Land cover (LC) (and land use (LU)) maps are an essential components for regional decision-making.They inform policy-makers about the structure of their territory and frame policies including spatial and urban planning, environmental management, transport optimization, risk assessment, etc. The LC data currently available in the Walloon region in Belgium date back over a decade and an update was thus needed. The regional administration decided to launch a research project to develop a robust, automatized, scalable and reproducible method for creating these data, principally based on the available VISNIR orthoimagery at 0.25 m resolution, as well as height information derived through photogrammetry.The ultimate aim of the project is not only to provide recent (2018) maps, but also to elaborate a method that would make it easier for the region to reproduce such data at higher temporal frequency than in the past. The size of the data set (several TB) also provoked a specific focus on scalability while ease of application for a regional administration was another priority.Whereas in urban areas an object-based (OBIA) approach has been the privileged path in the last years as it allows taking into account shape information relevant for the characterization of man-made constructions, such an approach has its limits in the rural and more natural areas the structure of which does not fit as well into the OBIA paradigm, thus calling for a pixel-based approach. In addition, many of the more natural land cover classes have temporal profiles which cannot be detected in a one-date orthoimage. We therefore decided to also analyze Sentinel 1 and 2 data in order to profit from their higher spectral and temporal resolution.All methods were trained using existing regional databases. In a second step, we combined the different LC classification results by fusioning them into one high-accuracy (over 90% OA) product, using a series of different approaches ranging from rule-based to machine learning, passing by more statistical techniques such as Bayesian fusion.

The research teams involved have a long tradition of working with FOSS in image analysis and the choice for a purely FOSS approach was quite obvious and clearly encouraged by the regional administration. Complementary experiences working, for one, with Orfeo Toolbox, and for the other with GRASS GIS, allowed the combination of these different software in the overall framework. This paper will present the details of the respective LC classification chains, including some improvements to the software that happened during the process. Individual LC results as well as results of the different approaches tested for fusioning the different LC products will allow to highlight the advances made, but also some difficulties encountered during the work. In a final section we will present future steps of the work, such as the passage from LC to LU based on alphanumeric databases and the use of LC landscape metrics.