Dear OpenIFS experts,

I am Abhishek and am very new to using the OpenIFS model, and have a question regarding the SST forcing we use to force the OPenIFS model (i.e., atmospheric components). So, I ran two simulations with the Tco95 OpenIFS configuration with different time steps, namely 1 hour and 30 minutes. Interestingly, when I compare the SST that is saved as a model output in both simulations (monthly only), there are some differences between them, which do not make any sense to me at all. I mean, why does the SST forcing change between these two simulations? According to my experience, SST should be the same between these two simulations because both simulations were forced with SST forcing. I am also attaching the figure in which you can see the changes are noisier and can mostly be seen in the Southern Ocean. I would really appreciate it if somebody could give some rationale behind it. Many thanks in advance.

\

10 Comments

  1. Marcus Koehler

    Hi Abhishek Savita, 

    For my understanding, is this a coupled experiment (with NEMO) or do you read SSTs from the ICMCL file?  How fequently do you write out the SSTs in either of the simulations. Note that the output frequency of model fields is linked to the timestep, so if you change the time step length then you will output at different times during your forecast.  How do you save monthly SST, do you calculate an average from the output steps?

    Regards,

    Marcus

  2. Abhishek Savita

    Thank you Marcus for your prompt response.

    First of all, these are not coupled simulations. That is, I forced the OpenIFS model with SST by ICML file.

    If I understood correctly, then I set the output at every 6 hours and then save it monthly by taking an average of each simulation.

    Just for a note: I do set in the context_ifs.xml file a different value of fullpos_freq that is 

    1)  1 hour time step  

    <variable_group id="fullpos_freq">
          <variable id="nfrhis" name="NFRHIS" type="int"> 6 </variable>
          <variable id="nfrpos" name="NFRPOS" type="int"> 6 </variable>
        </variable_group>

    2)  30 minutes time step

    <variable_group id="fullpos_freq">
          <variable id="nfrhis" name="NFRHIS" type="int"> 12 </variable>
          <variable id="nfrpos" name="NFRPOS" type="int"> 12 </variable>

    Please let me know if you requires any further information.

    1. Marcus Koehler

      Hi Abhishek, yes this looks all right to me, with 30 mins time step you would need NFRHIS & NFRPOS set to 12 to get 6-hourly output. 

      I am not sure what the answer is, but I believe that the key to the changes that you see is due to the fact that you are not forcing OpenIFS at each time step with the same SST value. 

      The ICMCL file contains usually SSTs with daily frequency and OpenIFS will interpolate between these daily forced values.  If you change the model time step then the interpolation will have a slightly different outcome (because the time steps are at a different time during the forecast).  If you then calculate mean whiles from these then I would expect small variations which look like noise.  And, looking at your figure, the changes look like noise and are mostly very small (fraction of a K if I see this right?).  Over land the changes are larger because the diurnal range of the land surface temperatures is bigger and if you sample at different times of day (due to the change in time step) I would expect the effect to be larger.  I have no idea why the southern ocean is more visible.

      At least this would be my explanation.  I can ask someone who knows more about how IFS deals with SSTs.

      Cheers,  Marcus

      1. Abhishek Savita

        Thank you very much Marcus Koehler and yes, the difference shown in the attached figure is in K. I was also wondering when I saw these differences why these differences are mostly visible in the Southern Ocean only. Although these differences are small compared to land, which you already explained.

        It would be good if you discussed this and maybe someone has an idea about it. 

        1. Marcus Koehler

          Hi Abhishek Savita, I did not receive any particular insights into the southern ocean location from my colleague, but they also consider these differences not as an indication of a problem (as also Unknown User (gdcarver113@outlook.com) indicated below).  If you look at the instantaneous output at a specific time instead of monthly averages (and, as recommended below, output them to the log file instead of writing to GRIB output) these differences should hopefully be smaller or disappear.  I also don't think that there is actually a problem. 

          1. Abhishek Savita

            Great!

  3. Unknown User (gdcarver113@outlook.com)

    Hi Abhishek,

    Your SST fields look virtually identical to me. Bear in mind that the model outputs GRIB and this reduces the data precision much less than the 64bit precision the model uses. You should therefore expect to get differences to the 3/4th significant figure.   I suspect if you look at the values of the SST rather than the plot, you'll see this.  I don't think you have a problem with the SST at all.

    Cheers.

  4. Abhishek Savita

    Thank you Ryan Harry and I do agree with you that the differences are small, but I am just wondering why these differences are visible in the Southern Ocean only. 

    Just in case, I noticed something mentioned in the user guide about cool skin temperature, and perhaps you guys Marcus Koehler  and Unknown User (gdcarver113@outlook.com)  will be able to comment if it is related to cool skin temperature.

    You can find the explanation at https://www.ecmwf.int/file/267309/download?token=gu4TQEB0 in chapter 8 at page number 162 and equation 8.151.

    Cheers

    Abhi

    1. Unknown User (gdcarver113@outlook.com)

      Hi Abhishek,

      I don't understand what you mean about differences visible in southern ocean only. In your difference plot, I can see differences in the N.Atlantic and the Pacific Ocean. They seem same density and order of magnitude as the Southern Ocean. There's just more ocean points in the SH.  The cool skin effect is not relevant here.  The SST field in IFS is stored in the same array as the first soil layer, that's why you see differences over land.  The models' land-sea mask is used to extract the values of SST just over the ocean.

      You're just looking at round-off errors in the difference plot because the GRIB packing has reduced the data precision to 12 bits (or thereabouts).

      As Marcus says, take a look at some instantaneous values too.  Unfortunately, one of the not so fun things about IFS is trying to debug small errors by looking at the model's GRIB output  (smile)

      Best.

      1. Abhishek Savita

        Thank you Ryan.

         I mean was that Southern Ocean has large number of red noise than other Ocean.

        Thanks for the explanation. 

        Cheers

        Abhi