Multiple Country Approach to Improve the Test-Day Prediction of Dairy Cows' Dry Matter Intake

Area of Science:

• Animal Science
• Dairy Science
• Agricultural Engineering

Background:

• Accurate prediction of dry matter intake (DMI) in dairy cows is essential for optimizing herd nutrition and management.
• Mid-infrared (MIR) spectroscopy offers a rapid and non-invasive method for analyzing milk composition, potentially enabling DMI prediction.
• Existing DMI prediction models may have limitations in diverse geographical and management conditions.

Purpose of the Study:

• To develop and validate predictive models for dairy cow dry matter intake (DMI).
• To evaluate the performance of partial least square (PLS) regression and artificial neural network (ANN) models using milk MIR spectral data.
• To assess model generalizability through cow-independent and country-independent validation.

Main Methods:

• Utilized a dataset of 10,711 milk samples from 534 dairy cows across Australia, Canada, Denmark, and Ireland.
• Developed PLS regression models and a one-hidden-layer ANN, incorporating milk MIR spectra and derived components.
• Employed 10x10-fold cross-validation (CV) and country-independent validation (CIV) to assess model performance.

Main Results:

• Cow-independent cross-validation yielded root mean square errors (RMSE_CV) of 3.27 ± 0.08 kg for PLS and 3.25 ± 0.13 kg for ANN.
• Country-independent validation (RMSE_CIV) ranged from 3.73 to 6.03 kg for PLS and 3.69 to 5.08 kg for ANN.
• ANN models generally demonstrated slightly better performance than PLS models, particularly in country-independent validation.

Conclusions:

• PLS regression and ANN models utilizing milk MIR spectral data can accurately predict dairy cow DMI.
• The developed models show good generalizability, although performance varies across different countries.
• These models offer a promising tool for real-time DMI monitoring and management in diverse dairy farming systems.