NIRS Calibration Strategies for the Botanical Composition of Grass-Clover Mixtures

Abstract

In literature, different calibrations to predict the species composition of grass legumes mixtures or mixtures of different grass species are described. Mostly, these calibrations were developed using so called “artificial samples”. These artificial samples are obtained by mixing pure (ground) material of the species for which the calibration is developed in known proportions. The plant material used for these artificial samples may have been grown in mixtures or in pure stands. Calibrations based on artificial samples mostly have very good calibration statistics but fail to predict real validation samples. “Real samples” are obtained by hand separation of species mixtures into the different species followed by recomposition. The advantage of the use of artificial samples relative to real samples is that a lot of calibration samples with a different composition can be obtained with a relative small labour input. We built calibrations to predict the white clover content in grass clover mixtures, based on real and artificial samples with the same composition, and validated them with the same independent samples. Calibrations based on real samples performed far better than calibrations based on artificial samples. The failure of the latter can be explained by the lack of environmental variation in their spectra. We recommend a calibration strategy based on fewer but more diverse hand sorted samples, rather than making a lot of artificial samples that contain relatively little spectral information.