Shandu et al. (2025) Comparative analysis of groundwater potential zones using machine learning and hybrid method in Nelson Mandela Bay Metropolitan Municipality, South Africa

Identification

• Journal: Next research.
• Year: 2025
• Date: 2025-09-17
• Authors: Irvin D. Shandu, Sifiso Xulu, Michael Gebreslasie, Iqra Atif
• DOI: 10.1016/j.nexres.2025.100821

Research Groups

• School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, Westville Campus, Durban, South Africa
• Department of Geography, Archaeology and Environmental Studies, University of Witwatersrand, Johannesburg, South Africa

Short Summary

This study developed and applied a hybrid machine learning meta-model to map groundwater potential zones (GWPZs) in the data-scarce Nelson Mandela Bay Metropolitan Municipality, South Africa. The meta-model accurately identified areas with moderate-to-very high groundwater potential, providing a replicable framework for sustainable groundwater management.

Objective

• To derive groundwater potential zones (GWPZs) in the Nelson Mandela Bay Metropolitan Municipality using machine learning and hybrid methods to address limited knowledge on sustainable groundwater assessment in semi-arid regions.

Study Configuration

• Spatial Scale: Nelson Mandela Bay Metropolitan Municipality (NMBM), South Africa.
• Temporal Scale: Groundwater discharge records (2011–2022), groundwater level records (2011–2014), and inconsistent measurements between 2021 and 2023.

Methodology and Data

• Models used: Bagged Classification and Regression Tree (BagCART), Boosted Regression Tree (BRT), Random Forests (RF), a weighting approach (ensemble), and a stacking RF-based meta-model (RF-BRT-BagCART).
• Data sources:
  • Fourteen groundwater-influencing factors: precipitation, land use and land cover, normalised difference vegetation index, lineament density, topographic wetness index, drainage density, slope, aspect, profile and plan curvature, distance from streams, distance from fault lines, lithology, and soil texture (processed in ArcGIS).
  • Borehole discharge data (2011–2022).
  • Groundwater level data (2011–2014).
  • Data gaps interpolated using Inverse Distance Weighting (IDW).
  • Model training and validation performed in RStudio (70 % training, 30 % validation).

Main Results

• The stacking RF-based meta-model (RF-BRT-BagCART) significantly outperformed individual machine learning learners.
• The meta-model achieved high Area Under the Curve (AUC) scores: 0.963 for groundwater discharge and 0.982 for groundwater level.
• The Nelson Mandela Bay Metropolitan Municipality was classified into GWPZs: 28.2 % very low, 19.6 % low, 17.0 % moderate, 22.1 % high, and 13.1 % very high potential.
• Overall, 54.8 % of the municipality was identified as having moderate-to-very high groundwater potential.

Contributions

• Demonstrates a novel approach by integrating long-term groundwater discharge and level data with hybrid machine learning for mapping groundwater potential in data-scarce regions.
• Provides a replicable framework for sustainable groundwater management and mapping, particularly beneficial for semi-arid areas.
• Offers valuable insights for more informed and strategic borehole siting, enhancing water security.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Shandu2025Comparative,
  author = {Shandu, Irvin D. and Xulu, Sifiso and Gebreslasie, Michael and Atif, Iqra},
  title = {Comparative analysis of groundwater potential zones using machine learning and hybrid method in Nelson Mandela Bay Metropolitan Municipality, South Africa},
  journal = {Next research.},
  year = {2025},
  doi = {10.1016/j.nexres.2025.100821},
  url = {https://doi.org/10.1016/j.nexres.2025.100821}
}