hunzikp
commented
7 years ago You're confusing two concepts: The storage mode of the data, and the datatype attribute of the raster object. velox typically preserves the storage mode of the data, even if you cast to a matrix or a raster object. To continue your example:
mat <- matrix(1:9, 3, 3)
class(mat[1,1])
# "integer"
vx <- velox(mat, extent=c(0,3,0,3), res=c(1,1), crs="+proj=longlat +datum=WGS84 +no_defs")
class(vx$as.matrix()[1,1])
# "integer"
class(as.matrix(vx$as.RasterLayer())[1,1])
# "integer"The raster package uses float32 as the default data type (unless the default was changed using raster::rasterOptions), regardless of the storage mode of the actual data. E.g.
mat <- matrix(rep(1L, 9), 3, 3)
storage.mode(mat)
# "integer"
ras <- raster(mat)
dataType(ras)
# "FLT4S"Consequently, the fact that the raster object you get from vx$as.RasterStack() has data type float32 is due to the raster package, rather than velox. Nonetheless, I've implemented an extra argument that allows assigning appropriate data types to raster objects created from velox rasters:
mat <- matrix(1:9, 3, 3)
vx <- velox(mat, extent=c(0,3,0,3), res=c(1,1), crs="+proj=longlat +datum=WGS84 +no_defs")
ras <- vx$as.RasterLayer(assign_data_type = TRUE) # New argument
dataType(ras)
# "INT1U"Note, however, that the raster::writeRaster function doesn't automatically respect the datatype attribute. You still need to do
writeRaster(x = ras, filename = "/foo/bar.tif", datatype=dataType(ras))Alternatively, you could just use vx$write(), which always uses the smallest possible data type when writing to disk.
When transferring data from Velox to Raster package the data type is not maintained. E.g. integer values become floats. This increases the file size of the resultant raster.
Simple example: