基于分位数回归与极值统计视角的农业巨灾风险评估研究

Evaluating agricultural catastrophic risk is an imperative practical problem of China's steady development of agriculture under the background of global climate change. Additionally, it is also an imperative scientific research problem of agricultural risk assessment. According to the previous studies, the theories and methods which have been used to evaluate agricultural catastrophic risk fall into three camps, from the perspective of risk factor, risk loss, and risk mechanism. Unfortunately, it will have misleading or biased if resolving the problems of estimating for agricultural production loss caused by extreme weather event and obtaining the probability distribution of agricultural catastrophic loss using above theories and methods. Aimed at these problems, this project will construct the basic framework of evaluating agricultural catastrophic risk based on the theories and methods of agricultural risk assessment, and will apply quantile regression, Monte Carlo simulation, Peak over threshold model, Value at risk as the basic analytical tools, and will surmount from the aspect of quantile regression model of agricultural production loss caused by extreme weather event and from the aspect of EVT model of probability distribution of agricultural catastrophic loss. On this basis, this project will take drought catastrophe in the Northeast region of China as an example and assesses grain-production risk using the theory and method proposed in this project. It is very important and urgent for this project to improve and develop the theory and method of evaluating agricultural catastrophe risk, and also to promote agricultural catastrophe risk management practices and promote the steady development of agriculture in China.

农业巨灾风险评估是在全球气候变化大背景下我国农业稳定发展急需解决的现实问题，同时又是农业风险评估学科亟待解决的科学问题。目前学术界主要从基于风险因子、基于风险损失和基于风险机理的不同视角对农业巨灾风险评估的理论和方法加以研究，但对极端性天气事件对农业生产造成的损失程度和农业巨灾损失的概率分布问题未能给出明确答案。本项目针对该问题，基于现代风险分析和评估理论，构建了一个农业巨灾风险评估的基本框架，利用分位数回归、蒙特卡罗模拟、超越阀值模型、风险值等为基本分析手段，从极端性天气事件对农作物产量影响的分位数回归模型和农业巨灾损失概率分布的极值统计模型两方面加以突破，在此基础上，针对中国东北粮食主产区粮食生产所面临的旱灾巨灾风险进行实证研究。本项目对于完善和发展农业巨灾风险评估理论和方法，进一步推进中国农业巨灾风险管理实践、促进农业稳定发展，显得十分重要和迫切。

农业巨灾风险评估已经成为全球气候变化背景下社会关注的热点和理论研究的难点。目前学术界主要从基于风险因子和基于风险损失的不同视角对农业巨灾风险评估的理论和方法加以研究，但对巨灾事件对农业生产造成的损失程度和农业巨灾损失的概率分布问题未能给出明确答案。我们针对上述问题，基于现代风险分析和评估的理论框架，进行了深入探讨。我们研究认为，农业巨灾对农作物产量的影响程度主要是由致灾因子危险度和承灾体易损度相互作用所决定的，其中致灾因子对农业生产系统产生外部冲击，其结果直接表现为农作物的受灾面积，通过分位数回归技术可以刻画出农业巨灾与处在分布尾端的农作物受灾面积之间的关系，进而通过生产函数可以反映农作物受灾面积在农业自然资源禀赋、农业生产要素投入和农业抗灾能力等内部稳定机制作用下与农作物最终产量之间的关系，从而解决了农业巨灾事件究竟会对农业生产造成多大程度损失的难题；我们尝试引入基于现代极值统计理论（EVT）的POT和BMM模型，提出了“灾害损失估计→蒙特卡洛模拟→广义极值分布拟合（广义Pareto分布拟合）→巨灾风险度量”的新方法，从而解决了农业巨灾损失究竟服从什么概率分布的难题。在此基础上，我们分别针对吉林省、江苏省和河南省粮食生产所面临的干旱和洪涝巨灾风险开展了实证案例研究，取得了很好的效果，充分显示出本研究所构建的基于分位数回归理论和基于极值统计理论的农业巨灾风险评估模型的合理性。本项目研究形成了一本有特色的农业巨灾风险评估专著，发表了一些高质量的学术论文，以期为综合管理中国农业巨灾风险以及合理指导农业生产抗灾减灾提供参考。