《保险研究》20260104-《基于多模态数据的长三角城市群洪涝灾害风险机制分析》(张连增、罗来娟、肖宇谷)

[中图分类号]F842.64[文献标识码]A[文章编号]1004-3306(2026)01-0043-18 DOI:10.13497/j.cnki.is.2026.01.004

资源价格:30积分

  • 内容介绍

[摘   要]洪涝灾害是全球最常见且影响广泛的自然灾害之一,准确预测其发生与否具有重要意义。本文从保险精算视角出发,以长三角城市群为例,融合多模态数据,采用Logistic回归、随机森林和LightGBM三种方法,对洪涝灾害的发生情况及其风险机制进行了系统分析。研究结果表明:第一,综合比较三种方法,LightGBM具有最优的预测性能。第二,基于SHAP(SHapley Additive exPlanations)的可解释性分析结果显示,相比于脆弱性因素而言,洪涝灾害受危险性因素的影响更大。第三,洪涝灾害风险呈现显著的空间自相关性,将空间滞后项纳入模型可以有效提升模型的预测精度。第四,随着气候情景恶化,未来洪涝灾害风险有可能进一步加剧,且局地极端性突出。相关研究结论可为洪涝巨灾保险的精算定价和风险管理提供科学依据和技术支撑。

[关键词]洪涝巨灾保险;多模态数据;莫兰指数;空间滞后项;SHAP

[基金项目]国家社会科学基金重点项目“巨灾债券定价与风险管理的统计建模研究”(22ATJ005)、教育部人文社会科学重点研究基地重大项目“数字时代风险管理与精算模型研究”(22JJD910003)的支持。

[作者简介]张连增,南开大学南开-泰康保险与精算研究院教授、博士生导师,研究方向:精算与风险管理、精算数据科学;罗来娟(通讯作者),南开大学金融学院博士研究生,研究方向:精算数据科学;肖宇谷,中国人民大学应用统计科学研究中心研究员,中国人民大学统计学院教授,研究方向:精算数学、量化风险管理、农业保险。


Risk Mechanism Analysis of Urban Flood Disasters in the Yangtze River Delta Based on Multimodal Data

ZHANG Lian-zeng,LUO Lai-juan,XIAO Yu-gu

Abstract:Flood disasters are among the most frequent and impactful natural hazards globally,and accurate prediction of their occurrence is essential.This study,from the perspective of actuarial science,takes the Yangtze River Delta urban agglomeration as a case and integrates multimodal data to analyze flood occurrence and its risk mechanisms.Logistic Regression,Random Forest,and LightGBM methods are employed for comparative analysis.The results show that:(1) LightGBM outperforms the other methods in predictive performance;(2) SHAP (SHapley Additive exPlanations) analysis indicates that hazard indicators have a greater influence on flood risk than vulnerability indicators;(3) the flood risk exhibits significant spatial autocorrelation,and introducing spatial lag terms improves model performance;(4) under more severe climate scenarios,future flood risk is projected to further exacerbate,characterized by increasingly localized and extreme events.The findings provide a scientific basis and methodological support for actuarial pricing and risk assessment in flood catastrophe insurance.

Key words:flood catastrophe insurance;multimodal data;Moran’s Index;spatial lag term;SHAP