The performance of multivariate calibration depends on the spectral resolution and quality. For Fourier Transform (FT) NIR spectrometry, decreasing resolution suppresses spectral noise by sampling shorter interferograms. However, lower resolution has possibility to lose useful spectral information for quantifying analytes. The spectral resolution has not too much effect on Raman spectrometry equipped with charge coupled device (CCD), as long as enough spectral data points allow the latent variables to be accurately estimated. The impact of spectral preprocess methods is more complicated and relies on specific sample matrices.

The issue of spectral resolution and preprocess were studied by glucose determination in a set of standard mixture solutions of glucose, urea and lactate. Raman spectra were collected with a HoloLab 5000 R Raman spectrometer over the spectral range of 3450 – 100 cm-1. Near infrared spectra were collected with a Thermo-Nicolet Magna 550 Fourier transform spectrometer and these spectra covered the combination spectral range of 5000 – 4000 cm-1. A series of different spectral resolution and preprocess methods were applied to the collected spectra before they were used for quantification. All calibration models were generated from these modified spectra by partial least square regression analysis coupled with optimized spectral range and number of latent variables.

Results demonstrate optimized spectral resolution exists for NIR spectra to provide the best glucose determination but Raman spectra give very similar calibration performance when spectral resolution changes. However, spectral preprocess methods have relatively bigger effect on glucose measurement from Raman spectra than that from NIR spectra.