How about the parallelDist package in R: https://cran.r-project.org/package=parallelDist

Thank you for your suggestion Dave, did not know this package. I will take a look.

If you are using R on Windows, it might help to use a version optimized for Windows. Microsoft acquired Revolution Analytics and maintains their own version of R: https://mran.microsoft.com/open

R is developed in a Linux environment. In my experience benchmarking different R - OS combinations, running Linux on a virtual machine on Windows can be faster than running R natively on Windows. It's even faster to dual boot Linux--you can boot Linux on a flash drive, for example. If you are new to Linux, I suggest Ubuntu or Mint.

Hi Francesco, did you find a solution to the issue you raised? I would be interested as I'm facing the same problem...

No, I did not... I tried to use microsoft R open without a significant improvement. Also I tested the parallelDist package sugested by Dave; however, the size of my rasters collapse my RAM during the process.. I ended by use ArcGis and model builder to efficiently do this task.

Ok thanks for your quick feedbacks. I have just decided to switch to QGIS...or when R-Spatial reaches its limit;)

Hi Kevin,

I generally suggest folks use qGIS to perform such an operation as (clearly) R raster is quite slow. Mind you, I haven't tried this yet, but I wonder if converting the raster to a tibble with xy-coordinates using R package tabularaster could speed this up for you. So, a function the performs the following might work quite well:

1) Convert to a tibble

2) Convert xy of each cell centroid to an imaginary number for REALLY fast distance calculations

3) Pull the distance from each cell to the nearest cell with the value of your choice

4) Convert the tibble back to a raster using the new distances rather than the old cell values

I will need to perform a similar operation in the near future and will post code if I find it to be more efficient. I suspect it will be, particularly because it will also be quite easy to parallelize.

Cheers,

Peter

I do these kind analyses with GRASS GIS (which can be called directly from R). On my system the following took < 10 seconds.

library(raster)

library(rgrass7)

library(sp)

library(tictoc)

tic()

# Start up grass

initGRASS(

gisBase = "/usr/lib/grass74/",

home = tempdir(),

override = TRUE)

# Create a dummy raster

r2

Thank you Johannes Signer !,

Its could be the best approach (calling from R). I will try it!

If you have access, you could import the raster into ArcMap (part of the ArcGIS suite) and calculate Euclidian Distance using the appropiate tool?

Dear Francisco,

I guess you need the 'gridDistance' function in the Raster package, try something like this:

> railways.dist