Wildfire Mitigation through Explainable Risk Predictions

The proposed innovation is to add explanations into Terrafuse’s existing wildfire risk prediction model to provide localized, actionable guidance on the effectiveness of different wildfire mitigation efforts. Mitigations are actions like clearing defensible space around buildings or installing fire-resistant vents. Explanations describe the relative importance of each input to the wildfire prediction model, or feature, to the overall wildfire risk. We refer to these explanations as feature contributions. We will implement them by building on our prior work with the Shapley Additive Explanations (SHAP) technique from game theory. Feature contributions will provide guidance on which mitigation actions will be most efficacious. For example, where wildfire risk is driven primarily by the amount of fuel, clearing defensible space around structures will be more effective. However, for locales where wildfire risk is strongly influenced by wind, structural improvements that protect against wind-blown embers, like enclosed eaves and fire-resistant vents, may be more efficacious. This localized mitigation guidance will support more optimal resource allocation and better decision-making during pre-wildfire planning for the insurance industry and the public.