Cubed-Sphere Grids (GEOS / GCHP)

In addition to regular latitude/longitude grids (RectangularGridSpec), SatelliteGridding can grid Level-2 footprints directly onto a cubed-sphere output grid in the GEOS / GCHP convention via CubedSphereGridSpec. This is useful for comparing satellite observations against GEOS-Chem (GCHP) output on its native grid without an intermediate regridding step.

Convention

The default geometry is GMAOCubedSphereDefinition — the grid used by GEOS-IT / GEOS-FP and by GCHP when driven by GMAO meteorology:

• equal-distance gnomonic edge coordinate law (grid_type = 0 in GEOS/FV),
• four-corner-normalized cell centers (GEOS cell_center2),
• GEOS-native panel ordering (panels 1–2 equatorial, 3 north pole, 4–5 equatorial, 6 south pole), and
• a final −10° longitude shift.

An EquiangularCubedSphereDefinition (classic tan-spaced equiangular gnomonic, no offset) is also available for synthetic targets.

A grid is named C<Nc>, where Nc is the number of cells per panel edge; the cube has 6·Nc² cells total. For example, Nc = 360 is C360 (~25 km).

Library usage

using SatelliteGridding, Dates

config    = load_config("examples/tropomi_sif.toml")
grid_spec = CubedSphereGridSpec(; Nc = 360)             # C360, GMAO/GCHP default
time_spec = TimeSpec(DateTime("2020-07-01"), DateTime("2020-07-16"), Day(16))

grid(config, grid_spec, time_spec, SubpixelGridding();
     outfile = "tropomi_sif_c360.nc", keep_going = true)

The data-source config (tropomi_sif.toml) is grid-agnostic — only the grid_spec changes between a regular and a cubed-sphere run. See examples/tropomi_sif_cubedsphere.jl for a runnable C24 demo.

CLI usage

julia --project=. bin/grid.jl l2 \
    --config examples/tropomi_sif.toml \
    --gridType cs --Nc 360 \
    --startDate 2020-07-01 --stopDate 2020-07-16 --dDays 16 \
    -o tropomi_sif_c360.nc

--csConvention selects gmao (default) or equiangular. For long global runs, bin/run_tropomi_c360.sh fans the work out across time chunks (one Julia process per chunk) and concatenates with ncrcat:

NC=360 JOBS=8 bash bin/run_tropomi_c360.sh

How footprints are gridded

Because the lon/lat → cube map is nonlinear and discontinuous across panel edges, footprints are subdivided in lon/lat space (not in fractional-index space as for rectangular grids). Each footprint corner quadrilateral is split into n × n sub-points, each projected independently onto its (panel, cell) via lonlat_to_panel_xy, contributing weight 1/n². A footprint straddling a cube edge therefore correctly deposits weight onto both panels. Corner longitudes are unwrapped to a common branch first so the subdivision is sane for near-pole footprints that cross the ±180° seam.

The default oversampling factor on the cube is a small fixed value (n = 4); override it with --nOversample / SubpixelGridding(n_oversample=...).

Output layout

The NetCDF output matches native GEOS / GCHP files, so it opens directly in Panoply, xarray, and GCHP tooling:

In xarray you can plot a face directly on its 2D coordinates:

import xarray as xr
ds = xr.open_dataset("tropomi_sif_c360.nc")
ds["sif_743"].isel(time=0, nf=2).plot(x="lons", y="lats")   # one panel