[OSM-talk] ODbL-clean Coastlines
penorman at mac.com
Sat Mar 24 09:34:42 GMT 2012
I have been running a nightly coastline generation on my server, using the
latest data from my jxapi server. Tonight I switched it over to filter out
data that WTFE reports as dirty. This is somewhat more aggressive than the
rebuild will be, but the results are worrisome.
A PNG showing the OSM coastline post-transition along with red dots for
where the coastcheck program found errors is at
The red dots represent places where coastcheck found an error
A few things can be seen:
No continent will generate a satisfactory rendering. Australia, the US, most
of Africa and most of Eurasia from 50 degrees south will be completely
There are 3471 points where coastcheck found errors. Normally I would expect
The following areas are particularly bad:
The west coast of the US
The US coast of the Great Lakes
Antarctica (which has significant parts completely missing with no error
points since there's nothing left to even try to close)
When examining the data, I found most of the ODbL-dirty status was mainly
from the following accounts:
Blars (US West coast)
I see a few ways for dealing with the coastlines:
- Remapping. Obviously this is the only option where the coastline was
actually created by the non-acceptor.
- Adopting changesets. Many of the dirty ways and nodes appear to be
imported. If the imported just imported PD data than they have no IP in the
ways and they can be retained.
- odbl=clean. It's generally easy to verify that a way is a coastline from
My coastline files are available in the usual location of
http://pnorman.dev.openstreetmap.org/coastlines/ but are named coastlinec_p
and processedc_p. processedc_p has not yet uploaded but will be complete in
about an hour.
Additional detail of some regions can be found at
If opening processedc_p in QGIS be sure to make a spatial index or it will
A refresher on coastlines for those who haven't tagged them in awhile:
Direction *matters*. Land on the left, water on the right.
Each end of each way should join with exactly one other natural=coastline
I am not sure when each processedc_p file will be completed. Cron starts the
job at 5 AM PST and I'd expect it to take about 5 hours, mainly depending on
WTFE server speed and my upload speed to errol. The coastlinec_p files will
be uploaded first so arrive at the server earlier.
The results of the jxapi way[natural=coastline] is filtered by cleanway to
remove any objects reported as severity=normal, leaving behind 0 or 1 node
ways if necessary. It then is passed to the coastcheck programs and the
A CT-clean way to which a decliner added a tag other than natural=coastline
would be removed by this algorithm, but not by the rebuild.
More information about the talk