[GRASS-user] r.resamp.rst dynamic tension?
hamish_b at yahoo.com
Wed Mar 30 19:21:28 EDT 2011
I am working with some u,v component wind data on a ~30 km grid which I
wish to resample to a 3 km grid before r.what extractions at various
fixed points. Often the winds have smooth transitions, so the small
smoothing towards the mean that the spline does is not really a problem.
Results so far with overlap=5 and tension=10 are very very nice.
Stunning actually.. most of the time.
One thing I am worried about though is that it will smooth out the peaks,
which can be very important and who's effects are non-linear - e.g. in
fronts or the eye-wall of a hurricane. Conserving the area under the
curve is not enough, it must preserve the peaks as well (due to dramatic
effects of velocity^2, etc.).
the r.resamp.rst module has a smooth=raster_map option, I thought to
maybe take the r.slope.aspect slope (1st deriv) of the overall magnitude
map and apply it as dynamic smoothing, so calm areas allow a relaxed fit
while areas of great change pull tighter. I guess that's more dynamic
tension than dynamic smoothing though.
the manual says:
For noisy data it is possible to define spatially variable
smoothing by providing a raster map named by the smooth option
containing smoothing parameters. With the smoothing parameter set to
zero (smooth is not given or contains zero data), the resulting
surface passes exactly through the data points.
which sounds like it should already try to pass through outliers exactly
as long as I don't use a smooth= map? As mentioned above, currently I'm
using tension=10 and overlap=5 to get rid of outliers, probably
introduced from an earlier resampling..
a comparison of r.univar mean and std.dev. shows a slight reduction in
overall range in the RST output map, but still closer to the original
than a r.resamp.interp bilinear or bicubic interpolation.
Losing the localized peaks due to grid aliasing is another concern, but
there's less I can do about that AFAICT.
More information about the grass-user