Scholarlink

<p>Freshwater scarcity is a critical challenge in arid regions, where groundwater supplies are increasingly stressed by limited recharge, intensive abstraction, and climate variability. Bahrain’s principal aquifers have experienced a long-term decline due to sustained over-extraction and constrained natural replenishment. This study presents a comparative machine-learning (ML) framework to predict monthly groundwater-level (GWL) variability using an integrated predictor set combining climatic variables (temperature, humidity, vapour pressure, evaporation, and rainfall) with groundwater extraction and a transboundary groundwater inflow variable. Six algorithms were benchmarked under a consistent hyperparameter optimization and validation workflow: multilayer perceptron (MLP), support vector regression (SMOreg), Gaussian process (GP), random forest (RF), and two tree-based ensemble models (bagging and additive regression with regression-tree base learners). Models were trained on 2011–2019 data and evaluated on an independent 2020–2021 test period; the same predictor set was further assessed by retraining models for Bahrain’s major aquifers (Khobar, Alat, and Umm Er-Radhuma). For the Khobar aquifer, the MLP achieved the strongest performance with a testing correlation coefficient (CC) of 0.8 and a root-mean-squared error (RMSE) of 0.24&#xa0;m. The optimal algorithm varied by aquifer; bagging (M5P) yielded the lowest testing RMSE for Alat, whereas the MLP ranked first for Umm Er-Radhuma. Sensitivity analysis showed the largest RMSE increases after removing groundwater from neighbouring countries (GNC), evaporation and rainfall, whereas vapour pressure and groundwater extraction had negligible effects when removed. Overall, the proposed framework provides an accurate and data-efficient tool for groundwater assessment and management in hyper-arid 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

Machine-Learning-Based Modelling of Groundwater Levels in Bahrain’s Aquifer System

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Reem Abdul-Rahman,
  • Sawsan AbdulRahman,
  • Badriya Altaweel,
  • Abdallah Accary

摘要

Freshwater scarcity is a critical challenge in arid regions, where groundwater supplies are increasingly stressed by limited recharge, intensive abstraction, and climate variability. Bahrain’s principal aquifers have experienced a long-term decline due to sustained over-extraction and constrained natural replenishment. This study presents a comparative machine-learning (ML) framework to predict monthly groundwater-level (GWL) variability using an integrated predictor set combining climatic variables (temperature, humidity, vapour pressure, evaporation, and rainfall) with groundwater extraction and a transboundary groundwater inflow variable. Six algorithms were benchmarked under a consistent hyperparameter optimization and validation workflow: multilayer perceptron (MLP), support vector regression (SMOreg), Gaussian process (GP), random forest (RF), and two tree-based ensemble models (bagging and additive regression with regression-tree base learners). Models were trained on 2011–2019 data and evaluated on an independent 2020–2021 test period; the same predictor set was further assessed by retraining models for Bahrain’s major aquifers (Khobar, Alat, and Umm Er-Radhuma). For the Khobar aquifer, the MLP achieved the strongest performance with a testing correlation coefficient (CC) of 0.8 and a root-mean-squared error (RMSE) of 0.24 m. The optimal algorithm varied by aquifer; bagging (M5P) yielded the lowest testing RMSE for Alat, whereas the MLP ranked first for Umm Er-Radhuma. Sensitivity analysis showed the largest RMSE increases after removing groundwater from neighbouring countries (GNC), evaporation and rainfall, whereas vapour pressure and groundwater extraction had negligible effects when removed. Overall, the proposed framework provides an accurate and data-efficient tool for groundwater assessment and management in hyper-arid environments.