A sequential HT-Bayesian method offers deep insights for precise groundwater contaminant characterization

Abstract

Precise characterization of aquifer heterogeneity and contaminant concentration distribution is essential for effective groundwater pollution management and environmental protection. Many fields face sampling difficulties, resulting in insufficient or minimal data, creating a "Small data". Developing methods to achieve precise characterization based on the limited data available has become a critical and urgent challenge in the field. Therefore, a sequential HT-Bayesian method integrating Hydraulic Tomography (HT) with Bayesian optimization to achieve precise characterization of hydraulic and contaminant concentration field in aquifers with small data was presented in this study. This method achieved an acceptable characterization of aquifer hydraulic conductivity fields with just four pumping tests. Compared to traditional interpolation techniques, this proposed method significantly outperforms in capturing the complexity of aquifer systems with R² > 0.94. The robustness and transferability of the sequential method were further validated in scenarios with well density of 9 × 10⁻⁵ n/m², where it maintained high accuracy and precision even in limited data conditions. This study evaluates the feasibility of using small datasets for contamination characterization, providing a theoretical basis for subsurface investigations. The findings suggest potential benefits under data-limited conditions but remain constrained by model assumptions and limitations.