Using signed Bytes with GDAL

Question

In the list of raster pixel data types that are available with GDAL, the "smallest" types are Bytes, UInt16 and Int16. I would like to store integers from 0 to 100 in the smallest image type, but with the constraint of using -1 and -2 for NoData value and other specific value. GDAL Bytes are unsigned, so the smallest type would be Int16. http://www.gdal.org/gdal_8h.html#a22e22ce0a55036a96f652765793fb7a4

I use this type to create rasters, with the Rasterize function in Python bindings.

Do you know if there is a trick to use signed Bytes ?

Answer 1

If you want to use signed 8bit integers for GeoTIFFs use datatype Byte together with creation options PIXELTYPE=SIGNEDBYTE

https://gdal.org/drivers/raster/gtiff.html

gdal_translate -ot Byte -a_nodata -1 -co PIXELTYPE=SIGNEDBYTE input.tif output.tif

Answer 2

This is an interesting problem. To my knowledge, GDAL does not support interpretation of signed 8-bit data. Even raster formats that can handle signed bytes SDAT, for instance are coerced to handle GDT_Byte I think...

The obvious solution to your underlying problem is to use a nodata value like 255 if you really care about the file storage size. Otherwise use a signed 16-bit image and some compression if the format supports it. You may find it plenty suitable.

• gvatin: Thanks a lot for your answer. I actually used a 16bit image, but due to the large size I guess we'll switch to Bytes with 255 for NoData, which is the simplest way. Anyway, it would be an interesting feature for future of GDAL to have signed Bytes !