Automatically identifying scatter in fluorescence data using robust techniques

First and second order Rayleigh and Raman scatter is a common problem when fitting Parallel Factor Analysis (PARAFAC) to fluorescence excitation–emission data (EEM). The scatter does not contain any relevant chemical information and does not conform to the low-rank trilinear model. The scatter complicates the analysis instead and contributes to model inadequacy. As such, scatter can be considered as an example of element-wise outliers. However, no straightforward method for identifying the scatter region can be found in the literature. In this paper an automatic scatter identification method is developed based on robust statistical methods. The method does not demand any visual inspection of the data prior to modeling, and can handle first and second order Rayleigh scatter as well as Raman scatter in various types of EEM data. The results of the automated scatter identification method were used as input data for three different PARAFAC methods. Firstly inserting missing values in the scatter regions are tested, secondly an interpolation of the scatter regions is performed and finally the scatter regions are down-weighted. These results show that the PARAFAC method to choose after scatter identification clearly depends on the data, for example signal to noise ratio and overlap between signal and scatter.