Abstract: Investigating the spatial patterns of forest fires could aid in better understanding the feedbacks between fire disturbance and forest ecosystems, providing a scientific basis for making rational decisions of forest fire management. In this study, we explored the spatial patterns of forest fire regimes in China using spatial point pattern analysis, cold and hot spot analysis and landscape spatial pattern analysis. We further analyzed the spatial nonstationarity of factors influencing the pattern of forest fire regimes in China using the global Poisson regression model (GPR) and geographically weighted Poisson regression model (GWPR). Our results showed that the interquartile range of the regression coefficients in the GWPR model was greater than twice the standard error of the GPR model, indicating spatial non-stationarity in the regression relationship between forest fire regimes and influencing factors. The GWPR model had a better fitting effect (deviance, AIC value, AICc value, percent deviance explained) than the GPR model, suggesting that considering spatial non-stationarity can improve model fitting. Road density and forest coverage were the main factors influencing the frequency of forest fires, while NDVI, road density and forest cover were the main factors influencing the size of burned patches. These influencing factors exhibited significant spatial non-stationarity, indicating that the dominant environmental factors for forest fire regimes in China were regionally specific. Our results could provide a scientific basis for making region-specific forest fire management decisions.

Key words: forest fire regime, geographical weighted Poisson regression model, spatial non-stationarity