GitHub

TM5 grid and advection references

This document points to TM5 sources and how this model aligns with TM5 conventions for lat/lon grids, cell centers, and pole/CFL handling.

TM5 repositories and docs

  • TM5 on SourceForge (Cycle 3 4DVAR) https://sourceforge.net/p/tm5/cy3_4dvar/ci/default/tree/ Contains: base/, analysis/, input/, proj/, rc/, user/, vpp/. Core transport is in base/.

  • TM5 physics (advection, grid) https://tm5.sourceforge.net/about-tm5/physics/ Advection: slopes scheme (Russell and Lerner, 1987); convection: Tiedtke (1989). Operator splitting for advection, convection, sources, chemistry.

  • KNMI TR-294: "Towards advection on a full reduced grid for TM5" P.B. Hooghiemstra, 2006. https://www.knmi.nl/research/publications/towards-advection-on-a-full-reduced-grid-for-tm5 PDF: KNMI TR-294. Describes:

    • Regular grid: same number of zonal cells on every latitude -> cells shrink near poles -> CFL violation risk unless time step is reduced.
    • Reduced grid: poleward cells are combined (e.g. one cell at the pole) so cell sizes are more uniform; ECMWF data is on a fully reduced grid.
    • Cell properties: cell center coordinates (xi, yi), cell length/width (dxi, dyi) per cell; linear subgrid tracer distribution; symmetric operator splitting X(dt/2), Y(dt/2), Y(dt/2), X(dt/2).

  • UU TM5 overview https://www.projects.science.uu.nl/tm5/TM5_overview.html Resolutions: global 6x4 deg, intermediate 3x2 deg, zoom 1x1 deg (and 0.5x0.25 deg in high-res); vertical layers from ECMWF.

Alignment in this model

  • Vertical coordinate: All met sources use hybrid sigma-pressure coordinates (p = A + B * p_surface). A/B coefficients are stored in TOML files (config/era5_L137_coefficients.toml, config/geos_L72_coefficients.toml). The HybridSigmaPressure type provides a universal interface regardless of whether the met data originated from spectral harmonics (ECMWF) or gridded output (GEOS, MERRA).

  • Grid: Regular latitude-longitude grid. All horizontal coordinates and spacings use cell-center convention: tracers and diagnostics at cell centers; Dx = Rcos(lat)Dlon, Dy uniform.

  • Poles / CFL: TM5-style reduced grid for x-advection (CPU) or CFL-adaptive subcycling (GPU and fallback). Polar cells are clustered to avoid CFL violations from shrinking zonal spacing. Subcycling also handles y and z CFL adaptively.

  • Source/observation placement: Point sources and lon/lat -> grid indexing use cell-center coordinates via find_nearest_ij(grid, lon_deg, lat_deg).

  • NetCDF output: Written lon/lat are the grid cell-center arrays.

Differences from TM5

  • Reduced grid: TM5 runs on a fully reduced grid; we support TM5-style reduced grids on CPU and fall back to subcycling on GPU.
  • Advection scheme: Same family – Russell & Lerner (1981) slopes scheme (see src/Advection/slopes_advection.jl). GPU kernels via KernelAbstractions.jl for x, y, and z directions. Optional minmod flux limiter; Strang splitting. TM5 also offers the second-order moments (Prather) scheme; we have slopes only.
  • Mass fluxes: TM5 computes mass-conserving fluxes via spectral integration. We use gridpoint winds directly. Comparison and validation are ongoing.
  • Met data: TM5 requires ECMWF spectral GRIB data. Our model reads standard NetCDF from ERA5, GEOS-FP, or MERRA-2 via TOML-configured readers. Adding a new met source requires only a TOML config file and (if different) a coefficient TOML – no Julia code changes.