Re: [ferret_users] @FNR in 2D fills up x dimension first

[Hella Riede <Hella.Riede@xxxxxx>]

Hi Ansley,

I had looked into fill_xy before trying @FNR, but it is "leaking" 
interpolation, see attached figure fill_xy_6iter_interpolation_leak.png.

fill_xy_iter1.png
shows that already at iteration = 1 (iter), fill_xy does not correspond 
to filling up with nearest-neighbor values. See for instance x=2 y=1: 
should be filled with 1.

fill_xy_iter1_longaxes.png
Same happens when x and y axes span from -2 to 10 to be sure not to cut 
any area of the created grid object, not shown.

The mask was '1' everywhere on the grid for all examples.

def ax/x=-2:10:1 xax
def ax/y=-2:10:1 yax
def sym lev = (0.5,10.5,1)
let a   = {3,3,3,5,5,5}
let b   = {1,2,3,1,2,3}
let val = {1,2,3,4,5,6}
let gridded=SCAT2GRIDGAUSS_XY(a,b,val,x[gx=xax],y[gy=yax],0.8,0.8,0.8,0)
let mask = IF gridded THEN 1 ELSE 1
let iter = 1
let filled = FILL_XY (gridded, mask, iter)
shade/lev=($lev) filled
go polymark poly/lev=($lev)/key/ov/line/fill a b val circle 2



Best regards,
Hella Riede



P.S. Regarding figure fill_xy_6iter_interpolation_leak.png: I had made a 
mistake in my previous handmade "expected" figure at position x=3 y=2, 
which is corrected in the attached version.






Am 29.06.2018 um 20:10 schrieb Ansley C. Manke:
> Hi Hella,
>
> The function fill_xy does  fill in two dimensions together, but with one 
> caveat. Here is your script, with the xax and yax axis definitions 
> changed so that the points are not in the corners of the domain
>
>     yes? let a   = {1,1,8,8,2}
>     yes? let b   = {1,8,8,1,4}
>     yes? let val = {2,4,6,8,10}
>
>     yes? def ax/x=-2:10:1 xax
>     yes? def ax/y=-2:10:1 yax
>
>     yes? def sym lev = (1.5,10.5,1)
>
>     yes? !! assign values from dots to grid cells
>     yes? set win 1
>     yes? let gridded =
>     SCAT2GRIDGAUSS_XY(a,b,val,x[gx=xax],y[gy=yax],0.8,0.8,0.8,0)
>     yes? shade/lev=($lev) gridded
>
>     yes? show function fill_xy
>     FILL_XY(DATA,MASK,N)
>          fills missing values with average nearest neighbour values
>          DATA: this arg is filled
>          MASK: mask (1 -> fill, missing -> do not fill)
>          N: the maximum number of fill-passes
>
>     yes? ! define a mask that's 1 everywhere
>     yes? let mask = 0*missing(gridded,1)+1
>     yes? set window/new
>     yes? shade/lev=($lev) fill_xy(gridded, mask, 1)
>
>     ! also look at fill_xy(gridded, mask, 2), fill_xy(gridded, mask, 3) ...
>
> The function avoids filling singleton points that lie in a corner of the 
> domain. We'll have to look into why that happens, but otherwise this 
> function does what you need.
>
> Ansley
>
>
> On 6/28/2018 7:37 AM, Hella Riede wrote:
> > Hello ferret community,
> >
> > I am using @FNR in two dimensions and got an unexpected result (see 
> > ferret script at end of this mail):
> >
> > In the attached ferret_before_FNR.png, the original 2D grid cells are 
> > shown (values inherited from dots by scat2grid, see below).
> >
> > ferret_after_FNR.png shows the result after applying @FNR (fill with 
> > nearest neighbor) in 2 dimensions.
> >
> > As an example for the unexpected result, lets take the grid cell at 
> > (x=8,y=5). Its value before @FNR was undefined. The nearest neighbor 
> > is (x=8,y=8). Yet, the value is inherited from (x=2,y=4).
> >
> > Is @FNR not supposed to be applied in 2D or is the logic behind 
> > different from what I expected? It seems the missing points are always 
> > filled up in x dimension first, and only remaining missing values are 
> > then filled up in y direction ...
> >
> > I attached a manual version of what I would have expected as 
> > ferret_FNR_expected.png.
> >
> >
> > Thank you very much in advance for helping to clarify this.
> >
> >
> > Best regards,
> > Hella Riede
> >
> >
> >
> >
> > can var/all
> > can sym/all
> >
> > let a   = {1,1,8,8,2}
> > let b   = {1,8,8,1,4}
> > let val = {2,4,6,8,10}
> >
> > def ax/x=1:8:1 xax
> > def ax/y=1:8:1 yax
> >
> > def sym lev = (1.5,10.5,1)
> >
> > !! assign values from dots to grid cells
> > set win 1
> > let gridded = 
> > SCAT2GRIDGAUSS_XY(a,b,val,x[gx=xax],y[gy=yax],0.8,0.8,0.8,0)
> > shade/lev=($lev) gridded
> > !! plot original scattered values for comparison
> > go polymark poly/lev=($lev)/key/ov/line/fill a b val circle 2
> >
> >
> > !! fill up undefined values with nearest neighbor
> > set win 2
> > let nn = gridded[x=@FNR,y=@FNR]
> > shade/lev=($lev) nn
> > go polymark poly/lev=($lev)/key/ov/line/fill a b val circle 2

--

Hella Riede

Deutscher Wetterdienst
Forschung und Entwicklung - Verfahrensentwicklung Fernerkundung (FE23)
Projekt Flottenwetterkarte (FloWKar)
Frankfurter Str. 135, 63067 Offenbach / Main
Tel. +49-69-8062-2414

https://www.bmvi.de/SharedDocs/DE/Artikel/DG/mfund-projekte/flotten-wetter-karte-flowkar.html

Attachment: fill_xy_6iter_interpolation_leak.png
Description: PNG image

Attachment: fill_xy_iter1.png
Description: PNG image

Attachment: fill_xy_iter1_longaxes.png
Description: PNG image