Communicating probability with real-time calibrated forecasts

Ensemble forecasts of surface parameters are often under-dispersed. However, statistical post-processing such as Bayesian model averaging (BMA) [Raftery et al, 2005, Mon Wea Rev] can provide calibrated probabilistic forecasts. Our group aims to develop new and better methods of communicating this information.

The UW BMA page (http://bma.apl.washington.edu) demonstrates a framework for visualizing real-time probabilistic forecasts, using the University of Washington MM5 ensemble as a testbed. For spot and min/max temperatures as well as precipitation, this provides a choice of several forecast maps derived using BMA output: the BMA forecast (a weighted average of the ensemble forecasts), high and low likely limits of the outcome, uncertainty, and probability the outcome will be less than or greater than a threshold (e.g. probability of freezing or probability of precipitation). A forecast consumer can view the forecast distribution at any grid point as a boxplot or a probability distribution by clicking a location on the map. Real-time verification information is available. This product calibrates and summarizes the forecast ensemble, using fewer and more transparent displays than most ensemble-based systems, while losing very little information.

This application represents a collaboration between meteorologists, statisticians, psychologists and engineers, funded by the DOD MURI Program. It builds on cognitive and ethnographic studies of the forecasting process. This work is in collaboration with Eric Grimit, David Jones, Susan Joslyn, Clifford Mass, Tilmann Gneiting, Adrian Raftery and McLean Sloughter.