Application of near infrared reflectance spectroscopy for mineral analysis in hay

The use of near infrared reflectance spectroscopy (NIRS) was explored to predict macro and micro mineral concentrations in hay samples. Hay samples (n = 37) from different locations in Ireland representing a wide distribution of grass species, season, time of harvest and soil types were used to examine the impact of sample preparation and presentation procedures (presence of residual moisture (RM) and particle size (PS) variation i.e. 0.5 and 1 mm) on resultant NIRS calibration and prediction statistics. The samples were scanned in reflectance mode using NIRS DS2500 (1,100 - 2,500 nm) and analysed for Ca, P, Mg, S, Na, Mn, Fe, Cu and Zn using inductively coupled plasma mass spectrometer (ICP - MS) for computation of reference data. Calibration models (n = 24) were developed using modified partial least squares regression (MPLS) based on cross-validation and tested using a validation set (n = 13). To optimise calibration accuracy, mathematical treatments (first and second derivatives) of the spectra and scatter corrections (SNV and Detrend) were applied. Better calibration statistics were obtained at 0.5 mm particle size without re-drying samples for Mg, S, Na, P and Fe. However, calibrations for Mn and Cu performed better at 0.5 mm with re-drying. The predictability of each mineral seems to be more affected by particle size than residual moisture. Overall, these results showed that NIRS prediction accuracy for different minerals vary with the sample preparation and presentation procedure.