Incorporating Time of UH Occurrence in the Generation of Ensemble-Derived Tornado Probabilities

Generation of convection-allowing model (CAM)-based tornado probabilities using updraft helicity (UH) as a proxy for tornado occurrence often results in over-forecasting when simulated non-supercell convective modes produce high values of UH. Simulated nighttime mesoscale convective systems (MCSs) are particularly problematic, because they are often associated with high values of UH, but produce tornadoes far less frequently than right-moving supercells do. Thus, a method of computing model-derived tornado probabilities from UH was developed that accounts for the time of day that the UH occurs. Incorporating the time of UH occurrence adds skill to the probabilistic forecasts by reducing the response of probabilities to the nocturnal UH swaths, reflecting that nocturnal tornadoes from right-moving supercells are far less common than their daytime counterparts. Aggregated seasonal statistical metrics of forecast skill from implementing normalized probabilities into the NSSL-WRF ensemble show large ROC area decreases. However, reliability remains high and false alarm area is greatly reduced. The false alarm reduction is specifically focused in areas of nocturnal MCSs that produce UH swaths without producing tornadoes, better highlighting the area threatened by tornadoes. Additionally, when these forecasts are verified against right-moving supercellular tornadoes, they can achieve a higher POD and lower POFD than the SPC’s 0600 UTC tornado forecasts, making them a useful first-guess tornado forecast.