热带海洋学报

• •    

基于贝叶斯网络和GIS的热带气旋灾害风险评估

田城1, 黎鑫1, 杜洋1,李明1,谢勇2,夏际炉3   

  1. 1.  国防科技大学气象海洋学院, 湖南 长沙 410005;

    2.  江西省气象局, 江西 南昌 330038;

    3.  95903部队气象台, 湖北 武汉 430000

  • 收稿日期:2022-04-29 修回日期:2022-07-12 出版日期:2022-07-19 发布日期:2022-07-19
  • 通讯作者: 黎鑫
  • 基金资助:

    国家自然科学基金项目(41605051); 国家自然科学基金项目(41976188); 国防科技大学重点科研计划(ZK17-02-010)

Tropical Cyclone Disaster Risk Assessment Based on Bayesian Network and GIS

TIAN Cheng1, LI Xin1, DU Yang1, LI Ming1, Xie Yong2, Xia Jilu3   

  1. 1. College of Meteorology and Oceanography, National University of Defense Technology, Changsha 410005, China;

    2. Jiangxi Meteorological Bureau, Nanchang 330008, China;

    3. 95903 Troop Meteorological Observatory, Wuhan 430000, China

  • Received:2022-04-29 Revised:2022-07-12 Online:2022-07-19 Published:2022-07-19
  • Contact: Li
  • Supported by:
     National Natural Science Foundation of China (41605051); National Natural Science Foundation of China (41976188); Key Research Program of National University of Defense Technology (ZK17-02-010)

摘要: 针对热带气旋灾害的复杂性和不确定性,本研究基于贝叶斯网络和地理信息系统(GIS)提出了一种新的热带气旋灾害风险评估模型。该模型能够从客观历史数据中自动挖掘灾害影响因素间的因果关系,并以概率形式进行表达,从而对不确定灾害风险进行评估预测。基于1980-2016年中国东南沿海三省(广东、福建、浙江)的热带气旋灾害历史数据进行风险评估实验,选取致灾因子危险性、孕灾环境敏感性、承灾体脆弱性三个方面共计12个评估指标作为模型输入,直接经济损失量化为灾害风险等级作为模型输出,进行贝叶斯网络模型构建和参数训练,发现该模型的准确率高达87.68%。在此基础上,利用2017-2021年热带气旋灾害数据进行模型检验,评估预测的准确率为80.75%。模型预测的极低、低、中、高和极高风险的相对误差分别为27.72%,8.45%,18.58%,16.52%,19.12%,风险预测值的区划结果在空间形态上与实际灾害损失分布高度一致。这表明,本研究建立的热带气旋灾害风险模型具有较高的准确率和可信度,为热带气旋灾害风险评估提供了一种新的方法途径和技术支撑。

关键词: 热带气旋灾害, 风险评估, 贝叶斯网络, GIS

Abstract: To address the complexity and uncertainty of tropical cyclone disaster, this study proposes a new tropical cyclone disaster risk assessment model based on Bayesian network and Geographic Information System (GIS). The model can automatically mine the causal relationships among disaster influencing factors from objective historical data and express them in the form of probabilities to assess and predict uncertain disaster risks. Based on the historical data of tropical cyclone disaster in three southeastern coastal provinces (Guangdong, Fujian, and Zhejiang) of China from 1980 to 2016 for risk assessment experiments, a total of 12 assessment indicators in three aspects, namely, hazard of disaster-causing factor, sensitivity of disaster-inducing environment, and vulnerability of disaster-bearing bodies, were selected as model inputs, and direct economic losses were quantified as disaster risk levels as model outputs for Bayesian network model construction and parameter. The accuracy of the model was found to be as high as 87.68%. On this basis, the model was tested using tropical cyclone disaster data from 2017-2021, and the accuracy of the assessment prediction was 80.75%. The relative errors of very low, low, medium, high and very high risks predicted by the model were 27.72%, 8.45%, 18.58%, 16.52% and 19.12%, respectively, and the zonal results of risk prediction values were highly consistent with the actual disaster loss distribution in terms of spatial patterns. This indicates that the tropical cyclone disaster risk model established in this study has high accuracy and credibility, which provides a new methodological approach and technical support for tropical cyclone disaster risk assessment.

Key words: tropical cyclone disaster, risk assessment, Bayesian network, GIS