Increases in atmospheric carbon dioxide are forcing climate, which in turn are forcing wildfire activity and amplifying uncertainty in wildfire management. Fire management organizations need to anticipate and plan to adapt to changes in extreme weather to create organizational resilience, as well as to promote ecological and community resilience. This may include the enhanced use of analytics to support decision making regarding wildfire preparedness, response, and mitigation actions. Models to inform management need to be based in a fundamental understanding of fire management systems and the influences of weather and climate on fire phenomena at appropriate scales. In this thesis I develop predictive models to inform daily preparedness planning following a systems approach. First, I show that fire weather, fire phenomena, and fire management decisions are complex systems connected by scale, and outline a hierarchical spatio-temporal fire decision making framework to decompose a complex system into tractable decision spaces. Second, I develop a framework to link statistical models of fire phenomena at different scales; many fire characteristics such as fire size, area burned, and fire frequency are compounded across temporal and spatial scales. Because drivers of fire such as human and lightning ignition sources, weather, and fuel properties are either random, or cannot be precisely known over any period or precisely represented in models, fire prediction is inherently probabilistic. Third, I introduce novel methods and covariates to improve the accuracy of daily human and lightning-caused fire occurrence prediction in British Columbia, and to rank variable importance. Fourth, I develop novel models of the probability of using aircraft in wildfire initial attack, conditional on a fire occurring. These can be combined with daily fire occurrence models to estimate the number and location of daily aircraft initial attack targets. Collectively this research highlights the importance of collaboration between fire researchers and statisticians to develop skillful predictive models grounded in domain specific knowledge.
Copyright is held by the author(s).
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Member of collection