Tuning of Leung_2024 dust emissions for clm6_0_cam7.0 and on namelist

[ekluzek]

@dmleung has tested some F cases with ctsm5.2.019 and found some issues that will need some tuning to give good results. He previously did some tuning for clm5_0_cam6.0 for the ctsm5.2 surface datasets, and will need to do more for cam6_0_cam7.0. This is part of the normal tuning exercise that needs to be done for all of CESM3.0.

@dmleung is working on this now and will get us some changes to bring in as soon as possible. Hopefully, this won't need to be done multiple times as the model should be stabilizing as we move forward. However, as other tuning changes happen in both CAM and CTSM it may need to be done more than once. Mostly this is work that @dmleung will do and ask us to bring in when ready. Code changes for this are likely to be small.

From @dmleung: Also a minor point is that dust_moist_fact could be in both the moisture threshold functions in Zender2003 and Kok2014 in SoilStateInitTimeConst.F90. So, it'd be great if you could add the same parameter into the Kok14 function as well if that's easy.

Definition of done:

• Tag a branch tag with a simple change with some additional parameters hardcoded in (Tuning Leung_2023 dust emissions for clm6_0_cam7.0 #2803)
• Put the hardcoded tuning changes in a tag on master
• On b4b-dev put those parameters on the parameter file -- show identical answers with correct values
• Handle dust_moist_factor for the Kok2014 function as well and add it to the unit-test
• Bring those b4b-dev change to master

[dmleung]

I tried to use the ctsm5.2.019/cam6_3_137 tag to run an FLTHIST case for year 1991 and saw some very high annual mean dust AOD values (>1) over Argentina (Patagonia) and Australia. I tried to look into this issue and understand what were changed to cause these blowups, since we did not see them in older tests. The older test used the ctsm5.2.006/cam6_3_160 tag to run a FHIST case for year 1991.

I am showing a side-by-side comparison of the two cases/tests. See the two CAM history variables below, upper is dust AOD and lower is dust emission (plotted in log scale).

---

Since CTSM dust emission is dependent on meteorological and land-surface fields, I looked into the differences between these fields in the two model tags. My primary conclusion is that the main cause is the enhanced friction velocity (or ustar) over the two countries that made dust emissions explode. See the two CTSM variables below, upper is FV from CTSM and lower is the modified friction velocity used in dust emission module. FV has some visible increases over places like Argentina, Australia, and the US. Increases are also seen over other places like boreal forests, but it is less relevant to dust simulations. Meanwhile, FV has some decreases over northwestern China, causing further low bias of dust over China.

---

There could be various reasons that caused FV changes from one model version to another. But the major concern here is more about tuning. An increase in annual mean FV by 0.1 m/s could easily cause a > 10 time increase in dust emission. This level of FV change might be not so important for other land processes, but dust emission is very sensitive to FV. Since future model physics will keep evolving and met fields will keep changing, we might not want such strong sensitivity of dust emission susceptible to all these changes in physics and met fields. We might prefer a more robust dust emission scheme that gives satisfactory results given different physics, grid, input datasets and model tags.

[dmleung]

So, the current idea is to reduce dust emission sensitivity to met field changes. In the Zender_2003 emission scheme, dust emission ∝ WND_FRC_SOIL^3 approximately. In the Leung_2023 scheme (built upon Kok_2014), dust emission ∝ WND_FRC_SOIL^(2+frag_expt), where frag_expt is the fragmentation exponent that increases with soil moisture. frag_expt is < 1 over major deserts but could go to > 5 over marginal/semiarid sources like Argentina.

Leung et al. (2023) has a discussion on globally capping frag_expt at 3-5 to limit the dust emission sensitivity to ustar. Since the frag_expt dependence on soil moisture was a fitting using local measurement data by Kok_2014, it could be relaxed and kept at higher values for high-resolution models that resolves local meteorology. But, it might be better capped at low values for 100km-resolution models.

---

The first proposed change is to cap frag_expt at smaller values in the dust emission module. Leung_2023 in ctsm5.2.006 caps frag_expt at 5, which works well in that tag. But it can be a smaller number, such as 1. This forces dust emission ∝ WND_FRC_SOIL^3 at most over marginal/semiarid sources, while changing little over major deserts. See the dust emission results below by capping frag_expt at frag_expt_thr = 5 (default) versus capping frag_expt at 1, both using the ctsm5.2.019 tag and FLTHIST compset. The results look acceptable to me.

---

For now I propose that we use frag_expt_thr = 1 for native resolution grids. We might want to make it a namelist variable and become resolution dependent (so users could elect to relax it to 2 or 3 for regional refinement), but this could be more optional. frag_expt_thr is more of a regional tuning parameter since lowering it tends to suppress dust from secondary/marginal sources more than from major deserts (vice versa, increasing it promotes more dust from secondary sources). Using frag_expt_thr = 1 could drastically suppress dust emission sensitivity to winds/ustar over marginal/semiarid dust sources and make the scheme more robust against changing model tags, so this seems a good solution for me. I am still exploring other possible solutions, since I think we could do better in the R.H.S. figure (e.g., suppress Australian dust further).

[dmleung]

I'm tagging @tilmes to follow the updates on this issue.

[dmleung]

Testing a new, corrected surface dataset:

Background:
Issue #2744 showed that the surface dataset has missing values for multiple variables in different areas, mainly over Africa. Software engineers later came up with a corrected surface dataset. Here we do a test to see the difference in dust using the two surface datasets.

Default fsurdat = '/glade/campaign/cesm/cesmdata/inputdata/lnd/clm2/surfdata_esmf/ctsm5.2.0/surfdata_ne30np4.pg3_hist_1850_78pfts_c240216.nc'
New fsurdat = '/glade/campaign/cesm/cesmdata/inputdata/lnd/clm2/surfdata_esmf/ctsm5.3.0/surfdata_ne30np4.pg3_hist_1850_78pfts_c240908.nc'

Remarks:

1. I am using ctsm5.2.019/cam6_3_137 with clm6_0/cam7_0 physics, CSLAM grid SE dycore, and FLTHIST compset.
2. frag_expt_thr is still 5 here; we didn't bring the abovementioned proposed changes here yet..
3. Using the new surface dataset, I encountered an error asking me to use check_dynpft_consistency = .false. in user_nl_clm, so i did it. For now I am not sure why this happened or how this would impact the simulation results below. I am not sure how to avoid this, but the error seemed to suggest there was something to do with the flanduse_timeseries. (Both fsurdat and flanduse_timeseries start in 1850 so I wasn't sure what's wrong)

Results:

1. The main result is that dust further reduced in Africa. This seems to mainly be because the friction velocity (FV) was reduced. I am not sure why FV would decrease when using a surface dataset without missing values. @wwieder may have some ideas?
2. I also saw a better soil moisture map (at the topmost soil layer) over Africa: areas with low PCT_CLAY has lower H2OSOI than the neighboring regions. It definitely looks better than the one when we use the default surface dataset, which gives too high soil moisture over Africa.

Thoughts:

1. Using a corrected surface dataset did not help us fix the dust bias. The dust overestimations over S hemisphere and the underestimations over Africa still exist, and somehow the underestimations over Africa is worsened given small FV values. I am not sure why FV decreases over Africa, but hopefully switching to frag_expt_thr = 1 will alleviate a lot of the dust overestimations over the S hemisphere and overall reduce the dust sensitivity to FV changes in different CESM updates (as discussed in previous comments).
2. The improvements in H2OSOI actually gives another way to tune dust. We might be able to reduce the soil moisture threshold (gwc_thr_col in SoilStateInitTimeConst.F90) using a global tuning constant. This increases the soil moisture impacts on reducing dust over semiarid regions (like the S hemisphere) while not limiting hyperarid (e.g., African) dust emissions much. This is something I will test next.

[dmleung]

Testing dust emissions using different CTSM roughness schemes:

Background:
CTSM's roughness scheme has been switched from z0param_method = 'ZengWang2007' to 'Meier2022'. Since we saw friction velocity (FV) changes caused big changes in dust emissions (see above comments), we tried to inspect changes in different elements in the CTSM that might contribute to the changes in FV. We also wanted to see if using the older scheme could help alleviate the dust biases in the late CTSM tag.

Description:

1. We compared a run using ctsm5.2.006/cam6_3_160 (z0param_method = 'ZengWang2007' in CTSM) and ctsm5.2.019/cam6_3_137 using both z0param_method = 'Meier2022' (default) and z0param_method = 'ZengWang2007' (we switched back to the older scheme) and see how FV and dust changed.
2. This test used the surface dataset with missing values (surfdata_ne30np4.pg3_hist_1850_78pfts_c240216.nc) since this test was done a few weeks ago (before the issue of missing values was discovered).

Results:
Using z0param_method = 'ZengWang2007' in the ctsm5.2.019/cam6_3_137 tag helped clear a lot of high dust AOD values over semiarid regions (right). It makes the dust distribution look a lot closer to the dust in the older CTSM tag (left). Using Meier2022 in the late CTSM tag (middle) did contribute to the issue of the dust biases across the world.