A novel approach for estimating and predicting uncertainty in water quality index model using machine learning...

Area of Science:

• Environmental Science
• Water Resource Management
• Statistical Modeling

Background:

• Water Quality Index (WQI) models are increasingly used globally, but lack standardized guidelines, leading to accuracy and reliability issues.
• WQI models face uncertainties in indicator selection, sub-index calculation, weighting, and aggregation, impacting overall assessment.
• Assessing and quantifying these uncertainties is critical for reliable water quality management.

Purpose of the Study:

• To develop and present a statistically sound methodology for assessing uncertainties in Water Quality Index (WQI) models.
• To identify the primary sources of uncertainty within WQI models and their contribution to overall reliability.
• To evaluate the performance of different aggregation functions for coastal water quality assessment.

Main Methods:

• Employed Monte Carlo simulation (MCS) to estimate overall model uncertainty.
• Utilized Gaussian Process Regression (GPR) to predict site-specific uncertainties.
• Analyzed eight different WQI models, focusing on sub-index functions, indicator selection, weighting, and aggregation methods.

Main Results:

• Sub-index functions were identified as major contributors to WQI uncertainty (12.86% in summer, 10.27% in winter).
• Water quality indicator selection and weighting processes yielded low uncertainty (<1%).
• Significant statistical differences were observed between various aggregation functions; the weighted quadratic mean (WQM) and unweighted root mean squared (RMS) showed promise for coastal water quality.

Conclusions:

• The methodology provides a robust framework for understanding and quantifying WQI model uncertainties.
• Careful selection of sub-index functions is essential to enhance the reliability of WQI assessments.
• The weighted quadratic mean (WQM) and unweighted root mean squared (RMS) aggregation functions are recommended for coastal water quality evaluation due to reduced uncertainty.