[gdal-dev] Discrepancy when tiling on MAC and Linux

[Even Rouault]

On vendredi 8 novembre 2019 08:18:11 CET Grégory Bataille wrote:
> Hey all,
> 
> So I have a strange issue that I somewhat isolated, but I don't know how to
> move further.
> I have dataset in EPSG 31468-15949

I would check if that wouldn't be related to availability of the BETA2007.gsb 
grid. osgeo/gdal:alpine-normal-3.0.2 has it. Does your MAC install have it ?

That said, your TIFF file has a hardcoded TOWGS84=598.1,73.7,418.2,0.202           
,0.045,-2.455,6.7 . When using

gdalwarp Investigation_tile_alignment/ClippedProject.tif out.tif -overwrite -
t_srs EPSG:4326 --debug on

I see locally that BETA2007.gsb is used. This is not so bad, but a bit 
unexpected since I'd thought that forced TOWGS84 would have take the 
precedence.

Checking further in ogrct.cpp, I see that as the GeoTIFF has also the EPSG:
31468 code, and when instanciating the PROJ coordinate transformation, this is 
the preferred initializer, and thus the hardcoded TOWGS84 is ignored. So this 
explains what I observe. I'm not completely clear of what would be the desired 
behaviour here. In some use cases, one might prefer to honour the possibly 
suboptimal TOWGS84 from the geotiff info (the simple fact of doing 
gdal_translate -a_srs EPSG:31468 hardcodes it. Even with GDAL 3, which is 
probably no longer a good idea...). In other cases, just trust PROJ to use the 
best transformation method.

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com