Zhi Dou

In the post-earthquake engineering setting, tunnel water inrush is controlled not only by water sources, but also by how fault and fracture pathways, excavation boundaries, and operation conditions reshape groundwater pressure. Here, we proposed a source-pathway constrained framework integrating an Almon-type distributed lag model, hydrochemical and isotopic constraints, mineral saturation screening, and three-dimensional groundwater simulations to characterize post-earthquake groundwater hydraulic behavior in the Mozi–Chayuan Tunnel, Southwest China. The event-adjusted distributed-lag model identified a primary significant rainfall-drainage response window of 35–76 d in tunnel 1/tunnel 1', whereas tunnel 2 showed no significant delayed response, confirming that rainfall acted mainly as delayed recharge rather than an immediate trigger. Hydrochemical facies, isotope-derived recharge elevations, and saturation-index patterns indicate the inrush water was predominantly meteoric-derived water supplied by adjacent catchments and shallow fractured aquifers, without evidence for a dominant deep saline or hydrothermal contribution. The simulations reveal a clear fluid-mechanical transition: excavation converted the tunnel into a near-zero-pressure drainage boundary, whereas lining and 8 m full-section grouting established a pressure-bearing boundary with pressure heads of approximately 70–110 m in the route-change section during operation. The fault-related damage zone and route-change section remained the main zones of pressure-head redistribution across all simulated conditions. Local piezometer validation at 4100 m yielded relative errors of 2.09% and 5.45%, supporting the reliability of the framework for post-seismic tunnel groundwater response assessment. This framework supports the safe operation and pressure-control design of the Mozi–Chayuan Tunnel and provides a transferable reference for assessing tunnel groundwater responses in earthquake-prone regions.

Predicting solute and heat transport in highly heterogeneous aquifers remains challenging. Two major limitations hinder borehole-based subsurface characterization: (1) limited capability of downhole logging tests to capture vertical hydraulic conductivity (K) variations across a broad range of particle sizes; (2) conventional interpolation or simulation methods cannot faithfully capture the nested heterogeneity pattern of layered porous media between boreholes. Here, we establish an integration framework that uses nuclear magnetic resonance (NMR) logging to provide decimeter-scale K profiles spanning over five orders of magnitude. To characterize the nested heterogeneity within a glaciofluvial deposit, we performed sequential Gaussian simulation (SGS) within each clustering-defined hydrofacies, yielding equiprobable realizations that honor the layered geological structure. Drawdown responses to pumping are used to select K realizations that best reproduce groundwater flow dynamics. Results show that a standard kriging model conditioned on NMR data captures the general pattern of complex three-dimensional plume behavior. The proposed framework further improves transport predictions, outperforming the kriging model by 33.9% and the expected field of multi-layer SGS by 41.8%. This framework circumvents complex model calibration and can be extended to incorporate other independent state variables, offering an operable approach to high-resolution site characterization for groundwater remediation and geothermal energy use.

Monitoring well network design is crucial for reliable solute transport predictions in contaminated aquifers and is essential for AI-driven hydrological digital twins. Existing methods rarely achieve sequential data assimilation, computational efficiency, and prediction-oriented uncertainty reduction simultaneously while remaining consistent with the underlying flow and transport processes. To address these challenges, we propose a surrogate-driven progressive look-ahead framework for monitoring network design. It integrates a Variational Autoencoder, dual Fourier Neural Operators, and a latent-space Ensemble Smoother with Multiple Data Assimilation for iterative posterior updating. Virtual assimilation is performed at each candidate location to estimate the expected reduction in global concentration prediction uncertainty ({\bar{\sigma}}_C). The candidate with the greatest reduction is selected, added to the network, and assimilated before the next decision. The framework was tested on a two-dimensional synthetic heterogeneous aquifer across 100 realizations. The look-ahead strategy reduced {\bar{\sigma}}_C to 0.074, representing reductions of 18.7%, 17.8%, and 20.4% relative to the greedy, LHS, and random strategies, respectively (p<0.001), while also achieving the highest R_C^2, the coefficient of determination for concentration prediction. Look-ahead outperformed LHS in {\bar{\sigma}}_C and R_C^2 despite comparable R_K^2 (for hydraulic conductivity), indicating that concentration prediction reliability depends on where, not just how accurately, the hydraulic conductivity field is constrained. The greedy strategy clustered wells in high-uncertainty regions, resulting in redundancy, whereas look-ahead maintained a better balance between informative sampling and spatial complementarity. This framework provides an effective prediction-oriented approach to monitoring network design and a proof of concept for configuring sparse observations in digital-twin-oriented groundwater management.

Heterogeneous porous media, characterized by contrasting interfaces such as those between sand and clay, are ubiquitous in vadose zones. However, the transport mechanisms of Dense Non-Aqueous Phase Liquids (DNAPLs) across these interfaces remain poorly understood. This study uses an innovative stratified nuclear magnetic resonance (NMR) methodology to investigate the spontaneous infiltration behavior and underlying mechanisms of DNAPL across variably configured sand-clay interfaces. Through experiments simulating distinct interfacial scenarios, real-time, non-destructive monitoring of DNAPL spatial distribution was achieved. The results show that DNAPL transport depth and vertical distribution are critically governed by both the direction of interfacial crossing and the lens type within stratified media. Specifically, when migrating from clay to sand, DNAPL accumulates above the interface, while the reverse direction leads to predominant retention below. Notably, clay lenses exert a more pronounced capillary barrier effect than sand lenses, significantly reducing vertical migration. The study quantitatively establishes that interfacial capillary barrier effects, represented by newly proposed metrics such as retention rate and capillary retention efficiency, critically determine the redistribution patterns of DNAPL. This study clarifies the governing role of interfacial capillary barriers and provides quantitative metrics necessary to predict DNAPL fate and optimize mitigation efforts in heterogeneous subsurface environments.

《地球科学》(ISSN 1000-2383;CN42-1874/P) (原刊名《地球科学——中国地质大学学报》ISSN 1000-2383;CN42-1233/P)，创刊于1957年，是由教育部主管、中国地质大学主办的综合性地学学术理论期刊，月刊，2015年改名为《地球科学》,她以反映中国地质大学教学与科研最新的、高水平的基础地质、应用地质、资源与环境地质及地学工程技术研究成果为主要任务。

Many hydrogeological processes are intricately linked to both the spatial heterogeneity and temporal dynamics of the involved parameters. The variability in hydrogeological parameters across both space and time domains is defined as the spatiotemporal heterogeneity. This study introduces a partitioning geostatistical inversion approach to characterize the spatiotemporal heterogeneity, and further applies this methodology to the long‐term multi‐extensometer data from the second aquitard beneath Qingliang Primary School, Changzhou City, China. The Principal Component Geostatistical Approach (PCGA) is used to capture the temporal variability in the hydraulic conductivity ( K ) and specific storage ( S _s ) across the three sub‐layers of the aquitard of interest. Comparative analysis reveals that models accounting for spatiotemporal heterogeneity achieve significantly higher accuracy than those considering only spatial heterogeneity. PCGA effectively captures the temporal variations in aquitard K and S _s , with sub‐layer 3 dominating the temporal dynamics. Additionally, neglecting the initial delayed drainage within the aquitard leads to a notable overestimation of K . Moreover, utilizing a stage‐wise distribution of parameters as an initial guess significantly enhances inversion accuracy. The proposed methodology not only deepens the understanding of aquitard deformation but also holds broad potential for advancing the characterization of spatiotemporal heterogeneity in 4D hydrogeological modeling.