[gdal-dev] how can a Geotiff occupy three times the disk space of an ASCII Grid?

[Even Rouault]

Le samedi 05 juillet 2014 20:35:45, G. Allegri a écrit :
> Given that an ASCII Grid doesn't provide a data type, maybe the driver
> choose the larger one, while the utilities guess it trying to fit the
> values into a smaller type?

No, GDAL does not do that sort of magic. Are you sure the GDAL version used by 
Python is the same as by the utilities ?

> 
> giovanni
> 
> 2014-07-05 20:31 GMT+02:00 G. Allegri <giohappy at gmail.com>:
> > I've debugged it, and it gets Float64 directly from
> > GDALGetRasterDataType( GDALRasterBandH ).
> > I've also tried opening it with Python GDAL and I get Float64.
> > gdalinfo tells me Float32.
> > I will try to send you a subset of it. It's not easy, as it is a heavy
> > ASCII Grid and I don't want to alter it.
> > 
> > giovanni
> > 
> > 
> > 2014-07-05 20:22 GMT+02:00 Even Rouault <even.rouault at mines-paris.org>:
> > 
> > Le samedi 05 juillet 2014 20:18:50, G. Allegri a écrit :
> >> > Finally I've reached the point: GDAL's driver is opening the ASCII as
> >> > Float64, consequently QGIS treats it this way.
> >> > I wonder why gdalinfo and gdal_translate treat it as Float32
> >> > instead....
> >> 
> >> If gdalinfo opens it as Float32, then it is Float32 for everybody. I
> >> guess QGIS probably promotes Float32 to Float64.
> >> 
> >> > giovanni
> >> > 
> >> > 2014-07-05 18:53 GMT+02:00 G. Allegri <giohappy at gmail.com>:
> >> > > The problem was simpler then it appeared: I didn't realize that QGIS
> >> > > output is Float64.
> >> > > Yet I don't know why QGIS chosed to use this data type...
> >> > > 
> >> > > giovanni
> >> > > 
> >> > > 
> >> > > 2014-07-05 17:56 GMT+02:00 Even Rouault
> >> > > <even.rouault at mines-paris.org
> >> > > 
> >> > > Le samedi 05 juillet 2014 17:25:48, G. Allegri a écrit :
> >> > >> > > QGIS is usually just calls to gdal, which makes this even more
> >> > >> > > mysterious.
> >> > >> > 
> >> > >> > Yes, in the end it uses gdal, but it chooses the way blocks are
> >> 
> >> read
> >> 
> >> > >> > and written.
> >> > >> > If I remember correctly geotiff final sizes may depend on block
> >> > >> > ordering and memory alignment.
> >> > >> 
> >> > >> In that instance, the target geotiff has a natural block dimension
> >> 
> >> which
> >> 
> >> > >> is a
> >> > >> raster line.
> >> > >> If QGIS writes the geotiff by 256x256 blocks (this is just a guess.
> >> 
> >> I've
> >> 
> >> > >> not
> >> > >> verified), the same raster line will be written several times. As
> >> > >> it
> >> 
> >> is
> >> 
> >> > >> a compressed geotiff, the resulting raster line will be each time
> >> 
> >> being
> >> 
> >> > >> bigger
> >> > >> since the initial zeros will be replaced by actual values. And if
> >> > >> the new size
> >> > >> of the line is bigger than its previous size, the new line will be
> >> > >> rewritten
> >> > >> at the end of the file, losing the space previously occupied.
> >> > >> 
> >> > >> > Maybe this is the case, QGIS raster provider not doing the best
> >> > >> > at this level? Don't know, but this discussion is for the QGIS
> >> > >> > ml ;)
> >> > >> > 
> >> > >> > giovanni
> >> > >> > 
> >> > >> > > On 7/5/2014 9:09 AM, G. Allegri wrote:
> >> > >> > > > I agree with you David, I'm surprised too.
> >> > >> > > > Anyway, gdal_translate is run without compression options.
> >> > >> > > > I've written to the QGIS devs (it was the software) to verify
> >> > >> > > > what's happening with its raster file writer code...
> >> > >> 
> >> > >> --
> >> > >> Geospatial professional services
> >> > >> http://even.rouault.free.fr/services.html
> >> > > 
> >> > > --
> >> > > Giovanni Allegri
> >> > > http://about.me/giovanniallegri
> >> > > Twitter: https://twitter.com/_giohappy_
> >> > > blog: http://blog.spaziogis.it
> >> > > GEO+ geomatica in Italia http://bit.ly/GEOplus
> >> 
> >> --
> >> Geospatial professional services
> >> http://even.rouault.free.fr/services.html
> > 
> > --
> > Giovanni Allegri
> > http://about.me/giovanniallegri
> > Twitter: https://twitter.com/_giohappy_
> > blog: http://blog.spaziogis.it
> > GEO+ geomatica in Italia http://bit.ly/GEOplus

-- 
Geospatial professional services
http://even.rouault.free.fr/services.html