欢迎访问林草资源研究

林草资源研究 ›› 2023›› Issue (5): 56-62.doi: 10.13466/j.cnki.lczyyj.2023.05.007

• 科学研究 • 上一篇    下一篇

广西林火驱动因子及预测模型研究

巨文珍(), 韦龙斌(), 彭泊林, 李常诚, 潘婷   

  1. 广西壮族自治区林业勘测设计院,南宁 530011
  • 收稿日期:2023-08-14 修回日期:2023-10-14 出版日期:2023-10-28 发布日期:2023-12-20
  • 通讯作者: 韦龙斌,高级工程师,主要从事森林防火、森林资源监测研究。Email:271687731@qq.com
  • 作者简介:巨文珍,高级工程师,主要从事森林防火、森林资源监测研究。Email:249753114@qq.com
  • 基金资助:
    广西自筹经费林业科技项目“广西森林可燃物载量特征及林火发生规律研究”(2023GXZCLK69);广西林业设计院自选科技项目“广西森林可燃物特征及林火发生规律研究”(林勘科研[2022]01-03)

Study on Driving Factors and Prediction Model of Forest Fire in Guangxi

JU Wenzhen(), WEI Longbin(), PENG Bolin, LI Changcheng, PAN Ting   

  1. Forestry Survey and Design Institute of Guangxi Zhuang Autonomous Region,Nanning 530011,China
  • Received:2023-08-14 Revised:2023-10-14 Online:2023-10-28 Published:2023-12-20

摘要:

了解林火最主要的驱动因子并对林火进行预测,能为当地森林火灾的预防与管理提供科学依据。基于2011—2020年的历史火灾数据集,以及气象、地形、人为活动和可燃物载量等数据构建Logistic回归模型和机器学习模型来探究广西林火发生最主要的驱动因子,同时选择最优模型对研究区内森林火灾发生概率进行预测。研究表明:月平均降雨量、月平均相对湿度和林区建筑物数量是影响广西森林火灾发生最显著的因子;Logistic回归模型和机器学习模型均取得了较好的拟合效果,AUC值均在0.85以上,机器学习模型的精度要优于Logistic回归模型,随机森林模型精度最高(SAUC=0.92)。通过随机森林模型对全区林火发生概率进行预测,结果显示桂西北、桂北、桂西南地区的林火发生风险最大,预测结果契合广西实际,能够为广西的林火预测预报提供参考。今后,应加强对野外火源的管控力度并提高对极端天气的预警防范能力,以降低森林火灾发生的风险。

关键词: 林火驱动因子, 林火概率预测, 机器学习模型, logistic回归, 广西

Abstract:

Understanding the main driving factors of forest fires and predicting the occurrence of forest fires can provide scientific basis for the prevention and management of local forest fires.Based on the historical fire data sets,meteorology,topography,human activities and combustible load data from 2011 to 2020,we built Logistic regression models and machine learning models to explore the main driving factors of forest fires in Guangxi.Additionally,the optimal model was selected to predict the probability of forest fire occurrences in the research area.The result showed that average monthly rainfall,average monthly relative humidity and the number of forest buildings were the most significant factors affecting the occurrence of forest fires in Guangxi.Both Logistic regression model and machine learning model had excellent prediction accuracy,with AUC values above 0.85.The accuracy of the machine learning model was better than Logistic regression model,and the random forest model had the highest accuracy(SAUC=0.92).The random forest model showed that the risk of forest fires was greatest in northwest,northern and southwestern regions of Guangxi.The overall results of the model were consistent with the actual situation in Guangxi and can provide a reference for forest fire prediction and forecasting in Guangxi.In the future,managers should strengthen the control of wild fire sources and prevention capabilities of extreme weather to reduce the risk of forest fires.

Key words: forest fire driving factors, forest fire probability prediction, machine learning model, logistic regression, Guangxi

中图分类号: