NeuralGCM Pre-trained Model Checkpoints (v1)
- Provider
- Google Research (NeuralGCM team)
- License
- CC-BY-SA-4.0 (source)
- Stored at
- Google Cloud Storage, gs://neuralgcm/models/v1 (us)
- Access
- Free and public. No account with the data provider is needed.
- DOI
- 10.1038/s41586-024-07744-y
- Category
- Machine Learning & AI
- Source
- https://doi.org/10.1038/s41586-024-07744-y
About NeuralGCM Pre-trained Model Checkpoints (v1)
Trained weights for NeuralGCM, the hybrid model of Kochkov et al. (Nature, 2024) that couples a differentiable dynamical core for atmospheric dynamics with learned neural network physics, trained on ERA5 to produce both medium-range weather forecasts and multi-year climate simulations. Measured 2026-11-14: this folder (models/v1) holds deterministic checkpoints at 0.7, 1.4 and 2.8 degree resolution and a stochastic 1.4 degree checkpoint, 4 pickle files totalling about 0.31 GB; the sibling folder gs://neuralgcm/models/v1_precip holds two stochastic 2.8 degree checkpoints that predict precipitation and evaporation (one trained on precipitation, one on evaporation). The NeuralGCM documentation recommends the 1.4 degree stochastic model for weather forecasts, the 0.7 degree deterministic model for the first few forecast days, and the 2.8 degree stochastic models for multi-decadal atmosphere-only climate runs, which need external sea surface temperature and sea ice forcing. Load a checkpoint with pickle and pass it to neuralgcm.PressureLevelModel.from_checkpoint (JAX), then initialise from ERA5 such as the ARCO-ERA5 Zarr. Checkpoints are distributed by Google under CC BY-SA 4.0; the code is Apache 2.0.
How to access NeuralGCM Pre-trained Model Checkpoints (v1) in Python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('neuralgcm/models/v1/', detail=False)[:10]Lists the files with anonymous access. Runs on any machine, with no cloud account.
Data structure
Files
Example file: models/v1/
Analyze this dataset without downloading it
On Scigantic, NeuralGCM Pre-trained Model Checkpoints (v1) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Machine Learning & AI datasets
- Deutsche Bahn Train Delay and Cancellation Data
- Nemotron RL Bixbench Hypothesis
- Amazon-PQA
- ASL 1000
- The Massively Multilingual Image Dataset (MMID)
- RxR (Room-Across-Room): Multilingual Vision-and-Language Navigation Instructions in English, Hindi and Telugu with Pose Traces for Matterport3D (Google Research)
- Milling Tool Wear Time Series across Three Machine Tools (Flank Wear VB, Forces, Spindle Torque)
- Answer Reformulation