UW Aquatic & Fishery Sciences Quantitative Seminar
Statistics and Sociology, UW
Probabilistic Weather Forecasting Using Ensemble Bayesian Model Averaging
Probabilistic weather forecasting consists of finding a joint probability distribution for future weather quantities or events. Information about the uncertainty of weather forecasts can be important for decision-makers as well as the public, but currently is routinely provided only for the probability of precipitation, and not for other weather quantities such as temperature, wind or amount of precipitation. It is typically done using a numerical weather prediction model, perturbing the inputs to the model (initial conditions, physics parameters) in various ways, and running the model for each perturbed set of inputs. The result is viewed as an ensemble of forecasts, taken to be a sample from the joint probability distribution of the future weather quantities of interest. This is often uncalibrated, however.
We review a principled statistical method for postprocessing ensembles based on Bayesian Model Averaging (BMA), that models the predictive distribution conditionally on the ensemble by a finite mixture model. We describe applications to precipitation, wind speeds, wind directions, visibility and winter road maintenance, a multivariate decision problem. For probabilistic forecasting of an entire weather field, we describe a spatial extension of the BMA method that perturbs the outputs from the numerical weather prediction model rather than the inputs. Forecasts are available in real time at www.probcast.washington.edu, and the R packages ensembleBMA and ProbForecastGOP are available to implement the methods.
This is joint work with Tilmann Gneiting, Veronica Berrocal, McLean Sloughter, Le Bao, Chris Fraley, William Kleiber and Richard Chmielecki.