Dead Arctic Vegetation

[wwieder]

Summary:

We have persistent dead Arctic vegetation in CLM.

• This was an issue in CESM2 because with dead vegetation soil C pools are also zero, leading to lower than expected soil C stocks in permafrost regions.
• Similar patterns appears to be occurring in tests for CESM3 simulations too.
  These plots are for land only BGC vs. SP runs
  

Causes:

• Late melting snow (and cold soil temperatures?) prevents plants from growing.
• Without vegetation this creates a biophysical feedback where snow continues to persist too late in the summer
• Preliminary results suggest that living vegetation (forced by using CLM-SP simulations) can accelerate snowmelt and extend the snow free period.

Question:

• Can we initialize coupled model runs with living arctic vegetation such that regional climate allows for plant survival and slightly warmer conditions? Specifically, we're hoping to create CPL_HIST output from simulations initialized with offline CLM-BCGcrop-GWSP3 simulations with living arctic vegetation. Output from the CPL_HIST can then be used to spinup CLM6 for CESM3?

Potential solutions:

• Plant parameters (FUNfracFixers, # 2; frootleaf # 7; leafcn, others) and discussion
• Phenology modifications: allow later greenup # 5
• Phenology modifications: relax seasonally decid triggers # 15
• Snow thermal conductivity (Sturm, # 4) & # 13, Coupled) and discussion
• Soil hydrology parameters (fff # 16, d_max # 17) and discussion
• snow parameters (upplim_destruct_metamorph # 22)
• combination of most impactful changes (Sturm + phenology and parameter changes: # 24 & # 25, coupled)
• revert to CLM4.5 snow density scheme (this is not a preferred option).

Background from the coupled model discussion

Identifying the dead vegetation feature:

This thread was initiated on the CTSM repository regarding coupled model runs, but migrated here to focus on land-only simulations.

@olyson posted:
The next in this series of simulations is b.e23_alpha16b.BLT1850.ne30_t232.033, as discussed in NCAR/amwg_dev#356.
From the land viewpoint, we are initializing using the spunup initial file recently generated from a coupler history forced case, the tag is cesm2_3_alpha16b which uses ctsm5.1.dev130, and SourceMods to accommodate the new surface dataset.
I ran some diagnostics before cheyenne went down (8 years) compared to the CESM2 piControl here

Regarding the lack of vegetation at high latitudes in Russia, the vegetation looks even less well-established in this most recent simulation (top plot) compared to a segment of the piControl (bottom plot) (JJA). Both have a "hole" at 60-90E. The coupler history spinup looks similar to b.e23_alpha16b.BLT1850.ne30_t232.033.

CLM5.2 forced with CESM3* CPL_HIST

CESM2 PI control

The "hole", as defined below, is 98% vegetated and is comprised of roughly 50%/40% deciduous boreal shrub and C3 arctic grass. Timeseries plot shows that snow cover is not near zero until Aug-Sep.

Just to start a discussion. I'll be able to do more analysis once cheyenne is back up.

[wwieder]

Regional climatology of land only and coupled runs

CPLHIST (10 years at end of 1850 coupler history spinup).
CESM2 BHIST.001, PI control
CTSM51_GSWP, yrs 1851-1860 from historical.

posted by @olyson
annual cycle of snow cover fraction, snow depth, snow water equivalent, exposed leaf area index for the dead vegetation "hole".

~1850 plots:

1995-2014 plots:

And ideas contributed in several threads by @dlawrenncar:
Thanks. These are all helpful. I think these plots support the idea that there is more snowfall in the coupled simulations than GSWP3, especially in the fall. We should obviously be cautious with GSWP3 snowfall since there wouldn't be much data going into it. There is also an indication that snow melt starts earlier (quite a bit of snow melt in May in land-only, but little snow melt in May in coupled model). So, it seems like we could focus in on the surface energy budget in May as one thing to examine more closely.
Since the CESM2 present day plot indicates that the excessive snow (and snowfall) persists to present day, maybe we can use present day obs to try to diagnose the problem against observations.

Trying to make sense of all the plots is hard, but one thing that I think stands out is that the surface albedo is lower in land-only
runs, even when snow cover fraction is pegged at one for both coupled and uncoupled models. That difference in surface albedo is about 0.1, which would equate to about 20W/m2 more energy absorbed at the surface, which is about the amount of energy going into the melt flux. What's interesting is that even in present-day conditions, there is greater snow melt flux (and lower albedo) in land-only, even when snow depth is actually larger and SWE about the same between coupled and land-only. So, this kind of
possibly implies that the snow is melting more in land-only because the albedo is lower. Now, is the albedo lower because there are trees growing in the land-only run for part of the domain and therefore there is some snow masking, or is it because the aerosol dep is different or is it because the snow grain size growth across the season is different. The albedo difference is there throughout the year, which implies that it is the trees that is reducing the albedo, but that could be checked. So, to diagnose further, we might need to focus in on regions without any trees (if that is indeed the source of the albedo different) or take a look at the snow albedo itself (rather than surface albedo) to see if that could explain things. We may want to have a small meeting to discuss. Seems worth pursuing this now while we are thinking about it since this does substantially mess up the permafrost carbon stocks across much of the Arctic.

Prior to our chat on Thursday, Keith, could you try to see if you could confirm whether or not the differences in albedo between land-only and coupled are due to vegetation masking or to the snow albedo itself. Our hypothesis is that the difference in albedo is due to shrubs (and maybe a few trees) surviving spinup in land-only runs, but not in coupled runs, rather than it being a difference in actual snow albedo.

A test, which would be a bit challenging, would be to run with CLMSP mode in coupled (or maybe just CAM-CLM) simulations. If we see the same earlier melt out, then that would support the idea that the vegetation not surviving is leading to the lengthy snow-on-ground season. If that is what is happening, then we need to think about why the vegetation is dying in coupled run and if there is something we could do to fix that.

[wwieder]

Does snow melt earlier in SP simulations?

Again, these are CESM3-CLMsp simulations

posted by @olyson
Some plots for discussion that include results from the CESM3 SP simulation (b e23_alpha16b BLT1850 ne30_t232 034SP).

The earlier melt-out looks a bit more pronounced later in the simulation (years 5-6):

[wwieder]

Initialize coupled model with healthier Arctic vegetation?

@wwieder , @dlawrenncar , and @olyson met on 08/17/23 to discuss ways we might initialize the coupled simulations with conditions that are conducive to increased vegetation viability at high latitudes. As shown in the plot below, our new coupler history spinup does not have much live vegetation at high latitudes in boreal summer:

Initially, we thought we could try a coupled simulation in which we initialize with conditions from a recent GSWP3V1 forced simulation in which the vegetation is healthier. However, as shown in the plot below, even the vegetation in this GSWP3V1 forced simulation is somewhat anemic when compared with a recent GSWP3V1 forced historical simulation for present day with MODIS data as below:

We don't seem too far off, except in far eastern Siberia. Interestingly, we see a small hole in the MODIS data located in about the same spot as our larger hole, possibly due to the presence of lakes (I haven't looked at lake fraction here). The observations have viable vegetation further north in the Canadian Arctic than in our simulation. There is another interesting lack of vegetation in our simulation compared to MODIS at 115-135E, near 60N, and also further south and west.

What are 'simple' ways to try and initialize with healthier vegetation?

[wwieder]

Modify penology to allow leaf-out after summer solstice, #5

@wwieder proposed that we try a GSWP3V1 forced simulation with a change to the SeasonalDecidOnset code so that onset can occur after summer solstice, by commenting out this code:

if (onset_gddflag == 1._r8 .and. ws_flag == 0._r8) then
onset_gddflag = 0._r8
onset_gdd = 0._r8
end if

I've done a AD, pAD spinup with this. I don't see too much of an improvement. The pAD spinup is not quite done but I doubt it's going to improve much. The Siberia hole is filled in a bit, but I also did a control with the same code and it looks quite similar to this so I think any improvement here is the result of a newer code base.

Given these modest results @rosiealice asked:
Is there anything in the PPE that might help here? I remember that the root:leaf ratio and root turnover were very influential on the health of the shrubs. Though maybe these were already dialed to the high end of plausible when we were dealing with similar issues in CLM5/CESM2?

[wwieder]

PPE insights

@Oleson started looking at the web site cited by Daniel in the PPE paper draft:

@swann noted: You are welcome to look at the coupled PPE runs for insight which cover 18 parameters with 1850 conditions. They are located here: /glade/campaign/cgd/tss/czarakas/CoupledPPE/coupled_simulations

@djk2120 and @katiedagon peaked at the latin hypercube PPE and found that our default simulation in the latin hypercube has peak LAI of 1.22, and there's lots of ways to increase it, up to and above 3.0. Lowering leafcn and lowering froot_leaf look like the best bets.
/glade/campaign/asp/djk2120/PPEn11/AF1855/hist/PPEn11_AF1855_OAAT0034.clm2.h0.2005-02-01-00000.nc

Here's the rankings plot, for August LAI in the subset under cold forcing, updated with Katie's fix. They are sorted by the +LAI effect

code here:https://github.com/djk2120/oaat_clm5_ppe/blob/main/pyth/template.ipynb

@dlawrenncar noted:
Interesting. I guess you may have already looked to see if any of the parameter sets that produce the higher LAI in 1850 have a less strong increase in LAI over the historical period so as to not produce such a large LAI in present day. I guess overall I wouldn't be surprised if this is just a limitation of the resolution of the number of pfts and that there is more adapted vegetation as you move northward and furthermore that there has likely been species shifts as climate has changed.

Also, the zsno is interesting, but looking at Ronnie Meier's paper for the new roughness param that is coming in soon, he is actually recommending zsno getting smaller (0.0007), but also that there is an increase in zsno as snow starts to melt, which might accelerate snow melt during the melt season. https://gmd.copernicus.org/articles/15/2365/2022/ (see section 2.4)

Do we need to test this with new roughness parameterization?

[djk2120]

updated plot, with corrected sorting algorithm

[katiedagon]

Awesome, thanks Daniel! @olyson now leafcn and froot_leaf are showing up at the top.

[wwieder]

Modify parameters based on PPE suggestions, #1 & #2

@olyson posed:
Results from some spinups. CPLHIST is the coupler history spinup we ran to provide initial conditions for fully coupled simulations, GSWP is a land-only control forced by GSWP, and then there are two perturbed parameter runs informed by the PPE. I see improvements in TLAI in both of the parameter runs, in particular a boost in TLAI in July and August in the Siberian hole (good job PPE!).

[wwieder]

I kind of agree with @dlawrenncar here, the changes to TLAI in JJA in regions that were previously dark blue (<0.1) are pretty subtle- maybe a big in Canadian Arctic and the impacts on albedo and snow melt are pretty negligible. Do we want to:

1. Dig in with additional diagnostics that @dlawrenncar is suggesting?
2. Push this farther in land-only simulations to get leafier conditions in the high Arctic?
3. Try running coupled simulations with current, land only initial conditions that have more leafy vegetation?
4. other ideas?

[wwieder]

@olyson posted:
I agree we're not seeing much movement in albedo and snow melt. Good ideas, sounds like we'll be talking about this more this week...
I've added HTOP to the annual cycle plot below. FUN_fracfixers has about a 10% increase in summer, less so for froot_leaf.

Defining polar as 60-90N and using a monthly max LAI threshold of 0.2 (definition of pretty much dead) and PCT_GLACIER < 50%, the amount of area is 3.28442e+06 km^2. The area decreases by 5.7% in FUN_fracfixers and 12.8% in froot_leaf.

[wwieder]

I like this 'dead area' metric and the HTOP plot, Keith. Thanks.

In the regional climatology plots it's interesting that HTOP is > Snow Depth in all of the offline cases. Given the dead area in the region, it makes me wonder what HTOP, LAI and snow dynamics look like in dead vs. undead grid cells over this domain? Are growing verdant Arctic tundra in more favorable grids to try and get vegetation to grow at all in less favorable climates? Alternatively, are the grasses, that can be buried by snow, (but not shrubs) responding more strongly to these parameter changes?

To clarify, I'm not asking for additional anaysis here, Keith. Just typing to avoid rubbing my head to hard...

[olyson]

Annual cycle of soil temperature at layer 3, requested by @wwieder

[olyson]

Script to generate polar plots: On cheyenne, /glade/u/home/oleson/misc_programs/CTSM5.1DEV/polar.ncl

[wwieder]

Lots of ideas in the meeting yesterday.
@samsrabin suggested removing or relaxing requirements for any/all of these switches in the seasonally deciduous phenology triggers may help diagnose the problem.

         else if (season_decid_temperate == 0 .and.  onset_gddflag == 1.0_r8 .and. &
                soila10 > SHR_CONST_TKFRZ .and. &
                t_a5min > SHR_CONST_TKFRZ .and. ws_flag==1.0_r8 .and. &
                dayl>min_critical_daylength_onset .and.  snow_5day<snow5d_thresh_for_onset) then
           do_onset = .true.
        end if

@rosiealice noted it may be helpful to get a closer look at climatologically conditions for dead grids to compare with adjacent ones that seem OK could also help identify potential causes for the dead grids.

@lawrencepj1 suggested also looking at CLM5 results to see if these holes persisted.
Not sure this is the right comparison to make, @olyson but it does look like summer LAI in CLM5 may be higher than with a 5.1 tag

[samsrabin]

Thanks for the summary, Will! Just wanted to add that I'm not sure straight-up removing such switches would be a good solution, but it might help us understand what the issue is sensitive to.

[wwieder]

agreed, just to diagnose the problem, corrected above

[wwieder]

maybe towards this end, @olyson can you extend one of your runs adding SOIL10, and A10TMIN, and SNOW_5D to history outputs, as these are the triggers that have to be satisfied. It may be helpful to look at these with daily frequency?

[rosiealice]

I agree it would be broadly interesting to know where and by how much these controls are delaying the onset? I guess with frozen soil we would have Btran=0, and so no gpp, but the leaves would at least nave fully expanded by the time the snow melted?

[aswann]

Insight from the coupled PPE

I took a look at how Arctic LAI is modified in the coupled PPE (includes the atmosphere) under pre-industrial conditions.
The coupled PPE runs are located here: /glade/campaign/cgd/tss/czarakas/CoupledPPE/coupled_simulations

As Dave predicted, our control run has dead plants in Siberia and the Canadian shield (summer JJA LAI is zero) because it was initialized from the PI control. Despite this, there are a few parameters in the coupled PPE which do boost summer LAI (nstem, jmaxb0, fff) at least over Siberia.

@wwieder was perplexed about why nstem increases LAI - I would guess that this is a feedback from nstem increasing surface temperatures, but would need to dig into it more. Nstem is part of biomass heat storage and we found that it had some large temperature impacts.

PI Control run summer (JJA) LAI

Delta summer (JJA) LAI for each parameter perturbation

[katiedagon]

Thanks for looking at this @aswann. Are the min parameter perturbations listed first the panel plot, then the max perturbations? If so, reducing nstem and fff to boost LAI matches what we found in the land-only simulations.

[aswann]

Yes, generally it's the min value first and the max second, but a few of the parameters are out of order (I'm not sure why!). So min nstem and min fff.

[dlawrenncar]

Nice plots. I had the same question as @katiedagon. Puzzling that nstem is so impactful. @aswann Could you make a plot of spring time albedo and or snow melt. That might tell us whether or not any specific parameter change, especially ones that lead to more LAI, feeds back onto the rate that snow melts in the spring. If it does, then that would provide some support for our hypothesis that if we could get vegetation to survive in some of these regions, it would help produce a better climate for the plants to live, a virtuous cycle!

[aswann]

Sure - What variable should I look at for snowmelt?

[rosiealice]

I feel like the impact of nstem is a good indicator that warming up the soil would help here...

[dlawrenncar]

FSM is snow melt heat flux. We are looking for a signature of earlier snowmelt. See the timeseries plots above, so not totally sure how to best plot that. You might need to look month to month in the spring (April to June).

…

On Fri, Sep 8, 2023 at 12:09 PM Abby Swann ***@***.***> wrote: Sure - What variable should I look at for snowmelt? — Reply to this email directly, view it on GitHub <#3 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFABYVBZ3KJBVS3XSJRQPPTXZNNL7ANCNFSM6AAAAAA4NT723Y> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[wwieder]

FSNO (snow covered fraction) or SNOWDP (snow depth) could also be good analogues. Maybe focusing on a single month (e.g. June)? Thanks Abby

[aswann]

We don't have those variables processed yet (FSNO, SNOWDP, FSM), but can try to get to it next week .

[olyson]

Regarding CLM5, this is a plot of the difference in JJA TLAI between CLM5 at the end of the post-SASU spinup (that was used to initialize the CLM5 historical) and the GSWP3 CTSM simulation:

In the Siberian hole, it looks like the northern part is somewhat better in CLM5 and the southern part is somewhat worse compared to CTSM.
CLM5 looks like it generally has less TLAI to the east.

[olyson]

Some preliminary (the pAD is not quite spunup but I ran an 1850 anyway) results from the Sturm snow thermal conductivity simulation. The first plot shows the difference between Sturm and the control for DJF TSOI layer 3. Definitely warmer. The second plot shows the difference in JJA TLAI. Some improvements but less than we got with FUN_fracfixers and froot_leaf. The third plot shows the annual cycle. Very slightly earlier snow melt. The "pretty much dead" area decreases by 14.6%.
We could try an nstem simulation as suggested by the coupled PPE, and/or start looking at phenology triggers by extending one of the runs adding SOIL10, and A10TMIN (this should actually be A5TMIN as A10TMIN is only used for crops as far as I can tell), and SNOW_5D to history outputs as suggested by @wwieder , or maybe it's time to set up some single-point simulations for dead and ok grid cells as suggested by @rosiealice ?

[wwieder]

This is the first indication that snowmelt is occurring earlier, which is encouraging. If I'm not mistaken, the reduction in %dead (14.6%) is greater than the previous PPE changes reported above where "The [dead] area decreases by 5.7% in FUN_fracfixers and 12.8% in froot_leaf". Here are my suggestions:

• @olyson maybe you can extend this simulation with some daily frequency history output to look at what's potentially still blocking the phenological triggers? At this point, I'm assuming it's still snow depth, but having more information would be helpful, and effectively let us look at the same kind of output one could get from single point runs.
• I think it's helpful to try a new simulation that reduces nstem, given results provided by @aswann, especially since this parameter also shows up in the OAT land only results. What magnitude perturbation should we try here (a 20% reduction)? I'm assuming this a PFT specific parameter? I'll create a new issue for this, reduce nstem (enter the name of the run here) #9.
• Finally, I feel like we need some independent methods to evaluate the changes in soil temperature that are caused by the snow conductivity changes. I'll start this in a new discussion thread Evaluate effects of Sturm snow conductivity changes #10.

[olyson]

@djk2120 , @katiedagon , I'd like to confirm that the PPE minimum value for nstem was 0.03.

[djk2120]

I'm realizing that we actually uncovered a bug with the nstem ranges. 0.03 is not a reasonable value for the grass PFTs. I would say default -50% is a more reasonable perturbation. We did +/- 30% for our transient/LHC ensemble. @aswann do you know what was used for the coupled PPE?

[olyson]

I've extended the control and Sturm simulations to include daily output of A5TMIN, SOIL10, SNOW_5D, and ONSET_FLAG, as shown below for one year. The horizontal dash lines represent the onset thresholds. Consistent with the above results, we can see these thresholds should be triggered earlier in the Sturm simulation (except for A5TMIN which is associated with the 2-m temperature). I'm a bit confused by why TLAI starts increasing earlier than the thresholds would indicate, but I also see that ONSET_FLAG becomes non-zero at about the same time, so maybe each deciduous pft is triggering at different times or the onset flag isn't consistent throughout a given day. Although for the first hypothesis, the SOIL10 and SNOW_5D variables are column-level and should be shared by all of the pfts. Might be useful to drill down to hourly pft-level output.
Another possible way to get earlier onset would be to increase the SNOW_5D threshold from 0.1 to something larger.

[wwieder]

@olyson this is great! Can you point the script to focus on a single living and dead grid in the domain now? I wonder how much we're confusing things by averaging across a region, when the issue is at individual points?

[olyson]

Oh, well, that's true, we are averaging across soil columns in different grid cells that could have different forcing. It's a bit difficult to find a pair of gridcells with similar pft distributions and close enough in location to have similar forcing. But as an example, see the plots below (it would be better if these were on the same single plot, which I will get to at some point). The plots below show output for two single grid cells with similar pft distributions (top plot has 87% BL deciduous boreal shrub+C3 grass and 11% bare soil, while the bottom plot has 96% BL deciduous boreal shrub+C3 grass and 4% bare soil) and are at the same latitude and about 5deg apart in longitude. The top plot/gridcell has anemic vegetation in summer while the bottom plot/gridcell is relatively healthy.
The A5TMIN (calculated from the 2-m temperature) appears to be similar in the two gridcells which would seem to indicate the temperature forcing at least is similar. The main difference seems to be that the thresholds are achieved earlier in the year and hence the vegetation has more time to grow in the bottom plot/gridcell. This might simply be because there is more snow to melt in the anemic gridcell as shown in the top plot. So, somewhat inconclusive. In both cases, Sturm does appear to cause the thresholds to be reached sooner. I'll try to look for a better pair of gridcells, but I imagine that having similar forcing will be key, e.g., temperature, precipitation.

[rosiealice]

Do I remember right that there is snow covered fraction as well? Is this a function of only snow depth?

I feel like IRL this is all about the spatial dynamics of some areas melting and leafing out and maybe changing the albedo as they do so (at least, that's how it seems to work around the shrubs in my almost-Siberian garden at 250m a.s.l. in SW Oslo ;) )

Which is a roundabout way of saying I think we should try relaxing the control of grid mean snow depth on leaf out... I guess it is there for a reason, to stop things resorting to death under thick snowpack, and there is presumably an optimum value (that is probably also related to the rate of change in snow depth in spring).

[olyson]

Since we are mainly interested in the effects of the snow albedo changes in the ABCDE_blk_ABCD simulation (see #37), I've generated some diagnostics and plots for the end of the AD for that simulation.
Standard diagnostics of this compared to the ABCDE simulation:
https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD/lnd/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD.271_280-ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_1850.11_20/setsIndex.html

Polar plots of this (top) compared to ABCDE (bottom):

Annual cycle plots:

Percent decrease in dead vegetation for the ABCDE and ABCDE_blk_ABCD:
(0) Percent Decrease in ABCDE_GSWP dead_veg (km2): -43.8176
(0) Percent Decrease in ABCDE_blk_ABCD_GSWP dead_veg (km2): -59.1189

So, some improvement there, and we see a small reduction in snow albedo for ABCDE_blk_ABCD and a bit earlier snow meltout compared to ABCDE, as we would expect from the single-point snow albedo simulations (#36)

[wwieder]

Thanks @olyson,

It seems like we can try to reduce albedo farther. I'm hoping to try to juice some of the snow albedo changes, per Cenlin's suggestions from last week's meeting. It's good to see that results from single point simulations are a good indications of what we'll see in the full model.

What variables go into the snow albedo calculation? Do I even need this for single point runs where we don't have the subgrid variability in snow depth?

[olyson]

We calculate snow albedo from the reflected solar components of snow:
snwalb1= (inptr1->SNOFSRVD + inptr1->SNOFSRVI + inptr1->SNOFSRND + inptr1->SNOFSRNI)/inptr1->FSDS
What happens in your single-point runs? Is FSNO always equal to 1 when there is any amount of snow?

[olyson]

I've generated some diagnostics and plots for the end of the AD for the ABCDE_blk_ABCDE simulation compared to ABCDE_blk_ABCD. These show the effects of raising snw_rds_min from its default of 54.526 to 100.
Overall, there are some decreases in polar albedo in winter/spring. Snow albedo decreases a bit, maybe by a percent or two. Marginal improvements in TLAI and dead area.

(0) Percent Decrease in ABCDE_blk_ABCD_GSWP dead_veg (km2): -59.9121
(0) Percent Decrease in ABCDE_blk_ABCDE_GSWP dead_veg (km2): -60.9082

Standard diagnostics:
https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCDE_1850AD/lnd/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCDE_1850AD.191_200-ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD.191_200/setsIndex.html

Annual cycle plots:

TLAI difference:

[olyson]

There still seems to be some room for decreasing snow albedo:

[wwieder]

Thanks for posting Keith. I agree, it seems like we have more room to reduce albedo.

These results seem to reflect findings from the single point run I did. I'm trying another ABCDE case with snw_rds_min_tbl = 100 in SnowSnicarMod.F90 (previously set to 30). Cenlin didn't think this needed to be changed, and after a few years I'm not sure it will do much for albedo, but we'll see...

[wwieder]

This modification setting snw_rds_min_tbl = 100 actually increase albedo. I'll switch this run with snw_rds_min=200 instead as Mark suggested in my ABCDE case.

[olyson]

I've generated some diagnostics and plots for the end of the AD for the ABCDE_blk_A5BCD simulation compared to ABCDE_blk_ABCD. These show the effects of raising xdrdt from 1 to 5 (and actually taking effect this time).
From the standard diagnostics we can see there are some decreases in polar albedo in winter/spring. The annual cycle plot shows snow albedo decreases a bit, 2-4% in winter/spring. Improvement in dead area.

(0) Percent Decrease in ABCDE_blk_ABCD_GSWP dead_veg (km2): -59.9121
(0) Percent Decrease in ABCDE_blk_A5BCD_GSWP dead_veg (km2): -69.5676

Standard diagnostics:
https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD/lnd/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD.271_280-ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD.271_280/setsIndex.html

Annual cycle plots:

Polar JJA TLAI:

[cenlinhe]

I am a little concerned if we give a too large xdrdt (e.g., =10), it might mess up the snow in mid-latitudes. Is there any evaluation result for mid-latitude snow for the current snow albedo parameter set to ensure it is still fine?

[olyson]

Based on what Mark said back in 2017 when we were experimenting with xdrdt, I would agree with @cenlinhe not to increase xdrdt further. We may need to back it off a bit. Although I'm not sure if this is still applicable in the new SNICAR scheme.

"From a snow microphysics perspective, xdrdt=5 is indeed uncomfortably large, though possibly defensible in extreme situations. A smaller value (2.0-2.5) would be preferable, perhaps as a compromise, if it attains a sufficient amount of high-latitude warming."

[olyson]

Mid-latitude (45S-45N) albedo for gridcells with fsno>0.5 looks pretty reasonable for the control (top plot). Still looks ok for the ABCDE_blk_A5BCD (with xdrdt=5) simulation (bottom plot). Sept in the ABCDE_blk_A5BCD looks low but there are only 7 gridcells that meet the fsno>0.5 criteria. Missing monthly values mean there are no gridcells meeting that criteria.

[wwieder]

Sorry if I missed it, but do you have plots of high Arctic snow albedo similar to this one, @olyson from the ABCDE_blk_A5BCD simulation (after you found and fixed the xdrdt bug)?

[olyson]

No, you didn't miss it. A small decrease. There are 65 gridcells meeting the fsno>0.5 for all months. The DJF and MAM albedo diagnostics show more detail on the changes:

https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD/lnd/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD.271_280-ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD.271_280/set2/set2_DJF_ASA.png

https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD/lnd/ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_A5BCD_1850AD.271_280-ctsm51_ctsm51d150_ne30pg3ne30pg3mg17_GSWP3V1_ABCDE_blk_ABCD_1850AD.271_280/set2/set2_MAM_ASA.png

[dlawrenncar]

This is all promising. I think at this point that we probably need to meet to go over things because I think I have lost track of all the changes and which ones do what. I do think that if we can get a broadly earlier snowmelt and a bit healthier Arctic vegetation that we could probably get something more realistic vegetation states through tuning other parameters or tuning some of the vegetation parameters that have been adjusted. It might be worth a test of one of the latest configurations, though, (xrdrdt =5 and maybe something a bit lower as well) in a coupled run to see what the feedbacks are.

…

On Mon, Dec 18, 2023 at 4:00 PM Keith Oleson ***@***.***> wrote: Mid-latitude (45S-45N) albedo for gridcells with fsno>0.5 looks pretty reasonable for the control (top plot). Still looks ok for the ABCDE_blk_A5BCD (with xdrdt=5) simulation (bottom plot). Sept in the ABCDE_blk_A5BCD looks low but there are only 7 gridcells that meet the fsno>0.5 criteria. Missing monthly values mean there are no gridcells meeting that criteria. plot_snowalbedo_moreobs_annual_cycle_Mid-latitudes_CTSM_1850_Con.png (view on web) <https://github.com/NCAR/LMWG_dev/assets/18671367/665ec613-af14-4e0e-8a0d-16ba2da0222e> plot_snowalbedo_moreobs_annual_cycle_Mid-latitudes_ABCDE_blk_A5BCD.png (view on web) <https://github.com/NCAR/LMWG_dev/assets/18671367/dc6ee845-c4ee-4233-8f7c-de0616c5a7ba> — Reply to this email directly, view it on GitHub <#3 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFABYVDQWEX5UIBKGAHICHDYKDDKVAVCNFSM6AAAAAA4NT7232VHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TQOJRGU3TK> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[olyson]

Per our discussion 01/04/23, I ran 4 more 1850 AD simulations. These were with ctsm5.1.dev160 with Meier2022 on. A Derecho control with ABCDE_blk_A5BCD (issue #47 ), another same as control but revert froot_leaf (C) and FUN_fracfixers (D) 50% of the way back to defaults (ABpt5Cpt5DE_blk_A5BCD, issue #50 ), another same as control but revert froot_leaf (C) and FUN_fracfixers (D) all of the way back to defaults (ABE_blk_A5BCD, issue #49 ), another same as control but using CRUJRA as forcing (issue #48).
Below are polar plots (left to right: GSWP3 control (no changes) on cheyenne, ABCDE_blk_A5BCD with ctsm5.1.dev150 on cheyenne, ABCDE_blk_A5BCD with ctsm5.1.dev160 on derecho, ABpt5Cpt5DE_blk_A5BCD, ABE_blk_A5BCD. In parentheses are the reduction in polar dead vegetation from the control.
Next plots are annual cycles for the GSWP3 simulations.

[wwieder]

@olyson you're amazing. Thanks.

It's hard to see a big difference in the ABpt5Cpt5DE_blk_A5BCD vs. the ABCDE_blk_A5BCD cases.
Let's discuss at the CLM meeting this week and see what the group think about moving forward.

[olyson]

[olyson]

I continued the simulations listed above (d160 at 1deg) in pAD mode with pft-level output for 80 years each. Here is an xlsx table of pft survivability for the last year of the pAD for each simulation. Individual plots of survivability and ANN max TLAI are shown after the table for each simulation.

PFT_Survivability_1.xlsx

[wwieder]

Thanks @olyson! Do you think you can drop the figures into a google directory somewhere? The figure headings are cut off in these uploads.

[olyson]

I've put them in the CTSM google drive folder, in the folder named Dead_Veg_Plots:
https://drive.google.com/drive/folders/1bno_FcsOg4sYmVfGL6ciT0lQuBN2Wv26?usp=sharing

[olyson]

Results from removing the CNPhenology changes (issue #51 ) from our base simulation (issue #48). Standard diagnostics for the end of the ADs are here:

https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d160_1deg_CRUJRA_FLDS_ACDE_blk_A5BCD_1850AD/lnd/ctsm51_ctsm51d160_1deg_CRUJRA_FLDS_ACDE_blk_A5BCD_1850AD_391_400-ctsm51_ctsm51d160_1deg_CRUJRA_FLDS_ABCDE_blk_A5BCD_1850AD_471_480/setsIndex.html

Below is a plot of JJA TLAI for the base simulation (second plot from left), the new simulation (third plot from left) and the difference from the base simulation (rightmost plot). Percent decrease in dead vegetation (as measured from a GSWP3 control simulation) is now -34.2 (compared to -55.7 in our base simulation).
Annual cycle plots follow after that.

[wwieder]

Thanks for posting these @olyson what's your suggestion on how to move forward with phenology?

[olyson]

Given the max LAIs are so high with the CRUJRA version of our kitchen sink simulation (7.2 for NL Deciduous Boreal Tree, 7.4 for BL Deciduous Tropical Tree, 5.9 for BL Deciduous Boreal Shrub, 5.6 for C3 Arctic Grass), and the fact that these are 1850, I wonder if we should back off the changes in CNPhenologyMod.F90 as we've tested in this most recent simulation (CRUJRA_ACDE_blk_A5BCD) and run a historical to see where we are at at present day, and possibly a coupled simulation using the most favored current coupled configuration. It would probably also be useful to generate some new coupler history files once the atmospheric group has converged on a near-final configuration climate-wise so that we can evaluate further changes in the context of the coupled model climate, as @dlawrenncar mentioned. We can discuss more at our afternoon meeting...

[olyson]

Per our meeting today, we will back out the changes to CNPhenology except for the snow depth threshold which we will set to 0.2m consistent with our current simulations. The snow depth threshold will be made a parameter on the parameter file. This is currently a local parameter called snow5d_thresh_for_onset and I propose we keep this name. It will likely need to be added to the CLM5 and CLM45 parameter file as well, keeping the default hardcoded value of 0.1m.

[olyson]

See ESCOMP/CTSM#2348 for the deadveg branch PR.

[olyson]

Results from our latest simulation (phenology changes backed out except for the snow depth threshold; ABsnoCDE_blk_A5BCD). First plot is polar JJA TLAI. Second plot is the latest simulation minus the simulation with all phenology changes backed out (ACDE_blk_A5BCD), so just showing the effects of including the snow depth threshold. A bit of an increase in TLAI in some regions, not much though. There are some decreases in TLAI in the southern part 0-60E but those appear to be due to the difference in tags in these two simulations (d166 and d160) which had some changes to crops, rather than the snow depth threshold. Percent decrease in dead vegetation from the GSWP control is -35.7% in the latest simulation compared to -34.2% the phenology-less simulation. The annual cycle plots don't show much so I'm not including them here. Standard diagnostics are here:

https://webext.cgd.ucar.edu/I1850/ctsm51_ctsm51d166deadveg_1deg_CRUJRA_FLDS_ABsnoCDE_blk_A5BCD_1850pAD/lnd/ctsm51_ctsm51d166deadveg_1deg_CRUJRA_FLDS_ABsnoCDE_blk_A5BCD_1850pAD_1311_1320-ctsm51_ctsm51d160_1deg_CRUJRA_FLDS_ACDE_blk_A5BCD_1850AD_251_260/setsIndex.html

A historical corresponding to ABsnoCDE_blk_A5BCD is underway (Issue #54 ).
And a 2000 F-case is underway as well (Issue #53 ).

[wwieder]

@olyson is there any chance you can generate polar view max LAI plots for the control and latest F-case, #53. for the LMWG meeting next week?

[olyson]

Yes, working on that now.

[olyson]

The F-case (#53 ) is complete. Standard diagnostics compared to the control (#14 ) are here:

https://webext.cgd.ucar.edu/F2000climo/f2000.ne30_t232.BGC_CROP.deadveg/lnd/f2000.ne30_t232.BGC_CROP.deadveg_77_96-f2000.ne30_t232.BGC_CROP_77_96/setsIndex.html

In general, we can see increased TLAI at higher latitudes with warmer temperatures in spring and summer and lower snow pack and earlier melt of snow in spring.
Polar plots of JJA TLAI top to bottom are the new simulation, the control, and new-control. I will produce some annual cycle plots soon.

[wwieder]

These are great. Thanks for working on your day off, Keith. Does your script also print change the dead area between the two cases? Don't worry about this if it's not immediately obvious.

[olyson]

Yes, the percent decrease in dead vegetation is -54%.

[olyson]

Annual cycle plots for the available F-cases, where F_CON is the control (#14 ), F_SStk is Sturm snow tk (#13 ), F_ABCD is Sturm snow tk, phenology triggers, FUNfracfixers, and frootleaf (#25), and F_ABsnoCDE_blk_ABCD is the current dead veg branch, which is Sturm snow tk, phenology snow depth threshold only, FUNfracfixers, frootleaf , new SNICAR, snicar_snobc_intmix = .true., xdrdt = 5, scvng_fct_mlt_sf = 0.5, snw_rds_refrz = 1500, and fresh_snw_rds_max = 400 (#53).

[olyson]

ILAMB results for historical simulations (some tuning definitely needed). CLM50, CTSM51 (ctsm51_cesm23a02cPPEn08ctsm51d030_1deg_GSWP3V1_hist), and the CTSM deadveg branch.

https://webext.cgd.ucar.edu/I20TR/ctsm51_ctsm51d166deadveg_1deg_CRUJRA_FLDS_ABsnoCDE_blk_A5BCD_hist/lnd/_build_newscore_CLM50CTSM51DeadVegBGC/

[olyson]

Note that CLM50 and CTSM51 are forced by GSWP3 and CTSM deadveg branch is forced by CRUJRA.

[olyson]

Update to ILAMB results that include regions (not just global):

https://webext.cgd.ucar.edu/I20TR/ctsm51_ctsm51d166deadveg_1deg_CRUJRA_FLDS_ABsnoCDE_blk_A5BCD_hist/lnd/_build_newscore_regions_CLM50CTSM51DeadVegBGC/