Just released analyses by USDA Forest Service researchers revealunderlying patterns in wildland arson. Research forester Jeff Prestemonand economist David Butry, both from the FS Southern Research Stationeconomics unit at Research Triangle Park, NC, have developed a modelthat can help law enforcement agencies better predict where and whenfires might be set in wildland areas and adopt strategies to reduce therisk of arson.
Over 1.5 million fires are set by arsonists each year in the UnitedStates, resulting in over $3 billion in damages. Arson is a leadingcause of wildfire in several heavily populated states, includingCalifornia and Florida. Since wildland arsonists often set fires nearhomes and roads, arson wildfires cause a disproportionate amount of thedamage attributed to wildfire in general. Several recent largewildfires were intentionally set, including the Hayman fire near Denverin 2002, which caused damages exceeding $100 million.
Criminals often commit multiple crimes in a short time frame, adocumented phenomenon that researchers call temporal clustering.Temporal clustering of arson events can also be a result of copycatbehavior. Property and violent crimes often have a spatial clusteringcomponent, also, with crime concentrated in neighborhoods. In a pair ofstudies, Prestemon and Butry set out to test whether spatio-temporalclustering also appeared in wildland arson. Their research, whichdescribed the firesetting process in the context of the economics ofcrime, also examined whether socioeconomic factors are related towildland arson in ways identified by criminologists.
"Even though the economic damages created by wildland arsonists areoften staggering, research into the factors that contribute to it hasbeen limited to a few published studies," says Prestemon. "Models ofwildland arson have mostly related firesetting to weather, seasonaltrends, and law enforcement, ignoring the role of the socioeconomicvariables used to predict other types of crime."
"At the same time, no one has been able to identify the spatio-temporaldimensions of wildland arson that we found," declared colleague DavidButry. "Our findings have uncovered a new avenue of fire research,deepened our understanding of arsonists' behaviors, and revealedanother way in which humans and society interact with our environment."
In one study, the researchers evaluated wildland arson as both anannual and a daily process. Their annual model revealed the influenceof law enforcement, fuels, poverty, and labor conditions on the ratesof ignitions observed in all of the counties in Florida during theperiod 1995 to 2001. The researchers uncovered the fine temporalpattern of arson by analyzing arson in nine high-arson countiesscattered across the state over the same time frame, but sliced intodays instead of years. This model identified temporal clustering thatlasted up to 11 days--implying higher risk of repeat arson ignitionsfor 11 days following the initial fire. This kind of pattern had neverbefore been found in any research into human-ignited wildfires.
In the second study, Prestemon and Butry measured the spatial as wellas temporal clustering of arson wildfires using daily information forthe six Census tracts in Florida with the greatest arson activity.Their statistical results showed that an arson event in one Censustract was related to arson in the previous two days in neighboringtracts and the previous ten days in the same tract.
"So not only did we confirm our finding from the previous study, thatarson clusters in time, but we also showed that it clusters in spaceover time," adds Butry. "In other words, arson events in one tract maybe used to predict future ignitions in the same or adjacent tracts forseveral days." When the two studies are taken together, a principalfinding is that wildland arson has spatio-temporal clustering similarto patterns found for other types of crime.
"Combining this principal finding with our data on law enforcement ledto our second finding, that there are strategies that law enforcementcan use to prevent wildland arson," says Prestemon. "They can closelymonitor areas where fires have been set before. They can also increasearson enforcement on days of the year when events are more common, andduring droughts."
A third finding was that locations with difficult economicconditions--low wages and high poverty rates--have higher rates ofwildland arson, which is consistent with an economic model of crime.Fourth, forest management activities are also related to wildlandarson, with fuel reductions from prescribed burning and other wildfirescorrelated with lower arson rates. "This finding is also consistentwith an economic model of wildland arson crime, where lower fuelsincrease the cost of successfully starting fires," says Prestemon.
The next step is to test the model in other locations to see if thestatistical results hold true outside of Florida and to conduct newresearch into arsonists' behavior. "While we have examined and foundsimilar firesetting patterns in California, we think we need to directsome of our effort into understanding who arsonists are and what,precisely, makes them behave as they do," says Butry. "If we could addfeedback from convicted wildland arsonists, we could really enhance ourunderstanding of how they choose where and when to set fires.Incorporating this information into the model would raise itspredictive value to law enforcement."
"Time to Burn: Modeling Wildland Arson as an Autoregressive CrimeFunction," by Jeffrey P. Prestemon and David T. Butry was published inthe August 2005 issue of the American Journal of AgriculturalEconomics, and is available in full text at http://www.srs.fs.usda.gov/pubs/20631
"Spatio-temporal wildland arson crime functions," by David T. Butry andJeffrey P. Prestemon, was presented at the Annual Meeting of theAmerican Agricultural Economics Association, July 26-29, 2005,Providence, Rhode Island, is available in full text at http://www.srs.fs.usda.gov/pubs/9537
Materials provided by Southern Research Station - USDA Forest Service. Note: Content may be edited for style and length.
Cite This Page: