Thanks, Craig: That worked magically!
--Doug
[email protected]
Software Engineer
UCAR - COSMIC, Tel. (303) 497-2611
On Mon, 9 Dec 2013, Craig DeForest wrote:
(As always, I hit "send" a moment too soon...)
That problem is a good example of the maxim that "when in doubt, add
more dimensions"... :-) Interpolators are particularly confusing
because they have to have a different number of active dims in different
parameters/terms -- hence Chris's recent example, and the
I-hope-now-well-known wart with index() threading...
On Dec 9, 2013, at 9:44 AM, Craig DeForest <[email protected]> wrote:
Hmmmm..
If I understand right, you've enumerated the altitudes in lgrid and hgrid.
Since you're interpolating over altitude, you want that moved to the pole
position in the dim list in your $x and $y for your dplinear call. Everything
else can be left alone. You want:
* active dim: lgrid-altitude
* thread dims: lat, lon, and hgrid-altitude
After that, it's easy:
$hg= Dutil::dplinear( $lgrid->mv(2,0), # (n: lo-alt, t0: lat, t1: lon)
$t->mv(2,0), # (n: lo-alt, t0: lat, t1: lon)
$hg # (t0: lat, t1: lon, t2: hi-alt)
);
I named the dimensions of each term, for clarity: 'n' is the active dim in the
signature, the 't' dimensions are the thread dimensions. Note that $hg (the
'in()' term) is missing the active dim 'n', as the signature requires.
On Dec 9, 2013, at 8:56 AM, Doug Hunt <[email protected]> wrote:
Hi PDL folks: I just wrote a function that I think should be thread-able, but
I can't think how. Any ideas welcome.
This function is used to up-sample a weather model grid via linear
interpolation. I use the local PP function Dutil::dplinear to do this. It may
be better to use other PDL interpolation functions, but I've been using this
for years and I know just what it does. Here is its signature:
dplinear (x(n); y(n); in(); [o]out())
It is just a simple interpolation routine which takes $x and $y vectors as
input, along with one value in $x-space to interpolate into $y space.
The tricky bit here is that two sorts of threading are necessary:
1) Interpolating a column of input temperatures via the $x and $y input
to dplinear.
2) Threading over all the lat/lon columns in the input temperature grid $t.
The tricky bit is that the pressure values are different for each column--this
model does not use standard pressure levels, but the pressure
levels at each grid point depend upon the surface pressure.
Any ideas? Please let me know if this is not clear...
Regards,
Doug Hunt
--------------------------------------------------------------------------------------
sub upsampleGrid {
my $lgrid = shift; # low resolution (half) pressure grid: lat x lon x 92
levels
my $hgrid = shift; # high resolution (half) pressure grid: lat x lon x 138
levels
my $t = shift; # temperature grid to upsample: lat x lon x level
my ($nlat, $nlon, $llvl) = $t->dims; # llvl = 92
my $hlvl = ($hgrid->dims)[2]; # 138
my $ht = zeroes($nlat, $nlon, $hlvl); # high resolution output grid
# This could be sped up with threading, I feel sure...
for (my $lati=0;$lati<$nlat;$lati++) {
for (my $loni=0;$loni<$nlon;$loni++) {
my $x = $lgrid->(($lati),($loni),:);
my $y = $t->(($lati),($loni),:);
my $in = $hgrid->(($lati),($loni),:);
$hg(($lati),($loni),:) .= Dutil::dplinear($x, $y, $in);
}
}
return $hg;
}
--------------------------------------------------------------------------------------
[email protected]
Software Engineer
UCAR - COSMIC, Tel. (303) 497-2611
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl
