Using milk mid-infrared spectroscopy to estimate cow-level nitrogen efficiency metrics

Area of Science:

• Dairy science
• Environmental science
• Animal nutrition

Background:

• Dairy farming contributes to nitrogen pollution, impacting water quality and greenhouse gas emissions.
• Efficient nitrogen utilization by dairy cows is crucial for environmental sustainability.
• Breeding programs require accurate individual cow nitrogen efficiency data.

Purpose of the Study:

• To assess the potential of milk infrared spectral data to predict individual cow nitrogen efficiency metrics.
• To compare prediction accuracies using partial least squares regression and neural networks.
• To evaluate the impact of data variability on prediction performance.

Main Methods:

• Utilized 3,497 test-day records including milk yield, milk spectra, parity, and days in milk (DIM).
• Investigated nitrogen intake, nitrogen use efficiency, and nitrogen balance.
• Employed partial least squares regression and neural networks for prediction modeling.
• Performed 4-fold cross-validation and stratified validation by herd and year.

Main Results:

• Neural networks using milk spectra, milk yield, parity, and DIM achieved the best cross-validation predictions (R² of 0.61, 0.74, 0.58 for nitrogen intake, use efficiency, and balance).
• Partial least squares regression outperformed neural networks in herd- and year-stratified validation (R² < 0.29 and < 0.60, respectively).
• Prediction accuracy is highly dependent on the similarity between calibration and validation dataset variability.

Conclusions:

• Milk infrared spectral data, combined with milk yield and cow information, can predict nitrogen-related traits in dairy cows with reasonable accuracy.
• Model performance is sensitive to the representativeness of validation data compared to calibration data.
• Accurate prediction of nitrogen efficiency can support breeding strategies for a more sustainable dairy sector.