Conventions for Catalog Structure

To achieve a well structured and easily searchable catalog, we need to define some conventions for the structure of the catalog and the metadata of the collections and items.

Collections and Items

Each collection describes an overall experiment. This can be a climate simulation, a reanalysis dataset or forecasts from a specific institution or model. Each item is a full dataset with a specific vertical levels, and spatial as well as temporal resolution. For current climate simulations produced at DKRZ, this results in an item for each combination of healpix zoom and output frequency. For forecast runs or ERA5 each item is a specific vertical level type like pressure levels or model levels. Those datasets also contain surface variables.

An item has at least one asset, which provides access to the data. All items should offer an option to open them as Zarr and in the best case also as Zarr over HTTP for remote access.

STAC Extensions

Items within the catalog use mainly two extensions:

  1. Scientific extension to highlight DOI and citation information

  2. Datacube extension to give a detailed view on the axis and variables

Metadata Fields

All items and collections have a metadata field called Variable`, which contains a list of all included variables in long names. Those should not include any abbreviations, short names or grib codes to ensure brosability of the catalog.

Schema for Collection IDs

As already described within the section on user rights management, collection IDs have to start with the prefix of the institution responsible for the collection. The rest of the collection ID is up to the data provider. Collection IDs are not intended for human but for machine readability, so they can be a combination of letters and numbers. We recommend to include metadata fields from the underlying dataset to create a unique ID.

Schema for Item IDs

There are no regulations on item IDs. We recommend to use metadata information of the dataset to create a unique item ID.

Conventions for Asset IDs

At the moment, we do not have any strict conventions for asset IDs. We recommend to use descriptive keys, that include the data format or the access method, e.g. zarr or zarr-http.