Machine-learning-derived severe weather probabilities from a Warn-on-Forecast System

Probabilistic severe weather probabilities are derived from the Warn-on-Forecast System (WoFS) run by the National Severe Storms Laboratory (NSSL) during Spring 2018 using a machine learning approach known as the random forecast (RF). Recent work has shown that this method generates very skillful and reliable forecasts when applied to convection-allowing model ensembles for the “Day 1” time range (i.e., 12-36 h lead times), but it has yet to be tested for lead times relevant to WoFS (0-6 h). Thus, in this work, a set of 30 predictors from WoFS, which includes both environment and storm-based fields, are input into a RF algorithm and trained using the occurrence of severe weather reports within 6, 12, 24, and 39 km of a point to produce corresponding severe weather probabilities at 0-3 h lead times. We analyze the skill and reliability of these forecasts, sensitivity to different sets of predictors, and avenues for further improvements.