Real-world examples showcasing RasterIndex with 2D coordinate arrays

[benbovy]

One of the great potential benefits of RasterIndex is handling complex transformations where the pixel plane is not aligned with the world plane. A RasterIndex instance can indeed manage the coordinates of the two axes both represented as 2-dimensional lazy arrays.

This is explained more in detail here: https://discourse.pangeo.io/t/example-which-highlights-the-limitations-of-netcdf-style-coordinates-for-large-geospatial-rasters/4140/12, with possible examples including low-level data products (raw acquisition image data with arbitrary GCPs) but also intentional transformations (e.g., rotation) for specific data processing purpose.

[scottyhq]

One of the great potential benefits of RasterIndex is handling complex transformations where the pixel plane is not aligned with the world plane

In case it's helpful I put an example of a public raster with rotated affine here:
https://github.com/scottyhq/rasterix/blob/examples/notebooks/rotated.ipynb

At least for rotated affines this issue on rioxarray is relevant for describing the different approach taken by odc.geo currently corteva/rioxarray#847 .

[benbovy]

In case it's helpful I put an example of a public raster with rotated affine

Yes it is helpful!

I'm still trying to understand what are the current challenges with those datasets and how rasterix / RasterIndex may help solving it. Looking at the rioxarray issue it seems that the problem is about accurate computation of the raster bounding box and resolution, which may be something out of scope for RasterIndex (apart from providing a convenient API to get the affine transform parameters), or maybe not?

[benbovy]

or maybe not?

Probably not: https://github.com/dcherian/rasterix/pull/18#discussion_r2049241816.