I have a 32-bit GeoTIFF file I'm trying to convert to map tiles using gdal2tiles.py.

What the GeoTIFF is supposed to look like when loaded straight into QGIS (no scaling, styling, etc):

enter image description here

And here is the output of gdalinfo input.tif:

bash-3.2$ gdalinfo input.tif
Driver: GTiff/GeoTIFF
Files: input.tif
Size is 294, 192
Coordinate System is:
        SPHEROID["WGS 84",6378137,298.257223563,
Origin = (-83.400000000000006,35.200000000000003)
Pixel Size = (0.016666666666667,-0.016666666666667)
Image Structure Metadata:
Corner Coordinates:
Upper Left  ( -83.4000000,  35.2000000) ( 83d24' 0.00"W, 35d12' 0.00"N)
Lower Left  ( -83.4000000,  32.0000000) ( 83d24' 0.00"W, 32d 0' 0.00"N)
Upper Right ( -78.5000000,  35.2000000) ( 78d30' 0.00"W, 35d12' 0.00"N)
Lower Right ( -78.5000000,  32.0000000) ( 78d30' 0.00"W, 32d 0' 0.00"N)
Center      ( -80.9500000,  33.6000000) ( 80d57' 0.00"W, 33d36' 0.00"N)
Band 1 Block=294x3 Type=Float64, ColorInterp=Gray
  Min=0.005 Max=0.759 
  Minimum=0.005, Maximum=0.759, Mean=0.057, StdDev=0.076
  NoData Value=-1.69999999999999994e+308

What a sample tile looks like when converted using gdal2tiles.py input.tif:

enter image description here

OK, so clearly there's something going on with floating points and possibly number signing. So I try to convert it:

gdal_translate input.tif output.tif -ot UInt16

... And then convert to tiles:

Result: Only values over 0.5 carry over as black. Everything else is white. enter image description here

I feel like there is something obvious that I'm missing. Does anyone have any insights into what's going on here?

Things I've done:

-Try all the different convert options:

Byte: Monochrome conversion

Int16: All black

UInt16: Monochrome conversion

UInt32: Monochrome conversion

Int32: All black

Float32: TV snow noise

Float64: TV snow noise

CInt16: All black

CInt32: All black

CFloat32: TV snow noise

CFloat64: TV snow noise

  • If you could use gdal_translate, maybe this post will help.
    – mgri
    Jan 13, 2017 at 9:42
  • You say that there is no scaling in QGIS. However, the metadata informs that the image is Float64 type with Min=0.005 Max=0.759. If metadata is correct the un-stretched image should look all black. Are you sure that QGIS is not doing automatic LUT-stretch for you?
    – user30184
    Jan 13, 2017 at 9:55
  • 1
    What do you expect the output to look like? Should tiles retain the original values of the raster or should they be "ready to present" and scaled from min/max to 0/255? Jan 13, 2017 at 13:18

1 Answer 1


My guess is that you have a far out of range nodata value, like -9999, which is not set as the nodata value in the .tif. Then, when gdal_translate rescales your values from whatever into 0 to 255, you get one bin with the nodata at 0 and one bin with everything else at 255. So, you need to check what value your nodata pixels have and then set the nodata value in your .tif using gdal_edit -a_nodata -9999 or similar.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.