[gdal-dev] How to retain double precision in GDAL?

[Ole Nielsen]

Dear group

We are using GDAL - both command line and Python bindings -  for much of our modelling work. However, we find that the library wants to round floating point numbers down to single precision (Float32 or DataType=6) where what we need for scientific computing is double precision (Float64 or DataType=7). We have been digging around the documentation and the Python binding source codes, but are running out of ideas.

Please find attached a small Python script with an ASCII test grid that reveals the issue. Data which is originally in double precision gets rounded to 6 decimal places if read directly by Gdal and also if converted first to GeoTIFF using gdal_translate -ot Float64 ... .


The output of the attached script is


ASCII datatype: 6 float32
TIF datatype:   7 float64

Data GDAL (ASC): 50.814723968505859
Data GDAL (TIF): 50.814723968505859
Data REF:        50.814723686393002

Error (ASC):     0.000000282112858
Error (TIF):     0.000000282112858




Grateful for your help
Cheers
Ole


Dr Ole Nielsen
Numerical Modeller
Australia-Indonesia Facility for Disaster Reduction
Mobile: +62 811 820 4637 | Phone: +62 21 398 30088 x1007 | Fax: +62 21 398 30068




-------------- next part --------------
A non-text attachment was scrubbed...
Name: gdal_precision_test.tgz
Type: application/x-compressed
Size: 1377 bytes
Desc: gdal_precision_test.tgz
Url : http://lists.osgeo.org/pipermail/gdal-dev/attachments/20100806/89c8271b/gdal_precision_test.bin