Predictive capability of THM models for drinking water treatment and distribution

Abstract

Research and practice suggest markers of drinking water quality such as trihalomethanes (THMs), can change during treatment and distribution, potentially elevating health risk of end users. Models have been developed to predict THM formation at drinking water treatment plants (DWTP), in drinking water distribution systems (DWDS), and to a lesser extent, building premise plumbing (PP). The goal of this research was to evaluate the performance of published THM models and their development methodology, with the purpose of improving future THM model development. Water quality variable data were collected from literature and used as inputs for collected models. Mean and variance of model prediction values were used to measure THM model performance compared to THM data trends from literature. The research found differences in model formulation, water quality variable selection, and model development practices, despite evaluated models being statistical in nature. These differences lead to substantial inconsistencies in model output behavior. Diversity of data used for model development was found to be the most important factor for generalizable model prediction capabilities. Following these findings, a new framework was proposed to encourage novel strategies, data sharing, and collaboration among researchers and practitioners to improve THM model development, application, and performance. Potential use of machine learning techniques for future model development was also discussed based on findings.