Neili et al. (2026) Machine learning algorithms for estimating basin-scale groundwater levels based on GRACE satellite data

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-07
• Authors: Khouloud Neili, Fatma Trabelsi, Shakil Jiwa, Sarah Durrani, Amir AghaKouchak
• DOI: 10.1016/j.ejrh.2026.103942

Research Groups

• Research Unit Sustainable Management of Water and Soil Resources, Higher School of Engineers of Medjez El Bab (ESIM), University of Jendouba, Tunisia
• Center for Hydrometeorology and Remote Sensing, Department of Civil and Environmental Engineering, University of California, Irvine, CA, USA
• Department of Earth System Science, University of California, Irvine, CA, USA

Short Summary

This study aims to predict groundwater levels in data-scarce environments by integrating Earth observation (EO) products with machine learning algorithms. The results demonstrate that Random Forest was the best-performing model for predicting groundwater levels.

Objective

• To develop and evaluate robust machine learning models for accurate GWL prediction using EO multi-source data.
• To integrate ML with EO data from GRACE, GLEAM, GLDAS, and in-situ measurements to predict GWL.

Study Configuration

• Spatial Scale: The study focuses on the Lower Valley of Medjerda (LVM) Basin in northern Tunisia, covering an area of 1656 km2.
• Temporal Scale: Monthly time series data from April 2002 to December 2019 were used for model development and evaluation.

Methodology and Data

• Models used:
  • Random Forest (RF)
  • XGBoost (XGB)
  • Support Vector Regression (SVR)
  • Long Short-Term Memory (LSTM)
• Data sources:
  • Gravity Recovery and Climate Experiment (GRACE) data for Total Water Storage (TWS)
  • Global Land Data Assimilation System (GLDAS) data for soil water (SW) and canopy water content (CWC)
  • Global Land Evaporation Amsterdam Model (GLEAM) data for actual evapotranspiration (AET)
  • Meteorological data from Tunisian rainfall stations and the National Institute of Meteorology of Tunisia
  • In-situ groundwater level data from 62 piezometric wells

Main Results

• The results demonstrate that RF was the best-performing model, yielding R² values between 0.7604 and 0.9475, NS values between 0.7604 and 0.9460, WI values between 0.7812 and 0.9722, and a minimal RMSE of 0.0210 m.
• The LSTM model effectively captured long-term trends despite being susceptible to overfitting.

Contributions

• This study highlights the potential of integrating downscaled EO data with machine learning algorithms to enhance groundwater assessment in data-scarce regions.
• The results demonstrate that RF was the best-performing model for predicting groundwater levels, providing accurate predictions and uncertainty estimates.

Funding

• Not specified.

Citation

@article{Neili2026Machine,
  author = {Neili, Khouloud and Trabelsi, Fatma and Jiwa, Shakil and Durrani, Sarah and AghaKouchak, Amir},
  title = {Machine learning algorithms for estimating basin-scale groundwater levels based on GRACE satellite data},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103942},
  url = {https://doi.org/10.1016/j.ejrh.2026.103942}
}