[Talk-us] Kansas extract

[David ``Smith'']

On Sat, Jul 3, 2010 at 5:06 AM, Toby Murray <toby.murray at gmail.com> wrote:
> I just downloaded the Kansas extract from cloudmade:
> http://downloads.cloudmade.com/north_america/united_states/kansas
>
> After rendering I noticed that the northern border was missing.
> Looking at it some more, the extract seems to be cut off about a half
> mile too far south. The cutoff for the other 3 borders looks to be
> about a half mile or so outside of the border which is fine, it is
> just the northern border that is off. I suppose the other option is
> that the extract is correct and the state boundary is off in OSM...
> Can anyone confirm either of these theories?

Neither is correct.  At least, not with absolute accuracy.  But the
state boundaries in OSM come from TIGER, and usually are pretty darn
close — certainly better than a half-mile error.  CloudMade has been
doing these extracts since before OSM imported state boundaries from
TIGER.  (Before that, state boundaries could be inferred by county
boundaries, which had been imported from USGS.  That particular
boundary source can at times be a half mile off...)  I actually don't
know where CloudMade gets its state boundary polygons for the purposes
of data extracts.  It's probably not a very detailed polygon, anyway.

Here's a suggestion for CloudMade: When creating these boundary
extracts, use a generous state boundary polygon that gets an extra
margin all around the state by a few miles.  Then, look for the actual
state boundary multipolygon in that data.  If it's a closed figure
with no errors, then use that to further trim the dataset.  (If not,
optionally, use the last error-free version of the state boundary
multipolygon from previous extracts.)  Since these are only done once
a week, this shouldn't be too huge a processing load.

-- 
David "Smith"
a.k.a. Vid the Kid
a.k.a. Bír'd'in

Does this font make me look fat?