[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: spherical gridding problem




Does anyone know exactly how the spherical gridding in IDL really works?  There is mention under QUINTIC of a paper by Renka, and JD was kind enough to find me an different article by Renka in the ACM titled "Interpolation of data on the surface of a
sphere" (http://www.acm.org/pubs/citations/journals/toms/1984-10-4/p417-renka/)
though I can't be sure that is the method used by IDL.  If it is, I would suspect that the problem is somehow in the calculation of the gradients at the corners of the triangles.  Though to tell the truth, I only really followed the idea and not the actual
details.  Not that knowing the reason would solve my problem.

I guess for now, I'll take your idea Craig and try adding some random scatter in several passes, and try comparing the results to find problem areas.  I think I need several passes, because one random scattering is bound to turn up new problems in other
places.

-Jonathan

Craig Markwardt wrote:

> 
> Hi Jonathon--
> 
> The results of the output are definitely not right.  I can get the
> artifact to disappear or at least decrease by shifting that center
> point about 0.1 degrees in any direction.
> 
> However I think this may come back to a problem some people have been
> seeing regarding TRIGRID.  [ TRIGRID is the underlying routine of
> SPH_SCAT. ]  When passed data points that are colinear then TRIGRID
> actually crashes.  In spherical coordinates the problem must manifest
> when points lie on a great circle.  I can imagine that if points are
> very *nearly* colinear then some kind of cancellation error occurs,
> which might give you the blow-ups you are seeing.
> 
> That center point appears to be at a crossing of two sets of nearly
> colinear points, so that may indeed be a problem.  However, I have to
> admit that there are a lot of other points like that.
> 
> How to deal with it?  Beats me.  This is really something that the RSI
> people should try to fix.  You could test for the error by putting
> some random scatter in your input lat/lon points and looking for major
> deviations in the result.
> 
> Sorry I can't help more,
> Craig
> 
> --
> --------------------------------------------------------------------------
> Craig B. Markwardt, Ph.D.         EMAIL:    craigmnet@cow.physics.wisc.edu
> Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
> --------------------------------------------------------------------------

-- 

-Jonathan