Human activities may have, through land use and management changes, an impact on the large amounts of carbon sequestered in soils. Increasingly, inventories of Soil Organic Carbon (SOC) stocks are requested at the regional/national level for countries involved in the Kyoto Protocol and as input in biogeochemical models. However, SOC is characterised by a high spatial variability and a slow temporal dynamic, which reduce our ability to detect and explain SOC stock changes. At the regional scale, local (e.g. land use, topography) and global (e.g. climate) driving factors interact in a complex way and are likely to generate high uncertainties. This thesis represents an effort in assessing and detecting SOC stocks and SOC stock changes at the regional scale. Specifically, our objective was to explore two possible strategies to cope with spatial variability: (i) increase the sampling density using a fast analytical technique (reflectance spectroscopy) and (ii) use information on land use change history to decrease spatial variability in landscape units. In the first part of the thesis, we investigate the capabilities of laboratory, portable and imaging spectroscopy to determine SOC concentrations in croplands. It is shown that laboratory and portable spectroscopy can reach accuracy levels comparable to a standard analytical method (Walkley & Black). Imaging spectroscopy, even with somewhat lower performance, can enhance the estimation of mean SOC stocks by providing a very large sampling density. Outside of the controlled environment of a laboratory, portable and imaging spectroscopy fail to demonstrate reasonable calibration stability over a range of soil types and surface conditions. In the second part of the thesis, we found after re-sampling a soil database collected in 1950-1960 that land use change history has a significant impact on SOC dynamic in forest. Building on this, a bookkeeping model based on historical land use maps (1868-2005) and SOC response curves was developed to assess the temporal ...