Ocean BGC choices

[aekiss]

We've been discussing what ocean BGC model to use for ACCESS-OM3. Here's a summary for future reference, which combines contributions from @hakaseh, Tyler Rohr, Matt Chamberlain and Jess Lou.

Background

The Linkage proposal specifies that we will use WOMBAT in ACCESS-OM3. This is not currently available in MOM6 and would require porting. However, MOM6 already has several BGC modules available (miniBLING, BLING, TOPAZ, COBALT, etc) https://github.com/NOAA-GFDL/ocean_BGC/tree/master/generic_tracers

The question

Should we proceed with the plan to port WOMBAT to MOM6, or would one of the existing MOM6 BGC models offer equivalent functionality and performance for less effort?

Discussion summary

Of the BGC models already available in MOM6, only BLING, COBALT and ERGOM seem to be actively maintained (judging from their modification times).

• BLING: Dunne et al. (2020; details in supplement): simpler than WOMBAT: 6 tracers, no explicit phyto- or zooplankton. Designed for large scale ocean mediated carbon cycling in hi-res fully coupled runs. Used in GFDL’s CM4.0.
• WOMBAT: Oke et al (2013, appendix B): an NPZD model with ~10 tracers. Used in OFAM3, ACCESS-ESM 1.0 and ACCESS-ESM 1.5. Used in ACCESS-OM2 and available via the master+bgc branch in the supported configurations; initial conditions, forcing files and parameter files have been created via https://github.com/COSIMA/input_om2-bgc; parameters match ACCESS-ESM1.5 (Ziehn et al., 2020), except for nat_co2.
• COBALT: Stock et al., (2020): much more complex than WOMBAT: 33 tracers. Used in GFDL’s ESM4.1.
• ERGOM: https://ergom.net/: similar complexity to COBALT; mainly used in Baltic Sea so probably needs testing and tuning for global applications - so don't consider further.

Slides from discussion at COSIMA meeting: COSIMA-meeting-2022-07-14-WOMBAT.pdf

• MOM6 supports long tracer timesteps, making it much cheaper than MOM5 to carry many BGC state variables. So is it now feasible to use a complex BCG (like COBALT) globally at high resolution?
• Tuning parameters and setting up input files can be a big job, and we already have that in place for WOMBAT.

[aekiss]

Here are some nice summary figures from Kearney et al., 2021:

[aekiss]

from Kearney et al., 2021:

Boxes indicate state variables, and solid arrows represent fluxes between state variables; dashed lines indicate that the rate of change of one state variable is calculated proportionately to another. Colors indicate the base element for each state variable, with blue, green, pink, orange, and gray representing nitrogen, phosphorous, silicon, iron, and carbon, respectively. State variable boxes are positioned vertically based on functional role, indicating whether the functional group includes producers (and within that, subcategories of nitrogen fixers, Si-users, and CaCO3-users), consumers, or detritus. Horizontal position indicates the approximate mean size of the cells/bodies/particles represented by each state variable.

[aekiss]

I feel like BLING is too simple, and I suspect COBALT will be too slow to run globally at high resolution and will require a lot of setup (initial conditions, forcing files, etc), so the easiest path forward could be WOMBAT, which we already have working in ACCESS-OM2. I have the impression that it wouldn't be too hard to put WOMBAT into MOM6, but perhaps it's easier said than done?

[aekiss]

@hakaseh took the opposite view, that porting WOMBAT would be harder than using one of the existing MOM6 BGC models, and somebody would need to be tasked with it. If COBALT is used, it'd be nice to also couple this to sea-ice BGC in CICE6.

[aekiss]

Tyler Rohr commented:

• COBALT is much more useful for work where the focus is the BGC. Particularly if the goal is to learn about the mechanisms through which BGC cycling will be modified by Climate Change. I’d further note that the level of complexity here (~similar to what’s used in CMIP6 runs of CESM, UKESM, IPSL CNMR, CMCC-ESM2) is still far more on the BGC side of things than the higher-ecosystem (and especially fisheries) side of things. For instance including 3 phytoplankton groups (instead of 0 or 1) is more about getting nitrogen cycling (by including n2-fixers) and silicate cycling (with a large diatom-like group) than it is about accurately building out the base of the food chain for ecosystem purposes. Similarly, their assumed size and composition differences becomes important for particle aggregation and sinking, and thus getting export and the biological pump right.
• Both can probably be tuned to get the big picture relatively right, but including more mechanisms and functional diversity in the plankton means the response to environmental and physical change in projections is likely going to be more robust. Obviously, this comes with computational trade-offs.

I think some of the questions moving forward are:

• How important is having ownership over WOMBAT? Both from a practical perspective (more familiar, easier to tweak) as well as an institutional perspective (copyrights, impact, momentum, etc)
• Who else is going to be interested in the BGC output (besides me…)
• What are the types of experiments we envision running with light vs. heavy BGC?
• How light is too light? (i.e. BLINGv2 vs. WOMBAT).
• With the numerical improvements to MOM6 how heavy is too heavy (WOMBAT v. COBALT)
• What advances in Bio-physical coupling does MOM6 v. MOM5 present

Probably, a combination of more than one would be ideal for different applications.

[aekiss]

Summary of comments from Jess Lou:

• BLING is much more barebones than what you might want if you were interested in any sort of plankton ecosystem dynamics.
• See paragraph preceding the section "Phytoplankton Growth Limitation Terms" in Kearney et al., 2021 for a summary of the advantages of more detailed models (e.g. COBALT) for ecological studies.
• WOMBAT has the advantage of adding independence to the models used in MIPs. Perhaps elements of TOPAZ/COBALT could be brought into WOMBAT to obtain new features while still maintaining some independence?

[aekiss]

Summary of comments from Matt Chamberlain:

• These GFDL packages (BLING/TOPAZ/COBALT) will work out-of-the-box for straight MOM-SIS configurations, though not in ocean-only mode, so work would be required to enable them in ACCESS-OM.
• Test case setups were available to download and try for MOM5; I guess they exist as well for MOM6(?)
• If these packages are enabled, it would also make the GFDL's CFC tracer available.
• Porting WOMBAT into MOM6 is probably relatively straight forward by comparison, and then would be useful for comparing and evaluating the new packages.

From the point of view of the ACCESS-ESM, I expect that MOM5 and WOMBAT will be used for some years yet, due to the need for long stable runs with realistic, balanced BGC states.
This took years to prepare leading up to ESM1.5 for CMIP6.
Looking forwards, if/when the coupled model and ESM start adopting MOM6 that may be a time that other tracer packages could be considered.

Those are useful comments from Jess, thanks for forwarding Tyler.
I see importing elements of TOPAZ/COBALT into WOMBAT as a potential way forward, at least for WOMBAT in the ESM perhaps, and maintaining the independence/distinctiveness of ACCESS.

[aekiss]

I think we have a consensus that WOMBAT is the way to go, but I suggest we build ACCESS-OM3 in a way that minimises obstacles to switching to a different BGC component in future, e.g. COBALT for marine ecological studies, or BLING for very lightweight BGC.

[micaeljtoliveira]

@aekiss Thanks for the summary of the discussions. As was agreed during the technical workshop we had a couple of months ago, I would be willing to take on the task of adding WONBAT to MOM6.

[aekiss]

Great, thanks!