Collaborative Research: Multimodel Bayesian Data-Worth Analysis for Groundwater Remediation Design

地下水污染补救包括去除地下污染物，使其不会对人类和环境构成不可接受的未来风险。虽然已经开发了各种修复方法，但它们都面临着一个共同的挑战，即地下水环境复杂，在有限的时间和资源下无法充分了解和表征地下水环境。因此，鉴于地下水环境特征的不确定性，挑战是设计一种具有最大成功概率或同等最小失败概率的补救战略。补救失败的一个常见原因是忽视了模型的不确定性。使用单一模型进行补救设计可能会导致对模型预测能力的过度自信，从而增加失败的概率。这项拟议的研究将使用基于多模型的数据价值分析来重新审视模型不确定性下的补救设计问题。换句话说，多个模型被用来识别和指导最有价值的数据的收集，以用于模型评估、改进和重建。由于模型、修复设计和数据之间的综合，所提出的修复设计的多模型数据价值分析将为科学家、工程师和决策者提供一个系统地研究地下水修复所涉及的所有组成部分的变革性平台。该项目还将为水文学、计算科学和土木工程领域的本科生和研究生提供跨学科培训的机会。此外，该项目将使暑期学校的高中教师和学生获得实验室和计算经验，以了解地下水污染物迁移和补救的概念。提出的研究有两个目标：考虑模型不确定性的地下水修复数据价值分析，以及打破修复设计所需的模型和模型分析之间的计算障碍。为实现第一个目标，将把数据价值分析纳入多模型分析框架(也称为模型平均)，该框架将发展成为与多模型数据价值分析相兼容的补救设计的新程序。为了实现第二个目标，将开发一个准确但评估成本低的模型替代模型，然后用于不确定情况下的数据价值分析和补救设计。将使用贝叶斯方法(理论和计算)来实现这两个目标。虽然所提出的多模型贝叶斯数据价值分析方法是通用的，可以应用于任何修复方法，但它将与最近开发的工程注入和提取方法相结合，这是一种很有前途的原位修复技术。建议的方法将以双管齐下的策略进行评估，使用合成和真实世界的建模问题。现实中的问题包括科罗拉多州纳图里塔遗址的铀污染，以及佛罗里达州印第安河县的氮污染。综合研究将尽可能地模拟真实世界的问题，以便从综合研究中获得的见解可以直接用于真实世界的建模。该项目将为两个实地地点以及其他受污染地点正在进行的环境补救和监测提供科学支助。

Groundwater contaminant remediation involves removing subsurface contaminants so that they do not pose unacceptable future risks to humans and the environment. While various remediation methods have been developed, they all face a common challenge that the groundwater environment is complex and cannot be fully understood and characterized with limited amount of time and resources. Hence, given the uncertainty in the characterization of the groundwater environment, the challenge is to design a remediation strategy that has the maximum probability of success or equivalently the minimum probability of failure. A common reason for remediation failure is ignoring model uncertainty. Using a single model for remediation design may lead to overconfidence in the predictive capability of the model and thus to increased probability of failure. The proposed research will reexamine the problem of remediation design under model uncertainty by using a multimodel-based data-worth analysis. In other words, multiple models are used to identify and guide the collection of the most valuable data for model evaluation, improvement, and reconstruction. Because of the synthesis between models, remediation designs, and data, the proposed multimodel data-worth analysis for remediation design will provide a transformative platform for scientists, engineers, and decision-makers to systematically investigate all components involved in groundwater remediation. This project will also provide an opportunity for interdisciplinary training of undergraduate and graduate students in the areas of hydrology, computational science, and civil engineering. In addition, the project will engage high school teachers and students in summer schools to gain laboratory and computational experience for understanding the concepts of groundwater contaminant transport and remediation. The proposed research has two objectives: to reformulate data-worth analysis for groundwater remediation with consideration of model uncertainty, and to break computational barriers between models and model analysis needed for remediation design. To achieve the first objective, a data-worth analysis will be integrated into a framework of multimodel analysis (also known as model averaging), which will be developed into a new procedure for remediation design that will be compatible with the multimodel data-worth analysis. To achieve the second objective, an accurate but cheap-to-evaluate surrogate of the models will be developed and then used for the data-worth analysis and remediation design under uncertainty. The Bayesian approaches (theoretical and computational) will be used for achieving both the objectives. While the proposed method of multimodel Bayesian data-worth analysis is general and can be applied to any remediation method, it will be integrated with the recently developed engineered injection and extraction method, a promising technique for in-situ remediation. The proposed methods will be evaluated in a two-prong strategy using synthetic and real-world modeling problems. The real-world problem involves uranium contamination at the Naturita Site, Colorado, and nitrogen contamination at the Indian River County, Florida. The synthetic study will mimic the real-world problem to the extent possible so that insights gained from the synthetic study can be used directly for the real-world modeling. This project will provide scientific support for on-going environmental remediation and monitoring at the two field sites as well as other contaminated sites.