Add the Equal Earth projection (ESRI:54035) and fix projected sub-areas at the edge of world projections

[corentincarton]

Summary

Adds the Equal Earth projection to Magics, and fixes the handling of areas given in projected coordinates (WMS tiles) that extend beyond the curved outline of world projections.

Release order: this PR must be merged and released first. The companion PRs depend on a Magics release that accepts ESRI:54035 as subpage_map_projection:

• Add Equal Earth (ESRI:54035) to the WMS CRS list magics-python#48 — adds ESRI:54035 to wmscrs()
• feat: support the Equal Earth CRS (ESRI:54035) and fix GetCapabilities bounding boxes skinnywms#74 — advertises the CRS and fixes the GetCapabilities bounding boxes

skinnywms gets the Magics binary through ecmwflibs, so that package needs a rebuild on the new Magics as well.

Commit 1: Add the Equal Earth projection as ESRI:54035

Equal Earth is an equal-area pseudocylindrical world projection (Šavrič, Patterson & Jenny, 2018). PROJ supports it since 5.2 (+proj=eqearth) and Magics already requires PROJ ≥ 6.1, so it is exposed like the other PROJ-based world projections (Robinson, Mollweide, ...).

It is registered under its ESRI code ESRI:54035 (WGS84 World Equal Earth) rather than a Magics-style name, so that WMS servers built on Magics can advertise it as a CRS that clients (QGIS, proj4js, ...) can resolve. Equal Earth has no EPSG code.

The changes mirror the ones made for EPSG:3035 (fbb5c93):

• share/magics/epsg.json: PROJ definition, whole-globe bounds, method geos (the same area handling as Robinson and Mollweide)
• subpage_map_projection option in FortranViewNode.xml, Proj4ESRI54035.xml, Proj4ESRI54035 class
• generated attribute/wrapper files (checked against the output of tools/xml2cc.pl and tools/xml2cc_mv.py)

Usage:

from Magics import macro
macro.plot(macro.output(output_formats=["png"], output_name="equal_earth"),
           macro.mmap(subpage_map_projection="ESRI:54035"),
           macro.mcoast(map_grid="on"))

Commit 2: Fix sub-areas in projected coordinates extending outside the projection domain

When the area is given in projected coordinates (subpage_coordinates_system = "projection", as skinnywms does for every WMS tile), Proj4Projection::projectionSimple() reverted the two corners to get the geographical extent, and later re-initialisations (fill(), revert(), ...) rebuilt the area from those geographical corners.

For world projections with a curved outline, the corners of a rectangular tile covering the edge of the outline have no geographical equivalent. PROJ then:

• reports an error for Robinson → Magics later raised Coordinate outside allowed range;
• returns infinite values for Mollweide → same error;
• silently returns wrapped values for Equal Earth → a wrong, mirrored area was drawn.

Even the plain "whole map" request (bbox = full projected extent) has its corners outside the outline: with the current code a Robinson full-extent request with data renders a blank image.

Now, when a corner lies outside the domain:

• the geographical extent is taken from the vertices of the visible part of the area (its intersection with the outline); when nothing is visible, from the closest point of the outline together with a degenerate envelope, so the tile comes out empty (transparent);
• re-initialisation rebuilds the area from the requested projected corners (new internal setting projected_corners) instead of the geographical corners, which cannot round-trip.

Areas whose corners are inside the domain (EPSG:3857, polar stereographic, tiles inside the outline, ...) follow exactly the previous code path.

Testing

Built from this branch on macOS arm64 against conda-forge dependencies (PROJ 9.4, cairo/pango, netCDF) and compared with Magics 4.16.0 from conda-forge:

Case	4.16.0	this branch
ESRI:54035, default (full) area, coastlines + grid	n/a	correct, drawn aspect 2.05 (Equal Earth: 2.055)
ESRI:54035, full extent in projected coordinates	n/a	correct
ESRI:54035, tiles crossing the outline (NE edge, SW corner)	n/a	correct, transparent outside the outline
ESRI:54035, tile entirely outside the outline	n/a	fully transparent
ESRI:54035, NetCDF field	n/a	correct
robinson / mollweide, tile crossing the outline	Coordinate outside allowed range	correct
robinson, full extent in projected coordinates, with data	blank image	correct
EPSG:3857 tile, EPSG:32661 tile, robinson inside tile and full globe	reference	pixel-identical
robinson full globe with NetCDF contours + grid	reference	1 differing pixel of 500 000

Also tested end to end through skinnywms (with the companion PRs): GetCapabilities advertises ESRI:54035, GetMap returns correct PNGs for the full extent and for edge tiles.

No regression test is added because it needs a reference image on the test-data server; regression/proj4/robinson-frame.json is the template to copy for equal_earth.

Notes / known limitations (pre-existing, not addressed here)

• Contour lines are clipped to the rectangular bounding box, not to the curved outline, so they can spill slightly outside the map near the poles. This is identical for Robinson and Mollweide today, and is more visible on WMS tiles because of subpage_gutter_percentage.
• Only the ESRI:54035 name is added. A friendlier alias (equal_earth) would be one more option line + class if wanted.

🤖 Generated with Claude Code

[iainrussell]

This is excellent! In fact, I had already started this work today, but I called it "EPSG:8857". Is there a good reason for using one name over the other, do you think @corentincarton?

[pmaciel]

The portable, reference name is EPSG:8857. ESRI is vendor-specific

[iainrussell]

Thanks Pedro, I agree, and Magics already uses EPSG codes for other projections, so if we can change the name then we'll be good.

[iainrussell]

Responding to

"Contour lines are clipped to the rectangular bounding box, not to the curved outline, so they can spill slightly outside the map near the poles. This is identical for Robinson and Mollweide today, and is more visible on WMS tiles because of subpage_gutter_percentage."

This is not the case for Robinson and Mollweide today, it seems to be unique to this Equal Earth projection. The case I refer to is where we plot the whole globe on a map, with no tiling.

[iainrussell]

@corentincarton, I've just merged my own #121 for this, including a bug fix for data that was plotted outside the Earth, so I wonder if your bug fix is still necessary?

[iainrussell]

I've merged my version with yours, including your fix for the tiling, all now in the develop branch. The projection is called "EPSG:8857".