Evaluation and correction of the TRMM 3B43V7 and GPM 3IMERGM satellite precipitation products by use of ground-based data over Xinjiang,China

Satellite retrieval of precipitation is a great challenge in arid regions where light precipitation prevails. To improve the accuracy of satellite precipitation products in Xinjiang, which is the driest region in China, this study first evaluated the performance of two widely used monthly satellite products from April 2014 to August 2017. The first was the Tropical Rainfall Measuring Mission (TRMM) 3B43 version 7 (hereafter 3B43V7), and the second was the Integrated Multi-Satellite Retrievals for the Global Precipitation Measurement (GPM) mission (IMERG) (hereafter 3IMERGM). The 3IMERGM product was then corrected by a stepwise regression model using topographic variables derived from digital elevation model (DEM). A comparison between satellite estimates and in situ measurements indicates that both 3B43V7 and 3IMERGM overestimate precipitation overall, but that 3IMERGM performs better than 3B43V7. The relative biases (RB) of 3B43V7 and 3IMERGM were found to be 10.24 and 7.76%, respectively. The correlation coefficient (CC) between 3IMERGM and the gauges is 0.68, which is higher than the CC observed (i.e., 0.62) between 3B43V7 and the gauges. To comprehensively evaluate the performance of the corrected model, a tenfold cross-validation method was used. The results showed that the corrected 3IMERGM (C-3IMERGM) performed much better than 3IMERGM. Specifically, CC was increased from 0.68 to 0.73, and RB is decreased from 7.76 to ? 1.65%. Furthermore, C-3IMERGM achieves a better precipitation distribution than the uncorrected satellite product and even than scarce gauge measurements. The actual spatial pattern of precipitation represented that the precipitation bands in the Kunlun Mountains located in southern Xinjiang were captured by C-3IMERGM, but missed by the other products.