Scholarlink

<p>The degradation of groundwater quality due to human activities and environmental changes has become a global challenge, especially in semi-arid regions that are highly dependent on these resources. This study presented a comprehensive mapping of Groundwater Quality (GQ) distribution in Northwest Iran, utilizing a combination of Deep Learning (DL) and Machine Learning (ML) algorithms alongside the Borda scoring method for validation. Based on research methodology, the study focused on a range of WQ parameters, including potassium (K<sup>+</sup>), sodium (Na<sup>+</sup>), magnesium (Mg<sup>2+</sup>), calcium (Ca<sup>2+</sup>), sulfate (SO<sub>4</sub><sup>2−</sup>), chloride (Cl<sup>−</sup>), bicarbonate (HCO<sub>3</sub><sup>−</sup>), pH levels, Total Dissolved Solids (TDS), and Electrical Conductivity (EC). After creating raster maps of the chemical parameters of GQ, critical and non-critical quality points were identified using the Borda scoring algorithm based on Game Theory (GT). Once these points were established, DL and ML algorithms were employed in Python to spatially map GQ. Finally, water quality classes were categorized into five levels—very low, low, moderate, high, and very high—using ArcGIS software. Key quantitative findings included Total Hardness (TH) values ranging from 2.06&#xa0;mg/L in Abgharm to 31.92&#xa0;mg/L in Hamid Bolaghi, and Sodium Adsorption Ratio (SAR) values between 0.1 and 1.89, indicating varying soil sodicity risks. Strong correlations were observed among specific ions, notably a correlation of 0.93 between Na<sup>+</sup> and Cl<sup>−</sup>. Borda scoring highlighted significant groundwater concerns, with the highest score of 260 at sampling point 26, while points 21, 25, and 33 showed lower scores of 32, 39, and 70, respectively. Based on the results, LSTM emerged as the most effective model among the DL algorithms, exhibiting the lowest MAE (0.12) and MSE (0.02), along with a high R² value of 0.89 and an AUC of 0.94. Spatial zoning identified areas of very low GQ, particularly in the central and eastern regions, which necessitated targeted management interventions. This research underscored the effectiveness of combining ML and DL methodologies with scoring systems to enhance GQ assessments, thereby fostering improved water management practices in the region.</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

Mapping groundwater quality distribution in Northwest Iran: combining machine and deep learning and Borda scoring algorithms

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Ali Nasiri Khiavi,
  • Mir Masoud Kheirkhah Zarkesh,
  • Bagher Ghermezcheshmeh,
  • Bayramali Beyrami

摘要

The degradation of groundwater quality due to human activities and environmental changes has become a global challenge, especially in semi-arid regions that are highly dependent on these resources. This study presented a comprehensive mapping of Groundwater Quality (GQ) distribution in Northwest Iran, utilizing a combination of Deep Learning (DL) and Machine Learning (ML) algorithms alongside the Borda scoring method for validation. Based on research methodology, the study focused on a range of WQ parameters, including potassium (K+), sodium (Na+), magnesium (Mg2+), calcium (Ca2+), sulfate (SO42−), chloride (Cl−), bicarbonate (HCO3−), pH levels, Total Dissolved Solids (TDS), and Electrical Conductivity (EC). After creating raster maps of the chemical parameters of GQ, critical and non-critical quality points were identified using the Borda scoring algorithm based on Game Theory (GT). Once these points were established, DL and ML algorithms were employed in Python to spatially map GQ. Finally, water quality classes were categorized into five levels—very low, low, moderate, high, and very high—using ArcGIS software. Key quantitative findings included Total Hardness (TH) values ranging from 2.06 mg/L in Abgharm to 31.92 mg/L in Hamid Bolaghi, and Sodium Adsorption Ratio (SAR) values between 0.1 and 1.89, indicating varying soil sodicity risks. Strong correlations were observed among specific ions, notably a correlation of 0.93 between Na+ and Cl−. Borda scoring highlighted significant groundwater concerns, with the highest score of 260 at sampling point 26, while points 21, 25, and 33 showed lower scores of 32, 39, and 70, respectively. Based on the results, LSTM emerged as the most effective model among the DL algorithms, exhibiting the lowest MAE (0.12) and MSE (0.02), along with a high R² value of 0.89 and an AUC of 0.94. Spatial zoning identified areas of very low GQ, particularly in the central and eastern regions, which necessitated targeted management interventions. This research underscored the effectiveness of combining ML and DL methodologies with scoring systems to enhance GQ assessments, thereby fostering improved water management practices in the region.