Islam et al. (2026) Predicting groundwater recharge potential across various physiographic divisions of Bangladesh using generative data augmentation

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-08
• Authors: Md. Ashraful Islam, Syed Nazmus Sakib, Almahmud Taha, Mahfuzur R. Khan, Monira Jahan Tania, Tanvir Hossain, Sheikh Touhiduzzaman, Shifat E. Arman
• DOI: 10.1016/j.ejrh.2026.103909

Research Groups

• Department of Geology, University of Dhaka, Dhaka, Bangladesh
• Department of Robotics and Mechatronics Engineering, University of Dhaka, Dhaka, Bangladesh
• University of Twente, the Netherlands

Short Summary

This study employed generative artificial intelligence (GenAI) to predict groundwater recharge across various physiographic divisions in Bangladesh. The results showed that active floodplain areas exhibited the highest recharge potential, followed by terraces and older floodplains.

Objective

• To investigate how multiple hydro-environmental factors influence the spatial distribution of groundwater recharge potential.
• To evaluate whether advanced machine learning approaches can improve predictive accuracy compared with conventional mapping techniques in data-scarce settings.
• To develop a sustainable water management strategy for Bangladesh by identifying potential recharge areas.

Study Configuration

• Spatial Scale: The study area covers 1239 km² within the Bengal Basin, encompassing six physiographic divisions: Madhupur Tract, Jamuna Floodplain, Old Brahmaputra Floodplain, Meghna Floodplain, Tippera Surface, and Surma Basin.
• Temporal Scale: The study used data from 2000 to 2024.

Methodology and Data

• Models used: Conditional Tabular Generative Adversarial Network (CTGAN) and Tabular Variational Autoencoder (TVAE).
• Data sources: Google Earth Engine (GEE), Bangladesh Water Development Board (BWDB), USGS, Geological Survey of Bangladesh (GSB).

Main Results

• The results showed that active floodplain areas exhibited the highest recharge potential, followed by terraces and older floodplains.
• The use of GenAI-based data augmentation improved model performance, with relative test-R² gains ranging from 8.3% to 24.2%.
• Gradient Boosting with TVAE augmentation achieved the largest relative gain (+24.2%), while XGBoost with CTGAN augmentation achieved the highest absolute test R² of 0.716.

Contributions

• This study provides clear insight into the applicability of machine-learning models for groundwater-recharge prediction under limited-data conditions.
• The results demonstrate the strategic value of GenAI for addressing data scarcity and advancing groundwater-recharge modelling.

Funding

• This research was funded by [insert funding agency/project name] with reference code [insert reference code].

Citation

@article{Islam2026Predicting,
  author = {Islam, Md. Ashraful and Sakib, Syed Nazmus and Taha, Almahmud and Khan, Mahfuzur R. and Tania, Monira Jahan and Hossain, Tanvir and Touhiduzzaman, Sheikh and Arman, Shifat E.},
  title = {Predicting groundwater recharge potential across various physiographic divisions of Bangladesh using generative data augmentation},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103909},
  url = {https://doi.org/10.1016/j.ejrh.2026.103909}
}