Tuesday 25 Feb 2014: Probabilistic Weather Forecasting: Recent Developments in Bayesian Model Averaging
Professor Adrian Raftery - University of Washington
Matrix Lecture Theatre, Business School Building: One 14:00-15:00
Probabilistic weather forecasting consists of finding a joint probability distribution for future weather quantities or events. This is often done using an ensemble of forecasts, obtained by perturbing the inputs to a numerical weather prediction model (initial conditions, physics parameters) in various ways, and running the model for each perturbed set of inputs. The spread of the ensemble is often related to the absolute forecast error, but ensemble forecasts tend to be uncalibrated, particularly for surface quantities. It is widely accepted that statistical postprocessing of ensembles improves the quality of the probabilistic forecast.
I will review Bayesian Model Averaging (BMA), a way of doing this that models the predictive distribution conditionally on the ensemble by a finite mixture model. I will outline some recent developments of BMA for weather forecasting, including to wind speed, wind direction, wind vectors, visibility, functionals of weather fields and multivariate weather quantities. I will also show how BMA can be applied to multimodel ensembles and ensembles in which subsets of members are exchangeable. Illustrative probabilistic forecasts are available in real time at www.probcast.washington.edu, and the R packages ensembleBMA and ProbForecastGOP are available to implement the methods.