应用机器学习模型对中国云贵川区域林火风险预测  

Prediction of Forest Fire Risk in Yunnan,Guizhou and Sichuan Using Machine Learning Model

作　　者：王姊辉 董恒[1] 赵洋甬 何思聪 袁艳斌[1] 张力文 Wang Zihui;Dong Heng;Zhao Yangyong;He Sicong;Yuan Yanbin;Zhang Liwen(Wuhan University of Technology,Wuhan 430070,P.R.China;Zhejiang Spatiotemporal Sophon Bigdata Co.,Ltd.;Wuhan Huaxia University of Technology)

机构地区：[1]武汉理工大学,武汉430070 [2]浙江时空智子大数据有限公司 [3]武汉华夏理工学院

出　　处：《东北林业大学学报》2023年第5期113-119,共7页Journal of Northeast Forestry University

基　　金：国家自然科学基金项目(52079101);宁波市科技局智团创业计划(2021ZH1CXYD060013);宁波市奉化区科技计划项目(202209204)。

摘　　要：加强云贵川区域林火管理,预测云贵川地区森林火灾的发生概率,对保护森林资源和森林生态系统结构和功能具有重要意义。利用MCD64A1火灾数据集、气象数据、地形数据和可燃物含水率数据构建模型训练数据集,结合逻辑回归、随机森林与极端梯度提升等机器学习模型,对中国云贵川区域日尺度林火燃烧概率进行预测。结果表明:单独使用气象数据集对林火燃烧概率进行预测时,极端梯度提升模型表现最好,预测精度为88.6%,其次为随机森林模型与逻辑回归模型,预测精度分别为86.0%与76.7%。地形因素及植被水分指数的引入对机器学习模型有一定的优化效果,极端梯度提升模型与随机森林模型的预测精度的提升均超过1%。Strengthening forest fire management and predicting the probability of forest fires in the Yunnan-Guizhou-Sichuan region is of great importance to protect forest resources and forest ecosystem structure and function.The MCD64A1 fire dataset,meteorological data,topographic data and combustible material water content data were used to construct a model training dataset and combine machine learning models such as logistic regression,random forest and extreme gradient boosting to predict the daily-scale forest fire burning probability in the Yunnan-Guizhou-Sichuan region of China.The results show that the extreme gradient boosting model performs the best when using the meteorological dataset alone for predicting forest fire burning probability,with a prediction accuracy of 88.6%,followed by the random forest model and logistic regression model,with prediction accuracies of 86.0%and 76.7%,respectively.The introduction of topographic factors and vegetation moisture indices had a certain optimization effect on the machine learning model,and the improvement in prediction accuracy of both the extreme gradient boosting model and the random forest model exceeded 1%.

关 键 词：森林火灾 植被水分指数 机器学习 随机森林 极端梯度提升 

分 类 号：S762.2[农业科学—森林保护学]