OSM exists now since more than 20 years. During this time, a tagging scheme evolved. It has undergone several changes since, is partly approved by the community, but never anything was forced. The freedom to invent new tags is considered a strength of OSM.

The OSM tagging scheme is clearly a result of swarm intelligence, which is sometimes said to be superior. I wondered how well this worked and thus started to analyse it. I’ll probably blog about this in a loose series of blog posts, starting today with looking at the highway tag, which according to the wiki identifies “any kind of road, street or path.”

In most cases, OSM elements are categorised in a hierarchical manner, with every step narrowing the meaning of the former step: highway=service, service=driveway, driveway=garage for example. A top level type=highway is missing and has to be derived implicitly.¹

The values of a certain tag should ideally be choosen in a way, that there is a match for every object in the real world belonging to this category. And there should be only one match. So, a highway=service must not qualify for a highway=footway at the same time.

Restricting my study further, I choose to look only at values used for linear features, that is, way elements. There are about three dozen of common values in the database. They match the values listed at the wiki page.²

 

The Present: Criteria used for Highway Classification

When you are faced by a highway feature on the ground, which is not yet in the database, you have to decide about the value of the highway tag.

While it’s sometimes really easy to decide which value is correct, other situations lead to endless discussions in the forums. Such discussions are a hint that there is something wrong with these values. So I asked myself, by which criterion the values have been decided. As it turns out, there are severeal criteria in use, and they are mixed.

The following figure shows a venn diagram containing most of the common values. I tried to sort them by the criteria used for distinguishing values.

First of all, there are some values that classify the highways by (destinated) access: motorway, footway, cycleway, bridleway, pedestrian and busway. They are often recognizable by a sign at the beginning of the way (at least in Germany). There is one more value, which is recognizable by a sign, but does not fit well in this diagram: living_street. It differs from the other values in not defining a certain destinated access. (And in Germany trunk is also recognizable by a sign, but that is not the case internationally.)

For (motor-)vehicles there are additional values which classify highways by importance³: trunk, primary, secondary, tertiary, unclassified, service and track.

There are also some additional values for pedestrians. These are classified by make-up: steps, ladder and elevator.

Two values are defined by the surroundings of the highway feature: residential and corridor. The first one is (in my oppinion) identical to unclassified in a residential area and the second one is just an indoor footway.

Two values are lifecycle values: proposed and construction.

Then there are two joker values that could be used almost everywhere: road and path. While the value road is needed, because there is no main tag (it replaces a type=highway without a highway tag), path never got a clear definition and is abused (in my oppinion) for stuff like mixed foot- and cycleways.

Finally there are some values with very special meaning: escape, bus_guideway, emergency_bay and via_ferrata.

As you can see, there is no single criterion that determines which value is used. If there were, there would only be one circle in the diagram above. Instead there are several criteria that overlap. This is far from ideal.

 

The Past: Historically Grown

Two questions come to mind: First: How did this come to be?

In the beginning of OSM, people wanted to create a map. A map, like the ones you could buy in a shop. These maps where car centered, like most of the thinking of our society at these days and thus it is no surprise that the tagging scheme evolved was somewhat car centered too.

And you have to keep in mind: Back then, there was nothing but a blank sheet of paper. I remember, when I started in 2007: I lived in Freiburg (on the edge of the Black Forest). Only a handful of Freiburg’s streets were mapped at that time. That was it. I started cycling all the other streets using my old Garmin Gecko device, which could only save the tracks of a little bit more then one hour of survey. After that I had to go home to enter the streets (adding highway and name tag) and then I set off again.

After several weeks (or months), I decided that I had completed Freiburg (meaning I had added all the primary, secondary, tertiary and residential highways). I’m not sure, but I think I also added some cycleways and some footways. At least I remember that, as a member of the germans cyclists federation (ADFC), I was keen to do so.

The tagging scheme of the highway tag evolved in these days. Could we have done better? Maybe. At least I think I can safely say that creating a group of experts to develop a tagging scheme would not have been better. OSM would probably have died before the experts had finished. It was very important that people could just start right away and use what’s there and invent what wasn’t there.

What could have been improved is a consolidation phase. I mean: Taking a step back, reviewing what we did, and deciding how to improve things. This happened, but it typically involved a lot of discussion, often resulting in no decision or, worse: contradictory results. I think swarm intelligence failed to some degree in this respect. Perhaps an expert group doing this consolidation would have been a better idea? I don’t know.

 

The Future: Will the Mapping Scheme be Improved?

The second question is about improving the current state. Should we attempt to do so, or is it already too late?

In politics, conservative parties typically establish stability, which is a healthy thing. The downside of stability is a lack of adaptation to new environments, which can lead to disruption from time to time.

A better approach would be to do something in the middle: Adapting in small steps. This maintains stability while avoiding disruption. However, it’s a balancing act.

The same is true of the OSM tagging scheme: I recently read somewhere that an OSM contributor is waiting for an OSM fork, before continuing to contribute. Other mappers have left the project completely, either complaining about the tagging scheme or doing so in silence. These are signs of too much stability and too little change.

Specifically, with the highway tag, my first step would be to get rid of the values proposed and construction. They are clearly superseeded by the lifecycle prefixes and could be automatically transformed into proposed:highway=road and construction:highway=road immediately.

Next, I would replace residential and corridor with unclassified and footway, but I can predict that this will cause some resistance from part of the community. Alternatively, residential and unclassified could be merged into a new category, such as normal street or so. This might be better, because people confuse unclassified with “no value” anyway.

After that, things will get more complicated. I think, that the chaos with path should be improved next. Paths should only be paths. Foot-, cycle- and bridleways, as well as combinations of these should be grouped into a single category and further differentiated using the access tags.

That’s three steps already. Enough for the time being. Don’t take on too many steps at once. I would be happy if at least the first of these step were taken in the near future.

---

1. One may think that deriving this top level tag is a simple thing. This may be true for a human in most cases, but consumers of OSM data are quite often computer programs. They have to check all tags that could qualify as the so called main tag - at the time of writing this are roughly 30 tags. Further more: This main tag can have a lifecycle prefix, which multiplies the number of tags to be checked: In practice you have to do about 300 string comparisons for every element in an all purpose application. String comparisons are expensive (which is computer scientists way of telling that there is no fast way to do so). Even with the use of hashtables, which can speed the whole thing up I estimate deriving the main key(s) of all elements from planet.osm to take about 2 hours at least. ↩

2. I omitted the variants ending in _link for simplicity. ↩

3. The term “importance” here is defined by the role they play for connecting places (when using motor vehicles). ↩