crop or mask large raster on inport

On Tue, 24 Nov 2020, Dexter Locke wrote:

Dear list,

Is there a way to reduce the spatial extent of an impractically-large
raster (~50Gb) on import? Can a crop or mask be applied as the large

raster

is read into R to create a more manageable subset?

rgdal::GDAL.open() simply opens the data source but does not read it,
returning an object with a handle to the open file - close with
rgdal::GDAL.close(). rgdal::getRasterData() has a complete set of
arguments for the kinds of things you mention, see:
https://rgdal.r-forge.r-project.org/reference/GDALRasterBand-class.html

The same arguments are used by rgdal::readGDAL(): offset, region.dim,
output.dim, band. These all permit rectangular cropping, and output.dim=
can decimate/resample/aggregate input cells to coarser output cells. These
are the same mechanisms that GDAL utility programs use. They are also used
by raster internally.

The example in rgdal::readGDAL() includes:

SP27GTIF <- readGDAL(system.file("pictures/SP27GTIF.TIF",
  package = "rgdal")[1], output.dim=c(100,100))

doing resampling on-the-fly. Lower down,

fn <- system.file("pictures/erdas_spnad83.tif", package = "rgdal")[1]
erdas_spnad83 <- readGDAL(fn, offset=c(50, 100), region.dim=c(400, 400),
  output.dim=c(100,100))

uses all of offset=, region.dim= and output.dim=. The example files are
tiny, but the same GDAL code applies to larger files. Note that the axis
ordering of the argument vectors is often counter-intuitive and needs some
patience.

The GDAL utilities can be run file-to-file using the gdalUtils package,
in addition to being run from the command line.

Otherwise, see the stars package and its proxy=TRUE read functionality,
which leaves the data on disk until required in possibly subsetted and
resampled form.

Hope this helps,

Roger

I've looked at raster and rgdal helpfiles, but have not found this
functionality. This would be akin to the sf::st_read() 's new "query"
argument that takes SQL statements for filtering records while reading

data

into R.

The ultimate use case is to summarize a large categorical raster into

many

thousands of polygons, and I thought reading small subsets in at a time
could be looped through or parallelized for summarizing smaller pieces of
the data. If there are other approaches to zonal functions (like
equivalents of ArcMap's Tabulate Area) that could also work.

Thanks for your consideration, Dexter

      [[alternative HTML version deleted]]

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
https://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en