In this study, we used the cumulative distribution function transform to conduct a bias correction for simulations from different regional climate models (RCMs) driven by one global climate model (HadGEM2-ES). We divided the historical period into two time-frames, i.e., the calibration period and the validation period. These two periods are 1986–1998 and 1999–2011, respectively. We then choose the period from 1986 to 2005 as the calibration period. The data for the future 2006–2098 were revised and used to explore future climate change under the RCP8.5 scenario. The difference before and after bias correction were compared. The results show that the cumulative distribution function transform method can improve the simulation accuracy of RCM in terms of the average precipitation and seasonal precipitation can improve in north arid regions. For extreme precipitation and different rainfall levels, the root mean squared errors of most indexes are reduced by about 60–80% in China, and the correlation coefficients are close to 1. For future precipitation, the bias correction method could reduce the overestimation of RCM simulations, but cannot change trends of precipitation variation. Compared with the simulations before bias correction, the predicted future precipitation indicates some differences in different regions. After correction, the spread of the precipitation and the most extreme precipitation indexes was smaller than those before correction. The predicted future daily precipitation intensity was also smaller. The reduction of drought days in the arid areas is more than before the correction, and the increase days of R50 in the southern regions is larger than before the correction.

Based on the outputs of the global climate models (GCMs) HadGEM2-ES, NorESM1-M and MPI-ESM-LR from Coupled Model Intercomparison Project Phase 5 (CMIP5) and the downscaling results with the regional climate model (RCM) REMO, the ability of the climate models to reproduce the extreme precipitation in China during the current period (1986–2005) is evaluated. Then, the future extreme precipitation in the mid (2036–2065) and the late 21st century (2066–2095) is projected under the RCP8.5 scenario. The results show that the RCM simulations have great improvements compared with the GCMs, and the ensemble mean of the RCM results (ensR) outperforms each single RCM simulation. The annual precipitation of the RCM simulations is more consistent with the observation than that of the GCMs, with the overestimation of the peak precipitation reduced, and the ensR further reduces the bias. For the extreme precipitation, the RCM simulations significantly decrease the underestimation of intensity in the GCMs. The RCM simulations and the ensR can greatly improve the simulations of Rx5day and CWD compared with the GCMs, decreasing the wet bias in North China and Northwest China. In the future, the consecutive dry days (CDD) will decrease in the northern arid regions, especially in North China and Northeast China. However, the southern regions will experience longer dry period. Both the amount and the intensity of precipitation will increase in various regions of China. The number of wet days will decrease in the south and increase in the north area. The significantly greater Rx5day and R95t indicate more intensive extreme precipitation in the future, and the intensity in the late 21st century will be stronger than that in the middle. Attribution analysis indicates that the extreme precipitation indices especially the R95t have significant positive temporal and spatial correlations with the water vapor flux.