Analyzing the effects of low level boundaries on tornadogenesis through spatiotemporal relational data mining

Multiple theoretical, observational, and numerical modeling studies have shown that the presence of low-level boundaries, including mesoscale fronts and outflow boundaries, have influenced tornadic supercell thunderstorms. Due to limitations associated with available surface data, such studies have not been able to analyze the large number of tornadic and non-tornadic supercell thunderstorms needed to determine general trends associated with boundary interactions and storms. This study combines Oklahoma Mesonet surface observations with a ten-year database of supercell thunderstorms in Oklahoma and tornado reports from SPC and NCDC into a spatiotemporal relational data framework. The framework consists of thunderstorm and air mass objects joined by spatial relations describing their relative positions over time. Air mass objects are created by applying k-means clustering to a set of Oklahoma Mesonet observations at a particular time, which groups similar observations together. Boundaries can be found along the edges of these observation groups. The spatiotemporal relational probability tree (SRPT) and the spatiotemporal relational random forest (SRRF) predict the probability that a tornado will form from a given supercell thunderstorm. The SRPT is a decision tree that predicts probabilities by randomly sampling and choosing questions concerning various spatial and temporal aspects of a particular dataset. The SRRF is an ensemble of SRPTs that follow properties similar to the random forests algorithm. Both are trained and tested to determine which algorithm provides more skill on the dataset. Analysis of the nodes of the SRPT and the SRRF provides insight into what factors the algorithms find most important for differentiating tornadic and non-tornadic storms based on boundary location and type. From these factors, a more general understanding between low-level boundaries and tornadic supercells can be acheived.