A Bayesian neural network for severe-hail size prediction

The National Severe Storms Laboratory has developed algorithms that compute a number of Doppler radar and environmental attributes known to be relevan t for the detection/prediction of severe hail. Based on these attributes, t wo neural networks have been developed for the estimation of severe-hail si ze: one for predicting the severe-hail size in a physical dimension, and an other for assigning a probability of belonging to one of three hail size cl asses. Performance is assessed in terms of multidimensional (i.e., nonscala r) measures. It is shown that the network designed to predict severe-hail s ize outperforms the existing method for predicting severe-hail size. Althou gh the network designed for classifying severe-hail size produces highly re liable and discriminatory probabilities for two of the three hail-size clas ses (the smallest and the largest), forecasts of midsize hail, though highl y reliable, are mostly nondiscriminatory.