Khan et al. (2026) Physics-informed Bayesian Neural Network for groundwater recharge estimation in data-scarce arid regions

Identification

• Journal: Frontiers in Water
• Year: 2026
• Date: 2026-03-25
• Authors: MD Shaibaz Khan, Marwan Fahs, Ahmed Hadidi, Husam Musa Baalousha
• DOI: 10.3389/frwa.2026.1787659

Research Groups

• Department of Geosciences, College of Petroleum Engineering and Geosciences, King Fahd University of Petroleum and Minerals (KFUPM), Dhahran, Saudi Arabia
• ITES UMR 7063, CNRS, ENGEES, Université de Strasbourg, Strasbourg, France
• Department of Applied Geosciences—German University of Technology in Oman, Muscat, Oman
• Interdisciplinary Research Center for Membranes and Water Security, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia

Short Summary

This study developed a Physics-Informed Bayesian Neural Network (PI-BNN) to estimate groundwater recharge and its uncertainty in the data-scarce South Al Batinah (SAB) Basin, northern Oman. The PI-BNN significantly reduced uncertainty bounds by approximately 50% compared to Latin Hypercube Sampling (LHS) while maintaining physically consistent and realistic recharge estimates.

Objective

• To quantify groundwater recharge and its uncertainty in the South Al Batinah (SAB) Basin, northern Oman, using a Physics-Informed Bayesian Neural Network (PI-BNN) integrated with a soil-moisture mass balance framework.

Study Configuration

• Spatial Scale: South Al Batinah (SAB) Basin, northern Oman, spanning approximately 5,280 km².
• Temporal Scale: Monthly data from January 1990 to December 2023 (34 years; 408 monthly time steps).

Methodology and Data

• Models used:
  • Physics-Informed Bayesian Neural Network (PI-BNN)
  • Bayesian Neural Network (BNN) (for ablation study)
  • Latin Hypercube Sampling (LHS)
  • Soil-water mass balance model
  • Noah Land Surface Model (version 3.6.1) (underlying FLDAS)
• Data sources:
  • FLDAS (Famine Early Warning Systems Network Land Data Assimilation System) remote sensing and land-surface model data (Noah Land Surface Model v3.6.1).
  • CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data) for precipitation.
  • MERRA-2 (Modern-Era Retrospective Analysis for Research and Applications, version 2) for meteorological forcing data (air temperature, humidity, wind speed, shortwave radiation).
  • Hydrological variables: soil moisture (0–10 cm, 10–40 cm, 40–100 cm, 100–200 cm), precipitation rate, actual evapotranspiration, surface runoff, subsurface runoff/baseflow.

Main Results

• Groundwater recharge in the SAB Basin is highly seasonal and episodic, with negligible recharge during dry months (May, June, September, October).
• Peak monthly recharge occurs in December (approximately 5–6 mm/month), with moderate values in February–March and July.
• Annual groundwater recharge is estimated at approximately 16 mm/year by PI-BNN and 7–11 mm/year by LHS (7.2 mm/year for correlated LHS).
• Precipitation variability accounts for more than 70% of recharge uncertainty during wet months.
• The PI-BNN reduced uncertainty bounds by approximately 50% compared to independent LHS, while maintaining physically consistent estimates.
• An ablation experiment confirmed that physics-informed constraints, not just the Bayesian architecture, were crucial for physically plausible recharge recovery.

Contributions

• Developed and applied a novel Physics-Informed Bayesian Neural Network (PI-BNN) for groundwater recharge estimation in data-scarce arid regions.
• Demonstrated that embedding physical constraints (water balance equation, recharge limits) into the neural network's loss function significantly reduces uncertainty and improves physical realism compared to traditional stochastic methods (LHS) and unconstrained BNNs.
• Provided a robust and transferable framework for quantifying groundwater recharge and its uncertainty in challenging arid environments.
• Quantified the seasonal and episodic nature of groundwater recharge in the South Al Batinah Basin, Oman, over a 34-year period using remote sensing data.

Funding

• King Fahd University of Petroleum and Minerals (KFUPM)

Citation

@article{Khan2026Physicsinformed,
  author = {Khan, MD Shaibaz and Fahs, Marwan and Hadidi, Ahmed and Baalousha, Husam Musa},
  title = {Physics-informed Bayesian Neural Network for groundwater recharge estimation in data-scarce arid regions},
  journal = {Frontiers in Water},
  year = {2026},
  doi = {10.3389/frwa.2026.1787659},
  url = {https://doi.org/10.3389/frwa.2026.1787659}
}