NeuralGCM-ENS Stochastic 50-Member Ensemble Forecasts for 2020 (WeatherBench 2)

Provider
Google Research (Kochkov et al., NeuralGCM)
License
CC-BY-SA-4.0 (source)
Format
Zarr (in sampled files)
Stored at
Google Cloud Storage, gs://weatherbench2/datasets/neuralgcm_ens (us)
Access
Free and public. No account with the data provider is needed.
DOI
10.1038/s41586-024-07744-y
Category
Climate & Weather
Source
https://doi.org/10.1038/s41586-024-07744-y

About NeuralGCM-ENS Stochastic 50-Member Ensemble Forecasts for 2020 (WeatherBench 2)

Probabilistic 15-day forecasts for 2020 from NeuralGCM-ENS, the stochastic 1.4 degree configuration of the NeuralGCM hybrid physics/machine-learning atmospheric model (Kochkov et al., Nature 2024), published in WeatherBench 2 for comparison with ECMWF IFS ENS and GenCast. Measured 2026-11-16: 2020-256x128.zarr holds 50 ensemble members (realization), 797 initialisation times from 2019-12-15 to 2021-01-16, 32 lead times from 0 to 372 h, and 37 pressure levels on a 256x128 grid, about 49 TB uncompressed. Variables: geopotential, temperature, specific humidity, cloud ice and liquid water content, u and v wind and wind speed. The folder also has regridded 240x121 and 64x32 conservative stores and _mean stores with the ensemble mean, which are what most CRPS and RMSE skill comparisons need. Load with xarray.open_zarr (token='anon'); the checkpoint is gs://neuralgcm/models/v1/stochastic_1_4_deg.pkl. Google releases the NeuralGCM simulation datasets under a Creative Commons licence (docs name CC-BY 4.0, link CC BY-SA 4.0; treat as CC BY-SA 4.0). Cite Kochkov et al. 2024.

How to load this dataset in Python

import os
import xarray as xr
ds = xr.open_zarr(os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], '2020-240x121_equiangular_with_poles_conservative.zarr'))
print(ds)

Run in a Scigantic notebook on 2026-11-21 against 2020-240x121_equiangular_with_poles_conservative.zarr: xarray Dataset dims [realization=50, time=732, prediction_timedelta=32, level=37, longitude=240, latitude=121], 8 variables e.g. ['geopotential', 'specific_cloud_ice_water_content', 'specific_cloud_liquid_water_content', 'specific_humidity', 'temperature', 'u_component_of_wind', 'v_component_of_wind.

From any machine

import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('weatherbench2/datasets/neuralgcm_ens/', detail=False)[:10]

Lists the files with anonymous access. Runs on any machine, with no cloud account.

Data structure

File types in sampled files: Zarr

Files

Example file: datasets/neuralgcm_ens/2020-240x121_equiangular_with_poles_conservative.zarr/geopotential/.zarray

Analyze this dataset without downloading it

On Scigantic, NeuralGCM-ENS Stochastic 50-Member Ensemble Forecasts for 2020 (WeatherBench 2) 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 Climate & Weather datasets