Study On Driving Factors And Forecast Models Of Forest Fire In Large-scale Areas

In view of the status that the research on forest fire driving factors and forecast models in China is usually limited to small-scale regions,this study uses 2010-201 8 national satellite monitoring hot spot data,2010-2017 daily meteorological data,basic geographic data and DEM data to analyze the distribution and the driving factors of forest fires in large-scale areas in China with statistical analysis,logistic regression,random forest algorithm and other methods.Then,according to the results of driving factors selection,a forecast model of forest fire occurrence in large-scale areas in China was established with logistic regression and random forest algorithm,the two models were evaluated with ROC,AUC value and the classification accuracy.Finally,this study drawn the forest fire danger grade distribution map in China based on the more accurate model,hoping to provide a reference for the forest fire prevention department in China.The main conclusions of the study are as follows:(1)Distribution Characteristics of Forest Fires in ChinaThe number of forest fires' occurrences from 2010 to 2018 showed a decreasing trend.There were obvious seasonal and regional differences in the number of forest fires in China,which the number of forest fires in spring and winter was significantly higher than that in summer and autumn,and the number in southeast China was significantly higher than that in other regions.In terms of meteorology,the number of forest fires initially increases and then decreases with the increase of average daily temperature,average daily relative humidity and average daily wind speed.In terms of topography,the number of forest fires decreases with the increase of altitude and slope,and that on sunny slope is more than that on shade slope.In terms of human factors,the number of forest fires decreases with increasing distance from roads,and initially increases then decreases with increasing distance from residential areas.(2)Selection and Analysis of Forest Fire Driving factors in Large-scale Areas in ChinaThis study selected and analyzed the main forest fire driving factors in large-scale areas in China based on logistic regression and random forest algorithm.Using multicollinearity diagnosis and logistic regression significance test,there were 15 forest fire driving factors selected,among which longitude,average air pressure,daily sunshine hours,average daily wind speed,maximum daily wind speed,fire point-road distance,fire point-residential area distance and special festivals had positive correlation with forest fire occurrence.The seven factors of latitude,altitude,slope,average daily surface temperature,20:00-20:00 accumulated precipitation,average daily relative humidity and minimum daily relative Humidity have negative correlation with forest fire occurrence.The forest fire driving factors selected by random forest algorithm and their order of importance were:latitude,minimum daily relative humidity,longitude,average air pressure,average daily temperature,average daily surface temperature,average daily relative humidity,daily maximum surface temperature,altitude,20:00-20:00 accumulated precipitation,daily sunshine hours and daily maximum air temperature.The main driving factors of forest fires selected by the both two models were longitude,latitude,altitude,average daily surface temperature,20:00-20:00 accumulated precipitation,average air pressure,average daily relative humidity,minimum relative humidity and daily sunshine hours.These 9 factors were considered to be the main driving factors of forest fires occurrence in large-scale areas in China.(3)Forest Fire Forecast Model in Large-scale Area in ChinaBased on the results of forest fire driving factors selection,Using logistic regression and random forest algorithm established forest fire occurrence forecast model for large-scale areas in China.The results showed that the model prediction accuracy of using random forest algorithm was better than logistic regression,and was more suitable for forest fire forecast in large-scale areas in China.(4)Forest Fire Risk Grade Distribution in ChinaBased on the prediction occurrence probability of forest fires with random forest model,used ArcGIS to draw the map of forest fire risk grade distribution in China.As can be seen from the map,the forest areas in northeast China,southeast China and parts of Yunnan province are high-risk area,and the forest fire risk rating in north China and central China are relatively low,while there is basically no fire in most parts of the west.