Kernel Density Estimation (KDE) is used produce raster images representing the mean density of the features. The KDE image, as indicated in Figure 5, can in future in incorporated into the model. In instances where our classifier mis-categorizes (for example) water as land, the KDE of OSM features is expected to improve accuracy, since the Random Forest decision trees can make additional cuts based upon the density of tags (captured in the raster pixel values from 0 - 255) or the lack of any tags on expanses of water.