Machine Learning with Feature Importance Analysis for Tornado Prediction from Environmental Sounding Data

Abstract

Tornadoes pose a forecast challenge to National Weather Service forecasters because of their quick development and potential for life-threatening damage. The use of machine learning in severe weather forecasting has recently garnered interest, with current efforts mainly utilizing ground weather radar observations. In this study, we investigate machine learning techniques to discriminate between nontornadic and tornadic storms solely relying on the Rapid Update Cycle (RUC) sounding data that represent the pre-storm atmospheric conditions. This approach aims to provide for early warnings of tornadic storms, before they form and are detectable by weather radar observations. Feature analysis of a Random Forest machine learning model uncovers that the pressure variable has little impact on the classification process, which is consistent with known key physical attributes of tornado formation, demonstrating the ability of machine learning techniques to provide insight solely based on the data.

A Powerful Prediction Framework of Fracture Parameters for Hydraulic Fracturing Incorporating eXtreme Gradient Boosting and Bayesian Optimization

“…There are a variety of methods for analyzing feature importance, of which the common ones include [46][47][48][49][50] the following.…”

How Are Tornadic Supercell Soundings Significantly Different From Nearby Baseline Environments?

“…Past studies have identified the importance of environmental parameters and how they can aid in the prediction of tornadic supercells (Davies‐Jones, 1984; Klemp, 1987; Markowski et al., 2002; Rotunno, 1981), and since then many studies have focused on the evolution, the spatial distributions, and the temporal distributions of those environmental parameters to characterize regional, diurnal and seasonal differences in tornado‐favorable environments (R. L. Thompson et al., 2012; R. Thompson et al., 2013; A. Anderson‐Frey et al., 2016; Reames, 2017). In recent years, with the rise in popularity of machine learning and deep learning, there is a substantial body of work that uses environmental parameters derived from near‐storm proximity soundings to classify the shared characteristics of tornadic storms that occur in different regions, seasons, and times of day (Lu et al., 2015; A. K. Anderson‐Frey et al., 2017; Warren et al., 2021), in addition to providing probabilistic prediction of tornadoes (Coffer et al., 2020; Shield & Houston, 2022). Most studies assume that tornadic environments are substantially different from the typical environments for a specific location at a certain time of year and time of day in which tornadic weather does not occur (hereafter referred to as baseline environments), but studies rarely quantify the differences between the two.…”

“…

In recent years, with the rise in popularity of machine learning and deep learning, there is a substantial body of work that uses environmental parameters derived from near-storm proximity soundings to classify the shared characteristics of tornadic storms that occur in different regions, seasons, and times of day (Lu et al, 2015;A. K. Anderson-Frey et al, 2017;Warren et al, 2021), in addition to providing probabilistic prediction of tornadoes (Coffer et al, 2020;Shield & Houston, 2022). Most studies assume that tornadic environments are substantially different from the typical environments for a specific location at a certain time of year and time of day in which tornadic weather does not occur (hereafter referred to as baseline environments), but studies rarely quantify the differences between the two.

…”

Bayesian Analysis of Spatial Model for Frequency of Tornadoes

“…Moore [12] analyzed the spatial distribution characteristics of tornadoes in the United States; they [13] looked at the relationship between tornado activity in the United States and the El Nio/Southern Oscillation in all four seasons and across multiple regions. Coffer [14] used Random Forest classification to predict tornadoes. Allen [15] employed Kernel Density Estimation for spatial pattern analysis and space-time cubes to visualize the spatiotemporal frequency of tornadoes and potential trends.…”