Application of FT- NIRS to estimate chemical components of freeze-dry herbages of Tuscany natural pasture

Near infrared spectroscopy has been successfully applied at analysis of animals feed, but not many authors have tested NIRs technologies on natural pasture and almost all the researches consider NIRS scansion of oven-dried samples. Nevertheless some parameters, such as vitamins or polyphenols, can be modified by the high temperature applied during the analysis. The object of this study is to investigate the FT-NIRs capacity to estimate chemical composition of natural pasture herbage scanned under freeze-dry condition. Eighty herbage samples, collected in natural and naturalized pastures of Tuscany (Italy), were used. Each sample was freeze dried, grounded and analysed for: Dry Matter, Crude Protein, Ash, Ether Extract, Crude Fibre and fibrous fractions (Neutral Detergent Fibre, Acid Detergent Fibre, Lignin). For each sample, three aliquots were scanned using FT-NIRs Antaris II (Thermo Scientific). Mathematical pre-treatments (Multiplicative Scatter Correction, Standard Normal Variate, 1st and 2nd derivate) were applied and outliers’ spectra were identified and removed when necessary. Partial least square regression was used on the average spectrum and then the models were fully cross validated. Results are evaluated in terms of coefficient of regression and root mean square errors in calibration (R2 -RMSE) and in cross validation (r2-25 RMSECV). Calibration results, excluding EE and Ash, obtained R2 26 higher than 0.87 for all parameters, with RMSE lower than 1.3 recorded for DM, CP and Lignin. Error values between 2.6 and 3 were achieved for CF, NDF and ADF, probably due to the broader range of variation in the primary analysis. In calibration, EE obtained a R2 29 of 0.76 that need of further evaluation, while Ash showed R2 values lower (>0.6), probably due to the absence of absorption in the near infrared region for the minerals. The cross-validation models achieved both lower coefficients of determination and higher RMSECV than the previous models; however error differences between calibration and validation steps were always smaller than the limit of 20% proposed by Moya (1993). Hence, the FT- NIRS applied on freeze dried samples seems able to estimate the principal parameters of chemical components on samples of natural pasture herbages, characterized by wide variability. Further researches should aim to evaluate additional parameters and they can improve the accuracy of FT- NIRS on chemical constituents.