GDAL: Set an entire band to nodata (make empty)?

Question

I have grayscale DEM that I've needed to convert to RGBA in order to play well with how blender addresses alpha channels. I'm using this script to create 2 more bands and then stick my alpha band at the end.

gdal_translate -of GTiff -co -b 1 -b 1 -b 1 -b 2 Filein fileout

It's working great, the only issue is that the file is bloated by about 2 gigs because it now has 2 bands with tons of data that are not used (blender is only pulling data from bands 1 and 4 but still seems to need them there). I'd like to clear out only bands 2 and 3 of all data to keep the files smaller. Any tips? I've looked through all the GDAL documentation and can't see a way to set only one band's values to nodata rather than applying an operation to the whole raster.

Best

Answer 1

I would rather start by preparing bands that can be combined directly into desired result instead of updating pixels of that 4 band image.

You need

• band 1 to be used as band 1
• band 2 to be used as band 4
• two dummy bands to be used as bands 2 and 3

You know already how to get the first two

gdal_translate -of GTiff -b 1 Filein.tif band1.tif
gdal_translate -of GTiff -b 2 Filein.tif band2.tif

A dummy band can be created with gdal_create utility https://gdal.org/programs/gdal_create.html. Notice that you need GDAL version 3.2 or newer. What you want is an image that takes most features from the source image, but has only 1 band with all pixels having a fixed value, let's say "0".

gdal_create -bands 1 -burn 0 -if Filein.tif dummy.tif

Finally you can combine the ingredients with gdal_merge https://gdal.org/programs/gdal_merge.html. Notice that you must have GDAL Python bindings installed for running gdal_merge because it is a python script. You want to take 4 input files and write them into one GeoTIFF, each input on a separate band.

Notice also that your assumption that updating a whole band into nodata would automatically save space. It does not because in an uncompressed TIFF every pixels requires the same amount of space. Fortunately a band with all equal pixels is compressing extremely well even with lossless compression methods like "deflate". Thus the gdal_merge command would look as follows. Notice that the dummy TIFF is selected twice.

gdal_merge -o 4bandtest.tif -of GTiff -co compress=deflate -co tiled=yes -separate b1.tif dummy.tif dummy.tif b2.tif

Check with gdalinfo proves that the result has 4 bands and bands 2 and 3 have just zeroes.

gdalinfo 4bandtest.tif -stats
Driver: GTiff/GeoTIFF
Files: 4bandtest.tif
...
Band 1 Block=256x256 Type=Byte, ColorInterp=Red
  Minimum=0.000, Maximum=255.000, Mean=63.804, StdDev=31.731
  Mask Flags: PER_DATASET ALPHA
  Metadata:
    STATISTICS_MAXIMUM=255
    STATISTICS_MEAN=63.803541618056
    STATISTICS_MINIMUM=0
    STATISTICS_STDDEV=31.731235967103
    STATISTICS_VALID_PERCENT=100
Band 2 Block=256x256 Type=Byte, ColorInterp=Green
  Minimum=0.000, Maximum=0.000, Mean=0.000, StdDev=0.000
  Mask Flags: PER_DATASET ALPHA
  Metadata:
    STATISTICS_MAXIMUM=0
    STATISTICS_MEAN=0
    STATISTICS_MINIMUM=0
    STATISTICS_STDDEV=0
    STATISTICS_VALID_PERCENT=100
Band 3 Block=256x256 Type=Byte, ColorInterp=Blue
  Minimum=0.000, Maximum=0.000, Mean=0.000, StdDev=0.000
  Mask Flags: PER_DATASET ALPHA
  Metadata:
    STATISTICS_MAXIMUM=0
    STATISTICS_MEAN=0
    STATISTICS_MINIMUM=0
    STATISTICS_STDDEV=0
    STATISTICS_VALID_PERCENT=100
Band 4 Block=256x256 Type=Byte, ColorInterp=Alpha
  Minimum=0.000, Maximum=255.000, Mean=68.127, StdDev=32.036
  Metadata:
    STATISTICS_MAXIMUM=255
    STATISTICS_MEAN=68.126588152778
    STATISTICS_MINIMUM=0
    STATISTICS_STDDEV=32.036317455231
    STATISTICS_VALID_PERCENT=100

Answer 2

Another approach than to prepare the 4 bands as ready to use is to update the pixel values of an existing GeoTIFF with gdal_rasterize https://gdal.org/programs/gdal_rasterize.html.

For updating a whole band user must have a polygon geometry that totally covers the image. Such polygon can be created for example with QGIS be digitizing and saving the result into GeoTIFF or any other vector format.

Usage:

gdal_rasterize -b 2 -b 3 -burn 0 big_polygon.json 4bandimage.tif