基于LightGBM的台风直接经济损失评估与预测——以福建省为例  

Assessment and Prediction of Typhoon-Related Direct Economic Loss in Fujian Province Based on LightGBM

作　　者：张之夏 杨剑 陈思孝 林森 Zhang Zhixia;Yang Jian;Chen Sixiao;Lin Sen(School of Economics and Management,China University of Geosciences,Wuhan 430078,China;School of Civil and Environmental Engineering,Harbin Institute of Technology,Shenzhen 518055,China;National Disaster Reduction Center of the Emergency Management Department,Beijing 100124,China)

机构地区：[1]中国地质大学(武汉)经济管理学院,武汉430078 [2]哈尔滨工业大学(深圳)土木与环境工程学院,深圳518055 [3]应急管理部国家减灾中心,北京100124

出　　处：《热带地理》2025年第4期648-659,共12页Tropical Geography

基　　金：国家重点研发计划(2024YFC3016800);国家资助博士后研究人员计划B档(GZB20230966)。

摘　　要：台风是中国沿海地区最严重的自然灾害之一,常导致重大经济损失。准确评估和预测台风直接经济损失对于提升防灾减灾能力和优化资源配置至关重要。文章以福建省84个区县为研究对象,基于2009—2021年影响福建省的30场台风灾害数据,结合致灾因子、孕灾环境因子和承灾体暴露度因子,共计20个关键影响因子,采用LightGBM方法构建台风直接经济损失预测模型,对台风风险进行定量评估并通过实际案例探讨模型在台风实际直接经济损失动态预测中的适用性。模型重要性分析表明,日最大风速、河网密度、日最大降水、累计降水、单位面积GDP和城市化率是影响福建省台风直接经济损失的主要因素。文章构建的模型在训练集上Pearson相关系数R达到0.836、可决系数R2达到0.66,通过4个超强台风案例验证模型性能,预测直接经济损失与实际的损失相关系数在0.6~0.71,表明模型具有较好的应用潜力。以超强台风“莫兰蒂”为例,利用所建模型开展动态预测应用,结果显示模型较好地模拟了台风动态变化过程中直接经济损失的动态分布变化,可为福建省及其他沿海地区的台风灾害损失评估和应急管理提供科学支持。Typhoons are among the most destructive natural disasters affecting China's coastal regions,often resulting in substantial economic loss and casualties.The annual average Direct Economic Loss(DEL)caused by typhoon disasters in China exceeds 60 billion yuan,accounting for 10%-30%of the DEL caused by all disasters each year.Consequently,the accurate assessment and prediction of typhoon-induced DEL are essential for improving disaster mitigation strategies and optimizing resource allocation.Rapid development of artificial intelligence and the growth of multi-source spatiotemporal big data have introduced data-driven methods for assessing disaster losses.These methods have the advantage of using large samples to improve adaptability and consider more risk factors.In this study,DELs of 30 typhoon events in Fujian Province at the county level and a total of 911 samples were collected from 2009 to 2021 to establish an assessment model.Owing to the large range of the DEL in different districts and counties during the same typhoon,the logarithm of the DEL was used as the model output.This study included three steps for constructing the model.First,24 influencing factors of typhoons,including disaster-inducing factors,disaster-forming environmental factors,and disaster-bearing body exposure factors,were calculated using the Pearson correlation coefficient and variance inflation coefficient to analyze the multicollinearity effect,and 20 key factors were selected to assess the DEL.Second,a LightGBM-based model is developed using the selected indicator factors as model inputs.Of the 911 samples,734 were used to train the model,and 177 were used for validation.Finally,Super Typhoon Meranti was used as a case study to evaluate the applicability of the model in the dynamic DEL assessment of a typhoon.This study evaluated predictive performance of the model using five indicators:the Pearson correlation coefficient(R),coefficient of determination(R2),mean squared error,mean absolute error,and median absolute error.The importance o

关 键 词：台风灾害 直接经济损失 机器学习 风险预测 LightGBM 福建省 

分 类 号：P429[天文地球—大气科学及气象学] X43[环境科学与工程—灾害防治]