"Flooded" map

[nlehuby]

I've built my own map for this region in OSM: https://www.openstreetmap.org/#map=14/5.3248/-4.0158

But the resulting map in A/B Street looks like the whole neighborhood is under water 😆

It seems only cosmetic, I couldn't see any impact on the traffic simulation.

[dabreegster]

Guess it's a flood simulator now. >_< Turning clipped OSM data into water polygons is a bit hard. https://github.com/dabreegster/abstreet/blob/9aa4fda837ab813fd2e6d7e4381208ee03ab55c2/convert_osm/src/osm_reader.rs#L347 is the relevant code. This roughly follows https://wiki.openstreetmap.org/wiki/Relation:multipolygon/Algorithm, then tries to deal with disconnected ends by finding a spot where the half-formed ring intersects the boundary polygon, and tracing along it.

I think the problem here is that when tracing along the boundary polygon, little bits of land reappear, and the code doesn't deal with that case yet. I don't know if I'll be able to look at this soon, but if you want to try to tackle it, run with --dev and then enter debug mode with control+D. You can select the water polygons and press x or F2 to look at the points and triangles.

[dabreegster]

I happened to hit this problem in Manhattan and found a possible workaround -- flip the order at https://github.com/dabreegster/abstreet/blob/41e80b48a16c624eda533de21131f2b8fb947407/geom/src/ring.rs#L94 to get the inverse polygon. Always picking the shorter connection doesn't make sense. There might be a way to decide automatically whether to pick the shorter or longer connection.

[dabreegster]

Stumbled across https://github.com/QwantResearch/osm_boundaries_utils_rs/blob/master/src/boundaries.rs, which could be the start to something more robust

[matkoniecz]

I had similar issues - maybe process water area data by some external software such as mapshaper and load already clipped water polygons? ( https://github.com/mbloch/mapshaper )

[dabreegster]

I had looked around a while ago for existing water polygons and didn't find much. I missed mapshaper; I'll see if it could be an easier option. Thanks!

[dabreegster]

Another case that we need to handle is gluing two pieces of coastline together that're expressed in different ways. In Lisbon we have a natural=coastline way (https://www.openstreetmap.org/way/975228400) that ends right where a natural=water, type=multipolygon (https://www.openstreetmap.org/relation/2805665) begins:

But maybe not... I also see another section of natural=coastline that seems connected: https://www.openstreetmap.org/way/692883618. And it seems to keep going. Need to dig into glue_multipolygon to see what's happening here. The multipolygon is a red herring maybe.

[Darellur]

From https://wiki.openstreetmap.org/wiki/Tag:natural%3Dcoastline
"The direction the ways are drawn is very important! They must be drawn so that the land is on the left side and water on the right side"
Has this been considered?