| Forest fires can have devastating consequences on the environment and human life.Predicting forest fires is of great importance in preventing them from happening.This article compares the applicability of four machine learning methods(Logistic regression model,LR;Gompit regression model,GR;Random forest model,RF;Boosted regression tree model,BRT)in predicting forest fires in Daxing’an Mountains of Inner Mongolia using historical fire data from 1981-2020 and multiple sources(meteorological conditions,terrain,vegetation,human activity and socio-economics)data.The best prediction model was found to be Boosted Regression Tree model.The fire data was divided into four periods(1981-2020;March 14,1981-March 15,1988;March 15,1988-2008;2009-2020)based on China’s old and new Forest Fire Prevention Regulations1.The study attempted to explore the impact of key historical events on the driving factors of forest fires in Daxing’an Mountains of Inner Mongolia by establishing a probability model for forest fire occurrence and drawing maps of possible fire occurrence and fire risk level1.The main research results are as follows:1.Meteorological factors are the main factors affecting forest fires in the Daxing’an Mountains of Inner Mongolia.They mainly include daily temperature difference and minimum relative humidity.Studies have also found that some meteorological elements during the autumn fire prevention period of the previous year,such as the average temperature and surface temperature of the autumn fire prevention period,will affect the occurrence of forest fires in the second year.Altitude in topography also significantly affects the occurrence of fires.Human activities and socio-economic factors also have some impact on fire occurrence,but their influence is not as high as that of the first two factors.The high fire risk areas in the Daxing’an Mountains of Inner Mongolia are mainly concentrated in the east and southeast.The northern part of the study area(the original forest area in the north of Daxing’an Mountains of Inner Mongolia)under foreign fire invasion(Sino-Russian border)and lightning fire impact,and the southwest under foreign fire(Sino-Mongolian border)also have a certain risk of fire.The prediction effect and fitting degree of BRT model and RF model are better than those of other two regression models.BRT model is the most suitable forest fire prediction model for Daxing’an Mountains of Inner Mongolia.The prediction ability of LR model and GR model is similar.When the proportion of fire points to non-fire points is 1:1.5,the fitting degree of LR model is higher.2.From the modeling results of different periods,the modeling effect of 1981-2020(all periods)is not better than that of 2009-2020(recent period),which shows that it is very important to choose the appropriate modeling time series length and combine key historical events when modeling forest fires in the future;altitude and distance from the observation tower are determined as important predictive factors affecting fire occurrence;there are obvious changes in the high-risk areas of fires in three periods(March 14,1981-1988;March15,1988-2008;2009-2020),and the eastern and southeastern parts of the study area have high fire risks in all three periods,while the original forest area in the north has fewer highrisk areas before the formulation of the old and new “Forest Fire Prevention Regulations”,which has changed significantly to a high-risk area after the revision of the new “Forest Fire Prevention Regulations”.The results of the study can provide theoretical support for the selection of future forest fire prediction models and optimize forest fire management strategies for forest fire prevention and suppression departments and emergency management departments.It also helps to understand the drivers and fire risk distribution of forest fires in the Daxing’an Mountains of Inner Mongolia under the influence of key historical events,and helps decision makers to optimize fire management strategies to reduce potential fire hazards. |
PDF Full Text Request