Supporting all Cartopy Projections

[andrewdnolan]

Below I'll list all cartopy projections and my notes about what type of wrapping/manipulation will be needed to support spherical plotting on the given projection.

[andrewdnolan]

PlateCarree ✅

• 1D wrapping and masking of patch w/ center at projection boundary

AlbersEqualArea ❌

• 1D wrapping in both x/y direction helps, but still something weird with the boundary patches...

AzimuthalEquidistant ❌

• same issue as AlbersEqualArea

EquidistantConic ❌

• don't need to 1d wrap, but something strange is going on for cells that appear on both sides of the top anti-meridian.

LambertConformal ❌

• same issue as above, but does calculating the projection period as a function of latitude (and lon) work ?

LambertCylindrical ✅

• 1D wrapping and masking of patch w/ center near projection boundary

Mercator ✅

• 1D wrapping, don't need to mask poles but one cell is periodic is y??

Miller ✅

• 1D wrapping, and masking of patch w/ center at projection boundary

Mollweide 🤷

• 1D wrapping (kinda), no need to mask poles, but b/c of narrowing of projection of period near poles the diff / 2 * np.sqrt(2) threshold is not sufficiently small

ObliqueMercator ❌

-1D wrapping is not sufficent, useful to plot in projected coordinates but not a geoaxes

Orthographic ❌

• projection results in lots of nan coordinates, which cartopy does not handle....

Robinson ✅

• 1D wrapping, don't need to mask poles

Sinusoidal 🤷

• 1D wrapping kind of works, but same problem as Mollweide

Stereographic ✅

• No wrapping or masking of poles needed....

TransverseMercator ❌

• 1D wrapping does not work, b/c period calculation is no good. Since cell centers extend well past projection boundaries that are used to determine the period

UTM ❓

• .......

InterruptedGoodeHomolosine ✅

• 1D wrapping and masking of cell at the y proj limit

RotatedPole ✅

• 1D wrapping works, don't need to mask poles

OSGB, LambertZoneII, EuroPP ❌

• NO

Geostationary ❌

• Same problem as Orthographic

NearsidePerspective ❌

• Same as Geostationary and Orthographic

EckertI -- EckertIVI ✅

• 1D wrapping works, don't need to mask poles

Aitoff 🤷

• same issue as Sinusoidal and Mollweide

EqualEarth ✅

• 1D wrapping works, don't need to mask poles

Gnomonic ❌

• IDK strange projection....

Hammer 🤷

• same issue as Sinusoidal, Mollweide, Aitoff

LambertAzimuthalEqualArea ❌

• same issue as AlbersEqualArea and AzimuthalEquidistant, should I be using the hypotenuse of the x/y period??

NorthPolarStereo ✅

• 1D wrapping works, don't need to mask poles

SouthPolarStereo ✅

• 1D wrapping works, don't need to mask poles

[xylar]

The only unsupported projection in this list that would be important to me is LambertConformal. When we use this in MPAS-Analysis, we always zoom in on a defined range in x-y space. So as long as that is possible, we should be good.

We also currently remap to a comparison grid, so this remains as a plausible alternative solution for analysis plots. We need to compare with observation in any case, so one or both datasets have to be remapped.