Influences on Wildfire Hazard Exposure in Arizona's High Country

Abstract

Based on the case of wildfire hazards in Arizona forests, this article addresses the question: What influences hazard exposure? Like other locales in the U.S. West, the study area has developed as large wildfires have occurred with increasing frequency. Management interventions have traditionally been based on the hypothesis that unsafe conditions result from inadequate residential knowledge of wildfire hazards. Findings from a household-level multiple regression analysis using structured survey, hazard exposure, and secondary data provide little support for this approach and underlying hypothesis. Results reveal that other variables—corresponding to amenity values, reliance on fire insurance, place dependency, and housing contextual factors—are important predictors of household exposure to wildfire hazards. Findings have implications for theoretical understandings of wildfire hazards and hazard reduction efforts in private community landscapes.

Keywords:

• Arizona
• environment
• forest
• hazard
• risk
• vulnerability
• wildfire

Notes

^a A = strucutred household survey; B = Navajo County (Arizona) parcel data; C = in-field property assessment.

Note. HH, household; w/o, without. Significance indicated by: ∗∗∗p < .001 (two-tailed); ∗∗p < .05 level (two-tailed); ∗p < .05 level (1-tailed).

^a Reference category is single-family residential.

Budgetary constraints necessitated this approach.

The property assessment was used to construct the dependent variable, “hazard exposure.” It is a continuous variable scaled 0–40 that gives equal weight to the four independent measures of wildfire hazard exposure. Home structure hazard is the sum of ignitability measures for roof (1), structural setback (1), siding (1), decks (1), eaves (1), windows (1), skylights (1), and window exposure extent (1). Defensible space hazard is the sum of exposure measures for roof vegetation (1) and roof debris (1), chimney vegetation (1), deck vegetation (1), 0.6 m perimeter flammability (1), 3 m surface perimeter flammability (1), near home landscaping (2), 3 m home perimeter flammable tree presence (1), and isolation from forest canopy (1). Property landscape hazard is the sum of crown fire potential measures for number of flammable trees/acres (<30 = 0, 30.01–60 = 1, 60.01–100 = 2, 100.01–200 = 3, >200.01 = 4), crown closure (1), branches below 3 vertical m (1), small tree ladders (1), accumulation of dead branches and downed trees (1), surface fuel separation (1), dead, diseased and damaged trees (1), surface-canopy ladders (1), grass near flammable materials (1), 8 inch landscape grass height (1), vegetation hydration (1), and 1 inch surface pine needle accumulation (1). Fire suppression hazard is the sum of fire suppression capacity measures for emergency driveway width (1), driveway vegetation clearance (1), address visibility (1), roadway vegetation clearance (1), road allowance for two-way traffic (1), road accessibility for emergency vehicles (1), road sign postage (1), and street naming and numbering (1).

Dave Ostergren from the Ecological Restoration Institute at Northern Arizona University allowed the author to use seven items related to knowledge about ecological restoration in ponderosa pine forests from a survey he developed. Respondents were asked to generate true/false responses to the following statements (in “true” form): restoration reduces fire hazard; ponderosa pine forests are fire-dependent; prescribed fire is a restoration tool; restoration benefits wildlife; restoration helps reestablish native plants; large fires result in part from fire suppression; and removing most pine needles reduces fire hazard. Responses were then converted to correct/incorrect and an ecological knowledge score (number correct/number of questions) was calculated.

Preferences for “environment” include three more specific survey items: shade, dense forests, and climate/weather. Preferences for “cost” and “privacy” were each constructed from individual survey items. Preferences for “social relations” include three survey items: proximity to work, close to schools, and close to family. Preferences for “aesthetics” include three items: good views, attractive home, and attractive lot. Preferences for “property fire safety” include four items: fire safe roof, defensible space, landscaping, and fire history. Preferences for “fire suppression capability” include four items: quality fire fighting, wide roads, visible road signage, and emergency water supply.

The specific fire safety items included as self-protection measures correspond to hazards associated with: the home structure (remodeling exterior walls = 1, remodeling roof = 1, removing/rebuilding decks = 1, and installing dual-paned windows = 1); the defensible space (moving combustibles = 1, clearing flammable vegetation within a 3-m perimeter = 1, regularly removing pine needles from roof = 1, and regularly raking most pine needles = 1); the property landscape (regularly trimming trees = 1, regularly watering trees = 1, removing trees = 1, re-landscaping with fire-safe plants = 1, and regularly mowing grass = 1); and fire suppression (maintaining an easily accessible driveway = 1, an easily identifiable address = 1, and a water supply for fire-fighting purposes = 1). The household “mitigation” measure is the sum of home structure, defensible space, landscape, and fire suppression hazard adjustments that were implemented during the time that households occupied their home sites, while the “prevention” measure is the sum of those wildfire hazard adjustments that had been implemented before respondents occupied their home sites.

Multiple diagnostic tests were used for multicollinearity; test results indicate that inferences from the regression are not affected by multicollinearity problems. First, the variance inflation factor (VIF) value reflects the presence or absence of multicollinearity. A high VIF indicates multicollinearity of the explanatory variables. A general rule is that the VIF should not exceed 10; a VIF value higher than 10 means the variable is likely affected by multicollinearity (Belsley et al. 1980). This cutoff point is above the highest VIF value (i.e., 4.00 for housing tenure) found in the regression analysis. Second, the higher the intercorrelation of the independent variables, the more the tolerance value will approach zero. As a rule of thumb, a problem with multicollinearity is indicated if tolerance is less than .20. This cutoff point is below the lowest tolerance value (i.e., .25 for housing tenure) found in the regression. Third, the collinearity diagnostics table in SPSS is an alternative method of assessing if there is too much multicollinearity. High eigenvalues indicate dimensions (factors) that account for much of the variance in the crossproduct matrix. Eigenvalues close to 0 indicate dimensions that explain little variance. Multiple eigenvalues close to 0 indicate that there may be a problem with multicollinearity and that the condition indices should be examined. Two factors in my analyses have eigenvalues less than .05, prompting my examination of condition indices. Condition indices are used to flag excessive collinearity in the data. A condition index (CI) over 30 suggests possible collinearity problems. The collinearity diagnostics for my regression analyses revealed one factor with a CI above 30 (31.353). If a factor has a high CI, one looks in the variance proportions column to see if that factor accounts for a high proportion of variance in two or more variables (i.e., if two or more variables are most heavily loaded on that factor). If this is the case, these variables have high linear dependence and multicollinearity is indeed a problem. The variance proportions for the factor of concern did not indicate multicollinearity problems in the regression analyses (the highest variance loadings were .18 and .13). Thus, according to VIF, tolerance, and condition index criteria, inferences from the regression appear to have not been affected by multicollinearity problems.