Run your first Mars GCM¶
Complete installation, including the GCM extra and MOLA staging. Run commands from the repository root using the installed environment.
1. Export a first map¶
rtk proxy .venv/bin/python -m cli.main mars maps \
--scale fast --truncation T21 --layers 12 --dt 300 --steps 100 \
--name first-gcm-map
This short T21/L12 run uses daily-mean grey radiation and energy-limited CO₂
exchange. Outputs appear under outputs/gcm3d_maps/first-gcm-map/ as PNGs and
NetCDF. It verifies the workflow; it is not a climatology or the full browser
physical baseline. fast is a resolution preset, distinct from the deprecated
global-mean model's Accuracy.FAST integrator.
2. Understand the fields¶
| Output | Units | Interpretation |
|---|---|---|
| Temperature | K | Surface skin in forced runs; lowest-layer air in dry runs |
| Pressure | Pa | Terrain-dependent surface pressure |
u, v, wind speed |
m/s | Lowest sigma-layer winds, not fixed 10 m height |
| CO₂ frost | Pa-equivalent | Surface reservoir; mass per area is this value divided by gravity |
| Elevation | m | Regridded MOLA lower boundary |
Read temperature_kind, insolation_sampling, wind-height and climate-status
metadata before comparing references. The final map is an instantaneous sample.
3. Explore in the browser¶
Open http://127.0.0.1:8000 and select GCM. The server enables correlated-k CO₂
radiation, regolith, stability exchange, PBL diffusion and dry convection,
with spectral diffusion and available TES/seasonal Ames dust data. Its explicit
CO₂/dust longwave multipliers are 0.25 and surface exchange multiplier is 1.0.
This differs from the simple CLI map configuration.
4. Configure Python and restart¶
The Python GCM quickstart contains complete
forcing, map/export and restart examples. Use run_maps for reporting/deployment,
and parameterized step/rollout kernels for differentiated objectives.
Start with float64 and an explicit stable timestep.
5. Choose the next experiment¶
- Physical calibration and paired ablations.
- Neural radiation and bounded heating.
- Frozen temperature-only postprocessing.
- Reference comparisons.
- Validation and limitations.
A Mars year is about 668.62 sols. Equilibrium needs matched-phase convergence, including deep soil, rather than elapsed duration alone.
Deprecated global-mean model
Use GCM workflows for new simulations. tform mars run and torch ODE
examples are retained under deprecated documentation.