[OSM-newbies] editing TIGER data

[Shawn Rutledge]

It's very nice to have this data imported, despite the inaccuracies:
it saves a lot of trouble with entering metadata.

I assumed the idea is that it's a starting point, right?  The last
couple weeks I've been editing some of those ways to make them more
accurate... many intersections are not aligned with their actual
positions, nodes need to be added to make straight ways match curves,
and every street is tagged "residential" so I've been changing the
highway type to something more appropriate for the major 1-mile and
1/2-mile streets in Phoenix, AZ.  I've been using primary for streets
that are also state highways (there are just a few of those),
secondary for the 1-mile streets, and tertiary for the 1/2-mile
streets, with some exceptions when a tertiary street is actually more
important and is promoted to a secondary.  Motorways were already
tagged correctly but I've been editing them to make sure there are
dual ways, one for each direction.  For example this area is very
close to home and I spent a lot of time on it:

http://www.informationfreeway.org/?lat=33.362129805432126&lon=-111.99522509084402&zoom=14&layers=B000F000

I was using GPS traces with JOSM at first, but realized it's actually
more accurate to use the satellite imagery (and I don't have to plan
on driving every street eventually!), so I did a lot in Potlatch the
last couple days (and once I get started it's really hard to stop...
I've got to back off a bit now).  I also hike most weekends so have
been carrying a GPS to capture hiking trails, and I do that work in
JOSM.

So I hope my fixups will be persistent.  The TIGER import was only
supposed to happen once, right?

And I cases where there are multiple ways describing the same actual
street, I usually delete the non-TIGER one because there is less
metadata, and clean up the TIGER data to be at least as accurate as
the one I deleted.

I was also thinking, wouldn't it be nice if there were something like
OCR for satellite images?  i.e. an algorithm to pick out the streets
from the rest of the landscape.  The TIGER data would provide the
metadata, and this correction of it to match the actual shapes could
be mostly automatic.  That way the image recognition doesn't have to
be perfect... if the algorithm sees an edge that's not "close enough"
in position and direction to a TIGER way, just ignore it and let the
humans deal with that area of the map.