Forest Fire Risk Mapping by Kernel Density Estimation

Abstract

When evaluating wildland fires, well prepared forest fire risk maps are regarded as one of the most valuable tools for forest managers, and during the production stage of these maps, association between historical fire data and other factors, such as topographic, anthropogenic and climatic, are often required. One of the most encountered problems in forest fire risk analyses is the fact that historical fire data, the dependent variable, are generally in point format, whereas other factors, the independent variables, are often expressed in areal units and available in raster format. Kernel density estimation is a widely preferred method for converting historical fire data into a continuous surface. In this study, kernel density estimate of forest fire events in the Middle East Technical University (METU) campus, in Ankara, Turkey, between 1993 and 2009, were obtained by using different bandwidth choices. Kernel density maps with regard to seasons and years were also produced and the final result was expressed as mean density value in each polygon of the study area. Actions that should be taken in high-risk areas were given on the basis of the results obtained.