应用机器学习算法分析广西林火发生驱动因素及林火预测

Application of Machine Learning Algorithms in Analyzing the Driving Factors of Forest Fire Occurrence in Guangxi and Prediction of Forest Fires

森林火灾对生态环境和人类社会构成了严重威胁,当前全球气候不断变化、人类活动日益频繁,森林火灾的影响也日益凸显。以广西壮族自治区为研究区,根据2006—2020年研究区的卫星监测森林火点数据,结合气象数据、地形数据、植被数据和人为活动数据,应用反向传播神经网络(BPNN)、梯度增强决策树(GBDT)、随机森林(RF)、支持向量机(SVM)和极限梯度提升模型(XGBoost)等机器学习算法对广西地区的森林火灾建模,对林火发生概率进行预测;利用最优模型绘制了季节性森林火险区域图,分析森林火灾发生的驱动因素和潜在的森林火灾风险。结果表明:(1)XGBoost模型在预测广西地区森林火灾风险方面表现最佳,其准确率为92.33%,精确度为92.89%,召回率为91.88%,F_(1)值为92.38%,A_(UC)值为97.68%。(2)广西地区森林火灾的主要驱动因素为气象条件与植被因素,主要因素为潜在蒸发量(Pes)、大气压(Sfp)、总初级生产力(GPP)和增强型植被指数(Evi)等。(3)广西地区的春季和冬季是森林火灾的高发季节,中高风险区主要集中在桂东、桂中南和桂西地区。

Forest fires pose a serious threat to the ecological environment and human society.With global climate continuously changing and human activities becoming more frequent,the impact of forest fires is increasingly prominent.Taking Guangxi Zhuang Autonomous Region as the study area,satellite-monitored forest fire point data from 2006 to 2020 were used,along with meteorological,topographic,vegetation,and anthropogenic activity data.Machine learning algorithms such as Back-Propagation Neural Network(BPNN),Gradient Enhanced Decision Tree(GBDT),Random Forest(RF),Support Vector Machine(SVM),and Extreme Gradient Boosting(XGBoost)were applied to model forest fires in Guangxi and predict the probability of fire occurrence.The optimal model was used to generate seasonal forest fire risk zone maps,analyze the driving factors of forest fire occurrence,and assess potential fire risks.The results showed that:(1)The XGBoost model performed best in predicting forest fire risk in Guangxi,with an accuracy of 92.33%,precision of 92.89%,recall of 91.88%,F_(1) value of 92.38%,and A_(UC) value of 97.68%.(2)The primary driving factors of forest fires in Guangxi were meteorological conditions and vegetation factors,including potential evapotranspiration(Pes),atmospheric pressure(Sfp),gross primary productivity(GPP),and enhanced vegetation index(Evi).(3)Spring and winter were the peak seasons for forest fires in Guangxi,with medium to high-risk areas mainly concentrated in Gui Dong,Gui Zhongnan and Gui Xi regions.