AI-driven research explores predictive modeling for Canada’s forestry resilience under climate change
Algoma University Master of Computer Science (Thesis-Based) student, James Gu, recently presented his research project on forest mortality prediction and climate impacts at the Science Carnival held at the Canadian Bushplane Heritage Centre in Sault Ste. Marie. The presentation highlighted the use of artificial intelligence (AI) to better understand and forecast changes in Canada’s forestry resources.
Under the supervision of Dr. Rashid Khokhar, Faculty of Computer Science and Technology (FCST) Assistant Professor at Algoma University, the project focuses on developing advanced predictive tools for understanding forest dynamics and tree mortality across Canada. This work contributes to ongoing research on improving forecasting methods in the context of climate change impacts on forest ecosystems.
Building on an initial dataset provided through an earlier collaboration with the Ontario Forest Research Institute (OFRI), and the Ontario Ministry of Natural Resources and Forestry, the study was independently expanded by combining the available forestry data with publicly available climate datasets from 1952 to 2024, in order to construct the training dataset for the model. Through this, historical trends, such as the 2001–2002 drought in western Canada that resulted in widespread aspen mortality, provided context as to the importance of predictive tools for assessing large-scale forest loss and ecosystem vulnerability.
“Forest mortality is driven by complex interactions among climatic, biological, and environmental factors,” said Dr. Khokhar. “Understanding these interdependencies is essential for improving prediction accuracy and identifying underlying causal mechanisms. Canada’s forestry resources represent a strategic economic asset, and climate change continues to impact forest health, making reliable forecasting increasingly important for informed policy and management decisions. National forests are also subject to large-scale disturbances, including insect outbreaks, which further emphasize the need for improved predictive tools for forest health assessment.”
The research employs a neural network-based forecasting model enhanced with graph neural network structures to identify relationships between environmental variables. Domain-informed constraints were applied to ensure physically meaningful relationships, and the model was trained on multi-decade forestry and climate data. The system is currently being evaluated under data perturbation and data poisoning scenarios as part of ongoing robustness analysis.
Initial experiments under these conditions suggest that key relationships identified by the model show a degree of stability, indicating potential robustness; however, further validation and analysis are ongoing.
The presentation also provided attendees with insights into how advanced AI techniques, such as causal modeling and graph-based learning, can be applied to real-world environmental challenges, sparking interest in their potential application across other domains.
The project presented at the Science Carnival exemplifies Algoma University’s dedication to community sustainability, collaborative research, and innovative methods of learning while addressing climate impacts on Canada’s natural resources.
Share Article
