Scholarlink

<p>Groundwater is a critical resource in Iran’s border regions, where surface-water scarcity has intensified reliance on subsurface reserves, leading to overextraction and rapid depletion. Sustainable management in these arid areas demands high-resolution, continuous data, yet field-based monitoring remains limited by cost and logistical challenges. Satellite remote sensing, particularly the GRACE mission, provides essential large-scale terrestrial water storage anomaly (TWSA) estimates but suffers from coarse spatial resolution that constrains local applications. This study introduces a machine learning-based framework to downscale GRACE-derived terrestrial water storage anomaly and simulate groundwater level change at a finer resolution of 0.25° (~&#xa0;25&#xa0;km). A random forest model was applied to refine GRACE data from 1 to 0.25° resolution using predictors such as precipitation, evapotranspiration, land surface temperature, and vegetation indices. The downscaled dataset, combined with ancillary hydrological variables, supported the development of a second random forest model for monthly groundwater level change prediction, validated against in situ piezometric data. Results indicated strong model performance, with <i>R</i><sup>2</sup> values of 0.90 and 0.74 for training and testing phases, respectively, confirming the framework’s ability to capture groundwater fluctuations across diverse aquifers. The study highlights the potential of integrating downscaled satellite observations with machine learning to enhance groundwater assessment and support data-driven management in water-stressed environments.</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

A machine learning framework for downscaling and predicting groundwater levels using random forest and satellite data: a case study in Iran

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Mohammad Rostami Khalaj ,
  • Hamzeh Noor,
  • Mohmood Arjmand Sharif

摘要

Groundwater is a critical resource in Iran’s border regions, where surface-water scarcity has intensified reliance on subsurface reserves, leading to overextraction and rapid depletion. Sustainable management in these arid areas demands high-resolution, continuous data, yet field-based monitoring remains limited by cost and logistical challenges. Satellite remote sensing, particularly the GRACE mission, provides essential large-scale terrestrial water storage anomaly (TWSA) estimates but suffers from coarse spatial resolution that constrains local applications. This study introduces a machine learning-based framework to downscale GRACE-derived terrestrial water storage anomaly and simulate groundwater level change at a finer resolution of 0.25° (~ 25 km). A random forest model was applied to refine GRACE data from 1 to 0.25° resolution using predictors such as precipitation, evapotranspiration, land surface temperature, and vegetation indices. The downscaled dataset, combined with ancillary hydrological variables, supported the development of a second random forest model for monthly groundwater level change prediction, validated against in situ piezometric data. Results indicated strong model performance, with R2 values of 0.90 and 0.74 for training and testing phases, respectively, confirming the framework’s ability to capture groundwater fluctuations across diverse aquifers. The study highlights the potential of integrating downscaled satellite observations with machine learning to enhance groundwater assessment and support data-driven management in water-stressed environments.