Hi Doug,
> 1) Any suggestions on why I am getting missing data from my resamplings?
> The original data seem to be OK but the grid is not particularly
> regular. (This is some MM5 output.) I defined the original domain as a
> Gridded3DSet by giving it a list of coordinates that aren't necessarily
> monotonically increasing. This field displays fine, but with 60*60*31
> wind vectors, the scene is a bit crowded (and slow to rotate). My
> resampling domain is a simple Linear3DSet (20x20x10). The result is
> mostly OK but there are enough holes (i.e. missing data) that attempts
> to do IsoContours are not pretty. As expected, NEAREST_NEIGHBOR doesn't
> leave as many holes as WEIGHTED_AVERAGE, but there are still plenty.
I ran some tests and believe that the cause of your missing
points is the nature of resampling and iso-surface computation
rather than a bug. When you resample from your Gridded3DSet to
a Linear3DSet, any sample locations of the Linear3DSet that lie
outside any "cubes" of the Gridded3DSet will have Field values
set to missing. Then, the iso-surface computation will not draw
in any "cubes" of the Linear3DSet that have a missing Field value
at any vertex. A low-resolution Linear3DSet will have large
"cubes", and some with corners outside the region of the original
Gridded3DSet will extend far into it and create holes.
A better approach might be to resample to a lower resolution
version of the original Gridded3DSet. Assume the original
Gridded3DSet and FlatField are constructed by:
float[][] samples = ...
float[][] values = ...
Gridded3DSet set = new Gridded3DSet(type, samples, nx, ny, nz);
FlatField field = new FlatField(ftype, set);
field.setSamples(values);
Then down-sample by an integer factor of nn, as follows:
int mx = (nx + nn - 1) / nn;
int my = (ny + nn - 1) / nn;
int mz = (nz + nn - 1) / nn;
float[][] new_samples = new float[3][mx * my * mz];
float[][] new_values = new float[1][mx * my * mz];
int m = 0;
for (int k=0; k<mz; k++) {
for (int j=0; j<my; j++) {
for (int i=0; i<mx; i++) {
int new = i + mx * (j + my * k);
int old = i*nn + nx * (j*nn + ny * k*nn);
new_samples[0][new] = samples[0][old];
new_samples[1][new] = samples[1][old];
new_samples[2][new] = samples[2][old];
new_values[0][new] = values[0][old];
}
}
}
Gridded3DSet new_set
new Gridded3DSet(type, new_samples, mx, my, mz);
FlatField new_field = new FlatField(ftype, new_set);
new_field.setSamples(new_values);
There won't be any missing values (unless there were in the
original field values). I haven;t run this code, so it may
have bugs. But its the basic idea.
By the way, in working on this problem I did find and fix a
bug in the iso-surface computation, that only occurred with
missing data.
Cheers,
Bill
----------------------------------------------------------
Bill Hibbard, SSEC, 1225 W. Dayton St., Madison, WI 53706
whibbard@xxxxxxxxxxxxx 608-263-4427 fax: 608-263-6738
http://www.ssec.wisc.edu/~billh/vis.html
