Chemometric modeling and prediction of trihalomethane formation in Barcelona's water works plant

Formation and occurrence of trihalomethanes (CHCl3CHCl3, CHBr3CHBr3, CHCl2BrCHCl2Br, and CHBr2ClCHBr2Cl) are investigated in water chlorination disinfection processes in the Barcelona's water works plant (WWP). Twenty-three WWP variables were measured and investigated for correlation with trihalomethane formation. Multivariate statistical methods including principal component analysis (PCA), multilinear regression (MLR), stepwise MLR (SWR), principal component regression (PCR) and partial least squares regression (PLSR) have been used and compared to model and predict the complex behavior observed for the measured trihalomethane concentrations. The results, obtained by PCA as well as the evaluation of the statistical significance of the coefficients in the linear regression vectors, revealed that the most important WWP variables for trihalomethane formation were: water temperature, total organic carbon, added chlorine concentrations, UV absorbance and turbidity at different sites of the WWP, as well as other variables like wells supply flow levels and carbon filters age. Overall, MLR and PLSR methods performed the best and gave similar good predictive properties. Best results were obtained for the total sum of trihalomethane concentrations, TTHM, with average modeling and prediction relative errors of 12% and 16%, respectively. Among the individual trihalomethanes, the concentrations of CHBr3CHBr3 were the worst predicted ones with average modeling and prediction relative errors between 21–25% and 29–31%, respectively, followed by CHCl2BrCHCl2Br with 23–26% and 25–27%. Better predictions were obtained for the concentrations of CHBr2ClCHBr2Cl with relative modeling and prediction errors varying between 14–17% and 21%, and for the concentrations of CHCl3CHCl3 with 21–24% and 23–25% errors, respectively.