Optimized sampling design for groundwater contaminant source zone architecture characterization using Bayesian experimental design.

Precise characterization of dense non-aqueous phase liquid (DNAPL) source zone architecture (SZA) is crucial for effective site remediation and environmental protection. To achieve accurate SZA characterization, the traditional uniform sampling method requires considerable intrusive boreholes, resulting in high costs and low feasibility. Therefore, optimizing the monitoring network is essential for minimizing the number of wells while maximizing information yield. However, DNAPL SZA exhibits a highly irregular morphology, posing great challenges to conventional optimization methods that determine all well locations in a single step. To address these challenges, this study proposes an adaptive sampling strategy that sequentially and jointly optimizes horizontal well placement and vertical sampling depths, while iteratively updating SZA estimates with measurements collected from newly selected locations. Relative entropy within a Bayesian experimental design framework is used to determine the next most informative sampling configuration. We evaluated the proposed approach through a 2-D aquifer model representing site-relevant conditions, and compared the performance of traditional uniform and adaptive sampling strategies under the same number of wells. Results show that the adaptive sampling strategy outperformed the traditional uniform sampling in characterizing the SZA, achieving a 40% reduction in DNAPL mass estimation error. This improved characterization further enabled more accurate simulation of DNAPL depletion dynamics and longevity, providing a more reliable initial condition for predictive remediation modeling. This study develops a novel framework for monitoring network optimization, offering a practical strategy for more efficient contaminated site management and remediation.