a bug in @modvar function?

[Muyin Wang <muyin@xxxxxxxxxxxxx>]

Dear Ferreters:
 I am trying to compute the variance for a time series. The @modvar
transformation works well for the grid points when there were no missing data. 
However, if there is a missing point in the time series, I ran into trouble.
Here is a sample of what I got 


yes? let skv = sktemp[gt=month_reg@modvar]
yes? list/l=1:12 skv
             SKTEMP[GT=MONTH_REG@MODVAR]
             Y: -0.03
             DATA SET: /home/aegir2/muyin/cdf/tovs_monthly_80-98.cdf
                      -0.03    
                       33
 16-JAN      /  1:  5.910E+08
 15-FEB      /  2:  5.710E+00
 17-MAR      /  3:  8.167E+00
 16-APR      /  4:  1.661E+01
 16-MAY      /  5:  2.528E+00
 16-JUN      /  6:  1.534E+00
 16-JUL      /  7:  6.520E+00
 16-AUG      /  8:  5.164E+00
 15-SEP      /  9:  3.358E+00
 16-OCT      / 10:  6.217E+00
 15-NOV      / 11:  7.806E+00
 16-DEC      / 12:  4.575E+00


It is obvious that the result for Jan is not right. Then I used the @modngd
transformation, and found for Jan, the answer is 18, well for the rest of the
months is 19.  So if I specifically get rid of the bad values by using @sum,
@ngd together, I can get a correct answer, which is 7.017 in this case. I was
surprised to find this as I thought Ferret will not include the bad values in
the calculation, whici is true for the functions like @sum, and @ave, etc... Why
not in @modvar?
Is there a way to solve this easily?
Thanks for the help!
Muyin