Azghandi et al. (2026) Machine Learning–Based Characterization of Groundwater Recharge in Semi-Arid Drylands

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-03-01
• Authors: Shadi Askari Azghandi, Ehsan Behnamtalab
• DOI: 10.1007/s11269-026-04557-8

Research Groups

• Department of Civil Engineering, Hakim Sabzevari University, Sabzevar, Iran

Short Summary

This study characterized groundwater recharge dynamics in the semi-arid Karkheh Plain (Iran) from 2001–2024 using satellite-based water balance and machine learning, finding that ΔSoil Moisture is the dominant driver and that positive recharge peaks have significantly declined, indicating increasing groundwater vulnerability.

Objective

• Assess long-term groundwater recharge variability.
• Identify dominant hydrometeorological controls on recharge.
• Classify groundwater recharge regimes using clustering techniques.

Study Configuration

• Spatial Scale: Karkheh Plain, southwestern Iran (within the Karkheh River Basin, approximately 51,000 km²).
• Temporal Scale: Monthly resolution from 2001 to 2024.

Methodology and Data

• Models used:
  • Water balance equation: Recharge = Precipitation - Evapotranspiration - Runoff - ΔSoil Moisture
  • Machine Learning Regression: Linear Regression, Random Forest, AdaBoost
  • Unsupervised Clustering: K-Means algorithm
  • Statistical: Pearson correlation, Principal Component Analysis (PCA), Univariate Linear Regression, RReliefF algorithm
• Data sources:
  • Satellite: CHIRPS (Precipitation), MODIS (Evapotranspiration, Land Cover MCD12Q1 Type-1), SRTM (Digital Elevation Model).
  • Reanalysis: GLDAS (Runoff, Soil Moisture).

Main Results

• Monthly groundwater recharge exhibits substantial fluctuations, with a notable decline in positive recharge peaks in recent decades (e.g., from +71 mm in 1974 to less than +20 mm in recent years).
• The long-term mean monthly recharge is close to zero or slightly negative, indicating a near-equilibrium state between inputs and outputs.
• Spatially, most of the plain shows near-zero or slightly negative recharge, with localized positive recharge hotspots occurring during maximum-recharge years.
• Correlation and feature-importance analyses consistently identified ΔSoil Moisture as the dominant driver of monthly recharge variability (Pearson correlation coefficient, r = -0.673, statistically significant).
• Precipitation is a secondary but significant factor (r = +0.235), while runoff and evapotranspiration show minimal direct influence on monthly recharge.
• Linear Regression achieved the highest predictive accuracy (R² = 0.999, RMSE = 1.168 mm, MAE = 0.805 mm, MAPE = 15.53%), outperforming Random Forest and AdaBoost, suggesting that recharge processes primarily follow linear hydrological relationships at the monthly scale.
• K-Means clustering identified three distinct recharge regimes: wet (high variability), transitional (moderate), and dry (uniform low/negative recharge), reflecting seasonal and interannual hydroclimatic variations.
• Land-cover changes, specifically the expansion of croplands and reduction of natural vegetation (shrublands, rangelands) from 2001 to 2024, likely contributed to increased water demand and decreased natural infiltration capacity.

Contributions

• Integrated satellite-based water balance modeling with advanced machine learning techniques (Linear Regression, Random Forest, AdaBoost, K-Means clustering) for comprehensive groundwater recharge characterization in a semi-arid dryland.
• Provided a long-term (2001-2024) assessment of groundwater recharge variability, highlighting a significant decline in positive recharge peaks and increasing groundwater vulnerability.
• Identified ΔSoil Moisture as the primary hydrometeorological control on monthly recharge variability, clarifying its dominant role over precipitation, runoff, and evapotranspiration in semi-arid conditions.
• Demonstrated the superior predictive performance of Linear Regression for monthly recharge in this context, suggesting that recharge dynamics predominantly follow linear relationships.
• Classified distinct hydrological recharge regimes, offering valuable insights for targeted monitoring and management strategies.
• Emphasized the combined impact of climatic pressures and anthropogenic land-cover changes on groundwater resources, underscoring the urgent need for adaptive land-water management.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Citation

@article{Azghandi2026Machine,
  author = {Azghandi, Shadi Askari and Behnamtalab, Ehsan},
  title = {Machine Learning–Based Characterization of Groundwater Recharge in Semi-Arid Drylands},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-026-04557-8},
  url = {https://doi.org/10.1007/s11269-026-04557-8}
}