A novel framework to predict water turbidity using Bayesian modeling.

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• High water turbidity poses a significant global challenge to aquatic ecosystems and water management.
• Accurate and cost-effective turbidity measurement is crucial, especially in resource-limited settings.
• Existing methods may lack the speed or accessibility required for rapid water quality assessment.

Purpose of the Study:

• To develop a novel, cost-effective framework for predicting water turbidity in various aquatic ecosystems.
• To quantify prediction uncertainty using Bayesian modeling.
• To implement a model-update mechanism for enhanced performance with new data.

Main Methods:

• Developed a Bayesian modeling framework to predict water turbidity from smartphone images.
• Collected 120 paired records of smartphone images and standard turbidity measurements from diverse Chinese aquatic environments.
• Implemented a model-update method to refine model structure and parameters as more data become available.

Main Results:

• Achieved high prediction accuracy with Nash-Sutcliffe efficiency (NS) >0.87 during training and NS >0.73 during validation.
• Demonstrated a decreasing trend in model uncertainty and stable model fit with the model-update method.
• Validated the framework's robust performance across rivers, lakes, and ponds.

Conclusions:

• The developed Bayesian-based framework provides a robust and accessible method for predicting water turbidity.
• The framework is valuable for water management, particularly in regions with limited resources.
• The model-update strategy enhances predictive power and reduces uncertainty over time.