Rapid and uncertainty-aware source tracing of accidental river pollution using surrogate-assisted Bayesian inversion

Area of Science:

• Environmental Science
• Water Quality Management
• Computational Hydrology

Background:

• Accurate and rapid identification of accidental pollutant releases in rivers is critical for mitigating ecological damage.
• Traditional inverse simulation methods are often too slow for effective emergency response.
• Existing models struggle with real-time data assimilation and uncertainty quantification for pollution source tracing.

Purpose of the Study:

• To develop a rapid, surrogate-assisted Bayesian inversion framework for identifying sudden surface water pollution sources.
• To enhance the speed and accuracy of emergency river pollution source tracing.
• To quantify conditional uncertainty associated with pollution source estimations.

Main Methods:

• A process-based hydrodynamic-water quality model was employed to create a virtual accident database.
• A re-optimized Jellyfish Search-based kernel extreme learning machine (JS-KELM) was utilized as a fast surrogate model.
• The surrogate model was integrated with a Bayesian-Particle Swarm Optimization (PSO) inversion scheme.

Main Results:

• The JS-KELM surrogate model demonstrated high accuracy (R²=0.9565, Pearson's r=0.9781) in predicting pollutant source-response relationships.
• The Bayesian inversion framework achieved accurate source variable estimation (mean relative error of 5.12% under multi-section monitoring).
• The framework successfully reconstructed pollutant concentrations (R² > 0.996) even under reduced monitoring conditions.

Conclusions:

• The developed surrogate-assisted Bayesian inversion framework offers a rapid, physically informed, and uncertainty-aware solution for emergency river pollution source tracing.
• The JS-KELM surrogate provides a robust and efficient alternative to traditional methods for complex environmental modeling.
• The framework's transferability was validated, indicating its potential applicability to various river systems.