OpenStreetMap: Addressable?

How on Earth are we going to add addressing in to OpenStreetMap?

Today OSM is a great display map. It’s routable too if you squint. But it’s essentially not geocodable, you can’t turn an address in to a location.

If we fix that then there’s really not a whole lot of point to ever using a proprietary map ever again.

Here in the United States there are essentially two readily available sources. TIGER data has address ranges between intersections and counties (all 3,000 odd of them) have parcel data.

TIGER is public domain but it’s kind of crappy. There is a problem importing it because automatically taking TIGER ranges and putting them in our map is non-trivial. In lots of places new roads have been added, old ones deleted and so on. So, getting an address range in TIGER and then figuring out where to put it in OSM isn’t always easy.

Parcel data is much better but it’s all over the place. There are companies which will aggregate it together for you and sell it to you, but that’s millions of dollars of cost. And, they have no incentive to make it all available.

And it’s even worse in Europe. And even more worserer in Japan where addresses are assigned according to the age of the house and the block they are on, which may as well be random().

So what the hell are we going to do?

I say import the TIGER ranges and slap them on top of the map. They won’t impact the rendering. It will be kind of painful to go and fix all those ranges but it’s much better than what we have today (which is nothing).

All the other solutions are basically horrible. We could crowd-source it but that might take 100 years. We could try and raise the money to purchase the data. We could go visit all 3,000 counties. None of these is palatable.

So, let’s just import what we have available and make the most of it. It worked for the road network, it can work for addressing.

27 Responses to OpenStreetMap: Addressable?

  1. Russ Nelson November 28, 2012 at 6:52 am #

    Agreed. Throw in the TIGER addressing data. We can fix it on the fly. But FFS, use the 2012 data.

  2. Komяpa November 28, 2012 at 7:16 am #

    Did you just forget Belarus and Russia?

  3. joe November 28, 2012 at 7:39 am #

    Importing the ranges from TIGER would give broader ‘base’ to begin addressing in OSM – but eventually, do you have any thoughts about using address information within building ways? Wouldn’t this be more precise, better? Definitely slower – yes. Related Imports list post

    • Steve November 28, 2012 at 7:42 am #

      Yes eventually every house should be addressed. TIGER ranges are a stop-gap.

  4. spara November 28, 2012 at 7:54 am #

    How do you propose to addresses from parcels to open streetmap? As nodes?

    Another thought is that additional processing is needed for parcels, centroids won’t work. They need go be snapped to the road network in some fashion. Some guidance/schema is needed.

  5. apmon November 28, 2012 at 8:08 am #

    I think it would be good, if the tiger address ranges could me made available as osm files for manual merge first. Then we can manually merge those address ranges in areas where we have active contributors.

    That would allow to gain some experience with quality and merging issues and identify what aspects are particularly tricky. It would also allow to see how mappers react to the imported data and if people will improve it?

    Eventually, once the above has gone smoothly and we learnt the lessons, one can think about a full scale import for those counties that don’t have active mappers yet.

    Overall, I support the idea of imports, but we should make sure we have learnt the lessons from the many issues of previous imports!

  6. ToeBee November 28, 2012 at 8:20 am #

    I’m a little hesitant on another TIGER import. Not completely opposed… but hesitant. Nominatim already uses the TIGER address range data (not sure how recent though) as a secondary source to fall back on if no matching address information is found in OSM data. Could this concept be used in other applications like mkgmap for garmin? Or what about creating a US extract with TIGER address ranges added in?

  7. pnorman November 28, 2012 at 9:41 am #

    I don’t see that there would be any benefit to importing the TIGER data, which is intentionally made worse by the census bureau.

    In the short term it would do nothing to help geocoding. Right now geocoders can and do use TIGER as a fallback so there is no difference between the results now and the results if TIGER addresses were imported.

    In the long term it doesn’t help either. The TIGER interpolation ways would make it more difficult to map better addresses and because the update problem is not yet solved address results would be worse then if we hadn’t imported when TIGER 2013 becomes available.

    • Steve November 28, 2012 at 10:31 am #

      In the short term it absolutely does help since I can correct the data if its I’m OSM.

      • AndrewBuck November 28, 2012 at 10:53 am #

        But you can already “correct” the tiger address data by just putting in the correct addresses into OSM anyway, which will cause it to override the tiger fallback in the geocoders.

        Importing the roads made sense because they could easily be fixed up by aligning them to imagery, and they allowed a quick way to make a routable road network by connecting the “islands” where people had mapped into one contiguous system. We also got the road names, which could not be gotten from imagery alone.

        Address data is not like this, however. It does not need to be made “routable” so having gaps where we fallback to a different source is fine, and furthermore, the only way to “correct” address data is by just walking around and collecting the data yourself. This is exactly what is required to just create the data ourselves, so there doesn’t seem to be an advantage to importing the data.


        • Steve November 28, 2012 at 10:57 am #

          Right so I can go address all of a city myself and it would override, theoretically, one geocoder. What about the hundreds of others?

          Importing gives me a head start to fix the same city for everyone at once.

          • AndrewBuck November 28, 2012 at 12:52 pm #

            But the import doesn’t actually make the addressing easier. The only way to check and/or improve addresses is to go building by building and add each house number. This is exactly what you have to do with no addresses in the DB.

            From what I understand the tiger address is only interpolation ways and is intentionally obfuscated. This means that we still need the same amount of work to gather proper addresses so the import gains us nothing, and just discourages people from actually going and gathering them.

            With the road data, they did discourage some mapping, but the also made other mapping easier. With the address data there are only the discouragements and no advantage to mapping.


          • Steve November 28, 2012 at 12:55 pm #

            No because the tiger ranges need to be split in to blocks and pushed off the centerline towards the houses. For most of American grid-like cities that instantly improves things for all the geocoders.

          • AndrewBuck November 28, 2012 at 12:55 pm #

            PS. I agree that having the data in the DB would be nicer for endusers, however I don’t think that argument alone justifies its import. There are lots of datasets we could merge in to make endusers jobs easier, however that argument has never held sway. Imports have only been justified on their own merits.

            There are other map projects that act as aggregation services to do this, we shouldn’t pollute the DB with marginal data just to make it easier for downstream users unless the data is useful to mappers.


          • Steve November 28, 2012 at 12:58 pm #

            Good then we are mostly agreed.

            You can’t call addresses marginal. It’s he last piece of the puzzle and critical for people switching to OSM.

          • AndrewBuck November 28, 2012 at 12:59 pm #

            Hmm, I didn’t realise that the data could be imprved by splitting it and putting in offsets. This does change my opinion of it as it does offer an avenue to improve the data without having to resort to doing a complete manual survey.

            If this can be done then I guess I would support the import, assuming it is done properly, etc.


          • AndrewBuck November 28, 2012 at 1:01 pm #

            I didn’t mean that addresses themselves were marginal, just that the tiger data is of marginal quality. It does sound like we are in agreement now though.

            You should probably add an addendum to your blogpost about the splitting/shifting concept as I think that is where most people would be confused about how useful the data could be to the project.


          • Steve November 28, 2012 at 1:14 pm #

            Makes sense

  8. Brian May November 28, 2012 at 5:03 pm #

    I think we should start with states that have favorable open records laws and start with accurate address points that are available from many county governments. If that is not available, see if parcels and building outlines are available and conflate site addresses from parcels to buildings. If that is not available, then go for parcel centroids. And yes, there’s the multiple addresses per single parcel issue to deal with – that will need to be a more manual thing. If you can’t get parcels, then go with county government streets for interpolated addresses and last look to tiger 2012. Tiger has been greatly improved since the original import, but is still lacking in many areas, especially compared to data available from local governments. There is a LOT of high quality local government data out there. At least that is what I see in Florida. FYI: Florida has very liberal open records laws.

  9. Jake November 28, 2012 at 8:06 pm #

    “If we fix that then there’s really not a whole lot of point to ever using a proprietary map ever again.” Except for the OdBL.

    It’s still not clear (to me anyway) that someone could geocode a large number of addresses and limit the distribution outside of the OdBL. As long as companies are afraid, they will continue to pay premiums to proprietary map providers instead of encouraging and contributing to OSM.

  10. Chaos99 November 28, 2012 at 11:27 pm #

    So we have learned nothing from the first TIGER import then?

    Importing the complete road data made the US the best mapped country by then, as we had only crowd-sourced a few ‘islands’ in other places. Just watch the old ITOworld animation. Boom – the US was mapped in just one import, while Europe was still struggling with just minor coverage.

    But now, some years later, we have a (perceived) 98% coverage in Europe, with details way surpassing commercial providers, because we pushed the survey-based mapping to its full extend. (Non-Mappers now happily give us detailed POIs and address data through phone apps and such.)

    But how about the US, that made that bold step years ago? It hasn’t evolved any further. The full coverage discouraged mappers to actually survey and now we have just the crappy TIGER data as we had years ago. No wonder so many cried out when popular sites switched to OSM. I’ve contributed to the recent Operation Cowboy mapping party: even in populated places the TIGER data is still untouched and way off.

    And now we want to do that again. Import data with full coverage, but bad quality. Discouraging survey for years.

    So it comes down to this: Do you want crappy address data NOW? Then go on and import, but don’t think the quality will increase anytime soon. Or wait and have address data in some years, but with a quality that can surpass your competitors.

    I would have ended with ‘Your call’, but I sincerely hope not.

  11. Alex-7 November 29, 2012 at 9:15 am #

    I tried to use OSMpad application for iPhone to collect addresses. There is also a version for Android.

    It shows the current location, the street name from the OSM, and all one has to do is to point the pointer to the building and type a house number.

    Then the resulting *.osm file is loaded into the JOSM. And houses are numbered.

  12. Mark Gray November 29, 2012 at 9:31 am #

    I am concerned that importing address range ways makes fixing alignment of roads much harder. In places where TIGER address ways have already been imported, I have been frustrated that instead of just moving the road to where it belongs, you have to also move addressing ways on each side of the road and addressing ways of all intersecting roads. There is also the additional difficulty that they should be kept a nice uniform distance from the road which is more of a challenge than it sounds.
    Address ways also complicate adding other features such as buildings because editors want to create inappropriate intersections with them. These factors have made me not a fan of address ways. Accurate address points and addresses on buildings are fine, but address ways seem to be more trouble than they are worth.

  13. Coleman McCormick November 30, 2012 at 1:10 pm #

    I don’t see why import of relevant content discourages survey so much, even if some of its qualities are subpar. Are there hard examples of this being the case in places where imports have been done? Were there places with huge, active communities of field surveyors where an import was done and nuked community contribution thereafter?

    I personally feel much more confident in my own edits, and in the long-term success of the project if participation is focused around correcting and augmenting imported data than relying on contributors to field survey 3.5+ million square miles of roads and places (in the US) on a blank canvas. Realistically, the levels of detail required for commercial success (like address data) won’t be achieved with volunteers having to do every shred of the work from scratch.

    In the evangelism I’ve done in our FL community, and with friends and colleagues, there are enough technical hurdles to overcome with getting casually-interested users to map without also throwing people onto a blank canvas and saying “okay, start mapping things”. Casual contributors need to have meaningful, small bits of things they can contribute, such as fixing alignments of streets or correcting road names and addresses. Almost every newcomer at our mapping parties not familiar with OSM has the question – “What kinds of things should I map? What does the map need?” If I can show them how to fix incorrect data, rather than create whole areas from scratch, it seems far more realistic that they’ll stay actively involved.

    I know there are technical issues with another national TIGER import to get it right, but don’t really know what the best approach is…

  14. Peter October 14, 2013 at 11:58 am #

    Use the TIGER data! Sure, maybe it’s low-quality and has to be manually edited, but in the meanwhile, it gets us something. The difference between having low-quality data and no data at all is that people might use something with low-quality data. With no data, I can’t even put in my home address and navigate there, so I uninstall whatever OSM app I might have been trying and use a proprietary map (Waze, Google, Bing) instead, and OSM never grows.

    I did a scan of the OSM database once, and found that 75% of roads imported from TIGER are on version 1… they’ve never been edited by anyone. If those roads didn’t exist, the US would be a mostly blank area on the map.


  1. Offline Android Maps Just Start Of Skobbler’s Ambitious Roadmap | Now - December 4, 2012

    […] in some of the gaps that the crowdsourced map still hasn’t yet done on its own, such as comprehensive address and street number data for the U.S. Not to mention the fact that OpenStreetMap on its own doesn’t provide anything as fancy as […]

  2. Warum wollen wir Hausnummern erfassen? | OSM Bremen - August 26, 2013

    […] Steve Coast, der Gründer des OpenStreetMap Projektes, hat in einem Blogartikel darauf hingewiesen, das diese Identifizierung bei OSM noch im Argen […]

Powered by WordPress. Designed by WooThemes