[GRASS5] r.resamp.rst

[Helena]

There is an explanation for the behavior of rst for lat long.

lat long coordinates are angles not distances and rst is not
designed to handle that. The distance in rst is computed
as r**2=(xi-xj)**2+(yi-yj)**2. If you are in a location where
you don't need the wrap-around and the distance is small this will work
(but obviously not for longer distances).

Expressed as decimal degrees, the values are extremely small,
rst thinks that it is in meters so the size of segment that is usually
around hundreds of meters to kilometers is now around cm, the normalization
is not working properly and distances fall into the interval where the approximation
function used in rst is much more complex and takes much longer to compute.

So in general, rst as it is now should not be used for lat/long data
(or use it at your own risk).
It can be modified to approximatelly handle it by checking the PROJ
information and computing the real distances, but it would be much
better to include an option for interpolation on sphere using angles
(we have the math. formulation for that), but segmentation would be
a little bit more complex (with proper handling of poles and wrap around)
and it would be a good project for a thesis
rather than a quick fix. We have already looked
at that at CERL and as there weren't many people who needed it we
did not do it.

Helena



> On Mon, Apr 26, 2004 at 08:59:35AM +0100, Paul Kelly wrote:
> 
>>On Mon, 26 Apr 2004, Markus Neteler wrote:
>>
>>
>>>On Sun, Apr 25, 2004 at 10:57:34PM +0200, grass at intevation.de wrote:
>>>
>>>>Author: paul
>>>>
>>>>Update of /grassrepository/grass/src/CMD/lists
>>>>In directory doto:/tmp/cvs-serv1526
>>>>
>>>>Modified Files:
>>>>	GRASS
>>>>Log Message:
>>>>re-enable compilation of r.resamp.rst as bug may be fixed now
>>>
>>>Hi Paul,
>>>
>>>did you try it by chance in a LatLong location? I am currently
>>>facing problems with s.surf.rst in LatLong (Helena is looking into
>>>it), so the problem may apply also to r.resamp.rst.
>>
>>No but I can give some anecdotal evidence about s.surf.rst in a latlong
>>location: it seemed to take a very long time to run (much longer than
>>s.surf.idw) and kept outputting the warning about taking to long to find
>>points (despite them being very evenly distributed across the region), but
>>it got there in the end and gave results that looked fine.
>>
>>http://www.stjohnspoint.co.uk/gis/france.htm (brief mention of it half-way
>>down the page)
>>
>>Paul
> 
> 
> So the s.surf.rst seems to work well in latlong under certain
> conditions. I was using 'r.fill.null' script to close holes in SRTM
> data which lead to some matrix errors (keeping some holes).
> 
> 
> Markus
> 
> _______________________________________________
> grass5 mailing list
> grass5 at grass.itc.it
> http://grass.itc.it/mailman/listinfo/grass5