Earth observation data is a software problem

Earth observation archives are often described as data problems. In practice, most of the friction is software: metadata that does not agree with itself, formats that require whole-file downloads, and access patterns that assume a local filesystem.

What cloud-native changes

Chunked, self-describing arrays make partial reads practical. A workflow can open a dataset, inspect coordinates, and request one time slice without downloading the archive.

import xarray as xr

ds = xr.open_zarr(
    "https://example.com/store.zarr",
    consolidated=True,
    chunks={},
)
print(ds)

The hard part is not the API. The hard part is consistency across collections, clear provenance, and clients that fail gracefully.

A short checklist

• Does the dataset expose stable, documented dimensions and coordinates?
• Can a user discover the variables without downloading the whole object?
• Are missing values and scale factors handled explicitly?
• Is there a small, runnable example in the documentation?

Those questions sound mundane. They are also the difference between an archive that is technically online and one that is actually usable.