gdal_translate fails to update nodata tag on convert 32bit to 16bit

[maphew]

When converting an image from 32bit float to 16bit signed integer the nodata tag from the source image is retained. The nodata values within the raster are converted from -3.40282346638529e+38 to -32768. but not in the header. Example:

gdal_translate -ot int16 my32bit.tif my16bit.tif

gdalinfo -mm my16bit.tif
...snip...
Band 1 Block=1800x2 Type=Int16, ColorInterp=Gray
    Computed Min/Max=-32768.000,5973.000
  NoData Value=-3.40282346638529e+38

[maphew]

source and output images

[maphew]

This is using GDAL 1.4.4.0, released 2007/11/23, on ubuntu 8.04

[Even Rouault]

As a workaround, you can add "-a_nodata -32768" to your gdal_translate command line

[warmerdam]

Matthew,

This is the intended behavior. I agree it can be surprising and confusing but the automatic alternatives get very complicated and I'm not prepared to pursue them at this time. Possibly someone else might feel differently.

[maphew]

I appreciate an automatic approach can quickly get too complicated for an issue relatively minor compared to the rest of gdal development. Perhaps a warning could be issued?

Personally, now that I've tripped over it and know it's there I'll know to step around in future. A warning might save someone else the trouble of trying to figure out what happened.