longitude shift is wrong when going from TripolarGrid -> ClimaCore cubed sphere

[juliasloan25]

I ran an example from test/usecases/climacore.jl and plotted the remapped fields to verify they look ok visually in addition to the conservation tests we have. I noticed that the longitude values are correctly shifted when we go from ClimaCore -> Oceananigans, but when we remap back they aren't shifted.

ClimaCore has longitude on [-180, 180], while Oceananigans Tripolar is [0, 360].

Note also that the test is missing one last step we should also be testing, which is to put the vector of remapped values onto the ClimaCore space using set_value_per_element!. I'll start a PR to add that.

To reproduce

Example

using ConservativeRegridding
using ConservativeRegridding: Trees
using Statistics
using Test
import GeometryOps as GO, GeoInterface as GI, LibGEOS

using ClimaCore:
    CommonSpaces, Fields, Spaces, RecursiveApply, Meshes, Quadratures, Topologies, Domains, ClimaComms
using Oceananigans

using CairoMakie
using ClimaCoreMakie

const ClimaCoreExt = Base.get_extension(ConservativeRegridding, :ConservativeRegriddingClimaCoreExt)

latlon_grid = LatitudeLongitudeGrid(size=(360, 180, 1), longitude=(0, 360), latitude=(-90, 90), z = (0, 1), radius = GO.Spherical().radius)

    cubedsphere_space = CommonSpaces.CubedSphereSpace(;
        radius = GO.Spherical().radius,
        n_quad_points = 2,
        h_elem = 64,
    )

    # Define a field on the first space, to use as our source field
    field = Fields.coordinate_field(cubedsphere_space).long

    fig = Figure();
    ax = Axis(fig[1, 1])
    hm = fieldheatmap!(ax, field)
    Colorbar(fig[1, 2], hm)
    save("field_src.png", fig)

    ones_field = Fields.ones(cubedsphere_space)
    cubed_sphere_vals = zeros(6*cubedsphere_space.grid.topology.mesh.ne^2)
    ClimaCoreExt.get_value_per_element!(cubed_sphere_vals, field, ones_field)

    latlon_field = Oceananigans.CenterField(latlon_grid)
    latlon_vals = vec(interior(latlon_field))
    set!(latlon_field, (x, y, z) -> 0)

    regridder = ConservativeRegridding.Regridder(latlon_grid, cubedsphere_space; threaded = false)

    ConservativeRegridding.regrid!(latlon_vals, regridder, cubed_sphere_vals)
    fig = Figure();
    ax = Axis(fig[1, 1])
    hm = heatmap!(ax, interior(latlon_field, :, :, 1))
    Colorbar(fig[1, 2], hm)
    save("field_oc.png", fig)

    # Test that the integral is conserved
    @test isapprox(
        sum(abs, latlon_vals .* ConservativeRegridding.areas(GO.Spherical(), Trees.treeify(latlon_grid))),
        sum(abs, cubed_sphere_vals .* ConservativeRegridding.areas(GO.Spherical(), Trees.treeify(cubedsphere_space))),
        rtol = 1e-13
    )

    # Test the other way
    set!(latlon_field, (x, y, z) -> x)

    ConservativeRegridding.regrid!(cubed_sphere_vals, transpose(regridder), latlon_vals)

    @test isapprox(
        sum(abs, cubed_sphere_vals .* ConservativeRegridding.areas(GO.Spherical(), Trees.treeify(cubedsphere_space))),
        sum(abs, latlon_vals .* ConservativeRegridding.areas(GO.Spherical(), Trees.treeify(latlon_grid))),
        rtol = 1e-13
    )

    # TODO lat/lon shift is incorrect in this direction
    # Convert the vector of remapped values to a ClimaCore Field with one value per element
    field_remapped = Fields.zeros(cubedsphere_space)
    ClimaCoreExt.set_value_per_element!(field_remapped, cubed_sphere_vals)
    fig = Figure();
    ax = Axis(fig[1, 1])
    hm = fieldheatmap!(ax, field_remapped)
    Colorbar(fig[1, 2], hm)
    save("field_remapped.png", fig)

Plots

Source field: ClimaCore CubedSphere

Dest field: Oceananigans TripolarGrid

Dest field: back to ClimaCore CubedSphere

[asinghvi17]

So it turned out the issue was that the ClimaCore long goes from -180..180 but the Oceananigans grid in this example went from 0..360....so the grids being compared were actually different.

Goes to show we probably need a quadrature method for arbitrary functions on spherical polygons - that would have avoided this issue

[juliasloan25]

Put another way - we can regrid between ClimaCore and Oceananigans even though they use different longitude ranges. This example is just misleading because the line set!(latlon_field, (x, y, z) -> x) overwrites the remapped field with the Oceananigans longitude (instead of the original ClimaCore longitude). Then when we remap back to ClimaCore and look at it, we're actually comparing a different field from the previous plots.