Automated 12–36 Hour Probability Forecasts of Thunderstorms and Severe Local Storms

Abstract

Probability equations were developed for 12?36 h forecasts of thunderstorms, severe local storms, and major or family tornado outbreaks. The equations were derived by applying screening regression techniques to relate manually digitized radar (MDR) data and severe storm reports to large-scale meteorological predictors obtained from numerical forecast models. The probability forecasts are prepared once daily on the NOAA computer system and are transmitted over facsimile and teletypewriter circuits. One of the more important innovations in the forecast equations was the development of an interactive predictor which takes into account the seasonal variations in thunderstorm occurrence, yet is responsive to the daily synoptic situation. This predictor is formed by combining the K stability index with daily mean thunderstorm relative frequencies estimated from MDR data. Local variations in thunderstorm occurrence over the MDR grid were also included in the generalized equation by incorporating probability estimates for each MDR grid block, based on statistical relationships between the interactive predictor and thunderstorm occurrence. Verification of the thunderstorm probability forecasts against independent MDR data and observed thunderstorms during the 1977?78 convective seasons have shown the probability forecasts to be highly reliable. During the spring months, severe local storm probabilities near 40% were forecast with good reliability, even though the climatological probability for the same period was only 7%. Verification scores show that categorical forecasts based on the conditional severe local storm probabilities compare favorably with operational convective outlooks.