Tom Bond, Nigel Graham

Quantitative methods which link molecular descriptors for recognized precursors to formation of drinking water disinfection byproducts are scarce. This study aimed to develop a simple mathematical tool for predicting chloroform (trichloromethane) yields resulting from aqueous chlorination of model organic precursors. Experimental chloroform yields from 211 precursors were collated from 22 literature studies from 1977 onwards. Nineteen descriptors, some established and others developed during this study, were used as inputs in a multiple linear regression model. The final model, calibrated using five-way leave-many-out cross-validation, contains three descriptors. Two novel empirical descriptors, which quantify the impact of adjacent substituents on aromatic and enolizable chlorine substitution sites, were the most significant. The model has r(2) = 0.91 and a standard error of 8.93% mol/mol. Experimental validation, using 10 previously untested precursors, showed a mean discrepancy of 5.3% mol/mol between experimental and predicted chloroform yields. The model gives insight to the influence that specific functional groups, including hydroxyl, chlorine and carboxyl, have on chloroform formation and the relative contributions made by separate substitution sites in the same molecule. It is anticipated that the detailed approach can be updated and extended as new experimental data emerges, to encompass additional precursors and groups of disinfection byproducts.