Using Machine Learning to Determine Convective Storm Modes and Warning Verification for Simultaneous and Collocated Tornado and Flash Flood Warnings

Independently, tornadoes and flash floods are a challenge for meteorologists to forecast and alert the public. When occurring simultaneously, the challenge to the forecaster and the risk to the public is greatly increased. This is because when a tornado and flash flood threaten the same location at the same time, members of the public are presented with a dangerous dilemma: to shelter from one of these hazards requires exposing oneself to the other. Another challenging aspect of severe weather forecasting is that of convective mode, which can generally be described as the form in which thunderstorms take shape and their subsequent appearance on weather radars. Convective mode affects the accuracy of tornado warnings, which in turn affects public preparedness and ability to take shelter. This study lies at the intersection of these two challenges: convective modes of storms producing tornadoes and floods. Here, we assess the utility of supervised classification algorithms in determining convective storm modes and warning verification of storms that produce simultaneous and collocated tornado and flash flood (TORFF) warnings by training the algorithms on ERA-5 reanalysis data on sounding-derived measurements and parameters.