On the Convective Environments, Modes, and Warning Verifications of Tornado- and Flash Flood-Warned Storms in the Southeast United States

Abstract

Thunderstorms can produce hazards to society such as tornadoes and flash floods, occasionally at the same time. These storms can be categorized by their convective mode, largely through their appearance on radar. Convective mode is an important factor in how forecasters analyze these threats and warn the public when necessary. This study uses a random forest classification technique to categorize two sets of storms in the Southeastern United States: one comprised of storms warned for tornadoes and flash floods at the same time, and the other warned for tornadoes without necessarily having a concurrent flash flood warning. The goal of these classifications was to use information about each storm's meteorological environment to identify (1) its mode and (2) whether the hazard warning(s) issued by the National Weather Service verified, or whether the warning was a “false alarm.” The models predicting mode generally exhibited more skill and identified differences between discrete modes and linear modes, particularly in upper-level humidity, lapse rates, and low-level wind speeds. The models predicting whether the warnings verified exhibited less skill, but indicated that environments favorable for tornadoes were characterized by stronger wind speeds, lower upper-level moisture, and higher supercell composite parameter, while environments favorable for flash floods were characterized by greater moisture, lower wind speeds, and slower storm motion. These results are of note to researchers and forecasters seeking to better anticipate hazards, identify hazards, increase warning accuracy, and minimize false alarms as the implementation of artificial intelligence into the forecasting process continues.