• Created by Unknown User (gdcarver113@outlook.com), last modified on Oct 10, 2017

I'm using OpenIFS 40r1 for some long integrations. I have been testing using a different model timestep but found that the amount of low cloud simulated by the model is sensitive to the timestep. I'm using the recommended timestep from the OpenIFS FAQ but have also tried longer steps but then warnings about 'trajectories underground' appear in the log file.

Can you please advise on the correct timestep and why the low cloud changes?

Thanks,  Ryan

2 Comments

  1. Unknown User (nagc)

    Hi,

    The OpenIFS FAQ recommends timestep values in order to avoid the warnings from the semi-Lagrangian scheme. However, at ECMWF we routinely run IFS/OpenIFS with much longer timesteps. For example T255 is run with a 45 mins timestep, T511 with a 15 min timestep. These are set according to a balance between speed and quality of the forecast. ECMWF needs to stretch the timestep as large as it can go for the operational forecast, but only if the forecast is not degraded. With these step choices, there may be some warnings appearing in the model log from the semi-Lagrangian calculation about trajectories going underground but the results will still be satisfactory.

    You may notice that the model also reports the "optimum" timestep for resolution it's running. This value is written to the NODE_001.01 output file. Look for the line:
          PTSTEP_OPT (optimal timestep computed in SUTSTEP) 0.36000000E+04
    to see the recommended value in seconds. However, please note that this code is at best a rough guide and in later versions of IFS it has been removed.

    As an interesting aside, resolution updates in the latest versions of IFS (not available yet as OpenIFS) and 'stretching' of the timestep for computational efficiency have shifted the mean ratio of timestep/grid-spacing upwards:

     Resolution         tstep      dt/dx
Tl399L91 (50km)     1200        0.02
Tl511L91 (40km)     900         0.02
Tl1279L91 (16km)    600         0.04
Tco1279L137 (9km)   450         0.05

    In order to compensate for the higher ratio above T511, the number of semi-Lagrangian departure point iterations was increased in order to maintain the forecast accuracy.

    As for the OpenIFS FAQ numbers,  after this discussion I am inclined to change that text to match the defaults we use in-house.

    Best regards,  Glenn


  2. Unknown User (nagc)

    Regarding the dependency of the low cloud amount with timestep, this is a known issue. Our scientists are aware that the low cloud amount will increase with a shorter timestep.

    It is under investgation and a fix is likely in the next release of OpenIFS (which will be in 2018).

    In the meantime, if you want to try tuning the low cloud amount, the most appropriate place would be tuning the cloud edge erosion term. That would changing the factor 5 in the lines below from cloudsc.F90:

    DO JL=KIDIA,KFDIA
   ZLDIFDT(JL)=RCLDIFF*PTSPHY !original version
   !Increase by factor of 5 for convective points
   IF(KTYPE(JL) > 0 .AND. PLUDE(JL,JK) > ZEPSEC)&
   & ZLDIFDT(JL)=5._JPRB*ZLDIFDT(JL) 
ENDDO

    Parameter RCLDIFF instead influences all clouds.

    Changes in the low cloud amount will unfortunately affect the model's energy budget, though the differences are small over medium range forecasts.

    Best regards,  Glenn