r: zApply in parallel computing

Question

I need to aggregate a rasterbrick into monthly values. Normally, this would be easy by using zApply function from raster package. However, I have a large rasterbrick and this would take a very long time.

So basically, I am wondering if this would be easy to do it with some libraries like parallel or clusterR but I have no clue how to parallelize this process

# create a random raster stack

library(raster)

lay <- stack()

for (i in 1:365){
  print(i)
  ras <- matrix(rnorm(500, mean = 21, sd = rnorm(21, 12, 4)))
  ras <- raster(ras)
  lay <- addLayer(lay, ras)
}

dats <- seq(as.Date('2000-01-01'), length.out = nlayers(lay), by = 'days')

lay <- setZ(lay, dats)

monthlies <- zApply(lay, by = format(dats,"%m"), fun = 'mean') # aggregate from daily to monthly.

Thanks!

Answer 1

Use foraech and doParallel packages

You can use foreach and doParallel to achieve your result. You will need to:

• Detect the number of your CPU cores with detectCores()
• Initialize DoParallel to work with your CPU cores with registerDoParallel(numCores)
• Setup the foreach loop with the needed packages, any init variable, and a method to combine the results.

Your code will look like this:

library(foreach)
library(doParallel)
library(raster)

lay <- stack()

## Loading required package: iterators

numCores <- detectCores()
registerDoParallel(numCores)  # use multicore, set to the number of our cores

lay <- foreach (i=1:365, .init = lay, .combine = addLayer , .packages = "raster") %dopar% {
  print(i)
  ras <- matrix(rnorm(500, mean = 21, sd = rnorm(21, 12, 4)))
  ras <- raster(ras)
}

dats <- seq(as.Date('2000-01-01'), length.out = nlayers(lay), by = 'days')
lay <- setZ(lay, dats)
monthlies <- zApply(lay, by = format(dats,"%m"), fun = 'mean') # aggregate from daily to monthly

# When you're done, clean up the cluster
stopImplicitCluster()

Measuring speed improvement

You can test the speed improvement using System.time(). These are my results:

#Time with a standard for loop
system.time({
  for (i in 1:365){
    print(i)
    ras <- matrix(rnorm(500, mean = 21, sd = rnorm(21, 12, 4)))
    ras <- raster(ras)
    lay <- addLayer(lay, ras)
  }
})

user  system elapsed 
66.29    0.09   67.15 

#Testing foreach loop time
system.time({
  lay <- foreach (i=1:365, .init = lay, .combine = addLayer , .packages = "raster") %dopar% {
    print(i)
    ras <- matrix(rnorm(500, mean = 21, sd = rnorm(21, 12, 4)))
    ras <- raster(ras)
  }
})

user  system elapsed 
21.72    0.09   25.58

As we can see, there was an effective speed improvement by using this method.

Hope this helps.