TOWARD INTELLIGENT WILDFIRE SYSTEMS: INTEGRATING MACHINE LEARNING ACROSS MITIGATION AND SUPPRESSION

Abstract

Wildfire management involves human judgment and decision-making across multiple temporal and operational scales, from long-term fuel treatment planning to real-time suppression under changing environmental and operational conditions. This dissertation examines different wildfire management activities through data-driven approaches that combine machine learning, spatial analysis, and historical datasets. Rather than prescrib ing optimal actions, this research focuses on identifying patterns in observed fuel treat mentandaerialsuppression decisions and relates them to measurable environmental and operational conditions. The first chapter examines hazardous fuel treatment decisions in Colorado’s national forests using historical treatment records and spatial variables related to wildfire hazard, terrain, infrastructure, and human influence. Neural network models are used to ana lyze associations related to treatment type occurrence. The second chapter develops a deep learning approach for identifying likely aerial suppressant drop activity, providing a new way to infer aerial suppression execution from publicly available telemetry data. The third chapter uses ADS-B data to analyze cross-agency aerial response timing during initial wildfire response in California, quantifying how environmental and social condi tions are associated with aircraft arrival patterns. Together, these studies illustrate how machine learning and open operational datasets canmakedifficult-to-observecomponentsofwildfiremanagementmoremeasurable. Col lectively, they contribute new methods and empirical insight for studying wildfire man agement across operational scales.