[talk-au] Manipulating ABS suburb data, and related data..

[Darrin Smith]

On Fri, 17 Apr 2009 14:10:32 +1000
Franc Carter <franc.carter at gmail.com> wrote:

> I totally agee that it's a good idea to work this out, I've been
> silent on the matter because I'm far from clear
> as to what is a good approach.
> 
> I'm currently wrestling with trying to get a handle on how we can tell
> whether the ABS data is more geographically
> accurate than yahoo or other data (not necessairly whether it is an
> accuate reflection of the boundaries)
> 
> So, yes - thoughts please

Here's how my thinking/practice has panned out so far:

- If an ABS way obviously runs along an existing OSM way in some manner
  then it makes sense to merge the two, suburb boundaries nearly always
  run down the middle of a features and the rare cases they don't need
  to be correctly researched and documented to show they deviate from
  the norm, until they are it's saner to merge them.

- If the original OSM way is landsat based then the ABS version is
  nearly always of a higher quality better accuracy, so I've been
  splitting the ways in relevant places to obtain parallel chunks,
  aligning points at one end, merging the 2 ways (which combines tags
  and relation memberships) then splitting them again and removing the
  less accurate one (landsat one in this case). The murray below has
  come up looking much more smooth and accurate with respect to existing
  GPS tracks when I used this method. The ABS ones also tend to have
  more smoother points on curves so this makes a nicer way in general.

- If the original OSM way is a GPS track the reverse seems to be the
  case in most cases, the GPS track will be the smoother nicer
  'looking' track (in most cases you end up with 2 ways sitting nearly
  on top of each other anyway), so you are just refining the ABS
  details when you do this.

- When it's a case of Yahoo high-zoom vs ABS it's a lot trickier and I
  think depends on how accurately alligned yahoo is in the area. It's
  never aligned well enough in Adelaide for me to come down in favour
  of the yahoo option yet. Other areas look to be better aligned so
  they may have better luck (or may just find the near-parallel way
  case arises I suspect)

- I leave the "attribution=Based on data from the ABS" or whatever it
  is tag in because even though we've adjusted the way (possibly
  chopping it up or moving it slightly) we are still using the lay of
  the ABS way to tell us it's running down said feature. I think this
  is important both legally and fairness wise.

On Ian's question:

> If so, and in a particular case we are confident that the ABS data
> isn't correctly aligned to a feature - say a coastline - how do we
> indicate that it is altered on the relation?

I haven't really come up with this in the case of a coastline yet, but
I have already found 4 missing suburbs in Northern Adelaide, and about
a dozen cases where the ABS boundary is a street or two off according
to signage or addresses of relevant establishments. In those cases I
remove the attribution=* for the part of the way that's no longer
following the ABS method because it's actively going against their data.

In the case of coastlines however it's a bit murkier, the ABS data is
effectively telling you it's going down the coastline, if you have a
better trace of the coastline you're still basing the fact that
suburb's boundary goes down that coastline on their data, so I guess
leaving the attribution=* in place is wise. Perhaps changing the
abs:reviewed=* tag to something not "no" to flag the relation has been
changed in some way is appropriate, I've been using 'yes' when I've
done all the boundaries of a suburb to any relevant common ways and
'partial' when I've done part of the boundary but still have some part
to review.

That's my 2.2c worth.

-- 

=b