Smoothing transformation and missing_value

[<hengkek@xxxxxxxxxx>]

Hi ferreters

I'm trying to perform some smoothing transformation (provided by Ferret)
on my gridded oceanographic data (to produce smoother contours) with
unexpected results. The problem is, this transforms some originally
valid datapoints to become "missing_value" points, namely around the
land/ocean boundaries.  The "missing_value" in the original dataset is
set to zero (0.0f) and denotes land.

I took the easy way: I used the ferret gui and clicked on the buttons
under "Transform" and tried all of the available schemes (boxcar,
binomial, etc).  I modified the .jnl file to do the following:

LET W_S = W_T[X=@SBN:3,Y=@SBN:3]
use all3.nc  ! unsmooth dataset
save/file=all_smooth.nc W_S

It appears that the transformation is taking into account the grid
points marked as "missing_value" in the original netCDF dataset.

How do I prevent grid points marked as "missing_value" from being
included in the transformations?
Or how do I prevent originally valid data from being transformed into
"missing_value" or "_Fillvalue"?
Is there another way to smoothen my data without the above limitation?
Is my use of zero for "missing_value" in the original dataset
problematic?

Thanks for any hints.

-chk
Physical Oceanography
Singapore.