| triangulating over undefined space in irregular grids [message #59446] |
Wed, 26 March 2008 08:45  |
bjelley
Messages: 11
Registered: November 2006
|
Junior Member |
|
|
Hi folks. Longtime lurker, but now I have a challenging (to myself
anyway) problem to post.
I have a need for some more details about triangulate than I can seem
to find.
The setup: I am trying to contour plot a number of variables for a
storm surge grid that has up to 10 km resolution in the open Atlantic,
yet has 8 meter resolution in New Orleans area waterways. The grid
obviously includes most anything at or below sea level for the
Northwest Atlantic domain, but also includes many (200k or so) land-
based grid points (or it would be useless as a storm surge model). I
mention the land-based nodes to express that a coastline mask will not
fix the problem.
The problem: When plotting the results using the triangles returned,
the result includes vast areas that are outside of the model domain.
For example, the plot shows data in trangles running from central
Louisiana to New England, yet the grid does not include any land with
an elevation greater than 20 meters. So the grid does include some
land for a range of distances from the coast, from ~5 to ~100 km
inland based on elevation.
For example, when plotting the wind field from a hurricane, the
magnitude plot ends up with streaks from higher wind speeds near the
storm stretched out along the triangles towards areas not
impacted...such as New England during Katrina. Clearly not acceptible.
More about the grid: Areas of Florida are excluded and should be null,
but of course with a peninsula like that the triangles cover it. Cuba
and the Yucatan landmasses are not included in the grid. Most of the
land based nodes are along the Northern Gulf of Mexico. And I must
reiterate, it is not that the triangles are including data where the
data is null, but that the triangles are extrapolating data where the
grid is undefined.
Also of importance is that the grid is numbered and ordered in a
counter-clockwise fashion, not in any order of south to north and the
like. This is native to the model which uses a finite element method
for calculations allowing us to successfully resolve conservation of
momentum, velocity, etc at very high resolutions in areas of interest
while maintaining the ability to carry out computation at lower
resolutions over larger areas where mass conservation is
required...such as the Gulf of Mexico. Could I reorder for purposes of
plotting? Sure, but do I need to and where would it get me?
I have been through Dr. Bowman's book, Liam Gumley's book, David
Fanning's site, the astro site, the online help, and the German
library of IDL routines with no light shed on the solution.
I have fiddled with the tolerance variable passed to triangulate which
has not changed anything until I make it too large at which time...
% TRIANGULATE: Points are co-linear, no solution.
Trial and error isn't fun as the wind file for this is 18 GB and the
spherical method for 700,000 highly irregular points takes a very long
time (hours) on top flight equipment (3.0GHz Xeons) and I have yet to
get good results from it.
Can anyone help with a successful determination of what tolerance
should be set to? Or is that not the answer?
Any other ways to plot a very irregular grid? Especially interested in
one much faster than spherical triangulation. Regular triangulation is
fast enough, but...see above.
Any pointers would be very welcome and appreciated at this point.
Cheers,
-Ben
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
