Green's function experiments with the Ai2 Climate Emulator (ACE)
To install the necessary packages, run
conda create --name fme python=3.10
conda activate fme
pip install fme
conda install -c conda-forge xcdat
pip install cartopy
pip install jupyterlab
conda install conda-forge::cmcrameri
See ACE documentation for more information about ACE.
Checkpoint and forcing data for ACE can be found at the Ai2 Hugging Face repository
Initial conditions and output data for the Green's function simulations can be found on Dryad
First, download the necessary data for the model checkpoints, forcing files, and initial conditions. To create the climatology for ACE2-ERA5, run ./forcing/create_ERA5_control.py.
Forcing files should be stored in ./forcing/<model>/control/forcing.nc
Initial conditions should be stored in ./initialization/<model>/20191231_1800.nc
To get the net ToA radiation and to plot the Green's function, the incoming solar radiation and ocean masks should be stored in
./output/<model>/rsdt.nc
./output/<model>/ocean_mask.nc
All model settings are stored in _model_settings.py. Make sure the checkpoint name is correct.
To perform all patch simulation for a certain <model> and <patch_amplitude>, change the model and amplitude in _GF_driver.py and run
python _GF_driver.py
Patch perturbations will be stored in ./forcing/<model>/GFMIP_patches_<patch_amplitude>K.nc and the global-mean output stored in ./output/<model>/R_gm_monthly_<patch_amplitude>K.nc.
Finally, make_GF_yearly.ipynb contains the code to make the Green's functions and the reconstructions. For ease of comparison, the processed output for ACE2-ERA5, ACE-FV3, and ACE-EAM are available for download on Dryad. To recreate the historical timeseries for ./forcing/historical.nc.
This work is a collaborative effort between Senne Van Loon, Maria Rugenstein, and Elizabeth A. Barnes.
Senne Van Loon, Maria Rugenstein, & Elizabeth A. Barnes (2025), Reanalysis-based Global Radiative Response to Sea Surface Temperature Patterns: Evaluating the Ai2 Climate Emulator, arXiv:2502.10893
This project is licensed under an MIT license.
MIT © Senne Van Loon