14

Is there a straightforward way of randomly sampling a raster so that the output of the process is a raster?

I'm using an example that I found on the r-sig-geo list and I have also tried the sampleRandom function in the raster package. Both of these approaches produce an output that I am not certain how to transform into a raster. I was not able to find an approach after searching for several combinations of "SpatialPointsDataFrame raster".

library(raster)

# read in raster
rasterSource <- 'landsat.TIF'
r <- raster(rasterSource)

# convert to spatial points data frame
r.spgrd<-as(r,"SpatialPointsDataFrame") 

# elminate NA values
r.spgrd = r.spgrd[!is.na(r.spgrd[[1]]),] 

# sample points
selectedPoints = sample(1:length(r.spgrd[[1]]), 1000)
r.sampled = r.spgrd[selectedPoints,]

# try to make spgrd into a raster
r.test <- raster(r.sampled)

When I run r.test I get the output:

class       : RasterLayer 
dimensions  : 10, 10, 100  (nrow, ncol, ncell)
resolution  : 28617, 14766  (x, y)
extent      : 1838505, 2124675, 2328685, 2476345  (xmin, xmax, ymin, ymax)
coord. ref. : +proj=aea +lat_1=29.5 +lat_2=45.5 +lat_0=23 +lon_0=-96 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0 
values      : none

So that the following line which tries to write a raster produces the message:

# write out as ascii file
writeRaster(r.test, filename="test1.ASC", datatype="ascii", overwrite=TRUE)

Error: hasValues(x) is not TRUE

My main objective is to produce some type of raster after the sampling process. I'm also fine with just changing the values within my raster (I'm just not certain how to do that).

PolyGeo
  • 65,136
  • 29
  • 109
  • 338
djq
  • 16,297
  • 31
  • 110
  • 182

2 Answers2

21

You can adapt examples from the Raster package vignette, section 5.2. Here's one way:

r <- raster(ncol=30,nrow=20)
r[] <- 1:(30*20)              # Raster for testing
#plot(r)                      # (If you want to see it)
r[runif(30*20) >= 0.30] <- NA # Randomly *unselect* 70% of the data
plot(r)

Raster selection

whuber
  • 69,783
  • 15
  • 186
  • 281
  • 10
    This answer makes me want to learn R... – SaultDon Nov 25 '11 at 18:21
  • 2
    This may be extremely obvious, but it took me some time to realize that the value of any cell can be referenced by r[r "condition"]. So if you want to set all value of the raster that are 100 to be 1 you can write r[r == 100] <- 1. Thanks @whuber - extremely useful example! – djq Dec 11 '11 at 20:21
  • @whuber Any clue why the Error: hasValues(x) is not TRUE popped up? – csheth Jun 20 '15 at 20:05
3

You can use the sampleRandom function:

library(raster)
r <- raster(ncol=30,nrow=20)
r[] <- 1:ncell(r)

x <- sampleRandom(r, ncell(r)*.3, asRaster=TRUE)
Robert Hijmans
  • 10,683
  • 25
  • 35