Spikes on tile edges

[kleunen]

I tried the latest head version now, but I am seeing this spikes now, at lower zoom levels:

[systemed]

I've seen this very occasionally - I think it might be an artefact of simplification. In theory we have a clippingBox with a small margin beyond the tile edge to prevent problems like this - see https://github.com/systemed/tilemaker/blob/master/src/coordinates.cpp#L80-L85 .

So it could be a result of simplifying the source coastline polygons, which are themselves tiled.

(One possibility to fix this would be to have an option in the simplification algorithm that doesn't simplify pure east-west or north-south lines - i.e. where lat1==lat2 or lon1==lon2.)

[kleunen]

Yes, I think it is related to the simplification. When you zoom in/out, you can see it appear and disappear

[kleunen]

So there is a straight line in the source data? which gets simplified ?
But the distance between these two points should be really large right ?

well, you can look at the angle within the node to be simplified. If it is close to 180 degrees, it can be simplified. A 90 degrees angle, should not be simplified.

But how is this possible ? because the simplification algorithm looks if the line does not move too much from the original line, when the node is removed ?

[kleunen]

Hmm, i think it may be this condition:
https://github.com/systemed/tilemaker/blob/master/src/geom.cpp#L48

[systemed]

Hmmm, interesting. I thought it would be the source data but in this case it isn't. If you look at the edge of the source polygons it's nowhere near this island:

So it's to something to do with simplifying the geometries once we've clipped them to tile bounds.

At present we hardcode the tile margin to 1/200th of the tile width, but maybe we should make this configurable per-layer. https://blog.cyclemap.link/2020-01-25-tilebuffer/ has a good explanation of buffers/margins.

[kleunen]

Maybe the simplification is not deterministic. So you have the island, clipped into two polygons. On one side different nodes are removed in the simplification process as the part of the island on the other side. The render might not find the same nodes on the different sides of the island ?

[kleunen]

I tried changing:

 if(boost::geometry::distance(input[start], input[end]) < 2 * max_distance || distance < max_distance) {

to

 if(distance < max_distance) {

but it makes no difference. I think the simplification should happen before the clipping to the tile edges

[systemed]

I was wrong - it is the source data after all:

(I was looking at the EPSG:3857 shapefile previously, but we use the EPSG:4326 one.)

[kleunen]

Ah ok. Maybe just not simplify nodes which are on the outer edge of the polygon ? So on min lon, max lon, min lat, max lat.

[systemed]

Yes, that would work!

[kleunen]

I will give this a try

[kleunen]

hmm, i tried, but did not give an improvement.

https://github.com/kleunen/tilemaker/blob/master/src/geom.cpp#L31-L33

I used this in simplification:

  boost::geometry::model::box<typename GeometryType::value_type> box;
    boost::geometry::envelope(input, box);

    std::priority_queue<std::size_t, std::vector<size_t>> pq;
    for(std::size_t i = 0; i < input.size() - 2; ++i) {
        if( input[i].x() != box.min_corner().x() && input[i].x() != box.max_corner().x() &&
            input[i].y() != box.min_corner().y() && input[i].y() != box.max_corner().y()) {
            pq.push(i);
        }
    }

[dieterdreist]

sent from a phone

On 3 Jul 2021, at 10:15, Richard Fairhurst ***@***.***> wrote: So it could be a result of simplifying the source coastline polygons, which are themselves tiled.

could the untiled version of coastline polygons be used instead? I have solved similar issues in rendering by switching from tiled to untiled coastlines.

[kleunen]

The water polygons:
https://osmdata.openstreetmap.de/data/water-polygons.html

Do say:
Where polygons have been split they overlap slightly to help avoiding rendering artefacts at the seams.

So maybe, the buffer in tilemaker should be slightly larger, such that they do not overlap with the buffer of the water polygons ?