workflow recommendations for simple spatial mean computations about stars HOT 5 OPEN

Have you tried aggregate?, as in

aggregate(tile_proxy, tt, mean, na.rm = TRUE)

from stars.

Maybe it will be even better to use the exact = TRUE argument to engage exactextractr (it is very fast for summarizing values)?

Edit: I checked the source of the aggregate() and basically exact_extract() is not used, so you need to test {terra} as backend.

from stars.

No luck with aggregate - similar behaviour to st_apply - slowly consumes RAM/virtual memory until crashing.

I'll look into terra and exact_extract

from stars.

exactextractr::exact_extract works well, crunching through the test set in under a minute while using minimal RAM

library(exactextractr)
a <- raster::raster(tifname)
g <- exact_extract(a, tt, "mean")

# should work with either terra or raster loaders
a <- terra::rast(tifname)
g <- exact_extract(a, tt, "mean")

terra::extract had the same ram problem as the stars approach:

a <- terra::rast(tifname)
g <- terra::extract(a, tt, mean, na.rm = TRUE)

raster::extract didn't appear to have a RAM problem, but took a really long time - I gave up after over an hour.

a <- raster::raster(tifname)
g <- raster::extract(a, tt, mean, na.rm = TRUE)

from stars.

If you are interested in the highest efficiency, I think the best option will be to use exact_extract() with SpatRaster, it should be faster than for RasterLayer. See these benchmarks:

• https://github.com/isciences/exactextractr?tab=readme-ov-file#performance
• https://kadyb.github.io/raster-benchmark/report.html#Zonal_statistics

from stars.