hirlam

Dear all,

In investigating the evolution of perturbation in Harmonie I have used R and xtool to externally add a perturbation (P) to an analysis (A) (as opposed to reading in the files and adding them in the source). To simplify things, both A and P have the same geometry.

Although both R and xtool yield the same result, which is reassuring, there is one feature that I do not understand:

When visualizing (A+P) - A, the original P returns BUT also very small scale structures (also amplitude is factor 10 smaller) emerge from the boundaries. These are absent in the original P file. These structures appear both when plotting the gribfile (made with gl) associated with (A+P) - A and when plotting the differences between the grib files associated with (A+P) and A.

Has anybody a clue (packing?)

Many thanks,
Jan

I'm not sure how the R version works, but xtool does the computations in grid point space. So repeated spectral transforms and additional I/O doesn't necessarily reproduce the initial result. Does it make sense or do you see the ripples also in pure grid point fields?

Ulf

Hi Ulf,

Thanks for your reply. The observation that (A+P)-A shows small-scale structures especially near the boundaries is also valid for gridpoint fields (e.g. q=par51) and in this case P=0. Structures emerge from the boundaries and they remind me of so-called Arnold tongues. See attached plot, which shows (A+P)-P for P=0 and A an boundary ELS file. xtool has been used for the computation)

Best regards,
Jan

Jan,

I've tried to reproduce your problem, but whatever I do I can't produce anything but white noise. I've tried both spectral/gridpoint and conversion to GRIB before/after the computations. Do you have any example files? Where are you running your computations?

Ulf

Hi Ulf,

This is odd. The data I used is in /gpfs/scratch/ms/nl/nlf/evo1
As A I used an LBC file: ELSCFHARMALBC000 and as P a singular vector: SVARPE001+0000

To add the fields I used R and "add_ic_sv" on c1a: /home/ms/nl/nlf

To visualize data (e.g. (A+P)-A)) I used metview on ecgate

Best regards,
Jan

Jan,

I repeated the exercise with your data on c1a and calulcated both A+P and (A+P)-A with xtool and converted to grib. The result is as below i.e. without any ripples.

Ulf

Jan,

I tried with your data and produced A+P and A+P-A (with xtool) without any ripples at the boundaries. Very strange.

Ulf

Ulf,

This is indeed strange. I used Metview to plot the differnce fields. What happens if you use the same contour interval 0.0005 as in my plot?

Best regards,
Jan

Hi,

Sorry to interrupt your discussion. Is Metview that Ulf and You use based on the same graphics libraries? If you use it at Ecmwf platform should be fine or at least comparable then.
That's just my guess, because i've seen some quite simmilar differences when using metview to plot same fields built on different Magics++ (that comes included as far as i remember, but the built itself seems to be different on different platforms) versions locally.

Hi,

Actually I haven't used Metview at all and further I have produced the differences with xtool directly.

Ulf

Hi Ulf and Martynas,

The ripples keep annoying me and it could be that the plotting software (in my case it is Metview) is the culprit. Ulf, in case you still have this difference field somewhere lying around in grib, that would be great.

Regards,
Jan

Jan,

This is how I created the files

#!/bin/bash

cat > namjan_fa << EOF
&naminterp
readkey%ppp=51
readkey%lll=56,
readkey%ttt=109
output_format = 'FA',
/
EOF

cp ELSCFHARMALBC000 A
cp SVARPE001+0000 P

xtool -f -f1 A -f2 P -op SUM -o AP -igd -n namjan_fa -s || exit
xtool -f -f2 A -f1 AP -op DIFF -o AP-A -igd -n namjan_fa -s || exit
xtool -f -f1 AP-A -f2 P -op DIFF -o AP-A-P -igd -n namjan_fa -s || exit

You can also find the AP-A file as:
/scratch/ms/spsehlam/hlam/4jan/AP-A.grib

Ulf

Thanks Ulf,

I basically did the same with xtool, only without the namelist. This seems crucial. Adding this indeed removes the ripples around the boundary for par=51.

However running the script with readkey%ppp=11 does not change the result for this parameter. Perhaps we have to live with this (due to the spectral to grid transformations).

Regards,
Jan