As noted, the classifier currently mis-categorizes certain pairwise combinations of classes for which atmospheric reflectivity is similar. Distinguishing high-density urban areas from low-density urban areas is difficult, as is distinguishing low-density urban from non-urban land. We also note that in several images, such as the 1999 classifier output, water is mis-classified as urban land. To further improve the classifier's performance, we propose to use density of urban features shops, houses, intersections hospitals - as an addition the feature space.To demonstrate the production of these inputs, we produce raster images of urban features OpenStreetMap (OSM) tags. Accessible through OSM Application Programming Interface and its Overpass web interface, we acquire point locations of urban feature categories.