[Gdal-dev] gdalwarp interpolation question

[Chris Pennock]

To all you gdalwarp pros:

I'm experiencing some possibly odd results from
gdal_warp, and wondered if someone here could tell me
what the behavior *should* be.

When I warp a dem into a GTiff using -rcs for cubic
spline resampling, the edges of the resultant GTiff
are "rolled off". For example, on the east side of the
tiff, where I would expect the height data to go
immediately from say 1000ft to 0 (because there was no
dem data further to the east), I instead get, say a
point at 1000, then one at 500, then one at 0. So it
seems to be interpolating the edges of the dem data
dwon to 0. This makes it hard to merge these tiffs,
since little valleys occur betwen them.

On the other hand, using -rn (for nearest neighbor
resampling) causing no such rounding, so I can merge
the dems, but the resulting terrain is too jagged, due
to the -rn resampling.

So, my question is: Is such rolling off expected? Is
there some way to achieve nice resampling (like -rcs)
but to have a clear cut off at the edge of the
resulting tiff, to allow merging.

Many thanks in advance!
Chris

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com