CABLE4 planning: meeting notes

Land Surface CABLE
101

March 5th

For discussion

  • Continue discussion on plans and goals for 2026

  • Latest ESM1.6 results for Australia

Notes

102

I’m an apology for today. I have another meeting I need to attend at 11.30am.

103

Hey everyone, here is a quick look at the latest historical simulations and their impact on the Australian carbon cycle. It seems we have increased our cumulative carbon sink, which sits about +1sigma above the ESM1.5 40-member ensemble average. Mean historical GPP has increased by 20% relative to ESM1.5. Interannual variability has also increased.

Presentation_historical_runs_notes.pdf (646.5 KB)

104

Thanks @alexnorton - nice to see the comparison with the ESM1.5 ensemble.
I am curious about the hot spots in the ESM1.5 NEE average that go away in ESM1.6. I wonder if they are from the crop tile but not sure why they’d be strongly positive in the ESM1.5 case. Trying to remember if we changed some of the c3 crop parameters as well as adding the c4 crops.

105

Good question. I have a feeling it may be due to the land-use change. I haven’t looked specifically. Unfortunately I don’t have per tile output from ESM1.5

106

March 19th

Attendance: Alex, Rachel, Tilo, Ian, Ben, Juergen, Lachlan, Jhan, Claire

For discussion

Notes

  • ESM1.6 Results: Concentration-driven historical runs with CMIP7 forcings are being run, CNP configuration, three simulations going from there points off the picontrol (each 20 years apart). No initial wood product pool.

  • ESM1.6 Results: Emissions-driven historical runs show fairly good results when comparing to observed atmospheric CO2 concentrations. With a larger initial wood product pool (50 Pg C in 100yr turnover pool) we better match the atmospheric CO2 overall BUT we end up with overshooting atmospheric CO2 by 2014 (~8 ppm). With a 25PgC initial wood product pool we overshoot by ~5 ppm by 2014. However, the rate of increase in modeled atmospheric CO2 in the most recent decade (2010-onwards) is higher in ESM1.6 for all test runs (including the test with 0 initial wood product) compared to the observations, suggesting modifying the initial wood product pool won’t fix that.

    • Action item: Revisit Tammas’ optimization script (see * * here) to see how we might adjust pool sizes, turnover times or wood harvest allocation fractions to correct these biases.

  • Wood harvest flux and harvest area from LUH3 as implemented in ESM1.6:

    • If we ignore thinning, we are missing some regionally important wood harvest fluxes, even though the impact on global total land flux is too large. We end up getting huge wood harvest fluxes (which ultimately respire to the atmosphere) as many areas in LUH3 indicate wood harvest area fractions > 0.1 and even up to 0.5, year after year, which is physically unrealistic. In particular, Europe has quite a large amount of wood harvested due to forest management i.e. thinning (much less due to land-cover change) – if we ignore thinning we get very little wood harvest occurring there at all which is possible not realistic.
    • Ben Smith will contact European modeling colleagues to set up discussions/meeting with us on how they interpret wood harvest area from LUH3.
107

March 26th

Attendance: Alex, Rachel, Tilo, Ian, Juergen, Lachlan, Jhan

For discussion

Notes

  • ESM1.6 results - historical runs: Rachel has compared the modeled CO2 concentration from the emissions-driven historical run against observed historical CO2. With 40 Pg C in the wood product pool (pool with 100 yr turnover time), the model produces a pretty good inferred CO2 concentration compared to the historical CO2 concentration. The absolute bias is <4 ppm across the whole run. Some variation is expected across an ensemble of runs (preliminary tests suggest ~2ppm variation at the start of the run).
  • Potential high-impact science paper: Comparing an ensemble of concentration-driven vs an ensemble of emissions-driven simulations. It is unknown how an interactive carbon cycle may impact the ensemble of climate simulations - ACCESS-ESM1.6 may be the first modeling center to produce these simulations and be able to address this question.

Action Items

  • ESM1.6: We should do a carbon conservation check with latest historical runs. Refer to Gang Tang’s * paper and the recommendations at the end.
  • Alex to reach out to Claire Carouge about NRI LWG allocation for next 2 quarters. Is there compute and storage available to support the CABLE-POP simulations for D-BEN?
108

Yes, there is. We have lots of extra. It won’t be a problem even with the rAM3 runs that Mat and Siyuan are doing.

109

That’s great thanks Claire. Is there anything else we need to do now or shall we just get the simulations ready to go?

110

Fill in an experiment topic

So there is a trace of what the resources are being used for.

111

April 2nd, 2026

Attendance: Alex, Rachel, Tilo, Ian, Lachlan, Jhan, Juergen

For discussion

ESM1.6 topics for discussion

  • CASA output - monthly averaging subtleties
  • Pools output as end of month values - approximate monthly mean with mid-month value. Testing? Post-processing effort?
  • emissions-driven historical - impact of CO2 drift in esm-picontrol
  • which runs to prioritise over long weekend

Notes

  • Pools output as end of month values - approximate monthly mean with mid-month value. Testing? Post-processing effort? Post-processing may be most straightforward. In order to do this we must ensure we have the first restart file to get the first beginning-of-month value in the run.
  • ~600 years of esm-picontrol shows a drift in global mean atmospheric CO2 (just looked at the surface layer) of +0.58ppm per 100years (linear fit is y=0.0058x + 283.82, R2=0.53). Tilo notes that, in the grand scheme of things, this is pretty small drift and other ESMs will likely show a drift in the esm-picontrol. Question is whether we should correct for this drift when spinning off esm-historical runs (e.g. ensemble runs taken at different points in the esm-picontrol which would mean different atmospheric CO2 starting point in each ensemble).

Action items

  • Alex to add scripts to CMIP7 Inputs repository that outline our translation of the historical LUH3 data into land-cover types and change.

  • Rachel to continue to run esm-historical runs that branch off different points in the esm-picontrol to evaluate impact of atm CO2 drift on a longer-term run ( > 20 years that Rachel has already tested).

    • Alex to create another restart with 40PgC wood products from later in the esm-picontrol (closer to the 600 year mark).
112

I looked at the picontrol with the latest N dep. Plot is GPP (new minus old) on the left for a 50 year mean, and the right is the difference between two 50 year periods from the current picontrol. The differences from the change in N dep don’t look any larger than within-control differences. I also looked at the mean seasonal cycle of NEE but only tiny changes.

image
image1073×512 104 KB

I assume if we’re going to see the impact of a change in N deposition anywhere it would be in carbon - or are there other fields that I should check?

113

Hi Rachel, that’s interesting to see. Might also be worth checking some pools like soil mineral N and plant labile C. Can you point me to the runs and I’ll look at those?

114

A copy of Martin’s control run output has been put on p66:
/scratch/p66/mrd599/access-esm/archive/PI-conc-nitrogen2-test-preindustrial+concentrations-FZJ-CMIP-nitrogen-2-0-7c8debc1
My historical run output (starting from 1971) is here:
/scratch/p66/rml599/access-esm/archive/hist-ndep2-hist-ndep2-e1fe82be

115

April 8th, 2026

Attendance:

For discussion

ESM1.6 topics for discussion:

  • Land carbon cycle evaluation: Land-use change emissions too low compared to GCB2024. One obvious explanation is that we are not running with thinning so we are missing any forest management e.g. harvested wood products from forests.
    • Alternative interpretation of LUH3 harvested biomass: Reduced area-based thinning fractions data has been created. This potentially resolves the issue of ESM1.6 underestimating land-use change emissions.
  • Latest emissions-driven historical tests (comparison of different initial conditions, initial wood product pool)
  • Anything further we wanted to discuss about nitrogen deposition?

Notes

Action items

116

April 16th, 2026

Attendance: Alex, Tilo, Ian, Juergen, Lachlan, Claire

For discussion

ESM1.6 topics for discussion:

  • Land-use change with new thinning input (derived from LUH3 harvest biomass instead of harvest area)

Updates on Land Cover for AM3 (Lachlan)

Notes

ESM1.6 Land-Use Thinning:

  • Output data for emissions-driven historical run with new forest thinning forcing and 20PgC initial wood product is here: /scratch/p66/ajn563/access-esm/archive/esmhist-thinbioh-20-dev-historical+emissions-63d1e9c1/
  • Results look promising up to 1975. Performance for rest of the run will be crucial. So far, we see improved representation of harvested wood products (compared to OSCAR model output, Gasser et al. 2020) and land-use change emissions over the historical period, with no negative impact on the net land sink NEE or NBE (compared to GCB2024) or inferred atmospheric CO2.

Land Cover for AM3 from CCI (Lachlan):

  • Working out how to separate grass in C3 and C4 types, also how to get Tundra. CCI doesn’t differentiate these types.
  • Taking inspiration from CLM4 / NCAR. They have information on C3/C4 grass fractions which can be used as relative weightings and applied to the CCI grass to get global maps of C3/C4 grasses.

Action items

Alex to check with Paul Leopardi on the LUH data for the CMIP7 Scenarios.

117

April 23th, 2026

Attendance: Alex, Ian, Jhan, Claire

For discussion

ESM1.6 topics for discussion:

  • Impact of new forest thinning on historical run
118

April 30th, 2026

Attendance:

For discussion

ESM1.6 topics for discussion:

  • Impacts of forest thinning sensitivity tests on historical runs (sensitivity tests run from 1981 onwards, off restart130 from esmhist-thinbioh-20) with four cases:
    • No change in thinning (but there is a fix in the executable which makes this different to the esmhist-thinbioh-20 run)
    • 30% reduction in thinning scaling factor
    • 50% reduction in thinning scaling factor
    • 70% reduction in thinning scaling factor
  • Processing ESM1.6 outputs with MOPPy and feeding this into ILAMB (Romain and Rhaegar)