A Bernoulli-Gamma hierarchical Bayesian model for daily rainfall forecasts

We consider stochastic weather models originally developed for rainfall simulations to build a hierarchical Bayesian mixture model daily forecasts using endogenous and external information. as seasonal-varying of Bernoulli distribution occurrence gamma the amount. The scheme allows inclusion predictors reduce bias variance forecasts, while framework promotes better understanding reduction in parameter uncertainties, especially gauges with short records, well supports estimation regional parameters that could be employed at ungauged sites. was tested 47 years (1973–2019) data from 60 South Korea. Climate indices derived low-level wind over region were analyzed Principal Component Analysis (PCA) embodied into enhance its forecast skills. structure based on detailed exploratory analysis, which included application Self-Organizing Maps (SOM) examine spatio-temporal patterns rainfall. Cross-validated results reveal improved skills reference climatology persistence up three days lead time. average gains metrics such Brier Winkler skill scores vary 5% 50%, correlation between predictions observations reach values 0.55. beyond time are marginal, but underlying proposed still encourages use models, being step forward improving real-time region. It has also great potential combined applied other places across world.