Comparison of near infrared reflectance spectroscopy calibration strategies for the botanical composition of grass-clover mixtures

Abstract

Predicting species composition in mixed swards by near infrared reflectance spectroscopy (NIRS) can save labour in grassland research, provided equations are available. This study compares calibration strategies to predict species composition in swards with tall fescue (Festuca arundinacea), perennial ryegrass (Lolium perenne) and white clover (Trifolium repens). The compared calibration strategies were based on either real or artificial samples. Real samples are samples taken from multispecies swards; the species composition is known by hand separating; after separating, the original samples are recomposed. Artificial samples are samples obtained by mixing single species grown in pure stands in known proportions. The performance of the equation based on real samples was significantly better than the performance of the equation based on artificial samples. We hypothesized that the weak performance of the equation based on artificial samples was due to a lack of environmental variation in the spectra of the artificial samples. The hypothesis was supported by the good performance of a novel calibration strategy based on the spectra of the artificial samples with added variation. According to the obtained results, a calibration strategy based on few but diverse calibration samples is discussed.