Study on the Optimization of Monitoring Well Placement Using Bayesian Model Averaging and Bayesian Maximum Entropy

　　Certain limitations arise when utilizing the Monitoring Efficiency Model (MEMO) and Monitoring and Remediation Optimization System (MAROS) to evaluate monitoring well placement at contaminated sites. MEMO is restricted to one-dimensional groundwater flow analysis, while MAROS can only handle two-dimensional spatial distribution of contaminants. These constraints hinder the ability to account for variability in the three-dimensional spatial distribution of contaminants, leading to suboptimal monitoring well configurations. In particular, factors such as geological heterogeneity and contaminant characteristics (e.g., biodegradation, chemical degradation, and physical adsorption) may lead to contaminant omissions or inappropriate monitoring well density distribution, ultimately limiting the efficiency and accuracy of monitoring well placement.

　　To address these challenges, this study proposes an optimized approach for monitoring well placement at three-dimensional groundwater contamination sites. The method integrates Bayesian Model Averaging (BMA) and Bayesian Maximum Entropy (BME) to delineate contaminant plumes more accurately and provide optimal recommendations for monitoring well placement. BMA, utilizing Markov Chain Monte Carlo (MCMC) simulations and Bayesian inference, calculates the posterior distribution of multiple potential Conceptual Site Models (CSMs) by evaluating discrepancies between observed and simulated contaminant concentrations.

　　Using the weighted CSM, the relative positions between existing monitoring wells and the contaminant plume can be evaluated. During the numerical simulation process, virtual observation points are added to enhance the richness and completeness of data distribution within the contaminated area, further improving the interpolation accuracy of BME. Through this improvement, BME can integrate simulated data with existing monitoring data to precisely predict the locations of additional monitoring wells, supplement critical monitoring data, and optimize the overall monitoring well placement strategy.

　　Additionally, this study incorporates monitoring well-installation costs, the value of information (VOI), and trans-information entropy (TE) into a multi-objective optimization framework. By minimizing the objective function, Pareto-optimal solutions are obtained. The Preference Ranking Organization METHod for Enrichment Evaluations (PROMETHEE) is then applied to rank these solutions, enabling decision-makers to balance monitoring efficiency with cost considerations and implement flexible and effective monitoring configurations. It also verifies the feasibility of retaining a significant portion of critical monitoring information through VOI-based quantitative analysis, even with a reduced number of monitoring wells.

　　The proposed optimization method has been validated through numerical simulations, demonstrating improved model accuracy under complex site conditions. The results offer adaptable, site-specific solutions that maximize both monitoring efficiency and economic viability.

 

Keywords: Bayesian Model Averaging, Bayesian Maximum Entropy, groundwater contaminant transport, optimization of monitoring well placement