Innovative interpretable AI-guided water quality evaluation with risk adversarial analysis in river streams...

Area of Science:

• Environmental Science
• Water Resource Management
• Computational Hydrology

Background:

• Industrial pollution threatens water security, necessitating robust water quality assessments.
• Seasonal fluctuations and influential factors require effective strategies for decision-makers.
• The Han River system in South Korea faces complex water quality challenges.

Purpose of the Study:

• To introduce a novel, AI-driven approach for evaluating water quality in the Han River basin.
• To identify significant seasonal trends and factors influencing water quality.
• To develop tailored strategies for water quality enhancement and early warning systems.

Main Methods:

• Utilized monthly water characteristic data (2013-2022) from 14 locations.
• Calculated sub-index water quality index (s-WQI) and applied Hampel filtering and feature selection.
• Employed machine learning, specifically Light Gradient Boosting (LGB), for combinatorial prediction.
• Conducted seasonal Monte-Carlo simulations and SHAP analysis for risk assessment and factor contribution.

Main Results:

• Light Gradient Boosting (LGB) demonstrated superior predictive accuracy for s-WQI.
• Identified key water characteristics influencing water quality through SHAP analysis.
• Monte-Carlo simulations defined risk bounds for water quality parameters.
• The novel approach effectively evaluates water quality and identifies influential factors.

Conclusions:

• AI-based prediction and feature selection offer a transformative approach to water quality assessment.
• The study provides practical implications for water quality enhancement and early warning systems.
• This research advances data-driven environmental science in water resource management.