Multiple linear regression models for predicting trihalomethane formation from chlorinated natural waters

Abstract

A USEPA survey in 1975 found trihalomethanes (THMs) to be widespread in chlorinated drinking water. A subsequent study identified humic substances as probable precursors. In 1976, the National Cancer Institute found chloroform to cause cancer in laboratory rats. Consequently, the USEPA set a maximum contaminant level (MCL) of 100 pg/1 for total THMs in drinking water systems. THM formation reactions between chlorine and humic substances have been studied but are not completely understood. Parameters found to affect THM formation rate and concentration are raw water pH, temperature, and bromide level; precursor source and concentration, chlorine dose, and chlorine contact (reaction) time. This research involved the formulation of empirically derived models, enabling THM prediction as a function of time and parameters, for natural waters. Kinetic experiments, designed to vary parameters, were conducted on seven geographically diverse natural waters. THM concentrations were analyzed by gas chromatography. Models were developed using multiple linear regression. Three predictive models were chosen based on statistical analysis and were tested on four natural waters. All models predicted well for ambient water conditions. No single model predicted better overall. Variations in the "reactivity" of waters due to precursor source were not adequately predicted, nor were chlorine-limited conditions.