B0213
Title: Simulation-based comparison of multivariate postprocessing methods for ensemble weather forecasts
Authors: Roman Schefzik - Zentralinstitut fuer Seelische Gesundheit (Germany) [presenting]
Abstract: Contemporary weather forecasts usually rely on ensembles, resting on different runs of numerical weather prediction models and accounting for major sources of uncertainty. Typically, ensemble forecasts require statistical postprocessing, and particularly, accurate modelling of spatial, temporal and inter-variable dependencies is crucial in many practical applications. State-of-the-art multivariate ensemble postprocessing methods developed to address this need are reviewed. The focus is on generally applicable two-step approaches in which ensemble predictions are first postprocessed for each location, look-ahead time and weather variable separately and multivariate dependencies are then restored using copula functions. Specifically, a Gaussian copula approach (GCA) is considered, as the Schaake shuffle and variants of ensemble copula coupling (ECC). These methods are compared using simulation studies tailored to mimic challenges occurring in practical applications. Particularly, the comparisons allow ready interpretation of the effects of different types of misspecifications in the mean, variance and covariance structure of the ensemble forecasts on the performance of the postprocessing methods. Overall, the Schaake shuffle provides a compelling benchmark, whereas the performances of GCA and the ECC variants strongly depend on the misspecifications at hand.
