Modeling spatial patterns of forest fire in Heilongjiang Province using Generalized Linear Model and Maximum Entropy Model

Forest fire distribution models are the powerful tools to map the spatial patterns of forest fire in larger scale, and to quantify the relative importance of the major factors controlling forest fire occurrence. Based on the forest fire ignition data in Heilongjiang Province in 1996-2006, and by using Generalized Linear Model (GLM) and Maximum Entropy Models (Maxent), this paper analyzed the factors controlling the forest fire occurrence in the Province, including topography, human activity, and land vegetation type, and compared the modeling accuracy, variable importance, and ignition probability map. Both the GLM and the Maxent had intermediate predictive performance, with the Maxent performed slightly better. Overall, the variables related to human activities were the most important predictors of forest fire ignition locations, followed by topographical variables. Despite the two models had similar modeling accuracy, the ignition probability map generated by Maxent was noticeably different from that generated by GLM. It was suggested that to make a comparison of or to selectively assemble different type models to produce integrated prediction results would be more desirable to more accurately identify the hotspots of forest fire occurrence, and thus, to provide more reasonable and higher efficient comments for forest fire prevention.