Rasters.jl has some basic rasterization tools, and I’m pretty close to merging this https://github.com/rafaqz/Rasters.jl/tree/methods_cleanup which will add much better, and pretty fast standardisted rasterization and masking for all GeoInterface.jl vector objects, as well as other vector formats and dataframes. I’m kind of aiming for rasterize and mask from Rs terra, but generic to all multidimensional objects and hopefully a lot faster.
You can also use RasterStack (e.g. a NetCDF file or list of tifs) directly as a Tables.jl table in most stats packages. It should be fast, but needs read from disk before use. A point extraction/rasterize round trip is also pretty fast if you do that.
GeoTiff compression should work with ArchGDAL.jl, and you can pass compression arguments to write in Rasters.jl. It probably needs some testing, if you want to try that would be great.
And for JSON this should work fine?
https://github.com/JuliaGeo/GeoJSON.jl