Abstract
Prediction models of different types of forage were developed using a dataset of near-infrared reflectance spectra collected by three handheld NeoSpectra-Scanners and laboratory reference values for neutral detergent fiber (NDF), in vitro digestibility (IVTD), neutral detergent fiber digestibility (NDFD), acid detergent fiber (ADF), acid detergent lignin (ADL), crude protein (CP), Ash, and moisture content (MO) from a total of 555 undried ensiled corn, grass, and alfalfa samples. Data analyses and results of models developed in this study indicated that the scanning method significantly impacted the accuracy of the prediction of forage constituents, and using the NEO instrument with the sliding method improved calibration model performance (p < 0.05) for nearly all constituents. In general, poorer-performing models were more impacted by instrument-to-instrument variability. The exception, however, was moisture content (p = 0.02), where the validation set with an independent instrument resulted in an RMSEP of 2.39 compared to 1.44 where the same instruments were used for both calibration and validation. Validation model performance for NDF, IVTD, NDFD, ADL, ADF, Ash, CP, and moisture content were 4.18, 3.86, 6.14, 1.10, 2.75, 1.42, 2.71, and 1.67 for alfalfa-grass silage samples and 3.22, 2.21, 4.55, 0.38, 2.07, 0.50, 0.51, and 1.62 for corn silage, respectively. Based on the results of this study, the handheld spectrometer would be useful for predicting moisture content in undried and unground alfalfa-grass (R(2) = 0.97) and corn (R(2) = 0.93) forage samples.