Interpolation considerations

[jerstlouis]

As discussed for the WCS Interpolation Extension update, we should:

• Allow (or require?) servers to specify the default interpolation being used globally, per field and/or per axis
• Consider defining a new parameter for clients to request the use of a specific interpolation method globally, per field and/or per axis

[pebau]

@jerstlouis sorry, what is "default orientation"? I missed something here I believe.

[jerstlouis]

@jerstlouis you did not miss anything -- my brain is just quirky ;) Fixed, thanks!

[KathiSchleidt]

From my experience with derived products, it would be essential to either have the interpolation specified for each field (band) and axis, or it must be possible to specify this in the request. Different types of data require different interpolation, e.g. counts (population data) or classifications (e.g. land cover) cannot be interpolated like raw imagery.

[chris-little]

The classic examples from meteorology and oceanography are that in the vertical, interpolation usually should be logarithmic (either base 10 or e)
Horizontally, the interpolations for, say, pressure, wind velocity, and precipitation would be completely different, and ideally may actually change across the coverage, such as on different sides of a weather front. HTH.

[jerstlouis]

SWG 2024-12-11:

We probably need to add default interpolation method, and recommended interpolation methods both per dimension and per field.

For the dimensions, we should add this to the Common Collection Description dimensions ( https://github.com/opengeospatial/ogcapi-common/blob/master/collections/openapi/schemas/common-geodata/extent.yaml ).

For the fields, we should add this as JSON Schema vocabulary extensions (#173, https://github.com/opengeospatial/ogcapi-common/wiki/Table-of-extensions-to-the-JSON-Schema-vocabulary ).

It should at least be a recommendation to specify these.

As for an interpolation parameter for the client to explicitly request a particular interpolation (whether global, or for a specific dimension and/or field), we could delay this to a future part.

What would these things be called? defaultInterpolationMethod and recommendedInterpolationMethods ?

How would the intersection of per-dimension and per-field interpolation methods work?

Some interpolation methods are registered on the definition server:

http://defs.opengis.net/vocprez/object?uri=http://www.opengis.net/def/interpolation/OGC/1/

[chris-little]

@jerstlouis @KathiSchleidt Apologies for closing by accident/inattention.

[jerstlouis]

SWG 2025-02-05:

• We discussed interpolation a bit today with @fmigneault and @joanma747
• We already have the variableType describing additional dimensions in the Collection Description https://docs.ogc.org/DRAFTS/20-024.html#_uml_class_model which can provide information
• We will delay to an extension the ability for client to pick a particular interpolation method
• Servers supporting the Scaling requirement class will do their best to return resampled data at the resolution requested by the client (specified as number of cells with scale-size or width/height, scale-factor relative to native resolution, or potentially at a specific resolution)
• We'll include recommendations providing guidance as to what servers should do and when in terms of resampling
• Perhaps we could add an x-ogc-variableType to the JSON Schema extensions to describe the variable type of fields, which could provide information about what kind of interpolations is possible?

[jerstlouis]

SWG 2025-02-12:

We added x-ogc-variableType to the JSON Schema extension, and agree to the way forward proposed last week.