Skip to content

Mars GCM configuration

Mars supplies body constants, forcing and boundary data to the generic JAX GCM. The primary spatial workflow lives in src.celestials.planets.mars.maps; it does not advance the deprecated torch Mars.compute_derivatives state.

From planet to spatial state

  1. Build spectral/sigma coordinates and body-based units from MARS_BODY_3D.
  2. Load checksum-verified MOLA terrain and transform it to the modal lower boundary.
  3. Initialize a terrain-balanced atmosphere or load a compatible full-state restart.
  4. Attach radiation, surface properties, dust and CO₂ forcing.
  5. Integrate coupled dynamics and surface reservoirs, optionally with diffusion.
  6. Export dimensional fields, sampling metadata, diagnostics and restart state.

The initial hydrostatic pressure uses Mars gravity, gas constant and temperature: p_s(z) = p0 * exp(-g*z/(R*T)). Winds and circulation subsequently evolve under the primitive equations; terrain is not just a plotting overlay.

Physics configuration

Process Selection and behavior
Solar geometry Keplerian season/orbit; diurnal or daily-mean illumination; solar slant path
Radiation Grey, explicit compact multiband, or bundled Ames correlated-k CO₂
Surface exchange Conservative sensible heat and momentum drag; optional stability dependence
Regolith Optional multilayer conduction and TES thermal-inertia/albedo fields
PBL/convection Conservative diffusion and smooth dry convective relaxation
CO₂ Pressure-dependent frost point, surface latent heat and atmosphere/frost exchange
Dust Prescribed opacity, optionally separate visible/IR seasonal climatology
Diffusion Optional spectral damping selected by physical e-folding timescale

The factory's generic defaults are deliberately smaller than the browser baseline. run_maps() with no forcing is dry dynamics. tform mars maps enables grey daily-mean radiation and CO₂ by default. The server enables correlated-k radiation and advanced surface physics, longwave scales 0.25/0.25, surface exchange 1.0, and a 0.1-sol diffusion timescale, using staged TES/Ames data when available. See the configuration quickstart.

Outputs and comparison

Forced temperature maps are surface skin temperature; dry maps use lowest-layer air temperature. Winds represent the lowest sigma layer rather than fixed 10 m. Frost is pressure-equivalent (Pa); divide by gravity for kg/m². Full-state comparison exports also provide atmospheric profiles and forcing metadata. Match season, averaging, height and dust before interpreting reference differences.

Calibration, neural components and limits

Physical calibration fits separate CO₂/dust longwave and surface-exchange multipliers through coupled trajectories. Neural radiation replaces directional fluxes with constrained boundary conditions; neural heating is bounded but can add net energy. Cached temperature-only fitting is a separate downstream workflow.

The model has no water/cloud cycle, interactive dust lifting or photochemistry. Composition-dependent high-pressure terraforming physics is not validated. Recorded multi-year atmospheric/surface repeatability passed selected thresholds, but deep-soil equilibrium did not. See validation.

Deprecated model and intervention semantics

The torch global-mean derivation is retained for existing users and is deprecated. Current multi-year browser intervention trajectories still use it, with independent GCM diagnostic spin-ups at selected years. They are not continuous 3-D intervention forecasts. Use coupled GCM experiment workflows for new spatial-model research and disclose this limitation for old runs.