PCA using raster datasets in R

Question

I have several large rasters that I want to process in a PCA (to produce summary rasters). I have seen several examples whereby people seem to be simply calling prcomp or pr

Answer 1

Here is another approach that expands on the getValues approach proposed by @Daniel. The result is a raster stack. The index (idx) references non-NA positions so that NA values are accounted for.

library(raster) 
r <- stack(system.file("external/rlogo.grd", package="raster")) 
r.val <- getValues(r)
idx <- which(!is.na(r.val)) 
pca <- princomp(r.val, cor=T)

ncomp <- 2 # first two principle components
r.pca <- r[[1:ncomp]]
  for(i in 1:ncomp) { r.pca[[i]][idx] <- pca$scores[,i] } 

plot(r.pca)