High-resolution characterization of complex groundwater systems using wireline logs analyzed with machine learning classifiers and isometric mapping techniques

Abstract

Listen

Translate article

Characterizing the lithological and hydraulic behavior of heterogeneous groundwater systems presents a significant challenge in hydrogeology. Traditional methods often rely on sparse data points that lead to inaccurate representations of the complex systems. This study presents an innovative approach to the characterization of the heterogeneous groundwater systems using wireline logs analyzed by machine learning (ML) techniques to infer the lithological variations and estimate aquifer parameters within the Quaternary aquifer system in the Debrecen area, Eastern Hungary. Initially, Manhattan distance-based k-means analysis as an outliers-resistance clustering method is employed to identify distinct lithological clusters based on the well logs responses. The results of the k-means clustering were then used to train ML classifiers including linear discriminant analysis, gradient boosting, random forest, and support vector machine for automated mapping of the lithofacies distribution. Additionally, the study introduced the first application of isometric map (IsoMap) to estimate the shale content and hydraulic conductivity within the aquifer system. The IsoMapping extracts latent components that capture essential features of the wireline logs and correlate them to the aquifer parameters. The regression between the latent component and the deterministically estimated shale volume and hydraulic conductivity showed significant exponential relationships resulting in universal equations that can be used independently to estimate these parameters. For more robust estimation, genetic algorithm global optimization was applied to refine the regression parameters governing these relationships to overcome the limitations associated with linearized estimations. The proposed approach provided a fast, automated, and effective alternative for characterizing heterogeneous groundwater, offering reliable inputs for groundwater flow and contaminant transport models.