Scholarlink

<p>This study investigates groundwater recharge dynamics in the semi-arid Karkheh Plain of southwestern Iran using an integrated approach that combines satellite-based water-balance modeling with machine-learning analysis for the period 2001–2024. Monthly recharge was estimated using CHIRPS precipitation, MODIS evapotranspiration and land cover, and GLDAS soil-moisture and runoff datasets. Results show substantial month-to-month fluctuations, while the long-term mean remains close to zero. A notable decline in positive recharge peaks is observed in recent decades. Correlation and feature-importance analyses identify that ΔSoil Moisture as the dominant driver of recharge variability, while precipitation acting as a secondary but significant factor. Among the machine-learning models tested, Linear Regression achieved the highest performance (R² = 0.999), outperforming Random Forest and AdaBoost. K-Means clustering further identified three recharge regimes—wet, transitional, and dry—corresponding to seasonal and interannual hydroclimatic variations. These findings highlight increasing groundwater vulnerability driven by climatic pressures and land-cover changes, emphasizing the need for adaptive land-water management strategies under semi-arid conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning–Based Characterization of Groundwater Recharge in Semi-Arid Drylands

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Shadi Askari Azghandi,
  • Ehsan Behnamtalab

摘要

This study investigates groundwater recharge dynamics in the semi-arid Karkheh Plain of southwestern Iran using an integrated approach that combines satellite-based water-balance modeling with machine-learning analysis for the period 2001–2024. Monthly recharge was estimated using CHIRPS precipitation, MODIS evapotranspiration and land cover, and GLDAS soil-moisture and runoff datasets. Results show substantial month-to-month fluctuations, while the long-term mean remains close to zero. A notable decline in positive recharge peaks is observed in recent decades. Correlation and feature-importance analyses identify that ΔSoil Moisture as the dominant driver of recharge variability, while precipitation acting as a secondary but significant factor. Among the machine-learning models tested, Linear Regression achieved the highest performance (R² = 0.999), outperforming Random Forest and AdaBoost. K-Means clustering further identified three recharge regimes—wet, transitional, and dry—corresponding to seasonal and interannual hydroclimatic variations. These findings highlight increasing groundwater vulnerability driven by climatic pressures and land-cover changes, emphasizing the need for adaptive land-water management strategies under semi-arid conditions.