Preliminary Evaluation of a Fused Algorithm for the Prediction of Severe Storms

Extreme weather such as severe thunderstorms and their associated hazards of large hail, strong winds, and tornadoes occur on small spatial scales and short time scales. Such phenomena are often difficult to forecast due to their rapid evolution and complicated interactions with environmental features that may both enhance and diminish storm intensity, and which may be challenging to observe. Thus, this paper presents a probabilistic approach to the forecast of severe convection, fusing together data from several sources as input to a statistical model (naïve Bayesian classifier). These sources include derived products from Geostationary Operational Environmental Satellites (GOES), the Next Generation Weather Radar (NEXRAD) network, and the mesoscale Rapid Refresh (RAP) numerical weather prediction (NWP) model. Each observation source provides information during different periods of storm development (i.e., the pre-storm environment, storm initiation and growth, and hydrometeor intensification). The algorithm is meant to provide warning guidance to forecasters in the near-term (0-60 min), by monitoring trends in developing convection. This probabilistic model has been run in real-time at the University of Wisconsin Cooperative Institute of Meteorological Satellite Studies (UW-CIMSS) from April through August of 2013. A preliminary evaluation of the model's skill is presented, measured against local storm reports and National Weather Service (NWS) severe thunderstorm and tornado warnings.