OHM Coastlines

[geohacker]

Following from #332, we should run a pilot of the osmcoastline pipeline for OHM. The goal is to first generate a coastline from existing coastline data, see what's missing or what's broken and then decide next steps. The medium term goal is to support #329 which would have to be some sort of import.

cc @jeffreyameyer @danrademacher @batpad @Rub21

[geohacker]

I managed to get the coastline process integrated as part of OSM Seed and used it to run on a latest OHM planet. Here are some results. Overall, the process was straightforward. I generated the land polygons and uploaded it to Mapbox to visualise. You can see it here

The coastline isn't complete so that's why you see a lot of non continuous polygons. There are also issues like this in the Caspian Sea:

@jeffreyameyer @danrademacher @tsinn In terms on next steps, I'd encourage you all to take a look at the map as well as the data below:

• Full land polygons geojson
• Point errors identified by the pipeline
• Line errors identified by the pipeline

I'm still figuring what the error terminologies exactly imply but I think @jeffreyameyer might be able to get to that better than me.

This is a good overview of all the errors and what one might need to do to resolve them https://wiki.openstreetmap.org/wiki/OSM_Inspector/Views/Coastline#Layers

cc @batpad @Rub21

[geohacker]

So I loaded up the Caspian Sea issue into JOSM and the error about directionality makes sense. I tried to reverse the direction to address and that should fix it. Haven't uploaded this yet.

[danrademacher]

Good progress! Only 26,000 errors in that point file.

I am feeling a bit daunted by that!

I'm also a little confused about this, emphasis added:

I generated the land polygons and uploaded it to Mapbox to visualise.

But when I look at the linked Mapbox map, I see red polygons scattered around the world that don't seem like "land polygons", or well some do, like Antarctica and Australia. To make sure I understand -- that Mapbox preview shows errors where water erroneously renders as "land polygons"?

I downloaded the linked GeoJSON files, but the Line Errors link is actually pointing to the Point Errors file, so we need a new link for that:

• Full land polygons geojson
• Point errors identified by the pipeline
• Line errors identified by the pipeline

I'll probably still be daunted, but tomorrow is a new day!

[geohacker]

@danrademacher haha yes coastline errors are daunting but the pipeline does very good job of trying address as much to construct a good polygon ready for rendering. All the files are uploaded to this drive

But when I look at the linked Mapbox map, I see red polygons scattered around the world that don't seem like "land polygons"

The coastline pipeline can produce land or water polygons. Land will have holes for water and that can be the base layer for styles for example. Water is the reverse. So the red polygons you see is correct — just that in OHM planet, the coastline information is not complete so there are lot of broken polygons where the water is bleeding into land.

I don't think we should try to fix these errors right now. If the goal is to import ice age time coastlines, then that's our next step.

[danrademacher]

Great! @tsinn has some solid ideas on how to produce the global glacial max lowest sea level coastline.

For the test @jeffreyameyer did over in #332, should we be able to see any change now that you ran OSM coastlines pipeline?

[tsinn]

Here's an example global glacial max lowest sea level coastline that I produced from 2021 GEBCO bathymetric data (GEBCO_2021 Grid (ice surface elevation)). I processed these geotiffs to show a shoreline that is 120 meters below present day sea level. I converted the resulting rasters to smoothed shapefiles to reduce the pixelization inherent in rasters. Lastly, I converted these shapefiles to geojson with a coordinate precision of 5 decimal places. https://drive.google.com/file/d/1TYQtXdAX8yQ6Tdh21NIcKlJFNKNShUJp/view?usp=sharing (96mb)

[svergeylen]

Hi. I am not sure it is the same bug, but I have moved the coastline to match the shore at its status in 1850 and it seems it is not regenerated. I suppose you are working on it.

Current coastline in JOSM (no harbour in 1850) :

Current coastline in OHM (the harbour is still visible) :

Can this bug get a higher priority please ?

[danrademacher]

This is the same issue. As shown in our vector tiles here, https://vtiles.openhistoricalmap.org/#14/46.98192/-2.29262, we are still carrying old coastline data probably from NAtural Earth:

The challenge we face is that OSM's whole coastline pipeline is separate from the rest of the data pipeline and made for infrequent changes. This is a priority for us, though it is also a large and complex task and a bit fraught with potential problems. So we don't have an ETA on this yet.

@geohacker and @jeffreyameyer are there any aspects of the pilot tests Jeff did in https://github.com/OpenHistoricalMap/issues/issues/332 that would provide a short-term solution for @svergeylen and others who want their coastal mapping to have immediate effect in the map?

[svergeylen]

We can maybe set up a manual trigger that regenerate a specific (part of) the coastline... from a given path id or from a geographic rectangle given....

[geohacker]

I've generated a list of latest coastline errors (points and lines) as GeoJSON here. ohm-coastline-nov-14-2023.zip

I think it would be good to go through this and come up with a method to prioritise what we should fix first. And then we can create a MapRoulette task or so.

cc @danrademacher @jeffreyameyer @batpad @Rub21 @1ec5

[geohacker]

@jeffreyameyer @danrademacher Martijn says we can use the current MapRoulette instance for this task. We can point it to the iD URL for OHM. Let me know if when you have had a chance to look at the list of errors and if you think of a prioritisation plan. I can help create the task.

[danrademacher]

I am not super clear on how we go about prioritizing these.

By far the most common issue is overlaps (in line errors) and intersections (in point errors)

Of the 26,354 point errors, fully 26,232 of them are associated with osm_id 0. That's 99.5% of all errors.

And of those, 26,011 (98.7% of all errors) are all along the West Coast of the US:

Is this some bad import way back at the down of OHM? Is there some automated way to fix this or do we really send himans chasing 26,000 errors in what must have been a bad coastline import?

If we could somehow resolve these, then the remaining issues -- like fixing the Caspian Sea -- start to look like fun things we could get folks to help with.

[danrademacher]

@jeffreyameyer is going to look at the 26K points on the West Coast and see what the source might be, and then maybe we can find a bulk way to fix them in JOSM