[GRASS-user] Models using r.buildvrt

[Moritz Lennert]

Le Wed, 18 Sep 2019 10:56:04 -0700 (PDT),
Rich Shepard <rshepard at appl-ecosys.com> a écrit :

> There are 29 individual DEM maps for the drainage basin. The
> r.buildvrt manual tells me that using r.buildvrt will produce a
> slower loading map than would loading each raster map individually
> unless only a small portion of the virtual raster will be used.
> 
> Here, small portions of some DEM maps extend past the watershed
> boundary and that vector map will be used as a mask.
> 
> For hydrologic, terrain, and geomorphic analyses might processing
> times be different using all individual maps or a virtual one?

AFAIK, one could summarize in the following way:

- Reading a small map is faster than reading a large map
- Reading one single map is faster than reading many maps

What the above citation from the r.buildvrt man page says is that it's
all about compromise and use case:

- If you have data covering a very large zone, but your actual
  processing always only actually concerns small portions of that zone,
  but these portions might be across more than one existing tile, than
  you are better off combining the tiles virtually using the VRT
  paradigm.

- If however, you always work on the entire zone at once, then tiling +
  VRT will slow things down compared to using one single big map.

Tile size will obviously play a role. Depending on the size of your
DEMs and of the zones you need to treat, you could possibly get better
results by dividing your DEMs into even smaller tiles.

Moritz