Comparison And Application Of Different Bias Correction Methods In Regional Climate Simulation Over China

The global and regional climate models are the primary tools for the simulation and prediction of climate change.Due to the complexity of the climate system and the level of present day science development,climate model simulations are usually biased compared to the observations,including the mean,probability density distribution,etc.Directly use of the climate model results to drive the impact,e.g.the hydrological or agricultural models,may lead to problems in their simulations.Thus bias correction of the climate model results are usually needed and sometimes considered as necessity.In the present study,we try to bias correct the simulated daily precipitation,mean temperature,maximum and minimum temperature over China as simulated by a regional climate model(RegCM4)driven by ERA-interim re-analysis,based on the probability distribution(Quantile-Mapping).The year is devided to four seasons in the study,as winter(December-January-Febuary,DJF),Spring(March-April-May,MAM),summer(June-July-August,JJA),and autumn(September-October-November,SON).Firstly,the daily precipitation is used to test different methods of the bias correction.Transfer functions(TF)are established from the reference period of 1991 to 2000.6 different TFs in total using parametric or nonparametric transformations are set up and compared against observations to validate their performances.Results show that while all the TFs effectively reduce the biases of the simulated precipitation,a better performance of the RQUANT method using nonparametric transformations is found.Thus the RQUANT is selected as the method to bias correct precipitation and temperature as simulated by RegCM4.Analysis of the RQUANT method in bias correcting the daily precipitation show that it greatly reduces the bias of the simulation.Diffrences of the bias corrected precipitation with observation is reduced to within-25% to 25% only in most China.With much larger spatial correlation coefficients and less SIDE,the precipitation agrees more with the observation both in the spatial distribution and the amount.The interannual variability is also found to be improved.Due to the better simulation of temperature compared to precipitation,the results of bias corrected temperature is much closer to the obversation,with the spatial correlation coefficient greater than 0.99.Improvements are found for all of the mean,maximum and minimum temperatures.The differences between corrected temperature and obversation are usually within±1°C,with larger spatial correlation coefficients and reduced SIDE.Correction of the interannual variability is not as good as the mean temperature,but large improvemsnt can be still found,e.g.the spatial correlation coefficient in MAM,as well as the less SIDE.The RQUANT method also improves significantly reduce the bias of the simulated extreme events as measured by different indices,such as consequtive dry days(CDD),simple daily intensity index(SDII),the maximum of daily maximum temperature(TXx),and the the minimum of daily minimum temperature(TNn).Larger values of the spatial correlation coefficients and less values of SIDE are found for the indices,indicating their better agreements to obversation.