Rising from the ashes.

For two weeks from late October to early November 2003, at least 10 major wildfires swept across Southern California from the Mexican border to north of Los Angeles. The two largest blazes, in San Diego and San Bernardino counties, combined to cause at least $2.04 billion in insured losses. These fires were by far the biggest driver in the property/casualty industry's fourth-quarter catastrophe losses--put at $2.6 billion--and ranked as the highest quarterly losses for a 10-year period.

As the flames scorched vast tracts of woodlands and reduced residential areas to charred rubble, reinsurance intermediary Benfield was at work creating a fire loss potential model based on the California wildfires. The idea was to deliver the latest information to clients on their loss potential, said Paul Budde, senior vice president. "We did not have a tool for this prior to the wildfires breaking out," he said. "Basically it was a case of the fires were burning, they continued to burn, and insurers had a need to understand their loss potential and identify affected policyholders, so we tried to do something to meet those needs for them."

Benfield already had property exposure data, down to street address level for the affected areas, that could be run through catastrophe models. It was this level of specificity that allowed Benfield to map wildfire exposures for the first time, Budde said.

"For a particular home, we know where it's located. We have tools to map each street address," Budde said. "That gives us a common frame of reference."

His company then relied on daily information issued by the United States Department of Agriculture Forest Service and the ignition points detected by the U.S. Forest Service's remote sensors. Once Benfield had the ignition points mapped, it created "burn zones"--polygonal representations of the fire boundaries for the seven main fires--and overlaid those areas with the latitudes and longitudes of the risks for a particular client. These zones were updated on a daily basis to include new detections, or a spread in each fire.

"We had a method where we could accumulate the number of risks and insured value of the properties within those burned regions. We even gave insurers policy IDs for potentially affected risks," Budde said.

Although it wasn't immediately clear if properties in the burn zones had been destroyed or if firefighters had been able to protect them, insurers still could get a sense of the potential loss by seeing the direction in which the fires were moving and their risks in relation to that path, Budde said.

Every morning, Benfield's staff refreshed the data, drew new fire boundaries and then ran more analyses. Clients with the greatest exposures received daily updates, Budde said.

The biggest challenge in developing the fire loss potential model was "the speed, the quick turnaround," he said, "because if you were to give this information to an insurer three days or a week afterwards, it really wouldn't mean much."

The big three catastrophe modeling companies do not yet have wildfire models on the market but are expecting to launch them soon. Risk Management Solutions, Newark, Calif., said it is actively working on one, spurred on by several clients that experienced significant losses in the California wildfires. Eqecat International, Oakland, Calif., has modeled wildfire for a number of companies and plans to integrate a wildfire model into its standard suite of products in July 2004. And Boston-based AIR Worldwide Corp. is using parent company Insurance Services Organization's extensive FireLine database to craft its own fully probabilistic wildfire model. FireLine is a software and database product that combines risk factors and satellite imagery to pinpoint hazards from wildfire.