Closed pairsa closed 1 year ago
Hi @pairsa
There was an on-going (a bit in stand by) discussion about the computational costs of computing the transitions. This is because gdistance
depends a lot on the packages raster
and Matrix
. We think that with the new package terra
and with data.table
we could solve or reduce the computational cost. But this will take some time to implement. Meanwhile, I quote a text send by Andy Nelson (UTwente) as his work around to produce the data for this publication https://doi.org/10.1038/s41597-019-0265-5
"I did not mosaic transition matrices, I mosaiced the resulting travel time surfaces. I cut up my study area (the world) into overlapping tiles, generating matrices for each one, ran the cost distance function on each one, and merged the outputs of that to get a global map"
I hope this can offer to you a solution in the short run, while we study the opportunities to implement a faster gdistance
K
I will be rewriting the transition()
and geoCorrection()
functions to be substantially more memory efficient. I'm going to go ahead and close this for now, and in the future, it can be revisited if the plan advancements fail to meet people's needs.
Hello! I am attempting to produce a transition layer from a very large raster, and am (expectedly) running into memory issues in R. Is there a work-around for using the "transition" function on large raster layers? My instinct is to tile the raster, produce smaller transition layers from these tiles, then mosaic them together somehow. Is there a way mosaic objects of class "TransitionLayer" in R?