Research On Forest Fire Drivers And Models In Heilongjiang Province

Forest fire is a highly destructive natural disaster.It not only causes irreversible damage to the ecosystem,but also threatens the lives and property of the country and people.China is a country with many forest fires.Heilongjiang Province,as an important forestry province in China,is also a high-risk area for forest fires.It is of great significance to do a good job of forest fire prevention in this region.Based on ArcGIS 10.2,SPSS 19.0 and R software,we use Logistic regression model to identify significant variables and use random forest algorithm to sort relative importance of variables.This paper use Logistic regression model and random forest algorithm to comprehensively analyze the relationship between forest fire occurrence and meteorological factors,non-meteorological factors and synthetic factors in Heilongjiang province and establish a prediction model.Based on the prediction probability of forest fire occurrence,establish a forest fire risk grade division in Heilongjiang Province.The research results provide scientific basis and theoretical reference for the prediction of local forest fires.The results of the study indicate that meteorological factors are important factors affecting the occurrence of forest fires in Heilongjiang Province.The regression analysis of meteorological factors shows that the significant meteorological factors identified by the Logsitic regression model include "daily maximum surface temperature","daily lowest surface temperature","cumulative precipitation of 24 hours","daily average station pressure","daily lowest station pressure","sunshine hours","daily maximum temperature" and "daily average relative humidity",a total of 8 variables.The important meteorological factors identified by the random forest algorithm include "daily maximum surface temperature","daily lowest surface temperature"," daily average station pressure","daily highest station pressure","daily lowest station pressure","daily average relative humidity","sunshine hours" and "daily maximum temperature",a total of 8 variables.The result of random forest variable importance ranking shows that "daily average relative humidity" is the most important driving factor.The second are the "daily maximum surface temperature" and "daily maximum temperature".The least affected variable is the "daily minimum local pressure".Compared with meteorological factors,the influence of non-meteorological factors on the occurrence of forest fire in Heilongjiang province is relatively small.The results of regression analysis of non-meteorological factors show:the significant non-meteorological factors identified by the Logsitic regression model include "Fractional Vegetation Cover","slope","distance from settlements","distance from railways","distance from road",”GDP per capita"and "population density",a total of 7 variables.The important non-meteorological factors identified by the random forest algorithm include "elevation","distance from railways","distance from settlements","distance from road","Fractional Vegetation Cover","GDP per capita","population density" and "slope",a total of 8 variables.The result of random forest variable importance ranking shows that "Fractional Vegetation Cover" is the most important driving factor,"elevation" and "GDP per capita" are the second.The least affected variable is"slope".The results of regression analysis of synthetic factors show:there are 15 driving factors in logistic regression model which are significant in a=0.05 level,and "Fractional Vegetation Cover","slope”,"distance from Railway","distance from road","GDP per capita","population density","daily maximum surface temperature","daily average local pressure","sunshine hours" and the "daily maximum temperature" are positively correlated with the occurrence of forest fire."Distance from settlements","daily lowest surface temperature","cumulative precipitation of 24 hours","daily minimum local pressure" and "daily average relative humidity" have negative correlation with forest fire occurrence.The importance of "daily average relative humidity" is ranked first in the 16 driving factors fitted by random forest algorithm,followed by "elevation","Fractional Vegetation Cover" and "daily maximum temperature",and the lowest affected variable is "slope".At the same time,the importance of meteorological factors and non-meteorological factors is distributed on an alternating basis.The two models were evaluated by the receiver operating characteristic curve,the AUC value and the classification accuracy of the model.In the meteorological factor analysis,the AUC value of RF(0.963～0.966)was higher than the AUC value of LR(0.837～0.839).The classification accuracy of RF(90.9%～92%)is higher than that of LR(75.5%～75.9%).In non-meteorological factor analysis,the AUC value of RF(0.871～0.876)is higher than the AUC value of LR(0.685～0.688),and the classification accuracy of RF(78.4%～79%)is higher than that of LR(62.9%～64.5%).In synthetic factor analysis,the AUC value of RF(0.969～0.973)is higher than that of LR(0.853～0.855),and the classification accuracy of RF(90.4%～91.2%)is higher than that of LR(76.7%-77.3%).In a general sense,except for the AUC value of non-meteorological factor analysis of LR,the AUC values of other models are all higher than 0.8.It shows that the Logistic model can not independently explain the relationship between non-meteorological factors and forest fire occurrence in Heilongjiang Province.The simulation results of the remaining models are good,but compared with the Logistic model,the random forest algorithm has a better simulation effect and a more accurate prediction accuracy.The random forest algorithm is more suitable for the prediction of forest fire in Heilongjiang province.Forest fire forecast probability map and fire risk zone map show that the area of high fire risk in Heilongjiang province accounts for 19%of the total area of the province.It is mainly distributed in Great Xing’an Mountain,Jiamusi,Shuangyashan,and most of Heihe.The fire protection strategy and management plan shall be based on the fire hazard area to strengthen the forest fire management work in the high fire risk area,taking into account other medium-risk areas and low-risk fire areas in the forest area.