gridSample: return spatialPoints object rather than matrix?

[plantarum]

I'm using gridSample to thin my occurrence records. The records are stored as a spatialPointsDataframe, which already includes environmental values as columns in the dataframe. gridSample returns a matrix with the coordinates of the retained occurrences, but drops all the other data columns. Is it possible to thin the data in a way that retains the non-coordinate columns? Alternatively, is there a way to use the matrix returned by gridSample to subset the original spatialPointsDataframe?

Thanks

[plantarum]

The following function does what I need.

Based on your original code for gridSample. I add an index to the spdf, and copy that over to the coordinates matrix xy. After the same processing you use to trim the records in xy, i use the index values that remain in xy to select desired points in spdf, then return that subset objects.

I also tried modifying the spdf object directly without extracting coordinates, but binding spatial data frames together in the for loop is much slower than binding matrices: 40 seconds vs 2 seconds in my tests.

If this looks reasonable and generally useful, I could prepare a pull request. Or if you know a more direct way to do this with existing functions, please let me know!

gridSampleTWS <- 
  function (spdf, r, n = 1) 
{
  ## add a new column with a unique index value for each
  ## record: 

  spdf$RECNUM <- 1:length(spdf)

  ## extract coordinates along with index:
  xy <- cbind(coordinates(spdf), index = spdf$RECNUM)

  r <- raster(r)
  cell <- cellFromXY(r, xy)

  ## uc contains the numbers for all non-NA cells (no duplicates)
  uc <- unique(stats::na.omit(cell))

  ## add the cell numbers to the table of coordinates 
  xy <- cbind(xy, cell = cell, rand = runif(nrow(xy)))
  
  ## drop missing cells:
  xy <- stats::na.omit(xy)

  ## seems unlikely there'll be unique rows, when one column
  ## is a newly-generated random number?

  xy <- unique(xy)

  ## sort by our random number:
  xy <- xy[order(xy[, "rand"]), ]
  xy <- as.data.frame(xy)
  pts <- data.frame(numeric(), numeric(), numeric(),
                   numeric(), numeric())
  names(pts) <- names(xy)
  for (u in uc) {
    ss <- subset(xy, xy[, "cell"] == u)
    pts <- rbind(pts, ss[1:min(n, nrow(ss)), ])
  }

  ret <- spdf[spdf$RECNUM %in% pts$index, ]
  ret <- ret[, names(ret) != "RECNUM"]
  return(ret)
}