Principal component regression model applied to dimensionally reduced spectral fluorescent signature for the determination of organic character and THM formation potential of source water

The characterization of dissolved organic matter (DOM) in source water not only is central to the study of precursors to disinfection by-products (DBPs), but can also aid in controlling the discharge of potentially harmful organic chemicals in water bodies. Rapid determination of six DOM fraction concentrations provides an added advantage in understanding the organic character of water in comparison to the measure of dissolved organic carbon (DOC), which is an aggregate parameter typically used by water purveyors. The experimental procedure for DOM isolation and fractionation by ionic resins is lengthy and tedious. Many attempts have been made towards the development of faster and reliable techniques including statistical analysis applied to spectral fluorescent signature (SFS). Fluorescence is a very sensitive technique and works best only at certain wavelengths that are different for different materials. It is therefore difficult to quantify a material using fluorescence technique, especially when the entire fluorescence matrix is considered. To address this difficulty, an innovative two-stage processing technique is developed in this research in order to build an enhanced, more robust empirical model. At stage I, the dimensionality of the input data is reduced by focusing on specific portion of the entire matrix obtained by applying scatter removal, peak analysis, and coefficient of variation (CV) analysis. Then statistical analysis in the form of principal component regression (PCR) follows as stage II. In addition, the same technique is applied to predict trihalomethanes formation potential (THMFP). This model provides better sensitivity and accuracy. while maintaining the advantages of the SFS technique for rapid identification and quantification of DOM fractions. (c) 2009 Elsevier B.V. All rights reserved.