Classification of GLM Flashes Using Random Forest Method

With the launch of the Geostationary Lightning Mapper (GLM) on the GOES-16 satellite, there now exists a way to look at total lightning continuously from space over both land and ocean. The GLM detects total lightning and does not currently have the capability to distinguish intra-cloud (IC) lightning from cloud-to-ground (CG) lightning. This research focuses on exploring how to differentiate CG and IC lightning flashes detected by GLM using a Random Forest machine learning algorithm. GLM flash and group characteristics, including the flash energy, child count, grandchild count, maximum number of events in a group (MNEG), maximum group area (MGA), flash duration, propagation, elongation, flash footprint, maximum group energy, and mean group energy are implemented into the Random Forest model and used to predict CG vs. IC flash classification. Results of model implementation are revealed and skill scores including probability of detection (POD), false alarm rate (FAR), critical score index (CSI), and percent correct (PC) are presented. The importance of each parameter as a predictor in the Random Forest algorithm is also discussed. Other applications include the spatiotemporal calculation of the cloud flash fraction evolution over the lifetime of hurricanes and storms. This research aids in broadening the scope for GLM data applications.