There are three component specifications that together make up the core SpatioTemporal Asset Catalog specification. Each can be used alone, but they work best in concert with one another. The STAC API specification builds on top of that core, but is out of scope for this overview. An Item is the core atomic unit, representing a single spatiotemporal asset as GeoJSON so it can be searched. The Catalog specification provides structural elements, to group Items and Collections. Collections are catalogs, that add more required metadata and describe a group of related Items. For more on the differences see the section below.
A UML diagram of the STAC model is also provided to help with navigating the specification.
Fundamental to any SpatioTemporal Asset Catalog, an Item represents an atomic collection of inseparable data and metadata. A STAC Item is a GeoJSON Feature and can be easily read by any modern GIS or geospatial library, and it describes a SpatioTemporal Asset. The STAC Item JSON specification uses the GeoJSON geometry to describe the location of the asset, and then includes additional information:
- the time the asset represents;
- a thumbnail for quick browsing;
- asset links, to enable direct download or streaming access of the asset;
- relationship links, allowing users to traverse other related resources and STAC Items.
A STAC Item can contain additional fields and JSON structures to communicate more information about the asset, so it can be easily searched. STAC provides a core set of Common Metadata and there is a wider community working on a variety of STAC Extensions that provide shared metadata for more specific domains. Both aim to describe data with well known, well defined terms to enable consistent publishing and better search. For more recommendations on selecting fields for an Item see this section of the best practices document.
A 'spatiotemporal asset' is any file that represents information about the earth captured in a certain space and time. Examples include Imagery (from satellites, planes and drones), SAR, Point Clouds (from LiDAR, Structure from Motion, etc), Data Cubes, Full Motion Video, and data derived from any of those. The key is that the GeoJSON is not the actual 'thing', but instead references files and serves as an index to the 'assets'. It is not recommended to use STAC to refer to traditional vector data layers (shapefile, geopackage) as assets, as they don't quite fit conceptually.
Before we go deep into the Catalogs and Collections, it is worth explaining the relationship between the two and when you might want to use one or the other.
A Catalog is a very simple construct - it just provides links to Items or to other Catalogs. The closest analog is a folder in a file structure, it is the container for Items, but it can also hold other containers (folders / catalogs).
The Collection specification extends the Catalog spec, so every Collection can serve as a catalog, requiring the same fields. But it has a number of additional fields: license, extent (spatial and temporal), providers, keywords and summaries. Every Item in a Collection links back to their Collection, so clients can easily find fields like the license. Thus every Item implicitly shares the fields described in their parent Collection.
But what should go in a Collection, versus just in a Catalog? A Collection will generally consist of a set of assets that are defined with the same properties and share higher level metadata. In the satellite world these would typically all come from the same sensor or constellation. It corresponds directly to what others call a 'dataset series' (ESA, ISO 19115), 'collection' (CNES, NASA), and 'dataset (JAXA, DCAT)'. So if all your Items have the same properties, they probably belong in the same Collection. But the construct is deliberately flexible, as there may be good reasons to 'break' the recommendation.
Catalogs in turn are used for two main things:
- Split overly large collections into groups
- Group collections into a catalog of Collections (e.g. as entry point for navigation to several Collections).
The first case allows users to browse down into the Items of large collections. A collection like Landsat usually would start with path and row Catalogs to group by geography, and then year, month and day groups to enable deeper. Dynamic catalogs can provide multiple grouping paths, serving as a sort of faceted search.
The second case is used when one wants to represent diverse data in a single place. If an organization has an internal catalog with Landsat 8, Sentinel 2, NAIP data and several commercial imagery providers then they'd have a root catalog that would link to a number of different Collections.
So in conclusion it's best to use Collections for what you want user to find as starting point, and then catalogs are just for structuring and grouping the data. Future work includes a mechanism to actually search Collection-level data, hopefully in concert with other specifications.
There are two required element types of a Catalog: Catalog and Item. A STAC Catalog points to STAC Items, or to other STAC catalogs. It provides a simple linking structure that can be used recursively so that many Items can be included in a single Catalog, organized however the implementor desires.
STAC makes no formal distinction between a "root" catalog and the "child" catalogs. A root catalog is simply a top-most catalog (which has no parent). A nested catalog structure is useful (and recommended) for breaking up massive numbers of catalog Items into logical groupings. For example, it might make sense to organize a catalog by date (year, month, day), or geography (continent, country, state/prov). See the Catalog Layout best practices section for more.
A simple Catalog structure might look like this:
- catalog (root)
- catalog
- catalog
- item
- asset
- item
- asset
- item
- item
- asset
- asset
- catalog
- catalog
This example might be considered a somewhat "typical" structure. However, Catalogs and Items can describe a number of different relationships. The following shows various relationships between catalogs and items:
Catalog->Item(this is a common structure for a catalog to list links to Items)Catalog->Catalog(this is a common tree structure to group sets of Items. Each catalog in this relationship may also include Item links as well as catalog links)
As all STAC Collections are also valid STAC Catalogs, all Catalogs described here could also be Collections.
The relationships are all described by a common links object structure, making use of
the rel field to further describe the relationship.
There are a few types of catalogs that implementors occasionally refer to. These get defined by the links structure.
- A sub-catalog is a Catalog that is linked to from another Catalog that is used to better organize data. For example a Landsat collection might have sub-catalogs for each Path and Row, so as to create a nice tree structure for users to follow.
- A root catalog is a Catalog that only links to sub-catalogs. These are typically entry points for browsing data. Often they will contain the STAC Collection definition, but in implementations that publish diverse information it may contain sub-catalogs that provide a variety of Collections.
- A parent catalog is the Catalog that sits directly above a sub-catalog. Following parent catalog links continuously will naturally end up at a root catalog definition.
It should be noted that a Catalog does not have to link back to all the other Catalogs that point to it. Thus a published root catalog might be a sub-catalog of someone else's structure. The goal is for data providers to publish all the information and links they want to, while also encouraging a natural web of information to arise as Catalogs and Items are linked to across the web.
The Catalog specification is designed so it can be implemented as easily as possibly. This can be as simple as simply putting linked json files on a file server or an object storage service (like AWS S3), or it can be generated on the fly by a live server. The first type of implementation is often called a 'static catalog', and any catalog that is not just files is called a 'dynamic catalog'. You can read more about the two types along with recommendations in this section of the best practices document, along with how to keep a dynamic catalog in sync with a static one.
In addition to information about different catalog types, the best practices document has a number of suggestions on how to organize and implement good catalogs. The catalog specification is designed for maximum flexbility, so none of these are required, but they provide guidance for implementors who want to follow what most of the STAC community is doing.
- Catalog Layout is likely the most important section, as following its recommendations will enable catalogs to work better with client tooling that optimizes for known layouts.
- Use of Links articulates practices for making catalogs that are portable (with relative links through out) and ones that are published in stable locations (with absolute self links).
- Versioning for Catalogs explains how to use STAC's structure to keep a history of changes made to Items and catalogs.
- STAC on the Web explains how catalogs should have html versions for each Item and Catalog, as well as ways to achieve that.
A STAC Collection extends the Catalog construct to provide additional metadata to describe the set of Items it contains. So every Collection is a Catalog, and can be used in the same ways. The required fields are fairly minimal - it inherits the 4 required Catalog fields (id, description, stac_version and links), and adds license and extents. But there are a number of other common fields defined in the spec, and more common fields are also defined in STAC extensions. These serve as basic metadata, and ideally Collections also link to fuller metadata (ISO 19115, etc) when it is available.
As Collections are Catalogs, they can be used just as flexibly. They can have both parent Catalogs and Collections as well as child Items, Catalogs and Collections. Items are strongly recommended to have a link to the Collection they are a part of. Items can only belong to one Collection, so if an Item is in Collections that is the child of another Collection, then it must pick which one to refer to. Generally the 'closer' Collection, the more specific one, should be the one linked to.
The Collection specification is used standalone quite easily - it is used to just describe an aggregation of data, and doesn't require links down to sub-catalogs and Items. This is most often used when the software does operations at the layer / coverage level, letting users manipulate a whole collection of assets at once. They often have an optimized internal format that doesn't make sense to expose as Items. OpenEO and Google Earth Engine are two examples that only use STAC collections, and both would be hardpressed to expose individual Items due to their architectures. For others implementing STAC Collections can also be a nice way to start and achieve some level of interoperability.