基于BMA的降低地下水污染场址评估不确定性研究

There exists significant uncertainties for the results of risk assessment due to parametric uncertainty and model uncertainty during the periods of groundwater contaminant site investigation and remediation. Most of the previous studies focused on the parametric uncertainty, however, the conceptual model uncertainty is usually omitted. In this study, we establish a muti-model analysis framework based on Bayesian Model Averaging (BMA) coupling with the groundwater contaminant transport simulation model, which is effective for reducing model and parametric uncertainties for the results of risk assessment. Meanwhile, the new sampling data-worth is quantified by updating the prior probabilities of the parameter distributions according to the Bayesian theory. The developed framework is validated by a groundwater contaminant site located in the Yanziji Area, Nanjing. This study could provide technical support for the decision-making on the sampling design and the remediation plan.

在地下水污染场址调查和整治等各阶段，参数取值和概念模型的不确定性往往导致污染场址的风险评估结果存在较大的不确定性。已有的研究较多针对参数的不确定性开展，而对概念模型的不确定性研究很少。本项目拟针对水文地质结构的不确定性，建立基于贝叶斯模型平均的多（概念）模型分析框架，通过将其与污染运移模型进行耦合，降低评估结果的不确定性，同时根据贝叶斯理论，通过采样获取的新增信息更新空间变异性参数的先验概率分布，量化不确定性的降低程度及由采样获取的新增信息的数据价值。选择南京市燕子矶地区某地下水污染场址，对已建立的贝叶斯平均多模型框架进行实例验证。研究成果可为污染场址的调查采样设计及污染场址整治决策方案的制定提供有力技术支撑。

根据项目计划书安排，经过三年研究，构建了基于贝叶斯模型平均的多模型分析框架，研发建立了污染场址风险不确定性降低程度及采样数据价值的量化方法，开发了一套基于BMA的污染物运移模拟的多模型分析代码，设计了理想案例进行了应用，为量化采样行为对降低污染场址风险评估不确定性程度和制定污染场址的修复整治方案提供可靠高效的新工具。针对水文地质参数反演的难题，拓展性的开展了基于向导点-正则化方法的探讨和分析。在实际污染场址上，开展采样行为数据价值的量化工作和污染场址风险评估的不确定性降低程度的分析工作，能为后续的修复整治提供科学依据和决策支持。