Backfill kind_detail on aerodromes based on iata codes

[nvkelso]

In #1277 we added kind_detail for aerodrome landuse (AOI), pois, and roads (runway) layers features. But it turns out aerodrome:type isn't very consistently filled in. We want to use this property to filter features at low zooms – there are a medium amount of not international airports at the low and mid-zooms that are large but not meaningful to show that early.

• What did you see? I saw public on SFO airport south of San Francisco, California, USA.
• What did you expect to see? I expected to see international, based on it having a 3-char IATA code (SFO)
• What map location are you having problems with? 15/37.6201/-122.3775
• Screenshot?

It is on the landuse:

It is passing to the POI:

I expect the runway lines to also have a aerodrome_kind_detail property via #1854:

[zerebubuth]

I expected to see international, based on it having a 3-char IATA code (SFO)

Would you also expect to see international for Arrabury Airport? Although it has a 3-char IATA code (AAB), it looks like it only serves a remote cattle station and I'd be quite surprised if it serves any international routes. Although it does have 2 runways, from aerial imagery it looks like neither are paved.

Perhaps we could use this list of international airports to override public → international?

[nvkelso]

Huh, that's odd! Maybe it's got an IATA code because it's an emergency landing strips for international flights that need a quick landing spot in the middle of the outback?

Rather than relying on an external list... what about allowing the OSM tags to drive this?

In this proposal, since the Arrabury Airport doesn't have an aerodrome:type tag, we'd default it because of the iata value being present to kind_detail of international... but that can be fixed by adding an aerodrome:type tag of airfield. Which is admittedly confusing since we generally use airfield in Tilezen's kind to indicate a military use, while this kind_detail use would be for rural airstrips.

Alternatively, we could opt to only do the backfill when the airport has a polygon? Because Arrabury is a node it'd fail that test.

[zerebubuth]

Bunbury airport has a polygon, aerodrome:type=public and iata=BUY, although the Wikipedia entry says it has no control tower and seems to suggest that it's mainly for flying clubs and light aircraft.

Bunbury has a aerodrome=regional tag, which we could use for excluding it (although I'm not convinced it's really a regional airport - seems more like an airfield).

However, Kerang airport has no tags which obviously exclude it from being an international airport. It has iata=KRA and aerodrome:type=public, but a look at the aerial imagery suggests that it probably lacks the facilities to do international immigration / customs checks (I don't see a terminal building).

I'm not sure that simply having an IATA code indicates that an airport is of international importance.

I agree that it would be better to drive this from OSM tags, but it seems like aerodrome:type is just totally inconsistent. We could fix the data by looking at the airports that the external list says should be aerodrome:type=international but currently aren't and double-checking them.

There are 913 such airports, according to my researches (probably not all of them are indeed international airports) and most (484) are simply lacking aerodrome:type. A further 306 are tagged as public and the rest are a mix of military/public, civil, regional, military, etc...

Ironically, a small number of airports tagged aerodrome:type=regional are actually called something "international", e.g: General Santos International Airport - although a look at Wikipedia suggests it no longer has any scheduled international routes, but operates as a diversion alternative for nearby international airports.

[nvkelso]

@zerebubuth can you take another pass at this, please?

It's working on the POIs (great!):

But I was expecting international as well on the runways, too:

I think that's not right because the POIs logic needs to be applied also to the landuse polygon?

[zerebubuth]

I think that's not right because the POIs logic needs to be applied also to the landuse polygon?

Ooops! Well spotted, thanks. Looks like I meant to apply the same logic to polygons, but just forgot. Should be fixed in #1893.

[nvkelso]

This is working on landuse polygons for runways, woot!

Noting here that we don't include this for runway lines or taxiway lines in the roads layer, but that's fine – those aren't shown at mid zooms and it's the mid-zooms that are most important to solve.