Het voorspellen van de melkproductie om de transitieperiode bij hoogproductief melkvee te monitoren

Analysing the global evolution of the dairy industry over recent decades reveals a consistent increase in milk production, accompanied by a growing awareness of sustainability challenges in this sector. While pioneering regions have made significant contributions in reducing green gas emissions, these efforts are often offset by the expansion of the dairy industry in other areas. This confirms the crucial leading role of Europe and the critical importance of continuing to organise research and innovation to further drive existing efforts towards sustainability while facilitating the introduction of sustainable milk production practices in developing industries. Technological advancements, such as automation and digitalisation, present promising solutions for enhancing efficiency and minimizing environmental impact in dairy production. Critically, animal health and welfare are central to the sustainability discussion, as diseases can lead to significant resource losses. Digitalisation and novel data analysis methods have provided useful insights to understanding and monitoring animal health, confirming the potential for developing data-driven tools to improve health management and foster more sustainable dairy practices. Central to this thesis is the observation that the transition period remains one of the most crucial periods in the productive life cycle of dairy cows. This period, typically defined as the six weeks surrounding calving, involves processes where animals successfully transition from one lactation to the other or struggle to adapt to the physiological changes required to support milk production in early lactation. Failure to cope with these changes can lead to diseases with enduring impacts throughout the lactation, including reduced milk production, impaired reproduction and involuntary culling. Despite the development of different monitoring strategies, effective control of transition-associated diseases on dairy farms has yet to be fully realised. Since the early 20th century, the dairy industry has undergone significant transformation through data-driven improvements, with dairy improvement programs (DHI) playing a pivotal role in this. This thesis builds on these established foundations and leverages modern technological advancements to explore the potential of a data-driven methodology for monitoring the transition period. The research was conceptualised within a framework based on precision livestock farming (PLF) principles, which guided the different steps needed to be taken. The first step involved developing a model capable of predicting healthy milk yields in the next lactation based on data from the previous lactation. This led to the creation of the nextMILKTD model, a random forest regression model developed in 3 variants. These variants based their predictions on an increasing number of parameters extracted from DHI data, herd data and reproduction data. Each variant could predict milk yields on the first test day of the next lactation with an RMSE of approximately 6 kg. Although there was no significant difference in performance among the variants, they all performed significantly better than benchmark models. The nextMILKTD was then complemented by the nextMILKMM model which, unlike the DHI-based model, utilises data from milk meters to predict a sequence of milk yields for the next lactation. For both models, a numeric value was derived from the milk residuals in the transition period (MRT). These values represent the differences between the predicted milk yield and the actual milk yield on the first test day for the MRTTD or averaged over the 21 DIM for MRTMM. The subsequent step in this thesis involved investigating the MRT as a potential monitoring tool in the transition period. This monitoring capability of the MRT is based on the premise that animals failing to transition would be identified by negative MRT. The presence of these negative values would indicate the presence of a perturbations that prevent milk production from reaching its the expected level. The initial steps towards investigating these assumptions involved examining the relationship between MRT and an animal's health status. In an observational experimental setup, clinically diseased animals had significantly lower MRT. Additionally, in animals without clinical diseases, MRT was negatively associated with indicators of unfavourable metabolic status such as dry matter intake (DMI) post-partum, insulin-like growth factor 1, and nonesterified fatty acids. Despite being based on different data sources of data, MRTMM and MRTTD showed similar associations. In the next steps, MRTTD was associated with more indirect measures of transition success, such as behaviour and herd management practices. In terms of behaviour, lower MRT was linked to lower rumination and eating times post-partum, further associating the MRT with the DMI post-partum. Similarly, lower MRT was also associated with higher standing times, which have been linked to the occurrence of diseases (metritis, rumen acidosis, mastitis, lameness) in various studies. Finally, in the association study of MRT and herd management practices, MRT was strongly associated with herd milk yield performances. While management practices showed weaker associations, the use of grassland products during the dry period had a negative effect on MRT. The insights derived from various direct and indirect associations underscore the potential of MRT as a transition monitoring metric. To realise this potential, a broad valorisation plan was developed, addressing the challenges and opportunities within the increasingly digitalised dairy industry. This potential monitoring metric was rebranded from MRT to lactation onset value (LOV), and from there, an outline of the key resources, partners, activities, and customer segments was presented. The valorisation plan particularly explored the envisioned methods for presenting the LOV to future customers. Moreover, it highlighted the challenges in establishing industry partnerships and the need for effective dissemination strategies. This valorisation plan concludes by emphasising the importance of demonstrating the LOV's value to farmers and other stakeholders in the dairy industry. In conclusion, this thesis provided a comprehensive examination of the development of a data-based key performance indicator. Diverging from industry standards, the MRT was developed through an open initiative aimed at creating a tangible metric to facilitate sustainable practices during the transition period and to advance the views on digital assets in a dairy sector. The importance of the FAIRness principle in this process is highlighted, and potential improvements to current industry methods are explored. Still, this thesis represents only the first step in this direction, and further research is needed to concretely transform the MRT into a robust evidence-based tool that can serve as a cow-side decision-making assistant. Nonetheless, the work presented in this thesis provides a solid foundation for future endeavours in this area.