Modeling source water disinfection byproducts formation potential using environmental variables

Predictive models of disinfection byproducts (DBPs) formation in treated drinking water have been widely used to guide operational decisions. However, very few studies have addressed the issue of managing DBPs through watershed protection programs and proactive management of water supply systems through predictive modeling of DBP formation potential in source waters. Here, we propose a two-component, simple statistical approach to predict the formation potentials of the sum of five haloacetic acids (HAA5fp) and total trihalomethanes (TTHMfp) in source water streams using environmental variables and ultraviolet absorbance at 254 nm wavelength (UV₂₅₄) as a surrogate for DBP precursors. In the first component of the model, using three feature selection regression models and cross-validation of subsets of the selected predictors, we identified three commonly monitored variables—streamflow, soil temperature, and total phosphorus for predicting UV₂₅₄. In the second component, HAA5fp and TTHMfp are predicted from UV₂₅₄. The approach is successfully demonstrated for two source water streams of the New York City water supply system (R² was 0.8, and 0.7–0.8 for the two model components). Long-term predictions of HAA5fp and TTHMfp showed distinct seasonal patterns that are linked to differences in land uses of the two watersheds. Moreover, sensitivity analysis showed that transport processes were important in one watershed whereas production processes were more important in the other.