Methods for unsupervised contribution analysis of raw EEM data in water monitoring. Contaminant identification and quantification

Fluorescence EEM spectra provide the fingerprint of water contamination and is a very efficient way to access the quality of water bodies. These multivariate datasets correspond to complex mixtures and are very rich in information. Graphical approaches have been used for decades to characterize and quantify different contamination sources. It is very important to resolve mixed signals in raw EEM spectra in terms of signal sources and respective composition profiles - signal sources allow the identification of contamination type, while concentration profiles quantify the respective contribution inside the mixtures. In order to be able to use robust modeling algorithms, the first task is to accurately estimate the number of contributions that are present. We demonstrate the ability of Singular value Decomposition (SVD) in accessing this information content in raw EEM datasets. To decompose raw EEM information, several algorithms are tested: PARAFAC, MCR-ALS and ICA. In this work we suggest a systematic unsupervised algorithm to process raw EEM spectra of water samples. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

Fluorescence EEM spectra serve as a fingerprint for assessing water contamination and quality, with rich information that requires resolving mixed signals and identifying contamination types. The study demonstrates the capabilities of algorithms such as SVD in processing information in raw EEM datasets.