Multi-modal, shape-based Inverse Methods for the Characterization of DNAPL Source Zone Architecture

项目摘要DNAPL源区结构表征的多模态和基于形状的反演方法地下水被致密非水相液体（DNAPLs）污染是美国乃至全世界的一个重大社会问题。旨在从地下完全清除污染物质量的昂贵的原位修复方法未能全面解决这一问题，因此，最近人们对旨在减少下游污染物质量通量的修复方法越来越感兴趣。对源区DNAPL污染的空间分布进行定量了解对于基于通量的修复和管理策略至关重要。事实上，已经证明这种源带结构与羽流的下游行为密切相关。不幸的是，源区结构的估计是一个非常具有挑战性的逆问题。我们提出了一种基于水文（下行梯度通量和浓度）和地球物理（电阻抗层析成像）数据的联合物理反演的源区表征方法。我们的处理方法通过采用一种新的源区表示来解决这个逆问题的病态性质。我们不是使用有限的数据来恢复DNAPL空间分布的精细尺度、像素化表示，而是将池、神经节和未污染区域的边界参数化。正在开发算法来估计这种几何形状以及污染区域中随空间变化的DNAPL饱和度。基于最近在图像处理和计算机视觉领域的工作，我们采用了一种新的参数化活动轮廓模型来描述池和神经节区域的边界。这些模型结合了传统水平集思想的拓扑灵活性和与蛇相关的低阶参数表示。我们的方法的性能是评估使用广泛的数值模拟套件，以及一套实验室规模的实验。模拟和实验将探索(a)将地球物理变量映射到水文变量的archie型混合规则的准确性和实用性，以及(b)该方法对地下电性和水文特性中未建模的体积非均质性的鲁棒性。智力优势：从本研究中获得的知识将通过将这些概念扩展到水文学中来改进基于几何的反演方法。除了推进水文学领域，我们的研究将通过开发新的、基于形状的多模态反问题方法来扩展数学成像。本研究还试图量化使用电阻抗和水文数据来表征准静态DNAPL源区结构的相关限制。更广泛的影响：这个项目有可能影响基础科学、工程和教育培训领域。正确识别源区域架构将为设计和选择适当的补救策略提供指导。本项目开发的方法具有应用于地球科学、医学成像和无损评估等领域的潜力，因为从高度异构的数据源中提取几何信息的问题在所有这些领域都广泛遇到。

Project AbstractMulti-Modal and Shape-Based Inverse Methods for the Characterization of DNAPL Source Zone ArchitectureGroundwater contamination by dense, non-aqueous phase liquids (DNAPLs) represents a major societal problem both within the United States and worldwide. Costly in situ remediation methods directed at the complete removal of contaminant mass from the subsurface have failed to provide a comprehensive solution to this problem thereby leading to recent, increased interest in an approach to remediation aimed at the reduction of downstream contaminant mass flux. Developing a quantitative understanding of the spatial distribution of DNAPL contamination in the source zone is critical to flux-based remediation and management strategies. Indeed, it has been shown that this source zone architecture is closely linked to downstream behavior of the plume. Unfortunately, the estimation of the source zone architecture is a very challenging inverse problem.We propose an approach to source zone characterization based on the joint, physics-based inversion of hydrological (down-gradient flux and concentration) and geophysical (electrical impedance tomography) data. Our processing approach addresses the ill-posed nature of this inverse problem by employing a novel representation of the source zone. Rather than using the limited data to recover a fine scale, pixilated representation of the spatial distribution of DNAPL, we parameterize the boundaries separating pools, ganglia, and non-contaminated regions. Algorithms are being developed to estimate this geometry along with the space-varying DNAPL saturation in the contaminated zones. Building on recent work in the image processing and computer vision fields, we employ a new form of parametric active contour models to describe the boundaries of the pool and ganglia regions. These models combine the topological flexibility of traditional level set ideas with the low order parametric representation associated with snakes. The performance of our approach is evaluated using an extensive suite of numerical simulations, as well as a set of laboratory-scale experiments. Simulations and experiments will explore (a) the accuracy and utility of Archie-type mixing rules for mapping geophysical to hydrological variables, and (b) the robustness of the method to un-modeled volumetric heterogeneities in both the electrical and hydrological properties of the subsurface.Intellectual Merit: Knowledge gained from this research will improve geometry-based methods for inversion by extending these concepts into hydrology. In addition to advancing the field of hydrology, our research will expand mathematical imaging through the development of new, shape-based methods for multi-modal inverse problems. This research also seeks to quantify the limits associated with using electrical impedance and hydrological data to characterize quasi-static DNAPL source-zone architecture.Broader impact: This project has potential to impact areas of basic science, engineering, and educational training. Proper identification of source zone architecture will provide guidance in designing and choosing appropriate remediation strategies. The methods developed in this project have the potential for application in fields such as earth sciences, medical imaging, and nondestructive evaluation, as the problem of extracting geometric information from highly heterogeneous data sources is widely encountered in all these areas.