해체 연구용 원자로 부지에서의 수리지질학적 민감도 및 불확실성 분석과 LHS 기반 용질 이동 시뮬레이션 = Hydrogeologic Sensitivity and Uncertainty Analyses with Latin Hypercube Sampling (LHS)-based Solute Transport Simulations at a Decommissioned Research Reactor Site

다국어 초록 (Multilingual Abstract)

Nuclear power plants worldwide are aging, and many are approaching closure, making decommissioning and remediation of radioactive contamination essential for site reuse. Numerical groundwater models are widely used in groundwater contamination studies, but few studies combine three-dimensional flow and transport modeling with sensitivity and uncertainty analyses at nuclear decommissioning sites, particularly with detailed layer-wise stratigraphy. This study used site investigation data and a numerical groundwater model of a decommissioned research reactor site to identify which hydraulic parameters and geological layers most strongly control contaminant migration under varying soil conditions, focusing on layer-specific sensitivities and spatial variability. Transport and concentration patterns of tritium and strontium-90 were simulated using uncertainty-based parameter ranges to estimate plausible concentration ranges. These results were used to characterize plume behavior and identify priority monitoring wells near the spent-fuel storage area, within the main reactor plume, and in the central downgradient part of the site. The proposed framework supports the design of remediation and monitoring strategies and can be applied to future decommissioning sites.
A base model developed by using site-specific field data was used for simulations at the study site. The model was configured with 7 geological layers, from hard rock at the bottom through medium rock, soft rock, weathered rock, weathered soil, and alluvial to backfill, and with 12 pumping and observation wells placed on a regular grid. Each well contained 7 observation points, giving 84 head targets. Sensitivity was evaluated from head changes caused by percent perturbations of hydraulic conductivity (K), porosity (φ), and specific storage (SS) at ±10, ±25, ±50, ±75, −99, and +100 %. Normalized head responses were used to compute parameter log variances from the covariance matrix, and these variances defined Latin Hypercube Sampling (LHS) ranges from which 1,000 parameter sets were generated. Two contamination sources representing the KRR-1 spent fuel storage room and the KRR-2 reactor were assigned tritium and strontium-90 at initial concentrations of 50 kg/m³, with the KRR-1 source located in the upper left of the model domain and the KRR-2 source on the right side. Three-dimensional transport was simulated for 30 years under annual precipitation conditions based on Seoul area averages from 2014 to 2024 using both the 1,000 sampled parameter sets and a baseline case without sampling. Results from the sampled runs were summarized as C/C0 with 1,000 realizations together with their ensemble mean and standard deviation.
Layer-wise sensitivity results showed that hydraulic conductivity (K) exerted the largest effect on heads across the model, while specific storage (SS) produced little sensitivity in all layers. Porosity (φ) sometimes produced large sensitivity values, but the associated 95 percent confidence intervals were wide and indicated substantial spatial variability. Among the seven layers, the medium rock layer (layer 2) and the weathered rock layer (layer 4) were most sensitive to changes in both K and φ. Well-based vertical analysis also confirmed that φ can be highly influential only at specific locations with wide confidence intervals. Overall, K ranked first in influence on heads, followed by φ and then SS.
Transport simulations without LHS sampling showed that contaminants moved along surface gradients and with groundwater flow. Tritium spread through the upper layers, infiltrated into the groundwater table, and reached the model boundary from both sources after approximately 1 year, whereas strontium-90 migrated much more slowly and remained concentrated near the sources after 30 years. The 1,000 LHS realizations produced results that were broadly consistent with the non-sampled simulations. Concentration time series at the 84 observation points indicated that monitoring locations near the KRR-1 spent fuel storage room on the side opposite the main groundwater flow, in the central downgradient part of the model, and within the core of the KRR-2 plume are suitable as priority observation wells. At the upgradient well near the KRR-1 source, tritium was detected despite the opposite flow direction, and small strontium-90 concentrations were observed in the weathered soil and alluvial layers, with mean values close to zero and only a small fraction of realizations showing measurable concentrations. Within the KRR-2 plume core, tritium concentrations reached up to about 40 kg/m³, and strontium-90 concentrations were mainly observed at this location. These patterns reflect the spreading of contaminants into surrounding shallow layers driven by precipitation and subsequent transport.
The results indicate that concentration can be observed under a variety of layered subsurface conditions, and that model outcomes vary with parameter values. The study identified K as the parameter that responds most sensitively in the modeling and found that accurate parameter values obtained from field investigations are important for model setup. Transport simulations conducted across a range of site conditions showed that contaminant behavior and concentration distributions change with subsurface properties, while uncertainty-based modeling is useful for detecting these changes and estimating plausible concentration ranges. Analysis of contaminant behavior and migration pathways at the research reactor site demonstrated that the combined sensitivity, uncertainty, and transport framework can support the selection of priority monitoring locations and more effective contamination management, thereby addressing the current lack of uncertainty-based three-dimensional transport analyses at nuclear decommissioning sites and providing an approach that can be transferred to similar facilities.

목차 (Table of Contents)

• 1. Introduction 1
• 2. Study site 8
• 3. Methods 21
• 3.1. Numerical model - HydroGeoSphere (HGS) 21
• 3.2. Modeling framework 27
• 3.3. Simulation setting for the pumping scenarios 33
• 3.4. Sensitivity analysis 38
• 3.5. Uncertainty analysis 41
• 3.6. Latin Hypercube Sampling (LHS) 44
• 3.7. Transport simulation 47
• 4. Results and discussion 49
• 4.1. Comparison of head sensitivity across parameters 49
• 4.2. Transport simulation of tritium and strontium-90 62
• 4.3. Transport simulation using LHS sampling 73
• 4.4. Implications of the study 85
• 5. Conclusions 88
• References 92
• 국문초록 102
• 감사의 글 106