Assessment of precipitation from the CRA40 dataset and new generation reanalysis datasets in the global domain

Abstract

AbstractThe China Meteorological Administration recently released its first reanalysis product, the CRA40 dataset covering the time period 1979–2018. We evaluate the performance of the CRA40 dataset in reproducing the observed precipitation climatology and variations over the global domain and typical monsoon regions during the period 1980–2018. In this study, four additional reanalysis precipitation datasets—the European Centre for Medium‐Range Weather Forecasts Reanalysis v5 (ERA5), Japanese 55‐year reanalysis (JRA55), Modern‐Era Retrospective Analysis for Research and Applications version 2 (MERRA2) and Climate Forecast System Reanalysis—are compared against the Global Precipitation Climatology Project (GPCP). For the global climatology, there are considerable differences in the performance of the five reanalysis products over south‐central Asia (85–95°E, 20–40°N), where CRA40 exhibits the wettest bias across all months especially for the south of 30°N bands. JRA55 produces more intense precipitation with a larger bias over the Atlantic and Pacific intertropical convergence zones (ITCZs). Meanwhile, CRA40 tends to slightly underestimate low rainfall, but overestimates high rainfall. In terms of trend, except MERRA2, the average global/hemispheric precipitation increased by 0.02–0.11 mm/day/decade in all the reanalysis datasets, which is attributed to clearly increased trends along the ITCZ. In contrast, a non‐significant increasing trend is observed. JRA55 generally has the empirical orthogonal function pattern most consistent with the GPCP, with the highest spatial and temporal correlations. Moreover, CRA40 correlates better with the Niño 3.4 index, producing precipitation patterns similar to the El Niño–Southern Oscillation. For six typical monsoon systems, CRA40 has a less consistent local precipitation average with a larger normalized standard deviation and centred root‐mean‐square error, and lower correlation coefficient than other reanalyses when compared with the GPCP.