Calibration transfer: PLS1 calibration model using pure components

vilas's picture


Will the piecewise direct standardization work in the following case of calibration transfer?

The calibration dataset has the near infrared absorption spectra of 10 pure components. A PLS1 model is developed for each component. The constraint is that only ONE transfer sample can be obtained, since the instruments are at customer sites. The wavelength scale variation between instruments is very small. The calibration spectra of components are quite different from each other.

If PDS is not the solution in this case, then what other approaches we can try? Can we somehow interpolate the absorbances to obtain the spectra for different wavelength scales (or instruments)? Any advice would be greatly appreciated.

Thank you!

hlmark's picture

Vilas - you don't say anything about the nature of the samples. If they're all clear liquids, then you might try applying the CLS algorithm (see, for example, Applied Spectroscopy, 64(9), p.995 (2010)). But also note the conclusions of that paper: the coefficients produced by the CLS algorithm correspond to the component concentrations only when the concentrations are expressed as volume fractions.




dardenne's picture


One sample will never correct for a slope effect.

If there is effective peak shifts, I would try an algorithm like COW (Correlation Optimized Warping) to correct for a X shift and apply a bias/wavelength to correct for the Y axis based on the common spectra.

It would be easier to advise by seeing the data.