03550naa a2200193 a 450000100080000000500110000800800410001910000170006024501610007726000090023852028830024765300210313065300270315165300110317865300240318970000180321370000180323177301070324911312302021-09-20       2021    bl uuuu     u00u1 u    #d1 aRECH, ??. F.  aNEAR-INFRARED SPECTROSCOPY (NIRS) AND MULTIVARIATE CALIBRATION TO DETERMINE THE BROMATOLOGICAL COMPOSITION OF GIANT MISSIONARY GRASS.h[electronic resource]  c2021  aThe NIRS technology is an advantageous alternative to conventional bromatological analysis. It is faster, less costly and less polluting, however good calibration models are needed to interpret the information contained in the sample spectra in the proximal infrared region. In the present work, near-infrared spectroscopy, combined with multivariate calibration methods, was used to develop models for predicting the contents of organic matter (OM), crude protein (PB), neutral detergent fiber (NDF) and acid detergent fiber (FDA) of giant missionary grass (Axonopus catharinensis). Samples from technical reference units and research experiments developed by Epagri were used (between 150 and 200 depending of the component analyzed). Bromatological analyzes were performed at the Animal Nutrition Laboratory-Epagri by reference methods between the period 2018-2021 and the results were used for calibrations. The reflectance spectra of each sample were collected in triplicate by the NIRFlex N-500 Solids spectrophotometer in the range between 4,000 to 10,000 cm-1 number of waves. The ranges of waves used were selected according to the functional groups of each component. The values of SEC (standard error of calibration), SEP (standard error of prediction), bias, number of latent variables (VL), coefficient of determination (R2) of calibration and internal validation were used to verify the models. For each component, more than one model was fitted using the NIRCal 5 BUCHI software, the PLS algorithm, and some mathematical pretreatments. Same points detected as ?leverage? outliers and high Student residuals were eliminated. The models were evaluated by external validation (VE), with samples not included in the adjustment, and the choice of calibrations with better predictive capacity was made after comparing the results of RMSEP (square root of the mean prediction error), RPD (ratio deviation prediction) and bias. The best results of VE were as follows: RMSEP: 0.37; 0.51; 1.31 and 0.59; RPD: 4.5; 4.6; 2.4 and 3.0; bias 0.18; 0.08; 0.52 and 0.64; respectively for the models of MO, PB, FDN and FDA. The mathematical pretreatments used in these models were: 3-point smoothing, normalization by closure and first BCAP derivative; 3-point smoothing and BCAP first derivative; multiplicative signal correction (MSC full) and first BCAP derivative; smoothing 9 points gap2 and first derivative BCAP, respectively for MO, PB, FDN, FDA. The best R2 of the calibrations were: 0.92; 0.95; 0.92; 0.91; respectively for MO, PB, FDN and FDA. The calibration models developed shows satisfactory results, meet the purpose of the study and can be used to predict the composition of samples of giant missionary grass from technical reference units and animal breeders. However, with the introduction of new samples, they will be constantly updated, validated and improved.  aanimal nutrition  aAxonopus catharinensis  aforage  amodeling prediction1 aWERNER, S. S.1 aFAVARO, V. R.  tIn: REUNI??O DA SOCIEDADE BRASILEIRA DE ZOOTECNIA, 56., 2021, Online. Resumos... Bras??lia: SBZ, 2021.