Fire Occurrence Probability Mapping of Northeast China With Binary Logistic Regression Model

Fire occurrence probability mapping provides a detailed understanding of the spatial distribution of the fire occurrence probability and it is useful in fire management. The binary logistic regression (BLR) can combine continuous and categorical variables together in the analysis. Here we use BLR analysis to map the fire occurrence probability of Northeast China which has the largest forest area in China. Ten predictor variables including altitude (Alt), slope (Sl), aspect (As), distance to the nearest village (Dv), distance to the nearest path (Dp), distance to the nearest water bodies (Dw), land cover (LC), Fuel Moisture Content (FMC), land surface temperature (LST) and Normalized Difference Vegetation Index (NDVI) are employed and multi-temporal random sampling methodology is used to create the training subset, and then the training subset is utilized to build the fire occurrence probability spatial model. Here, a backwards stepwise procedure based on the likelihood ratio estimation is used in the model development. Assessed by the area under a relative operating characteristic (ROC) curve (AUC–area under curve) procedure, the model's fitness accuracy is 84.2%. The interpretations of the estimated coefficients show that NDVI best explain fire occurrence in the region. Evaluated by the inner testing and independent validation, better reliability and discrimination capacity of the developed spatial model can be concluded from 17 fires among the total 18 fires. Good performance suggests that the developed model is valuable to fire managers or can be directly applied to fire management in Northeast China.