关键词:
Chinese mainland
FY-2
GPM
random errors
satellite precipitation estimates
systematic errors
摘要:
Characterizing error components (including systematic and random components) is essential to improve precipitation retrieval algorithms and develop bias adjustment techniques. Benchmarked by the Chinese merged precipitation analysis dataset derived based on automatic weather stations, the errors of satellite precipitation estimates from Fengyun 2 (FY-2F and FY-2G) and the mainstream Global Precipitation Measurement (GPM;IMERG-Final and GSMaP-Gauge) are decomposed into systematic and random components at the hourly scale. The comprehensive performance of the systematic and random errors of the four products are revealed from the perspectives of spatial distribution, temporal pattern, rainfall intensity distribution, and elevation distribution. (1) The systematic errors of FY-2G and FY-2H over Mainland China are close to those of IMERG-Final and GSMaP-Gauge, among which IMERG-Final has the lowest systematic errors. FY-2G and FY-2H adopt only infrared observations as data sources in their precipitation retrieval algorithms, whereas IMERG-Final and GSMaP-Gauge adopt not only infrared data but also high-accuracy microwave data as the data sources. However, the systematic errors of FY-2G and FY-2H reach the level of GPM precipitation products because of the relatively dense gauge networks in the satellite–gauge merging procedure of FY-2G and FY-2H. (2) GSMaP-Gauge has the lowest random errors over Mainland China among the four satellite precipitation estimates. The gap between FY-2G and FY-2H in relation to GPM precipitation products is mainly manifested in the random errors. Particularly, the random errors of IMERG-Final and GSMaP-Gauge over the eastern monsoon and northern arid regions are much lower than those of FY-2G and FY-2H in summer. (3) IMERG-Final has much lower systematic errors at different elevations compared with the three other precipitation products, and GSMaP-Gauge, FY-2G, and FY-2H have close systematic errors at different elevations. The random errors