Digital Chart of the World and Vmap

22 07 2008

Unfortunately the urban areas classified in the UMD land cover classification dataset I referred to yesterday are not classified using the satellite imagery as I believed before, the data simply come from another data source the Digital Chart of the World (DCW). Today has seen some further research of DCW (renamed Vmap0) and I have found out that some of Vmap0 has already been used for import into OSM, in particular the urban areas that I was trying to pick out and test. I am still trying to find a decent data source for Vmap0, the only ones I can find at the moment require me to use ArcView, I want to use Qgis! If anyone knows of a good way to download DCW/Vmap data, please let me know.

Just because the UMD land cover data for urban areas was derived from another source, that doesn’t mean that the other data in the project isn’t useful. Their process obviously clearly identifies different types of forests, grasslands and bare land and it will be interesting to understand how it does this. The process is in fact rather complex, as can be seen by the published paper. They analysed data over a whole year and calculated Normalized Difference Vegetation Index (NDVI) figures. These compare the spectral reflectance in the red and near-infrared bands and essentially detect areas where there is vegetation, (I wonder if its possible to do something like this for urban!). Anyway the UMD data uses in total 41 metrics using 5 channels on radiation and the NDVI figures and then uses a flowchart to breakdown which land use there is. This could help us classify large rural areas in the World in OSM if we think that would be useful but it is not going to help with my model for urban areas.

The model I envisage will not simply classify areas but will hopefully be able to predict the length of road, or the amount of OSM data we can expect in an area. The processes discussed earlier can accurately distinguish urban areas from everything else, the question is whether there is a correlation between the brightness of a pixel and the amount of ‘stuff’ in that area. If there is, then my models will be a lot more useful than the UMD datasets I discovered yesterday.