Re: [ferret_users] regridding etopo5

[mick spillane <mick.spillane@xxxxxxxx>]

Hello All,

   Here is an idea that might work for extracting the median of water 
depths (ie excluding land) for a series of grid boxes without the need 
for an external function. It is based on SORTI and SAMPLEI.  I'll 
illustrate the idea for one grid box 170W-166W, 53-56N.

! Define the data file and how to select "water" points
use etopo05 ; set mem/size=500
let sea=if(rose[d=1] lt 0)then rose[d=1]
! count the number of etopo05 water points that fall in the new grid box
let nsea=sea[x=@ngd,y=@ngd]

! ... and define a file to accept the median values
sp touch median.file

!Step 0. define the corners of the grid box desired ... for example
let my_xlo=190 ; let my_xhi=194 ; let my_ylo=53 ; let my_yhi=56

!--------------------------------------------------
!Step 1.  For a particular grid box write out the water depths with a 
flag 999 for land
region/x=`my_xlo`:`my_xhi`/y=`my_ylo`:`my_yhi`
list/nohead/clob/file=temp.file/form=(f10.2) missing(sea,999)
let ngood=`nsea` ; let midpoint=int(ngood/2)+1
can region/all

!Step 2. Read back the water depths and sort in increasing order
file/form=free/var=depth temp.file ; set var/bad=999 depth
let index=sorti(depth) ; let sorted=samplei(depth,index)

!Step 3. Pick off the mid point
let median=sorted[i=`midpoint`]

!Step 4. List the result (to a file) then go back to Step 0 for the next 
grid cell
list/app/nohead/form=(f10.2)/file=median.file median
can data 2 ! eliminate the temporary file
!----------------------------------------------------

The results check out: there are 1769 "water" points among the 1813 
etopo05 cells within the box.
The mid point is #885 and the 885th of the sorted depths (i.e the 
median) is -951m.

To create a whole grid of boxes a pair of nested loops can define the 
my_xlo,my_xhi, my_ylo and
my_yhi values in turn.  In the inner loop a .jnl file containing steps 
1-4 above is called to incrementally write the median values.  After 
this is done the ASCII file of medians is read into the new grid and the 
result output to a netCDF file.

Hope this helps,
Mick
------------------------
David Wang wrote:
> Hi Ansley,
>
> Thanks. That is supposed to work also. But I think that taking medians 
> is ideal for those adjacent to coastlines when a coarser grid box 
> needs to be determined as a wet box or a dry box.
>
> David
>
> Ansley Manke wrote:
>
> > Hi David,
> > You could also do another regridding operation from etopo5, using 
> > linear or averaging to regrid to your coarser grid.
> >
> > Ansley
> >
> >
> > David Wang wrote:
> >
> > > Hey Ansley,
> > >
> > > Thanks very much for your response. I don't think I have time 
> > > working on an external function right now. maybe some time in the 
> > > future, then I will contact you.
> > > This was to generate the bathymetry file for an ocean model. But 
> > > since currently I actually don't need a very accurate representation 
> > > of the sea floor and the coastlines, I may still use my present 
> > > crude one derived from Levitus.
> > >
> > > Cheers,
> > > -d.
> > >
> > > On 5/30/06, *Ansley Manke* <Ansley.B.Manke@noaa.gov 
> > > <mailto:Ansley.B.Manke@noaa.gov>> wrote:
> > >
> > >     Hi David,
> > >     I think this would require an external function. You'd pass in
> > >     the data
> > >     on the finer grid as one argument, and a second argument would be
> > >     some
> > >     dummy variable on the desired output grid.  Then the function
> > >     would need
> > >     to determine which grid points from the fine grid lie within the 
> > > grid
> > >     boxes of the output data, and compute the median value, assigning
> > >     it to
> > >     the "result" of the function, on the coarser grid.
> > >
> > >     There would be some things to decide. Would you weight input data
> > >     according to what portion of the input grid cell lies within the
> > >     output
> > >     grid cell?  That is, if you had the fine grid and the output grid
> > >     lining
> > >     up like this, with data at the o's and grid cell edges at the |'s,
> > >
> > >     |---o---|---o---|---o---|---o---|---o---|
> > >         
> > > |----------O----------|----------O----------|----------O----------|
> > >
> > >     then, would data at the first o in the fine data, which lies 
> > > right on
> > >     the edge of an output grid cell, be weighted the same as the next
> > >     two
> > >     data points which lie entirely within the output grid cell? or
> > >     would its
> > >     contribution to the value of the median be smaller?  (and similar
> > >     considerations due to working in 2 dimensions.)
> > >
> > >     Another consideration would be whether to apply the cosine latitude
> > >     correction when working on a latitude-longitude grid.
> > >
> > >     If you (or anyone!) would like to work on this, I can give you more
> > >     ideas and help you get started.
> > >
> > >     Ansley
> > >
> > >     David Wang wrote:
> > >
> > >     > Hello Ferreters,
> > >     >
> > >     > I'm thinking of regridding etopo5 bathymetry data onto a much
> > >     coarser
> > >     > grid (say 0.5x0.5) in such a way that the regridded value at
> > >     each new
> > >     > grid point is chosen to be the median of all etopo5 grid points
> > >     within
> > >     > each new grid box. Are there any tricks for doing this in FERRET?
> > >     >
> > >     > Thanks,
> > >     > - d.