Investigation on Extreme Rainfall Detection Capability of Various Satellite Precipitation in a Typical Arid Region of China

The frequency of extreme climatic events is increasing under the background of climate change, which leads to a significant impact on production and livelihood. Accurate assessment of extreme rainfall is necessary for agricultural production management and efficient utilization of water resources, especially in arid areas. In this study, the detection capability of extreme precipitation characteristics of three satellite precipitation products (TRMM 3B42, PERSIANN and CHIRPS) were evaluated at time scale and grid scale with 23 observed rainfall stations data in a typical arid region of China. The results showed that the correlation coefficient (R) of the three satellites are very poor (R<0.3) at daily scale; the R of TRMM 3B42 is the best (R=0.8) and the error is minimal [relative root mean square error (RRMSE)=1.0; mean absolute error (MAE)=7.65 mm] at monthly scale; on the annual scale, the CHIRPS has the highest R and the smallest error (R=0.84, RRMSE=0.53, MAE=55.88mm). TRMM 3B42 and PERSIANN show a certain degree of underestimation of daily, monthly and annual precipitation. CHIRPS only underestimates precipitation on the daily time scale. At the grid cell scale, three satellites all have good consistency with measured precipitation, and PERSIANN performances best. With the increase of rain thresholds, the equitable threat score (ETS) and probability of false detection (POD) of the three satellites are decreasing and the false alarm ratio (FAR) are increasing, which indicates the three satellite products have limitations on the ability to accurately predict heavy precipitation. The PERSIANN has the largest POD and ETS at small rainfall from 0.2~0.5 mm/day. When the rainfall is greater than 2 mm/day, the POD and ETS of CHIRPS is maximal and that of PERSIANN is minimum. The FAR of PERSIANN drops to 0 when the rainfall is greater than 15 mm/d.