How to create a raster from NoData values using GDAL?

Question

I have 1,000 very large single-band rasters and many of them have nodata values. I want to take those nodata values and create a binary map to visualize all of the areas where I do not have data coverage.

Even after researching for quite awhile I don't really understand alpha bands/masks, where they exist, how to examine them. I thought I could just add an alpha band in GDAL but when I export the band I just get a raster full of 255s (checking with np.unique(tif))

# Adds alpha band
gdalbuildvrt out_vrt src_dir/*.tif -b 1 -vrtnodata -9999 -addalpha  

# Deletes the first band, leaving just the alpha
gdal_translate in_vrt out_tif -b 2

My next idea was to use gdal_calc to create a binary tif where nodata=1 and data=0

# Both of these attempts left me with just a raster of zeroes
# Attempt #1
gdal_calc.py -A src_tif.tif --A_band=1 --calc="(A<=-9999)" --outfile out_tif.tif --co compress=Deflate --type=Byte

# Attempt #2
gdal_calc.py -A src_tif.tif --A_band=1 --calc="(A*0)" --NoDataValue=1 --outfile out_tif.tif --co compress=Deflate --type=Byte

Any tips? This seems like it should be so simple, it's driving me crazy

Answer 1

If the aim is just to divide the image into two values, one for data and another for nodata and exact values are not so important, a short gdal_calc https://gdal.org/programs/gdal_calc.html command can do it.

gdal_calc -A src.tif --outfile=out.tif --type=Byte --calc="255*(A>-9999)"

The command creates a single band, 8-bit GeoTIFF because there is no need to keep the higher bit depth. Adding a creation option for compressed output --creation-option=compress=deflate will save a lot of disk space.

The calc expression is changing all pixel values that are greater than -9999 into 255 that is pure white. This stands for data. The remaining pixels (-9999 or less) are getting zero as pixel value because that seems to be the default for gdal_calc.py. The result is a binary image that can be checked visually unlike if it was created to use pixel values 0 and 1 which both appear visually black. If vector data are needed the image can be converted into vector with gdal_polygonize https://gdal.org/programs/gdal_polygonize.html.