[talk-au] Suburb boundaries

[Darrin Smith]

On Fri, 13 Feb 2009 17:15:34 +1030
Darrin Smith <beldin at beldin.org> wrote:

> On Fri, 13 Feb 2009 14:44:50 +1100
> Franc Carter <franc.carter at gmail.com> wrote:
> 
> > After some nashing of teeth and swearing I have script that converts
> > the ABS data in to a set of non-overlapping ways with some minimal
> > info on the ways.
> > 
> > I'd like some volunteers who I can give some subset of the data to
> > (name your subrubs/areas) to have a look over and see if it 'looks
> > ok' (i.e correct enough and no pathological cases I have missed).
> > 
> > Then, we can start making some more solid decisions about exactly
> > what form we want the data uploaded in.
> 
> I'll have a look at the Northern Suburbs of Adelaide if you like,
> compare them to the existing data I've put together :D
> (Elizabeth *, Salisbury * and Munno Para * if you need a list of names
> to match :)

Franc has forwarded me the suburbs I requested and I've had a look.
From my first look over them I've come to the following conclusions:

- The redundant point removal code does a really good job on the
straight lines, in some places removing 4 or 5 unneeded points between
two others, and I haven't found a case yet where it makes any noticable
curve shape difference either, so a win there :D

- Where the boundaries follow roads they are all pretty close to
the roads, but generally down one side, which means the roads will only
appear in one of the suburbs doing any kind of search doing this data.
This can be fixed whichever path we choose by adjusting the
way/relation to follow the road.

- Found a number of cases (mostly in industrial areas, but not always)
where the boundaries are seem to follow some arbritary smooth boundary
which is not the actual suburb boundary (according to my own research
and comparing against atlas.sa.gov.au and directories). I can only
think this is because the ABS statistical divisions are crossing subrub
boundaries but are only allocated to one of these suburbs. (An example
is GMH Elizabeth (Vale) which shows up as being inside Elizabeth South
whos boundary is a block east of there. However in some cases this
involves 1/2 dozen streets being in the wrong suburb - so there's
definite need for review.

- There are a few subtle weird boundary things at some intersections
where it closely follows one edge of the road (even following the left
turn lane edges at one point of a side road off Main North Road), which
aren't really accurate against any other data I can compare it with.
Again there will be a need to review these and tidy things up.

This all leads me to the conclusion we need to consider some kind of
system akin to the TIGER import where all ABS imported data is flagged
with an abs:reviewed=no tag (Or some similar setup) which gets
changed/removed when someone tidies up these kinds of issues so that
the next ABS import in 4 years knows to not fiddle with these ways or
work out some other solution to flag for review, a problem for 4 years
time ;) Of course as soon as the way is moved it's not quite the ABS
data any more, but still derived/confirmed by it so we need a tag to
put on those ways to signify this to keep things attributed.

It does look like it's a good 90% start to getting all the suburb data
in there though, it puts us *way* ahead of where our current suburb
data is.

Another thought is that given a large portion of .au (all the major
cities afaik, happy to be corrected) have postcodes aligned with
suburb boundaries do we even need to do a data import of those or can
we build the postcode boundaries on top of the suburb boundaries? 

-- 

=b