Assessment of regulated and unregulated disinfection byproduct formation using chlorine dioxide pre-oxidation in...

Chlorine dioxide (ClO₂) has been emerging as an alternative to chlorine for disinfection due to the lower formation of regulated organic disinfection byproducts (DBPs). This pilot-scale study investigated the impacts of ClO₂ pre-oxidation and delayed chlorination on regulated and unregulated DBPs. These included trihalomethanes (THMs), haloacetic acids (HAAs), halonitromethanes (HNMs), chloral hydrate, chlorite, and chlorate. Both regulated and unregulated DBPs in the pilot system were found to be approximately 10 % of the levels seen at the full-scale plant that uses chlorine and chloramines. To assist in the prediction of future DBP levels (THMs, 5 HAAs, chlorite, and chlorate) after the adoption of ClO₂, an evaluation of the most effective modeling technique was conducted to compare artificial neural network (ANN), extreme gradient boost (XGBoost), and support vector machine (SVM) regression techniques to predict DBP levels based on raw water quality. A novel approach of using composite model predictions from leave-one-out cross-validation (LOOCV) sub-models was further developed to reduce variance and improve generalizability of data-driven models. The composite XGBoost model was selected as the best fit for all DBP species, with test root mean square errors between 0.11 - 3.32 μg/L of individual DBP species. The resulting models offer a foundation for full-scale application and decision support in optimizing treatment strategies to minimize DBP formation while maintaining regulatory compliance and adequate disinfection.