Model Comparisons for Predicting Grassland Fire Occurrence Probability in Inner Mongolia Autonomous Region, China
Abstract. Grassland fires threaten grassland ecosystem, human life and economic development greatly. However, there are limited researches focusing on grassland fire prediction, thereby it is necessary to find a better method to predict probability of grassland fire occurrence. Here we selected 16 environmental variables that may have impacts on fire occurrence, then built regression models of grassland fire probability based on historical fire points and variables in Inner Mongolia by three methods to identify the grassland fire drivers and to predict fire probability. The three methods were global logistic regression, geographically weighted logistic regression and random forest. The results showed that random forest model had the best predictive effect. The influence of 9 variables selected by geographically weighted logistic regression model on grassland fire was unbalanced spatially. The three models all showed that meteorological factors were of great importance to grassland fire occurrence. In Inner Mongolia, grassland fires occurred in different areas had various responses to the influencing drivers, and the areas with different natural and geographical condition had different fire prevention periods. Thus, the grassland fire management strategy based on local conditions should be advocated.
This preprint has been withdrawn.
China monthly Vegetation index (NDVI) spatial distribution dataset https://doi.org/10.12078/2018060602
Model code and software
Windows Application for Geographically Weighted Regression Modelling https://download.csdn.net/download/xiaodongfly/7027693
Viewed (geographical distribution)