Re: Regridding in space and time

[Chris Konig <konigc@xxxxxxxxxxxx>]

Jaison, (and Ferreters)

Thank you very much for the help, it's most useful.

The script how it is now, depends on the fact that I have an identical 
number of locations for each time step. For the actual data that I have, 
this is not true though.

Is it possible to not specify the length of the x-axis ('xscat' in 
Jaison's script below) or to just make it larger than anything I'll 
encounter, and then start 'filling the next time slot' as soon as the 
time stamp changes?

That probably would require some if-statements (?).


Ansley,

So do I assume correctly that there is no (for example) nearest 
neighbour or linear interpolation for 2D fields? I.e. I will have to use 
scat2gridXXX.

I don't mean to imply that this is bad, I just wondered if I had more 
options than Gaussian and Lagrangian interpolation.

Thanks again to everybody!

Chris



Jaison Kurian wrote:
> Hi Chris,
>            You reached there almost...but missed few points. I modified 
> your script (see the attachment:scat2grid.jnl, using the data 
> sample_data.txt)  to get proper time dimension for the variable based on 
> following details...
>
>      1. Your data is repeating for the common X-Y locations for 5 time
>           points ( 1990- 3/31, 4/30, 5/21, 6/14 &  6/21). But the lat-lon
>           point "80.50  -67.75" exists only for "1990 03 31". If this  
>           point is skipped we are left with 107 spacial points and 5 time
>           points ==> 107*5 = 535 values (or lines in the data file). 
>           Now read the variable in Ferret like...
>
>    define axis/x=1:107:1 xscat        ! spatial points  
>    define axis/t=1:5:1   tscat        ! time points
>    define grid/x=xscat/t=tscat gscat           
>    file/grid=gscat/var=..............
>
>     2. to define a time axis choose only the five values from year, month 
>           & day (skip all the repeated values). Get the time values using
>           days1900 function, define a time axis..assign to the variable 
>           using @ASN regridding.
>
>    3.  Use this new variable in scat2grid function. Now the variable is 
>          2D but one of the dimention is time. So you have to give only
>          the common lat and lon points as the arguments to the scat2grid.
>  
>        Define the final destination grid such that it is within the limits
>          of the original data. Dont forget to use the /UNITS attribute
>          while defining latitude or longitude axis
>
>    4.  Define a climatological monthly time  axis then regrid to this
>         usin @MODULO regridding to get the climatology.
>
>  Please go through the modified script..if anything need to be 
> explained..let me know.
>
>  As a "bonus" you can see how to use "POLYGON" command to look at the 
> original data in a spatial plot without any interpolation/extrapolation.
>
>
>  Hope this helps
>
>  With Regards
>
>  Jaison 
>
> On Thu, 18 Nov 2004, Chis Konig wrote:
>
>
> > Hi there,
> >
> > I have data that came in 2D-fields at irregular times, i.e. it is 
> > unevenly spaced in time (weekly - monthly) and space (i.e. 2D but was 
> > not on a lon-lat-grid). I'd like to make a climatology with monthly 
> > means etc. from it.
> >
> > Unless there is an elegant way of doing that all at once (3D 
> > regridding?) I think I have to achieve this in two steps (but correct me 
> > if I'm wrong):
> >
> > - First regrid it in space onto a lon-lat-grid for each time step.
> > - Secondly regrid it over time, i.e. make monthly averages etc.
> >
> > I might prefer a two-stepped approach anyway as it gives me more 
> > control, e.g. different types of interpolation for space and time.
> >
> > The data is in ASCII and in the form:
> >
> > yr mo dy lat lon var
> >
> > An example file can be found here: 
> > http://math.nyu.edu/~konigc/sample_data/sample_data.txt
> >
> > The first three variables (time) are non-decreasing but identical for 
> > several hundreds to thousands data points (all the data from one field).
> >
> > That's how far I got:
> >
> > !!!!!!!!!!!!!!!!!!!!!!!!
> >
> > define axis/x=1:1000:1 x_temp
> > define grid/x=x_temp g_temp
> > file/var="yr,mo,dy,lat,lon,var"/grid=g_temp sample_data.txt
> >
> > let gregday = days1900(yr,mo,dy)
> >
> > define axis/x=-75:-55:0.25 xax
> > define axis/y=79:83:0.25 yax
> > !Those axes would be different for the total data.
> >
> > let var_2=scat2gridgauss_XY(lon,lat,var,x[gx=xax],y[gy=yax],0.5,0.5,2,2)
> > !OR
> > let var_2=scat2gridlaplace_XY(lon,lat,var,x[gx=xax],y[gy=yax],5,2)
> >
> > shade var_2
> > go land
> >
> > !!!!!!!!!!!!!!!!!!!!!!!!
> >
> > My problems with that approach:
> >
> > - I lose the time dimension in the regridding process.
> > - The data is confined to the ocean. But with scat2gridXXX the 
> > information 'spills' onto the land. Is there a possibility to use a 
> > simpler interpolation (nearest neighbour, linear...)?
> >
> > Thanks for any help!
> >
> > Chris
>
>
> ------------------------------------------------------------------------
>
> !  
> ! delete the line containing data for 1990 03 31 80.50  -67.75  
> !   to bring all days to a common scattered points
> !
>       define axis/x=1:107:1 xscat   ! data points  
>       define axis/t=1:5:1   tscat   ! time points
>
>       define grid/x=xscat/t=tscat gscat
>  
>       file/var="yr,mo,dy,lat,lon,var"/grid=gscat sample_data.txt
>
>       ! let us have a cross check  
>       list/y=5 var
>  
>       !  get all the 5 dates  
>       let yr_5  =  yr[i=1] ! values need only for I=1,L=1:5     
>       let mo_5  =  mo[i=1] !     "
>       let dy_5  =  dy[i=1] !     "
>   
>       list yr_5, mo_5, dy_5
>
>       let gregday = days1900(yr_5,mo_5,dy_5)
>       
>       define axis/t/t0="01-JAN-1900:00"/units=days/from_data t5day=gregday
>       list t[gt=t5day]
>
>       ! get the variable onto the time grid instead of Y
>
>       let var_t    = var[GT=t5day@ASN]         
>       plot/l=5 var_t
>
> message
>
> ! now do the scat2grid
> !      choose min & max for defining the axes so that those
> !      values are within the actual range
>
>       let lat_1D = lat[l=1]  ! we need only the 107 common values 
>       let lon_1D = lon[l=1]  ! we need only the 107 common values
>
>       define axis/x=-61.5:-67.5:0.25/units=longitudes xlon  
>       define axis/y=80.5:81.75:0.25/units=latitudes ylat
>
>       let var_2=scat2gridgauss_XY(lon_1D,lat_1D,var_t,x[gx=xlon],y[gy=ylat],0.125,0.125,2,2)
>
>       fill/l=1 var_2
>       plot /ov/vs/color=7 lon[l=1],lat[l=1]
>       go land
>
> !  no you can define a monthly climatological axis (if needed..because now
> !     you have one value per month)..and make the climatology using var_t
>
> !      let us see what "polygon" can do
>
>  message
>
>       let xpts = XSEQUENCE(lon[l=1,i=1:107])  
>       let ypts = XSEQUENCE(lat[l=1,i=1:107])
>
>       let dxp=0.2 ; let bx=ysequence({`(-1)*dxp`,`dxp`,`dxp`,`(-1)*dxp`})
>       let dzp=0.05 ; let bz=ysequence({`dzp`,`dzp`,`(-1)*dzp`,`(-1)*dzp`})
>
>       polygon bx+xpts,bz+ypts,var_t[l=1]
>       plot /ov/vs lon[l=1],lat[l=1]
>       go land
--
  Christof Konig                    New York University
  251 Mercer St                          (212) 998 3207
  New York, NY 10012       http://math.nyu.edu/~konigc/
--