[Liwg-core] energy budget issue for glaciated landunits

Bill Sacks sacks at ucar.edu
Tue Jan 24 15:19:21 MST 2017


Hi Leo,

In going back to the code and corresponding comments, I remembered one of the main motivations for doing downscaling the same over vegetated columns as over glc columns, within the CISM domain: Imagine you are starting with a CLM cell that is entirely ice-covered, with:

(a) EC 1: 10% of grid cell (100 m; further assume this is very thin ice)
(b) EC 2: 50% of grid cell (300 m)
(c) EC 3: 40% of grid cell (500 m)

Now imagine that EC1 melts, with other ECs staying ice-covered (at the same elevation as before) so you have:

(a) Bare: 10% of grid cell (100 m - as noted above, we were assuming very thin ice, so the elevation is the same as before – but that's not crucial for the conclusions)
(b) EC 2: 50% of grid cell (300 m)
(c) EC 3: 40% of grid cell (500 m)

If I were analyzing the change due to the ice retreat, I would hope that the only significant difference between these cases is the land cover in (a) – ice vs. bare land. If you add to the mix that (a) has different downscaling before and after, it gets hard to figure out what's really responsible for any changes you're seeing. And actually, the problem is bigger than this: If longwave downscaling only operates over glc columns, then even the behavior of (b) and (c) will change when (a) goes from glacier to bare land – since the normalization factors will change. This, to me, is very undesirable behavior, and I would need to hear some very good arguments in favor of this treatment.

All that said: I think the experiments you're doing to investigate the impact are still worthwhile... I just want you and others to know that I'm very hesitant to actually change the model in this way.

Bill S

> On Jan 24, 2017, at 2:45 PM, Lipscomb, William Henry <lipscomb at lanl.gov <mailto:lipscomb at lanl.gov>> wrote:
> 
> 
> Hi Leo,
> 
> Thanks for the update.  Were you surprised that melting decreased when you turned off LW downscaling?  This suggests that some of the lower-elevation glacier ECs rely on the downscaled LW to melt.  Would you like to go ahead and try the ice-column-only downscaling?
> 
> I haven’t been able to think of a reason why the LW effect is much bigger in coupled runs than in offline runs.
> 
> A couple of questions:  First, could you remind me of the definitions of QSNOMELT, QSNOMELT_ICE and QICE_MELT?  Is QSNOMELT_ICE the refreezing?  Second, what is our target for QICE_MELT?  It seems like we’re still way below where we want to be, even in B3 with the adjusted ramp, the lowered ice albedo, and the SCF bug fix.  Is there any significant increase in melting over time?
> 
> Given that the surface forcing seems reasonable overall, I can’t help wondering if we’re missing something important.  What was the last simulation with a reasonable amount of surface melting?  Do you think we’re getting too much refreezing, maybe as a result of missing physics?
> 
> Thanks,
> 
> Bill L.
> 
> 
>> On Jan 24, 2017, at 2:13 PM, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>> 
>> Hi Bill and others,
>> 
>> I’ve been doing some more experiments with LW downscaling:
>> https://docs.google.com/document/d/1l6SU0PPba8eOuB8fiI3LYVDdyDMrwZkQf2lyh7Ztaqw/edit# <https://docs.google.com/document/d/1l6SU0PPba8eOuB8fiI3LYVDdyDMrwZkQf2lyh7Ztaqw/edit#>
>> 
>> As I mentioned in a previous email, I found that it has little effect on the energy balance in offline runs (only about 5 Gt/yr). 
>> However in coupled mode, there is a big response when I turn off LW downscaling : it is halved (cf. run B3 and B4). 
>> 
>> I can’t really explain this at the moment, it could even be caused by natural variability , but I suspect a real difference between coupled and offline runs. 
>> I’ll keep working on this a bit more, any suggestions welcome! 
>> 
>> Leo
>> 
>> 
>> 
>>> On 23 Jan 2017, at 16:48, Lipscomb, William Henry <lipscomb at lanl.gov <mailto:lipscomb at lanl.gov>> wrote:
>>> 
>>> Hi Leo,
>>> 
>>> Your finding about the downscaled LW is very interesting.  When we added the LW downscaling option, I was imagining giving greater LW (and greater melt) to lower-elevation ECs, relative to high-elevation ECs.  I think this suggestion originally came from Jeff Ridley at the Hadley Centre.  But I can see how the downscaling would redistribute LW energy from glaciated to bare-land columns, which could be a bad thing.
>>> 
>>> I’d suggest that we first try turning off the LW radiation downscaling entirely, and see what happens.  Then we can come back to the idea of an ice-column-only downscaling.  I would expect that an ice-column-only downscaling would maximize melt in low-lying glacier ECs, though as you say the physical basis for this (without a corresponding treatment of SW) is a little shaky.
>>> 
>>> Thanks,
>>> 
>>> Bill L.
>>> 
>>>> On Jan 23, 2017, at 8:13 AM, Bill Sacks <sacks at ucar.edu <mailto:sacks at ucar.edu>> wrote:
>>>> 
>>>> Hi Leo,
>>>> 
>>>> Okay, thanks for the explanation of your thoughts. I'll let Bill L chime in here if he has anything more to add, but my understanding is that the current parameterization implicitly assumes that cloud fraction, cloud height and most other properties such as cloud liquid water content are the same for different elevation classes / bare land. In that case, downwelling longwave radiation is a function of cloud temperature. The current adjustment is based on differences in surface temperature, which won't be exactly the same as cloud temperature, but seems like a reasonable approximation given the limited information we have.
>>>> 
>>>> Do you see problems with this?
>>>> 
>>>> If you feel this is problematic, then I'd probably argue for just turning off longwave radiation downscaling entirely, rather than applying it only over ice-covered regions. This can be done via the namelist parameter, glcmec_downscale_longwave.
>>>> 
>>>> Thanks,
>>>> Bill S
>>>> 
>>>>> On Jan 23, 2017, at 6:33 AM, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>>>>> 
>>>>> Hi Bill,
>>>>> 
>>>>>> I'm adding liwg-core to the cc list here, to include others in the discussion.
>>>>> 
>>>>> 
>>>>> that is probably a good idea, thanks. 
>>>>> 
>>>>>> Within the CISM domain, the elevation of the bare land portion of the grid cell comes from CISM. Jeremy and I discussed this extensively, and this felt like the most consistent way to handle downscaling. Are you suggesting continuing to do other aspects of the downscaling over bare land columns, but not the longwave downscaling? What would the physical justification be for not doing the longwave downscaling over the bare land columns?
>>>>> 
>>>>> 
>>>>> Downwelling longwave radiation is controlled primarily by the existence and thickness of clouds. So by downscaling longwave radiation you’re downscaling a cloud property. However, clouds also modulate shortwave radiation and this is not accounted for, which makes the current approach inconsistent. I believe this inconsistency currently leads to a net loss of radiation over the ice sheet, with a considerable magnitude (~ 0.5 W/m2 year round, equivalent to about 80 Gt/year of melt).  
>>>>> 
>>>>> Now that I think about it, an alternative approach could be to introduce a lapse rate for SW transmittance, based on .e.g. Figure 6 of this paper <http://www.the-cryosphere.net/10/2379/2016/tc-10-2379-2016.pdf>. This could compensate the LW radiation loss by an increase in SW radiation. Unsure how to relate the SW transmittance to SW downwelling though. 
>>>>> 
>>>>> Leo
>>>>> 
>>>>> 
>>>>> 
>>>>>> On 23 Jan 2017, at 13:43, Bill Sacks <sacks at ucar.edu <mailto:sacks at ucar.edu>> wrote:
>>>>>> 
>>>>>> Hi Leo,
>>>>>> 
>>>>>> I'm adding liwg-core to the cc list here, to include others in the discussion.
>>>>>> 
>>>>>> Within the CISM domain, the elevation of the bare land portion of the grid cell comes from CISM. Jeremy and I discussed this extensively, and this felt like the most consistent way to handle downscaling. Are you suggesting continuing to do other aspects of the downscaling over bare land columns, but not the longwave downscaling? What would the physical justification be for not doing the longwave downscaling over the bare land columns?
>>>>>> 
>>>>>> Bill S
>>>>>> 
>>>>>>> On Jan 23, 2017, at 5:30 AM, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>>>>>>> 
>>>>>>> Hi Bill,
>>>>>>> 
>>>>>>> I was wondering about the topographic height of the bare land portion of the grid cell. Is this set by CISM or is is equal to the grid cell mean height? Either way, I think in general this height will be lower than that of the glaciated portion, so more radiation goes the the bare land (low elev) than to the glacier part (high elev). To my current understanding, this is a design decision with big implications ( -5 W/m2 over the entire ice sheet) so we may need to reconsider this. My proposal would be to turn off downscaling for the bare land columns, just for the glaciated columns. Has this ever been tested? If not, I would like to test the effect of this. 
>>>>>>> 
>>>>>>> Thanks,
>>>>>>> Leo
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>>> On 21 Jan 2017, at 14:41, Bill Sacks <sacks at ucar.edu <mailto:sacks at ucar.edu>> wrote:
>>>>>>>> 
>>>>>>>> Hi Leo,
>>>>>>>> 
>>>>>>>> If I understand the problem right: You need to look at grid cell means for the longwave radiation downscaling to conserve: the adjustments are made over the bare land portion of the grid cell, too, so those need to be included in your averages to get conservation.
>>>>>>>> 
>>>>>>>> Bill S
>>>>>>>> 
>>>>>>>>> On Jan 20, 2017, at 6:12 PM, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>>>>>>>>> 
>>>>>>>>> Hi Bill,
>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> To make sure I understand:  It’s currently the case that in glacier columns with fractional snow cover, any shortwave radiation that ought to be going into the ice (or “soil”) is simply thrown away?  And the fix, at least for now, is to apply all this radiation to the snow cover until the snow is gone?
>>>>>>>>> 
>>>>>>>>> That is correct. Following this reasoning, the imbalance is greatest when the fractional snow cover is smallest, right before bare ice appears. 
>>>>>>>>> 
>>>>>>>>>> Looking at your pdf plot, it seems there is a mean ice-sheet-wide imbalance of ~0.1 W/m2 during summer and fall.  Based on a back-of-the-envelope calculation, that translates to a melting deficit of ~5 Gt/yr, which is less than I was hoping.  But obviously the imbalance is much bigger locally, which could lead to important feedbacks.  Also, with it being late on a Friday, I could have made a mistake somewhere on my envelope :-)
>>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> I get the same value as you do, with the same caveat. 
>>>>>>>>> 
>>>>>>>>> There was a second finding in my exercise which I did not mention in my email because I’m unsure what to think of this yet. That is, the longwave downscaling also causes a SEB residual but only when integrating the _ICE energy flux variables, not with the grid-cell average variables. According to the code documentation, this cannot be true because there is a normalisation in place that forces the downscaled longwave to exactly equal the gridcell value. However, when I turn off LW downscaling the residual disappears. 
>>>>>>>>> 
>>>>>>>>> That said, the residual is about 0.5 W/m2 throughout the year with your Stefan-Boltzmann downscaling approach, and over 0.6 W/m2 with a lapse rate approach. Using the same envelope as before, this translates to about 80-100 Gt/yr of additional melt that we’re missing out on, excluding feedbacks. I don’t have time to look into this over the weekend, but I’m happy to do some more analysis next week.
>>>>>>>>> 
>>>>>>>>> Cheers,
>>>>>>>>> Leo
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>>> Thanks again,
>>>>>>>>>> 
>>>>>>>>>> Bill L.
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>>> On Jan 20, 2017, at 3:21 PM, David Lawrence <dlawren at ucar.edu <mailto:dlawren at ucar.edu>> wrote:
>>>>>>>>>>> 
>>>>>>>>>>> Thanks Leo,
>>>>>>>>>>> 
>>>>>>>>>>> I read through your presentation and I think your logic is sound.  But, as Bill suggests, Sean is really the one who should review this because he knows this code really well.  I think he is in a meeting this afternoon, but I hope he can look at this on Monday at the latest.
>>>>>>>>>>> 
>>>>>>>>>>> Cheers,
>>>>>>>>>>> 
>>>>>>>>>>> Dave
>>>>>>>>>>> 
>>>>>>>>>>> On Fri, Jan 20, 2017 at 2:55 PM, Bill Sacks <sacks at ucar.edu <mailto:sacks at ucar.edu>> wrote:
>>>>>>>>>>> Hi Leo,
>>>>>>>>>>> 
>>>>>>>>>>> Thank you very much for your careful analysis and detailed write-up of this problem. I have only skimmed through this and will let others – particularly Sean – comment in more detail.
>>>>>>>>>>> 
>>>>>>>>>>> I'm not sure if this is related, but I have vague recollections of a discussion with Sean a few years ago about how the fractional snow parameterization should be treated for glacier_mec columns. I recall our feeling that at least some parts of  this parameterization should not apply to glacier_mec columns, since they already have some treatment of subgrid snow variability through the multiple elevation class scheme. I can't recall any more details right now, though maybe Sean remembers more?
>>>>>>>>>>> 
>>>>>>>>>>> Bill S
>>>>>>>>>>> 
>>>>>>>>>>>> On Jan 20, 2017, at 2:51 PM, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>>>>>>>>>>>> 
>>>>>>>>>>>> Forgot to attach the source code that contains the fix
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>> On 20 Jan 2017, at 22:32, Kampenhout, L. van (Leo) <L.vanKampenhout at uu.nl <mailto:L.vanKampenhout at uu.nl>> wrote:
>>>>>>>>>>>>> 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> Hi Dave, Sean and Bill,
>>>>>>>>>>>>> CC others,
>>>>>>>>>>>>> 
>>>>>>>>>>>>> While doing the surface energy analysis for Greenland, I discovered that the surface energy budget was not fully closed. This holds both at the grid-cell integrated level as well as the glacier level (for this, I use the  variables ending in ‘_ICE’). 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> See attachment seb.pdf, panel SEB residual. Note that the imbalance occurs in summer, during the melt season. 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> I reproduced this problem in a single column run, the results are attached. I believe the problem lies with the implementation of fractional snow and how glaciers are handled, namely by setting frac_sno_eff = 1.0. I hope you agree with the analysis. I also suggested a workaround which appears to work without problems in another ongoing BG run (2+ years). We are seeing slightly higher melt rates now over Greenland, which is good. How substantial the changes are is left to be seen.
>>>>>>>>>>>>> 
>>>>>>>>>>>>> In the long-term a better solution may be desired. 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> Cheers,
>>>>>>>>>>>>> Leo
>>>>>>>>>>>>> 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> 
>>>>>>>>>>>>> <swrad_bug_glaciers.pptx>
>>>>>>>>>>>>> 
>>>>>>>>>>>>> <seb.pdf>
>>>>>>>>>>>> 
>>>>>>>>>>>> <SurfaceRadiationMod.F90>
>>>>>>>>>>> 
>>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> ---
>>>>>>>>>> William Lipscomb
>>>>>>>>>> Los Alamos National Laboratory
>>>>>>>>>> Group T-3, MS B216
>>>>>>>>>> Los Alamos, NM 87545
>>>>>>>>>> 505-667-0395
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>> 
>>>>>>>> 
>>>>>>> 
>>>>>> 
>>>>> 
>>>> 
>>> 
>>> ---
>>> William Lipscomb
>>> Los Alamos National Laboratory
>>> Group T-3, MS B216
>>> Los Alamos, NM 87545
>>> 505-667-0395
>>> 
>>> 
>>> 
>>> 
>> 
> 
> ---
> William Lipscomb
> Los Alamos National Laboratory
> Group T-3, MS B216
> Los Alamos, NM 87545
> 505-667-0395
> 
> 
> 
> 
> _______________________________________________
> Liwg-core mailing list
> Liwg-core at cgd.ucar.edu <mailto:Liwg-core at cgd.ucar.edu>
> http://mailman.cgd.ucar.edu/mailman/listinfo/liwg-core

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cgd.ucar.edu/pipermail/liwg-core/attachments/20170124/0388c33b/attachment-0001.html>


More information about the Liwg-core mailing list