Prediction Of Grassland Fire Occurrence Probability Based On Geographically Weighted Logistic Regression Model

Grassland fire is an important influence factor of grassland ecosystem,which can be used rationally to serve human production and life,but at the same time,grassland fire is also one of the important disasters that destroy grassland resources and ecological balance,endanger the safety of life and property of grassland people,and threaten the sustainable development of national economy.Therefore,it is necessary to build a grassland fire probability prediction model that can prevent the occurrence of grassland fires and reduce the loss of grassland fires.This study proposes to use the daily fire product data of MODIS 1km resolution imaging spectra,global land surface assimilation system data,vegetation monitoring data of MOD13A3 vegetation index product,digital elevation data,and human impact index data provided by NASA(https://earthdata.nasa.gov/),with the help of R,MRT,Arc GIS,GWR4.0 and other software,through descriptive statistics methods,to analyze the interannual and monthly fluctuations and spatial distribution of grassland fires in Inner Mongolia,and to analyze the change trend;here On this basis,a probability prediction model for grassland fire occurrence was further established to obtain the spatial distribution of grassland fire occurrence probability in Inner Mongolia,so as to realize the spatial dynamic prediction of grassland fire risk,and provide technical support for the pre-disaster safety management of grassland-related fire emergency management departments.The following conclusions were drawn from the study:(1)Grassland fires in Inner Mongolia occur frequently and the spatial and temporal distributions are quite different.The number of fire points has been increasing before 2014,reaching the maximum value in 2014,and then showing a downward trend.The occurrence time of grassland fires is mainly in spring and autumn;the spatial characteristics of grassland fire distribution are more in the east and less in the west,more in the south and less in the north.(2)In this paper,Geographically Weighted Logistic Regression(GWLR)and global Logistic Regression(LR)were used to establish probability prediction models for grassland fire in Inner Mongolia.The comparison results show that the AUC value of GWLR model is significantly higher than that of LR model,the AICc value of GWLR model is significantly lower than that of LR model,and the prediction accuracy of GWLR model is significantly better than that of LR model.(3)The results based on the full-sample GWLR model show that the influence of each factor on grassland fire occurrence has strong spatial distribution characteristics.Among them,the daily maximum temperature and human influence index have significant positive effects on grassland fire occurrence in the whole study area,with the coefficients of daily maximum temperature on grassland fire occurrence ranging from 0 ～ 0.176 and human influence index on grassland fire occurrence ranging from 0 ～ 0.383;the daily minimum specific humidity has significant negative effects on grassland fire occurrence in the whole study area,with the coefficients ranging from-0.027 ～ 0.(4)The occurrence probability of grassland fire in Inner Mongolia has obvious spatial characteristics.The occurrence probability of grassland fire is low in the west and high in the east;while in the central region of Inner Mongolia,the occurrence probability of grassland fire increases from north to south.Among them,Wuhai city is special,its grassland fire occurrence probability is obviously higher than other surrounding areas.