Ensemble Postprocessing for Weather Trajectories

Over the past decade, powerful methods for the statistical postprocessing of ensemble forecasts have been developed. These methods correct for model biases and dispersion errors, and generate calibrated and sharp predictive distributions from ensemble output. However, they typically apply to a single weather variable at a single location and a single prediction horizon only, and thus ignore dependencies. In contrast, applications such as air traffic control or flood management require physically consistent postprocessed ensemble forecasts of temporal, spatial or spatio-temporal weather trajectories, whose dependence structures are to be captured by the predictive distributions. I will discuss and illustrate techniques for doing this.