[Liwg-core] latest #134 with reset snow thickness simulates N ablation areas

Miren Vizcaino M.Vizcaino at tudelft.nl
Thu Mar 23 09:15:25 MDT 2017

Hi Bill and All,

I copy Raymond’s response to (1)

"Just to clarify on point (1), what I meant with the minimum amount was not the ‘all-time-minimum’, but the location that has the least snowfall, has on average (over a 20-year period) 35mm (standard deviation: 11mm) of snowfall during Sep-Jan. For comparison, the location with the most snowfall in this region has on average 75mm of snowfall during this same period (standard deviation: 20mm). This means that if we initialize with 100mm, we initialize with a snow mass that is 1.3x - 3x more than what is realistic in this area, and also more than snowfall + standard deviation. Sure, there is also the possibility to pick a number between 35mm-100mm, but why not ensure that every location at least have the possibility to get to seasonal snow cover in the first year, rather than artificially impose a snow cover that represents extreme snowfall in this region?"

Now my response to (1):  Northern Greenland is a dry place, so don’t expect large amount of snowfall variations. The mass balance is very sensitive to accumulation, as we saw before.

As far as I understand, RACMO starts with zero snow thickness in September. Since we cannot do this, we cannot afford putting excessive snow thickness when starting in January.

Regarding (3), Raymond found out that the netCDF format for CISM (scale factor) was responsible for the discrepancy. Maybe worth to be warned of to other CISM users (?)

Regarding (4) I cannot attend.

And now an update: Leo will be testing n_melt and Raymond on repartition.

Regarding Jeremy’s question about stability of this result: we need to run past year 20, but first we need to do the testing for n_melt and repartition. At the moment the model is generating too much ice melt, (because of the SCF), and we need to work on that.

Thanks, Miren

On Mar 22, 2017, at 10:26 PM, Bill Sacks <sacks at ucar.edu<mailto:sacks at ucar.edu>> wrote:

Hi Miren, Raymond and others,

A few thoughts:

(1) Regarding initialization at 35 mm: If we truly find that initializing at 35 mm allows for ablation areas, whereas initializing at 100 mm causes permanent snow cover – if the model is really this sensitive to initial conditions, then I think we have bigger problems. My understanding is that you chose 35 mm because it was the minimum amount of snow fall from Sept - Jan in this region. But that suggests to me that, at many places and times, we'd expect greater than 35 mm SWE on the ground on Jan 1. If the model is truly this sensitive to initial snow cover, then presumably a single year with higher-than-usual Sept - Dec snowfall would lead to incurable permanent snow cover, following the logic in this email thread.

My point here is that I am hesitant to believe that it is important to initialize at 35 mm rather than 100 mm unless I see an clean experiment showing its effect. On the other hand, I also don't object to initialization at 35 mm if there is agreement to do that.

(2) Regarding the initialization procedure: Thanks for the clarification about doing this via setting h2osno_max. I definitely see how that is a better way to ensure self-consistency. However, I don't think this is something we can hope to allow out-of-the-box, and it may meet a bit more reluctance from Dave Lawrence and others in the land group: In order to avoid hitting the ocean with a massive runoff signal, we'd need to do this via an additional offline run that is run for a few days (or somewhat longer) to allow the runoff to make its way out of the system, before restarting the coupled run. I would need to check with others as to whether this is an acceptable step in the spinup procedure. If your additional experiments show that resetting the snow pack (without any other changes) has the benefits we want, then I'll pursue this further.

(3) Regarding the discrepancy that Miren mentions between the integral of (SMB < 0) and QICE_MELT: Without knowing exactly how you did these integrals, it's hard for me to say. In general, I don't think I'd expect these to be exactly equal: I don't think the current elevation class interpolation makes any guarantees about this – though this changes with the new scheme that Bill Lipscomb has implemented. But I'd still expect the integrals to be closer than what you gave. I wonder if this is due to missing some weighting factors in computing the integral on the CLM side? For example, since these QICE variables apply only over glacier landunits, you need to multiply by the area of the glacier landunit (PCT_LANDUNIT(4) on the CLM history files).

(4) There will be a co-chairs meeting tomorrow (Thursday) at 10 am MDT, where I hope to learn more about where we are on the timeline of the CESM2 spinup, and thus how much more time we have to finalize all of this. Feel free to call in if you'd like to contribute to the discussion.

Bill S

On Mar 22, 2017, at 1:54 PM, Lenaerts, J.T.M. (Jan) <j.lenaerts at uu.nl<mailto:j.lenaerts at uu.nl>> wrote:

Hi all,

Thanks Raymond and Miren for the great work.

I am a bit cautious on using this 35 mm. In this case, i.e. without repartitioning and with N_MELT=0.5, this seems to work, but I wonder what happens as soon as one of these two is reset to default values.

Switching off repartitioning is undesirable, since that would create a huge discrepancy between land and glaciers (assuming we only switch it off for glaciers). We still have some room to play with N_MELT.



On 22 Mar 2017, at 17:39, Jeremy Fyke <garmeson.lanl at gmail.com<mailto:garmeson.lanl at gmail.com>> wrote:

Hi Miren

This is great.  To what extent do you think this represents an equilibrated snow pack/ablation area?  Ie are you guys confident the snow pack isn't 'filling in' still, and that this ablation area would persist indefinitely if this simulation was continued?


On Wed, Mar 22, 2017 at 10:13 AM Miren Vizcaino <M.Vizcaino at tudelft.nl<mailto:M.Vizcaino at tudelft.nl>> wrote:
Hi Bill, and All

The latest simulation is able to sustain N Greenland ablation areas and seasonal snow over ice-free areas


Raymond is running another resetting simulation with the repartition on.

For the resetting technique, please check with Leo if anything remains unclear. Raymond was applying Leo’s capping approach that I assume will adjust all variables.

One important remaining issue is the mismatch between downscaled SMB<0 and QICE_MELT. Bill and Bill, do you have any ideas of what is going on here? Could it be related to still having acab=0 over the

accumulation areas?

- integrated ablation for the downscaled SMB (sum of SMB<0) = 70 Gt; however, integrated QICE_MELT=300 Gt (maybe due to differences in how the 1-SCF is included in the computation?)

Regarding refreezing, the spatial map and amounts seem to largely agree with RACMO (please see slide 6 of the PDF). The % using melt as the sum of QICE_MELT and snow melt is in the right ballpark.

I’d like very much to hear from Jan, Leo and/or Michiel on the thickness initialization amount. My take here is that, giving the delicate mass and energy balance in northern Greenland, and that we don’t start the runs in September, it is critical to follow Raymond’s recommendation.

Thanks, Miren

On Mar 22, 2017, at 3:28 PM, Raymond Sellevold - CITG <R.Sellevold-1 at tudelft.nl<mailto:R.Sellevold-1 at tudelft.nl>> wrote:


Yes, I will also do an experiment with n_melt = 1, to see the effect of resetting the snowpack more clearly.

There have been several experiments by Leo where the snow has been reset to 100mm over the tundra, with the outcome that the snow keeps on growing. The reason seems to be because the total melt in this region is the same as the yearly snowfall and when areas that have a total snowfall of 35mm w.e. between end of melt season and January are initialised with 100mm in January, you end up with more snow than the model is able to melt in the summer, thereby causing the permanent snow cover. The same story applies for the northern ablation areas.

Currently, I have adapted Leo’s method of doing this: h2osno_max is set to 35.0 and I run the model for one day, such that the model itself calculates the snow. I figured this was a better idea then editing the restart files since h2osno is not the only variable related to snow.


On 21 Mar 2017, at 22:16, Bill Sacks <sacks at ucar.edu<mailto:sacks at ucar.edu>> wrote:

Hi Miren and others,

Thanks a lot for this.

This sounds promising, but it also sounds like we need a clean experiment to just see the effect of resetting the snow pack. In addition to turning the repartitioning back on, should n_melt be kept at 1 for this?

I'm also wondering if a reset value of 100 mm would have the desired effects, since that was the value that I thought we agreed on. I guess we could use 35 mm, but it seems confusing to have this reset value differ from the cold start value, which is currently set at 100 mm. Should we also switch the cold start value to 35 mm in this case? I'm interested to hear what others think about this – particularly Leo and Jan, who had thoughts in the past about the appropriate snow initial conditions to avoid undesirable heating of soil under snow.

Also: I realized that there could be some subtleties with how exactly we should reset the snow pack. If we decide to move ahead with this solution, I'd like a more detailed description of how this should be done. Specifically: it's not clear to me that just resetting h2osno is enough, because there are a lot of different snow variables, and resetting h2osno without resetting some other variables might cause things to be in an inconsistent state.

Related variables that are integrated across the snow column are int_snow_col, snow_depth_col, and possibly others. And then there are many variables that are stored layer-by-layer for the snow pack. Does anyone know if any of these variables need to be reset for things to be in a consistent state? (From a quick look through the code, it looks like it should be okay not to reset int_snow_col, but I'm not positive of that, and I'm not sure about the other variables.)

Bill S

On Mar 20, 2017, at 10:41 AM, Miren Vizcaino <M.Vizcaino at tudelft.nl<mailto:M.Vizcaino at tudelft.nl>> wrote:


Raymond did a new run with #134 physics and reset of initial snow thickness to 0.35mm over the GrIS and n_melt=0.5 and rainfall repartition off.

The simulations has gone through 20 years. Ablation areas of GrIS are well captured and amount to 10% of ice sheet area.

Other highlights:

- Seasonal snow cover over tundra

- Some high accumulation areas in the SE have reached maximum snow thickness (10 m w.e.)

- integrated ablation for the downscaled SMB (sum of SMB<0) = 70 Gt; however, integrated QICE_MELT=300 Gt (maybe due to differences in how the 1-SCF is included in the computation?)

- refreezing amounts to 40% of snowmelt + QICE_MELT + rainfall and 57% of snowmelt + downscaled ice melt + rainfall.  Still a high ratio. Will this ratio change as snow thickness equilibrates ?

Next steps:

-  sensitivity to switching on rainfall repartition

Thanks, Miren

Liwg-core mailing list
Liwg-core at cgd.ucar.edu<mailto:Liwg-core at cgd.ucar.edu>
Liwg-core mailing list
Liwg-core at cgd.ucar.edu<mailto:Liwg-core at cgd.ucar.edu>

- - - - - - - - - - - - - - - - - - - - - - - - - - -
Jan Lenaerts
IMAU, Utrecht University || University of Colorado
@lenaertsjan<https://twitter.com/lenaertsjan> || website<http://www.colorado.edu/lab/icesheetclimate/>

Liwg-core mailing list
Liwg-core at cgd.ucar.edu<mailto:Liwg-core at cgd.ucar.edu>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cgd.ucar.edu/pipermail/liwg-core/attachments/20170323/1a4c66f4/attachment-0001.html>

More information about the Liwg-core mailing list