Evaluation and quality control of environmental analytical data from the Niagara River using multiple chemometric methods

The use of artificial neural networks (ANN), principal component analysis (PCA) and universal process modelling (UPM) to identify the source of water samples based on the variation of chemical data from these samples has been investigated. Chromatographic data sets generated from three locations on the Niagara River were used in this research. The concentrations of target organic compounds were chromatographically determined and used as classification features. Chromatographic variation between three sampling sites was determined over a one-year period and included 149 separate samples. Variation within sampling sites was evaluated otter a seven-year period. ANN and UPM techniques correctly identified the source of 95% of the water samples based on minor differences in the chromatographic data. PCA and UPM gave direct visualization of differences within chemical data sets. PCA and UPM were also found to be useful tools for the detection of chromatographic outliers from within sampling sires. The correlation between target compounds and surrogates are discussed. The results show that these methods are useful for the determination of the variation of target organic compounds over time both within and between sampling sires. The potential of these systems for monitoring analytical quality control based on entire data sets is also presented.