| Wildfire threats to homes are increasing in the wildland-urban interfaces in the Western United States. However, little research has been done in community planning issues related to wildfire risk mitigation and systematic evaluation of risk factors based on fire experiences. The objective of this research is to move this process forward by studying a wildfire-burned site integrating high-resolution remote sensing data and GIS data. Research conceptualizes that the wildfire vulnerability of a house is a characteristic of its built environment which is inextricably linked to the natural environment, socioeconomic conditions and planning policies. Accordingly, independent variable are: vegetation density, defensible space, road width, road type, parcel size, adjacency of a parcel to public land, morphology, proximity of a house to a fire station, land value, elevation, slope, and aspect. The dependent variable is the probability of a house burning whose response is dichotomous---1 for burned houses and 0 for unburned houses. Data are extracted from IKONOS imageries taken before and after fire, parcel data, and Digital Elevation Models (DEMs) and analyzed in logistic regression. Seven variables are significant in univariate analysis, out of which vegetation density, defensible space, proximity of a house to a fire station and slope are significant in the regression model. Model performance is tested by Receiver Operating Curve (ROC) where Area under Curve (AUC) is 0.91. Research findings are congruent with widespread wildfire mitigation practices on: (1) vegetation management; (2) defensible space requirements; and (3) slope. Univariate analysis results emphasizes the need for further study on the relationship between the probability of a house burning and (1) proximity of a house to a fire station; (2) adjacency of parcel to public land; (3) morphology; and (4) road width. The research also recommends the inclusion of roofing material, house density and boundary walls in future studies. Findings of this research add our understating on how the probability of a house burning changes when a unit-change is made in independent variables. The method applied in this research demonstrates a process of integrating high-resolution remote sensing and GIS data for planners and policy makers to focus in the evaluation of wildfire risk and to develop mitigation measures. |
