Risk modellers grapple with wildfire risk in wake of record losses

In 2017 California was rocked by what, at the time, were thought to be the worst wildfires for many years. Then 2018 arrived, and with it even worse wildfires that scorched the north of the state.

The causes of the wildfires remain, even now, a point of contention. There were multiple fires, with multiple causes and the finger of blame has been pointed at widely different triggers. The Carr Fire of July 2018, for example, was allegedly caused by sparks from a car towing a trailer with a blown tyre, while the deadly Camp Fire of November 2018 is thought to have stemmed directly from malfunctioning power equipment owned by PG&E.

The impact of the fires has been huge, with the total amount of damage being estimated as at least $11.4 billion in damages, a figure that might increase as further information and claims come in. The damage in some areas was so bad that PG&E has been forced to file for bankruptcy protection to shield itself from losses.

The re/insurance industry has watched the wildfires with real concern due to the losses they have inflicted on balance sheets for two years running. How can the risk industry now judge and calculate the risks associated with any such future events, especially against the backdrop of climate change?

A complex peril

Events such as the Californian fires have been looked at in some detail by modelling agencies such as AIR Worldwide and RMS. The AIR wildfire model for the US currently covers the 13 westernmost contiguous states which, according to AIR, comprise the majority of wildfire activity in the country. For RMS, its wildfire modelling scenarios cover the contiguous US (the lower 48 states minus Alaska and Hawaii) with a Canada scenario forthcoming.

According to Tammy Viggato, senior scientist at AIR Worldwide, wildfire is a complex peril to model, with a number of interdependent factors. AIR first calculates the annual area burned in the model domain for each of the model’s 10,000 stochastic years, considering the relationship between weather and fire activity, temperature, precipitation, and drought impact vegetation lifecycles as well as temporal and spatial wildfire variability.

“We then model ignition locations, paying close attention to population density and roadways, as humans directly or indirectly ignite approximately 85 percent of wildfires,” Viggato says. “To model wildfire spread, we consider topography, fuels, and wind, while also taking into account how wildfire suppression would likely occur.”

Chris Folkman, senior director, product management at RMS, agrees about the difficulties of modelling this unique but deadly form of risk. He points out that wildfire modelling is challenging because there are multiple ways a wildfire can cause damage—by radiant heat, by burning embers, or by heavy smoke—and that all must be explicitly modelled. Because of this, RMS has to create three different hazard models and then deftly combine them.

Furthermore, Folkman says, getting it right is computationally intensive—simulating millions of fires, over thousands of years, at high resolution is a truly a big data exercise.

As a result, what he describes as a critical mass of real-world damage data (insurance claims, damage reports, observations) is needed to validate that the model behaves correctly.

“Fire behaviour is complex and depends on many different interacting variables: fuel patterns, wind conditions, relative humidity, temperature, and terrain to name a few,” he says.

The terrain issue is an interesting one. There is a tendency to view most wildfires as being rural but they can also infringe into urban areas, as the town of Paradise found out to its cost in November 2018, when it was totally destroyed by the Camp Fire.

Urban vs rural wildfires

“Wildfire activity varies significantly depending on fuel types and the impact of suppression, which change depending on where the fire occurs and spreads,” says Viggato.

“The majority of wildfire ignitions occur close to where humans are—for either direct or indirect reasons.”

Fires in the wildland-urban interface (WUI) will cause the most insured damage as they encroach on homes and other structures, propelled by fuels and embers. Urban conflagration is possible in the case of severe winds, as the industry saw in the Californian town of Coffey Park which was badly damaged in the Tubbs Fire of 2017, after floating embers ignited rooftops in quick succession. Fires in rural areas may spread quickly without being noticed, or without being immediately extinguished for fuel management purposes.

Folkman agrees, and points to lessons learned from the same event. In the wildlands, fire spread depends largely on fuel and local weather conditions.

“Wildfires are very frequent events, so modelling companies have a lot of observations to draw from. Urban conflagrations, by contrast, are much rarer and typically require extreme weather conditions because there is little burnable fuel other than the structures themselves,” he says.

“Fire spread in urban environments depends on things like exposure density and street width,” he adds. “The Coffey Park urban conflagration in the Tubbs Fire (2017) was a wake-up call for the insurance industry and showed the need for a better understanding of that phenomenon.”

According to Viggato, catastrophe models are intended for a long-term view of risk so they inherently consider climate variability. AIR’s wildfire model directly considers two years of antecedent weather patterns to account for how temperature, precipitation, and drought impacted activity in consecutive years. Models are updated every few years to incorporate the latest science and technology as well as current fuel layers and industry exposures.

“Weather variability is a fact of life, which is why it needs to be simulated for long periods of time to provide a credible view of the risk,” says Folkman.

“We sample from 30 years of weather data to simulate the 50,000 years’ worth of ignitions our model is based upon. We refresh the model as often as necessary to provide an up-to-date picture of the wildfire risk landscape.”

Both companies will doubtless be watching the 2019 wildfire season closely, especially in California, where the odds of a third year of record wildfire losses are being carefully assessed.