In the context of Climate Change, the potential impacts for viticultural terroirs pose a number of questions, especially the likely risks (variability in wine characteristics and quality) and implied challenges. This thesis aims to develop a methodology for measuring and modeling the spatial variability in temperature in viticultural terroirs of the Loire Valley, in order to define the current climate and to provide some answers about the future consequences of Climate Change. The temperature trends in the Loire Valley have therefore been analysed since the middle of the 20th century, using the regional weather stations network. The results indicate an increase of temperatures and bioclimatic indices for all the regional stations. Experimentations at local scales have been realized in the viticultural terroirs of the middle-Loire Valley thanks to an important network of weather stations and data loggers. Bioclimatic indices, as the growing-degree days, underline the spatial variability in temperature and show that this variability can sometimes be as significant at fine scale as at larger scales. A multicriteria modeling has been applied on an experimental site in the Coteaux du Layon vineyards and highlights which local factors prevail in the temperature variability. Climatic modeling has been carried out at fine scales in the last stage of this study with the Regional Atmospheric Modeling System (RAMS). Downscaled simulations delivered at high resolution (5 km resolution) for the control period (1991-2000) were compared with ARPEGE-Climat simulations (50 km) for a few important months during vine growth. Results showed that RAMS contributed to reducing the large-scale induced bias and gave better simulations of extremes temperatures. The RAMS assimilation of the SRES A2 scenario for the 2041-2050 period projected frost risk to decrease in April but an increase in the frequency of hot days (>30°C) and very hot days (>35°C) during veraison and berry ripening. Finally, extreme weather simulations revealed all the complexity of this kind of modeling, with contrasted results according to the different simulated days
