Emission-driven spinup

[adagj]

Questions:

• dependence on the land carbon cycle, which is not ready yet

• initiated from quasi-equilibrium concentration driven coupled simulation

• important to have near balance TOA radiation to avoid runaway feedback

• Warming => higher soil respiration + low CO2 solubility => CO2 outgassing

• Concentration driven land-air and sea-air CO2 fluxes should be close to zero or
  combined to <0.1 [Pg C yr-1] drift (10Pg/100yr)

• Minimum of 300 years quasi-equilibrium atmospheric CO2 (esm-piCtrl)

• what about N2O emissions?

Relevant issue:NorESMhub/NorESM#665

@rosiealice @kjetilaas @tjiputra

[DirkOlivie]

I looked at drifts in 4 different CO2-emission driven simulations. These simulations are :

1. n1850esm.ne30_tn14.hybrid.3.0a01.20250410 : a relatively short NorESM3 simulation (16 year)
2. N1850_f19_tn14_20190730esm : part of the CMIP6 piControl (450 year)
3. NHIST_f19_tn14_20191104esm : the CMIP6 historical simulation (165 year)
4. N1850frc2esm_DMS_noBio_noRiv_f19_tn14_20231027 : (2251 year)

The main conclusion from the analysis below is that all these simulations showed unexplained CO2 tendencies in the atmosphere on the order of +/- 1 ppmv[CO2]/century.

I compared the atmospheric CO2 burden difference between start and end of the simulation, with the net amount of CO2 emitted into the atmosphere (fossil fuel emissions + exchange with land biosphere + exchange with ocean) over the period of the simulation.

The change in burden between the start and end of the simulation :

Simulation	start year	end year	CO2 burden start	CO2 burden end	Difference	Difference per year	Difference expressed in ppmv
			Pg[CO2]	Pg[CO2]	Pg[CO2]	Pg[CO2]/year	ppm[CO2]/year
(1)	1	16	2206	2245	39	2.4375	0.31266
(2)	1751	2200	2192	2173	-19	0.04222	-0.005416
(3)	1850	2014	2186	3142	956	5.7939	0.74319
(4)	1851	4101	2191	3538	1347	0.5984	0.07675

The mean emission rate of CO2 over the period (expressed as Pg[CO2]/year and as a change in ppmv[CO2]) :

Simulation	Mean emission rate over whole period	Induced concentration change
	Pg[CO2]/year	ppmv[CO2]/year
(1)	2.409	0.30900
(2)	0.0002314	0.00002968
(3)	5.629	0.72203
(4)	0.6709	0.086057

So the unexplained difference in ppmv[CO2]/year or ppmv[CO2]/century is (difference between last column of the two tables above) :

Simulation	Unexplained trend	Unexplained trend
	ppmv[CO2]/year	ppmv[CO2]/century
(1)	0.0037	0.37
(2)	-0.0055	-0.55
(3)	0.0212	2.12
(4)	-0.0093	-0.93

The mean emission amounts (averaged over the whole period) split into contributions from fossil fuel emissions, land emissions and ocean emissions:

	SFCO2	SFCO2_FFF	SFCO2_LND	SFCO2_OCN
	Pg[CO2]/yr	Pg[CO2]/yr	Pg[CO2]/yr	Pg[CO2]/yr
(1)	2.409	0	0	2.409
(2)	0.0002314	0	0.3842	-0.3840
(3)	5.629	8.921	0.4101	-3.702
(4)	0.6709	0	-0.5547	1.226

Conversion factor from Prather et al. [2012] : 2.086 PgC/ppm

[JorgSchwinger]

Hi @DirkOlivie , thanks a lot for putting this analysis together! I do not quite understand the "Mean emission rate over whole period", though. Are these the air-sea and air-land C-fluxes or something else? For the N1850 compsets, the CO2 emissions should be zero, shouldn't they?

[DirkOlivie]

Hi Jörg, the "Mean emission rate over whole period" is based on the net CO2 fluxes diagnosed/reported by the atmosphere (variable SFCO2). I agree that for the N1850 compsets in equilibrium ((2) and (4)), these fluxes should be close to zero. They are not zero, but considerably smaller than in NHIST (3). Simulation (1) (also N1850) is probably still quite out of equilibrium.

[JorgSchwinger]

Ah ok, thanks for clarifying. I was confused by the word "emissions" because I was thinking of "anthropogenic CO2 emissions", but in SFCO2 everything (land, ocean, FF-emissions) is merged together, right?

[DirkOlivie]

Hi Jörg, yes SFCO2 contains the contribution of all 3 sources (FF, land and ocean). I have added a table with the split of the emissions.

[DirkOlivie]

I have made some plots (based on Jörgs analysis) for 4 available experiments. The left figures represent the TM variables (emissions integrated during model simulation), the middle figures shows the off-line integrated emissions (based on SM variables), and the right figures show both.

The concerning message is that purple and red do not always overlap, and that full and dashed lines do not always overlap. The red curve is obtained by offline summing up the contribution from fossil fuel, land and ocean.

The simulations are :

1. N1850_f19_tn14_20190730esm : emission-driven NorESM2-LM piControl for CMIP6
2. N1850frc2esm_DMS_noBio_noRiv_f19_tn14_20231027 : emission-driven NorESM2-LM piControl
3. NHIST_f19_tn14_20191104esm : emission-driven NorESM2-LM historical for CMIP6
4. NSSP534frc2esm_f19_tn14_20230106+extension : emissiondriven NorESM2-LM SSP5-3.4

For interpreting the variables, one should look at the legends (not at the figure title - which always contains FFF).

The different contributions are :
FFF = fossil fuel flux (CAM reads these emissions from external file)
LND = flux from exchange with the land model
OCN = flux from exchange with the ocean model
FFFLNDOCN = off-line sum of above three fluxes

[JorgSchwinger]

Hi @DirkOlivie, thank you for making this figure! Could you explain which simulations we look at? I assume the last row is SSP5-3.4?

[DirkOlivie]

@JorgSchwinger I have now put a bit more information in the text above the figures on the experiments and on what the variables are. Indeed, the last line is from your SSP5-3.4 experiment. Could you post also your figure? (I can also do it if you allow me). Your results would show an increasing trend in the orange curve of the plots above. Maybe we should to try to find out why that differs.

[DirkOlivie]

The figure from Jörg :

[gold2718]

I have been through the CAM code and found only one place where any of these tracers is updated in a way that is not tracked by the current methods. Aircraft emissions include CO2 as ac_CO2 which is read in from files using the tracer_data functionality (advance_trcdata).
This data is stored in a pbuf field and added in the co2_cycle code to the CO2_FFF and CO2 constituent fields.

The aircraft emissions module converts the emission to a mass mixing ratio and also outputs the MMR value to history (I think as ac_CO2 but I cannot find an existing simulation log file that has aircraft emissions turned on). I think this field should be output and added to the emissions side of the budget when trying to track down the CO2 discrepancy.

[tjiputra]

the ac_CO2 is written out in the SSP534 standard scenario (2015-2100):
/projects/NS9560K/noresm/cases/NSSP534frc2esm_f19_tn14_20230106

but not on the extended scenario (2101-2299):
/projects/NS9560K/noresm/cases/NSSP534frc2esmext_f19_tn14_20230510/

Here, i plot the vertical profiles of CO2 3d variables (CO2, CO2_FFF, CO2_LND, and CO2_OCN) from the atmospheric outputs. I think these variables reflect changes associated with fossil fuel, land, and ocean fluxes. Strangely, the CO2_FFF doesn't go to zero even though, the emission forcing files (surface and aircraft indicated zeros).

[JorgSchwinger]

To clarify: for the above plot (CMIP6 SSP5-3.4 scenario), I took into account the aircraft emissions (the red dotted line is calculated from the two emissions files used in NorESM2, the surface and the ac emissions). Not sure what @DirkOlivie did for his plots.

[adagj]

The aim of today’s meeting was to define a strategy and establish a priority list for addressing the carbon closure problem in NorESM. We also want to quantify the size of the non-closure problem before starting an emission-driven spinup.

Below is the agreed prioritization of simulations

---

General Setup for All Simulations

• Simulation type: Emission-driven with SE history output for CAM
  We will use the noresm3_0_beta10 NorESM#754 tag.
  Mariana will create a recipe below (in Emission-driven spinup #105 )

• Initial conditions:

  refcase="n1850.ne16pg3_tn14.noresm3_0_beta09.Run4.2026-01-06"
  refyear="0251"

• CAM output settings (IMPORTANT):

  • Ensure CAM outputs directly on the SE grid.

  • In user_nl_cam, remove the following lines:

    interpolate_nlat           = 96
    interpolate_nlon           = 144
    interpolate_output         = .true.,.true.

• Mesh verification:

  • Check nuopc.runconfig in your run directory.
  • This file lists all mesh files in use, including the CAM mesh.
  • You can use elementArea in fv1.9x2.5_141008_ESMFmesh_c20191001.nc for diagnostics on SE grid

---

Simulation Priority List

1. Pre-industrial: all fluxes enabled (fossil fuel = 0)

Responsible: Jerry / @tjiputra
Issue: #318

• Configuration: ne16pg3_tn14
• Duration: 30 years
• Purpose: Diagnose the magnitude of the carbon closure problem under pre-industrial conditions and assess readiness for starting the emission-driven NorESM3 spin-up.

---

2. Pre-industrial: all fluxes set to zero in the CAM cap

Responsible: Dirk / @DirkOlivie
Support: Mariana / @mvertens (will explain how to set up the run in the issue)
Issue: #317

• Configuration: ne16pg3_tn14
• Duration: 10 years
• Check before starting: verify that initial CO₂ conditions are not homogeneous.
• Purpose: Test whether transport of CO2 in CAM is conservative.

---

3. Fossil fuel experiment (SSP534-OS) with Surface fossil emissions only

Responsible: Ada / @adagj
Issue: #321

• Configuration: ne16pg3_tn14
• Duration: 75 years
• Setup:
  • Fossil fuel emissions: surface only
  • Land and ocean fluxes: set to zero
• Purpose: Diagnose carbon conservation in CAM under strong positive and negative emissions, focusing on surface vs. 3D emissions.

---

4. Fossil fuel experiment (SSP534-OS) with Aircraft emissions only

Responsible: Ada / @adagj
Issue: #322

• Configuration: ne16pg3_tn14
• Duration: 75 years
• Setup:
  • Fossil fuel emissions: aircraft only
  • Land and ocean fluxes: set to zero
• Purpose: Diagnose carbon conservation in CAM under strong positive and negative emissions, focusing on surface vs. 3D emissions.

---

Additional / Follow-up work

• Testing land emissions:
  • Only land emissions sent to CAM (to be done later).
  • Known issue: there is a water balance error associated with transient land use at the moment, please see nhist.ne16pg3_tn14.nhist_cmip7.constLU.20260118 #313 and errh2o in BalanceCheck fails in NorESM-FATES historic simulations CTSM#186 for details

@MichaelSchulzMETNO @matsbn @JorgSchwinger @kjetilaas

[adagj]

Yes, the intention is to start all simulations from the same initial state

refcase="n1850.ne16pg3_tn14.noresm3_0_beta09.Run4.2026-01-06"
refyear="0251" 

Do you have other suggestions? @JorgSchwinger

[JorgSchwinger]

No, I was just wondering, sounds good.

[oyvindseland]

Can anyone point to an issue with the compset information for prognostic CO2 set-up including running it from a CO2 prescribed experiments as the refcase above

[adagj]

I don't fully understand you question @oyvindseland. Do any of these issues help? #321 #322 #317 #318

[oyvindseland]

Those are good, thank you

[DirkOlivie]

Hi @adagj , in the description above of the second experiment ("2. Pre-industrial: all fluxes set to zero in the CAM cap"), you link to experiment issue #313. Shouldn't the link be to #317? (Possibly I misunderstood it).

[adagj]

Thanks @DirkOlivie ! I have updated the description

[DirkOlivie]

Hi,
some results from experimets 1, 2 and 3.

The figures show change in global CO2 burden (w.r.t. tot the initial conditions). Full lines indicate the mass change based on the column-burdens output. Dotted lines show offline calculated mass change based on emissions output. For the land model, we have also added an estimate of the mass change based on emissions when taking into account the land area fraction.

Variables used in this analysis:

• TMCO2_FFF, TMCO2_LND, TMCO2_OCN and TMCO2
• SFCO2_FFF, SFCO2_LND, SFCO2_OCN and SFCO2
• area [fractional area - sums globally up to 1]
• LANDFRAC

Annual totals based on monthly output :

• the actual length of the months (31, 28, 31, 30, ...) is taken into account when calculating annual averages or totals

Observations:

1. All experiments: there is for all 3 experiments in general agreement between the 4 global column burden values values TMCO2_FFF, TMCO2_LND, TMCO2_OCN and TMCO2 : summing up the first 3 terms offline (represented by the red curve) gives the fourth term (purple curve). In all figures, the red and purple full line overlap. That was not always the in earlier simulations (e.g.,NorESM2). However, in all the experiments here, no aircraft emissions have been used.
2. All experiments: the same is true for the integrated emissions based on SFCO2_FFF, SFCO2_LND, SFCO2_OCN and SFCO2. In all figures, the red and purple dotted line overlap.
3. Experiment 1: there is a discrepancy for the land between the column burden output (TMCO2_LND) and the offline integrated emissions (based on SFCO2_LND). The dotted green and full-line green disagree. Taking into account the land area fractions (dashed green line) seems to improve the agreement initially, but not in the second part of the experiment.
4. Experiment 1 : for the ocean emissions, there is good correspondence between the mass-based evaluation and emission-based evaluation : blue dotted and full line overlap.
5. Experiment 2 : when there are no emissions into the atmosphere, the mass of CO2 in the atmosphere seems to be well conserved.
6. Experiment 3 : when using only anthropogenic surface emissions, the mass-based evaluation and emission-based evaluation seem to agree.

Experiment 1:

Experiment 2:

Experiment 3:

[DirkOlivie]

Experiment from Jerry with only ocean emissions : n1850esm.ne16pg3_tn14.noresm3_0_beta10_onlyOcnFx_20260123. Shown here are annual (upper figures) and monthly values (lower figure).

[DirkOlivie]

Experiment where we sum up the emissions from the land component and the emissions from the ocean component into one flux in the coupler, which we then pass to the atmosphere as one flux. We still see the discrepancy.

The error is maybe related to the fact that fluxes from the land can be large (negative or positive) at some grid points. Maybe the updating of the mass in the atmosphere tops off large fluxes to avoid negative concentrations?

[adagj]

Oh no, I was really hoping this experiment will give us some leads...

[adagj]

@DirkOlivie Are the blue lines below the purple (since they are identical)?

[adagj]

But I think it is now very clear that the discrepancy is related to how the land emissions of CO2 are treated in CAM. However, there must be something with the land emissions since disguising them as ocean emissions didn't help. Like @DirkOlivie suggested, we should check if there are some constraints for large fluxes in CAM.

[MichaelSchulzMETNO]

Would it be possible to multiply the CO2 concentations in the model by a factor of 4 in the restart file and then do a test. I wonder if the carbon "in"closure is proportional to the CO2. Wondering whether there is a hidden flux between land and atmosphere.

[DirkOlivie]

Hi @adagj , I assume that the blue, red and purple lines are equal (blue and red not visible). I assume that the orange line is below the green line.

Currently I am doing 2 new test simulations (after discussing with Øyvind):

1. one where only positive land fluxes are allowed to be passed to the atmosphere.
2. one where only negative land fluxes are allowed to be passed to the atmosphere

[adagj]

@gold2718 @mvdebolskiy

on betzy, in Dirk's run folder : /cluster/work/users/olivie/noresm/

You will find these cases:
n1850esm.ne16pg3_tn4.noresm3_0_beta10.atmco2fluxzero.20260120
n1850esm.ne16pg3_tn4.noresm3_0_beta10.atmco2fluxzero.20260121
n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxviacplocn.20260127
n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxonlypos.20260127
n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxonlyneg.20260127

[mvdebolskiy]

In /cluster/work/users/olivie/archive/n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxviacplocn.20260127/logs/atm.log.1415970.260128-114435.gz:

 2868 CO2_OCN_qneg3                    kg/kg              58 I  CO2_OCN QNEG3 error (cell)
 2869 CO2_OCN_qneg3_col                kg/kg               1 I  CO2_OCN QNEG3 error (column)
 2870 CO2_FFF_qneg3                    kg/kg              58 I  CO2_FFF QNEG3 error (cell)
 2871 CO2_FFF_qneg3_col                kg/kg               1 I  CO2_FFF QNEG3 error (column)
 2872 CO2_LND_qneg3                    kg/kg              58 I  CO2_LND QNEG3 error (cell)
 2873 CO2_LND_qneg3_col                kg/kg               1 I  CO2_LND QNEG3 error (column)
 2874 CO2_qneg3                        kg/kg              58 I  CO2 QNEG3 error (cell)
 2875 CO2_qneg3_col                    kg/kg               1 I  CO2 QNEG3 error (column)
 QNEG3 from Gravity wave drag:CO2_OCN Min. mixing ratio violated at        14 points. Worst =   -0.3E-03
 QNEG3 from Gravity wave drag:CO2 Min. mixing ratio violated at        14 points. Worst =   -0.3E-03

[gold2718]

So these are probably land point issues that show up in the ocean because of Dirk's manipulation? Looking at several runs, it does seem to be consistently due to gravity wave drag that is causing the problem.

[adagj]

@gold2718 Can we make a run turning off gravity wave drag?

[mvdebolskiy]

Yeah, that was just an example from one of the logs. I suspect there are more occurrences than just this.

[DirkOlivie]

This is a new figure from the old NorESM2-LM SSP534 simulations. One obtains better agreement for the fossil fuel CO2 burden if one takes into account the output field ac_CO2 (aircraft emissions) when integrating the emissions over time : the orange full and orange dotted line now overlap better. This output field only exists up to the year 2100 (and is not available in the extension of SSP534). This might also explain why Jörg saw smaller deviations in the IMPOSE simulations as those simulations don't have aircraft emissions.

[DirkOlivie]

A histogram of the CO2 fluxes into the atmosphere from land and ocean. The results are from an older simulation n1850esm.ne16pg3_tn14.noresm3_0_beta03b.BLOMalbAndBrine.2025-10-30, using the monthly output of the years 271-330.

[DirkOlivie]

Difference for the CO2 column burden between the two approaches (monhtly frequency): outputed columnburden minus integration of emissions is shown.

Experiment 1 :

Experiment 3 :

Experiment where atmosphere only receives positive fluxes from land :

Experiment where atmosphere only receives negative fluxes from land :

[gold2718]

For CAM7, the various CO2 fields we are tracking are output to history buffers at (slightly) different parts of the CAM timestep. All of these outputs are after CAM returns from the coupler and before the dycore is called.

• The surface emission fields (fluxes into CAM) are output first, before any physics is called.
• The column burden fields (including CO2) are output in tphysac after mixing with the surface fluxes, CLUBB and microphysics, aerosol wet processes (including wet deposition), and deep convection.
• Addition of aircraft emissions to the CO2 tracers (FFF and total) happens in tphysac, after the column burden fields are output.

Each output is slightly inconsistent but I suppose it should average out over many timesteps. Tracking CO2 conservation internally might be a better way to understand if and how any problems occur in CAM.

[mvertens]

I am taking a different approach at looking at this. I am asking for qneg3 output every time step and I have put the following calculations in physpkg.F90 so that output is available for every physics time step:

• I am calculating the global TM for the four co2 tracers BEFORE the call to tphysac.F90 (call this global_beforeTM)
• I am calculating the global TM for the four co2 tracers AFTER the call to tphysac.F90 (call this global_afterTM)
• I am also calculating the global SF in after the call to tphysac.F90 (call this global_SFLX)
• I am then computing global_TMafter - globalTMbefore - global_SFLX = diffco2 for all tracers
  The run I am doing is starting from year 266 of Dirk's n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxviacplocn.20260127.
• Over a period of almost 3 years and 3 months, other than a handful of points (which I'll describe below) the minimum difference in diffco2 for the total co2 is -0.33157D-13 and the maximum is 0.33326D-13
• Values of global_SFLX are normally on the order of +/- 1.e-4.
• There are a handful of points that exhibit differences that are much greater that 1.e-13 - and these are triggered by qneg calls.
  tphysac: diff TM-SF at nstep 20711 CO2 0.56046D-05
  tphysac: diff TM-SF at nstep 20716 CO2 0.90203D-05
  tphysac: diff TM-SF at nstep 20718 CO2 0.14337D-04
  tphysac: diff TM-SF at nstep 20712 CO2 0.16389D-04
  tphysac: diff TM-SF at nstep 20714 CO2 0.26576D-04
  tphysac: diff TM-SF at nstep 20717 CO2 0.34022D-04
  tphysac: diff TM-SF at nstep 20713 CO2 0.43961D-04
  tphysac: diff TM-SF at nstep 20715 CO2 0.54075D-04
  Note that in these cases TM is always > SF (and this is expected given what qneg3 is doing)
• In summary - I am seeing conservation of CO2 on a physics time step basis - that is not showing up in how the offline post-processing is being done.
• My output is currently in /cluster/work/users/mvertens/noresm/n1850.ne16pg3_tn14.noresm3_0_beta10.global_sums.20260128/run.
• I'm happy to enter this into the noresm3_dev_simulations issues - but I would have to include SourceMods - which I'm wary of adding at this point.

[DirkOlivie]

Hi @mvertens , many thanks for doing this. Can you indicate the units you used for the mass and the flux: e.g. for the values

• the minimum difference in diffco2 for the total co2 is -0.33157D-13 and the maximum is 0.33326D-13, and
• values of global_SFLX are normally on the order of +/- 1.e-4 ?

[adagj]

Hi @DirkOlivie , one thing - in these calculations @mvertens is using a total area of 4*pi - so the values will be much less than what we obtain using the area of Earth.

[mvertens]

@DirkOlivie - I believe that the units are in kg. A code snipped in physpkg.F90 follows:

    total_column_burden_after(:) = 0._r8
    total_sflx(:) = 0._r8
    do m = 1,ncnst
       mm = c_i(m)
       do c = begchunk, endchunk
          ncols = get_ncols_p(c)
          call get_area_all_p(c, ncols, area)
          do icol = 1,phys_state(c)%ncol
             total_column_burden_after(m) = total_column_burden_after(m) + &
                  sum(phys_state(c)%q(icol,:,mm) * phys_state(c)%pdeldry(icol,:), dim=1) * rga * area(icol)
             total_sflx(m) = total_sflx(m) + cam_in(c)%cflx(icol,mm)*ztodt * area(icol)
          end do
       end do
    end do

After this an MPI_REDUCE is done with a SUM in order to have the global values calculated.
I think it would be helpful if all post-processing calculations were clearly spelled out in this issue as well - not just the plots.

[mvertens]

@DirkOlivie @gold2718 - below is a sample output with additional information for time step 19 (where I am also showing the values of TM TM before and after:

tphysac:         TM before at nstep   19  CO2_OCN  0.54467195747815D+02
tphysac:         TM after  at nstep   19  CO2_OCN  0.54467190572850D+02
tphysac: delta   TM        at nstep   19  CO2_OCN  -.51749652953959D-05
tphysac:         SF        at nstep   19  CO2_OCN  -.51749652888288D-05
tphysac: diff TM-SF at nstep   19  CO2_OCN  -.65670176409424D-14

tphysac:         TM before at nstep   19  CO2_FFF  0.54371166592412D+02
tphysac:         TM after  at nstep   19  CO2_FFF  0.54371166592412D+02
tphysac: delta   TM        at nstep   19  CO2_FFF  0.00000000000000D+00
tphysac:         SF        at nstep   19  CO2_FFF  0.00000000000000D+00
tphysac: diff TM-SF at nstep   19  CO2_FFF  0.00000000000000D+00

tphysac:         TM before at nstep   19  CO2_LND  0.54373632107406D+02
tphysac:         TM after  at nstep   19  CO2_LND  0.54373659768733D+02
tphysac: delta   TM        at nstep   19  CO2_LND  0.27661327145267D-04
tphysac:         SF        at nstep   19  CO2_LND  0.27661327154041D-04
tphysac: diff TM-SF at nstep   19  CO2_LND  -.87730387191010D-14

tphysac:         TM before at nstep   19  CO2      0.54469661262810D+02
tphysac:         TM after  at nstep   19  CO2      0.54469683749171D+02
tphysac: delta   TM        at nstep   19  CO2      0.22486361878293D-04
tphysac:         SF        at nstep   19  CO2      0.22486361865212D-04
tphysac: diff TM-SF at nstep   19  CO2      0.13081631047504D-13

As you can see - because of the fact that there is many orders of magnitude difference bewteen TM and SF, several decimal places of the surface flux will be lost in adding to the existing CO2.

[DirkOlivie]

Hi @mvertens, thanks for this.

The inconsistency which you find for one half-an-hour timestep (e.g., -.65670176409424D-14 kg[CO2]/radians^2) would correspond with a drift of 4602 kg[CO2]/yr on an annual basis for the whole atmosphere:
(0.65e-14 kg[CO2]/radians^2 * (6.37e6 m)^2 * 365 days/year * 48 (timesteps/day) = 4620 kg[CO2]/year ).
Also, 4620 kg[CO2] corresponds with 1260 kg[C]/year.

In the post processed data we find a drift of around 1Pg[C]/(30 yr) =0.03 Pg[C]/yr = 30 Tg[C]/yr =3000e6 kg[C]/year. So in the post-processed data we have a drift 1-3 million times stronger than in the online verification.

[tjiputra]

@DirkOlivie, @mvertens,
(i) Mariana calculates [CO2] while Dirk [C]
(ii) the drift that we see in the post-processed is non-linear (i.e., fluctuates between positive and negative). So we need more than one time step in Marianna's calculation (e.g., one year). Or maybe redo the calculation for summer months, when the drift is largest.

[mvertens]

@DirkOlivie @tjiputra - I have 3 years of log file output in /cluster/work/users/mvertens/noresm/n1850.ne16pg3_tn14.noresm3_0_beta10.global_sums.20260128/run/atm.log.1422710.260204-202734.gz. The log file has entries like the following:

tphysac: delta   TM before at nstep   0  CO2_FFF  0.54371166592599D+02
tphysac: delta   TM after  at nstep   0  CO2_FFF  0.54371166592599D+02
tphysac: delta   TM        at nstep   0  CO2_FFF  0.00000000000000D+00
tphysac:         SF        at nstep   0  CO2_FFF  0.00000000000000D+00
tphysac: diff TM-SF at nstep   0  CO2_FFF  0.00000000000000D+00

However - the first two entries have a typo and it should appear as

 tphysac:         TM before at nstep   0  CO2_FFF  0.54371166592599D+02
 tphysac:         TM after  at nstep   0  CO2_FFF  0.54371166592599D+02
 tphysac: delta   TM        at nstep   0  CO2_FFF  0.00000000000000D+00
 tphysac:         SF        at nstep   0  CO2_FFF  0.00000000000000D+00
 tphysac: diff TM-SF at nstep   0  CO2_FFF  0.00000000000000D+00

What is concerning to me is that periodically there are entries like the following:

tphysac: delta   TM before at nstep   5  CO2_FFF  0.54371166592544D+02
tphysac: delta   TM after  at nstep   5  CO2_FFF  0.54371166592544D+02
tphysac: delta   TM        at nstep   5  CO2_FFF  0.71054273576010D-14
tphysac:         SF        at nstep   5  CO2_FFF  0.00000000000000D+00
tphysac: diff TM-SF at nstep   5  CO2_FFF  0.71054273576010D-14

I'm not sure I understand how delta TM at nstep 5 CO2_FFF 0.71054273576010D-14 can be non-zero where the input SF is zero.
I'm also not sure what to do with all of this output at this point.

[oyvindseland]

I remember there was a discussion on running a CO2 emission driven test using MAM. Was that done in case there are some array overwrites in the emissions routine?

[gold2718]

We have not had success getting that to run. NCAR is not using or supporting this scheme and the scheme they are using (ghg_mam4) includes CO2 and they are not using the emissions co2_cycle code that we are using (at least not yet).

[mvertens]

I have done a 5 year run of case n1850esm.ne16pg3_tn4.noresm3_0_beta10_20260120 as a hybrid starting from year 266 of n1850esm.ne16pg3_tn4.noresm3_0_beta10.lndco2fluxviacplocn.20260127. I have put in new global sums that are cumulative for TMoffset and SF for LND, OCN, FFF and total. Note that there is no FFF input in this run co2_readflux_fuel = .false.


The first plot shows the difference between accumulated SF and TMoffset for the LND, OCN and FFF for the first ~500 days - and they lie on top of each other. Then something very strange happens on time step 23212 - the TMoffset-SF (LND) jumps up by 2 orders of magnitude and effectively stays there for the duration of the run. You can see this in the second plot on the right - where I have extended the time for the duration of the run. There is clearly a problem here. In addition, even for FFF - there seems to be roundoff level changes that are occuring between tphysbc and tphysac that should be zero but are not.

However, despite this - the errors are so small that you do not see this in the plot showing TMoffset and SF (the two curves for each component effectively lie on top of each other). I am also now output ing in units of Pg. I am not seeing anything close to the difference that @DirkOlivie was seeing. I can try another run that is identical to the one that @DirkOlivie ran. @DirkOlivie - can you point me to the best way to try to duplicate your results - but using my inline versus offline calculations?

[tjiputra]

@mvertens , could you add the 'LND SF' and 'OCN SF' calculated from the netcdf outputs together in your lower left figure? Alternatively, if you provide me with the case name and location in nird, i can calculate and plot them. Thank you.

[mvertens]

@tjiputra - my data is in /nird/datalake/NS9560K/mvertens/n1850.ne16pg3_tn14.noresm3_0_beta10.global_sums.20260128/atm/hist.
You want to look at 1850.ne16pg3_tn14.noresm3_0_beta10.global_sums.20260128.cam.h2a* (which contain daily averaged output of the fields you want to look at). In addition, 1850.ne16pg3_tn14.noresm3_0_beta10.global_sums.20260128.cam.h3i* contains the daily instantaneous values of those same fields.

[tjiputra]

@mvertens , i tried to reproduce (only) cumulative SFCO2 fields as in your lower left figure (bottom panel in my figure). There is discrepancies whether I use daily or monthly netcdf outputs. The daily outputs (dashed lines) look more similar to yours.

I then calculated the difference between these daily and monthly fields only for lands (red line, top panel, convert it to ppm unit) and compare that with the atmospheric CO2 drift (mass imbalance), equivalent to Dirk's purple lines. They both has same drift.

If my calculations are correct, then the issue is in the SFCO2 'monthly' (averaging routine) outputs. The daily outputs looks more comparable with your online calculations.

[mvertens]

@tjiputra - thanks for the plots. I have added both my plots and your plots for tomorrow's discussion. Please feel free to edit these!