Integrating geotargets tile support

[sigmafelix]

https://www.r-bloggers.com/2024/11/geotargets-0-2-0/

geotargets 0.2.0 seems to be supporting raster tiling. It is worth integrating this functionality into prediction grid computation.

[sigmafelix]

https://github.com/sigmafelix/workbenches/blob/ad41fddecb9a5139d0c9c8600b515ad58c899aab/targets_split/_targets.R demonstrated chopin and geotargets applications for distributed processing in a targets pipeline. Native tiling in geotargets supports exhaustive only, thus I will add chopin way of tiling with overlaps to the current pipeline soon.

[sigmafelix]

It might be a bit off-topic, but I want to record that my current implementation of dividing the prediction grid points into subsets works well with pre-processed MODIS files in the Apptainer container (I want to ask your understanding of the local interactive run since my system is not configured for SLURM):

Fifty workers are working for 48+ hours in total to calculate MOD06 with three radii and 40 by 20 grid splits for one month (*it doesn't match the system record of btop in my machine (uptime: 1d 01:38), which is because of a sudden crash leading to system reset). My estimation for MOD06 calculation is around 11 days, which needs a significant reduction. There is a point of consideration regarding setting the proper number of grid dimensions:

• Preferable (if not optimal) settings considering memory usage per worker and the total runtime considering parallel overhead
  • Trying coarser grids (e.g., 20 by 10 or 10 by 5)
  • Estimating impacts of adjustment to other prediction grid calculation targets (e.g., vector calculations such as ecoregion)