Understanding and modeling the formation of disinfection by-products in urban domestic sewage under intensified chlorination

Abstract

Intensified chlorination during public health emergencies significantly increases disinfection by-products (DBPs) formation in urban domestic sewage, yet their occurrence, toxicity and predictability remain poorly understood. In this study, real sewage was subjected to batch chlorination under varied conditions to investigate trihalomethanes (THMs), haloacetonitriles (HANs), and residual chlorine (RC) in urban domestic sewage. High DBP concentrations were observed, with THMs and HANs reaching 62.0–441.1 and 43.6–254.2 μg/L respectively. HANs drove the comprehensive toxicity (1–2 orders of magnitude higher toxicity than THMs), with the primary toxicity shifting dynamically from bromochloroacetonitrile (≤2 h) to dibromoacetonitrile (DBAN, ≥10 h). Meanwhile, RC remained as high as 27.4 mg/L after 192 h, potentially threatening downstream biological treatment. Among four evaluated models, Random Forest (RF) model achieved the best DBP prediction performance (R2 of 0.722–0.976). Crucially, coupling this model with interpretable tools, Partial Dependence Plots and Shapley Additive exPlanations, successfully decoded the complex in-sewer mechanisms. The interpretable framework revealed that extended contact times in sewage were associated with a data-driven shift to a chloramine-dominated system, driving the late-stage accumulation of DBAN under high Cl₀ and intermediate Cl/N. This study demonstrates that intensified chlorination elevates DBP toxicity and sustains high RC, and that integrating this risk characterization with interpretable machine-learning models provides a practical basis for risk-oriented management of chlorination practices in sewage networks. Limited by static data, the RF model is applicable to urban sewage under uniform hydraulic conditions; extrapolation to more complex systems requires validation with dynamic or field data.