Update

Co-authored-by: Benoit Bovy <[email protected]>

Author: dcherian

design_notes/raster_index.md

@@ -61,11 +61,12 @@ Below is a proposal for (2).
 ## Proposal for transform index
 
 We design RasterIndex as a wrapper around **one** of many transform based indexes:
-1.  AffineTransformIndex ↔ GeoTransform
-2.  ModelTransformationIndex ↔ ModelTransformationTag
-3.  ModelTiepointScaleIndex ↔ ModelTiepointTag + ModelPixelScaleTag
-4.  GCPIndex ↔ Ground Control Points
-5.  RPCIndex ↔ Rational Polynomial Coefficients
+1. AffineTransformIndex ↔ GeoTransform
+2. ModelTransformationIndex ↔ ModelTransformationTag
+3. ModelTiepointScaleIndex ↔ ModelTiepointTag + ModelPixelScaleTag
+4. GCPIndex ↔ Ground Control Points
+5. RPCIndex ↔ Rational Polynomial Coefficients
+6. Subsampled auxiliary coordinates, detailed in [CF section 8.3](https://cfconventions.org/Data/cf-conventions/cf-conventions-1.12/cf-conventions.html#compression-by-coordinate-subsampling) and equivalent to GDAL's [geolocation arrays](https://gdal.org/en/stable/development/rfc/rfc4_geolocate.html) with `PIXEL_STEP` and/or LINE_STEP` > 1. 
 
 Each of the wrapped index has an associated transform:
 ```python
@@ -92,6 +93,7 @@ class RasterTransform:
 
 ### Read-time
 These transforms are constructed by **popping** the relevant information from a user-provided source.
+This is analogous to an "encode/decode" workflow we currently have in Xarray.
 ```python
 transform = rasterix.RasterTransform.from_geotransform(ds.spatial_ref.attrs)
 # transform = rasterix.RasterTransform.from_tiepoints(ds.band_data.attrs)