The Observed Relationship Between Pretornadic Mesocyclone Characteristics and Tornado Intensity with Machine Learning Applications

Abstract:

Trapp et al. (2017, JAS) used idealized model simulations of supercell thunderstorms to demonstrate support of their hypothesis that wide, intense tornadoes should form more readily out of wide, rotating updrafts. Observational data were used herein to test the generality of this hypothesis, especially to tornado-bearing convective morphologies such as quasi-linear convective systems (QLCSs), and within environments such as those found in the southeastern U.S. during boreal spring and autumn. A new radar dataset was assembled that focuses explicitly on the pre-tornadic characteristics of the mesocyclone, such as width and differential velocity: the pre-tornadic focus allows us to eliminate the effects of the tornado itself on the mesocyclone characteristics. GR2Analyst was used to manually analyze 102 tornadic events during the period 1 April 2011 to 1 May 2019. The corresponding tornadoes had damage (EF) ratings ranging from EF0 to EF5, and all were within 100 km of a WSR-88D. A key finding is that the linear regression between the mean, pre-tornadic mesocyclone width and the EF rating of the corresponding tornado yields a coefficient of determination (R²) value of 0.75. This linear relationship is the higher for discrete (supercell) cases (R²=0.82), and lower for QLCS cases (R²=0.37). Overall, we have found that pre-tornadic mesocyclone width tends to be a persistent, relatively time-invariant characteristic that is a good predictor of potential tornado intensity. These findings have motivated us to explore tornado-intensity prediction approaches using pre-tornadic mesocyclone characteristics and other data through machine learning applications. Several classification machine learning algorithms such as Logistic and Lasso Regression, Random Forest, K-Nearest Neighbor, Naïve Bayes, Decision Trees, and Support Vector Machines have been implemented and are being used to examine their skill in predicting potential tornado intensity, either non-significant or significant, for a given storm.