In a supervised classification problem, training data often comes from field inventories acquired by domain experts. However, location of these inventories may be approximate (due to intrinsic precision of portable GPS used). This spatial inaccuracy is particularly problematic when these data are used to train a classifier on very high resolution (VHR) satellite images. Indeed, in some cases, spatial accuracy of inventories may be much lower than the one of images. In this paper, we propose three preprocessing methods to correct this spatial inaccuracy of field inventories. The principle of these approaches is to exploit available VHR satellite images to spatially correct field data. Our experiments highlight the interest of these pretreatments and compare the proposed approaches on a dataset consisting of 24 habitat inventories of coral reefs and a VHR satellite image (WorldView-2) . 2