Meta-analysis of trihalomethane formation models and application to bromide intrusion

This research evaluated previously published trihalomethane (THM) formation models for their statistical robustness to be applied outside of their original calibration data set, and be used as a predictive tool at the water utility scale. All models predicted THM4 (i.e., the sum of the four chlorine- and bromine-containing THM species) based on the chlorination of natural waters and were developed using different combinations of precursor types (i.e., organic carbon concentration, UV-absorbing substances, and bromide concentration) and chlorination conditions (i.e., chlorine dose, pH, temperature, and time) as explanatory variables. All models were log (base 10) transformed into a common format, and were evaluated using a nationally representative water quality and THM4 formation data set based on the statistical metrics standard error, mean absolute percentage error, R2, and adjusted R2. The most robust log10(THM4) formation models had standard error equal to 0.226â€"0.262 and adjusted R2 equal to 0.696â€"0.783. The THM4 formation models that included bromide as an explanatory variable tended to under predict THM4 formation as a function of increasing bromide concentration. Overall, the results of this research show that several previously published THM4 formation models were developed with the appropriate explanatory variables and calibrated with a sufficiently broad data set such that the models may give a reasonable prediction of THM4 formation, depending on the level of accuracy desired, for chlorinated water conditions separate from the original calibration data set. However, the majority of THM4 formation models were not sensitive to changing bromide concentration, and as such, under predicted THM4 formation at bromide concentrations greater than 500 μg/L.