J5.2 A Neural Network for Detecting and Diagnosing Tornadic Circulations using the Mesocyclone Detection and Near Storm Environment Algorithms

A Mesocyclone Detection Algorithm (MDA) and a near-storm environment (NSE) algorithm have been developed at the National Severe Storms Laboratory. The MDA analyzes azimuthal shear in Doppler velocity data in 3 dimensions to identify storm-scale circulations. Sometimes, though not always, these circulations are precursors to tornadoes. Marzban and Stumpf (1996) developed a neural network based on the MDA parameters to identify which of the circulations would be tornadic using a small set of data cases. That work was extended to cover 43 storm days in Marzban (2000) using a more robust methodology. In this paper, we further extend the work to use 83 storm days and introduce some variations that improve neural network performance over that achieved by Stumpf and Marzban (2002). We also evaluate whether the incorporation of near-storm evironment (NSE) data from those days can improve the predictive capability of the neural network. On an independent test set of 27 storm days, we achieve a Heidke Skill Score (HSS) of 0.41 using just the MDA parameters and a HSS of 0.45 using a combination of MDA and NSE parameters. The Critical Success Index (CSI) for the MDA-only neural network is 0.29, while the CSI for the neural network with both MDA and NSE parameters is 0.32.