Wildland Arson Crime Functions

1
Wildland Arson Crime Functions

• David T. Butry
• National Institute of Standards and Technology
• Gaithersburg, MD

Jeffrey P. Prestemon Southern Research
Station USDA Forest Service Research Triangle
Park, North Carolina
and
2
Introduction

• There are 500,000 arson fires/year (wildland plus
  structural) in the US, 3 billion in damages
  (National Fire Protection Association).
• Wildland arson is the leading single cause of
  wildfires in Florida.
• Arson ignitions on national forests have trended
  down over the past 1-2 decades, as have all
  causes.
• Area burned by accidental fire starts has trended
  upward over time, apparently, in aggregate,
  although arson area burned has not trended.
• Few have rigorously evaluated the underlying
  causes of short- or long-term temporal patterns.

3
Number of Ignitions by Fire Source on National
Forests
4
Area Burned by Ignition Source on National
Forests (FS Protection)
5
Crime and Arson

• It is apparent that arson is following patterns
  similar to major crimes committed in the U.S.
• Recent research shows that wildland arson is
  similar to violent crime in its response to law
  enforcement, criminal sanctions, and economic
  variables.

6
Crime Trend Nationwide, Plus Wildland Arson on
National Forests
7
Changes in Crime in Florida, 1972-2004
8
Todays Presentation

• Provide background information on Floridas arson
  situation
• Outline our econometric models
• Report results
• Describe implications for fire forecasting

9
Why is Learning About Arson Important?

• Arson fires threaten large values
• More often in the WUI
• Arson wildfires are part of a larger ecological
  process
• Behave similarly in response to management,
  weather, fuels
• Evidence suggests that arson fires appear to be
  clustered in both time and space.

10
Background on Arson

• Wildland arson has a long history
• Especially in the South
• Florida has over 1,400 arson ignitions and 45,000
  arson-ignited burned acres per year
• Wildland arson is linked to demographic factors
• Old research quantifying the role of law
  enforcement
• Research identifying some links to socioeconomic
  factors

11
Background on Arson

• Recent research links Arson to physical factors
• Arson fires follow other fires in timing
• Mostly during fire season (January-July)
• Peak in ignition rate in mid-afternoon
• Are more common in dry weather
• Respond to previous wildfires in the area
• But arson fires differ from others
• More ignitions on weekends
• Concentrated in spatial distributionperhaps,
  closer to roads and urbanized areas

12
Arson wildfire theory

• Serial and copycat arson behaviors imply a
  contagion process.
• Current arson could be explained by previous
  arson ignitions.
• Other research identifies these behaviors for
  other kinds of crimes.
• Law enforcement may play a role.
• Floridas number of police officers per capita
  increased 12 between 1982 and 2001 but has
  declined by 2 since 1995 trends vary by county.
  
• Much recent research identifying a negative
  relationship between law enforcement and crime.
• Weather and land management may affect it.
• Dry weather makes firesetting easier
• Fuels management can affect success rates and
  opportunities
• Leisure time could help explain it.
• Socioeconomic factors should explain some of it.
• Population level should be relatedmore people,
  more arsonists?
• Poverty has been linked to other crimes. Arson
  models should control for this.
• Labor factors might explain itwages,
  unemploymentaffecting crime opportunity costs
  (Becker, new research in AER, elsewhere).

13
Crime Model (Becker)
The decision to commit a crime is described as
Oi is the number of offenses committed pi is the
probability of being caught and convicted fi is
the wealth loss experienced by the criminal if
caught and convicted ui measures other factors
influencing the decision and success of
completion of the crime
14
Arsonists Expected Utility from a Successful
Ignition (Becker)
Oi is the number of offenses committed pi is the
probability of being caught and convicted gi is
the arsonists psychic and income benefits from
illegal firesetting ci is the production cost
for the firesetting fi(Wi,wi) is the loss from
being caught and convicted of the crime is a
positive function of income while employed Wi is
the employment status wi is wage
15
TERM DEFINITION FUNCTION OF
? Probability of being caught Law enforcement
f Loss from being caught and convicted Wage rate Employment status
c Production cost of firesetting Time available Unemployment status Fuels and weather Variables related to ignition success
g Psychic and income benefits from illegal firesetting
16
Arson Poisson Autoregressive Model

• PAR(p) Daily Ignition Model

yj,t is a vector of daily arson ignitions for
location j xj,t is a vector of independent
variables (including a constant) ßj is a vector
of associated parameters ?j,is are the
autoregressive parameters
17
Empirical Models

• County-level daily time scale Poisson
  Autoregressive models of order p, PAR(p)
• Five high-arson county pairs in Florida
• 1994-2001
• Locational daily time scale PAR(p) with
  spatio-temporal components
• Six high-arson Census tracts in Florida
• 1994-2001
• Annual fixed-effects cross-section time series
  panel Poisson model
• Most Florida Counties
• 1994-2001
• California national forests daily time scale
  PAR(p)
• 1993-2002

18
Study Locations
Spatio-temporal Analyses
19
Daily Time Series Model Spatio-Temporal Analysis

• The PAR(p) relates current days fires to
• Previous days fires,
• Presence of neighboring arson
• Localarson in surrounding Census tracts
• Regionalarson in Census tracts in same and
  surrounding counties
• Long-term annual wildfires in the area (1-12 yr),
  
• Prescribed fire permits in the area (0-2 yr
  lags),
• Current fire danger index (KBDI),
• Seasonal factors days of the week, months
• Socioeconomic factors population, full-time
  equivalent police officers per capita, poverty
  rate

20
Data

• Wildfire and prescribed fire from the Florida
  Division of Forestry
• Socioeconomic data
• U.S. Bureau of the Census
• University of Florida-Bureau of Economic and
  Business Research
• Florida Department of Law Enforcement
• Climate and weather from NOAA

21
Daily Locational Model Results

• Broadly significant variables
• (significant across 3 or more models)
• Previous ignitions (up to 4 days)
• Previous local ignitions (up to 11 days)
• Previous regional ignitions (up to 4 days)
• KBDI
• Some months
• Previous wildfire area (up to 5 years)

• Significant variables
• (significant across 1 or 2 models)
• Weekend days
• Poverty rate
• Unemployment rate
• Retail Wage
• Police
• Some months
• Previous prescribed fire

22
Daily Pooled Model Results

• Significant variables
• Previous ignitions (up to 10 days)
• Previous local ignitions (up to 11 days)
• Previous regional ignitions (1 day)
• KBDI
• Saturday
• Most months
• Previous prescribed fire (up to 1 year)

• Insignificant variables
• Sunday
• Population
• Poverty rate
• Unemployment rate
• Retail wage
• Previous wildfire

All variables interacted with population except
AR terms, local ignitions, and regional
ignitions.
23
Daily Model Results Daily Autocorrelations
24
Simulated Outbreak Response

• Assume one unexpected arson ignition occurred on
  April 30, 2005
• Analyze using the pooled model results and with
  continuous variables set at the pooled model
  means
• Examine variation in response when outbreak
  occurs at different locations
• Same Census tract
• Local Census tract
• Regional Census tract

25
SimulationResponse of an unexpected arson
ignition on April 30, 2005.
Day after outbreak
26
Response to Outbreak

• 15.7 additional arson ignitions when outbreak
  occurs in same Census tract
• 18.3 additional arson ignitions when outbreak
  occurs in a neighboring local Census tract
• 17.6 additional arson ignitions when outbreak
  occurs in a neighboring regional Census tract

27
We Also Evaluated Effects of Law Enforcement
Saturation Strategies

• Ongoing work is seeking to develop hot-spotting
  models for law enforcement

28
Summary

• We have extended results from newly published
  work in AJAE wildland arson, at least in
  Florida, is spatially and temporally
  autoregressive.
• Hence, wildland arson is a predictable process
  after an ignition occurs, potentially allowing
  for successful and effective law enforcement
  action.
• Also implies that ignitions should be modeled
  that recognizes at least temporal and probably
  spatio-temporal autocorrelation (depends on the
  spatial scale of modeling) within daily time
  frames.

29
Questions
30
Law Enforcement Saturation

• Given an outbreak, examine how varying levels law
  enforcement saturation affects future arson
• Levels of saturation are consecutive days,
  following the outbreak, of arson prevention
• Saturation supposes perfect ability to control
  arson ignitions (i.e., when theres saturation,
  no arson ignitions occur)

31
Law Enforcement Saturation
32
Effect of Saturation

• Although an outbreak can have long-lasting
  effects (several weeks), eleven days of
  saturation prevents any new arson ignition
• Saturation has different effects depending on
  locational source of the outbreak (significance
  of differences across neighboring locations not
  evaluated)
• On average, the following number of ignitions
  are prevented for each day of saturation
• 1.4 if outbreak occurred in same Census tract
• 1.7 if outbreak occurred in neighboring local
  Census tract
• 1.6 if outbreak occurred in neighboring
  regional Census tract

33
Law Enforcement Implications

• Focus enforcement on locations with recent and
  nearby arson fires.
• Concentrate enforcement where arson fires have
  been ignited in last ten days.
• Concentrate enforcement around where arson fires
  have been ignited in last 2 days.
• Pay attention to weather trends.
• Periods of hot, dry weather associated with
  higher arson risk
• Perhaps this is associated with the success of
  ignition, lower expected time and effort needed
  to obtain a successful ignition.
• There is a Saturday effect.
• Count on Saturdayslower opportunity costs of
  firesetting?
• This result is consistent with an economic model
  of crime, at least for this variable.

34
Fire Management Implications

• The use of prescribed fire is not found to be
  associated with lower arson risk
• Locations with lots of wildfire are at lower
  arson ignition risk.
• As other ignition risks, arson risk is closely
  tied to time of year and fuel flammability.