Study on the effect of spectral difference coefficient on the precision of quantitative spectral analysis

Abstract

Due to noise, spectral multicollinearity (or overlapping) between an analyte and interferents seriously influences the precision of quantitative spectral analysis for the analyte in a multicomponent mixture. Removing the effects of the interferents is crucial to improving the precision of analysis. In order to allow the process to be more targeted, in this paper, the influence mechanism of the interferents is explored. Firstly, a spectral difference coefficient was proposed as the index that measured the degree of the spectral multicollinearity between the analyte and the interferents, then through simulations, the effectiveness and feasibility of the spectral difference coefficient were verified and the relation between it and analysis precision was investigated. The results showed that effects of the interferents on analysis precision related exclusively to spectral multicollinearity between them and the analyte, i.e., the spectral difference coefficient, and not to their number or shape of their spectra. The RMSEP (Root Mean Square Error of Prediction) was decreased by half when the spectral difference coefficient doubled. Moreover, the spectral difference coefficient was proved to be independent of the noise intensity, sensitivity of the analyte and the number of modeling wavelengths in affecting the analysis precision, which was more evident for the universal validity of the relation and offered the possibility of combining these factors together to improve the analysis precision by weighing their respective effects on the analysis precision. Thus, a new wavelength selection method based on the influence mechanisms of different factors was proposed. Finally, actual quantitative analysis of the ethanol in the ethanol–water solution was conducted, compared with full spectra and the other promising variable selection method, synergy interval PLS (siPLS); this new wavelength selection method showed better prediction ability. The study is helpful to understand the influence mechanism of interferents on model precision and so help to reduce their influence accordingly.