# How can I transfer a multivariate model between spectrometers without the use of standard samples?

Our lab has created several multivariate models to predict sample properties from infrared spectra. However, different spectrometers often give different responses, and often the model created on one instrument cannot be used with another without first using some method to transfer the calibration to the second instrument. I need a good method for transferring a multivariate model to a second spectrometer so that it still gives accurate predictions.

I have already come across a few methods:

• create entirely new model for second instrument
• piecewise direct standardization
• adding samples to the model that have been measured with the second instrument
• linear fit to the model output for the second instrument to correct the output (slope-bias)
• transfer using virtual standards

I have tried these, but the only ones that work for our models require several standard samples (samples with known property values or spectra) to be measured on the second instrument, which ends up taking almost as much time as simply creating a new calibration model. I would prefer a method that does not require this hassle. and I am hoping that someone can offer a better solution.

If you know the sets of coefficients $R_k$ which adjust the output for each wavelength band for both spectrometers, just take the corrected set from one spectrometer, divide by its correction coeffs, and multiply by the correction coefs from the other one. (or apply the full inverse of whatever correction algorithm was used).