[HOT] Latest re: Increasing Somalia Coverage

Schuyler Erle schuyler at nocat.net
Fri Aug 12 01:37:51 BST 2011

Hey gang,

TL;DR: I am doing preliminary work on importing GNS and MapMappers roads into OSM. I will eventually need help. Someone who is intimately familiar with Somalian geography would be helpful right away.

Last night I did some initial research towards importing data into OSM to improve our coverage in Somalia. I looked at three datasources:

* UNHCR refugee locations
* NGA GNS points of interest
* MapMappers transport and hydrology layers

(1) The UNHCR refugee location data is sadly unexciting. It's a KML file that consists mostly of UNHCR assets in Kenya and Ethiopia, with fewer than a couple dozen refugee locations, half of which are from 2007 or even earlier. None of the data is actually in Somalia. I don't see a reason not to import it, but it doesn't feel very high priority.

(2) The NGA GNS data consists of 6,700 populated places, 4,200 assorted localities, 4,600 terrain features, 8,700 hydrographic features (including 3,300 wadis), and 450 other points of interest. Somalia is one of the few places where I would seriously consider attempting this kind of import, because of the sparsity of the data. I expect to import only a fraction of the feature classes.

Unfortunately, GNS imports have fallen out of favor in OSM, and all the code I can find to work with them is old and does very little for actually transforming the feature types into meaningful OSM tags. I haven't been able to find a good crosswalk between OSM tags and GNS feature codes, so I tried to do some statistical analysis of the distribution of OSM tags and previously imported GNS features in OSM but didn't come up with much. There are only 112 feature codes used by GNS in Somalia, so I'll probably just break down and do the crosswalk by hand.


My plan for GNS is to do automated imports only of PoIs that are further than 100m from anything currently in the database. The remainder I plan to split up into separate files (per Kate's recommendation) and import manually using JOSM.

(3) The MadMappers road data appears to mostly have been traced from imagery from 2008 and a bit before. Here are the sources:

  SOURCE (String) = 200k topo maps (note: I think these are Russian - SDE)
  SOURCE2 (String) = Formosat January 2008
  SOURCE2 (String) = Landsat ETM+
  SOURCE2 (String) = Tracklog

The feature types are interesting:

  fin_type (String) = Geological Trace/Track
  fin_type (String) = Major road
  fin_type (String) = Road
  fin_type (String) = Streets
  fin_type (String) = Track/Trail

This Geological Trace/Track layer is curious, because it appears to contain a ton of straight  lines that run at right angles through the middle of the desert:

  http://iconocla.st/hot/so/madmappers1.png (QGIS screenshot; the green lines are existing OSM roads)

HELP: I'd like someone who actually knows something about the Somalian outback to reality-check us on this. I can see some of the depicted lines in the Bing imagery -- what the heck are they? They appear man-made. Are they actual roads (or trails)? Should we include them? Or should I just leave them out, to be on the safe side?

We also get paving type:

  SURFACE (String) = Loose
  SURFACE (String) = Paved

This is good, because there seem to be whole provinces in Somalia without a major paved road.

We get some naming data, but not a ton.

Again, per Kate's suggestion, I'd like to generate the .osm files and then parcel them out to volunteers for manual conflation, to avoid stomping on the work that's already there. If you look at the respective datasets for Mogadishu, you can see that both are woefully incomplete for a city of 2 million people, but the overlap isn't complete, so there's useful work to be done manually:


Fortunately, the Bing imagery is *really* good there. So there's that.

I strongly welcome any constructive advice that anyone has to offer. Otherwise, I'll let you know when I've got the data carved up and ready for people to look at. I'm hoping it'll be sometime over the coming weekend.


More information about the HOT mailing list