[MITgcm-devel] another bug in growth.F ?

Wed Nov 29 19:51:40 EST 2006

Hi Guys,

J.M. just steps in my office. And guess what he says?
"Will take some strong beer to convince them
that there is a pkg that has those snow issues solved?
(and we had planned to switch to it)"
Well, it's not exactly what he said, but...

-p.

On Nov 29, 2006, at 11:59 AM, Martin Losch wrote:

> Hi Dimitris,
> I have tried this:
>> CML          IF(FICE(I,J,bi,bj).GT.ZERO) THEN
>>           IF ( AREA(i,j,2,bi,bj).GT.0. _d 0
>>      &        .AND. atemp(i,j,bi,bj).LE.273.16 _d 0 )  THEN
>> C FREEZING, PRECIP ADDED AS SNOW
>>            HSNOW(I,J,bi,bj)=HSNOW(I,J,bi,bj)+SEAICE_deltaTtherm*
>>      &            PRECIP(I,J,bi,bj)*AREA(I,J,2,bi,bj)*SDF
>>           ELSE
>> C ADD PRECIP AS RAIN, WATER CONVERTED INTO equivalent m of ICE BY  
>> 1/ICE_DENS
>>              SEAICE_SALT(I,J,bi,bj)=SEAICE_SALT(I,J,bi,bj)
>>      &            -PRECIP(I,J,bi,bj)*AREA(I,J,2,bi,bj)*
>>      &            SEAICE_deltaTtherm/ICE_DENS
>>           ENDIF
> to make snow a function of air temperature instead of FICE (ice  
> growth rate?), but that does not change the results very much.  
> (after 10 years, more tomorrow)
>
> However, what seems to be more important is my little flooding  
> algorithm. Turning that off decreases HEFF by almost a a factor of  
> 2 at the cost of a having too much snow (>5m in 10years) in  
> isolated places again (all without advection of snow).
>
> I don't know what to say. I'll wait until tomorrow to have a closer  
> look after 100years of integration, but a first look implies that  
> the snow alogithm is still not working the way we expect it to work  
> and the flooding algorithm takes all the snow and converts it into  
> ice. Which leads us back to the original problem of too much snow  
> in the first place. (I am using the latest code).
>
> M.
>
> On 28 Nov 2006, at 20:29, Martin Losch wrote:
>
>> Hi Dimitris,
>>
>> since I have been doing all this sort of "by the way" I am no  
>> longer absolutely sure, what exactly run34 is. But most likely it  
>> is with growth.F pre 1.29, and with flooding turned on, but NO  
>> advection of snow (the last two I know for sure). So the flooding  
>> algorithm takes care of the snow over open water, but in a pretty  
>> simple way, thermodynamically speaking. run38 is the same with  
>> growth.F 1.34
>>
>> This forward sensitivity is frightening. I am not sure if I am as  
>> optimistic about it as you are, but I have nothing to offer in  
>> terms of explanations. I observe in my 100y runs that the ice  
>> thickness builds up within the first 10years and then stays nearly  
>> stationary (with seasonal cycle of course). The ice cover does not  
>> retreat very far in summer. Does the non-linear effect that you  
>> describe explain such an "equilibrium" at higher mean ice  
>> thicknesses?
>>
>> I have not had the time to get my head around the thermodynamics  
>> in growth, but your suggestion (making snow/rain depend on forcing  
>> fields) sounds good to me. Currently I have not time to play with  
>> this (maybe starting again next Friday)
>>
>> Martin
>>
>> On 28 Nov 2006, at 17:08, Dimitris Menemenlis wrote:
>>
>>> Martin, I also notice increased ice thickness around Antarctica  
>>> in the high-res cubed sphere integration and increased sea-ice  
>>> extent, much more summer ice extent than observed.
>>>
>>> What is you run34?  In terms of CVS repository,
>>> http://mitgcm.org/cgi-bin/viewcvs.cgi/MITgcm/pkg/seaice/growth.F
>>> which version of growth.F did you use: pre 1.29, 1.29 with  
>>> flooding and advection turned on, 1.30, or 1.34?  So far as I  
>>> have looked in my own tests, results from 1.30 and 1.34 are not  
>>> that different.  But then your run34 does not have snow over open  
>>> water, so it cannot be pre 1.29?
>>>
>>> Optimistically, the too-much ice in run38 may be a bug in the  
>>> NCEP/CORE forcing fields rather than in growth.F.  That is, too  
>>> much precipitation is converted into snow, which extracts heat  
>>> from ocean when it melts.  The effect would be highly non-linear  
>>> since more ice/snow extent means higher albedo, which leads to  
>>> cooler ice/ocean surface temperature, which in turn leads to more  
>>> precipitation being converted to snow, since in present treatment  
>>> of precipitation, rain to snow conversion depends on  
>>> thermodynamic ice growth (rain) or melt (snow).
>>>
>>> Optimistically again, the high tangent linear sensitivity noted  
>>> by Patrick and also in the verification experiments that you  
>>> report is also due to above effect.  Incidentally, with growth.F  
>>> prior to 1.30, the verification/lab_sea domain is at all times  
>>> 100% covered with snow (but not ice).  So low forward  
>>> sensitivities prior to 1.30 are almost certainly for wrong reason.
>>>
>>> One possible way of reducing the forward sensitivity (I have not  
>>> yet tried it) would be to remove the snow/rain dependence on ice  
>>> growth rate and instead make it depend on forcing field, e.g.,  
>>> surface air temperature and fresh water freezing point, as is  
>>> done in pkg/thice.  Should I give this a try?
>>>
>>> Dimitris
>>> _______________________________________________
>>> MITgcm-devel mailing list
>>> MITgcm-devel at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>>
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel

Dr Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
MIT | EAPS, 54-1518 | 77 Massachusetts Ave | Cambridge, MA 02139, USA
FON: +1-617-253-5259 | FAX: +1-617-253-4464 | SKYPE: patrick.heimbach