Mise en place de l'évaluation génétique sur les contrôles élémentaires en bovins laitiers et perspectives d'utilisation des résultats en appui technique

Genetic evaluation models for milk production traits and cell counting in cattle have evolved considerably over the last 10 years. Most countries (Germany, Canada, Denmark, Italy, New Zealand, The Netherlands ...) have developed or are currently settling (France) evaluations that directly use each test-day production instead of using cumulated production over 305 days. The main interest of test-day models is in the fact that they have an increased consideration for systematic environmental effects of each test-day. The estimated genetic values are therefore more accurate. France has chosen an original model that is easy to interpret and flexible. The lactation curves are made up of a sum of cow's (genetic and non-genetic traits) and environment's effects (herd, breed, region, lactation number, age and calving month, length of dry period, pregnancy...). These effects are determined for the days in milk from 7 to 335 days and are corrected for all other factors of variation making them directly interpretable. However, the models are costly in terms of computing needs due to their sophistication and the large number of data to be analysed simultaneously. The first objective of this thesis was to validate the software for genetic evaluation developed for this application and to study the possibilities of simplifying the model in order to limit computing needs. The exactness of the calculations performed were validated using a new approach based on the rebuilding of simulated data fulfilling specific conditions so that the effects estimated by the software are exactly those simulated if the evaluation software is correct. Considering the computing limits, the total number of random effects of the model was reduced using a reduced rank approach on the (co)variances matrix of these random effects without leading to important variation of the estimated genetic values. We also show that only the level (and not the form) of the fixed effects modelled by days-in-milk curves depend on test-day year. This observation allows a pre-correction of the data for the effects associated with lactation curves depending on age, calving month, length of dry period and pregnancy. This modification has allowed to increase the stability of the effects estimated and to decrease in an important way the computing time, by making two-step genetic evaluation possible: the pre-correction step is followed by an analysis of the pre-corrected data in a model including all the random effects and the fixed effects that depend on the year. Besides the gain in accuracy of the estimated genetic values - which are probably modest -, the main benefits of this sophisticated and highly computationally demanding models are related to prediction and herd management support which was the second objective of this thesis. The studies performed show the different perspectives of this view point. Indeed, due to the original modelling of the fixed effects as sums of specific curves, the genetic evaluation test-day model favours the development of new tools for herd management where production predictions will be made using estimated effects from genetic evaluation. These new tools could also include a dynamic analysis of the herd management via theLes modèles d'évaluation génétique des caractères de production laitière et de comptages cellulaires chez les bovins ont considérablement évolué depuis 10 ans. La plupart des pays ont mis en place (Allemagne, Canada, Danemark, Italie, Nouvelle-Zélande, Pays-Ba