Climate change is a major issue for the wine industry. Climate and in particular temperature plays a key role in vine physiology and phenology. Temperatures can be highly variable inside a winegrowing region and they are strongly related to local environment (topography, water bodies, vegetation, urban areas...). General Circulation Models (GCM) and dynamical regional models can not take into account this local variability due to their low resolution. For fine scale modeling, a classic option is to create model based on Multiple Linear Regression (MLR) using temperature as dependant variable and local parameters as predictor variables. Though efficient, the non-linearity assumption is a strong constraint that limits performances of spatial models at the vineyard scale. In this study, we compared two fully automated methods which estimate daily temperature and temperature sums at a very fine scale, based on linear (MLR) and non-linear (Support Vector Regression: SVR) assumptions. Data were registered using a network of temperature data loggers installed in 2011 in renown sub-appellations of the Bordeaux area, including Saint-Emilion and Pomerol. Three growing seasons were studied 2012, 2013 and 2014. Model validation showed that SVR presented better results in each case thanks to the non linear component, for an equivalent computing time. Our study has highlighted that a high density network produces maps with a wider range of temperatures compared to medium to low density networks commonly used at a regional scale. In this article, a replicable and highly accurate model was created to produce fine scale temperature maps. Assessment of precise temperature variability at fine scale is essential to allow wine industry to adapt to climate change.