[gdal-dev] Warping with GCPs at high latitudes

[Knut-Frode Dagestad]

Hi,

I am having trouble to perform accurate warping of images with GCPs 
(lon,lat) in the high Arctic (to "+proj=stere +lat_0=90 +lon_0=0"); 
problems starting already around 80 degrees of latitude.
Thin Plate Spline (-tps) performs better than polynomial fits, but 
output is often severely dragged at one or two edges, and sometimes it 
looks like the destination image is mirrored across a line.
Trimming images at the edges and adjusting warp options SAMPLE_STEPS and 
SAMPLE_GRID helps a little, but not sufficiently.

For the polynomial fits I assume the problem is related to precision due 
to the fact that lon and lat become singular at the pole. However, it 
should then probably be better if the algorithm could:
- convert first GCPs to the output projection (in my case stereographic, 
which has no singularity for the relevant images)
- make the polynomial fit GCP_stereo <-> pixel, line
- use this polynomial for the warping

I guess most GDAL users have a long wishlist of switches, but how about 
a switch to control whether GCP polynomial should rather be made in 
destination projection? Or a switch to gdal_translate to change the 
GCPProjection and GCP coordinates. A simpler and more pragmatic solution 
could be to make a separate Python script gdal_translate_gcps.py?

Are there other suggestion to deal with these problems?


Best regards from Knut-Frode