Exercise solution for raster format conversion

1. Selection of red, green, blue bands:

$ gdal raster select TDR.tif TDR_rgb.tif --band 1,2,3 --creation-option TILED=YES

or

$ gdal raster select TDR.tif TDR_rgb.tif --band red,green,blue --creation-option TILED=YES

2. Scaling to 8-bit with JPEG compressed tiled GeoTIFF output

$ gdal raster scale TDR_rgb.tif TDR_rgb_byte_jpeg.tif \
    --output-data-type uint8  \
    --creation-option COMPRESS=JPEG \
    --creation-option TILED=YES

3. Improving the visual result by clamping the input range to 2 standard deviations of the mean.

   First let's collect statistics:

   $ gdal raster info --stats TDR_rgb.tif
   

   [ ... snip ... ]
   Band 1 Block=256x256 Type=UInt16, ColorInterp=Red
      Minimum=0.000, Maximum=17143.000, Mean=1439.120, StdDev=582.568
   [ ... snip ... ]
   Band 2 Block=256x256 Type=UInt16, ColorInterp=Green
      Minimum=0.000, Maximum=17660.000, Mean=1469.245, StdDev=543.003
   [ ... snip ... ]
   Band 3 Block=256x256 Type=UInt16, ColorInterp=Blue
      Minimum=0.000, Maximum=18510.000, Mean=1324.608, StdDev=482.921
   

   So roughly all bands have a mean around 1400 and a standard deviation of 500, so using [1400 - 2 * 500, 1400 + 2 * 500] should roughly remove the 5% outliers.

   $ gdal raster scale TDR_rgb.tif TDR_rgb_byte_jpeg_clamped.tif \
       --input-min 400 \
       --input-max 2400 \
       --output-data-type uint8  \
       --creation-option COMPRESS=JPEG \
       --creation-option TILED=YES
   

   Note

   We have not yet marked the nodata value as being 0, so above statistics are not fully correct. We'll do that later.