Multi-model ensemble bias-corrected precipitation dataset and its application in identification of drought-flood abrupt alternation in China

High precision precipitation products are the basis of precipitation-related research. Based on 27 global climate models (GCMs) in the Coupled Model Intercomparison Project phase 6 (CMIP6), we designed eight schemes for comprehensively using the empirical quantile mapping (EQM) method and data ensemble method to conduct precipitation bias correction; then, we selected the scheme with the highest accuracy as the final bias correction scheme. Using the selected bias correction scheme, we created a monthly precipitation dataset with a 1° spatial resolution, which spans the historical period of 1961–2014 and the future period of 2015–2099 under three shared socioeconomic pathway (SSP) scenarios: SSP126, SSP245, and SSP585. The corrected precipitation data were validated using the CN05.1 grid precipitation dataset from the China Meteorological Data Sharing Network and were compared with the ERA5 precipitation data from the European Centre for Medium-Range Weather Forecasts. The dataset was also utilized for future prediction of alternating drought and flood events in China. The results show that this best bias correction scheme is the first to integrate precipitation simulation data from 27 GCMs using the random forest (RF) model and then the EQM method to further correct the integrated precipitation data. The corrected precipitation data are better than the original GCM precipitation data in terms of both the monthly precipitation and extreme precipitation. From the perspective of the monthly precipitation, the difference between the ERA5 and RF-EQM is small, but the extreme precipitation of the RF-EQM clearly outperforms the ERA5 extreme precipitation. For the annual maximum (minimum) monthly precipitation, the correlation coefficient, the RMSD (standardized), and the STD (standardized) between the ERA5 and CN05.1 are 0.925 (0.743), 0.474 (1.223), and 1.207 (1.765), respectively; the correlation coefficient, the RMSD (standardized), and the STD (standardized) between the RF-EQM and CN05.1 are 0.947 (0.735), 0.337 (0.837), and 0.849 (1.226), respectively. The occurrence frequency of DF (an abrupt change from drought to flood) events is continuously increasing in all scenarios, with the highest frequency observed under the SSP585 scenario. The increase in FD (an abrupt change from flood to drought) event frequency is not pronounced. This study expands the method for bias correction of meteorological data and provides a reference for other climate parameters and precipitation bias correction in other regions.