粱慧玲,郭福涛,苏漳文,王文辉,林芳芳,林玉蕊.基于随机森林算法的福建省林火发生主要气象因子分析[J].火灾科学,2015,24(4):191-200.
基于随机森林算法的福建省林火发生主要气象因子分析
Analysis of meteorological factors on forest fire occurrence of Fujian based on random forest algorithm
投稿时间:2015-05-17  修订日期:2015-07-03
查看全文  查看/发表评论  下载PDF阅读器
DOI:10.3969/j.issn.1004-5309.2015.04.02
基金项目:福建省教育厅资助省属高校专项(JK2014012);福建省自然科学基金(2015J05049);保险精算实验室建设项目(118310010)
作者单位
粱慧玲 福建农林大学计算机与信息学院福州350002 
郭福涛 福建农林大学林学院福建350002 
苏漳文 福建农林大学林学院福州350003 
王文辉 福建农林大学林学院福州350004 
林芳芳 福建农林大学计算机与信息学院福州350002 
林玉蕊 福建农林大学计算机与信息学院福州350002 
中文关键词:  气象因子  林火发生  福建省  随机森林算法
英文关键词:Meteorological factors  Fire occurrence  Fujian province  Random forest algorithm
摘要点击次数: 336
全文下载次数: 342
中文摘要:
      应用“随机森林”算法,以福建省22个国家级气象站每日气象数据和2000年~2003年间林火火点卫星解译数据为基础,对影响福建省林火发生的主要气象因子进行分析,并对2004年的林火数据进行独立检验。研究结果显示,“日最高地表气温”、“日最低地表气温”、“日照时数”、“日最高气温”和“日最小相对湿度”等5个气象因子是影响林火发生的主要驱动因子,且这5个气象因子对林火发生的影响大小为:“日照时数”>“日最高气温”>“日最低地表气温”>“日最小相对湿度”>“日最高地表气温”;随机森林算法的拟合结果显示:随机森林算法对福建省林火发生的预测精度为82.3%,表明随机森林算法对我国福建省林火发生的预测具有较高的预测能力,可用于基于气象因子的我国福建省林火发生的预测预报。本研究可为福建省林火的预测和决策工作提供一定的参考依据。
英文摘要:
      By Random Forest (RF) algorithm, an analysis of driving-factor on fire occurrence was performed based on the meteorological factors that were provided by the twenty-two national weather stations located in Fujian province and the fire dataset between 2000 and 2003 extracted from the satellite image, and used the dataset of 2004 to independent test. Daily maximum ground surface temperature, daily minimum ground surface temperature, sunshine hours, daily maximum temperature, daily minimum relative humidity were found to be the driving factors on forest fire occurrence. The importance test of predictors showed that the sunshine hours has the strong influence on the fire occurrence, followed by daily maximum temperature, daily minimum ground surface temperature, daily minimum relative humidity and daily maximum ground surface temperature. In addition, the result of model fitting revealed that RF approach performed very well in the prediction of fire occurrence in Fujian and the prediction accuracy reached 82.3%, which indicated that the RF method was suitable for the forest fire prediction of Fujian. Our study can benefit the fire prevention management and plan of Fujian.
关闭