In this work, however, we instead make use of Self-Organizing Maps (SOMs) to create statistically distinct two-dimensional clusters of STP values, using a dataset that consists of model-produced [Rapid Update Cycle (RUC) or Rapid Refresh (RAP)] near-storm environmental data corresponding to over 15,000 tornado events in the continental United States between 2003 and 2015. These two-dimensional patterns allow for the identification and clustering of patterns that can distinguish, for instance, between a prototypical U.S. dryline scenario and a scenario dominated instead by high convective available potential energy to the southwest of the tornado; tornadic activity does not always simply correlate with a "bullseye" of a given parameter.
The SOM approach is also used to differentiate between the environments in which tornado outbreaks are likely to occur, rather than occurrences of more isolated tornadoes, and SOMs are built for specific geographical regions of the United States (such as the Southeast, home to atypically high-shear, low-instability tornado environments). For a given geographical region, dynamic parameters of the near-storm environment (such as 0--1-km storm-relative helicity and 0--6-km vector shear magnitude) tend to be higher for outbreak EF1+ tornadoes than for isolated EF1+ tornadoes, whereas the spatial extent and pattern of these high-shear regions do not tend to vary for outbreak versus isolated tornadoes.
In this way, each cluster can be analyzed individually to gain insight into the climatology of tornadic near-storm environments: which two-dimensional patterns in STP correspond to a lower probability of detection? A higher false alarm ratio? A larger fraction of quasi-linear convective systems? More outbreak or high-fatality events?
By answering these questions, we provide a means by which a more nuanced climatology can be built for tornadic near-storm environments that can be applied in situations and locales extending beyond the prototypical U.S. Great Plains springtime early-evening events.
![](images/flag.png)
![](images/blue_ribbon.png)