Scholarlink

Loktak Lake, the largest freshwater lake in northeast India, is currently threatened by increasing pollution due to rapid urbanisation and uncontrolled anthropogenic activities. Given the critical state of the lake, the objective of this study is twofold: to develop optimised data-driven systems for rapid and accurate water quality assessment and to implement model interpretation techniques for effective water quality management. The methodology used the entropy-weighted water quality index (WQI) for a comprehensive water assessment and utilised self-organising maps (SOM) for visualising complex interactions between water quality parameters and the development of a robust data-driven system using gradient boosting machine (GBM), eXtreme Gradient Boosting (XGBoost) and Convolutional Neural Network (CNN) models optimised via a grid search. Subsequent interpretation of the best model utilised explainable artificial intelligence (XAI) techniques such as SHAP values, 3D interaction diagrams and locally interpretable model-agnostic explanations (LIME) for a nuanced site-specific assessment of water quality. The results show that the WQI at the different sites is between 80.59 and 100.00. The distribution of water quality across in the lake is quantitatively categorised into four categories as 22% ‘Good’, 31% ‘Slightly Polluted’, 24% ‘Moderately Polluted’ and 23% ‘Heavily Polluted’, illustrating the varying degrees of pollution in the lake. The XGBoost model proved to be the most accurate, with average precision, recall and an F1 score that was close to perfect at 0.99. Explainable AI techniques such as SHAP values highlighted the impact of turbidity, COD and BOD on water quality predictions, while LIME analysis provided site-specific insights that enabled targeted remedial strategies. The 3D interaction diagrams illustrate the complex relationships between key water quality parameters and show that a combined increase in dissolved oxygen (DO) and biological oxygen demand (BOD) significantly affects water quality predictions. Furthermore, the interactions between total dissolved solids (TDS) and chemical oxygen demand (COD) indicate that higher concentration of these parameters tend to synergistically exacerbate water pollution. Therefore, the study not only emphasises the urgency of the ecological threats facing Lake Loktak, but also demonstrates the effectiveness of sophisticated data-driven modelling and interpretive techniques in environmental decision-making. The use of such advanced tools provides policy makers and environmental managers with a scientific basis to develop precise and effective measures aimed at preserving the integrity of this vital aquatic ecosystem and to provide potable water in the region.

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

Water Pollution Assessment and Management Through Interpreting Black Box Deep Learning Algorithms in the Loktak Lake

  • 章节

  • 发表日期：

    2025年

查看全文
  • Swapan Talukdar,
  • Ishita Afreen Ahmed,
  • Shahfahad,
  • Mirza Razi Imam Baig,
  • Mohd Rihan,
  • Atiqur Rahman

摘要

Loktak Lake, the largest freshwater lake in northeast India, is currently threatened by increasing pollution due to rapid urbanisation and uncontrolled anthropogenic activities. Given the critical state of the lake, the objective of this study is twofold: to develop optimised data-driven systems for rapid and accurate water quality assessment and to implement model interpretation techniques for effective water quality management. The methodology used the entropy-weighted water quality index (WQI) for a comprehensive water assessment and utilised self-organising maps (SOM) for visualising complex interactions between water quality parameters and the development of a robust data-driven system using gradient boosting machine (GBM), eXtreme Gradient Boosting (XGBoost) and Convolutional Neural Network (CNN) models optimised via a grid search. Subsequent interpretation of the best model utilised explainable artificial intelligence (XAI) techniques such as SHAP values, 3D interaction diagrams and locally interpretable model-agnostic explanations (LIME) for a nuanced site-specific assessment of water quality. The results show that the WQI at the different sites is between 80.59 and 100.00. The distribution of water quality across in the lake is quantitatively categorised into four categories as 22% ‘Good’, 31% ‘Slightly Polluted’, 24% ‘Moderately Polluted’ and 23% ‘Heavily Polluted’, illustrating the varying degrees of pollution in the lake. The XGBoost model proved to be the most accurate, with average precision, recall and an F1 score that was close to perfect at 0.99. Explainable AI techniques such as SHAP values highlighted the impact of turbidity, COD and BOD on water quality predictions, while LIME analysis provided site-specific insights that enabled targeted remedial strategies. The 3D interaction diagrams illustrate the complex relationships between key water quality parameters and show that a combined increase in dissolved oxygen (DO) and biological oxygen demand (BOD) significantly affects water quality predictions. Furthermore, the interactions between total dissolved solids (TDS) and chemical oxygen demand (COD) indicate that higher concentration of these parameters tend to synergistically exacerbate water pollution. Therefore, the study not only emphasises the urgency of the ecological threats facing Lake Loktak, but also demonstrates the effectiveness of sophisticated data-driven modelling and interpretive techniques in environmental decision-making. The use of such advanced tools provides policy makers and environmental managers with a scientific basis to develop precise and effective measures aimed at preserving the integrity of this vital aquatic ecosystem and to provide potable water in the region.