Understanding the Clustering of Proximity Sounding Profiles and Sounding Derived Parameters for Near-Storm Environments

Proximity soundings have previously been used to explore how the vertical structure of atmospheric conditions affect the formation of severe convective storms. Furthermore, in severe convective storm research and forecasting, sounding derived convective parameters have often served as a proxy to identify the potential severe weather. Many previous studies have used different clustering techniques to group soundings based on either their full vertical structure or spatially local distribution of convective parameters. However, those studies have rarely discussed both scenarios and their relationship together. More specifically, we would like to understand if the clusters associated with convective parameters infer certain types of sounding profiles and vice versa. The dataset that we used consists of more than 15,000 model proximity soundings for tornadoes in the contiguous United States (CONUS). Our study uses self-organizing maps (SOMs) to group soundings based on either their full vertical structure or the surrounding convective parameters. A random forest classifier is trained using the full vertical structure of the soundings based on the spatial convective parameters’ SOM clustering. Silhouette score will serve as a metric to evaluate the discrimination skill of the random forest classification and SOM output. The combination of sounding derived parameters that results in the highest silhouette score indicates such clusters of sounding derived parameters infer a distinguishable pattern of sounding profiles. This spatial and temporal distribution of the sounding profiles from the cluster with the highest silhouette score could provide valuable climatological insight to forecasters and researchers.