Land cover plays an important role in the Earth's climate as it affects multiple biochemical cycles and is critical for food security and biodiversity. As land cover is continuously evolving, influenced by anthropogenic and other factors, the availability of temporally varying land cover data sets of large spatial domains is integral to understanding, monitoring, and informing environmental management efforts. Here we use classification trees to generate annual land cover maps of the European continent for 2001 to 2019 on a ∼250 m resolution. The classification trees are trained using gap-filled and smoothed MODIS normalised difference vegetation index (NDVI) satellite data, as well as CORINE reference land cover data. We apply the bagging ensemble technique on oversampled NDVI data, with an additional majority vote for overlapping segments over the continent-wide domain. We distinguish between 39 land cover classes, with a total classification accuracy of 75% and average precision of 76%. The accuracy varies between the classes, with common classes (e.g. agricultural and forest classes) performing better than rarer ones (e.g. artificial land cover). Over the entire continent, we find that artificial land cover, wetlands, and forests have increased on average by 0.76, 0.50 and 0.22%/year respectively, while the agricultural area has decreased by 0.21%/year. We also quantify these changes in land cover on a national and metropolitan level. Given the near-real-time availability of global NDVI data, we note the potential of the presented approach for generating ‘near-real-year’ annual land cover data sets of large geographic domains, for the continuous monitoring of land cover change and the effects of interventions.