Re: [ferret_users] Extrapolate over land.

[Martin Schmidt <martin.schmidt@xxxxxxxxxxxxxxxxx>]

Sangeeth,
which data are you using? If they are defined on a regular grid there 
are also grid points over land. I guess, the data are set to "missing" 
there. This may confuse your ocean model, since the "missing" may be 
interpreted as some valid number, say some really high wind speed. This 
could crash the model. If you interpolate such data onto your model grid 
the missing values have to be considered carefully.

So you may want to replace the missing values by some reaonable number, 
say 0, before you interpolate or read the data into your model. This is 
simple.

use your_file.nc

let wx = missing(windx,0)
let wy = missing(windy,0)

save/file=your_new_file.nc/clobber wx, wy

I expect the data set is large.  So you may want to save time step by 
time step or allocate more memory:
sp rm your_new_file.nc
repeat/l=1:some_large_number  save/append/file=your_new_file.nc wx, wy

This shows the principle. Getting the number of time steps is easy, just 
issue "show data". You may also read the maximum number of time steps 
from the data set, but I recommend to read through the manual and the 
tutorial anyway.

Recent versions of ferret write real*8 output. To save disk space you 
may specify

save/file=your_new_file.nc/clobber/outtype=float wx, wy

This may solve your problem. But I guess, your input data need 
interpolation to the model grid. Having zeros there, leads to reduced 
winds near the coast with a lot of consequences for near-shore currents.

So you need to make a decision what to do with coastal points:
- set the winds to some reasonable value for land points. How to, just 
described...
- extrapolate the ocean values into land, before you interpolate onto 
the model grid. ferret has nearest neighbour fillers @fnr but these are 
working only in one direction. See the manual. An alternative is 
fill_xy, which extrapolates into land by replacing land with averages 
over all valid neighbour points. Doing this repeatedly may give very 
smoothed extrapolation. Especially, value may never run away. This 
sounds difficult, but is simple. Use the coads winds as example

yes? use coads_climatology
! the function needs a mask, that is 1, where filling is allowed. Below 
we generate a field on the wind grid that is 1 everywhere.
yes? let mask = 0*missing(uwnd,0)+1
yes? let wx = fill_xy(uwnd,mask,1)                     ! fill in one 
cycle only

yes? shade /l=1 wx

shows the wind field, where one coastal point is filled. Now you may 
play around with the number of cycles:
yes? let wx = fill_xy(uwnd,mask,5)                     ! fill now in 
five cycles
yes? shade /l=1 wx

I think this shows the ideas and I hope this helps to get you started.

Greetings,
Martin