Good afternoon all

I have set up a coupled configuration with OpenIFS 43r3 + NEMO and run a first year. After using CDO to convert the grib output to netCDF I see a lot of "wiggles" in the surface pressure. I did not see any of this in 40r1. I'm using Tco95 (O96) for 43r3 and TL159 (N80) for 40r1.

Is this normal behaviour? I would expect some more wiggles due to the lower spectral resolution in Tco95, but it's quite a lot. Maybe it gets even worse due to the higher grid-point resolution?

I'm attaching two snapshots of my surface pressure fields and attach my post processing script.

Many thanks for any help

/Joakim

oifs-43r3-postprocess.sh

8 Comments

  1. Unknown User (nagc)

    Hi Joakim,

    Apologies for the delay in replying.  I don't use CDO but this doesn't look right to me. I wonder if this is a a result of using a Tco grid.  I'll ask around.

    Glenn

    1. Jan Streffing

      Hello Glenn, I don't think this has anything to do with Tco grids. I've seen these wiggles many times on ECHAM5 and ECHAM6 publications with regular grids. You can see them for example in the principal ECHAM6 model description and evaluation paper, Stevens et. al 2013


      For T95 this looks pretty normal to me. Usually above T159 they become less noticeable. I guess this was an advantage of TL grids with a higher spectral to grid point resolution ratio.


      Unknown User (de3j) If you plot the original grib file with, e.g. panoply, do you get the same stripes? As in, are they created by cdo or in the raw output?

  2. Unknown User (nagc)

    Spectral noise is not unusual around sharp features, such as around the Andes in the surface pressure. I used to eliminate some by carefully choosing which contour levels to use.

    Unknown User (de3j)  just checking, your plots above, the left one is the Tco and the right one is the Tl?  And both are using the same contour levels? (they look different).


  3. Unknown User (nagc)

    There might also be a small impact from different spectral transform codes. You could use the 'sptogp' program that comes with OpenIFS to use the same code for the Legendre & FFT transforms that OpenIFS uses.

  4. Unknown User (de3j)

    Hey guys

    Thanks for all the feedback. I would absolutely expect a lot more wiggles in the left plot (Tco95) than the right (TL159), especially around Greenland, Andes etc. But I'm getting wiggles almost everywhere.

    I should say this is in the surface pressure field ("SP") which comes in the ICMGG file. I thought this was from the ICMSH file, but it isn't. So this means the plots were written to disk on the octahedral grid, O96, and then I used "cdo -setgridtype,regular" which created a regular grid of size 400x192. I'm not sure how "SP" is done in OpenIFS, but I'm guessing it comes from "sp = exp(lnsp)" which is spectral and then Fullpos does the spectral transform before writing it to the ICMGG file. Hence any wiggles must come from the internal spectral transforms in OpenIFS, right?

    I'm attaching two plots of LNSP from the two ICMSH files. They show the same behaviour has the SP field from ICMGG, i.e. lots of wiggles with Tco95 (left) and no wiggles for TL159 (right).
    Maybe this is fine? I'll see if I can dig up some ECHAM6 Tq63 and Tq127 snapshots for comparison. Jan showed it's even visible in the climatologies, so that seems much worse than my Tco95 runs.

    I will give the sptogp from OpenIFS a go. I have only compiled the master.exe so far...


    Cheers!

    /Joakim


  5. Klaus Wyser

    Has there be any more insight into this problem? We also see these wiggles in many surface variables, and we think it comes from the spectral orography that is used by OpenIFS. Here is a plot of the geopotential from Tco95:

     

    Is there something that we can do about it? Reading a gridpoint filed with geopotential at initialisation?

  6. Unknown User (de3j)

    Hi Klaus

    I got this in an email some time ago (originally from Nils Wedi):

    Dear all,

    there is no question that there are more spectral ripples in the TCo, this manifests in low resolutions in particular,
    and that's why TCo is not the best choice for these. There is quasi a competition between spectral aliasing (from quadratic
    terms allowed in TL but not in TCo), and the increased spectral ripples in TCo, which is reduced with TL == doubling the wavenumbers (with the 
    same gridpoint representation) and especially with sharper features (such as orography and lnsp) where cubic truncation produces more ripples. 
    The linear grid 'absorbs' effectively the ripples from truncation over N/2 of the spectrum.

    Now at higher resolution, the aliasing becomes terrible and one needs to invent filters to remove it. These filters affect up to N/2 of the spectrum, and associated with tis is a high computational cost paying for an extra N/2 of the spectrum (in the transforms every time step) only to remove 
    ripples (and importantly where N/2 covers an increasingly smaller delta x == smaller ripples in a map ... ).

    There is an additional effect due to us choosing to push orography to the limit of truncation in TCo grids (possible because we don't have the aliasing issues), 
    again this choice is motivated by high resolution (>=TCo399).

    So in short, use TCo wisely, otherwise stick with the linear grid.

    Best,

    Nils


    I interpret this as Tco being a "less than fantastic" choice at low-res but really good at high-res. I see a big effect of the ripples in Tco95 but it's a lot less at Tco199 and Tco399.

    See attached plot of U10 bias (compared to ERA-5) in AMIP runs. It certainly gets better with Tco399 than Tco95, but I don't think it comes from the ripples...

    Best wishes
    Joakim


  7. Klaus Wyser

    Thanks a lot Joakim for the swift reply. Unfortuantely that's exactly what I was afraid off, there isn't much we can do about the wiggles.