LAUSR

Abstract Gridded precipitation datasets for the arid region of northwestern China (ARNC) are thought to have large discrepancies. This study systematically evaluated the performance of multiple datasets for the ARNC over the period 1980–2014 using the observed gridded precipitation dataset (OBS) from the China Meteorological Administration (CMA). The various datasets include two National Center for Environmental Prediction (NCEP) reanalyses (NCEP-1 and NCEP-2), two Climate Prediction Center (CPC) Merged Analyses of Precipitation (CMAP-1 and CMAP-2), the Interim European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA-Interim), the Global Precipitation Climatology Project (GPCP), the Climatic Research Unit, University of East Anglia, UK (CRU), and the second Modern-Era Retrospective Analysis for Research and Application system (MERRA-2). The results show that all the multiple datasets reasonably reproduce the climatology, seasonality, interannual variability, and spatiotemporal patterns of ARNC precipitation, but there are discrepancies in the long-term trends. MERRA-2 and CMAP-2 best capture the spatial distributions and inter-annual variability of mean annual precipitation (MAP). CRU best reproduced the seasonality and temporal variability of MAP. However, ERA-Interim and NCEP overestimated the precipitation amount in mountainous regions, and MERRA-2 underestimated precipitation in almost all seasons. Empirical orthogonal function (EOF) analysis indicated that GPCP and CMAP-2 better described both the spatial modes and principal components of the first two EOF patterns. The systematic evaluation in this study indicated that, overall, most of the satellite-merged datasets (GPCP, MERRA-2, and CMAP-2) perform better than gauge-based and reanalysis datasets alone. This study demonstrates that a systematic evaluation of the differences between multiple datasets is critical for reducing discrepancies.