MD Looking to the subsamples statistics

Normally when scanning ungrounded samples like wheat, barley, corn in a NIR instrument, we use a large cup which moves across the sample presentation plate, in some cases continuously and in others stopping and scanning several times. Finally we get the average results, but there are some software’s which give you also the subsample results, giving information of the standard deviation between subsamples which is a good indicator for homogeneity or other issues. It can give you also the Mahalanobis distance for each subsample.

Suppose the case of a large cup where we acquire 32 scans. The cup stop in 8 different positions to scan the sample, so we have 8 subsamples with 4 scans each. Lower numbers of scans mean a noisier spectrum, so the Mahalanobis distance will be normally higher in all the subsamples (independently) than the Mahalanobis distance of the 32 scans average spectrum.

We have build our score matrix with 32 subscans average spectra, so if we compre spectra of 4 subscans average spectra it is normal we get out of range MD values.