Statistical Post-processing of Ensemble Forecasts: Recent Developments and Current Issues (Invited Presentation)

Statistical post-processing uses forecast-observation pairs from the past to identify model errors and adjust the current forecast accordingly. While traditional methods were applied to deterministic NWP forecast to remove biases, the advent of ensemble prediction systems sparked an additional interest in the correct representation of forecast uncertainty which is, together with the absence of forecast bias, a prerequisite for reliable probabilistic predictions.

This talk will review some recent developments in statistical post-processing of ensembles and point out some active areas of research. Specifically, an overview will be given over parametric and non-parametric approaches for the statistical calibration of univariate quantities, and current efforts and challenges with the prediction of multivariate quantities (e.g. different weather variables considered simultaneously, or spatio-temporal trajectories of a single weather variable) will be pointed out. Another challenge that will be addressed is the issue of limited training sample size, which becomes especially critical when the interest is in rare and extreme events. All of these points will be illustrated with both synthetic data and actual case studies with GEFS and ECMWF ensemble forecasts.