An Examination of Outliers from Self-Organizing Maps of Severe Storm Environments

Self-organizing maps (SOMs) were further explored in this study for improving tornado forecasting. In previous work, severe weather events were split into the following storm types: nonsupercell, nontornadic, weakly tornadic and significantly tornadic supercells. SOMs were used to classify these events based on vertical profiles of variables from model proximity soundings. In this study, SOMs were explored to explain why outliers were matched to a particular node. To do this, other important variables such as convective available potential energy (CAPE), convective inhibition (CIN), lifting condensation level (LCL) height, and storm relative helicity (SRH) were calculated and plotted as box-and-whisker plots to better observe their variability between storm types in each node. Two outliers were picked out for more detailed case studies.

SOM nodes that were organized based on a kinematic variable, such as ground relative wind, had less variability between storm types in bulk wind difference (BWD) and SRH but more in CAPE and CIN. SOM nodes that were organized based on a thermodynamic variable had similar variability between the storm types for all calculated variables. This suggests that the kinematic variability is reduced more by the kinematic SOMs than the thermodynamic variability is reduced in the thermodynamic SOMs.