Scholarlink

<p>Groundwater vulnerability assessment is essential for sustainable water resource management. The widely used DRASTIC framework provides a practical approach for intrinsic vulnerability mapping ; however its reliance on fixed parameter weights and linear aggregation limits its ability to capture site-specific hydrogeological complexity and observed contamination patterns. This study develops an interpretable machine learning framework by integrating the typical DRASTIC framework with a Random Forest (RF) regression algorithm and SHapley Additive exPlanations (SHAP), applied to the Meshginshahr Plain aquifer in northwestern Iran. Using the seven DRASTIC parameters, the intrinsic DRASTIC Vulnerability Index (DVI) was calculated and calibrated using observed nitrate measurements to derive a nitrate-adjusted vulnerability index as the target variable of the RF model. Model performance was assessed using Pearson and distance correlation metrics. Results show that the RF-enhanced DRASTIC framework substantially outperforms the typical DRASTIC framework, increasing the Pearson correlation with nitrate concentrations from 0.18 to 0.72 and the distance correlation from 0.49 to 0.71 in the test dataset. The improved vulnerability map exhibits clearer spatial correspondence with observed nitrate contamination, particularly in high-risk aquifer zones. SHAP-based interpretation reveals Soil Media and Aquifer Media as the dominant controls, followed by Depth to Water and Vadose Zone characteristics, while Topography plays a negligible role at the study scale. The spatial variability of SHAP contributions highlights the importance of local hydrogeological conditions and confirms the limitations of uniform weighting schemes. These findings demonstrate the significant potential of interpretable machine learning methods to enhance groundwater vulnerability assessment and support more robust and reliable groundwater management strategies.</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

An interpretable random forest–enhanced DRASTIC framework for groundwater vulnerability mapping in the Meshginshahr Plain, northwestern Iran

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Fatemeh Kadkhodaie,
  • Asghar Asghari Moghaddam,
  • Rahim Barzegar

摘要

Groundwater vulnerability assessment is essential for sustainable water resource management. The widely used DRASTIC framework provides a practical approach for intrinsic vulnerability mapping ; however its reliance on fixed parameter weights and linear aggregation limits its ability to capture site-specific hydrogeological complexity and observed contamination patterns. This study develops an interpretable machine learning framework by integrating the typical DRASTIC framework with a Random Forest (RF) regression algorithm and SHapley Additive exPlanations (SHAP), applied to the Meshginshahr Plain aquifer in northwestern Iran. Using the seven DRASTIC parameters, the intrinsic DRASTIC Vulnerability Index (DVI) was calculated and calibrated using observed nitrate measurements to derive a nitrate-adjusted vulnerability index as the target variable of the RF model. Model performance was assessed using Pearson and distance correlation metrics. Results show that the RF-enhanced DRASTIC framework substantially outperforms the typical DRASTIC framework, increasing the Pearson correlation with nitrate concentrations from 0.18 to 0.72 and the distance correlation from 0.49 to 0.71 in the test dataset. The improved vulnerability map exhibits clearer spatial correspondence with observed nitrate contamination, particularly in high-risk aquifer zones. SHAP-based interpretation reveals Soil Media and Aquifer Media as the dominant controls, followed by Depth to Water and Vadose Zone characteristics, while Topography plays a negligible role at the study scale. The spatial variability of SHAP contributions highlights the importance of local hydrogeological conditions and confirms the limitations of uniform weighting schemes. These findings demonstrate the significant potential of interpretable machine learning methods to enhance groundwater vulnerability assessment and support more robust and reliable groundwater management strategies.