Search

In previous studies, long-wave near-infrared (LW-NIR, 960 to 1,690nm wavelength range) spectroscopy accurately characterized the main components of raw milk (fat, protein and lactose). These components contain information on the udder and metabolic health of dairy cows, as milk production has a critical role in their metabolism. In current practices, milk composition is monitored post-hoc with a low frequency, or on-farm with a poor prediction performance. In this work, we present and evaluate an accurate analyzer for the on-farm, real-time monitoring of milk composition. For every milking performed by an automatic milking system (AMS), the analyzer extracts a milk sample automatically. After stabilization, the sample is introduced into a flow-through cuvette, and reflectance and transmittance LW-NIR spectra of the sample are acquired. Reflectance and transmittance reference and dark reference spectra are also measured after the sample. During a continuous trial of 34 weeks, the analyzer measured 1,926 reflectance and transmittance spectra from raw milk samples. For these, laboratory reference values were obtained for fat, protein and lactose. Prediction models for fat, protein and lactose were trained exclusively with samples acquired during the first six weeks of the trial (n=600). The prediction models were evaluated with subsequent samples (n=1,326). These models had an error (root-mean-square error of prediction, RMSEP) lower than 0.16% (% in weight/weight) for fat (range 2.01-7.95%), 0.18% for protein (2.55-4.48%) and 0.12% for lactose (4.03-5.18%). The presented analyzer can be used for accurate autonomous milk composition monitoring, with a prediction performance within ICAR requirements for at-line milk analyzers (RMSEP < / 0.2%). However, drift was observed in the predictions over time. Therefore, further research and development of calibration maintenance techniques is required to correct model drift and further increase the accuracy.

Plaatsspecifiek bemesten volgens de waargenomen bodemtoestand en gewasbehoefte laat toe om meststoffen efficiënter in te zetten en zo duurzamer te telen. Het samenbrengen van bodemdata en gewasmetingen van diverse proef- en praktijkpercelen gaf in 2020 extra inzichten in de lokale stikstofbehoefte van het gewas. Via simulaties met een gekalibreerd gewasgroeimodel willen we in 2021 taakkaarten opstellen die toelaten om tijdens het groeiseizoen de bemesting op preipercelen plaatsspecifiek bij te sturen.