[talk-au] Tracks flagged as missing from government data

[Andrew Harvey]

On Sun, 22 Aug 2021, at 5:39 PM, Little Maps wrote:
> Andrew, thanks for the super fast reply, and for the overpass query which I'll cut and paste from! A few thoughts…
> 
> AH: 1.98% of tracks have public vehicle access and 8.7% of tracks have no public vehicle access (of all tracks). So where we know the vehicle access then 18% are public and 81% are not public access.
> 
> This makes sense. To date, in Vic at least, most mapped tracks are on public land or on the public road network (the opposite trend may exist in outback areas). So existing mapping is strongly biased to public tracks, and access tags have mostly been used to indicate restrictions.

I'm totally guilty of this too. Warin mentioned that in NSW, tracks in National Park are usually no access and State Forests are usually accessible. I mostly encounter tracks in NSW National Parks so generally I just assumed all are private and public are the exception. We are all biased by our own experience, the only sensible way forward is to encourage always setting explicit access tags.

> I just re-read the Aus tagging guidelines and it has a similar emphasis. It explains how to add access restrictions but doesn't say that public access isn't a default on tracks or that access=public is a worthwhile tag to add. I'll put together some draft text to add to the page and will circulate for comment in a day or two.

Sounds great. Though access=yes is the tag for public access, access=public was discouraged as a duplicate of access=yes.

> I have a different take, but I think you'd be happy with my ideal router. It would give me 2 options: (1) use all available tracks (public + unknown) vs (2) only use known public tracks. Given how few tracks have an access tag, most users would default to "show me all of them", but they'd have a choice. Globally, only 3.8% of tracks have an access tag: 20.7 million of 21.5 million tracks don't. Any app that only used known public tracks would be viewed as crippled by users and would go broke. The market would force developers to show all tracks, regardless of their personal intentions.

I don't think there is any perfect solution until all tracks have an access tag, only compromises. You could decide to route on tracks including without an access tag set, with a warning or just accept there will be some bad routes and encourage users to report or fix those in OSM.
 
> Luckily for me, the strong bias of osm mappers for mapping public rather than private tracks is why routers that do assume that access is public unless indicated otherwise actually work pretty well in Vic (prob not in central Aus). As more and more private roads are added we can expect this convenient correspondence to fall apart though. That's why I was so concerned about the Challenge adding lots of private tracks without having an access tag on them, as it will be the first major influx of untagged private roads to Vic.

> A question: I don't understand how the "default value" approach differs from Joe's suggestion, which as I understood it, was that if access is assumed to be no, then he wouldn't have to bother adding access tags (inc access=unknown) when doing armchair mapping. Doesn't this have the same outcome as a default position of not needing to add a tag? However, despite the fact that I don't comprehend the distinction, I don't think it matters a great deal.

Not sure I understand, but if the access is known then I'd encourage it to be set, if it's not known, then leaving the tag value empty signals that it's not yet mapper let's data consumers make their own decision about how to treat that, and prompts other mappers to complete this data.

> If there was a discussion to try to reach consensus on whether we should assume that access=yes or no when there is no access tag, I would take one of two positions: support access=yes or continue to make no assumption about access. I wouldn't support an assumption that access=no for the reasons I've described above. I think I'd probably just take the long term view and say, avoid the debate and tag everything.

If you say access=yes is assumed default, then you run into the exact problem here, someone who knows there is a track and wants to map it, but isn't yet sure about the access get's scolded for making a contribution without the access tag. One of OSM's core strengths is that one person can make a positive contribution and someone else can come along and build upon their contribution. Like someone adding the track from remote sources so a local surveyor can confirm access and tag it.

I'm all for always encouraging tagging access (or the mode specific access tags, motor_vehicle=*, bicycle=*, foot=*) when it's known, treating it a bit different to other highway=* values which are generally only mapped when not =yes.

> By analogy, until recently the Aus community took the view that there was no need to add paved surface tags on roads and only unpaved tags needed to be added. Paved was taken as the default value. As lots of roads had no tags it was impossible to know which were actually paved and which just hadn't been tagged. Same problem to here. Fortunately, heaps of mappers added paved tags anyway, which enabled us to get to the stage this year where virtually every road down to tertiary level across the whole country now has a surface tag (except in Melb and Perth). Soon every unclassified road in Vic will have one as well. Keep chipping away at the job is my suggestion.

Exactly, and overall I think OSM data is in a much better place because of this.

> If we want to make progress on access tags, I suggest we need to discuss loosening the restrictive (IMO) approach that we currently take to adding access tags, which is to avoid adding them unless we see it on the ground. That's unscaleable across Australia in any meaningful timeframe. I'd be happy to support well-designed imports and challenges that used reputable datasets that contain access restrictions (e.g. Vic transport data; Dry WO, MVO, seasonal closures, etc.) and (perhaps) to use these datasets to indicate access=public, which is where we have the biggest gap in our data. This way we could make much faster progress. We'd make some mistakes but the system is iterative and editors continue to do an awesome job to refine an amazing map.

I don't have any strong opinion on this. I guess it comes down to the data, if it's reliable enough. Maybe I could try to compared existing tracks in Victoria which have an access value with Vicmap to see if it's usually in agreement or not. How do you rate the accuracy of the Vicmap data when it comes to track access?

> Ultimately, I'm with you in that, we can develop the best map if we accurately tag access everywhere. Thanks again, Ian

Yeah we are on the same page about that, it's just about how to bridge that gap in the meantime before all access tags can be set.

I've tried to get StreetComplete to ask about access but it was rejected https://github.com/streetcomplete/StreetComplete/issues/2930