Could the California Wildfires Have Been Predicted’

Lessons for Agents and Insurers

In late October 2003, major wildfires broke out in five southern California counties, burning more than 750,000 acres, or an area the size of Rhode Island. These devastating fires burned for several weeks before being brought under control. Insured losses were reported to be more than $2 billion dollars, and more than 3,500 homes were totally destroyed. Although these fires may not have been preventable, as an insurer, could technology be used to predict these fires, and thus avoid risky areas?

Most insurers have guidelines that limit the amount of brush exposure they will accept. Properties that are in or very close to brush are usually declined by insurers, but generally accepted in the non-standard/excess markets. For example, a company may decide that they do not wish to insure properties that are within 1,000 feet of brush. While insurers have had these guidelines for years, they have not had an accurate, cost effective manner to identify brush hazards.

Current methods
Most companies had been using two methods to determine if the property was in close proximity to brush: 1) Brush atlases and 2) inspections. Both methods are flawed.

Insurance companies and agents have used brush atlases for years to screen for brush. In fact, this method has been the de-facto standard. Every insurance company and most agencies have a copy of these brush atlases handy.

Unfortunately, there are two fundamental flaws in this process. First, these atlases only cover six counties in Southern California. This means there are high risk areas of both brush and forest fires throughout California that have not been mapped. Secondly, these atlases are based upon “Designated Brush Areas” as identified by the California Fair Plan. The designated brush areas are not a comprehensive list of all brush areas. This means that many areas which are covered by brush are not identified in the atlases! These are significant problems.

Insurers who use these books for brush risk mitigation have a false sense of security. Not only do these books miss the wildfire hazards in over 50 counties, but even where the maps are available, they do not identify all brush areas; only designated brush areas, which are politically motivated.

Many other companies use property inspections to check for brush. This method is also flawed for several reasons. First, inspections are performed at the end of the underwriting process. By the time a property is inspected, the agent and underwriter have both done work, and the property owner thinks that they have obtained coverage. It is an unpleasant surprise for everyone involved to find out after the inspection that the insurance company does not want the risk.

Secondly, the insurer is still relying on the judgment of the inspector. Can the average person really judge distance? Do they recognize the different types of vegetation and what types should be of concern?

Thirdly, because inspections are expensive, many companies only inspect a portion of applications. Unfortunately, many companies use the brush atlases to identify which properties should be inspected. If the property is not shown to be in or close to brush hazards, it is not inspected. However, due to aforementioned omissions in the brush areas, many houses close to brush are never inspected!

How mapping technology can help
Several companies have built wildfire models using Geographic Information Systems (GIS) technology. These models predict the risk of wildfire hazards. The models are generally based upon vegetation data taken from satellite and aerial images. The vegetation data classifies land cover into hundreds of different types of vegetation. This is far too many for the average user to comprehend. Therefore, based upon extensive research, the data is reclassified into much broader categories based upon how well each type of vegetation or “fuel” burns. The goal is to make the information, “actionable” for both agents and underwriters.

The proof is in the pudding
The wildfires in the fall of 2003 provided an excellent opportunity to evaluate the performance of these wildfire models. An analysis was performed to determine how well brush models performed during actual fires. The goal of the analysis was to know if wildfires could indeed be predicted.

Fire extent data was obtained from GeoMac, an interagency government group that provides information on wildfires. Fire extents show the areas that burned in the fires. The fire extents were laid on top of the brush models to determine what percentage of the area that burned was predicted to be a fire risk.

The analysis showed that these wildfire models were 85-90 percent effective at predicting the areas that burned! Basically, this means that of the areas that burned, the GIS model had identified most of these areas as fire risks. The analysis was done on a county-by-county basis. The lowest score was 73 percent in Ventura County, and the highest was San Bernardino County at 91 percent.

Using the areas that were identified by the Fair Plan as designated brush areas, the average predictability was less than 25 percent. These scores ranged from a high of 80 percent in San Bernardino County to a low of 0 percent in San Diego County (there are no designated brush areas in San Diego). To put this in perspective, in Riverside County, only 1 percent of the area that burned was identified as at risk by the Fair Plan! Conversely, the GIS model identified 81 percent of the area that burned as at risk! So older methods have been proven to be unreliable while the new brush models were proven to be highly effective.

If we couldn’t stop the fires, why is this important?
Insurers are not required to write every risk. If insurers had better information as to where the brush was located, they could make more informed decisions. Insurance companies that have done analysis after the fires have in some cases reported that if they had used a tool like this, they could have avoided as many as half of their losses.

Agents can benefit from this technology, too. Using this type of mapping technology, agents can effectively screen for brush prior to submitting an application to carriers. This will save them the time and expense associated with driving to the property to look for brush. If they do not screen for brush, they risk the policy being cancelled by the carrier after an inspection is performed. Then the agent has a disappointed customer. Additionally, many agents are given bonuses for writing profitable business for carriers. Avoiding wildfire losses is one way to increase profitability. While this technology has for years been very expensive, the advent of Internet mapping technology and Web-based services has brought the cost down to as little as $1 or less per address!

For decades, the innovative use of technology has been known to be an effective source of competitive advantage. Using geographic information systems technology, insurers definitely could have avoided many of the riskier properties. And agents could have spent more time producing profitable business.

This technology is not only applicable to wildfires, but can be used to identify other natural hazards, too. Mapping technology and natural hazard modeling has been used to evaluate risks such as flood, hurricanes, earthquakes, tornadoes and hail.

Although 2003 was the worst fire season in history, the danger has not passed. The 2004 fire season is expected to be as bad as last year. Employing good risk management techniques and mapping technology now can lead to significantly lower losses later!

Dan Munson is the vice president of Sales & Marketing at CDS Business Mapping, LLC. CDS specializes in GIS mapping solutions for insurance companies. The company runs the RiskMeter Online, an online hazard mapping service for agents, brokers and insurers.