SSH output of GLBa0.08

[shao...@gmail.com]

Hi ,

    I got a question about SSH output when I rerun the GLBa0.08 in realtime with all the setting mentioned in Web and Forum. As an example, the restart file is from  /datasets/GLBa0.08/expt_91.2/hindcasts912_archv_2016041818_2016041800.tar.gz, and exp starts in 2016.04.18. The force is from NAVGEM with the corresponding time. Then, I make a five days run and get the SSH with the format of Netcdf  from the output. As a validation, so I compare it with analysis data delivered in web. But, the result is not good, and the SSH from output is overall less than the analysis data about 10cm. 

   I have tried to modify some settings. I find that the variables 'psikk'(Montgomery potential in bottom layer)  'thkk'(Virtual potential density in bottom layer)  and 'pbot'(Bottom pressure at t=0)  are written from another restart file when I make the initial file. I cannot make clear that the mistake is from these variables or not. And, are there other reasons can make this error in usual.

Looking forward to your reply. Thanks a lot.

Carl

[alan.wa...@hycom.org]

As you say, some of the fields in the restart file are from model day zero and this is why we need a "template" retstart file to convert an archive file to a restart file.

In your case, the template restart came from runnning the model starting from climatology to get an initial restart file.  This won't have exactly the same "day zero" state as our runs, but it should be close if you start from the January climatology.  Note also that you can always add a constant everywhere to SSH without changing the result (because only its gradient matters).  So I would not worry much if your SSH differs by a global constant offset from the one here.  I would worry if the gradients are different, e.g. if the range of SSH over a region is smaller than the original or if the patterns are very different.

Alan.

[shao...@gmail.com]

Alan,

Thanks first for your reply.

And other issues for this problem. These variables such as 'psikk' only exit in restart file, so I make a short model run in climatology to get a restart file and use it as a template. Maybe it is too casual.  According to your reply, I think, I need to run the model for a long time in climatology (spin up) to get the restart files of each month. And then use these restart files as template of new restart file in realtime for different month. Is this?

And I will check the SSH gradients of the results according to your suggestion later.

However, I have an idea to solve it, I do not know whether it is correct. The variables of 'psikk' 'thkk' and 'pbot' can  caculate from other variables in archive file, e.g. 'psikk' is the function of the gravity acceleration, density and deep. So, can I caculate these variables by myself and write these to restart file instead of template(other restart file). I think it is good "day zero" state for each realtime run. Is it possible to do so?

Carl.

[alan.wa...@hycom.org]

You only need *one* template restart file and it should be for *January*.  This is because all our simulations start from climatology in January.  They then run for years, but, as I said in my last post, all the restart files contain the same few fields from *the very first day of the model run* (even if it was model-decades ago).  This also means that you don't have to run from climatology for a long time (we typically run 1 model day when generating a template restart).

It is probably possible to generate synthetic 'psikk' 'thkk' and 'pbot', e.g. from a long term mean archive file (from the reanalysis perhaps) but I have never looked into what this would require because the "template" approach has worked well enough.  It is likley also possible ot correct a template restart file to allow for a region-wide SSH offset, but I have never tried that either.

Alan.

[shao...@gmail.com]

So, if I need to simulate other month instead of January, can I also use template restart file from climatology one day run in January?

Carl.

[alan.wa...@hycom.org]

Yes.  You should use the one day run in January as the template for all restarts from an archive file.

Note that we would not usually setup a restart file from an archive multiple times.  Usually the restart from January (say) is used to start February, etcetera.  If you are constraining your solutions to archives that include data assimilation, then multiple archive to restart steps may be appropriate.

Alan.

[shao...@gmail.com]

Then, can you give me some advice of the correct method to calculate the offset of SSH output. And is it different in different place and different time?

Carl.

[alan.wa...@hycom.org]

It should be a constant in space and time.

You don't need to apply an offset at all, but if you want to estimate its size I suggest plotting the global SSH from here and from your run.  It should be easy to eyeball a "good enough" constant offset from the plots.

Alan.

[shao...@gmail.com]

The offset of SSH is different in different place between the my results and the analyzed data, and there is a large deviation. Of course, the error of model run is the part of this deviation.

So, if the offset is constant in space and time, can I calculate the average of the global SSH difference between the model results and the analyzed data and use this average value as the offset?

Carl.

[alan.wa...@hycom.org]

Yes,

Note that there should not be a large difference between your initial state and the model run after (say) 1 model day.  So I would use the 1st SSH you have and compare it to the data assimilative SSH from here.  Once you add the global area mean offset (if you have .a files you can use ALL/bin/hycom_mass to calculate these) the plots after a short run should be very similar.

Alan.