Models for Conductance Measurements in Quality Assurance of Water Analysis

Models for the prediction of conductance in nonbrine water samples through the measurement of ionic concentrations and other parameters are compared. Such predictions are often used for quality assurance purposes by comparing them with actual measurements to determine whether gross analysis errors have been made. A currently recommended method for making such predictions is a semiempirical relation that adapts the Debye-Hiickel-Onsager equation to mixed electrolyte systems by incorporating modified definitions of ionic charge and concentration. The limitations of this model are examined, and extensions to it are considered. Other predictive methods, including multiple linear regression (MLR), principal components regression (PCR), partial least-squares (PLS) regression, continuum regression (CR), and neural networks (NN), are also considered. Models employ the concentrations of 10 ions as well as, in some cases, additional water quality measurements. Best results were obtained with an extended form of the Debye-Hiickel-Onsager equation and an optimized MLR model. PCR, PLS, CR, and NN did not offer significant advantages.