Abstract
Multi-species leys provide high-quality forage for livestock while reducing the nutrient requirements to achieve the targeted production aims. Accurate and timely estimation of forage quality is essential for optimising animal nutrition and agricultural practices. In a field experiment in South Tyrol (NE Italy), being part of the LegacyNet voluntary network, hyperspectral data were collected by proximal sensing coupled to quality analyses performed by NIRS analysis on forage samples, to assess the accuracy of hyperspectral-based quality predictions. The field trial was laid out as a simplex design including mixtures of increasing diversity of six species from three functional groups, and an equiproportional seed mixture of twelve species. Measurements were taken at the time of the second and third harvest of the second production year. Predictive models based on Partial Least Squares Regression linked spectral reflectance to forage quality parameters, including NDF, ADF, metabolisable energy and digestibility. Pre-processing techniques involving Savitzky-Golay filtering followed by first or second derivative enhanced accuracy. Digestibility had the highest accuracy, that of ADF and energy concentration was moderate, and that of NDF was unsatisfactory. Utilising Sentinel-2-aligned bands instead of the whole spectrum resulted in roughly comparable accuracy, except for metabolisable energy, showing a clear drop.