Postprocessing Ensemble Weather Forecasts for Introducing Multisite and Multivariable Correlations Using Rank Shuffle and Copula Theory

Abstract Statistical methods have been widely used to postprocess ensemble weather forecasts for hydrological predictions. However, most of the statistical postprocessing apply a single variable at location, thus neglecting intersite and intervariable dependence structures forecast variables. This study synthesized multisite multivariate (MSMV) framework that extends univariate method MSMV version by directly rearranging postprocessed members (post-reordering strategy) or latent variables in (pre-reordering strategy). Based on generator-based (GPP) method, two reordering strategies three reconstruction [rank shuffle (RS), Gaussian copula (GC), empirical (EC)] totaling six (RS-Pre, GC-Pre, EC-Pre, RS-Post, GC-Post, EC-Post) were evaluated precipitation temperature Xiangjiang basin China using 11-member from Global Ensemble Forecasting System (GEFS). The results showed raw GEFS tend be biased both ensembles dependencies. can improve performance mean spread but misrepresent among well utilize advantages also reconstruct Among methods, RS-Pre, EC-Post perform better than others with respect reproducing statistics multivariable dependences. post-reordering strategy is recommended combine (i.e., GPP) methods.