基于复合极值分布理论的重大灾害多要素联合概率预测及风险分析

In recent years, the frequency and losses of serious natural disaster events both increased significantly. This issue becomes one of the most important factors that constraints of regional economic sustainable development. How to assess the joint probability of multi-hazards and risk of serious disasters more accurately is one of the key issues to slow or reverse the continuing growth trend of disaster risk. This problem is still not solution because of some restrictions of multivariate joint distribution method. This project bases on multivariate compound extreme value distribution theory, takes serious marine disasters in coastal zone and near shore areas for example, analysis of synergy mechanism and probability characteristics and nonlinear correlation structure of multi-hazards in different development stages of serious disaster events. In order to forecast the joint probability of multi-hazards and risk of serious disasters more accurately, build multivariate compound extreme value distribution model which both considerate the discrete distribution of small sample of serious disaster events and the continuous joint distribution of multi-hazards. This study made up the defects of usual methods that cannot perform disaster mechanisms comprehensively and strengthen the multi-variable methodology of multivariate analysis in disaster mechanism study. The multivariate analysis method will contribute to improve the accuracy of quantitative description of risk and lay the theoretical and technical foundation of deeper applications in the risk analysis of floods, earthquakes, tsunami and other natural disasters.

近年来重大自然灾害发生频率、相应损失均明显增加，成为制约区域经济可持续发展的一个重要因素。如何准确评估多要素协同作用下的重大灾害发生概率和风险大小，是减缓或扭转灾害风险持续增长趋势的前提基础。目前多变量联合分布的研究方法受诸多不足的限制，尚不能很好解决这一问题。本项目基于多维复合极值分布理论，以海岸带及近岸区域重大海洋灾害为例，分析灾害事件在多要素协同作用下的成灾机理，明确多要素随灾害发生发展不同阶段的非线性非对称相关结构形态及联合概率特征，构建综合考虑灾害事件离散型分布和多要素连续型联合分布特征的多维复合极值分布模型，从而更加准确地评估要素极值的发生概率和灾害风险大小。该研究可以克服以往常用方法在全面表现灾害机理上的缺陷，从理论上为加强灾害机理研究中小样本多变量极值分布的方法论和提高风险概率定量化的精度做出贡献，该方法为在洪水、地震、海啸等重大自然灾害风险研究应用奠定理论和技术基础。

如何准确评估多要素协同作用下的重大灾害发生概率和风险大小，是减缓或扭转灾害风险持续增长趋势的前提基础。本项目以我国辽东湾地区海冰灾害和沿海大型工程等典型承灾体为例，分析了重大海冰灾害的主要破坏机制和成灾模式，得出在不同成灾模式中的主要致灾要素，分析了主要致灾要素间的相关结构和相关性变化特征。基于连续多年的观测数据，对冰厚、结冰范围和冰速等致灾要素进行了边缘分布拟合，根据不同Copula函数的适用区间，构建了多维混合Copula函数模型，进行了海冰灾害多维致灾要素的联合分布和重现期计算，并对重大海冰灾害风险进行了定量分析和表达。. 本项目研究得出，单变量重现期会使估值偏大，与结冰范围相比，最大冰厚的取值对于海冰灾害风险的影响更大。对于二维联合概率，在低值区的拟合效果要好于高值区，而三维联合概率拟合，对高值区拟合效果较好。因此对于一般强度的灾害事件，可以构建二维的联合概率分布进行风险分析，而对于极端事件的研究，出于极端值拟合精度的要求，则需要考虑构建多维的联合概率分布进行分析。. 该项研究通过多维非线性联合概率理论，更加准确反映了多要素在不同组合不同相关结构下的危险性特征，在海洋灾害风险评估研究中怎么实现多维度和定量化这些关键问题的解决上起到了有效的推动作用。该项目相关成果对于防灾减灾政策的制定和大型工程类海冰灾害设防标准的构建具有很大的参考价值，为加强海洋灾害机理和风险研究中多变量分析方法的应用做出贡献，也为这种新方法在海洋灾害风险分析中进行更深一步的延伸奠定了基础。