Improved Probabilistic Predictions through Bayesian Post-Processing and Bias Correcton

The probabilistic predictions produced by a mesoscale ensemble system can be substantially improved by post-processing. At the University of Washington, a high-resolution mesoscale ensemble system has been run in real time since 2000 and today it encompasses 17 members produced by initialization and physics uncertainty. It has been found that bias-correction increases the reliability and sharpness of the forecast probabilities and that further improvements are possible using Bayesian Model Averaging BMA), where the contributions of each model is weighted by past performance. This talk will describe the results of a joint project between the UW Atmospheric Sciences and Statistics departments to perfect and apply advanced post-processing techniques for probabilistic prediction.