Statistical postprocessing for ensembles of numerical weather prediction models

The past fifteen years have witnessed a radical change in the practice of weather forecasting, in that ensemble prediction systems have been implemented operationally. An ensemble forecast comprises multiple runs of numerical weather prediction models, which differ in initial and lateral boundary conditions, and/or the parameterized representation of physical processes. However, ensemble forecasts are subject to biases and dispersion errors, and thus statistical postprocessing is required, with Bayesian model averaging and ensemble model output statistics being state of the art approaches. Future work is called for to ensure that the postprocessed forecast fields show physically realistic and coherent joint dependence structures across meteorological variables, geographic space and look-ahead times.