BGC Validation: ACCESS_OM2 1/10 degree IAF

See here for the BGC summary slides:

Those that presented will be meeting soon for some follow-up synthesis and will share results here/set up some larger follow up meetings. Let me know if you’d like to be involved in the initial discussion, probably next week (Feb 27-Mar 2)

And please post any questions/rabbit holes here that you might want us to try and chase down :slight_smile:

1 Like

You can try uploading the power point slides directly if that helps (not entirely sure if it does or not)

Great that the discussion is continuing! Can you post the day/time/zoom link here?

1 Like

Here is the current plan to diagnostically diagnose explain the lower NPP and iron drift in the 1/10 run. Stay tuned.

2 Likes

Great, thanks @tylerrohr22 - could you also post this as text so it will be searchable and quotable?

My back of the envelope hypothesis is that the 1/10 run is permitted deeper mixing

MLD dependence on resolution in ACCESS-OM2 has been investigated by Treguier et al 2023 (in cycle 1 - not sure how similar that is to cycle 4).

Winter MLD relative differences (0.1°-1° resolution) are small at most latitudes in the zonal average, but MLD is shallower (less biased) at around 50S in SAMW formation regions and deeper (more biased) in Weddell. In summer 0.1° MLD is deeper at around 50S and shallower near the equator (in both regions the resolution increase reduces but doesn’t eliminate the MLD summer bias relative to obs). See fig below and maps here.

Of course, what’s important is the MLD difference in the regions and seasons of high productivity, and the MLD vs. photic zone depth, so a MLD study alone is insufficient.

Could look at .25 as well if were feeling ambitious.

AFAIK we don’t have any 0.25° runs with BGC

I’m guessing you’re proposing offline diagnostic calculations here, but many of these calculations are nonlinear, so the results will differ from what you’d get from an online calculation - i.e. feeding temporal (say monthly) means into a nonlinear formula will give a different result from the formula calculated on each model timestep and then (say monthly) time-averaged.

Is that cause for concern?

@PSpence - did you do a 0.25° ACCESS-OM2 run with BGC?

I don’t know if there is a new run, but during our BGC discussion I was referring to the “old” control run performed with the 0.25 deg and BGC, available at
/g/data/v45/pas561/mom/archive/wombat_jun16z

Sup Fig. 9 and 11 show DIC, ALK, O2 and CFC distributions compared to obs in that simulation:

Here’s plain text version (just lost its formattting… will write/format directly in the forum next time).

Diagnostic Triage Plan:
Confirm NPP is lower in the 1/10 deg vs 1 deg run.
Confirm Iron is higher in the 1/10 deg run
Confirm Nitrate is higher in the SO (and lower elsewhere?) in the 1/10 deg
Check if phytoplankton specific division rates are lower or higher
If higher, that mean cells are dividing faster but there is less biomass. So we are probably looking at a top-down control, which I think is unlikely with the simple Z scheme and same parameters
If lower, then something is limiting growth more.
The phytoplankton specific division rate is determine by a light and nutrient replete (but T-limited) ‘maximum’ growth rate (Tbio) which is scaled by the most limiting of light and nutrient stress. Each are represented by a limitation term that varies from 0-1. Note, a lower limitation term means MORE limitation. So, the next step is to compare the change in the light and nutrient limitation terms
If it is the light limitation term that went down then what we are seeing is probably that the high-res physics led to deeper MLDs, more light limitation, less NPP, less iron utilization, and higher iron concentrations.
If the nutrient limitation term went down then the next step is to figure out if it was due to iron of nitrate
Nutrient limitation is simply the minimum of a nitrogen and iron limitation term, which both vary from 0-1.
If the iron limitation term went down then there might be something going on with scavenging, but this seems unlikely seeing as the problem is the iron concentration went up.
If the Nitrogen limitation term went down then something (physical) could be stifling the supply of nitrate, limiting growth prior to iron limitation, explaining the low NPP and high iron. But this would imply surface Nitrogen would be down, which it sounds like is not the case, per Ram.

My back of the envelope hypothesis is that the 1/10 run is permitted deeper mixing which has caused the light limitation term to drop below the nutrient limitation term in deeper mixing regions. Since light and nutrient ARE NOT co-limiting, this will stifle nutrient uptake, lower NPP, and elevate iron and nitrate in deeper mixing regions, but not necessarily the lower latitudes/equator.

Methods:
Create comparable products from OM2 1/10 IAF and either OM2 1 deg or ESM 1.5 1 degree
Years: whatever you think is best to create the most comparable Climatologies, average out enough interannual variability, but not have to download too much data
Depth resolved variables to pull: Nitrogen, Iron, Phytoplankton Biomass, Temperature, PAR, NPP
Could look at .25 as well if were feeling ambitious. But note, parameters below are from ESM1.5 and OM OM2 1/10 IAF (the Param_list in input_om2-bgc/interpolate_to_access-om2.ipynb at 5ec65b4b6edefde82d2bdb0f8774a12ed9bcde86 · COSIMA/input_om2-bgc · GitHub)

Diagnostics
NOTE: All of these should be 2D fields at what ever time step your create the Climatologies (daily?)
NOTE: In all cases, confirm parameter units (noted) are consistent with outpuf fields and convert aas required.
Surface Nutrients (N and Fe)
Column integrated NPP
Biomass-weighted, Depth Averaged, Phytoplankton Temperature Limited (but infinite light and nutrients) Maximum Growth Rate
i. Create depth resolved terms using the temperature field
Tbio = abiobbio^(cbioTemp) % (1/d)
Abio = .27 % (1/d)
Bbio = 1.0666 (unitless)
Cbio =1.0 (C)
ii. Create Biomass-weighted, Depth Averaged Term — (Here, you’re just taking a column-wise average of Tbio weighted by the vertical profile of Phytoplankton, so that the result is the Temperature limited Maximumum growth rate of the mean phytoplankton in the water column.)

Tbio_mean = sum(Tbio.*Phyto)./sum(Phyto) % Operating in vertical direction

Biomass-weighted, Depth Averaged, Phytoplankton Nitrogen limitation
i. Create Depth Resolved terms using Nitrogen field
Lim_N= Nitrogen/(k1bio+Nitrogen) % (unitless)
k1bio = .7 % (mmolN/m3). → make sure this in the same units as the output Nitrogen Field
ii. Create Biomass-weighted, Depth Averaged Term
Lim_Nitrogen_mean = sum(Lim_N.*Phyto)./sum(Phyto) % Operating in vertical direction
% Same idea as above
Biomass-weighted, Depth Averaged, Phytoplankton Iron limitation
i. Create Depth Resolved terms using Nitrogen field
Lim_Fe= Irone/(k2bio+Iron) % (unitless)
K2bio = .1 % (umolFe/m3). → make sure this in the same units as the output Iron Field
% Note: I don’t this isn’t stored in the parameter file; just directly input at .1 in the source code.
ii. Create Biomass-weighted, Depth Averaged Term
Lim_Fe_mean = sum(Lim_Fe.*Phyto)./sum(Phyto) % Operating in vertical direction
% Same idea as above
Biomass-weighted, Depth Averaged, Phytoplankton Nutrient limitation
i. Lim_Nut = min(Lim_Fe, Lim_N)
ii. Lim_Nut_mean = sum(Lim_Nut.*Phyto)./sum(Phyto)

Biomass-weighted, Depth Averaged, Phytoplankton Light limitation
i. Lim_Light = 1.0-exp(a1bio*PAR/Tbio)
a1bio = .2560 % (W/m2)-1 (d)-1 → make sure W/m2 is the units for the output PAR Field

Lim_Light = sum(Lim_Light.*Phyto)./sum(Phyto) % Operating in vertical direction
% Same idea as above

Biomass-weighted, Depth Averaged, Phytoplankton specific divisions rates
i. u = min(Lim_Light, Lim_Nut)*Tbio
ii. u _mean = sum(u.*Phyto)./sum(Phyto)

Figures Ideas
Spatial distributions for each diagnostic averaged over winter and summer, for each run and the difference (after re-gridding the 1/10 deg onto the 1 deg grid)
Time series of each diagnostic, averaged across the Southern Ocean (and maybe other regions of interest), for each run and the difference

True. Anything that saturates will be overestimated offline if the mean is above the half saturation point. With grazing rates (similar nonlinearities) and monthly averages I’ve found the difference is order 10-20%. Where possible, it would certainly be better to compute these offline with daily averages. If it is the case that there is daily output saved for one of the runs but not the other, then I think we would want to compute both sets of offline diagnostic on monthly means to reduce the comparative bias. Even with the monthly offline diagnostic, I think there might be something to learn.

what about nonlinear functions of insolation - do they need to resolve the diurnal cycle? The forcing (incl. downwelling shortwave) is 3-hourly.

Interesting. Agree that the winter difference (.1-1) is probably insignificant to BGC, and getting BGC in Weddell is pretty tough regardless imo. But I think the summer SAZ, which amounts to a 10-20% difference I suspect (assuming summer MLD of 50-100m), could be quite consequential and still potentially consistent with the possibility of light limitation playing a relatively more important role than iron lim in the high res SO.

Also, as you say it is pretty hard sort out MLD diagnostics from the actual turbulance closure scheme and photic zone etc. Thats why I like looking at the biomass-weigthed light limitation term for phytoplankton. It tells you explicitly how much light the mean phytoplankton in the water column is getting and accounts for differences in mixing.

If we’re focusing on the SO summer (ie long days), and comparing diagnostics with the same diurnal cycle, I suspect that the bias wont be disqualifying. That said, I would certainly advocate for using the highest resolution available to compute any nonlinear offline diagnostic.

And also note, I am always a fan of saving these limitation terms explicitly for exactly these reasons. They always prove tremendously helpful in doing any BGC detective work or trying to say anything about mechanisms.

Agreed, but what’s most important for nonlinearity is capturing the timescale over which the insolation goes through its greatest range. I suspect the variation of daily-mean insolation over a month would usually be less than the variation of the 3-hourly insolation over a day, except well south of the Antarctic circle (66S) in summer or winter.

Yes I did with RYF forcing.

P

Some surface fields are available as 6-hourly means for 1 Jan 2014 to 31 Dec 2016 in 0.1° IAF cycle 4 (see here):

  • radbio1 (Photosynthetically active radiation for phytoplankton growth at surface)
  • surface_fe, surface_no3, surface_o2, surface_phy
  • mld, surface_pot_temp, surface_salt

These could be used to estimate the error in nonlinear calculations that use daily- (or longer) averaged data.

Hi all,

I’m late to the party, but I have a student working on nutrient and chl-a data from CTD casts of the coast of Brisbane. We were thinking of using output from this run to provide broader context to our study. It is possible that we could kill two birds with one stone here and use our data to validate (say) the seasonal cycle of NOx, phosphate and chl-a.

This is (obviously) very regionally and process specific, but could complement this effort.