A Bayesian hierarchical approach to ensemble weather forecasting

In meteorology, the traditional approach to forecasting employs deterministic models mimicking atmospheric dynamics. Forecast uncertainty due to partial knowledge of the initial conditions is tackled by ensemble predictions systems. Probabilistic forecasting is a relatively new approach which may properly account for all sources of uncertainty. We propose a hierarchical Bayesian model which develops this idea and makes it possible to deal with ensemble predictions systems with non-identifiable members by using a suitable definition of the second level of the model. An application to Italian small-scale temperature data is shown.