Scholarlink

<p>Escalating groundwater salinity threatens the sustainability of deep desert aquifers in Kébili, Southern Tunisia, where groundwater is the only dependable freshwater source. In this study, groundwater quality and salinity conditions were assessed using an integrated framework combining hydrochemical analysis, selected water quality indices, ensemble machine learning modeling, and GIS-based spatial analysis. Forty-five groundwater samples were analyzed for major ions and physicochemical parameters to evaluate suitability for domestic and agricultural use. Key water quality indices (including Water Quality Index (WQI), Sodium Adsorption Ratio (SAR), and Sodium Percentage (Na%) were calculated to characterize groundwater quality and irrigation suitability. To capture complex, nonlinear relationships governing groundwater salinity, ensemble learning models were developed to predict Total Dissolved Solids (TDS) and WQI, with model hyperparameters optimized through grid search. Feature importance analysis revealed the dominant hydrochemical controls on salinity evolution. In parallel, an unsupervised Kohonen Self-Organizing Map (K-SOM) was employed to classify groundwater samples into five distinct quality groups, ranging from very poor to very good, with the robustness of the classification confirmed by Principal Component Analysis (PCA). Spatial patterns of groundwater salinity and quality were mapped using ordinary kriging within a Geographic Information System (GIS). The results indicate that approximately 46% of the study area is affected by poor to very poor groundwater quality, primarily due to elevated salinity and nitrate concentrations, whereas only 24% exhibits good to very good quality associated with deeper confined aquifers. This study represents the first integrated application of hydrochemical indices, ensemble machine learning, unsupervised clustering, and GIS mapping to assess groundwater salinity in deep desert aquifers of southern Tunisia. The proposed framework offers a robust and transferable tool for groundwater management and salinity mitigation in arid and 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

Integrating Hydrochemical Indices and Machine Learning Models for Groundwater Quality Assessment in a Deep Desert Aquifer (Southern Tunisia)

  • 期刊论文

  • 发表日期：

    2026年7月4日

查看全文
  • Bilel Abdelkarim,
  • Latifa Dhaouadi,
  • Mohamed Haythem Msaddek,
  • Yahya Moumni,
  • IMHR Antunes,
  • Nabil Abaab

摘要

Escalating groundwater salinity threatens the sustainability of deep desert aquifers in Kébili, Southern Tunisia, where groundwater is the only dependable freshwater source. In this study, groundwater quality and salinity conditions were assessed using an integrated framework combining hydrochemical analysis, selected water quality indices, ensemble machine learning modeling, and GIS-based spatial analysis. Forty-five groundwater samples were analyzed for major ions and physicochemical parameters to evaluate suitability for domestic and agricultural use. Key water quality indices (including Water Quality Index (WQI), Sodium Adsorption Ratio (SAR), and Sodium Percentage (Na%) were calculated to characterize groundwater quality and irrigation suitability. To capture complex, nonlinear relationships governing groundwater salinity, ensemble learning models were developed to predict Total Dissolved Solids (TDS) and WQI, with model hyperparameters optimized through grid search. Feature importance analysis revealed the dominant hydrochemical controls on salinity evolution. In parallel, an unsupervised Kohonen Self-Organizing Map (K-SOM) was employed to classify groundwater samples into five distinct quality groups, ranging from very poor to very good, with the robustness of the classification confirmed by Principal Component Analysis (PCA). Spatial patterns of groundwater salinity and quality were mapped using ordinary kriging within a Geographic Information System (GIS). The results indicate that approximately 46% of the study area is affected by poor to very poor groundwater quality, primarily due to elevated salinity and nitrate concentrations, whereas only 24% exhibits good to very good quality associated with deeper confined aquifers. This study represents the first integrated application of hydrochemical indices, ensemble machine learning, unsupervised clustering, and GIS mapping to assess groundwater salinity in deep desert aquifers of southern Tunisia. The proposed framework offers a robust and transferable tool for groundwater management and salinity mitigation in arid and hyper-arid environments.