Determining the Likelihood of Severe Weather Based on Model Output

Because of their small scale and short lifecycles, severe convective storms feature low predictabilities and require a probabilistic approach to forecasting. Convective outlooks from the SPC (Storm Prediction Center) attempt to quantify the risk of severe weather for particular locations (all regions of the country) based on an educated assessment of current and model forecasted atmospheric conditions. The completion of the North American Regional Reanalysis provides a gridded data set of North American weather at 3 hour and 32 km horizontal resolution for the past 27 years. Together with the digitized Storm Data records of observed severe weather covering the period of NARR, the opportunity exists to build a definitive statistical relationship between gridded mesoscale analysis of weather and the occurrence of severe weather. The goal of this study is to build an objective quantification of the likelihood of tornadoes, damaging winds, and large hail, based on the gridded NARR which could then be applied to daily operational model output to objectively assess the likelihood of severe weather in real time. Objective quantifications already available include the supercell index (Wilt, 1994), which quantifies the likelihood of a supercell thunderstorm, and the hail parameterizations of Marglass and LaPenta (1996), which is regionally specific to upstate New York. This study will objectively quantify the likelihood of all 3 types of severe weather separately, based on state variables and derived parameters calculable from the gridded NARR output and operational mesoscale model forecasts. The statistical evaluation process will employ a Bayesian framework to assess the conditional likelihood of severe weather based on the derived variables. The NARR and SPC's storm data for the years 1979-2003 will be used to create a twenty-five year climatology of atmospheric conditions associated with severe weather. This climatology will be used to create probability distributions of severe weather as functions of selected parameters shown to be strong predictors of severe weather likelihood. For instance, Rasmussen (2000) suggested that EHI (Energy Helicity Index), is a strong indicator of severe weather and tornado potential. Weisman and Klemp (1984) suggested that the Convective Richardson number was also a good indicator. On the other hand, SPC forecasters have long known that the cloud base height above ground also is an indicator of tornado potential. The best performing independent (or uncorrelated) predictor variables for each of the three types of severe weather will be selected. Using the 25 years of NARR, this study will statistically evaluate up to 50 potential predictors, calculated from gridded output, for their skill in assessing the potential for each severe weather type. Skill is defined as high Probability of Detection (POD) and a small False Alarm Rate (FAR). Baysean analysis will then be used to combine 5-7 of the best predictors to form composite “super predictors” designed to maximize the overall Critical Success Index (CSI) for predicting each storm type. The result will be a set of indices each providing a quantitative probability estimate for each type of severe weather event based on how that index performed from similar gridded model output over the past 25 years on all events. The accuracy of this procedure will be tested using NARR data and SPC storm responds for the years 2004 and 2005.