Earth System Modeling for Integrated Multiscale Fire Behavior Prediction and Reinsurance Risk Assessment

Novel integration of generative AI with Earth system modeling to produce high-resolution fire risk metrics for reinsurance stakeholders, addressing the $394-893 billion annual costs of wildfire activity in the U.S.

Problem Statement

Medium-term wildfire risk projections (1-10 years) are critically needed by the insurance industry to assess community risks and economic impacts, as climate change and fuel accumulation drive unprecedented fire activity with annual costs reaching $394-893 billion in the U.S. alone. Current fire prediction capabilities face significant limitations:

• Traditional Earth system models operate at coarse spatial scales (50-100 km) that cannot resolve local fire weather conditions
• Existing fire weather indices provide only daily snapshots that miss critical synoptic changes driving extreme fire behavior
• Conventional downscaling approaches suffer from computational constraints, stationarity assumptions, and inability to capture complex non-linear relationships

Proposed Solution

To address these gaps, we propose a novel integration of generative artificial intelligence with Earth system modeling that will produce high-resolution fire risk metrics at the spatial and temporal scales required by reinsurance stakeholders, employing state-of-the-art machine learning-based downscaling to generate detailed medium-term climate projections up to 2040 from cutting-edge Earth system models. This research will provide the insurance industry with unprecedented capability to assess wildfire risks at the spatial and temporal scales needed for accurate pricing and risk management, potentially saving billions in losses and improving community resilience to wildfire threats.

Tools & Techniques

• Diffusion-based probabilistic modeling for uncertainty quantification in climate projections
• Medium-term climate projections (up to 2040) using stochastic ensemble approaches
• Multiscale modeling framework for wildfire risk assessment
• Fire weather index development and validation
• Probabilistic ensemble generation for risk assessment and uncertainty propagation

Research Team & Collaborations