Mediterranean cultivated landscapes are prone to floods, erosion and water pollution. These landscapes are also considered as hot spots for biodiversity. As hydraulic and linear artificial settlements, terrace fronts or terrace walls consisting in historic manmade levees on agricultural plot margins, can alter hydrological fluxes, limit erosion, favour soil water storage and biodiversity (habitat and corridor effects) [1]. More and more spatially explicit models and indicators need the map of these landscape linear features to diagnose the hydrological risk or biodiversity at catchment or region scale [2]. However, terraces walls are almost never available in any map agency databases. Therefore, the potential of very high spatial resolution remote sensing data, in 2D or 3D scenes, to map these linear elements, need to be addressed. Some recent studies showed that terraces from 2D VHR scenes, i.e images, can be poorly detected, due to both vegetation cover altering objects visibility having poor spectral signature or due to anisotropic shadowing effects. The use of 3D scenes, i.e. digital terrain models (DTM) and digital surface models (DSM), appears therefore as more reliable. A lot of literature exist in landscapes linear elements detection from remote sensing 3D scenes, e.g. for hedges corresponding to surface convexities or hydrographic networks corresponding in terrain concavities [3, 4]. However, very few remote sensing or geomorphometry literature exists for terrace walls linear detection, corresponding to terrain stairs or discontinuities. This paper aims to propose an innovative DTM processing methodology chaining existing algorithms to map terrace walls from optical VHR satellite data.