BIOMASS workflow with Gamma0 LUT

User @nmt28 is working on a science case using BIOMASS data and has run into a few tricky issues we can help with.

• Some of the ancillary files to BIOMASS are netcdf not geotiff
• The Geotiffs are not in ground coordinates and require use of GCPs align
• In some cases the entire granule may need to be interpolated for use

Repo is https://github.com/nmt28/ESA_MAAP/tree/main/MAAP/BIOMASS
Ask in slack for access (currently private)

If I understand correctly, @nmt28 is building a data cube with time dimension of all granules within an Area of Interest (AOI) within a date range.

• currently I read what I need to with rio, whether that be local (LUT) or the radar backscatter (fsspec/rio), but when I'm done interpolating the Gamma0 data from the LUT I store it as an xarray, which allows me to set the spatial info without writing to disk
• I think it a bilinear interpolation and I don't see why it cant be done over a window. I can share code for this
• by "All" the Gammas, I mean all radar images (once corrected to gamma) that fall within a certain roi between certain dates. Basically, I want the mean of multiple overlapping images. This ties back into he GCP issue as I need the spatial info to do the averaging based onthe ROI bounds
• I was thinking of getting all the data in 4326 (or 4979 if you're esa) and doing the averaging before doing the reproj?

Questions:

• What's the final projection, and AOI grid that will be used?
• Can or should we create Virtual cubes over some of these products with Virtual Zarr so that we can offload the complexity of opening and reading bytes to a lazy as needed method?
• The Bilinear interpolation, is this because of the reprojection or for some other reason?
• What needs to be saved to disk in the end? What format makes the most sense Zarr?

Possible Avenues:

• Switch to obstore+xarray(h5netcdf) instead of (fsspec+rasterio), or some other combination of things.