Optimizing water quality index using machine learning: a six-year comparative study in riverine and reservoir systems

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Climate change and human activities threaten water quality globally.
• Existing Water Quality Index (WQI) models face challenges in parameter weighting, aggregation, and transparency.
• Robust water quality assessments are crucial for effective water resource management and safe drinking water.

Purpose of the Study:

• To develop and validate an improved Water Quality Index (WQI) model framework.
• To compare the performance of different machine learning algorithms, weighting methods, and aggregation functions for WQI.
• To identify critical water quality parameters and optimize monitoring efficiency in surface water bodies.

Main Methods:

• A comparative optimization framework integrating three machine learning algorithms (including Extreme Gradient Boosting - XGBoost) and various weighting/aggregation methods.
• Analysis of six-year monthly water quality data (2017-2022) from 31 sites in the Danjiangkou Reservoir (DJKR) riverine and reservoir systems.
• Development of a novel Bhattacharyya mean WQI model (BMWQI) coupled with the Rank Order Centroid (ROC) weighting method.

Main Results:

• The XGBoost model demonstrated superior performance, achieving 97% accuracy for river sites.
• The proposed BMWQI model with ROC weighting significantly reduced uncertainty compared to other WQI models (17.62% for rivers, 4.35% for reservoirs).
• Key pollutants identified included total phosphorus (TP), permanganate index, and ammonia nitrogen for rivers, and TP and water temperature for reservoirs in the DJKR system.

Conclusions:

• Machine learning and innovative aggregation functions substantially improve surface water quality assessment methodologies.
• The developed framework offers practical value for water quality policy formulation by enabling targeted pollutant identification.
• The adaptable framework has broad applicability across diverse aquatic systems for optimized monitoring and management.