An integrated workflow combining machine learning and wavelet transform for automated characterization of heterogeneous groundwater systems

Musaab A A Mohammed ^1,2,3, Norbert P Szabó ⁴, Abdelrhim Eltijani ⁵

⁰Faculty of Earth and Environmental Sciences and Engineering, University of Miskolc, Miskolc, Egyetemváros, 3515, Hungary. musab20501@gmail.com.

Scientific Reports +

|

February 10, 2025

View abstract on PubMed

Summary

This study presents an automated machine learning approach to improve groundwater aquifer characterization. The method accurately maps lithology and hydrostratigraphy, aiding in effective groundwater management.

Area Of Science

Hydrogeology
Geophysics
Machine Learning

Background

Accurate aquifer characterization is crucial for groundwater management but traditional methods are costly and time-consuming.
Geophysical well logs offer an efficient alternative, yet manual interpretation can be challenging and ambiguous.
Automated interpretation of well logs is needed for large-scale hydrogeological studies.

Purpose Of The Study

To develop and validate an automated machine learning approach for enhanced aquifer characterization.
To improve the accuracy and efficiency of lithological and hydrostratigraphical mapping.
To create a 3D lithological model for groundwater flow and contaminant transport modeling.

Main Methods

Imputation of missing deep resistivity logs using a gated recurrent unit (GRU) neural network.
Application of Self-Organizing Maps (SOMs) for lithological unit distribution mapping.
Refinement of hydrostratigraphical boundaries using Continuous Wavelet Transform analysis.

Main Results

The GRU network effectively imputed missing log data, enhancing overall data quality.
SOMs successfully delineated three primary lithological units (shale, shaly sand, sand and gravel).
The integrated approach generated a 3D lithological model with four hydrostratigraphical zones, correlating with permeability and shale volume.

Conclusions

The automated approach provides a robust and efficient method for aquifer characterization.
The 3D lithological model supports groundwater flow and contaminant transport modeling.
This methodology can be extended to other regions for improved aquifer management and development.

Keywords:

Artificial neural networks Debrecen Pannonian Basin Self-organizing map Wireline log