Modeling the Probability of Dry Lightning-Induced Wildfires in Tasmania: A Machine Learning Approach

Abstract

Dry lightning is a prevalent episodic natural ignition source for wildfires, particularly in remote regions where such fires can escalate into uncontrollable events, burning extensive areas. In this study, we aimed to understand the interplay of environmental, fuel, and geographical factors in evaluating the probability of fire initiation following dry lightning strikes in Tasmania, Australia. We integrated dry lightning, active fire records, and gridded data on fire weather, fuel, and topography into a binary classification framework for both fire-initiating and non-fire-causing lightning strikes. Employing statistical and machine learning techniques, we quantified the likelihood of fire initiation due to dry lightning, with the resampled Random Forest model exhibiting notable performance with an ROC-AUC value of 0.98. Our findings highlight how fuel characteristics and moisture content associated with particular vegetation types influence fire initiation and provide an objective approach for identifying susceptible regions of dry lightning ignitions, informing associated fire management responses.