Forecasting Tornadoes by Combining Climatological Frequencies with the NSSL-WRF Ensemble

Using radar-identified convective modes, peak low-level rotational velocities, and near storm environmental data from the Storm Prediction Center’s RAP-based, surface objective analyses for a 5-year period, SPC forecasters have computed the climatological frequencies that severe-weather-producing, right-moving supercells produce tornadoes given specified levels of near-storm Significant Tornado Parameter (STP) values. These climatological frequencies are used to generate probabilistic tornado guidance from the 10-member, 4km NSSL-WRF ensemble using forecast updraft helicity (UH) as a proxy for severe-weather-producing supercells, and forecast near-storm STP. Creation of the probabilities involves (1) identifying right-moving supercells in the forecast using UH, (2) assigning the supercells a value of STP based on the near-storm environment from the previous forecast hour, (3) assigning a tornado probability based on the climatological frequencies, and then (4) averaging these tornado probabilities over all ensemble members and applying a Gaussian smoother.

Evaluation of these probabilities is done over the spring (April through June) of 2014 and 2015 and compared to the Day 1 tornado probabilities issued by the SPC at 0600 UTC and valid 1200 – 1200 UTC. The SPC tornado probabilities are issued at the same time the NSSL-WRF output becomes available and correspond to model forecast hours 12-36. Various smoothing levels of the ensemble probabilities are also tested, to determine if the smoothing of probabilities impacts the objective statistics of the forecasts as well as the subjective “look” of the forecasts. These probabilities are considered across all cases and on a day-by-day basis, to investigate situations where the forecasters added value to the model forecasts. Overall, when verified using local storm reports, the model-generated forecasts underforecast the magnitude of the tornado probabilities. However, all smoothing levels demonstrate cumulative ROC areas of greater than .7, despite remaining lower than the cumulative SPC tornado forecasts. Generally SPC forecasters had a higher ROC area than the model forecasts on a day-to- day basis, although this was not the case every day. On some days, smoothing had a very large effect on the ROC area, typically through the change in the POD.