STAC vs Open Datacube¶
The Open Datacube (ODC) project, on which this library is based, started before STAC spec existed. As a result ODC uses different terminology for otherwise very similar concepts.
STAC |
ODC |
Description |
|---|---|---|
Product or |
Collection of observations across space and time |
|
Single observation (specific time and place), multi-channel |
||
Component of a single observation |
||
Pixel plane within a multi-plane asset |
||
Alias |
Refer to the same band by different names |
Similarly to STAC, ODC uses several levels of hierarchy to model metadata. At the highest level there is Product which is a collection of Datasets. Each Dataset contains a set of Measurements and related metadata. Finally Measurement describes a single plane of pixels captured at roughly the same time. Metadata includes location of the “file” and possibly location within a file.
Multiple Bands per File¶
Multiple bands in a single file are supported by both ODC and STAC, but representation differs. In STAC another level of hierarchy is added below an Asset via the bands attribute of the EO extension. Resources pointed to by an Asset may contain more than one band of pixels, and an Asset contains descriptions of those bands. In ODC, Asset is not modelled explicitly, instead resource path and potential location within this resource are properties of a Measurement object. It is common in STAC to have one to one mapping between band and asset, and in that scenario ODC Measurement and STAC Asset can be seen as equivalent.
Geo Referencing Metadata¶
Precise geo referencing metadata is stored within a file pointed to by Asset/Measurement, but it can also be recorded within a STAC Item/ODC Dataset document. Having geo-referencing information at this level can enable more efficient data access by providing spatial information without needing to access the source (data file) itself.
In STAC, the Projection Extension is used to bring this metadata from file to Item document. In STAC each band might have different projection, but in ODC projection is a Dataset level property and has to be shared across all Measurements. In ODC individual bands can be of different resolution and have different footprints (usually with a lot of overlap), but must be in the same projection.
Consistency Assumptions¶
In STAC, Collection is a very loose term, in theory it can point to very heterogeneous set of Items. In practice Items are typically very similar in structure, most contain the same set of Assets and bands. ODC is more strict in that regard. ODC Product contains expected set of Measurements per Dataset as well as some basic common metadata per Measurement, specifically pixel data type, which is assumed to stay the same across all Datasets for a given Measurement.
STAC equivalent would be Item Assets extension with Raster Extension inside. It describes at the Collection level, expected structure of Items contained within.