WeidongLi/Stochastic_Ocean4BGC-ckpts

Stochastic_Ocean4BGC — Model Checkpoints

Pretrained checkpoints for the regional stochastic BGC ocean emulator (generative-model ensemble / data-assimilation bundle GenModel_DA).

Code: https://github.com/Weidong-Li-HydroMet/Stochastic_Ocean4BGC

File	Size	Model	Role
ConvNext_Forecast_dt1_best.pt	3.0 GB	ConvNextUNet	pixel-space deterministic one-step forecast (in 454 → out 451)
AE_best.pt	554 MB	AutoEncoder	pixel ↔ latent codec (in 451, latent_dim 128)
LatentGAP_ocean.pt	2.0 GB	ConUNet_1degV2	latent-space SDEdit diffusion

n_vars = 451 prognostic variables (temp/salt/dic/o2/no3/pp/chl/uo/vo × 50 levels + SSH), n_forcing = 3 (Qnet, tauuo, tauvo).

Download

pip install huggingface_hub
hf download WeidongLi/Stochastic_Ocean4BGC-ckpts --repo-type=model --local-dir checkpoints/
# verify integrity:
cd checkpoints && sha256sum -c SHA256SUMS.txt

Or in Python:

from huggingface_hub import hf_hub_download
p = hf_hub_download("WeidongLi/Stochastic_Ocean4BGC-ckpts", "AE_best.pt")

Then place the files under GenModel_DA/checkpoints/ and run the scripts as described in the GitHub README.