Using selective NIR wavelengths in portable devices to evaluate the chemical composition of cattle feeds

In modern cattle nutrition, the estimation of chemical composition must be accurate, rapid, and available throughout the farm feeding process. This goal can be achieved using portable near infra-red instruments applied at several positions in the feeding chain. Although portable instruments are generally more affordable than benchtop equipment, the current costs of purchasing and maintaining these tools may still be prohibitive for small and medium-sized animal husbandry businesses. This makes their widespread adoption economically unsustainable for many farms. Reducing the spectral data helps prevent multicollinearity issues, enabling the use of a minimal–optimal problem approach and supporting the development of a targeted and cost-effective instrument. This was achieved by evaluating various cattle rations and silages (including rations for cows and bulls and grass and corn silage), where we selected the most significant wavelengths for fibre characterisation using the Random Forest (Boruta) algorithm. The number of identified features varied based on the spectral pre-treatments applied or the use of a batch effect reduction algorithm (ComBat function). External validation was accomplished with spectra collected by different instruments, origins and sample types, and the highest adjusted coefficients of determination in validation were 0.95 for dry matter and 0.84 for alpha-amylase and sodium sulphite-treated NDF. Results demonstrated that using a few spectral bands is suitable for instrument calibration.