Spatial point analysis of fire occurrence and its influence factor in Huzhong forest area of the Great Xing'an Mountains in Heilongjiang Province, China

Forest fire is a key natural disturbance in shaping forest landscape pattern and dynamics, affecting tree species composition and age structure. Therefore, understanding spatial pattern of fire disturbance and its influencing factors is integral to adaptive forest management. Fire is a complex process, influenced by various environmental controls at different scales. The relative influence of environmental controls on fire occurrence can vary spatially and temporally. At the regional and continental scales, spatial pattern of fire is mainly influenced by ignition, biome distribution, and climate. At the landscape scale, spatial pattern of fire is mainly influenced by ignition, vegetation, and topography. At the stand scale, spatial pattern of fire is mainly influenced by fuel, micro-topography and weather. Because forest fire management is often conducted at the landscape scale, we focused on this scale in our study. Using the spatial point process analysis, this study examined the spatial pattern of fire occurrence and its influencing factors in Huzhong forest area of the Great Xing'an Mountains in Heilongjiang province, China during 1990-2005. A spatial point process (e.g., Poisson process) is any stochastic mechanism that generates spatial pattern of point locations. Reported fire occurence locations, recorded as geographically referenced spatial points, were used as a dependent variableand were mapped using GIS. Abiotic (e.g., elevation, aspect, and slope), biotic (e.g., vegetation type), and human factors (e.g., Euclidean distance to nearest road, Euclidean distance to nearest settlements) were used as explanatory variables (spatial covariates). The fire occurrence was modeled as inhomogenerous Poisson process. The residual analysis and AIC were used to determine the optimal inhomogenerous Poisson models that include different sets of spatial covariates (with transformation). A maximum pseudolikelihood method was used to estimate the coefficients of each spatial covariates. The results indicated that fire occurrence is not random but spatially clustered. There are some hotspots (i.e., areas with high fire occurrence density) as well as a few coldspots with low occurrence density across the landscape. The fire occurrence density map showed a spatial trend from southwest to northeast. The burned probability ranged from 0.004- 0.012/(km²·a), with average burned probability is 0.0077 /(km²·a) for the study area. Spatial point process analysis showed that distance to nearest settlement and road, elevation, slope, and forest type were the main influencing factors. The results are consistent with previous studies that human-related factors, topography and vegetation are the primary drivers for modern fire regimes, although their relative influence varies. Current forest fire management for this landscape has mainly focused on reducing human activities that may lead to fire ignitions. Our results suggested that, in addition to human activities, influences of topography and vegetation type on fire occurrence should also be considered in the future fire risk management.