Resolving uncertainty in chemical speciation determinations

Speciation determinations involve uncertainty in system definition and experimentation. Identification of appropriate metals and ligands from basic chemical principles, analytical window considerations, types of species and checking for consistency in equilibrium calculations are considered in system definition uncertainty. A systematic approach to system definition limits uncertainty in speciation investigations. Experimental uncertainty is discussed with an example of proton interactions with Suwannee River fulvic acid (SRFA). A Monte Carlo approach was used to estimate uncertainty in experimental data, resulting from the propagation of uncertainties in electrode calibration parameters and experimental data points. Monte Carlo simulations revealed large uncertainties present at high (>9–10) and low (<4) ?logH⁺], which result from larger instabilities in the proton balance function in these regions. Uncertainties in speciation parameters were compared for uniresponse fitting (linear programming and least-squares) and multiresponse fitting. Linear programming and least-squares approaches both fit the observed data, but suggest different mixtures of monoprotic ligands. Least-squares fit the data with 21 sites, whereas linear programming fit the data equally well with 9 sites. Multiresponse fitting, involving simultaneous fluorescence and pH measurements, improved model discrimination. Deconvolution of the excitation versus emission fluorescence surface for SRFA establishes a minimum of five sites. Diprotic sites are also required for the five fluorescent sites, and one non-fluorescent monoprotic site was added to accommodate the pH data. Consistent with greater complexity, the multiresponse method had broader confidence limits than the uniresponse methods, but corresponded better with the accepted total carboxylic content for SRFA. Overall there was a 40% standard deviation in total carboxylic content for the multiresponse fitting, versus 10% and 1% for least-squares and linear programming, respectively.