The recent Mapbox OSM metrics page shows the road coverage in Thailand to be 100% compared with the CIA factbook figures. It would be great to get a sense from the mappers how accurate the evaluation is. Even while browsing the map around Bangkok, I could easily spot minor roads in the city centre still untraced and unnamed, so assume its the same case for other cities.
Has anyone tried doing any kind of mapping coverage analysis to get an idea of the state of place and road coverage for various cities in Thailand? Are there any official statistics for roads in each city that we could use to compare the road lengths on OSM?
I don’t know how that 100% figure was derived but it is definitely inaccurate. Even if the thousands of smaller unpaved roads aren’t taken into account there are still a huge number of roads that have not yet been added to OSM. There are many reasons for that but probably the most important is that there are few mappers working in Thailand and most of those are foreigners. That means places that are of interest to or have a large population of resident foreigners get mapped with greater care than places that might be of interest to only Thais. The area where I live, Chiang Mai, is mapped pretty well but on a visit to the provinces northwest of Bangkok earlier this fall I stumbled across many main highways that were not yet on the OSM map.
Another reason is lack of good aerial imagery in certain regions. Many areas I visit have only low resolution Bing or Mapbox imagery available on OSM editing platforms. As one who does a lot of armchair mapping in addition to actual fact finding explorations, this severely limits the number of highways that can be added informally to be ground-truthed later. You cannot ground-truth a highway you don’t know about and can’t see in the satellite imagery. Imagery for much of northeast Thailand was added recently. Before that the OSM map was practically blank in that area. It’s getting slowly filled in now.
If you look at the relevant blog entry you’d find the number’s actually 147%. (164,660 miles on OSM vs 111,880 miles from the World Factbook). It’s rather obvious Mapbox’s tally is counting a lot of roads the World Factbook doesn’t. For example, highway=residential and highway=service are included; most of these are unlikely to be included in the World Factbook data.
I have pretty much no idea of the actual status. Obviously, the more major roads are more completely mapped, and 80% might be a fair estimate for the national highways. If one looks at the local level, however, swaths of the country remain essentially blank. And since most road length is in the finer zoom levels, the mapped roads probably amount to only a small fraction of the country’s total road length, when counted down to residential level.
Regarding the second question, stephankn’s tool at http://compare.osm-tools.org/ compares major features in OSM against Google Maps tiles. It shows that most of the major highways are fairly complete, though quite a few regional roads remain unmapped.
I also believe the Mapbox site is more a promotion page for their services than a useful tool. If I remember correctly counting roads was actually an example on how to use their vector data with a MapReduce algorithm to evaluate stuff.
As I had first seen the statistic it already looked wrong so I did not spent any more time into thinking why it was wrong. There are so many sources of potential errors involved here.
First I don’t trust any American three letter agency data. What is the source of their factbook? We don’t know what was counted there. And then we don’t know how old that data is. Roads are constantly built and lower category roads upgraded.
Counting major highway would probably mean not to count “unclassified” and lower.
Another factor is on how to handle dual carriage ways or frontage roads. Do they get counted? I would not, but this required extra processing which Mapbox probably didn’t do.
Thanks Paul for mentioning the compare site. With this you could at least visually compare to the data used by Google. Their data seems to be from more official sources. I think with regards to major highways it is complete. Sometimes googles highway classification differs from OSM. In case OSM has a ref number I would treat it as more reliable.
I’m not too happy with the compare site. Google does some simplification which can hide roads on lower zoom levels. So you have to actually zoom in a bit to spot missing roads. Zoom 14 should be sufficient.
I had thought about merging the bitmaps in the browser and counting colored pixels and by this estimating the missing road network length. But it would just have been a so rough estimate that I dropped the idea as not worth investing into it.
If all mappers doing armchair mapping in Thailand invested a few hours each week then we would soon have no road missing google would have classified as major.
I would like to add that currently the compare website does not pick up some of the more recent additions. Somehow the data update got messed up and I would need to re-import the database.
I have a slightly different implementation (nearly) ready. I might deploy during the Christmas holiday.
On the OSM site you can switch between the compare view, osm and bing arial to easily get an Idea whether the way is actually missing
Yes, I did build it a few years ago. The site is quite dated. I have already reworked the backend processing as it currently puts a quite high load on the server I would prefer to use for other things.
Front end still needs some work before ready for general public. One thing I definitely want to get rid of is the flickering when quickly panning the map.
Unfortunately there is no ideal mapping framework I could build on. All have the one or the other drawback.
It is now using updated libraries. I got rid of code bloat to handle outdated browsers. Also the backend was changed to free resources on the server for other purposes.
Red marked ways are considered major highways by Google and are missing in OSM. Blue marked are missing water features.
Zoom in to enable editing.
I suggest to aim at a distinct geometry to later spot the missing way in the editor. Use the crosshair to better aim. I always load the center point in the editor. So you can press “+” in the editor to zoom in to the same spot.