Migrate from data.dynamical.org

data.dynamical.org is going away — plan on August 31, 2026. In practice we will turn URLs off dataset by dataset over the following months rather than all at once, so nothing breaks the morning of September 1.
The replacement datasets are stored as Icechunk Zarr repositories.
Only how they are opened needs to change. The Icechunk versions are identical in structure, naming, attributes, chunking/sharding, and content — drop-in replacements for the deprecated Zarrs.

Replace each hard-coded https://data.dynamical.org/.../latest.zarr URL with one of the two supported ways to open a dataset. Dataset IDs replace data.dynamical.org paths: https://data.dynamical.org/noaa/gfs/forecast/latest.zarr becomes noaa-gfs-forecast. Both need Icechunk 2, and both resolve the current repository from our STAC catalog, so your code does not need to track storage locations or version changes.

dynamical-catalog

Option 1: open using dynamical-catalog, a thin STAC and Icechunk wrapper that ensures minimum dependency versions. Install dynamical-catalog>=0.8.0 on Python 3.12 or newer, and pass the dataset ID from its catalog page.

import dynamical_catalog

ds = dynamical_catalog.open("noaa-gfs-forecast", chunks=None)

STAC and Icechunk

Option 2: load the STAC and open directly with Icechunk. Resolve the icechunk-https asset from the dataset's STAC Collection each time your application opens a repository. Do not copy and permanently hard-code the asset URL. The example below uses pystac, but any HTTP client can read the Collection.

import icechunk
import pystac
import xarray as xr

catalog = pystac.Catalog.from_file("https://stac.dynamical.org/catalog.json")
collection = catalog.get_child("noaa-gfs-forecast")
asset = collection.assets["icechunk-https"]

repo = icechunk.Repository.open(icechunk.http_storage(asset.href))
session = repo.readonly_session("main")

ds = xr.open_zarr(session.store, chunks=None)

Option 2 is not Python-specific. Icechunk publishes a JavaScript/TypeScript client (@earthmover/icechunk, paired with zarrita) and a Rust crate (icechunk), both of which open the same icechunk-https asset over HTTPS.

Have an agent do it

Paste this into a coding agent running in your repository.

Replace every data.dynamical.org URL in this repository with dynamical.org's
supported access, and leave no hard-coded storage locations behind.

Context to read first:
- https://dynamical.org/llms.txt for current access patterns and dataset IDs.
- https://stac.dynamical.org/catalog.json is the source of truth. Confirm each
  dataset ID against it rather than guessing.

Rules:
- A legacy URL maps to a dataset ID by dropping the host and joining the path
  with hyphens: data.dynamical.org/noaa/gfs/forecast/latest.zarr ->
  noaa-gfs-forecast. Exception: /noaa/gfs/analysis-hourly/ is retired, use
  noaa-gfs-analysis.
- Two ways to open a dataset, equally supported. Use either, consistently:

  import dynamical_catalog

  ds = dynamical_catalog.open("noaa-gfs-forecast", chunks=None)

  or, resolving the asset from STAC yourself:

  import icechunk
  import pystac
  import xarray as xr

  catalog = pystac.Catalog.from_file("https://stac.dynamical.org/catalog.json")
  collection = catalog.get_child("noaa-gfs-forecast")
  asset = collection.assets["icechunk-https"]

  repo = icechunk.Repository.open(icechunk.http_storage(asset.href))
  session = repo.readonly_session("main")

  ds = xr.open_zarr(session.store, chunks=None)

  Any HTTP client reads the STAC; pystac is not required.
- The first option needs dynamical-catalog>=0.8.0.
- Either way you open an Icechunk 2.0 repository, so Icechunk 2 is required; in
  Python that means 3.12+. Resolve the asset at open time — do not hard-code the
  asset href, the S3 bucket, or a version number.
- Drop any ?email= query argument. It is no longer used.
- The datasets are identical to the legacy archive in structure, naming,
  attributes, chunking/sharding, and content — only the open call changes, so
  leave the surrounding analysis code alone.

Report every URL you could not map, and every place a dataset ID had to be
chosen rather than derived.

Help

If you have any questions or run into trouble migrating, we want to help. Email us at [email protected].