Pixel size for sampling tolerance in WGS84

[ChristianBeilschmidt]

I'm using T-Rex to generate WGS84 vector tiles for Natural Earth Data. For this, I import the geopackage into PostGIS and then use this as a datasource for T-Rex.

One layer is specified like this:

[[tileset.layer]]
name = "boundary_lines_land"
geometry_field = "geom"
geometry_type = "MULTILINESTRING"
fid_field = "fid"
srid = 4326
buffer_size = 0
simplify = true
make_valid = true

[[tileset.layer.query]]
minzoom = 0
maxzoom = 2
sql = """SELECT geom, fid, featurecla, name, min_zoom FROM ne_110m_admin_0_boundary_lines_land WHERE min_zoom::integer <= !zoom! AND geom && !bbox!"""

[[tileset.layer.query]]
minzoom = 3
maxzoom = 4
sql = """SELECT geom, fid, featurecla, name, min_zoom FROM ne_50m_admin_0_boundary_lines_land WHERE min_zoom::integer <= !zoom! AND geom && !bbox!"""

[[tileset.layer.query]]
minzoom = 5
maxzoom = 10
sql = """SELECT geom, fid, featurecla, name, min_zoom FROM ne_10m_admin_0_boundary_lines_land WHERE min_zoom::integer <= !zoom! AND geom && !bbox!"""

It seems to work in general 🎉 but unfortunately every line gets simplified to having only ~3 nodes.

Then I found the tolerance parameter, which is per default !pixel_width! / 2.
However, there was no difference when I changed the 2 to 4, 16 or 100 – which appeared very strange.

I look into the codebase about the database queries and the string replacement seems to be working just fine, so this left me to question what pixel_width is.
In more detail, I found the following code:

t-rex/tile-grid/src/grid.rs

Lines 185 to 193 in f2433f6

	/// Pixel width for 256x256 tile
	pub fn pixel_width(&self, zoom: u8) -> f64 {
	const METERS_PER_DEGREE: f64 = 6378137.0 * 2.0 * consts::PI / 360.0;
	match self.units {
	Unit::Meters => self.resolutions[zoom as usize],
	Unit::Degrees => self.resolutions[zoom as usize] * METERS_PER_DEGREE,
	Unit::Feet => self.resolutions[zoom as usize] * 0.3048,
	}
	}

So pixel width has something to do with 256-pixel tile width in meters for Web Mercator.

Then, I tried to divide the pixel width with the fixed value that was multiplied to our degree values.
This leads to the following setting:

tolerance = "!pixel_width! / 3507975.35 / 2"

And this seems to solve the simplification problem to reasonable country borders.

So I guess either there should either be different defaults for different projections for tolerance or the pixel size for degrees is debatable.

[pka]

Thanks for your investigations. Seems clear that the current tolerance default doesn't make sense for geographic coordinates. Thanks for documenting a tolerance configuration,which works better!