Following the devastating 1999-2000 fire season, the secretaries of Agriculture and Interior made the following ominous prediction in a letter to President Bill Clinton: “Explosive growth in the wildland-urban interface now puts entire communities and associated infrastructure, and the socioeconomic fabric that holds communities together, at risk from wildland fire.”

Last October, in the span of 30 days, five fires in southern California destroyed 750,000 acres of woodland; leveled more than 3,700 homes and 862 businesses; killed 22 people, including one Forest Service firefighter; injured another 185; and caused damages that are rapidly approaching the $1 billion mark.

These numbers are so staggering it's difficult to wrap your mind around them. Eerie images of entire neighborhoods leveled from fire — with nothing standing but chimney stacks and stumps of trees — and satellite photos of California smothered in a veil of hazy-red smoke captured the terrible drama of this event. Two months later, the same areas suffered devastating and fatal mudslides, the result of a denuded landscape incapable of retaining water.

To say that these fires affected “entire communities and their infrastructure, and the socioeconomic fabric that holds them together” is an understatement. Without question, the secretaries of Agriculture and Interior were on their game in their gloomy prediction.

AUSTIN'S PROBLEM

Travis County, Texas, provides a perfect example of rapid interface development. Driven by Austin's explosive economy and the resultant increase in wealth and population, the urban expansion of homes, neighborhoods and businesses into the hill country of west Austin and Travis County has exploded.

Today, Austin has numerous neighborhoods of million-dollar-plus homes situated along wildland preserves. Many of these homes are nestled along ridge-lines, cliff edges and other classic interface hazard zones.

Years of fire suppression in the area have significantly disturbed the renewal process of natural fire regimes. The result has been the gradual accumulation of understory and canopy fuels to levels of density where high-energy, intense wildfires can result, further increasing the hazardous nature and exposure of the interface problem.

Perhaps most alarming, while the fuel types are somewhat different, substituting Texas mountain cedar for California chaparral, the topography and interface conditions are very similar to those present in southern California prior to the devastating 2003 interface fires.

Although the danger was evident, any attempt to address the interface problem in Austin and Travis County via model codes or other mitigation strategies before the completion of a risk analysis would be premature. The Austin Fire Department needed clear criteria for identifying, measuring and cataloging risk in the interface before it could develop coherent strategies to mitigate it.

The problem went beyond mitigation strategies. Without such empirical evidence, any attempt to create political or social momentum to address the problem likely would fail. There was an urgent need to define risk in the interface through comprehensive hazard modeling/mapping, and to develop a fire management plan to reduce community vulnerability.

HOMEWORK TIME

The west Austin and Travis County interface area is spatially non-specific and includes numerous jurisdictions and stakeholder interests, such as wildlife preserve management areas. The first step to understand risk in the interface was to identify all the stakeholders who would be interested in or affected by the proposed risk analysis. Visits with the various stakeholder groups revealed that they shared the fire department's concerns.

Whether the stakeholder was a fire department, landowner, forester or habitat manager, each was deeply concerned about the wildfire potential in west Austin and Travis County, and all stakeholder groups agreed to lend a hand where possible in completing the study. This initial relationship-building proved vital to the later success of the study, as the department ultimately had to spend hundreds of hours in the field collecting data and ground-truthing assumptions.

Perhaps the greatest benefit to come from the comprehensive stakeholder assessment was learning there was significant funding available from the Federal Emergency Management Agency to support studies similar to the one the Austin Fire Department's fire marshal was proposing. Ultimately, FEMA's Project Impact Disaster-Resistant Communities Program supplied $20,000 to support the risk analysis.

The next step was to dive into a comprehensive review of the relevant literature and to identify benchmark studies of interface risk. The department was committed to developing a model of risk that was tailored to Austin's unique characteristics while remaining consistent with the accepted practices, language and understandings of the wildfire body of literature.

After a thorough review of the relevant literature and numerous conversations with subject-matter experts across the country, we decided to create a descriptive model of risk based on the confluence of three broad categories of variables: weather, spatial and human. This conceptual framework is designed to capture the essential elements of risk in the interface and includes variables to determine degrees of risk in three categories of potential:

• Potential for ignition,
• Potential to burn and
• Potential human consequence.

IGNITION POTENTIAL

Because weather patterns change over time and weather is the most important contributing factor to fuel moisture, any model designed to determine risk in the interface has to account for weather fluctuations. Each area of the country (and the world) experiences predictable times of the year when wildfire is more likely, hence the concept of fire seasons that are a product of cyclical variations in weather over time. A model of wildfire risk in the interface has to include time-scaled pictures of weather patterns and the effect of these patterns on the available fuels over a given length of time. Of course, combustibility is a function of fuel moisture, which is a direct result of weather patterns.

Fortunately for Austin, right in the middle of the 90,000-acre study area is a remote automated weather station. As part of the National Weather Information System network, this station collects daily information on important changes in weather, including wind speed, precipitation, barometric pressure, temperature and humidity. From these daily data inputs, fire managers can run a number of algorithms to determine changing conditions in wildfire danger; these algorithms are called outputs and are fundamental to the National Fire Danger Rating System. The outputs provide predictive information on wildfire potential and are used by fire professionals to categorize periods of the year as high or low risk.

Because each output uses different weather inputs to determine a variety of fire behavior predictions, we decided to use a set of four NFDRS outputs, the sum of which should offer a good picture of climate patterns for the study area. Based on seven years of data, the mean output level (average expected weather) was calculated based on 28-day cycles. For each NFDRS output, the Texas Forest Service ran a series of BEHAVE models to create a scale of values representing degrees of danger; breakpoints for the scale were based on the expected effect the change in output value would have on fuel fire behavior.

BURN POTENTIAL

Obviously, wildfires require combustible vegetation to burn. However, different fuels possess different burn characteristics. It would be grossly misleading to characterize an area as “combustible” simply because trees and grass are present. The arrangement, density, size, type and quality of combustible vegetation all play a role in the ignitability of fuels, as well as in a fire's intensity and rate of spread. Fuels must be categorized by predominant type and expected fire behavior.

Similar to the NFDRS, there already exists a widely accepted taxonomy of fuel types in the wildfire literature: the Anderson Fuel Model Classification. By separating fuels into the four broad categories of grass, chaparral and shrub, timber, and slash, Anderson provides a consistent categorization of fuel complexes that are commonly found throughout the United States. Distinguishing elements within the taxonomy, such as fuel size or height, further separates each of the four broad categories into 13 total classifications.

With the extensive expertise and assistance of fuel model experts from the Texas Forest Service, we were able to identify five predominant Anderson fuel classifications in the West Austin — Travis County study area. Based on the known fire behavior of each fuel classification, we further developed a hierarchy of risk by fuel type and mapped the results in GIS.

A very important but disturbing discovery during this process was that the predominant fuel classification in the study area is Anderson Type 4, chaparral (6 foot). The exact fuel is the Ashe juniper, commonly known as Texas mountain cedar, which is very similar to the California chaparral in size and burning characteristics. One difference does exist between the juniper and chaparral, however: Under extremely dry conditions, the Ashe juniper releases a resin that causes the shrub to burn explosively, significantly adding to the energy output of a wildfire.

Fire behavior in fuels also is largely influenced by spatial orientation. Specifically, aspect and slope play a key role in the intensity of a wildfire. Aspect is important because south-facing slopes dry out much more quickly than north-facing slopes, providing tender fuels for ignition.

Similarly, slope is important as it affects fuel arrangement. Hold a lit match sideways and it will burn slowly; take the same burning match and hold it upside-down and you will quickly scorch your finger. Like an inverted burning match, slopes arrange fuels vertically, allowing the transfer of heat and gas to occur much more quickly. The steeper the slope, the faster a fire will run. As a variable in the model, slope needs to be considered in terms of degrees.

A GIS layer was created for both aspect and slope using U.S. Geological Survey topographic maps, and hierarchies of risk were created for each based on expected fire behavior given a specific aspect or range of slope.

POTENTIAL CONSEQUENCE

The human category posed the greatest difficulty in determining which variables to consider in the model. Although there are a number of variables that influence the level of human exposure in the urban interface, the available literature doesn't shed much light on the subject. There's widespread disagreement among fire professionals on how to measure and prioritize human exposure in the interface.

Some of the models we reviewed were incredibly detailed and included a complete analysis of every structure in the interface and a complete inventory of firefighting capability/infrastructure. As our study area included more than 100,000 structures and spanned six jurisdictions encompassing more than 90,000 acres, it was impossible to think that we could approach the study with such a degree of rigor.

Furthermore, as the interface problem grows on a national scale, it's important for the industry to develop a methodology that can be used regionally, with feasible and affordable expectations in data collection and analysis. We decided that four variables would provide a balance between diluting their significance and providing an incomplete picture.

1. Defensible space

   For humans to be exposed to wildfire in the interface, they have to be close enough to combustible vegetation to be at risk.

2. Construction type

   The fire has to be able to jump from the vegetation to the home.

3. Response time

   Firefighters have to be close enough to intervene and protect structures.

4. Water supply

   Water has to be available to firefighters to make a stand.

Each variable in the human category was further divided to provide a more complete picture. Defensible space was measured in blocks of distance from the continuous canopy of fuels to the first combustible fixture, whether house, deck or shed. Construction type was split into roof material, siding material and existence of combustible accessories that could carry the fire from the fuel to the home and neighborhood.

Our logic with water supply was simple: Is there water within 1,000 feet? Because each pumper in the response area carries 1,000 feet of supply line, it was a logical conclusion to measure water supply in terms of carrying capacity of our fleet. Finally, response time was split in increments of five minutes, with any response over 15 minutes considered “extreme.”

WEIGHTED POLYGONS

Because it was beyond the capability of our staff and funding to conduct a complete analysis of every structure within the study area, we separated the entire study area into 26 GIS polygons, or mesoscale sampling areas. The polygons were based on groupings of neighborhoods that were built at about the same time in the hope that the predominant construction type would be consistent.

For each polygon, Austin firefighters conducted a random sampling of 40 structures, measuring and cataloging defensible space and type of construction. The mean value for each variable was calculated for a typical picture of what defensible space and construction looked like in each polygon. These values then were entered as layers in a GIS spatial analysis, which already included data on response time and water supply.

Once the data had been collected according to the conceptual framework, the model needed to be weighted. After presenting our model and defending the selection of variables, a panel of subject-matter experts debated the importance of each variable based on a simple question: Which one contributes most to risk in the urban interface?

The experts, including structural and wildland firefighters, foresters, habitat managers, fire ecologists, and scientists, ranked the variables after the debate, which were tabulated and averaged. The meeting facilitator presented the results and reopened the floor for debate. For each set of variables within each category, three votes and debates were allowed, creating a hierarchy of importance. We then followed the same process to give percent weights to each category of variables, resulting in the final weighting scheme for the model.

The weighting scheme is biased heavily toward the weather and spatial variables. It doesn't matter what the structure is built of or how strong the firefighting capacity is, if the weather and fuels are present in the proper combination of conditions and homes are nearby, they will burn. As a result, the human category of variables received a 20% weighting of the 1,000 available points.

GEOSPATIAL ANALYSIS

After we completed the field analysis and data collection, it was time to move everything over to geographic information systems software. For our study, we used ESRI's ArcView 3.3, Model Builder and 3-D Analyst. Model Builder was the product that allowed us to easily create and manipulate an arithmetic algorithm unique to our study.

Once we imported the data, we converted the entire study into evenly distributed grids with 300-foot cells. This allowed us to calculate total points within each grid and create a picture of risk for the study area. We began by adding the spatial and human data by grid to establish preliminary risk. Because topography, fuels and construction will not vary significantly in any 12-month period, these values are constants.

The 28-day weather analysis, however, is where we see fluctuations in overall points, as weather values change dramatically over the course of a 12-month cycle. To create a running picture of risk over time, we had to run 12 separate weather calculations with the constant spatial and human values. With the final creation of 12 monthly maps, we began to see the month-by-month picture of risk in the interface.

Based on the total 1,000-point value system, each grid within the overall map received a total value: the sum of all spatial, human and weather calculations for each month. By color-coding four ranges of values, we were able to create grades of risk for the model area. With the assistance of the Texas Forest Service, which ran a series of fire behavior models for each set of variables, we were able to define the following adjectival descriptors for each gradation in severity:

Low, 0-250

Expect-slow moving surface fuel fires only. Direct firefighting methods will easily contain these fires. Individual homes or structures may be at minimal risk.

Moderate, 251-500

Expect moderate rate of spread in surface fuels and woody species of shrubbery. Some torching of individual trees may occur. Most fires can be contained easily through direct firefighting methods. Individual homes and isolated structures may be at risk.

High, 501-750

Expect high rate of fire spread in fine fuels and moderate rate of spread in woody species of vegetation. Direct firefighting is possible on flanks; indirect is recommended on head fires. Torching is probable, with short-range spotting very likely. Individual homes, groups of homes and tracts of habitat may be at risk.

Extreme, 751-1,000

Expect very rapid propagation of flamefront with extreme rates of fire spread and high-intensity fires rapidly reaching crown. Torching and runs in crowns of trees are expected, with long-range spotting probable. Indirect firefighting methods only. Entire neighborhoods and habitat may be at risk.

THE BIG PICTURE

In addition to pinpointing Austin's areas of greatest risk should a fire occur, the maps tell a compelling story. While this model and its results provide specific direction to the fire service in terms of target areas for mitigation and education, it also demonstrates in a very graphic and easy-to-understand fashion that there really is a danger posed by the interface.

During October 2003, while southern California was burning, the Austin American-Statesman ran a front-page story on the results of the fire department's study. The Austin map was centered on the page with the headline, “Could it happen here? Where does your home lie?”

With the assistance of the fire department's GIS staff, the paper posted the map to its Web site. Residents of Austin and the surrounding area could determine their neighborhoods' risk by finding their homes and the color of their portion of the grid. Within days, the Austin interface problem had become the centerpiece of political and social debate, and the Austin Fire Department and surrounding fire districts fielded dozens of calls from concerned residents and homeowners associations.

The real result? The Austin Fire Department and surrounding jurisdictions have come together to achieve collaboration and cooperation in community education and response. Combining resources and brain power, a coalition of fire districts and Austin city agencies declared April Wildfire Awareness Month.

Using the results of the study and door-hanger brochures, firefighters from each jurisdiction canvassed the map of extreme-risk areas, driving the territory and walking the streets, communicating to citizens what they can do to become fire-wise and how they can protect themselves and their homes from the wildfire danger.

Most importantly, as community awareness has increased, the wildfire risk has become both accepted and understood, empowering departments to make substantive gains in wildfire mitigation. Without the graphic and statistical power of the risk model, it's arguable that the wildfire danger to Austin would continue to be a poorly understood and widely ignored phenomenon.

WHAT'S NEXT?

To respond to the complexities inherent in the emerging interface threat, fire departments and local authorities have to begin to use modern technology for spatial analysis to study and quantify risks. In tough economic times, calling political attention to a category of risk as poorly understood as the urban-interface hazard is difficult in the absence of scientific descriptive models.

It is hoped that other jurisdictions — local, state and federal — will pick up this model and use it to explore their own interface hazards, thereby creating a standard approach to quantifying and describing the emerging urban-interface threat. Indeed, it is arguable that no mitigation strategy to address the interface problem through model codes or otherwise will ever prove successful, either politically or practically, until fire departments and forest services can accurately describe and confidently explain the nature of the interface hazards and how it threatens local assets.

Should a disaster similar to the southern California fires occur in your interface, it's very likely that someone will point a finger your way and lament, “You knew there was a problem. What did you do about it?” The results of this study, and those like it, will give city and county leaders a powerful response to this question.

Kevin Baum is the assistant fire chief and fire marshal for the Austin Fire Department. A 20-year veteran of the fire service, Baum is a recognized leader in the fire service for his efforts in pioneering the implementation of performance measures into programs and services that are difficult to quantify and evaluate statistically. He is an adjunct faculty member of the Texas State University master of public administration program and guest lecturer for the U.S. certified public manager's program. Baum holds a bachelor of arts degree in fire service administration and a master of public policy and administration.

Christine Thies, GIS specialist for the Austin Fire Department, created the ARCView model to support this study. Her work has been recognized nationally, and she is frequently asked to present her methodology. For more information on GIS applications to the urban-interface problem, contact Thies at christine.thies@ci.austin.tx.us.

RISK MUST ACCOUNT FOR HUMAN FACTOR

The literature is overwhelmingly conclusive on one point. For a wildfire to exist, it needs a combination of the three key variables: fuels (combustible vegetation); topography (slope, aspect and relief); and weather (variations in heat, humidity, precipitation and wind).

However, it's the degree to which each of these variables are present that determines the extent of risk to an area. Any model designed to determine wildfire risk must look not just for the existence of these three key variables, but also for the degree to which they exist and how often.

For example, an area with steep slopes, heavy accumulations of dense vegetation and slash, and patterns of drought in the presence of strong winds and low humidity likely will suffer a wildfire of significant proportions. In fact, these exact conditions existed in California during last October's wildfires. Take the same area, however, and place it in the temperate rainforest of the Tongass National Forest in Alaska, where the fuels are almost always moist, and the likelihood of experiencing a wildfire is remote.

The same argument holds true for determining wildfire risk to humans. If humans and their assets aren't intermingled with wildland vegetation, there's no interface risk. However, simply adding a human variable does not complete the story. Risk to humans is a matter of degree that depends on a host of variables, such as type of construction, density of population, access and infrastructure to support firefighting operations.

Much like the three key variables in wildfire prediction, any model that proposes to determine risk in the interface has to include variables that determine the degree of human exposure. For example, a house of non-combustible construction with a defensible space of 100 feet and a landscape of succulent ornamental vegetation is likely to be safe from a wildfire, even though it sits atop a ridge surrounded by dense vegetation.

Any conceptual framework designed to model risk in the wildland-urban interface has to be scaled, so that degrees of risk in each category of variables can be measured. Without a method to scale the degree of risk, the entire study area will be given the same rating, skewing the results and offering minimal meaningful information on which to make intelligent mitigation decisions.

CONCEPTUAL RISK FRAMEWORK

Temporal Variables		Spatial Variables		Human Variables
KBDI drought cycle	11%	Slope fuel arrangement	36%	Combustible construction within GIS polygon	25%
ERC/Burn index reaction intensity	49%	Aspect mesic vs. xeric	17%	Mean response time within GIS polygon	20%
Ignition component fuel ignitability	15%	Anderson Fuel Model models 1, 2, 4, 6, 9	47%	Mean defensible space within GIS polygon	35%
1,000 hr/dfm long-term moisture cycle	25%			Water supply within 100 feet	20%
	380 points		420 points		200 points

In this framework, the temporal variables are based on current and historical climate data from NFDRS outputs. The spatial variables come from existing fuel and topographic data, and existing information on structures and infrastructure makes up the human variables. Hazard is the sum of all variables, appropriately weighted.