Abstract :
[en] Maintaining satisfying economic outcomes and limiting environmental impacts are key challenges in dairy farming today and this requires good decision-making regarding actions to make on farms. The analysis of milk by Fourier-transform mid-infrared (MIR) spectrometry provides valuable information on milk composition. This technique has already demonstrated qualities to support decision-making, for example, through the well-established predictions of milk fat and protein contents or the latest development of prediction models for novel traits. However, its full potential remains partly uninvestigated. Hence, the objective of this thesis was to contribute to the development of decision-support tools with economic and environmental interests for the dairy sector using milk MIR spectrometry. The research conducted in the framework of this thesis covered three different approaches of using MIR for decision support: (1) the development of a MIR calibration equation to predict a trait of interest, (2) the development of a test-day model to predict milk MIR spectra for management purposes, and (3) the combination of MIR-predicted data with other data streams as a means of providing additional information for decision-making.
First, we explored different strategies to predict the pregnancy status of dairy cows (pregnant vs. open) in Australia using milk MIR spectra and partial least squares discriminant analysis. Correctly identifying the pregnancy status of cows is imperative for a profitable dairy farm. Early pregnancy could not be detected satisfactorily, but promising results were obtained using MIR spectra recorded 151 days or more after insemination (i.e., mid- and late gestation), with the area under the receiver operating characteristic curve of 0.76 on the testing set. A potential application that needs to be explored further is the development of a screening tool to detect mid- to late-term fetal abortion.
Secondly, we studied the ability of a test-day mixed model to predict milk MIR spectra from first parity Holstein cows for management purposes (e.g., for the detection of problems, simulations, predictions of future data). The spectral data used for modeling originated from the Walloon milk recording database. The average correlation between observed and predicted values of each spectral wavenumber was 0.85 for the modeling set and ranged from 0.36 to 0.62 for different scenarios that corresponded to situations with more or less information known about the cows. Correlations between milk fat, protein and lactose contents predicted from the observed spectra and from the modeled spectra ranged from 0.83 to 0.89 for the modeling set and from 0.32 to 0.73 for the scenarios. These results demonstrated a moderate but promising ability to predict milk MIR spectra using a test-day model. Different improvements of the model are possible before potential practical applications that could have economic or environmental implications for dairy farming, depending on the MIR traits subsequently predicted from the modeled spectra.
Thirdly, we investigated the univariate relationships (correlations) between dairy cow enteric methane (CH4) production (g/day) predicted from milk MIR spectra and 42 technico-economic variables from 206 Walloon dairy herds over a period of 8 years. Enteric CH4 is an important part of the carbon footprint of milk production. Significant correlations ranged between |0.06| and |0.38|. Low MIR CH4 production tended to be associated with more extensive or suboptimal management practices, which could lead to lower profitability. The observed weak correlations suggest intricate interactions between MIR CH4 and technico-economic variables due to the use of real farm data with large variability in management practices. This implies the need for further research to unravel these complex relationships for a better understanding of factors associated with CH4 production on dairy farms in order to better target mitigation strategies.
Lastly, we discussed, in the light of the research carried out in this thesis, strengths as well as issues and considerations regarding the development of decision-support tools using milk MIR. In particular, key strengths of MIR are the low cost and rapidity of the technology as well as the standard procedures for milk sample collection and analysis, allowing the acquisition of MIR data on a large scale for the development of various customized tools to assist decision-making on dairy farms. Issues and considerations covered the prediction of indirect MIR traits, the quality and variability of spectral and reference data, the choice and validation of models, the utilization of MIR indicators, the study of MIR traits in the population, the timing of milk sampling, and the uptake of MIR tools by farmers.
In conclusion, this thesis contributed (1) to establish the first steps of the development of new MIR tools and studies to support decision-making in dairy farming with potential economic and environmental benefits; and (2) to gain insight into the benefits and considerations of using milk MIR for the development of decision-support tools.
