A Spatiotemporal Distribution of Deep Convection Initiation in the Eastern United States

Thunderstorms are one of the most widespread destructive weather phenomena in the United States. Understanding where and why these storms form is critical to mitigating the severity of their impacts. Machine learning has proven effective in advancing understanding of deep convection initiation (DCI) by exposing spatiotemporal patterns in DCI and identifying environmental parameters capable of differentiating between environments supportive of DCI from those that are not. Past research of spatiotemporal patterns in DCI occurring in the Eastern United States has only focused on smaller regions within the Eastern US, such as the Florida Panhandle or near the Appalachian Mountains. This study will be the first to cover the entire region east of the Mississippi River by using unsupervised machine learning to develop a climatology of deep convection initiation. The dataset will be created using the Thunderstorm Observation by Radar (ThOR) thunderstorm tracking algorithm to identify DCI points using WSR-88D radar data, North American Regional Reanalysis data, and cloud-to-ground lightning data from 2005-2015. This dataset will then be fed into an unsupervised machine learning program that will cluster initiation points based on location, intraseasonal variability, and diurnal variability. Revealing these spatiotemporal patterns in DCI will aid in exposing potential causes of initiation due to local influences.