How to improve wildfire modelling

Historical loss models and fire maps insurers use are in need of a tune-up, and experts have turned to artificial intelligence to do the job.

Traditional wildfire models rely on static historical data such as hectares burnt, the number of events, and the length of the fire season. But advances in AI, machine learning, and probabilistic modelling can be combined to predict the probability and intensity of a wildfire, as well as expected losses, says Mihalis Belantis, CEO of AISIX Solutions Inc.

Data inputs are based on ignition types (i.e. human or natural), region-specific vegetation (i.e. fuel), topography, how topography impacts fire spread, weather, and building information.

“We take weather and climate into consideration, and then we run it through something called Cell2Fire2,” which is a fire growth engine AISIX integrates with the Canadian Forest Fire Behavior Prediction (FBP) System, he says. “Then we run it through something called the Monte Carlo Simulation,” which simulates millions of wildfire scenarios to predict the probability of a variety of outcomes.

“We slap in all the building information we have, and it gives you the burn probability, risk score, and fire intensities.”

The AI also produces a 1-to-5 risk score with expected losses based on current and future climate conditions.

Most B.C. scores are close to a 5 due to its dense forests, “but you take the same forest area out east, like Muskoka, and it is probably sitting at 3 just because of the different types of trees and fuel available,” says Belantis.

He says the company’s scientists “saw a gap in wildfire modelling,” since historical models don’t get as granular or provide personalized insights.

“The key to de-risking your project, especially when you’re an insurer, is having the best wildfire data…[or] climate data possible that you’re when you’re doing this risk and hazard assessment,” he says. “If you’re just using the Fire Weather Index, I don’t think you have a clear picture of what the wildfire risks are.”

For Mihalis, better wildfire modelling is a matter of carriers making more informed underwriting choices so they can keep insuring clients. Poor modelling can increase the chance insurers may find it too risk to cover wildfires. In California, for example, he points out, many insurers have stopped offering home insurance policies amid more frequent and severe disasters and higher costs.

“The more robust model you have, and the more inputs you have going into it, the better that is going to be to help you understand what the real risks are,” he says.

Feature image by iStock.com/gorodenkoff

Alyssa DiSabatino

Alyssa Di Sabatino has been a reporter for Canadian Underwriter since 2021, covering industry trends, market developments, and emerging risks.