Wildfire prevention is one of the most effective and economical risk mitigation strategies. Human-started wildfires account for over 60% of all recorded wildfires across the western United States and are responsible for the vast majority of wildfire-related societal impacts, underscoring the value of effective wildfire prevention strategies. To address this need, we developed machine learning models that not only effectively predict spatial and temporal patterns of wildfire ignitions but reveal the nuanced interactions among physical, biological, social, and administrative factors that govern wildfire ignition outcomes. Annual temperature (climate), discovery day-of-year (seasonal pattern), fire year (trend), and national preparedness level (management and fire danger) were the primary governing factors in models of all ignitions, natural ignitions, and human-caused ignitions. Secondary governing factors of natural and human-caused ignitions, respectively, were weather-related attributes and weather and social attributes. Our results indicated that although daily ignition probabilities generally track weather patterns, they can remain persistently high in areas where human factors dominate. Our results also show that models relying solely on weather do not accurately predict wildfire ignitions, reinforcing the fact that ignitions are caused by complex interactions among diverse factors.
Speaker
Dr. Mojtaba Sadegh
