mdsumner
commented
7 months ago that in a context where show the warper resolving sentinel zones, regridding from rectilinear or curvilinear or regular, and ambiguities of ghrsst and similar
and, the generalized window read of RasterIO or GDALwarp
so exactextract in a scaleable tile context is (kind of trivial cont. below, paste fail)
what it makes me think is that R has sits, rsi, rstac, gdalcubes, terra, sf, gdalraster ... but only the last one really harnesses the lazy out of memory power of GDAL in a close way - the cubes are mostly about sentinel and landsat being in different crs, and not about the general capacity of the warper to resolve irregular grids
if you generate polygon masks, none of the R stuff has out of memory handling (except if you drive gdal with it directly, or my spinoff of fasterize "controlledburn" which needs attention to be useful)
exactextract is another case on its own, and efforts to get that into GDAL were hindered by voices in the rasterio and xarray camps
kind of trivial in R or Python reading windows, but it's more downstream duplication
xarray sits in this space and kind of drives perceptions, but it's a backward step to convert a regular grid to a rectilinear or curvilinear one, so I'm super interested where Ryan says this stuff about out of mem, and whether centre point is the right interp (on another thread), and about polygon overlays as an index for regridding and aggregation is a promising generalization (in a blog last year)