Good morning all

I've made some runs with Tco95L91 coupled to an NEMO (ORCA05) ocean and started investigating some of the biases. One thing that pops up is a strange behaviour of the QBO (see Figure).

I reproduced Fig 1 from Baldwin et al (2001), i.e. plotting de-seasonalised monthly-mean zonal-mean zonal wind over the equator. The top plot is ERA-Interim and the lower plot is years 210 to 240 of the pre-ind spin up. Clearly, OpenIFS has

1) Only QBO activity in the first 10-15 years, then almost nothing
2) Too high QBO frequency

If I understand the documentation correctly, the non-oro wave drag parameterisation is resolution dependent in the range 29km - 1km, so I guess I should increase the momentum flux?

Are there some recommendations for parameter choices outside the resolution range 29-1 km, in particular in the 50-100 km range?

Many thanks in advance for any help!

/Joakim


7 Comments

  1. Federico Serva

    Hello, out of interest, which value of GFLUXLAUN are you using? You can have a look at sugwwms.F90. In other IFS versions the scaling was done for higher

    resolutions probably in your case you should reduce the flux (depending on the launching level)?

    Regards

  2. Joakim Kjellsson

    Hi Federico

    I find this is sugwwms:
    GFLUXLAUN=3.75E-3_JPRB
    ZLAUNCHP(1)=45000.0_JPRB

    So I'll do a few attempts to reduce GFLUXLAUN and see how that goes.

    I see that it's also possible to alter launching level and also add up to three launching levels, but I'll stay away from this for now...

    Thanks for the tip!

    Best
    /Joakim


  3. Federico Serva

    Hi, yes these were the default in previous versions as well.

    At T95 you may need to reduce the flux, but take into account this should also strengthen the polar jets.

    For the launching height, if you move it towards the surface then you should increase the flux.

    Regards

  4. Tim Stockdale

    The QBO in the IFS is driven by a combination of resolved wave forcing (which increases with horizontal resolution, and appears to be unrealistically strong, even more so with stochastic physics active) and the parameterized non-orographic wave drag.  To adjust the strength of the NOOGWD, the recommendation is not to change GFLUXLAUN (which controls the overall value, used at mid and high latitudes), but to change instead the parameter GGAUSSB. If you use a value of -0.25, this gives a 25% reduction near the equator. If instead you use -0.75, you will git a 75% reduction, so the forcing of the QBO will be 3 times smaller and the period will then be longer and maybe more realistic. For SEAS5 (TCo319 plus stochastic physics) we run with GGAUSSB = -0.95, i.e. hardly any NOOGWD is present at the equator. The amount needed also depends on the model cycle.

    I've never tried to get a realistic QBO in the IFS at TCo95 L91 resolution. I don't know what the resolved forcing will look like at TCo95, which is much lower than I normally use. Also, the resolved forcing depends on whether the model is couled to the ocean or not, and if it is, might depend on what SST biases develop in the tropics.  Also, L91 might allow a QBO, but it will be substantially more relaistic with L137, especially in the upper levels (5-10 hPa), which might be relevant for sustaining the QBO in a long integration in some situations.

    My advice is first to try adjusting GGAUSSB (maybe start with -0.75) to try to get a reasonable period. If the amplitude is then too small (or the QBO just dies out), I might then try improving the vertical resolution if you are able to. The vertical diffusion in the stratsophere also damps the QBO quite strongly, so you may need to reduce this, too.

  5. Joakim Kjellsson

    Hi Tim

    Thanks for such a detailed reponse!

    I should say that getting a realistic QBO is not all that important, but I'm looking to set up as good of a version of Tco95L91 I can.
    There is a plan to run higher res, which means I would have to revisit this parameter for a new resolution.

    I'll take your advice to not change the launch flux, but instead the GGAUSSB.
    Might be worth testing L137, but I don't think we would use it for long coupled runs in the end...

    Thanks again!

  6. Joakim Kjellsson

    Hi Tim et al

    I've fiddled with GGAUSSB and attach plots of zonal wind bias (time- and zonal mean) and the monthly mean zonal wind anomalies associated with QBO. In both cases I compare to ERA-Interim.

    From what I can see GGAUSSB does not change the zonal wind bias much. Smaller GGAUSSB seems to be better but its not a linear response.

    But the change is dramatic for the QBO. Small GGAUSSB gives good QBO amplitude while large GGAUSSB gives a more realistic frequency.
    I guess GGAUSSB = -0.95 in combination with less diffusion in the stratosphere might be worth trying.
    At the same time, I guess there's a limit to how well this can go at Tco95L91.

    In any case, I'll leave my plots here in case they help someone else going down this path later on.

    Thanks all for the suggestions
    Joakim

  7. Tim Stockdale

    Nice set of plots to illustrate the behaviour. The zonal wind bias exists also in seasonal forecasts, and is very stubborn. The QBO is as expected, it is easy enough to tune the period, but very hard to try to get the amplitude and structure right at the same time!!  In the QBOi studies (see https://doi.org/10.1002/qj.3820 and its reference list at the bottom), all models have similar problems.