TIGER data quality

Minh_Nguyen · 2022-12-19T16:47:12.135Z

Some classes of problems were common to the entire TIGER import, but we largely fixed those. Off the top of my head: abbreviations in road names, disconnections at county lines, divided roads intentionally digitized as single carriageways, disconnections at railroad crossings. Probably the only one that’s still very widespread is the name_* tags, because that requires a lot of manual review.

Otherwise, the classes of problems that remain are specific to individual counties (albeit many counties) scattered throughout the country. For example, overnodedness happens in many counties but is difficult to clean up en masse. Many counties also have every private driveway tagged as a highway=residential. Some have every parking lot outlined as a highway=residential. It’s easy to perceive these problems as being common to the entire import because of our personal experiences focusing on counties where these things occur.

mvexel · 2022-12-19T16:59:39.910Z

Agreed that the problems are very localized. At this point the best we can do is therefore come up with local solutions. I’ve done state-wide cleanup challenges in #maproulette before but those prove too daunting…Tens of thousands of tasks. Here in Utah the main problem remaining, and it’s a big one, is the huge number of “made up” rural roads, geometry that does not represent anything currently existing on the ground, or I would not recommend anyone rely on for anything. I’ve considered proposing deleting everything in Utah that is TIGER imported and never looked at by a human. Either as a bulk operation or as something more targeted (not sure what that would look like). Any experience with localized “reverting” of TIGER import data?

ezekielf · 2022-12-19T17:26:54.043Z

Would it be possible to build some sort of heatmap overlay showing the density of potential remaining TIGER issues? This could help interested mappers find areas to focus on.

Richard · 2022-12-19T17:32:54.962Z

So the big issue is A41-class roads in rural areas, particularly (though not exclusively) away from the coasts. These were imported as highway=residential. Some are indeed residential highways! But most aren’t. Some would be better as highway=track, some as highway=unclassified, surface=unpaved (or something more nuanced), and some would be better simply deleted.

My interest is that this is particularly problematic for cycle routing. Car routing is less affected because it prefers the higher routes in the hierarchy (trunk, primary, etc.), which are mostly fixed. Bike routing is the opposite - it prefers unclassified/residential etc. - and if you try and route across the US on highway=residential, you will basically die of dysentery somewhere on the way to Oregon.

We’re unfortunately a long way past anything that can be sanely reverted. There have been lots of incremental little fixups over the years, plus the blizzard of corporate edits relating to driveways, that mean most ways have indeed been edited somehow over time. As an example, there are several counties where (say) a maxspeed=45 mph tag has been added to every highway in the county because that’s the local ordinance. This means you might find a drainage ditch tagged with highway=residential, maxspeed=45 mph

A smart(ish) data consumer can go a long way to alleviating these issues with several heuristics. cycle.travel is very distrustful of highway=residential with tiger:reviewed=no (and no surface tag) in rural areas. The upshot is that you can use it to route across the US with cycle.travel and you will probably not die of dysentery. But it would be better to improve the source data.

I’m not against carefully targeted automated edits when the point is to fix issues with an earlier automated edit, i.e. TIGER. For example, a few states (Colorado springs to mind) publish open data on road surfaces. This could be sanely brought into OSM. The imagery-derived surface detection mentioned in another thread also looks really promising. And, of course, maybe Overture Maps will be releasing some relevant open data… who knows.

But anything automated would have to be very carefully reviewed for fear of blatting the few usable heuristics we have at the moment.

mvexel · 2022-12-19T17:43:30.319Z

Richard:

cycle.travel is very distrustful of highway=residential with tiger:reviewed=no

You should be, and you’d know because you’ve looked into this pretty deeply. My heuristic for separating wheat from chaff is:

highway=residential
No name=*
existence of tiger:cfcc and / or tiger:reviewed=no
Last touched date long ago (I usually query for <2012)

There is a bunch of “advanced” overpass queries that have more elaborate criteria, @Minh_Nguyen would know where to find these.

Minh_Nguyen · 2022-12-19T17:52:54.964Z

mvexel:

Any experience with localized “reverting” of TIGER import data?

My last experience with this was in 2009 when I asked for all of Greene County, Ohio, to be deleted – because it had been imported twice, every road duplicating another copy of the road without any connections between them. I had already made many edits in the area and had haphazardly deleted many roads from one or the other import, but it only took me a couple months to recover. I don’t think we could’ve done something that clean in 2011. Granted, Greene County is much more developed than some of the counties in Utah where you map.

mvexel:

There is a bunch of “advanced” overpass queries that have more elaborate criteria, @Minh_Nguyen would know where to find these.

Here are some Overpass queries for unedited TIGER ways. As @Richard points out, there are many false negatives because of driveway editing and such. The public Overpass instance can’t handle querying for TIGER unedited ways beyond a small area.

A couple years ago, I developed some SPARQL queries to determine the most deserted TIGER desert counties, and even refined it down to individual ZIP codes. Unfortunately, Sophox is no longer reliable for these queries, but I made this snapshot in 2020 that might still be useful.

mvexel · 2022-12-19T17:59:07.262Z

This is an Overpass query I’ve used in the past that trades off reasonable speed for reasonable accuracy:

rel(161993);map_to_area->.a; // state of utah
way // consider ways that...
  [highway=residential] // are residential 
  [!name] // do not have name tag
  ["tiger:cfcc"] // have tiger:cfcc tag which was created as part of the import
  (if:timestamp() < "2013-01-01T00:00:00Z") // have timestamp before 2013
  (area.a); // are in the defined area
out meta geom; // output geometry and metadata

This yields 2176 ways. If you leave out the highway=residential criterion it’s up to 8000+.

Richard · 2022-12-19T18:49:50.355Z

There’s possibly an argument that some of these highway=residential roads should be automatically retagged as highway=road - which itself is effectively a fixme tag.

NM is probably the state with the most un-road-like A41s IME. Some of the geometry in WV is pretty shocking though…

MikeN · 2022-12-19T19:38:52.567Z

I can see that it might make sense to remove untouched TIGER in ‘wilderness’ areas, but even then by someone in the region who has familiarity.

But locally it would make no sense to take any sort of automated or mass action. I’ve been going over counties and setting the surface and geometry where not hidden under trees, using the US Tasking manager. Even that requires knowledge of regional road construction, soil, and any possible better local authoritative sources of road names and geometry. I’ve also found that the commercial ‘driveway mappers’. have often taken the time to improve roads nearby where the original TIGER was wonky.

mvexel · 2022-12-19T20:25:41.065Z

Welcome to the new forums, @MikeN !

As a compromise, I think it is fesaible to do MapRoulette challenges for smaller areas. I discovered an interesting dividing line at the Navajo Nation boundary where inside NN there’s many old, untouched TIGER residential roads, and outside NN (at least on the Utah side) almost none. I don’t know if this is just selective mapper activity or inconsistencies in the TIGER data coverage, or something else. But I made a MapRoulette challenge to encourage people to help with building a better road network for the Navajo Nation: Martijn van Exel: "Mapping Inequality The map below shows SE Utah w…" - En OSM Town | Mapstodon for OpenStreetMap

stevea · 2022-12-20T10:26:21.518Z

This is a neat little snippet of OT code, thanks Martijn. I’ve entered something similar for my county (in California) linked in our (county-level) wiki, it produces a bit “richer / deeper” a set of data (both nodes and ways).

I really, really miss the wonderful, deprecated (summer of '19?) ITO World “TIGER Cleanup” (I think it was called) renderer. I used this to clean (and clean, and clean, and clean…) my county until I got to something like 75% or 80% “done” (I might give my efforts a solid B-?!) and then that particular renderer quit. So sad.

I’ve looked for other “prettified” helpers / renderers to aid in TIGER cleanup, as in many cases, automation is quite ad hoc, specific to a county, state, aboriginal_land (again, Martijn, thanks for the tip about Navajo Nation, I’ll go take a look). Alas, there aren’t any renderers that suit my fancy, so what little work I now do (in my county, really) to improve TIGER is from my OT query. Somehow, because it isn’t as pretty as that ITO World version, I clean up TIGER less than I used to. I think it was the color-scheme (red, orange, light-blue, dark-blue, I think) and rather clever reasons (including “3 year aging since last edited”) that made it truly useful. I know if we got a replication or close to it, I (for one) would slash away towards 90%, then towards 100% (again, in my county, where I concentrate my mapping efforts, especially for TIGER fixup).

I do recall one august volunteer in this project (I have a lot of offline email conversations with him) calling TIGER, in many cases, “not much better than an hallucination.”

Anyway, I’m dedicated to improving TIGER data, locally, more widely (statewide, and indeed, there is a lot to be said for state-by-state “divide and conquer,” as we’ve done a decent job of whacking rail data from TIGER down, though there’s still tens of thousands of rail miles to go, and these aren’t getting easier to quantify). I’d love to know that better tools are available. OT queries are good, but they’re wonky and largely used by the more geek-inclined (no offense to geeks, I actually proudly have the word on a license plate of a car of mine).

Yes, it might be the 2040s before we clean it all up. My sleeves are rolled up, and have been for a while.

stevea · 2022-12-20T10:41:03.633Z

Richard, I don’t know if you’ve been to West Virginia or know much about it, but it’s an outlier among states in some interesting ways. For one example, it seems to be deliberately “radio signal quiet,” I believe part of that was or is for a radiotelescope near there that needs to attenuate interference, improving its signal-to-noise ratio. A number of things seem to “fall off the map” when you enter West Virginia, it’s hard to explain. I’m sure there are reasons for such things, they seem beyond me. Maybe there’s an article written about why.

Richard · 2022-12-20T11:12:03.990Z

stevea:

I’ve looked for other “prettified” helpers / renderers to aid in TIGER cleanup

cycle.travel can be used in some areas. It shows unfixed TIGER residentials in rural areas as a faint grey dashed line, like this:

Screenshot 2022-12-20 at 11.10.05661×685 99.1 KB

But it won’t be 100% reliable for this purpose - in many areas it has additional heuristics to guess what might be a usable road, and it updates roughly once a month so it’s not ideal for real-time fixing.

stevea · 2022-12-20T11:25:38.076Z

While I’ve perused cycle.travel on my little county before (I did develop and propose to the transportation commission the “CycleNet” bicycle local bike route numbering protocol), thanks to your “unfixed TIGER residential = faint grey dashed lines,” it visually now makes much more sense! I’m not sure how you determine / calculate “rural areas,” but my eyeballs are quickly getting retrained as they parse your semiotics. Thanks!

Minh_Nguyen · 2022-12-20T17:39:07.842Z

stevea:

For one example, it seems to be deliberately “radio signal quiet,” I believe part of that was or is for a radiotelescope near there that needs to attenuate interference, improving its signal-to-noise ratio. A number of things seem to “fall off the map” when you enter West Virginia, it’s hard to explain.

I think this is a reference to the National Radio Quiet Zone, which also extends into a good chunk of Virginia. It is indeed an area where you’re guaranteed to lose cell reception, but the data quality issues in West Virginia aren’t limited to this zone by any means. I’ve cleaned up many roads that corresponded to old mining roads or roads predating mountaintop removal. Even the many roads that legitimately exist have poor geometry because most roads follow winding rivers in dense woodlands within narrow hollows – tough for both GPS reception and aerial survey.

TIGER’s data quality issues are generally endemic to specific counties, but for West Virginia they seem to be pretty consistent statewide. I wonder if this is because West Virginia maintains the entire public road network outside incorporated cities, in contrast to most states that rely more heavily on local highway departments, which are typically responsible for sending road network data to TIGER.

stevea · 2022-12-20T21:51:56.934Z

Yes, Minh: the Green Bank (,West Virginia) radiotelescope et al. Thanks for your link, thanks for your mapping of the “quiet zone” polygons.

stevea · 2022-12-21T01:08:42.605Z

Making an explicit reply here because Zeke’s excellent suggestion got a like from me and resonates with my experience with that (no longer functional) ITO World render I mentioned. Goin’ through a bit of “already” ground here (for over five years of history) at TIGER Edited Map - OpenStreetMap Wiki. The topmost render is what I’m talking about. That “overview” or “heat map” (something about those red-and-orange turning to sky-blue, then darker-blue as we’re done) clinched it as “visually parsable semiotics which make a lot of sense to my mind,” driving forward TIGER cleanup.

It worked, is what I’m saying. Replication (or something close) WOULD rekindle that fire, fairly easily, I speak for myself.

ElliottPlack · 2022-12-22T13:33:29.706Z

I can speak generally about some local governments’ I consult for. The Census bureau maintains relationships with GIS managers in local administrative jurisdictions (call them Counties in Maryland). They routinely exchange data, not just centerline but also boundary annexations, things like that.

However, not all jurisdictions participate. Some smaller ones do not have the resources to pass quality data back and forth with the Census. Some really small ones may just have a single GIS person and all of their CL could be in an incompatible format. Lots of these tiny jurisdictions around.

ElliottPlack · 2022-12-22T13:44:24.078Z

This overpass query works to find all the untouched tiger imports. In areas that have not seen much improvements this works.

// gather all Dave Hansen TIGER import edits in an area
[out:json][timeout:25];
// gather results
(
  way(user:"DaveHansenTiger")({{bbox}});
);
out body;
>;
out skel qt;

SimonPoole · 2022-12-27T13:36:48.260Z

Seems as if MS has released new ‘road detections’ GitHub - microsoft/RoadDetections: Road detections from Microsoft Maps aerial imagery this would be completly unremarkable, except that they found ~800’000km of ‘missing’ roads in the US*. Now while there is likely to be lots of ‘not actually a road’ in that data, it still would indicate that they found lots of segments that didn’t match up with the existing data. My hypothesis is that might be due to TIGER derived OSM data with geometry issues that didn’t allow matching and that the MS data might allow pinpointing that.

Can’t test that right now as I’m on the road, but maybe it is worth looking in to.

* conventional wisdom is that OSM already has more roads than actually exist thanks to TIGER.