基于四维时移电阻率成像监测和数据同化的重非水相污染物源区识别研究

DNAPL (Dense Non-Aqueous Phase Liquid) contamination poses a major threat to the groundwater supply; thus, successful remediation of the contaminated sites is of paramount importance. When DNAPL migrates through saturated subsurface environments, it forms complex contaminant source architectures (CSAs) due to aquifer heterogeneity. As a result, identification of CSAs is an essential step that renders remediation successful and lowers the estimated remediation time and cost significantly. As a non-invasive technique, four-dimensional time-lapse Electricity Resistivity Tomography (ERT) is used to monitor DNAPL CSAs . A framework coupled ERT and multiphase forward model is developed to overcome the non-unique shortage of ERT inversion, which considers the linkage between hydrogeological and geoelectrical properties. Hydrogeological and geophysical parameters are jointly inversed through Ensemble Kalman Filter (EnKF) approach integrating geophysical and hydrological data. The developed approaches are expected to solve the inverse problem of DNAPL source identification with the challenge of high dimension and non-unique. Improving the identification of DNAPL CSAs is helpful to determine the contaminative responsibility and provides theories foundation and technical support for the monitoring and remediation of DNAPL contaminated sites.

重非水相（DNAPL）污染物已对地下水水质构成重大威胁。含水介质的非均质性导致含水层中形成的DNAPL污染源区结构非常复杂。识别DNAPL污染源区结构是污染场地调查和修复的难点和关键，决定了修复成本和效率。本课题以四维时移电阻率成像（Time Lapse ERT）作为污染源区结构的非侵入式监测工具；构建耦合DNAPL多相流数值模型和ERT正演模型的框架，以克服分开估计地球物理和水文地质参数造成的反演多解；采用集合卡尔曼滤波数据同化方法融合ERT数据和水文地质数据进行求解，以提高DNAPL污染源区结构的识别精度并降低其不确定性。所开发数值平台将通过室内实验和野外场址进行校核，预期可有效解决DNAPL污染源区识别逆问题中的高维参数和强不适定性难题。研究成果可直接应用于DNAPL污染源区识别，指导后续DNAPL污染场地的科学监测及修复整治，也可为国家相关部门进行污染责任认定提供有力科学依据。

针对DNAPL污染源区识别时观测数据稀少以及多相流反演计算量大的困难挑战，通过室内试验，数值模拟和现场研究相结合的研究手段，系统探索了DNAPL污染源区精细刻画的解决途径。已按计划完成项目研究工作。主要研究进展与成果包括：（1）基于室内二维和三维砂箱实验定量评估了非侵入式时移ERT刻画DNAPL迁移分布的有效性，并在野外场地结合钻孔取样进行验证；（2）考虑到地质非均质性对DNAPL污染源区的显著影响，构建了多相流模型-ERT 正演模型的耦合框架，联合推估DNAPL污染源区的饱和度分布和介质的渗透率分布；（3）为有效利用低成本的地球物理数据，开展了融合地球物理数据和水文地质数据的数据同化研究。为进一步提高多相流反演计算量负担，开展了基于深度学习的数据同化框架探索。上述研究进展已发表论文28篇（SCI收录17篇，其中水文学领域顶级期刊Water Resources Research 5篇和Journal of Hydrology 4篇），有效解决DNAPL污染源区识别逆问题中的高维参数和强不适定性难题。