Actuarial Science and Financial Mathematics seminar series
Jin Xing
TD Insurance
Room: M3 3127
GeoAI-based Wildfire Modelling for Actuarial Decision-Making: Opportunities and Challenges
Wildfire is an escalating challenge for Property & Casualty (P&C) insurance.
We quantify financial loss using Average Annual Loss (AAL), which integrates insurance exposure, the probability of burning, and expected damage. Climate change disrupts this framework by reshaping where and when fires occur, including in areas historically viewed as low risk. To address this non-stationarity, we developed a Canada-wide Geospatial AI (GeoAI)-based catastrophe model that forecasts ignition and spread using multi-source data (wildfire maps, weather, fuel maps, topography, and human proximity).
Experiments show promising results, but important gaps still remain, using our proposed GeoAI framework. This talk presents validation on recent Canadian wildfire events, highlight limits (e.g., data bias, tail uncertainty), and outline next steps. While actuarial applications (pricing, capital, claim and portfolio planning) motivate the work, the focus here is more scientific: how GeoAI can translate wildfire dynamics into decision-ready risk metrics without sacrificing transparency or accuracy. We aim to spark further research toward robust, AI-driven wildfire catastrophe modelling work across Canada.