[gdal-dev] Re: Strange things with gdalwarp ...

[Hermann Peifer]

Robert Coup wrote
> On Sat, Aug 15, 2009 at 10:57 PM, Hermann Peifer <peifer at gmx.eu> wrote:
> 
>> I have the same observation while working with ASTER GDEM tiles (1x1 degree
>> tiles, 3601x3601 pixel each). When warping/merging, say: 10 tiles into a
>> single outfile.tif, then it takes gdalwarp around 5 seconds per tile to do
>> the job:
> 
> 
> Does creating a tiled tiff (-co TILED=YES) make any difference? It should
> improve random access, which is what writing new data to image blocks is.
> 

Thanks for the hint, but I didn't make any substantial difference, as far as I could see.

Basically, my task is to warp/resample/merge some 1500 ASTER GDEM tiffs into a single output file in LAEA projection. My area of interest is around 5000 x 4000 km, i.e. my 100m output raster would be 50000 x 40000 x 16 bit/pixel = 4GB GeoTiff.

I thought this should be a doable task and would perhaps take some hours or so. But over the last days, I have been experimenting quite a lot with -wm, --config GDAL_CACHEMAX settings, etc. without getting anywhere apart from the insight that it would take 10 days and more to do the job. I have a vague feeling that I am overlooking something obvious.

A promising approach seemed to be to warp/resample the ASTER files one by one and create 1500 separate tiles in LAEA projection, as an intermediate step. Then I created a vrt file with gdalbuildvrt and gdal_translated it into tiff format. The result did however look odd, as the elevation data in most tiles was clipped by NODATA areas of their neighbouring tiles :-(

Hermann