PROBABILISTIC EXTREME TEMPERATURE FORECASTS USING THE BAYESIAN PROCESSOR OF ENSEMBLE OVER TAIWAN

A statistical post-processing (SPP) system called Bayesian Processor of Ensemble (BPE) is demonstrated in this study for the generation of extended-range probabilistic extreme temperature forecasts at selected weather stations in Taiwan. BPE is based on the Bayes’ Theorem, and comprises three main components: (1) the estimation of the prior, the climatic distribution of the predictand; (2) the generation of the likelihood distribution, capturing the relationship between the predictors and predictand, and (3) the fusion of the prior and likelihood distributions for the generation of the predictive (or posterior) distribution, given the latest operational ensemble forecasts. The Bayesian use of the prior distribution allows BPE to optimally calibrate, with a maximum level of informativeness, the predictive distribution, even under operational constraints such as limitations in the size of the reforecast data sample, and low skill in raw extended-range ensemble forecasts.