Projections of the earth in the examples on the front page

[jaakkor2]

Looking at the examples on https://github.com/JuliaPlots/Makie.jl ,

First earth example from https://gist.github.com/SimonDanisch/8f5489cffaf6b89c9a3712ba3eb12a84 simplified and with lighting adjusted (current documentation does not say too much about the light attribute :) )

using Makie, FileIO, Colors
earth = load(download("https://svs.gsfc.nasa.gov/vis/a000000/a002900/a002915/bluemarble-2048.png"));
scene = Scene(resolution = (500, 500), color = :black)
m = GLNormalUVMesh(Sphere(Point3f0(0), 1f0), 60)
light = Vec{3,Float32}[[1.0, 1.0, 1.0], [0.1, 0.1, 0.1], [0.9, 0.9, 0.9], [0.0, -20.0, 0.0]]
Makie.mesh(m, color = earth, light = light)

Antarctica looks big

Antarctica should look like this (takes from Wikipedia) https://upload.wikimedia.org/wikipedia/commons/thumb/f/f2/Antarctica_%28orthographic_projection%29.svg/250px-Antarctica_%28orthographic_projection%29.svg.png

This example is a nice one https://gist.github.com/SimonDanisch/58cb1050581658859f7a5df51fa1e5a1#file-earth-jl but it has two problems 1) Projection of the tempeture map has the same projection error as above. 2) The precipitation is not matching the temperature map, but is 90 deg rotated and scaling is distorted

These are low priority, but since these are on the front page, I thought to file an issue.

Also, gists as far as I know do not accept PRs to update them to Julia v1.0.

[mkborregaard]

I think this is because of https://github.com/JuliaPlots/Makie.jl/issues/205 . So the precipitation bars might actually be correct, but the heatmap is wrapped the wrong way around the globe. EDIT: No this idea was incorrect.

[mkborregaard]

No, now I agree it's unrelated. But the color does seem to wrap around weirdly. Check the code here (temperature[10] is a matrix)

m = GLNormalUVMesh(Sphere(Point3f0(0), 1f0), 200)
p = decompose(Point3f0, m)
scene = Makie.mesh(m, color = temperature[10], shading = true)
scatter!(scene, p, markersize = 0.1)

The decomposition shows that the sphere's vertices align with centers of a lat/long grid (conveniently). But the heatmap seems to have wrapped around "from the side", so that North on the heatmap is now facing to the right, in a different orientation in that revealed by the points.

[AshtonSBradley]

trying another equirectangular projection test:

using Makie, FileIO, Colors, GeometryTypes
scene = Scene(resolution = (500, 500), color = :black)
m = GLNormalUVMesh(Sphere(Point3f0(0), 1f0), 60)
moon = load(download("https://svs.gsfc.nasa.gov/vis/a000000/a004600/a004675/phases.0001_print.jpg"))
mesh(m, color = moon)

there is quite a lot of polar distortion in the image. Craters are elliptical:

(not sure if I am using this correctly though!)

[AshtonSBradley]

ok, needed correct coordinate spacing. I if I adapt the image on a sphere example, everything works:

using Makie, FileIO, Colors, GeometryTypes

n = 512
θ = LinRange(0,pi,n)
φ = LinRange(0,2*pi,2*n)
x = [cos(φ)*sin(θ) for θ in θ, φ in φ]
y = [sin(φ)*sin(θ) for θ in θ, φ in φ]
z = [cos(θ) for θ in θ, φ in φ]
pts = vec(Point3f0.(x, y, z))
moon = load(download("https://svs.gsfc.nasa.gov/vis/a000000/a004600/a004675/phases.0001_print.jpg"));
surface(x, y, z, color = moon)

[AshtonSBradley]

@SimonDanisch the distortion seems to be traceable to the way the sphere points are sampled by HyperSphere. Does this function generate an equirectangular coordinate grid for the 3-sphere? Should it be doing something like the image on a surface example adapted above?

[mkborregaard]

So the error is in the code wrapping the color on a spherical mesh. Where is that function defined in Makie?

[SimonDanisch]

It's basically here: https://github.com/JuliaGeometry/GeometryTypes.jl/blob/master/src/decompose.jl#L367 And here: https://github.com/JuliaGeometry/GeometryTypes.jl/blob/master/src/decompose.jl#L381 (uv coordinates are in 0..1 mapping to the coordinates of the image)

[mkborregaard]

But decompose works, right? Look at my image above - the points are correctly placed but the underlying heatmap isn't (or is it the other way around)? The "North pole" convergence of the decompose points are at the points that correspond to longitude-latitude (0, 0), the South Pole at (180, 0) on the heatmap. But the correspondence should be at (_, 90) and (_, -90) (where _ corresponds to any value as the longitudes collapse in the poles).

[SimonDanisch]

The UV map seems pretty wrong (decompose(UV, Sphere(...)))

[SimonDanisch]

This works:

n = 512
θ = LinRange(0,pi,n)
φ = LinRange(0,2pi,n)
ux = LinRange(0, 1, n)
x = [cos(φ)*sin(θ) for θ in θ, φ in φ]
y = [sin(φ)*sin(θ) for θ in θ, φ in φ]
z = [cos(θ) for θ in θ, φ in φ]
pts = vec(Point3f0.(x, y, z))
uv = [UV{Float32}(φ, θ) for θ in ux, φ in ux]
f = decompose(Face{3, GLIndex}, SimpleRectangle(0, 0, 1, 1), (n, n))
m = GLNormalUVMesh(vertices = pts, faces = f, texturecoordinates = vec(uv))
scene = mesh(m, color = moon, shading = false)

Maybe that's how I should define the decompose for sphere

[SimonDanisch]

Although that's still inconsistent with surface:

[SimonDanisch]

Ah reverse ux for x fixes that. I put some time into making sure that surface works like the others, so I guess we should just follow surface here ;)

[SimonDanisch]

Thanks everyone, I think this is fixed now with https://github.com/JuliaGeometry/GeometryTypes.jl/pull/158

using Makie, FileIO, Colors
earth = load(download("https://svs.gsfc.nasa.gov/vis/a000000/a002900/a002915/bluemarble-2048.png"));
scene = Scene(resolution = (500, 500), color = :black)
m = GLNormalUVMesh(Sphere(Point3f0(0), 1f0), 60)
light = Vec{3,Float32}[[1.0, 1.0, 1.0], [0.1, 0.1, 0.1], [0.9, 0.9, 0.9], [0.0, -20.0, 0.0]]
Makie.mesh(m, color = earth, shading = false)

[mkborregaard]

Awesome, you're a wizard!

[mkborregaard]

Note that this does not completely resolve the issue with the points decomposition: This is with GeometryTypes/sd-sphere checked out:

Note the polar position of the points wrongly at (0,0) and the clearly misplaced rainy belt over Sahara. The bar heights could be something in the meshscatter function but at least the point decomposition is not consistent with that of a globe.

[mkborregaard]

For info, the world-clim precipitation layer is only defined over land, so the correct projection would put all the flat bars over the ocean.

[SimonDanisch]

Yeah the uv look up code needs to be adapted for sampling the precipitation:

[mkborregaard]

Do you mean in the example or in GeometryTypes? That looks amazing btw.

[SimonDanisch]

The example needs an update ;)

[mkborregaard]

OK, so you just fixed the to_msize function? I wouldn't mind adding this unless you've essentially already done it for your plot above.

[SimonDanisch]

Yeah it's fixed ;) https://github.com/JuliaPlots/Makie.jl/blob/master/examples/bigdata.jl#L2 needed some julia 1.0 updates as well!

[mkborregaard]

Oh great. Yeah I'd fixed the other things locally because I wanted to update this example, that was how I fell over this to begin with.

[mkborregaard]

So, this can be closed I guess?