Icechunk stores

Icechunk

Zarr v3

/vsis3/

weather

ECMWF AIFS forecasts and ERA5 from public Icechunk repositories, when GDAL is built with Icechunk support.

Icechunk is a transactional storage engine for Zarr. GDAL reads Icechunk repositories through /vsiicechunk/ when built with it, which most packaged builds are not yet. This page runs only where the build has it; the store status below says whether it did this time.

"/vsiicechunk/" in gdal.GetFileSystemsPrefixes()

False

ECMWF AIFS forecasts

ECMWF’s AI forecast system, published by dynamical.org as an Icechunk repository on anonymous S3 (CC-BY-4.0). New initialisation times are appended as they are produced.

Not run in this build: ECMWF AIFS single forecasts (dynamical.org Icechunk)

Address: /vsis3/dynamical-ecmwf-aifs-single/ecmwf-aifs-single-forecast/v0.1.0.icechunk
Provider: dynamical.org
Rolling: the address is stable, the contents change over time.
Status: this GDAL lacks /vsiicechunk/.

dsn = "/vsis3/dynamical-ecmwf-aifs-single/ecmwf-aifs-single-forecast/v0.1.0.icechunk"
cfg = {"AWS_NO_SIGN_REQUEST": "YES", "AWS_REGION": "us-west-2"}
if live("aifs-icechunk"):
    aifs = xr.open_dataset(dsn, engine="gdalxarray", config_options=cfg)
    print(list(aifs.data_vars)[:8], dict(aifs.sizes))

The 10 m wind at Hobart through the first forecast:

if live("aifs-icechunk"):
    pt = dict(latitude=-42.9, longitude=147.3, method="nearest")
    u = aifs["wind_u_10m"].isel(init_time=0).sel(**pt)
    v = aifs["wind_v_10m"].isel(init_time=0).sel(**pt)
    fig, ax = plt.subplots(figsize=(9, 3.2))
    u.plot(ax=ax, label="u"); v.plot(ax=ax, label="v"); ax.legend()
    plt.show()

A single 2D face of the 4D array, as a classic raster any GDAL tool can use: :{0}:{0} fixes init_time and lead_time.

if live("aifs-icechunk"):
    face = xr.open_dataset(
        'ZARR:"/vsiicechunk/{/vsis3/dynamical-ecmwf-aifs-single/'
        'ecmwf-aifs-single-forecast/v0.1.0.icechunk}":/wind_u_10m:{0}:{0}',
        engine="gdalxarray", multidim=False, config_options=cfg)
    print(face["band_data"].shape)

ERA5, partly readable

Earthmover’s ERA5 Icechunk repository opens, and its coordinates read, but its data variables use the pcodec codec, which GDAL does not decode yet. The read below shows the error GDAL gives; when GDAL gains the codec, this cell starts producing values with no change here.

Not run in this build: ERA5 pressure-level spatial group (Earthmover Icechunk)

Address: /vsiicechunk/{/vsis3/earthmover-icechunk-era5/icechunkV2}/pressure/spatial
Provider: Earthmover public demo
Not yet classified.
Status: this GDAL lacks /vsiicechunk/.

dsn = "/vsiicechunk/{/vsis3/earthmover-icechunk-era5/icechunkV2}/pressure/spatial"
if live("era5-icechunk"):
    era5 = xr.open_dataset(dsn, engine="gdalxarray",
                           config_options={"AWS_NO_SIGN_REQUEST": "YES", "AWS_REGION": "us-east-1"})
    print(dict(era5.sizes))
    var = list(era5.data_vars)[0]
    try:
        era5[var].isel({d: 0 for d in era5[var].dims}).values
    except Exception as e:
        print(type(e).__name__, str(e).splitlines()[0])

Rendered with GDAL 3.13.3, xarray 2026.4.1.dev32+g71cfb13ac, numpy 2.5.4, gdalxarray 0.5.1, rasterix 0.2.4.