Inter-product biases in extreme precipitation duration and frequency across China

Accurately delineating the duration and frequency characteristics of extreme precipitation is vital for assessing climate change risks. This study reassesses the spatiotemporal variations in the frequency and persistence indices of extreme precipitation from 2001 to 2019 across thirteen widely used precipitation datasets. We quantify the inter-product biases using common accuracy indices based on reference data, derived from a national observation network of over 2400 stations. Regarding the duration of extreme precipitation, represented by consecutive dry days (CDD) and consecutive wet days (CWD), gauge-based datasets generally demonstrate better accuracy. Satellite retrieval datasets tend to overestimate CDD (4.58%) and CWD (60.50%) at continental scale. Meanwhile, reanalysis and fusion datasets tend to underestimate CDD (-30.27% and -15.39%, respectively) and overestimate CWD (148.44% and 93.41%, respectively). In terms of frequency indices, represented by the number of heavy precipitation days (R10MM) and the number of very heavy precipitation days (R20MM), gauge-based, satellite retrieval, and fusion datasets show weak biases in R10MM (all below 3.5%), while reanalysis datasets indicate substantial overestimation (33.62%). In the case of R20MM, there is an improvement in the performance of reanalysis datasets, while the performance of other datasets declines. However, almost all datasets fail to consistently capture variations in the Tibetan Plateau and Xinjiang regions, where gauge stations are limited and terrain is complex. Furthermore, multiple datasets present significant discrepancies in temporal trends from 2001 to 2019. Remote sensing datasets tend to overestimate CDD, while reanalysis datasets generally show persistent underestimation of CDD and persistent overestimation of other indices. This research contributes to guiding the application and improvement of global precipitation datasets in extreme precipitation studies.