Evaluation of High-Resolution Gridded Precipitation Data in Arid and Semiarid Regions: Heihe River Basin, Northwest China

The reliability of three satellite-derived precipitation products, Tropical Rainfall Measuring Mission (TRMM) 3B42 V7 and the Climate Prediction Center morphing technique (CMORPH) satellite-only (CMORPH-RAW) and gauge-corrected versions (CMORPH-CRT), and three gauge-based precipitation datasets, Asian Precipitation-Highly Resolved Observational Data Integration Toward Evaluation of Water Resources (APHRODITE), National Climate Center of China Meteorological Administration (CN05.1), and Institute of Tibetan Plateau Research, Chinese Academy of Sciences (ITPCAS), is evaluated via comparisons with rain gauge observations from stations over the Heihe River basin (HRB) for the period from 1998 to 2012. The results show that the observed climatology, interannual variability, the detection of precipitation events, and probability density functions (PDFs) are reasonably well represented by the high-resolution precipitation products (HRPPs), with APHRODITE presenting the best performance, CN05.1 and ITPCAS exhibiting similar performances, and CMORPH-CRT showing a poor performance. The bias-correction algorithms applied in CMORPH-CRT improve the accuracy of CMORPH-RAW slightly but fail to improve the rainfall detection skill. TRMM consistently outperforms CMORPH-CRT at various scales, whereas CMORPH-CRT is comparable to TRMM in summer. The spatial correlations, normalized root-mean-square error (NRMSE), and probability of detection (POD) show that all datasets perform better in summer than in winter. Except for CMORPH-RAW, the HRPPs could adequately reproduce the unimodal characteristics of annual cycle, although they overestimate the magnitude of the warm season precipitation. The HRPPs could capture the overall spatial distribution and decadal trend of extreme precipitation indices. However, the satellite-derived products overestimate the wet day precipitation and underestimate the consecutive dry days, although the TRMM generates relatively better results.