Projection and drawing issues on large rectangular boxes

[rysebaert]

When uploading a rectangular frame in Magrit with EPSG:3035 (ETRS89-extended / LAEA Europe), it seems that there is a mistake with the reprojection (cf below screenshot with associated layer, attached). Any idea ?

This layer in R or QGIS

Same in Magrit

Associated geojson file frame.zip

[rCarto]

I suspect that the frame has only four points. I think that adding intermediate points would solve the problem.

[neocarto]

It comes from the way d3.js handles projections. As Tim says, I think this can be fixed outside Magrit by densifying the polygon.

[mthh]

I confirm the problem is not in the reprojection per se (I guess the 4 points that describe the rectangle are reprojected the same / with the same coordinates in Magrit and QGIS) but in the way the lines between the points in question are drawn.

In Magrit (due to the underlying use of d3), lines between two points are not straight line on the plane but segment of great circles (geodesics).

Look at the example below which illustrates this (and for which I think the line drawing is correct).

Line between Spain and California in various projections

  

However I am not sure that the curves we get on your screenshot are correct and there must be something to do in Magrit as well ? (don't you think @neocarto ?)

In the meantime, the workaround on your side is to densify the geometries as advised by @rcarto and @neocarto.

[neocarto]

Well, if @rysebaert share the geometry of the box, I can make some tests.

[mthh]

It's in the "frame.zip" file attached to the first post. Beware that it's not a rfc7946-compliant geojson (as it contains geometries in EPSG:3035).

[neocarto]

I can't reproduce Ronan's image. When I import the geojson into Magrit and choose the EPSG:3035 projection, I get this result (see pj). However, I think that maybe we should add "none" or "already projected" in the projection menu to avoid these distortions. This is what I do in bertin.js

[mthh]

Yeah I saw that the frame.geojson was triggering some other issues in Magrit :-/

I'm not sure where these artifacts came from. It could be because there is no clipping when you ask for the epsg:3035 projection and because the coordinates of @rysebaert frame are outside the extent of the EPSG:3035 projection.

Frame GeoJSON :

minX = 2380000.0 maxX = 6550000.0 minY = 1350000.0 maxY = 5420000.0

EPSG:3035 extent :

minX_proj = 1896628.62 maxX_proj = 4629821.48 minY_proj = 1507846.05 maxY_proj = 6882401.15

>>> minX < minX_proj
False
>>> minY < minY_proj
True
>>> maxX > maxX_proj
True
>>> maxY > maxY_proj
False

Having a "none" or "already projected" option would be a major change in Magrit (which we have already discarded in the past - but I'm not that stubborn about it!) So, I wouldn't mind implementing this in a future version, but it would involve changes in many places (because until now all datasets had to be associated with a CRS, all datasets were converted to GeoJSON with WGS84 coordinates, etc. so many calculations like those needed for some representations are based on these assumptions).

[rysebaert]

Thks for the feedbacks and explanations guys ! Yes @neocarto , my Magrit's image was a zoom in. I have in fact the same than yours... And @mthh, agree with you, the geojson should be in WGS84 to be compliant...

Nevermind, my European template looks good without this frame...

[mthh]

Looks great !

[mthh]

Anyway, I'm a bit worried about the rendering of rectangular frames like the following :

{
"type": "FeatureCollection",
"name": "frame",
"crs": { "type": "name", "properties": { "name": "urn:ogc:def:crs:EPSG::3035" } },
"features": [
{ "type": "Feature", "properties": { }, "geometry": { "type": "Polygon", "coordinates": [ [ [ 2380000.0, 1500000.05 ], [ 4600000.48, 1500000.05 ], [ 4600000.48, 5420000.0 ], [ 2380000.0, 5420000.0 ], [ 2380000.0, 1500000.05 ] ] ] } }
]
}

(derived from the original frame of @rysebaert in order to have all coordinates within EPSG:3035 bounds, still not rfc7946-compliant but Magrit still supports geojson-2008)

Using "EPSG:3035" string in "Projection from WKT/Proj4/EPSG string..." option of Magrit (so it uses proj.4 under the hood) I have this weird extra-line :

Using the "More option..." value in the projection menu and selecting any of the projection (so it uses d3-geo-projection under the hood) I obtain the inverse of the desired polygon :

I'm letting this issue open, tagging it as a bug and I will have a closer look on Magrit's projections in the next few weeks.

[neocarto]

This is not specific to Magrit. I don't know how to do an automatic rewind of the geometries to avoid this. https://www.npmjs.com/package/@turf/rewind

[mthh]

The coordinates of the geojson in my last post are already counterclockwise.

[mthh]

Fixed by 326e3c8.