rain), while ERA5 and MERRA2 had strengths in capturing spatial correlation. Notably, ERA5 and MERRA2 had region-wide wet and dry biases, respectively, before 2002, which can be attributed mainly to the estimation of flood season precipitation (from April to June) and heavy rain. In contrast, the fractional contribution of light-moderate rain was underestimated and overestimated by ERA5 and MERRA2, respectively. However, after 2002, because of the improved reproducibility of flood season precipitation and heavy rain magnitude, the systematic biases of ERA5 and MERRA2 significantly decreased, especially for ERA5, which outperformed other RPDs on all aspects of precipitation. The conclusions of this paper suggest that the variation of applicability over time should be seriously tested before the utilization of RPDs in hydro-climate analyses.