Modeling residual chlorine and disinfection by-products (DBPs) dynamics in urban sewers during COVID-19 disinfection practices: A comparative analysis of process-based and data-driven approache

Abstract

The COVID-19 pandemic has intensified chlorine-based disinfection, elevating risks from residual chlorine and disinfection by-products (DBPs) in sewers. Using a pilot-scale sewer system with MS2 bacteriophage (SARS-CoV-2 surrogate), we stimulated wastewater collection and transportation process, and compared process-based (reaction kinetics) and data-driven models (random forest, decision tree, deep learning, general additive model, stacked model) under static (collection) and dynamic (transport) scenarios. The experimental results showed that the changes in residual chlorine and DBPs concentration in the dynamic scenario were more complex than in the static scenario, and higher residual chlorine dose didn’t accelerate the inactivation of the virus. According to analyses, data-driven models showed superior accuracy for residual chlorine prediction (R² +0.03) but poorer robustness for DBPs (MAE +0.35 vs. process-based), while process-based models exhibited smaller RMSE increases (2.91 vs. 5.31) when predicting DBPs versus chlorine, reflecting their adaptability to complex chlorine-organic matter interactions driving DBP formation. Uncertainty analysis revealed data-driven models’ sensitivity to high initial residual chlorine and DBPs doses. As the global situation is still rapidly evolving with a more frequent outbreak of epidemic events, our study provides a tool for estimating hazardous substances production caused by sterilization behavior for pandemic prevention.