Scholarlink

<p>Groundwater fluoride contamination in Tamil Nadu’s geochemically diverse aquifers necessitates high-fidelity, cost-effective monitoring. This study introduces a Geochemically-Aware Domain Adaptation (GADA) framework to predict fluoride concentrations in data-scarce regions.Query Using 2089 data samples from 32 districts of Tamilnadu collected from Central Ground Water Board, India, different domain adaptation models like Best Source Transfer (BST), Correlation Alignment (CORAL), and Feature-Weighted Data Augmentation (FWDA) strategy utilizing Gradient Boosting as a base estimator is developed. While direct model transfer showed poor performance, the GADA achieved a robust R<sup>2</sup>of 0.91 with a minimal standard deviation of 0.042, an MSE of 0.015, and an MAE of 0.094. Functioning as a Virtual Sensor, the model performs real-time inference using only three low-cost parameters: pH, Total Dissolved Solids (TDS), and Electrical Conductivity (EC). Field validation against Ion-Selective Electrode (ISE) measurements confirmed the system’s high predictive fidelity. This stable and accurate solution provides a scalable tool for continuous environmental monitoring and decentralized water governance in heterogeneous hydrogeological domains.</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

Geochemically aware domain adaptation (GADA) for real-time fluoride inference in heterogeneous aquifers of Tamil nadu, India

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Ananthi Sankararaman,
  • Kirubaveni Savarimuthu,
  • Gulam Nabi Alsath Mohammed

摘要

Groundwater fluoride contamination in Tamil Nadu’s geochemically diverse aquifers necessitates high-fidelity, cost-effective monitoring. This study introduces a Geochemically-Aware Domain Adaptation (GADA) framework to predict fluoride concentrations in data-scarce regions.Query Using 2089 data samples from 32 districts of Tamilnadu collected from Central Ground Water Board, India, different domain adaptation models like Best Source Transfer (BST), Correlation Alignment (CORAL), and Feature-Weighted Data Augmentation (FWDA) strategy utilizing Gradient Boosting as a base estimator is developed. While direct model transfer showed poor performance, the GADA achieved a robust R2of 0.91 with a minimal standard deviation of 0.042, an MSE of 0.015, and an MAE of 0.094. Functioning as a Virtual Sensor, the model performs real-time inference using only three low-cost parameters: pH, Total Dissolved Solids (TDS), and Electrical Conductivity (EC). Field validation against Ion-Selective Electrode (ISE) measurements confirmed the system’s high predictive fidelity. This stable and accurate solution provides a scalable tool for continuous environmental monitoring and decentralized water governance in heterogeneous hydrogeological domains.