Physically Consistent Sampling For Ocean Model Initialization - Dataset

This dataset gathers all data and codes needed to reproduce the figures in Physically Consistent Sampling For Ocean Model Initialization, submitted to Tackling Climate Change with Machine Leurning NeurIPS Workshop 2025. (Link to paper and poster). Description of the files: generated_npy_files.zip gathers the npy temperature and salinity fields generated by the diffusion model for the 4 different cases used in the paper: no conditioning, temperature and salinity conditioning, C1 conditioning and C2 conditioning. model_weights.zip gathers the weights of the trained diffusion model used to generate the npy files. dino_1_4_degree_coarse_240125.tar : the data used to train the diffusion model emp_climatology_1_4deg.nc is needed to run a NEMO simulation with our parameters. 1deg_restart.zip contains the reference files required for the generation of the 1deg restart files from the from the npy files. restart25_arch.zip contains the reference files required for the generation of the 1_4deg restart files from the from the 1deg restart files. before_10y_integration.zip and after_10y_integration.zip contains all the restart files generated from the npy files with the script (see convert_state.ipynb and in DINO-FUSION github). "Before" files are computed directly from the npy files and "after" files are the results obtained after 10 years of NEMO integration. DINO_12960000_restart.nc is the restart file used as reference (target) for long term stratification in the paper's long term integration Figure. Additionnally, the codes to produce the Figures can be found in the following repositories: https://github.com/Etienne-Meunier/DINO-Fusion/tree/dev-diffusion https://github.com/Etienne-Meunier/Dinonline/tree/diffusion The first link contains the code of the constraint diffusion model, the tutorial to create the restart files needed to run the NEMO simulation from the npy files (see Results/notebooks/convert_state.ipynb and first step of Results/Reports/Tuto_launch_nemo.md) and the code to plot the zonally averaged density figures both before and after NEMO simulation. The second link contains the files and scripts that are needed to run nemo simulations with our parameters. To reproduce the Figures: Results/notebooks/convert_state.ipynb allows to convert the generated npy files into first 1deg restart files and then interpolate to 1_4deg restart files. (For this script, generated_npy_files.zip, 1deg_restart.zip and restart25_arch.zip are required). Results/notebooks/compute_density.ipynb allows to reproduce the vertical density figures of the paper from initial and final states. For this, the previously converted 1_4deg restart files are needed (see restart_files_1_4deg.zip before), the restart files after 10 years of NEMO simulation