The transfer of NIR calibrations for undried grass silage from the laboratory to on-site instruments: Comparison of two approaches

As a first step in a project whose aim is to implement near infrared (NIR) analysis of animal feed on the farm, the present work has examined the possibility of transferring undried grass silage calibrations for dry matter, crude protein, and neutral detergent fiber from a dispersive laboratory NIR instrument (Foss NIRSystem 6500) to a diode array on-site NIR instrument (Zeiss Corona 45 visNIR 1.7). Because the samples are complex and heterogeneous and have high humidity levels it is not easy to establish good calibrations, and it is even more of a challenge to transfer them. By cutting the spectral range to 1100–1650 nm and treating with first or second derivative followed by standard normal variate (SNV) scatter correction, it was possible to obtain very similar spectra from the two instruments. To make the transfer, two approaches were tried. Simply correcting the Corona spectra by subtracting the mean difference spectrum from a transfer set met with only limited success. Making a calibration on the Foss using a calibration set of 503 samples with spectra orthogonalized to the all the difference spectra in the transfer set of 10 samples resulted in a successful transfer for all three calibrations, as judged by performance on two prediction sets of size 22 and 29. Measuring 5 replicate subsamples with the Corona allows it to see a similar surface area to that of 3 replicates in the Foss transport cell, and it is suggested that this is an appropriate level of replication for the Corona.