Open Collections

Abstract

Wildfires are an increasingly severe disturbance in British Columbia, posing significant risks to forest resources, timber supply, and the long-term sustainability of bioproducts production. In regions such as the Williams Lake Timber Supply Area (TSA), increasing wildfire frequency and intensity create substantial uncertainty in the availability of forest residues used for biofuel and bioenergy production. Accurately forecasting biomass supply under wildfire uncertainty is therefore essential for designing efficient and economically viable bio-based supply chains. Previous studies, including the baseline model used in this research, have developed optimization frameworks for forest-based bioenergy supply chains using deterministic biomass estimates, such as those provided by FPInnovations. These approaches assume stable harvesting conditions and neglect the effects of climate variability and wildfire disturbances, limiting their ability to represent real-world supply uncertainty. To address this gap, this thesis develops a data-driven modeling framework that explicitly incorporates wildfire uncertainty into biomass forecasting and supply chain optimization. The research is conducted in three phases. First, machine learning (ML) models including logistic regression, random forest, XGBoost, and artificial neural networks are trained using historical climate and fire data to estimate the annual probability of significant wildfire occurrence in the Williams Lake TSA. Second, these probabilities are integrated into biomass supply forecasting using Harvest Billing System data, Vegetation Resource Inventory data, and climatic variables. A random forest model is then used to forecast biomass availability from 2000 to 2034, capturing wildfire-driven variability in forest residue supply. In the final phase, a mixed-integer linear programming (MILP) model is applied to optimize the regional bioenergy supply chain by determining facility locations, biomass flows, and production levels. Results show that the ML-based biomass forecasts closely align with observed harvest trends while responding dynamically to wildfire disturbances. Compared to FPInnovations’ deterministic estimates, the data-driven approach better captures temporal variability and uncertainty. Overall, this research demonstrates that integrating wildfire uncertainty into biomass estimation improves forecast reliability and supports more robust and sustainable bioenergy supply chain design in British Columbia