Multivariate Analysis of Large µ-FTIR Data Sets in Search of Microplastics

Abstract

µ-FTIR spectroscopy is a widely used technique in microplastics research. It allows to simultaneously characterize the material of the small particles, fibers or fragments, and to specify their size distribution and shape. Modern detectors offer the possibility to perform two-dimensional imaging of the sample providing detailed information. However, data sets are often too large for manual evaluation calling for automated microplastic identification. Library search based on the comparison with known reference spectra has been proposed to solve this problem. To supplement this ‘targeted analysis’, an exploratory approach was tested. Principal component analysis (PCA) was used to drastically reduce the size of the data set while maintaining the significant information. Groups of similar spectra in the prepared data set were identified with cluster analysis. Members of different clusters could be assigned to different polymer types whereas the variation observed within a cluster gives a hint on the chemical variability of microplastics of the same type. Spectra labeled according to the respective cluster can be used for supervised learning. The obtained classification was tested on an independent data set and results were compared to the spectral library search approach.