Edzer Pebesma
Looks like more fall-out of using spherical geometry; setting `sf_use_s2(FALSE)` prior to `st_intersect()` should do. The reason might be that the horizontal "straight lines" you see as polygon boundaries you...
This is not a bug, it is a feature (of spherical geometry).
Could you provide a sample image, or a link to it?
```r library(stars) # Loading required package: abind # Loading required package: sf # Linking to GEOS 3.10.2, GDAL 3.4.3, PROJ 8.2.0; sf_use_s2() is TRUE read_stars(gdal_subdatasets('sample_tiff_8bands.tiff'), proxy = TRUE) # stars_proxy...
In case you can't fix this yourself, I would suggest trying to work with more up-to-date versions of PROJ, GDAL and GEOS.
I committed changes that I think do sth similar but for a `stars_proxy` object; could you do the benchmark again when reading `x1` with `proxy = TRUE`?
Did you update stars? I see ``` > aggregate(x1_proxy, nc, mean) although coordinates are longitude/latitude, st_intersects assumes that they are planar although coordinates are longitude/latitude, st_intersects assumes that they are...
Interesting is also the difference in memory footprint. Of course the gain with this approach is only large if `by` is geometrically small compared to `x`. I agree that sth...
Did you also try this on non-proxy data? Your benchmark would be most useful it was reproducible, as the one from OP.
Thanks, looks great, will try to run them!