The primary goals of this meeting were to update everyone on code status and present results from the: 3-degree coupled model effort, 1-degree biogeochemistry initialization run, and 3-degree ocean initalization studies.

Code is flowing, but only in "non-concurrent coupling" mode, which means that we can run the full CO2 system, but only serially on 30 processors, giving us a wall-clock time of only 8 model months per day. With concurrent coupling on 150 pes, we should be able to increase this to 3 years per day on the origin and twice that on the Altix. Here is a plot of CO2 fluxes with an atmospheric cold-start:



We have yet to verify that carbon is conserved in the full system, however, or to attempt to link the CO2 tracer in the atmosphere to the CO2 in the radiation feedback.

Postmeeting Followup on adding tracer fluxes from rivers:

Sarmiento had asked the question of putting DIC (and other tracers) in river fluxes from the land model to the ocean model. We will need to do this eventually, particularly as we seek to consider nitrogen explicitly and move to higher model resolution on continental shelves. Unfortunately, this feature is currently beyond our near-term stated goals or capabilities. It is a very important issue, however, and deserves an entire discussion of its own, particularly to determine:

1) the scope of the coding necessary in the hydrology model - I talked with Chris Milly, and he felt like adding advection and diffusion of tracers would be a fairly straightforward but coding-intensive task. He recommended not doing this activity based on the current LM2 infrastructure implemented in CM2, but rather the newer, more physically motivated code that he has been working on more recently. This code has yet to be merged with the vegetation model (LM3v), but it sounds like this merge will be done, I am hoping in the coming months, with Sergey Maleshev. Then someone would have to be dedicated to further adapting this code for generic tracers. Sergey maleshev has agreed that he is, in principle, a likely candidate for this task.

2) the feasibility of adding C and N losses to the land model (i.e. Pacala, Shevliakova, Maleshev, Levin). Stephen Gerber, a postdoc with Pacala, has

3) the scope of coding necessary to have variable concentrations in river outflow to the ocean model - I talked with Griffies and it seemed like a fairly straightforward code change. A separate issue is whether MOM will be suitable for the kind of high-resolution studies we would want to do with such a model.

4) whether or not this would need to take place in the context of continental weathering and marine sedimentation models, and how best to assure conservation of tracers in that case - Jorge had a postdoc, Jeff Greenblatt, working on the sediment model a couple years ago, but it only got as for as a toy, 2-D version. Someone would also need to investigate a way to implement weathering and sediment transport in rivers.

This is a large enough task that someone would have to be dedicated to the development of this capacity. While the nitrogen cycle will require this to happen eventually, this project has not yet been given a high priority and thus seems at least a year or so off in the future unless someone decides to take the problem on in earnest.

120 year Ocean-only spin-up in OM3
This was a foray into attempting to spin up the ocean offline. Here is surface Chlorophyll and Surface Nitrate from the model run out 120 years:




CO2 fluxes from the model look fairly reasonable After 120 years, the net CO2 flux is out of the ocean at about 0.5 PgCO2/yr... Our hope is to bring this down to below 0.1 for climate change studies. Because of the circulation differences and strong ice-cover - gas-exchange relationship, it was considered that a formal spinup would require the full model.

Post-meeting note on the ice-gas-exchange formulation:
The ESM appears to be indeed taking into account the fraction of open water in each grid cell in the appropriate way)
400 year initialization study (Toggweiler)

A concern in the development of the GFDL earth system model is the possibility that our simulations will be compromised by drift in the oceanic carbon system. The basis for this possibility is the fact that our early ocean-only carbon cycle models have tended to have more or less dissolved inorganic carbon (DIC) than observed. This means that any coupled model initialized with the observed distribution may have to run for more than 1000 years in order to produce an atmospheric pCO2 tendency that is free of drift.

Dunne and Toggweiler set out several months ago to determine whether or not the biases in our ocean biogeochemistry model could be anticipated through a series of short model runs. If so, the drift might be eliminated by through adjustments of the the initial DIC concentrations that would compensate for the biases.

Work by Murnane et al. (1999) and Toggweiler et al. (2003a,b) has shown that surface to deep DIC differences in solubility-only and organic-only versions of MOM are biased low and high, respectively, with respect to the real ocean. These results were used to set DIC targets that might anticipate the solubility-only and organic-only pump strengths of OM1p5, assuming some improvement over the old Murnane et al. model. Solubility-only, organic-only, and full carbon system versions of OM1p5 were set up to test whether the targets were realistic. The targets were uniform DIC concentrations of 2173 umol/l for the solubility-only model, 2281 umol/l for the organic-only model, and 2305 umol/l for the full model.

The three models were run for 400 years with a fixed pre-industrial atmospheric pCO2 and uniform initial DIC concentrations at the target values. If OM1p5 performed as expected, the deep concentrations would remain close to the initial values. This would be an indication that Dunne and Toggweiler might actually have some skill at anticipating the biases of a new model on the basis of past performance. If the individual results drifted away from the target concentrations it would be an indication that the situation is more complicated. Another round of adjustment might then be necessary to produce a better initial guess. Ocean-atmosphere pCO2 differences in the polar regions were monitored to see if they might indicate how the deep concentrations were trending. Toggweiler reported on these results in the meeting.

Solubility Pump

The strength of the solubility pump in OM1p5 was stronger than anticipated and probably much more like the real world than the solubility pump in the old Murnane et al. model. After 400 years, DIC concentrations below 3000 m had reached 2180 umol/l, well above the target of 2173. A model with a strong solubility pump should have small pCO2 deficits in the areas of deep-water formation, and indeed, the pCO2 deficits in the Weddell and Ross Seas in OM1p5 were only 40-60 ppm in relation to deficits in excess of 80 ppm in the Murnane et al. model. Toggweiler et al. (2003a) had estimated that solubility-only pCO2 deficits in the real Weddell Sea might be as low as 30-40 ppm. pCO2 deficits in the GIN Seas in the North Atlantic were about 60-80 ppm in both models in good agreement with observations. The surface to deep DIC difference in the real world due to solubility differences is about 120 umol/kg out of 165 umol/kg at full solubility (i.e. with fast gas exchange). This means that the real ocean's solubility pump is about 73% efficient. This estimate is consistent with initial pCO2 deficits of 60-70 ppm and 30-40 ppm in the areas of deep water formation in the north and south, respectively.

These particular figures, 120 and 165 umol/kg, were derived from the three-box model, which uses equilibrium constants that should be similar to those being used in OM1p5 and OM3. They do not, however, include the effect of salinity differences (because the three-box model doesn't have any). In order to come up with a comparable set of targets for our new models I need to know the surface-to-deep DIC difference in the solubility-only version of OM1p5 at full solubility with salinity differences after 1000-2000 years. This version of OM1p5 should suffice as a reference for OM3/CM2.1U if the temperatures and salinities in OM1p5 are similar to those in the coupled model.

My strategy for initializing OM3/CM2.1U with respect to solubility is as follows. a) We start with the surface to deep DIC difference from the long run of OM1p5 with fast gas exchange and salinity differences. This difference will be smaller than the 165 umol/kg figure above because it will include the salinity effect. b) We multiply this difference by 0.73 to account for the effect of finite gas exchange rates. c) We add this number to the average DIC concentration that we got in our original solubility-only run of OM1p5 for low-latitude surface waters (50S to 50N). This should give us an average deep DIC concentration that we can use to initialize a new solubility-only test run. d) We run OM1p5 for 200-400 years to see how its deep DIC concentrations deviate from the target. We examine the polar pCO2 deficits to see how they correspond to the expectations in Toggweiler et al. (2003a).

The sign and approximate magnitude of the deviation from the target is saved for use in initializing OM3/CM2.1U.

Organic Pump

Surface-to-deep DIC differences due to the production and remineralization of organic carbon in the Murnane et al. model range from 90 umol/kg with fast gas exchange up to 144 for normal gas exchange. The initial guess that we used for our organic-only test run in OM1p5 corresponds to 132 umol/kg on this scale. The organic pump in the real world is probably somewhat smaller.

The Murnane et al. runs were done with the effect of the virtual salt flux on DIC switched off. The alkalinities in these runs should have been the same everywhere. We cannot run OM1p5 without salinity differences. So the organic-only runs that we carried out back in August included the effect of salinity; the alkalinities were also scaled to salinity as I recall. Including the salinity effect should have increased the "organic-only" DIC concentrations in the upper ocean and reduced them in the deep ocean. Now that I am focusing on this aspect of the problem I can see that the "skewed" profile that OM1p5 produced is not due to nutrient/DIC trapping or weak ventilation, but is, in fact, simply a reflection of the salinity differences in the model!!

The strategy for initializing an organic-only run in any one of our new models is as follows. a) We start from the 132 umol/kg surface-to-deep DIC target above and scale it up by 1.025 to convert it to umol/l. b) We add this number to the average DIC concentration between 50S and 50N that we got in our original organic-only run of OM1p5. c) We then multiply this number by the global average salinity at each model depth and divide by 34.7 to construct an initial DIC profile that is based on the original target but includes the salinity effect. d) We re-run OM1p5 for 200-400 years to see how the model deviates from this target. We examine the polar pCO2 excesses to see how they correspond to the expectations in Toggweiler et al. (2003b).

The sign and approximate magnitude of the deviation from the target is saved for use in initializing OM3/CM2.1U.

The process in steps a-d above assumes that the initial PO4 concentrations in NADW and AABW in OM1p5 and OM3/CM2.1U agree with those in the Murnane et al. model and that the deep ocean in the new model is filled with similar proportions of northern and southern water. The process also assumes that the new models use the same Redfield ratio for organic matter as the Murnane et al. model. The C:P ratio assumed for organic matter in the Murnane et al. model was 120 while the value used in OM1p5 is variable as a function of nutrient stress and nitrogen fixation

Carbonate Pump

We haven't spent a lot of time worrying about the carbonate pump because the results from box models suggest that the air-sea pCO2 differences associated with the carbonate pump are small. This means that we can predict the surface- to-deep DIC difference from the surface-to-deep Alkalinity difference via the following relationship

DICd - DICs = 1/2*(Alkd - Alks + NO3d - NO3s)

where DICd - DICs is the surface to deep DIC difference due to the production and dissolution of CaCO3. A rough guess for a carbonate pump target is about 85 umol/kg. We should check the surface to deep alkalinity differences in OM1p5 to see if they look reasonable with respect to CDIAC.

Final Initialization

If we add up the target DIC differences above we get 337 umol/kg (120 solubility + 132 organic + 85 CaCO3). None of the target values includes the salinity effect, which should reduce the total DIC difference by some 24 umol/kg. Thus, the three pumps minus the salinity effect = 313 umol/kg. Volk and Hoffert give a value of 325 umol/kg for the real ocean (based on GEOSECS differences outside of the circumpolar region. We need to check all these numbers against the CDIAC data.

These target values should be adjusted so that the total is closer to the observed pre-industrial difference. Our targets do not have to be perfect. These targets will be used to adjust the initial DICs in OM1p5. We can then make new targets on the basis of what we learn from OM1p5 that will give us a basis for adjusting the initial DICs in OM3/CM2.1U. With each iteration the procedure should get better.

In our first round of tests with OM1p5 we initialized the full model with the global average CDIAC DIC concentration, 2305 umol/l (2249 umol/kg). The predicted deep concentrations stayed very close to the initial value but the upper ocean lost a great deal of DIC. It is not surprising that the upper ocean lost a lot of DIC since the upper ocean was initialized with the global average DIC and the atmosphere was held fixed at 280 ppm. I'm not sure that we learned anything from this run.

In future test runs we should initialize the full model with the CDIAC data. Then we will get a sense of how the full model drifts with respect to the observations. This is not necessarily a good basis for adjusting the model because there could be problems unrelated to the initial DIC, like bad initial PO4s, that are causing the full model to drift. Observing the drift of the solubility-only and organic-only models along with the drift in the full model should help us focus in on the problem areas.

We propose that the final initialization procedure for OM3/CM2.1U would look something like this. We produce new targets for the solubility- and organic-only runs based on the OM1p5 results. We then do simultaneous 100-year test runs in OM3/CM2.1U with the atmospheric pCO2 set to 280 ppm to evaluate the solubility-only, organic-only, and full versions of the model. On the basis of these runs we adjust the deep CDIAC data to compensate for the observed deviations. We should then be ready to roll with an atmospheric pCO2 that is allowed to vary.

So, we need to re-evaluate our targets and do one more set of test runs with OM1p5 (in addition to the long full solubility run) before taking on OM3/CM2.1U.



Post-meeting Note on DIC at 400 m:
The big model-data differences seen for DIC were due to a units problem in the comparison - model mol/m3 and data mol/kg. When the data is converted to mol/m3, the problem disappears.)

1) To our great disappointment, changing from the b-grid to the finite volume did not improve the simulation markedly. It improved the northern jet slightly, but degraded SSTs... other effects? Is it bad ocean parameters? Is it switching from 2-deg ocean to 3-deg ocean? Is it switching from 50 layer ocean to 28 layer ocean? Is it switching from 2-deg atmosphere to 3-deg atmosphere?
2) The simulation is not currently robust enough for scientific work (i.e. <10 deg SST bias, >10 Sv NADW formation).
3) Because the simulation is not an order of magnitude faster than CM2.1U, such that it is not even a worthwhile workhorse for code/model development/initialization.
4) While there are many intriguing possible fixes, no one has the time to do investigate them at this time.
5) No more runs with CM1p5 will be done in the near future under the auspices of ESMDT.
6) Rather, all runs will be done in the context of CM2.1U
The CM1p5 finite volume run can be found in:
/archive/jpd/khartoum/CM1p5_new_coupler_lm2
and the CM1p5 b-grid run can be found in:
/archive/jpd/postjakarta/cm_paleo_test3

A patch is in place that should allow us to run on the Altix once codes are merged.

1) Ascertain whether carbon is conserved in the model
2) Establishment of Standard Diagnostic Suite
3) Get model working in concurrent mode to bump up to 150pes
4) Wait for LM3v to be frozen
5) Spin up land model with fixed CO2 atmosphere
6) Spin up ocean model with spun-up land and fixed CO2 atmosphere
7) Develop code to allow radiation code to "feel" the atmospheric CO2 tracer
8) Start running on Altix when it frees up.

1) First results for CO2 fluxes in the fully-coupled and atmospheric CO2 data-override version of the ESM.
2) Establishment of Standard Diagnostic Suite

• About us
  • Welcome
  • Visiting GFDL
  • Overview
  • Tour
  • Gallery
  • People
  • Employment
  • Help
  • Notices
• Research
  • Atmospheric Physics and Chemistry
  • Biospheric Processes
  • Climate Diagnostics
  • Climate Dynamics and Prediction
  • Oceans and Climate
  • Weather and Atmospheric Dynamics
• Products and Services
  • Flexible Modeling System
  • Our Data Portal
  • Our GForge
  • Technical Services
  • Visualization
• Reference
  • Our Online Bibliography
  • Our Meetings and Seminars
  • Our Library
• GFDL Only
  • External forms and links
  • External web mail
  • Help Desk
  • Internal Administrative Services
  • Internal Technical Services
  • NOAA's Hot Items
  • webTA Login