Proposed formulation of surface water quality and modelling using gene expression, machine learning, and regression...

Area of Science:

• Environmental Science
• Water Resource Management
• Computational Hydrology

Background:

• Rising anthropogenic pollution necessitates advanced water quality prediction models.
• Existing models face limitations with data timespans and empirical expression derivation.
• Surface water quality in the Upper Indus River basin requires robust predictive tools.

Purpose of the Study:

• To model and derive empirical equations for surface water quality using a 30-year dataset.
• To identify the most reliable machine learning model for predicting river water quality (TDS and EC).
• To analyze parameter sensitivity and provide effective monitoring tools.

Main Methods:

• Utilized a 30-year dataset for surface water quality analysis.
• Employed machine learning techniques including Gene Expression Programming (GEP), Artificial Neural Network (ANN), and regression models.
• Evaluated model performance using Nash-Sutcliffe efficiency (NSE), RMSE, R², and MAE, with k-fold cross-validation.

Main Results:

• All developed models showed strong correlations (NSE and R² > 0.85).
• Gene Expression Programming (GEP) demonstrated superior performance over ANN and regression models for predicting TDS and EC.
• Bicarbonate was identified as the most sensitive parameter affecting TDS and EC.

Conclusions:

• GEP provides accurate and reliable empirical equations for predicting surface water quality (TDS and EC).
• The derived GEP equations can aid authorities in effective river water quality monitoring.
• This research contributes to addressing limitations in current water quality prediction models.