r.fillnulls needs a little patch to work, it is attached against
current CVS head.
No, the problem was r.buffer was expecting meters as input, but when in
a lat-lon location, the resolution as given by 'g.region -g' is given in
degrees. A resolution of 3 arcsec as from STRM data gives a resolution
of 0.00083333 which rounds to 0.00 in that awk command. This results in
'r.buffer distance=0.00' which doesn't work.
look for a fix in CVS soon.
.. I can do a test for proj=ll then multiply by 1852*60, but it might be
better to use "meters" from PROJ_UNITS in 'g.proj -p'? (which happens to
be 1.0?!)