B0513
Title: Comparison of multivariate post-processing methods using global ECMWF ensemble forecasts
Authors: Maria Nagy-Lakatos - University of Debrecen (Hungary) [presenting]
Sandor Baran - University of Debrecen (Hungary)
Abstract: Ensemble weather forecasts are obtained to quantify uncertainties about future atmospheric behavior, and due to the way of their generation, also capture spatiotemporal and/or inter-variable dependencies. Univariate statistical post-processing is an often applied tool to address the systematic errors of the NWP systems; however, such a form of calibration can result in the loss of correlation dependencies across marginals. These correlation structures can be reinstalled with the application of multivariate post-processing. In recent years many multivariate post-processing approaches have been developed, and a comprehensive comparison was given on simulated ensemble predictions. The aim is to extend that work and apply the aforementioned methods to real datasets, namely the global temperature, wind speed, and precipitation accumulation forecasts of the European Centre of the Medium-Range Weather Forecasts, in order to create temporally consistent forecast trajectories. The focus is on copula-based two-step approaches, and the findings indicate that there are generally minor differences in the predictive performance of the various multivariate post-processing methods, and the skill seems to be superior compared to the univariately post-processed benchmark models.
