general todo

[mdsumner]

• In the volcano, take the index coords of the matrix and transform them via the offset and scale to generate the image coords. Relate that to the affine transform,
• show that the vrt fix for GHRSST is truly language agnostic, run with rasterio, cli, sf/terra/gdalraster, gdalcubes
• take a chunk of GHRSST and extract it in native, but also a projected window, match that with image tile server, model output from CMIP or the precip Zarr, bathymetry, and so on see #3

[mdsumner]

chapters

rasterio is not gdal

I want to divert a little from rasterio ... I'll reply with the GHRSST grid example, show that osgeo.gdal will happily warp to a regular grid using the inexact rectlinear coords, OR simply translate it after assigning the ostensible transform that we infer it should have (with open qs ??)

show a curvilinear one, and that's the big choice, do we brute force a new regular grid from the coords, or do we assign what we believe should have been there as a regular transform - netcdf can't store that high-fidelity representation and neither can xarray atm

see gdal-dev thread, and parts of gists recently

extent + dim + crs vs affine

and I want to make clear that extent+dim+crs is complementary to the geotransform way, not saying that it needs to be adopted (there are shear-based files out there) but it's certainly an easier way to think about what the abstract spec actually is