Google WeatherNext 2 operational forecast, virtual
| Spatial domain | Global |
| Spatial resolution | 0.25 degrees (~20km) |
| Time domain | Forecasts initialized 2025-01-01 00:00:00 UTC to Present |
| Time resolution | Forecasts initialized every 6 hours |
| Forecast domain | Forecast lead time 6-360 hours (0.25-15 days) ahead |
| Forecast resolution | 6 hourly |
WeatherNext 2 is Google's global medium-range probabilistic weather forecasting model. It produces a 64-member ensemble on a 0.25 degree grid, initialized four times daily with forecasts extending to 15 days.
This dataset is the 2025-present archive of Google WeatherNext 2 forecasts, optimized for spatial (map) access patterns. Forecasts are identified by an initialization time (init_time) and one of 64 ensemble_member values, then step forward from 6 to 360 hours (15 days) along the lead_time dimension at a 6 hourly interval. Surface variables are at the dataset root; variables carried on pressure levels are in the pressure_level group.
Each forecast step is published once its valid time is at least one hour in the past, so recent initializations are partially filled: steps that were not yet eligible at the most recent update read as NaN until a later update publishes them.
Related Datasets
Examples
import dynamical_catalog # dynamical-catalog>=1.0.0
ds = dynamical_catalog.open("google-weathernext2-forecast-operational-virtual", chunks=None)
# Choose an initialization whose day-10 step is eligible.
init_time = ds.init_time[-1] - ds.lead_time.sel(lead_time="240h")
day10 = ds.sel(init_time=init_time, lead_time="240h", ensemble_member=slice(0, 3), y=slice(25, 75), x=slice(270, 359.75))
(day10["wind_u_100m"] ** 2 + day10["wind_v_100m"] ** 2) ** 0.5import icechunk
import pystac
import xarray as xr
catalog = pystac.Catalog.from_file("https://stac-staging.dynamical.org/catalog.json")
collection = catalog.get_child("google-weathernext2-forecast-operational-virtual")
asset = collection.assets["icechunk-https"]
authorize = icechunk.containers_credentials({"https://wn.dynamical.org/chunks/": icechunk.Credentials.HttpAccess()})
repo = icechunk.Repository.open(
icechunk.http_storage(asset.href),
authorize_virtual_chunk_access=authorize,
)
session = repo.readonly_session("main")
ds = xr.open_zarr(session.store, chunks=None)
# Choose an initialization whose day-10 step is eligible.
init_time = ds.init_time[-1] - ds.lead_time.sel(lead_time="240h")
day10 = ds.sel(init_time=init_time, lead_time="240h", ensemble_member=slice(0, 3), y=slice(25, 75), x=slice(270, 359.75))
(day10["wind_u_100m"] ** 2 + day10["wind_v_100m"] ** 2) ** 0.5Dimensions
| min | max | units | |
|---|---|---|---|
| ensemble_member | 0 | 63 | 1 |
| init_time | 2025-01-01T00:00:00Z | Present | seconds since 1970-01-01 |
| lead_time | 21600 | 1296000 | seconds |
| x | 0 | 359.75 | degree_east |
| y | -90 | 90 | degree_north |
| pressure_level | 50 | 1000 | hPa |
Variables
Access a variable by its name (the bold identifier, e.g. ds["pressure_reduced_to_mean_sea_level"]).
Dimensions: init_time × ensemble_member × lead_time × y × x
| variable | units |
|---|---|
pressure_reduced_to_mean_sea_level
Pressure reduced to MSL (prmsl)
|
Pa |
sea_surface_temperature
Sea surface temperature (sst)NaN over land where sea surface temperature does not apply. Value changes once every 24 hours.
|
degree_Celsius |
temperature_2m
2 metre temperature (2t)
|
degree_Celsius |
total_precipitation_surface
Total precipitation (tp)Accumulated over a six-hour forecast interval. Small negative values are raw model artifacts; set values < 0 to zero.
|
kg m-2 |
wind_u_100m
100 metre U wind component (100u)
|
m s-1 |
wind_u_10m
10 metre U wind component (10u)
|
m s-1 |
wind_v_100m
100 metre V wind component (100v)
|
m s-1 |
wind_v_10m
10 metre V wind component (10v)
|
m s-1 |
Pressure Level (6 variables)
These variables live in the pressure_level Zarr group (e.g. pass group="pressure_level" to dynamical_catalog.open() or xr.open_zarr()).
Dimensions: init_time × ensemble_member × lead_time × y × x × pressure_level
| variable | units |
|---|---|
geopotential_height
Geopotential height (gh)
|
m |
specific_humidity
Specific humidity (q)Small negative values are raw model artifacts; set values < 0 to zero.
|
1 |
temperature
Temperature (t)
|
degree_Celsius |
vertical_velocity
Vertical velocity (w)
|
Pa s-1 |
wind_u
U component of wind (u)
|
m s-1 |
wind_v
V component of wind (v)
|
m s-1 |
Details
License
Dataset licensed under CC BY 4.0.
Attribution and citation
Google requires this attribution: © 2025 DeepMind Technologies Limited's machine learning models used to create the experimental data made available at https://developers.google.com/earth-engine/datasets/catalog/projects_gcp-public-data-weathernext_assets_weathernext_2_0_0 under CC BY 4.0 licence terms. This data is intended for experimental modelling only and is not intended, validated, or approved for real world use. Use of the third-party materials referred to in the Acknowledgements section may be governed by separate terms and conditions or license provisions. Your use of the third-party materials is subject to any such terms and you should check that you can comply with any applicable restrictions or terms and conditions before use.
Or Google WeatherNext 2 from dynamical.org.
Source
This archive is built from Google's WeatherNext 2 experimental forecast dataset. WeatherNext 2 is a global, medium-range ensemble forecast produced by an operational version of Google DeepMind's Functional Network Generative weather model.
The terms on Google's Earth Engine catalog page for this dataset distinguish Real-Time Experimental Data, which relates to a time less than one hour in the past or in the future and is governed by separate GDM Real-Time Weather Forecasting Experimental Data Terms of Use, from Historic Experimental Data, which relates to a time one hour or more in the past and is licensed under CC BY 4.0. This product publishes a forecast value only once the time it relates to (initialization time plus lead time, the valid_time coordinate) is at least one hour in the past, so everything it serves is Historic Experimental Data; nothing Google classes as Real-Time Experimental Data is published here.
Google prescribes the following citation for disclosures of findings arising from the Historic Experimental Data, and this collection also carries it verbatim in its attribution metadata: "© 2025 DeepMind Technologies Limited's machine learning models used to create the experimental data made available at https://developers.google.com/earth-engine/datasets/catalog/projects_gcp-public-data-weathernext_assets_weathernext_2_0_0 under CC BY 4.0 licence terms. This data is intended for experimental modelling only and is not intended, validated, or approved for real world use." The data is experimental: it is not an official forecast, alert, warning or notice from a meteorological agency. CC BY 4.0 includes a disclaimer of warranties and limitation of liability.
Modifications. dynamical.org has modified the source data as follows: it is restructured into an Icechunk virtual dataset with dynamical.org's variable and dimension names, and read-time filters convert temperatures from kelvin to degrees Celsius, geopotential to geopotential height (divided by standard gravity), and total precipitation from metres of water to kg m⁻². Values are otherwise the unchanged model output.
Third-party materials. The third-party clause carried in this collection's attribution metadata refers to the Acknowledgements on Google's dataset page, which say that the experimental data was generated by models which communicate with and/or reference the following separate libraries and packages, and that use of these third-party materials may be governed by separate terms and conditions or license provisions:
- Data and products of the European Centre for Medium-range Weather Forecasts (ECMWF), as modified by Google.
- Modified Copernicus Climate Change Service information 2023. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains.
- ECMWF HRES datasets
- Copyright statement: Copyright "© 2023 European Centre for Medium-Range Weather Forecasts (ECMWF)".
- Source: www.ecmwf.int
- License Statement: ECMWF open data is published under a Creative Commons Attribution 4.0 International (CC BY 4.0). https://creativecommons.org/licenses/by/4.0/
- Disclaimer: ECMWF does not accept any liability whatsoever for any error or omission in the data, their availability, or for any loss or damage arising from their use.
These notices are reproduced from the Acknowledgements section of Google's dataset page.
Data availability
This product contains four daily initializations from 2025-01-01T00 UTC to the present. Forecast steps are published one at a time as their valid times pass: a step is added about an hour after its valid time, so every initialization of the last 15 days is partially filled, with its remaining lead times NaN until an update publishes them. Select by valid_time, or by an initialization old enough for the lead time you need, rather than the latest initialization at a long lead. The fixed 2022–2024 source has a different native chunk layout and is published separately as the historical WeatherNext 2 archive. The two products share the same variables, dimensions and coordinates, so they can be concatenated along init_time when a workflow needs the full record.
Variables
The dataset carries 8 surface variables at the root and 6 atmospheric variables in the pressure_level group. Every forecast contains 64 ensemble members, 60 lead times at 6 hourly intervals, and a global 0.25 degree grid. The spatial dimensions are named y and x; their coordinates are geographic latitude and 0–360 degree longitude, not a projected grid.
Storage
The Icechunk repository is served over public HTTPS by dynamical.org. Its virtual chunks reference the unchanged, compressed source Zarr chunks through the public wn.dynamical.org HTTPS endpoint; the unit conversions described under Source are applied as read-time filters.
Chunks
This dataset is stored in Zarr format, which splits each variable into a grid of chunks — the smallest unit read from storage. When possible, aligning your reads with this dataset's chunk grid can significantly improve data access speed.
The element count and coordinate span of this dataset:
| dimension | chunk |
|---|---|
| init_time | 1 (6 hours) |
| ensemble_member | 1 |
| lead_time | 1 (6 hours) |
| y | 721 (180.25°) |
| x | 1440 (360°) |
| uncompressed | 4.0 MiB |
Validation report
Review the validation report for variable availability, missing data, known quirks, fill values, value distributions, and sample plots.