Need help trying to solve a re-gridding problem

[P.R.]

Hi,
Im looking for help trying to devise a solution to the following problem:
I have a high-resolution lat/long dataset: 30second (0.008333333degrees)
resolution.

I want to create 2 re-gridded datasets from the hi-res data:
1) a low-res dataset that uses 'box averaging' to drop to a given lower
resolution (using regrid in grads)
2) a 'mask' grid, that indicates the percentage of each grid point in the
low-res grid that is above a specific threshold value, using a slightly
higher resolution as a reference.

Here is a more specific example of what Im trying to accomplish:
I'll call my 0.008333 hi-res data "grid_0.083"
Using regrid in grads, I create a 0.5 degree low-res data set from 'grid1'.
I'll call this 'grid_0.5'

Now, I want to create a 'mask' grid that has the same dimensions as
'grid_0.5', called 'grid_mask'.
'grid_mask' should indicate what percentage of each grid point in 'grid_0.5'
is above my threshold value of '-0.75'.
All points within 'grid_mask' that fall completely below the threshold
should simply have a 'zero' value.

However, 'grid_mask' should use a higher resolution than the source grid in
order to calculate the percentage...
i.e., if the source grid ('grid_0.5') has a 0.5 degree resolution, then
'grid_mask' should use a 0.125 resolution in order to calculate the
percentage values...

So, for example, if a given 'grid_mask' grid point has a value of 0.8, this
indicates that 80% of that specific grid point within 'grid_0.5' has a value
above the threshold '-0.75' value, and using a higher, 0.125 degree
resolution as reference for the 'sub-grid' percentage...

And, if a given 'grid_mask' point has a zero value, '0.00', then it means
that the related grid point within 'grid_0.5' has

I hope these examples made sense.

Can someome please help me devise a method of creating this mask?
I believe that it could be done using multiple calls to 're' in grads, and
perhaps also aave, but Im not clear on how to best implement a solution...

Please help,
Thanks,
P.R.