Benchmarking traditional interpolation and machine learning methods for 3D hydrogeological modeling in complex multi-aquifer systems
Year: 2026
Type: article
Abstract: Study region This study focuses on a geologically complex multi-aquifer system in the Berlin-Brandenburg region of the North German Basin. The area is shaped by repeated Quaternary glaciations resulting in a heterogeneous geological architecture characterized by glaciotectonic deformation, overdeepened valleys, erosion windows, and discontinuous aquifers. Study focus We present a scalable and transferable workflow to construct high-resolution three-dimensional hydrogeological models by integrating borehole data and interpreted geological cross-sections within a unified label-based framework. We systematically compare conventional interpolation techniques (inverse distance weighting and nearest neighbor) and machine learning approaches (decision trees, random forests, AdaBoost, XGBoost, k-nearest neighbors, and support vector machines), including deep learning models (artificial neural networks) to evaluate their ability to reconstruct complex heterogeneous subsurface structures. New hydrological insights for the region The Berlin-Brandenburg multi-aquifer system is controlled by highly discontinuous glacial structures, including overdeepened valleys, erosion windows, and fault-related discontinuities that govern aquifer connectivity and potential pathways for vertical groundwater exchange. Our results demonstrate the ability of the proposed artificial neural network to reconstruct key hydrogeological features while preserving their spatial continuity. It also outperforms the benchmark methods evaluated in this study upon achieving a test accuracy of 98% and a weighted F1-score of 0.98. By preserving the spatial continuity of aquifers and aquitards, the reconstructed geological architecture provides the structural foundation required for physically consistent assignment of hydraulic properties in future groundwater models. As such, it supports improved representation of structural heterogeneity controlling groundwater flow and potential salinity migration in the North German Basin.
Language: English
Subfield: Geology
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Keywords: multi-aquifer system, machine learning interpolation, artificial neural networks, inverse distance weighting, nearest neighbor interpolation, borehole data, geological cross-sections, glaciotectonic structures, aquifer connectivity, groundwater exchange, aquitard, North German Basin, Berlin-Brandenburg region
Open Access status: gold
Estimated APC paid: $2,600
OpenAlex ID: W7203750001
