摘要
【目的】建立四川省凉山彝族自治州(简称凉山州)林火预测模型,为该地区林火防控和管理决策提供参考。【方法】基于凉山州2001-2015年MCD14DL的火灾数据集,采用核密度评估和空间自相关分析揭示火灾时空变化规律;选取多年平均气温、多年平均降水量、高程、坡度、坡向、植被类型、归一化植被指数、土地利用类型、人口密度、距最近道路距离等10个林火驱动因子,采用随机森林(RF)算法探究各林火驱动因子的影响程度并构建林火预测模型,利用受试者工作特征(ROC)曲线对模型精度进行评价,并完成对凉山州的火险区划。【结果】凉山州林火年际变化波动较大,月季变化显示林火集中发生在12月到次年5月,占全年火灾的96.86%;林火在空间上呈极显著集聚分布,中高和高核密度区域主要集中在德昌县、盐源县、会理县、西昌市、会东县和木里县,二者总面积占比为7.84%。RF模型结果表明,多年平均气温、多年平均降水量、人口密度、高程、归一化植被指数和坡度是林火发生的重要驱动因子,其中气象因子贡献最大;预测模型的ROC曲线下面积(AUC)值为0.873,约登指数为0.495,模型预测精度达到78.70%。依据RF模型,凉山州低风险火险区面积占比为72.61%,中、高风险火险区面积高达27.39%。【结论】凉山州中、高风险火险区域主要集中在南部和中部,少部分位于东北部,其中会理县、德昌县、宁南县、会东县、盐源县和西昌市是高风险林火频发区域,应重点关注这些区域,强化火灾防控措施与管控能力。
【Objective】 A forest fire prediction model for Liangshan Yi Autonomous Prefecture,Sichuan was established to provide reference for forest fire prevention and control and management decision in this area.【Method】 Based on the MCD14DL fire data set from 2001 to 2015 in Liangshan Prefecture,the spatial and temporal variation of fires was revealed by kernel density assessment and spatial correlation analysis.Ten forest fire driving factors including average annual temperature,average annual precipitation,elevation,slope,slope direction,vegetation type,normalized difference vegetation index,land use type,po-pulation density and distance from the nearest road were selected.Random forest(RF) algorithm was used to explore the impact degree of each factor and build a forest fire prediction model.The model accuracy was evaluated by receiver operating characteristic(ROC) curve,and the fire risk zoning was completed.【Result】 The inter-annual variation of forest fires in Liangshan Prefecture fluctuated greatly,and forest fires mainly occurred from December to May,accounting for 96.86% of total fires.The forest fires were concentrated in Dechang,Yanyuan,Huili,Xichang,Huidong and Muli,with a total area of 7.84%.The RF model showed that average annual temperature,average annual precipitation,population density,elevation,normalized difference vegetation index and slope were important driving factors of fire occurrence,and mete-orological factors contributed the most.The area under ROC curve(AUC) value of the prediction model was 0.873,the Yoden index was 0.495,and model prediction accuracy reached 78.70%.According to the RF model,low risk risk areas in Liangshan Prefecture accounted for 72.61%,and medium and high risk areas were 27.39%.【Conclusion】 The middle and high risk fire risk areas in Liangshan Prefecture mainly concentrated in the south and central parts,with small areas in the northeast.Huili,Dechang,Ningnan,Huidong,Yanyuan and Xichang are high risk areas with frequent fires.Attention should be paid to these areas to strengthen fire prevention and control measures and capabilities.
作者
张甫桓
张斌
罗君
左自泉
ZHANG Fuhuan;ZHANG Bin;LUO Jun;ZUO Ziquan(School of Geographical Sciences,China West Normal University,Nanchong,Sichuan 637009,China;Sichuan Provincial Engineering Laboratory of Monitoring and Control for Soil Erosion in Dry Valleys,China West Normal University,Nanchong,Sichuan 637009,China;Liangshan Soil Erosion and Ecological Restoration in Dry Valleys Observation and Research Station,Xide,Sichuan 616753,China)
出处
《西北农林科技大学学报（自然科学版）》
CSCD
北大核心
2024年第11期48-60,共13页
Journal of Northwest A&F University(Natural Science Edition)
基金
国家自然科学基金项目(41971015)。
关键词
林火预测
林火驱动因子
火险区划
林火防控
凉山彝族自治州
forest fire prediction
forest fire driving factors
fire risk zoning
forest fire prevention and control
Liangshan Yi Autonomous Prefecture
