Application of a self-organizing map statistical technique to a RUC supercell proximity sounding database

We apply the self-organizing map (SOM) statistical technique to a supercell proximity sounding data set derived from RUC model analyses with the goal of better distinguishing and predicting supercell-forming and tornadic environments. SOMs employ a learning algorithm to display significant features of a data set onto a map consisting of a specified number of clusters. This technique is applied to profiles of thermodynamic and kinematic parameters that are considered relevant in discriminating tornadic from nontornadic environments as well as environments that are conducive to supercell formation from environments that are not, such as equivalent potential temperature, relative humidity, storm-relative winds, wind shear, crosswise and streamwise vorticity, and Richardson number. We discuss the ability of SOMs computed from each parameter to cluster soundings in a way that is useful in differentiating storm type (e.g. nonsupercell, nontornadic supercell, weakly tornadic supercell, and significantly tornadic supercell), location, and time of year. We also explore the impact of altering the depth of the profiles and the number of SOM nodes on the results.