森林是碳库,具有强大的固碳增汇功能,在应对气候变化中发挥着重要作用.然而,由于极端高温的影响,频繁发生可燃物自燃而引发森林火灾,除了影响区域水文大气循环过程以外,也给人类带来严重的人员伤亡和经济损失.现有森林火灾预测研究主要侧重可燃物研究和火灾监测等方面,较少关注大尺度地形、气象和人类活动对森林火灾的影响,但这些也是除可燃物外导致森林火灾发生的主要因素.以嘉陵江流域重庆段为研究区,区域内山地受自然火灾影响严峻.基于地理信息系统叠加地理空间因子与火灾分布点获得数据集,构建 4 种机器学习模型,测试模型性能,评价最优模型进行森林火灾灾害风险制图.研究结果表明,模型评估指标受试者工作曲线下面积(area under the curve,AUC)平均值为95.0%,模型性能梯度提升决策树最优,AUC值为98.3%.利用梯度提升决策树(gradient boosting decision tree,GBDT)模型预测森林火灾风险对防范大尺度森林火灾具有一定的可行性,对山城避灾规划起到借鉴作用,规划引导降低森林火灾风险,从而维护生态平衡和生态系统碳汇能力.
Watershed-scale forest fire risk prediction based on machine learning
Forests are crucial for sequestering and storing carbon to mitigate climate change.However,forest fire caused by spontaneous combustion of fuel and extreme temperatures not only disrupt the regional hydrologic and atmospheric cycle but also cause severe casualties and economic losses.Existing forest fire prediction research has primarily focused on fuel research and fire monitoring,neglecting large-scale terrain,weather,and human activities that are also the main factors leading to forest fire.This study centered on the Jialing River Basin of Chongqing where mountainous regions were severely affected by natural fire.Based on the geographic information system(GIS)superimposed geospatial factors and fire locations to obtain the dataset,four machine learning models were constructed and tested,and the optimal model was evaluated for forest fire disaster risk mapping.Results showed an average AUC of 95.0%,with the gradient boosting decision tree outperforming the optimal model with an AUC of 98.3%.The GBDT model used in this study demonstrated the feasibility of predicting forest fire risk to prevent large-scale forest fire.These findings can inform disaster avoidance planning and guide efforts to reduce forest fire risk in mountain cities and other regions,thereby maintaining the carbon sequestration capacity of nature and ecosystems.
forest firemachine learning(ML)gradient boosting decision tree(GBDT)disaster risk mappingplanning for mountain city disaster mitigation
NETL
NSTL
万方数据
