What do you think about adding RPC as proj:rpc here like proj:transform?

[bkavlak]

Some primary data is delivered with RPC files. Would it be a good fit to extend this with the metadata? Here is the specification: http://geotiff.maptools.org/rpc_prop.html

existent in rasterio as a model here https://github.com/rasterio/rasterio/blob/95f1f5fb55be6763bf987f42eb54c644604c6d3d/rasterio/rpc.py#L5

[m-mohr]

Additional files should be provided in the assets or links, depending on their usecase. We could define a relation type or asset role for this purpose in this extension. An additional field doesn't make a lot of sense to me, if it's files.

[bkavlak]

I think RPC is another method of projection, as opposed to an affine transformation, with coefficients defined under proj:transform. I understand the methodology is different in this case, so it might make sense to have a field. The RPC model would provide coefficients for any X, Y instead of the grid (again, as far as I understand).

We provide them as a separate asset (not sure if it should be link in the best practice), but sometimes map them in the relevant data asset extra fields temporarily so that I can do sth like this:

# mapping things to the asset application

data_asset = pystac.Asset(href="some_ungeoreferenced.tif")
data_asset.extra_fields["rpc"] = {...}

# in another processing application

rpc_model = rio.RPC(data_asset.extra_fields["rpc"])
transformer = rio.transform.RPCTransformer(rpc_model)

and then applying this transformation to the asset itself

[m-mohr]

If it's a per-pixel value, the number of values would be rather large so embedding them in the JSON seems lika a bad idea. Providing it as asset (with a yet to define role) would be a good option then. Do you already provide a specific role for your RPC assets?

[bkavlak]

Yes, we do that. Users usually prefer it to be embedded in the TIFF, but we map it anyway.