[MITgcm-devel] heff_max...more sea ice issues

Matthew Mazloff mmazloff at MIT.EDU
Mon Dec 18 14:57:07 EST 2006


Hi Jinlun,

I am not 100% sure what is happening.  The info I have comes from the  
monitor output.

with:     SEAICEuseDYNAMICS = .TRUE. the model crashes after 6  
months.  Heff_max is not a smooth function but jumps quite a bit with  
time
with:       SEAICEuseDYNAMICS = .FALSE. the model runs successfully  
for one year.  Heff_max is smooth in time and the values are at the  
low side of the oscillations I see with the dynamics on: I've  
attached a picture below

The advcfl_W_hf_max is similar in both models...perhaps it is the  
seaice model that is crashing.  Without ice dynamics, however, the  
ocean velocity fields have larger maxima.  But the cfl numbs  
associated with these maxima are still OK.

So maybe it is the ice model that is unstable.  I do not know what  
stability criteria I should consider with the ice model.  My setup:  
1/6 degree resolution, 900 second timesteps, using C-grid ice model  
configuration, LSR_ERROR = 1e-3.

I know this really isn't too much info....sorry about that.  Any ideas?

Thanks for the help,
Matt
-------------- next part --------------
A non-text attachment was scrubbed...
Name: heff_max.jpg
Type: image/jpeg
Size: 19993 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20061218/d0ce00a7/attachment.jpg>
-------------- next part --------------


On Dec 18, 2006, at 1:53 PM, Jinlun Zhang wrote:

> Hi Matt,
>
> When the model crashes, is it the ice that crashes, or the ocean  
> crashes? Generally, thick ice won't cause crashes (or make the  
> ocean more stable) because it moves slower and it does not create  
> big buoyancy change. So change heff_max generally won't help much  
> if the model would eventually blow up. And it is not a good idea to  
> set heff_max low because if would create lots of side effect  
> (buoyancy). One sensitivity study is to turn off ice dynamics and  
> run the model for sufficiently long to see what happens. Another  
> thing is to make sure the calculation of  ocean surface stress is  
> robust?
>
> Jinlun
>
> Matthew Mazloff wrote:
>
>> Hello,
>>
>> My upper layer is 10m thick.  When heff_max > 5m the model  
>> crashes.   (This only happens with dynamics on (LSR_ERROR = 1e-3  
>> in my  calculations so not very accurate).  With dynamics off the  
>> ice never  reaches this thickness.)  I am modeling the Southern  
>> Ocean, I am sure  Arctic modelers must have this problem to an  
>> even greater degree.   What is other people's experiences with  
>> thick ice and stability?   Should an effective thickness capping  
>> be implemented?
>>
>> Thanks
>> -Matt
>>
>>
>> fyi:  I tried the free drift sea ice hack (i.e. setting uc ~ uvel  
>> in  seaice_advdiff.F) and the model crashed.  The results were   
>> odd....lots of numerical noise in AREA and, it appeared, no   
>> significant advection.
>>
>>
>>
>>
>> On Dec 6, 2006, at 3:45 PM, Martin Losch wrote:
>>
>>> Hi Jinlun,
>>>
>>> in the end, we want to be able to use a flux-limited 3rd order   
>>> advection scheme, it's just not yet possible.
>>>
>>> Unfortunately my model stops at 80degN. I really need to set up  
>>> a  truly global ocean model. I do assume though, that there is   
>>> something fishy with the precipitation in the CORE data set.  
>>> Would  be interesting to see if some-one else has a similar  
>>> experience.
>>>
>>> In general, though, the ice extend is too large in winter (in   
>>> particular in the Drake Passage, isn't it?). That 's the only  
>>> thing  that's still worrying me.
>>>
>>> XKI: I am only playing with this parameter to find out how the  
>>> model (s) behave(s). In practice I will always use something  
>>> around 2.  XKI only appears in the denominator in budget.F, so I  
>>> don't quite  see what it does. I guess I have to dig up  
>>> Hibler79/80 and have a  look at the thermodynamics, right?
>>>
>>> M.
>>>
>>> On 6 Dec 2006, at 18:34, Jinlun Zhang wrote:
>>>
>>>>
>>>>
>>>> Martin Losch wrote:
>>>>
>>>>> Hi Jinlun,
>>>>> thanks for you input. I really enjoy this discussion!
>>>>>
>>>>> For clarification: I use the same advection for HEFF, HSNOW,  
>>>>> and   AREA. That is run47 has 1st order upwind for all three   
>>>>> variables,  while run41 has 2nd order central differences  
>>>>> scheme  (and not  flooding algorithm). All runs use a little  
>>>>> bit of  diffusion (the  default values of DIFF1=0.004), which  
>>>>> is probably  not good for run47.
>>>>
>>>>
>>>> Hi Martin,
>>>>
>>>> Good the ice and snow advection is consistent. When you use 1st   
>>>> order upwind, perhaps you do not have to use any diffusion (if  
>>>> you  have used any) since it is quite diffusive.
>>>>
>>>>>
>>>>>
>>>>> run48 uses only 10% of the snow precipitation, but uses  
>>>>> flooding   (it's just like 45). Are you saying that this should  
>>>>> not reduce  the  ice amount? One source of the ice is flooded  
>>>>> snow in the  flooding  algorithm in the current version of  
>>>>> growth.F (http:// dev.mitgcm.org/ cgi-bin/viewcvs.cgi/MITgcm/ 
>>>>> pkg/seaice/growth.F?  rev=1.34&only_with_tag=MAIN&content- 
>>>>> type=text/vnd.viewcvs- markup).  One of my problems was, that  
>>>>> the huge amounts of snow  that you see in  run40 (160m in some  
>>>>> places, no flooding, no  advection) are turned  into ice by  
>>>>> flooding and lead to ice  thicknesses beyond my  expectation.  
>>>>> Either there is too much snow  to begin with, or  something is  
>>>>> wrong in the handling of snow and  not enough snow is  melted.
>>>>
>>>>
>>>> I was looking at wrong figure and had mistakenly thought that  
>>>> when  you reduced snow, ice was pretty much gone. But actually  
>>>> ice was  only slight reduced with run48, not really bad. Sorry  
>>>> for the  mistake.
>>>> I don't know what is wrong with run40, but I wonder if this  
>>>> huge  snow depth also occurs in the Arctic. If that is the case  
>>>> in  Arctic also, then definitely something is really wrong with  
>>>> the  model.
>>>>
>>>>>
>>>>> 1D tests: As far as I understand the physics of ice formation:   
>>>>> Ice  forms because the atmospheric heat flux cools the ocean   
>>>>> surface below  freezing. Ice continues to grow as long a the   
>>>>> atmospheric surface  flux continues to cool the ocean. In the   
>>>>> presence of ice this  atmospheric heat is  
>>>>> "diffused" (conducted)  through the ice according  to the net  
>>>>> conductivity. In the  absense of snow this conductivity  should  
>>>>> be SEAICE_iceConduct  (XKI in budget.F). If the ocean provides   
>>>>> heat from below by  upward transport of warmer waters (by  
>>>>> vertical  convection), then  this heat flux can balance the  
>>>>> atmospheric heat  flux and stop  the ice from growing. When you  
>>>>> equate these fluxes  roughly at  equilibrium: Qocean =  
>>>>> conductivity*(Tair-Tsurfocean)/hice  you  get the ice thickness  
>>>>> that follows form this balance hice =   conductivity*(Tair- 
>>>>> Tsurfocean)/Qocean.
>>>>> Hypothetically I should be able to modify this "equilibrium    
>>>>> thickness" by playing with the conductivity (or Qocean or the    
>>>>> temperature difference). However I find that the model  
>>>>> parameter   XKI=SEAICE_iceConduct has no impact on hice (I use  
>>>>> 1e-6 instead  of  2!). That's puzzling, isn't? For the thsice  
>>>>> package, the   corresponding parameter does have an impact.
>>>>
>>>>
>>>> Your reasoning sounds ok. But I am not sure why ice is so   
>>>> insensitive to XKI in equilibrium. It does not make sense. Note   
>>>> that XKI is a physical term that is likely determined by lad   
>>>> experiments. Better not use a different number. Would be   
>>>> interesting to see how sensitive ice thickness/extent is to XKI  
>>>> in  real simulations.
>>>> Jinlun
>>>>
>>>>>
>>>>> Martin
>>>>>
>>>>>
>>>>> On 5 Dec 2006, at 18:17, Jinlun Zhang wrote:
>>>>>
>>>>>>
>>>>>> Martin Losch wrote:
>>>>>>
>>>>>>> Jinlun,
>>>>>>> thanks for your opinion. The thsice thermodynamics are   
>>>>>>> basically   Winton's (2000) model, but we have not yet fully   
>>>>>>> sorted out the   advection part.
>>>>>>> I have now a run47 with SEAICEadvScheme = 1 (1st order  
>>>>>>> upwind,   too  smooth) and no flooding, and and another one  
>>>>>>> (run48) which  is  just  like run45 but with only a 1/10th of  
>>>>>>> the snow fall,  just to  see what  happens, see
>>>>>>> http://mitgcm.org/~mlosch/run47.png
>>>>>>> http://mitgcm.org/~mlosch/run48.png
>>>>>>> As expected is run47 closest to what we expect. But run48 is   
>>>>>>> not  too  bad either, too little snow (of course) and as a   
>>>>>>> consequence  too  little ice. So either there is too much  
>>>>>>> snow/ precip in the   atmospheric forcing, or there is  
>>>>>>> something not  kosher in the snow   parameterizations. As the  
>>>>>>> problems are  similar with thsice I  would  agree that the  
>>>>>>> forcing may be the  problem ... I have to try  and find   
>>>>>>> different precipitation  fields.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Marin,
>>>>>> Yeah run47.png looks pretty good. The advection works ok. But  
>>>>>> I   wonder what ice advction you are using, 2nd order or 1st  
>>>>>> order?  The  one I installed is 2nd order. Ideally, the snow  
>>>>>> advection  should be  exactly the same as the ice advection so  
>>>>>> ice and snow  won't devorce  with each other.
>>>>>> It is not right with run48 that when the snow is turned off,  
>>>>>> ice  is  gone. Some thing is wrong here.
>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> I have also made another observation: I tried to run the    
>>>>>>> different  thermodynamics without any dynamics in a 1D case.  
>>>>>>> I   expect (and JMC  agrees with me) that for constant air   
>>>>>>> temperature  (say -30degC), ice  thickness should grow until   
>>>>>>> there is some  equilibrium thickness, when  the remaining  
>>>>>>> heat  flux out of the  ocean is balanced by the diffusive   
>>>>>>> flux of  heat through the ice.  I assume that the diffusion  
>>>>>>> is   controlled by "SEAICE_iceConduct"  for seaice and kice  
>>>>>>> for  thsice. The  equilibrium thickness can  roughly be  
>>>>>>> estimated by  hequil =  conductivity*(Tair-Twater)/ heatflux.
>>>>>>> I have only succeded yet in reaching some equilibrium   
>>>>>>> thickness  with  thsice (with an unrealistic value of  
>>>>>>> kice=1e-6  instead of  2). For  growth, this only works if I  
>>>>>>> turn on some  precipitation  (snow).  Without snow HEFF is  
>>>>>>> completely  independent of  SEAICE_iceConduct,  which I don't  
>>>>>>> think is right.
>>>>>>
>>>>>>
>>>>>>
>>>>>> I don't understand this equil. ice thickness )-:. As said   
>>>>>> above,  without snow-without ice thing or ice not working  
>>>>>> right  without  snow does not make sense to me.You might want  
>>>>>> to check  with  Thorndike (199?) for a toy model of equil. ice  
>>>>>> thickness.
>>>>>> Jinlun
>>>>>>
>>>>>>>
>>>>>>> M.
>>>>>>>
>>>>>>> On 5 Dec 2006, at 03:39, Jinlun Zhang wrote:
>>>>>>>
>>>>>>>> Martin,
>>>>>>>>
>>>>>>>> I would vote run45.png for best performance except that  
>>>>>>>> the   summer  ice is slightly overestimated. I would not  
>>>>>>>> vote  run41.png  because  of its weird snow distribution.  
>>>>>>>> The snow  pattern should  generally  follow the ice pattern  
>>>>>>>> (could mean  a problem with ice  advection). I  don't know  
>>>>>>>> why the snow  gets so thick with  run40.png, the precip   
>>>>>>>> forcing could be  way off. But obviously  snow advection  
>>>>>>>> helps a lot.  Snow  flooding, if it overestimates  ice, then  
>>>>>>>> turn it off, not big   deal (since what we do is to make   
>>>>>>>> the fields look like   observations). As for thsice, I don't  
>>>>>>>> know  what is going on.  But  for any ice thermodynamics  
>>>>>>>> that involves  ice salinity  (if thsice  uses ice salinity),  
>>>>>>>> there might be a  singularity  in the formulation  (I had  
>>>>>>>> such feeling before, but I  could  be wrong).
>>>>>>>>
>>>>>>>> Jinlun
>>>>>>>
>>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> MITgcm-devel mailing list
>>>>> MITgcm-devel at mitgcm.org
>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>>>>
>>
>
> -- 
>
> Jinlun Zhang
> Polar Science Center, Applied Physics Laboratory
> University of Washington, 1013 NE 40th St, Seattle, WA 98105-6698
>
> Phone: (206)-543-5569;  Fax: (206)-616-3142
> zhang at apl.washington.edu
> http://psc.apl.washington.edu/pscweb2002/Staff/zhang/zhang.html
>
>
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel



More information about the MITgcm-devel mailing list