Multiple doubts

General scientific issues regarding ROMS

Moderators: arango, robertson

Post Reply
Message
Author
rao_sc
Posts: 9
Joined: Tue Aug 04, 2009 3:00 pm
Location: Naval Physical & Oceanographic Laboratory

Multiple doubts

#1 Unread post by rao_sc »

Hi
I am having some doubts regarding roms:
1. We are implementing both ROMS and POM. When I am running same configurations for both the models, ROMS is taking 3 hr to run for one month on normal PC while POM is taking half an hour on the same PC (whithout MPI). Why this much time difference is there between the both ? However any options are there to decrease the time for ROMS to simulate Indian ocean circulation other than increasing the no. of processors?
2. How to use wind stress in bulk flux parameterization ?
3. Whether to spin-up the model, diagnostic mode is necessary ?

Thanks in advance.

Regards,
GNRao.

User avatar
kate
Posts: 4088
Joined: Wed Jul 02, 2003 5:29 pm
Location: CFOS/UAF, USA

Re: Multiple doubts

#2 Unread post by kate »

3. You do not need to spin up in diagnostic mode. I assume you mean TS_FIXED mode.
2. #define BULK_FLUXES will expect you to provide atmospheric fields and it will then compute wind stresses. Is that what you were asking?
1. ROMS may well just be that much more costly. Are you running both with the same timestep? Have you tried ROMS with longer timesteps? Have you looked at the character of the solution to see if they are giving the same answer? We like to think that ROMS now does a much better job over steep bathymetry and with not damping the free surface. There are things that can be done to speed up the open source ROMS, but they haven't been the priority thus far.

rao_sc
Posts: 9
Joined: Tue Aug 04, 2009 3:00 pm
Location: Naval Physical & Oceanographic Laboratory

Re: Multiple doubts

#3 Unread post by rao_sc »

Thanks for ur immediate response Kate.
1.U r right I am asking about the bulk_fluxes only. But I am having wind stress data, then how to feed to model since it looks for the wind speed?
2. How can I justify that diagnostice mode (TS_FIXED) is not required in ROMS?
3. At present I am not worried about the outputs from both the models. Now I am concerned about the time, why it is taking that much time with same time step, any specific reason is there? U have mentioned that there are methods to speed up the open source ROMS. Can u tell me how can I do that?
4. How to give the relaxing time in ocean.in file for the model to relax surface salinity for every month to climatology (WOA2005)?
Thanks in advance.

Regards,
GNRao.

User avatar
kate
Posts: 4088
Joined: Wed Jul 02, 2003 5:29 pm
Location: CFOS/UAF, USA

Re: Multiple doubts

#4 Unread post by kate »

rao_sc wrote:Thanks for ur immediate response Kate.
1.U r right I am asking about the bulk_fluxes only. But I am having wind stress data, then how to feed to model since it looks for the wind speed?
Make sure you don't have #define BULK_FLUXES and ROMS should look for wind stress instead of winds.
2. How can I justify that diagnostice mode (TS_FIXED) is not required in ROMS?
Try it and see? Is it really needed in POM?
3. At present I am not worried about the outputs from both the models. Now I am concerned about the time, why it is taking that much time with same time step, any specific reason is there? U have mentioned that there are methods to speed up the open source ROMS. Can u tell me how can I do that?
You may or may not be able to use a longer timestep with ROMS - it's worth a try. The other speedups are rather more invasive, involving timestepping schemes. They haven't been put into the open source ROMS because it would require that the adjoint and friends also get rewritten.
4. How to give the relaxing time in ocean.in file for the model to relax surface salinity for every month to climatology (WOA2005)?
The surface salinity relaxation uses the same timescale as the open boundary relaxation. If you want them to be different, hack in set_vbc.F in the SCORRECTION section and/or the SRELAXATION section. I've got:

Code: Select all

          stflx(i,j,isalt)=stflx(i,j,isalt)*t(i,j,N(ng),nrhs,isalt)-    &   
     &                     Tnudg(isalt,ng)*Hz(i,j,N(ng))*               &
#ifdef NEP5
!  60 days from Tnudg of 360 days
     &                     6.0_r8*                                      &
#endif
     &                     (t(i,j,N(ng),nrhs,isalt)-sss(i,j))

rao_sc
Posts: 9
Joined: Tue Aug 04, 2009 3:00 pm
Location: Naval Physical & Oceanographic Laboratory

Re: Multiple doubts

#5 Unread post by rao_sc »

Thanks Kate
1. I had run in diagnostic mode (TS_FIXED) upto 150 days. When I plotted kinetic energy vs time averaged over basin it increased upto 20 days and then it is almost constant(file attached). Whether now can we call the model is stabilized?
2 Actually I mean how to give tnudging values in ocean.in, it should be applicable to both boundary as well as SRELAXATION.
3 One more doubt is, I want to run the model with zero forcing. So I had given 0.0d0 scale factor in the varinfo.dat file for winds, tair,pair,qair,precip,lwrad,swrad. But model is getting blow up with any time step. So how can I run the model with zero forcings ?

Thanks in advance

With regards,
GNRao.
Attachments
Kinetic Energy Vs Time at first level
Kinetic Energy Vs Time at first level
ke_basin.gif (3.13 KiB) Viewed 4370 times

User avatar
kate
Posts: 4088
Joined: Wed Jul 02, 2003 5:29 pm
Location: CFOS/UAF, USA

Re: Multiple doubts

#6 Unread post by kate »

rao_sc wrote:Thanks Kate
1. I had run in diagnostic mode (TS_FIXED) upto 150 days. When I plotted kinetic energy vs time averaged over basin it increased upto 20 days and then it is almost constant(file attached). Whether now can we call the model is stabilized?
With TS_FIXED, the model doesn't really get to do a whole lot. You could have gone straight to free T,S after the 20 days - or even at day zero. The point of TS_FIXED is to spin up velocities in balance with the T,S fields if you only have say a T,S climatology to initialize from. My experience with using Levitus climatology this way is that it contains enough small scales that you spin up peculiar u,v fields. Setting T,S free to evolve then lets things settle into a more reasonable flow. My point last time is that u,v respond so quickly to the density, that it doesn't much matter whether you spin them up first or not. Now we happily initialize from a global ocean model, so we start off with u,v in balance and don't even think about the TS_FIXED option.
2 Actually I mean how to give tnudging values in ocean.in, it should be applicable to both boundary as well as SRELAXATION.
It's TNUDG in the input file. I have:

Code: Select all

 TNUDG == 2*360.0d0                    ! days
It is in units of days here, a value each for T and S. After it gets read in, it gets inverted for use later in the code.
3 One more doubt is, I want to run the model with zero forcing. So I had given 0.0d0 scale factor in the varinfo.dat file for winds, tair,pair,qair,precip,lwrad,swrad. But model is getting blow up with any time step. So how can I run the model with zero forcings ?
I would set them with ANA_SMFLUX, setting them explicitly to zero in the ana_xxx.h files. I would turn off BULK_FLUXES and only provide the stresses and the heat/salt fluxes. If it is still blowing up, it's probably something else.

feroda

Re: Multiple doubts

#7 Unread post by feroda »

kate wrote: We like to think that ROMS now does a much better job over steep bathymetry and with not damping the free surface.
Hi Kate,

I am quite interest in the performance of ROMS over steep bathymetry. Can you explain this point in more detail? What's your experience to handle with the steep bathmetry?

Specifically,
1. What are the key parameters in the *.in file needed to be carefully considered?
2. What are the appropriate values for theta_s, theta_b, hcline?
3. What do you think of the performance between different transformation equations as
well as stretching functions (Vstretching,Vtransform)?
4. In case of the water depth varies from 0m to 6000m within a 20x20
degree area, What do you think of the minimum depth in the model should
be? And, how about HCLINE in the *.in file?

5. Is it necessary to smooth the bathymetry? What the appropriate factor for maximum
slope of topography allowed in smoothing?


Thank you

User avatar
kate
Posts: 4088
Joined: Wed Jul 02, 2003 5:29 pm
Location: CFOS/UAF, USA

Re: Multiple doubts

#8 Unread post by kate »

Good questions all. Let us know what you discover. ;)

Seriously, ROMS reports two steepness parameters and I only look at the Beckmann one, aiming for around 0.4. For the other, I'm in the "insane" range according to Sasha, especially now that we've gone to N=60. :twisted:

I have a minimum depth of 10 m and I do smooth the bathymetry. I couldn't get the Beckmann number low enough without making the minimum depth off Kamchatka closer to 30 m, so I have a ramp in hmin. We do what we have to.

I haven't fully explored the vertical coordinate parameters. Enrique has even another set which is in my code. I've fallen into a rut of using theta_s = 5, theta_b = 0.4 for the old stretching - these values came from Bernard Barnier back in the last century.

rao_sc
Posts: 9
Joined: Tue Aug 04, 2009 3:00 pm
Location: Naval Physical & Oceanographic Laboratory

Re: Multiple doubts

#9 Unread post by rao_sc »

Sorry for the late reply Kate.
I would set them with ANA_SMFLUX, setting them explicitly to zero in the ana_xxx.h files. I would turn off BULK_FLUXES and only provide the stresses and the heat/salt fluxes. If it is still blowing up, it's probably something else.

I had did what u said and run the code but it was blown up even without completing one day. What would be the reason ? When I run same configuration with wind forcing, it is running successfully.
By doing what u said above, it will set only winds zero, then how to set fluxes also zero? Actually my problem is that to spin up the model with zero forcing (winds, fluxes) but only with initial conditions (currents=0, temp,sal from levitus climatology).
Thanks in advance

Regards,
GNRao.

Post Reply