For ground truth the River Authority’s photo-interpretation maps based on 2017 aerial images (false colour, 25cm pixel resolution) were used which were remapped into the Legger-classes, see Table 1. Training and test pixels were selected from patches (polygons) of the photo-interpretation maps . We excluded dynamic patches by comparing earlier photo-interpretation maps (2017, 2012, 2006, 1997) and selecting patches that remained stable over these years.