Vine Copula Ensemble Downscaling for Precipitation Projection Over the Loess Plateau Based on High‐Resolution Multi‐RCM Outputs

Abstract A vine copula‐based ensemble downscaling (VCED) framework is proposed to jointly downscale the projected precipitation from multiple regional climate models (RCMs). This approach can effectively reduce the biases inherent to precipitation projections from different RCMs and thus provide more reliable ensemble projections. The proposed approach was applied to RCM projections over the Loess Plateau of China, which features complex topography and various climatic zones. Precipitation projections from 7 RCMs were used, and 21 sets of downscaling results were obtained. The performance of the VCED in reproducing historical precipitation across the Loess Plateau was evaluated using mean absolute error (MAE), the Taylor diagram, and the rank histogram (RH). The proposed VCED approach was found to be more effective than quantile mapping and bivariate copula methods in achieving robust precipitation projections. Overall flat RH diagrams indicate that the ensemble prediction and observations have strong consistency in distribution. Future precipitation changes of two 30‐year periods (i.e., the 2050s and 2080s) under two Representative Concentration Pathway (RCP) scenarios (RCP 4.5 and RCP 8.5) over the Loess Plateau were then analyzed after postdownscaling processes. The results show that the average annual precipitation over the Loess Plateau may increase by 8.4%–11.4% under the RCP 4.5 scenario and by 9.3%–17.5% under RCP 8.5. The projected precipitation in the south‐central parts of the Loess Plateau would be significantly reduced whereas those of the other parts be significantly increased. Key Points A vine copula ensemble downscaling method is proposed to jointly downscale the projected precipitation from multiple climate models Postsimulation analyses of monthly precipitation projections over the Loess Plateau were conducted using seven regional climate models This method can generate more ensemble series than conventional bias‐correction methods, and thus more robust projections