How extensive is OSM Sidewalk data?

Question

Pardon the rookie question, but I've been investigating the OSM dataset for sidewalk data.

I've been looking to use the OSM sidewalk tag dataset to determine accessibility maps (i.e. which areas are walkable), but so far the queries that I've been running seem to be showing a very sparse set of sidewalk labels.

Is the sidewalk-tagged dataset known to be pretty thin, or am I just going about this totally wrong?

My current validation method is running the wiki-recommended (https://wiki.openstreetmap.org/wiki/Key:sidewalk) query:

/*
This query looks for nodes, ways and relations 
with the given key.
Choose your region and hit the Run button above!
*/
[out:json][timeout:25];
// gather results
(
  // query part for: “sidewalk=*”
  node["sidewalk"]({{bbox}});
  way["sidewalk"]({{bbox}});
  relation["sidewalk"]({{bbox}});
);
// print results
out body;
>;
out skel qt;

on Overpass Turbo (http://overpass-turbo.eu/), and moving to some heavily populated areas (e.g. Manhattan). It looks like there are very few actual edges that are returned from this query.

Please feel free to call me out if this is just a lack of knowledge or if I'm going about this totally wrong, but I'm wondering if the dataset actually is this sparse, or if the error is between the chair and keyboard here.

Any insight?

Answer 1

Sidewalk coverage in OSM is far from complete and obviously varies a lot depending on which part of the world you're looking at. However, I suspect that the poor results of your Manhattan query have a different reason: There are currently two competing approaches for mapping sidewalk information in OpenStreetMap.

• Adding the sidewalk as an attribute to the road using the sidewalk=* key (this is what your query looks for).
• Drawing the sidewalk as a separate way next to the road using the footway=sidewalk tag.

At the moment, the two styles are roughly equally common. The US community appears to disproportionately favour separate ways, so your particular query would be more successful looking for the second approach, or preferably both at the same time.

Both approaches have benefits and downsides. For example, separate ways allow more detailed geometry, but lose the semantic relationship with the road. I hope the OSM community will eventually arrive at an unifying standard that works for all use cases of OSM data.

Answer 2

In general, I think the answer from Tordanik already summarizes the current situation quite well. I just wanted to add some data points and diagrams to support and enrich some of his statements about the two variants of mapping sidewalks in OSM:

At the moment, the two styles are roughly equally common.

This can be seen in the chart below. What can also be seen is that while the "mapped as attributes of roads" style was more popular in the earlier days of OSM, the "mapped as separate ways" has more momentum at the moment:

In order to compare the two mapping variants the numbers were calculated in the following way:

• for the "mapped as attribute" variant, only "positive" attributes were counted (i.e. attributes like "sidewalk=no" or "sidewalk=none" were omitted) and the road length for sections having the tag "sidewalk=both" was multiplied by two.
• for the "mapped as separate way" variant, only ways with the tag "highway=footway" and "footway=sidewalk" were counted.

A similar graph can also be produced by counting OSM objects (instead of summing up their mapped length) which can be seen here.

The US community appears to disproportionately favour separate ways

For this, I have did basically the same analysis from before but once limited to North America and once again limited to Europe, which gives us the following:

PS: To get the data for these graphs, I used the ohsome dashboard which itself uses the ohsome API to analyze the historic development of OSM data.

Answer 3

Our experience in Denver is mixed. OSM has some old sidewalks, but the sidewalk data that was collected by the Regional Council of Gov'ts (DRCOG) has not yet made it into OSM (though they are trying). For this reason, we don't use OSM sidewalk data for any of our routing/walking applications.

I am also torn on how we would best use sidewalks - do we want the actual sidewalk line data to travel down, or do we want roads with SIDEWALK = YES as an attribute? Both have their pros and cons.

As a proxy, we use speed limits and a few other attributes as defined in the OSRM 'walking' profile.

Answer 4

The answer to the question really depends on the level of detail. sidewalk=* has been mapped very extensively, footway=sidewalk for less roads but with more occurrences due to the more detailed mapping style. See also @tyr's answer.

We use queries like this one to search for sidewalk tags that have been rarely used so far (some because of their experimental usage):

is_sidepath:of=residential

[out:csv(country, "count")];
area[boundary=administrative][admin_level=2]["name:en"]["name:en"!~" - |,|\\("]["name:en"!~" - |,|\\("];
foreach->.c(
  way(area.c)["is_sidepath:of"=residential];
  make count country = c.set(t["name:en"]), "count" = count(ways);
  out;
);

foot=use_sidepath

[out:csv(country, "count")];
area[boundary=administrative][admin_level=2]["name:en"]["name:en"!~" - |,|\\("]["name:en"!~" - |,|\\("];
foreach->.c(
  way(area.c)[foot=use_sidepath];
  make count country = c.set(t["name:en"]), "count" = count(ways);
  out;
);

We are still looking for a solution to query more than one tag count column at once, but already like presented here the queries reveal quite interesting results about different sidewalk tags.