Use of regression models to link raw water characteristics to trihalomethane concentrations in drinking water

Abstract

A study was conducted to examine the effect of raw water bromide on the formation and distribution of trihalomethanes (THMs) in finished drinking water. Twenty major water supplies in East Tennessee were selected for their significant levels of bromine-containing THMs, as demonstrated by previous studies. The cities were sampled quarterly for raw water pH, temperature, NVTOC, and bromide content, as well as finished water pH, NVTOC, and applied chlorine dose, coupled with conjunctive measurement for 7-day THMs. Few data for bromide levels in natural waters are currently available since most conventional colorimetric applications lack desired sensitivity, as does direct ion chromatography (IC). Although Fishman's kinetic permanganate method is adequate in the 10–100 μp 1 −1 range, it is time-consuming, and prone to certain interferences. An IC method using a sample preconcentration column has been evaluated. The method produced a 1 μg 1 −1 minimum detection level using deionized water based standards, with 3% relative precision completed at a standard concentration of 1000μg 1 −1. Bromide levels in the raw waters sampled were found to range from 10 to 225 μg 1 −1. Concurrent with the field sampling, laboratory chlorination experiments were conducted using Tennessee River water, under controlled laboratory conditions of bromide level, chlorine dose, pH, ionic strength, temperature, and organic precursor concentration. Resultant THM formations were monitored over a 96 h reaction period. The results concurred with observations made in previous work by such researchers as Bird and Rook. Nonlinear regression models for THM formation were generated using the laboratory chlorination data with respect to pH, temperature, chlorine dose, bromide, and NVTOC level. Actual values for these variables were substituted into the regression models, using the seasonal field data. Resultant predictive THM values were then compared with actual THM values for those data sampled. In general, these models were found to give acceptable fits. Overall 74.1% of the predicted values were within ± 15% of the measured values.