Attending: Olivier Aumont, Laurent Bopp, Patrick Brockmann, John Bullister, Kenneth Caldeira, Jean-Michel Campin, Scott Doney, Helge Drange, Jean-Claude Dutay, Nicolas Gruber, Akio Ishida, Philippe Jean-Baptiste, Fortunat Joos, Robert Key, Corinne Le Quere, Keith Lindsay, Ferial Louanchi, Harilous Loukos, Ernst Maier-Reimer, Nicolas Metzl, Patrick Monfray, Anne Mouchet, Raymond Najjar (minutes taker), James Orr, Kasper Plattner, Reiner Schlitzer, Steven Spall, Anne Stoens, Ian Totterdell, Jorge Sarmiento, Christopher Sabine, Christoph Voelker, Marie-France Weirig, Yasuhiro Yamanaka, and Andrew Yool
Jim Orr opened the meeting with a welcome and brief description of OCMIP-1 results. After some logistics, he noted that OCMIP is partly funded by JGOFS, as well as GAIM. In reviewing the timetable, Jim noted that we are well behind original plans for submitting model output. On the other hand, developement of the model analysis package is proceeding on scehdule. Jim asserted that a new timetable needs to be developed (see below).
John Bullister then gave an overview of CFC data sets from WOCE and other programs. Most CFC data are CFC 11 and 12, but there are also CFC 113 and carbon tetrachloride data available. John began by showing the atmospheric histories of the CFCs. These histories are based on direct measurements in recent decades and inferences from industrial emission patterns before that. John noted that the concentrations of CFCs have started to level off in recent years. Carbon tetrachloride (CCl4) is closer than CFC 11 and 12 to anthropogenic CO2 in terms of its time history, but, unfortunately, shows some non-conservative behavior (due to hydrolysis), particularly in warm waters. CCl4 may be most useful as an anthropogenic CO2 analogue, therefore, in deep waters. The atmospheric histories are also useful in that an apparent age can be estimated from the computed CFC partial pressure (from the measured CFC concentration and the solubility).
NSF has funded the analysis of the WOCE CFC data, which will be conducted by a few U.S. labs in collaboration with European and Canadian labs. About 10 labs in all were responsible for the CFC data collection. Part of the analysis will be to quality control the data, making sure, for example, that blank corrections were done properly. All the major ocean basins were covered by WOCE, with many transects in each basin.
Some plots of CFC sections were shown. Most surface waters are within a few percent of saturation, except in high latitudes where significant undersaturation can occur. The CFCs nicely trace the formation of deep water in high latitudes, subtropical convergence and equatorial currents at depth. The deep western boundary current is also seen in the Pacific as far north as 10 S.
John showed an analysis of the CFC budget in AABW. The CFC inventory was computed in AABW (defined by a certain density class) by interpolating sparse data and normalizing to a common time. By assuming knowledge of the preformed CFC concentration, an estimate of 8 Sv of AABW formation rate was computed.
Jean-Claude Dutay then showed results and a preliminary analysis of the OCMIP-2 CFC simulations. The analysis was done mostly with the GOSAC Analysis Package (GAP, see below). CFC output from the 12 3-D models have been received and all have been checked except for one model. A few models were not accepted because of mass conservation problems, and these problems are currently being resolved. Two of the models do not have seasonality. The models were briefly summarized in a table in terms of grid size, mixing parameterizations, mixed layer physics, sea ice parameterization, surface boundary conditions, etc. Najjar suggested that this table be put on the OCMIP web site because is a concise and useful summary of the OCMIP-2 models.
Jean-Claude showed a time series of the cumulative flux for all the models. At the end of the run, the standard deviation of this flux is 20% and the range is 60%. An interesting figure showing high correlations (r2 ~ 0.99) of the cumulative fluxes of the models versus the mean of all the models drew a lot of discussion. Most models showed large cumulative inputs in the North Atlantic, the Southern Ocean and the Equatorial Pacific, and small cumulative inputs in the subtropical gyres.
Jorge Sarmiento pointed out that the inventory differences among the models might simply be due to the different SSTs in the models. There was some discussion suggesting that a comparison of pCFCs would be better than CFCs, at least in terms of sorting out circulation differences. Inventory maps revealed that the highest values were in the North Atlantic and Southern Ocean for most models. Jean-Claude also showed maps of the inventory minus the cumulative flux. Maps showed smooth distributions for most models, except for the AWI (Alfred Wegener Institute) model, which appeared "speckled." Reiner Schlitzer noted that this might be due to the relatively low diffusivity in his AWI model.
Sections were then shown for the model and observations. All models revealed NADW formation in the CFCs, but most appeared to be too weak in this sense, and this generated much discussion. An east-west section in the North Atlantic showed tremendous differences among models; some models appeared to reproduce the CFC imprint of the deep western boundary current, some did not. The position of the subsurface maximum in the CFCs differed greatly among models. A comparison was made with CFC data from Ray Weiss on sigma-1.5 = 34.63 with model CFC at the depth of the maximum, which also revealed large differences, though some pointed out that the model output should also be presented on the same surface for a fair comparison. The AJAX section in the S. Atlantic also showed very large differences; some models had no penetration at high latitudes, others looked fairly good. The Princeton and NCAR models appeared to have too low penetration here. The AWI model appeared to stand out from the rest as being particularly good. This may not be surprising because the model circulation is computed diagnosticly from hydrographic and nutrient data (though not CFC data). It is then run forward in time as are the other models to compute oceanic invasion of CFCs.
Future analysis by Jean-Claude will be more quantitative, including a comparison of section inventories between observations and models.
Some general discussion followed. Scott Doney made the point that it is important, when analyzing the CFC model output, to distinguish between the impact of deep water formation and the propagation of the CFC signal away from the formation region. Jorge Sarmiento noted that the main purpose of the CFC runs is to help sort out the anthropogenic carbon budget in the ocean. We can use the models to understand relationships between tracers (such as CFCs and anthropogenic CO2) and then use the CFC data to get at the actual anthropogenic CO2 budget. (Some of these ideas were elaborated on in his presentation, described below.) He also noted that another important OCMIP goal is to pin down the air-sea CO2 (natural plus anthropogenic) flux. Many pointed out that though our goal is to study the marine carbon cycle, it is clear that CFC model runs provide a lot of information about the ocean circulation models themselves.
Jim Orr then led a discussion of the OCMIP-2 timetable. After much discussion, the following timetable was agreed to:
Ray noted that it is important for him to receive results from the equilibrium runs on time if he is to get the analysis done.
There was some discussion of whether to do a Biotic anthropogenic C12 run. It was decided not to do this run because of time constraints and because it is expected that the anthropogenic CO2 in biotic and abiotic runs will be very similar.
Olivier Aumont then described the proposed CO2 Injection HOWTO. The goals of doing the injection simulations are: to estimate how long injected CO2 stays in the ocean, what the dependence of this injection on site and depth is, and what the pH impact will be in the ocean if injection is performed.
Olivier Aumont noted that his simulations showed little difference between injection performed in biotic and abiotic background fields, so the proposed HOWTO has no biology. He also evaluated the nonlinearity in the interaction of plumes from the seven proposed injection sites and found it to be minimal. Thus the sum of the separate tracer fields produced by the seven sites is essentially equal to the tracer field produced by all seven sites together. Ten tracers were proposed. Tracer 1 has atmospheric CO2 fixed at 280 ppm in order to evaluate model drift. Tracer 2 has atmospheric CO2 specified according to the S650 IPCC stabilization scenario (identical to the future abiotic simulation). Tracers 4 through 10 are for the individual injection sites, each with atmospheric CO2 specified according to S650. Tracer 3 accounts for air-sea gas exchange due to CO2 injection.
The seven sites are near major metropolitan areas, and 3 injection depths will be simulated (800 m, 1500 m and 3000 m), making for a total of 26 tracers for the "future" runs. Also planned are "emission scenarios," in which CO2 emission rates would be specified instead of atmospheric CO2 values. The proposed schedule for the sequestration runs is as follows:
Fortunat Joos noted that IPCC has four emission scenarios that span a wide range. There was some discussion about which simulations to do first: specified atmospheric CO2 or specified emissions. Ernst noted that IPCC appears to be moving more towards emission scenarios as opposed to stabilization scenarios. Jorge suggested using WRE650 instead of S650 and to also consider going to a higher CO2 level, like 4X present day CO2, which many climate models suggest has a dramatic impact.
Fortunat Joos then described how CO2 pulse simulations might be done in OCMIP-2. In a linear system, the response to a pulse input can be used to characterize the response to any input. The CO2 system in seawater is nonlinear, however. Fortunat devised a scheme for analyzing the output of the pulse runs in such a way as to get around the problems with nonlinearities. The proposed Pulse component of the Abiotic HOWTO is to simply initialize the model at some steady state, double atmospheric CO2 instantaneously, and monitor the decrease in atmospheric CO2. Reconstructions of simulations with various input scenarios showed that the approach seems to work quite well. Corinne noted that just because it worked for one GCM doesn't mean it will work for all of them. Fortunat promised to analyze the OCMIP-2 model pulse runs and attempt to reconstruct the WRE650 simulations. Other scenarios could easily be done using the output from the pulse runs.
Patrick Brockmann then gave a demonstration of the package he has been developing for analysis of the OCMIP-2 model output and associated observational data sets. An important constraint he had to abide by was to work with the original horizontal grid of each model. The grids vary considerably, and include curvilinear, diamond, regular, semi-regular and unstructured grids. Tecplot 7.5 was chosen because it is the only graphical software package that includes this flexibility. The problem is that Tecplot is not free (unlike Ferret and GMT, for example) and each group will have to buy Tecplot in order to use the analysis package. The new GDT standard which relies on NetCDF was chosen for model output storage. GDT facilitates efficient storage of information for any model grid.
Tecplot is fast, flexible and interactive. It was originally designed for applications in computational fluid dynamics and was developed in C for UNIX operating systems. A user can launch GAP, a package developed by Patrick Brockmann as a Tecplot Add-on. Currently, GAP produces maps with spherical and cylindrical projections, maps of cruise tracks, and sections along any predefined cruise track. Multiple frames are easy to plot. Coming soon will be the capability to make tracer-tracer scatter plots, zonal averages and inventories. (Currently, Ferret is commonly used for data manipulation, e.g., to compute inventories, then the output is saved in GDT format and read in for graphical presentation by Tecplot.)
Patrick Brockmann showed some examples with the CFC model output. The interface allows one to easily choose a model, a particular time period, and whether or not to show the grid boundaries. The CFC observations are also part of the package. Zooming is easy, as well as displaying the numerical value at a given point in a field. Contouring and shading can be done. Details, like line thicknesses, contour intervals, etc., are easy to manipulate. In short, Tecplot has a lot of "tools."
Some movies were shown as well. Free software known as "Framer" (for PCs and UNIX machines), and also from the makers of Tecplot, was used to view Tecplot figures sequentially. Some amazing movies revealed the penetration of CFCs into the ocean interior in deep convective plumes.
Patrick Brockmann has never had a memory problem with Tecplot. Storage, however, may be an issue; for example, the whole CFC database (models plus observations) is 10 Gbytes. There was some discussion about how the model output should be distributed. CD ROM is a possibility. FTP is also possible if folks want only part of the data base.
Ray Najjar then described the justification for the Biotic runs. Originally planned were two biotic runs: a nutrient-restoring run and a model-specific biological run (such as an ecosystem model). The nutrient-restoring run was planned, in part, to help analyze differences between the model-specific biological runs (which have different circulations *and* biology). The nutrient-restoring runs also allow the estimation of fields of new production and air-sea CO2 fluxes consistent with surface nutrient observations. There was some discussion as to whether there would be time in this phase of the project for model-specific biological runs. Ray argued that it was important to go beyond the nutrient-restoring runs, and some groups expressed interest in doing so as part of OCMIP, though no formal plans were made.
Ferial Louanchi then showed the seasonal phosphate maps she created from the NODC data base that are used to force the nutrient-restoring runs. Seasonal variations are clearly seen at middle and high latitudes.
Ray then briefly showed results of the nutrient restoring runs from the NCAR group (Scott Doney and Keith Lindsay). The model was not yet completely spun up, though surface waters seemed to be close to equilibrium. Qualitatively, the model results for new production, air-sea CO2 flux, latitude-depth sections, etc., seemed reasonable. Ray noted that the NCAR group has been instrumental in pushing the development of the nutrient-restoring runs.
Bob Key then described the Pacific radiocarbon observations made as part of the WOCE program. As of now, all Pacific samples (about 10,000) have been measured and some of the data can be accessed on the WOCE web site. Some additional data are available through Bob. Indian and Atlantic data are not yet available. The errors in the WOCE radiocarbon data are similar to the errors for GEOSECS, TTO and SAVE: about 2 to 4 permil. Bob showed Pacific C-14 maps for the surface and several potential density surfaces. One interesting feature is that the C-14 minimum is off of California, whereas most models put it further north. The deep western boundary current can clearly be seen in the South Pacific as a tongue of high C-14 on potential density surfaces, but near-bottom values most dramatically show this feature. Two low C-14 bull's-eyes can be seen along 32 S, which also show up in silicate. Bob suggested that it will be a challenge for models to reproduce these features.
Some GEOSECS-WOCE differences were shown. There's not much difference in deep water, except around Antarctica. Surface values have generally decreased significantly from GEOSECS to WOCE, except near the equator, where little change is seen. In the East Pacific, thermocline values have increased. The GFDL model shows much larger differences between the two programs.
Some analysis was shown of the separation of the radiocarbon signal into natural and bomb components. Broecker and others recently used relationships between deep C-14 and silicate in such an analysis, but Bob showed that this relationship is not linear (when a lot more data are included), as Broecker et al. had assumed. Bob, working with Stephanie Rubin, found that the relationship between potential alkalinity (which is corrected for nitrate and salinity) and deep C-14 is a lot more robust. They believe they now have more accurate estimates of the bomb C-14 distribution in the Pacific, though the technique is still being improved. Bob showed some estimates of the bomb and natural C-14 distributions in the Pacific.
Chris Sabine then discussed the WOCE/JGOFS CO2 SURVEY DATA. Some gaps in the CO2 survey data are being filled in by using data from research groups in Australia, Japan, and France.
Chris Sabine, Bob Key, and Jorge Sarmiento (with other co-PIs) recently had four related proposals funded:
Chris outlined the uses for global carbon data in model development and improvement, including estimates of natural and anthropogenic CO2, decadal scale CO2 inventory changes, and property-property relationships. Chris showed some anthropogenic CO2 distribution estimates using the Gruber et al. technique, as well as comparisons with models. Also shown were changes in the anthropogenic CO2 distribution since GEOSECS.
Chris said the Indian Ocean CO2 data should be available by the end of the year, maybe even this summer. He will make an effort to have the anthropogenic CO2 estimates available for the global ocean by the end of the year, though the Atlantic results will be from earlier (non-WOCE) work.
Niki Gruber then presented an analysis of the errors in the computation of Delta C*, which is an approximation of the concentration of anthropogenic DIC in the ocean. There are two types of errors: random and systematic.
The random errors amount to about 6 or 7 umol/kg. But with more observations, the error in the integrated amount of anthropogenic DIC in the ocean goes down.
Of more concern are systematic errors. Niki wrote down the equation for Delta C*, and pointed out three terms that likely have the highest error associated with them:
In summary, Niki estimates the error north of 50 S, which holds about 90% of the anthropogenic DIC inventory, to be about 20%. South of 50 S, the error is estimated to be about 40%.
Philippe Jean-Baptiste was next up. He gave an overview of observations of primordial Helium-3 in the ocean and its potential as a tracer of deep ocean circulation. Primordial He-3 has several advantages as a tracer for OCMIP. First, it is probably the best tracer analogue to CO2 injected at depth. Second, as Scott Doney pointed out, it is perhaps the best (if not only) tracer to give information about flow at mid-depths. Third, WOCE has greatly increased the number of He-3 measurements.
Philippe showed a zonal section in the North Pacific that revealed strong primordial He-3 sources from the East Pacific Rise. He also showed the P17 line (135W), which revealed two distinct plumes, one on each side of the equator. Primordial He-3, with its subsurface source and surface boundary condition of essentially zero concentration, is like an inverse of primordial C-14, Philippe noted.
The main difficulty in simulating primordial He-3 and using it as a tracer is the rather crude understanding of its source. Primordial He-3 is emitted mainly at plate spreading centers. Though the locations of the spreading centers are exactly known, the He-3 injection rate is not. The best one can do at current time is to assume the injection rate is proportional to the spreading rate. That is what Farley and Maier-Reimer did in a modeling paper they wrote on this subject. However, they found that the best simulations in their model assumed no source in the Atlantic, even though spreading occurs there. Some of those present suggested taking this further: inverting the He-3 data to get the sources.
There was some brief discussion about other tracers in OCMIP-2, like age, dye and decay tracers. Jim Orr suggested that, at most, due to time constraints, only an age tracer be simulated. Nothing firm was decided about additional tracers.
Niki Gruber then described some work he has been doing with his Princeton colleagues on estimating the air-sea O2 flux by inverting ocean oxygen data with an ocean GCM. A new conservative tracer was developed, O2*, which is defined as O2 + 170*PO4. The inversion is applied to O2* to obtain the air-sea O2 flux. The technique is very similar to that used for estimating surface CO2 fluxes using atmospheric CO2 data and an atmospheric GCM.
Niki broke the surface ocean up into 15 regions and ran 15 corresponding dye tracers with a surface input until steady state was reached. These tracer distributions allow the inversion to be done. It was first applied to temperature data to extract the surface heat flux. By comparing with the observed heat flux, some information about the error in the GCM circulation could be gained. For O2, Niki found that the North Atlantic and Pacific are sinks of atmospheric O2 and that the tropics are a source. The North Atlantic and tropical results are in qualitative agreement with what Najjar and Keeling found using a surface ocean climatology of oxygen. Results for the Southern hemisphere are mixed.
Patrick Monfray then presented an outline of a possible next phase of OCMIP. The objectives of OCMIP are to (1) understand anthropogenic CO2 uptake by the ocean, past and future and (2) estimate the impact of climate change on marine productivity and the marine carbon cycle.
The first of these goals is addressed by OCMIP 1 and 2. Patrick suggested that OCMIP-3 attack the second goal. The way to go about this, he suggested, is to leverage the information about climate impacts found in the signal of interannual variability. Thus, a potential focus for OCMIP-3 is a study of the marine carbon cycle response to interannual variability as well as climate change. An effort is currently underway in Europe to obtain funding for a project called Future Ocean Carbon-cycle under Climate-change (FOCCI), which would address some of the above goals.
A "Joint JGOFS-GAIM Task Team" (JTT) for OCMIP is being formed. Current members include Patrick Monfray, Ray Najjar, Jim Orr and Jorge Sarmiento. Patrick suggested that for OCMIP-3 the JTT should have a greater representation from people with more knowledge about observations and biological processes. Since the time of the workshop, Ian Totterdell (JGOFS-UK, biological processes), Chris Sabine (JGOFS & WOCE database, USA) and Matthew England (WCRP/CLIVAR, AUS) have been added as members of the JTT.
Corinne Le Quere then gave a talk about her thesis work at IPSL focused on simulating interannual variability in the air-sea CO2 flux (with the OPA8 dynamical model and the HAMOCC-3 carbon cycle model). After a climatological spinup, the model was run from 1979 to 1997 with meteorological forcing from ECMWF, NCEP and ERS.
Corinne found that most of the interannual variability of the model comes from the Equatorial Pacific, on the order of +/- 0.4 Gt C/yr. The maxima in variability in the Equatorial Pacific were found in two regions, around 180W and 100W, in agreement with limited observations there. The model also seemed to do a reasonable job of simulating seasonal and interannual variability at the BATS site. This surprised some because non-Redfield processes are thought to occur at the BATS site, yet the model has Redfield stoichiometry. Comparisons were made with Seawifs-based export production estimates. The Antarctic Circumpolar Wave in the model was analyzed, though no data were available to evaluate the model in this respect.
Corinne found that her model tended to agree with the smaller estimates of interannual variability in the air-sea CO2 flux, which are mainly ocean-based. Estimates based on atmospheric data give much larger interannual variability.
Laurent Bopp then gave a presentation of a modeling study about the impact of future climate change on the marine carbon cycle. He also used HAMOCC-3 and OPA8. These were coupled to an atmospheric model forced with a 1%/year increase in atmospheric CO2, as well as with constant atmospheric CO2 (the control case). An NPZD model was also used in place of HAMOCC-3.
By the time CO2 had doubled, the mean SST increased by 2 deg. The strongest warming was found in equatorial regions. The mixed layer shoaled at high latitudes. Both biogeochemical models gave broadly similar results. Globally, export production decreased by about 6%. Export production increased in high latitudes and decreased in low latitudes. Changes were on the order of 10 to 20%. In the HAMOCC-3 model, the changes in export were largely due to changes in stratification and nutrients; the direct temperature effect was small. In the Equatorial Pacific, upwelling decreases, surface nutrients decrease, and export production decreases from 185 to 165 g C/m2/yr. Laurent found that the CO2 uptake by the ocean decreases significantly as a result of climate change.
Jorge Sarmiento discussed some ideas about the types of analysis that might be done within OCMIP. He began with his version of OCMIP goals: (1) to determine the magnitude of the ocean sink of anthropogenic CO2, (2) to determine the spatial and temporal distribution of the air-sea CO2 flux, and (3) to improve ocean biogeochemical models for future prediction. He stated that ocean tracers can be used to help achieve these goals.
For example, the relationship between bomb C-14 and anthropogenic CO2 could help pin down the first goal above. Jorge expanded on this example by presenting some box model (HILDA) work with Fortunat Joos showing that the bomb C-14 inventory is mainly affected by gas exchange and the anthropogenic CO2 inventory is mainly affected by vertical diffusion. This means that getting the bomb C-14 distribution right in a model doesn't necessarily guarantee a good simulation of anthropogenic CO2.
What was found, however, was that the penetration depths of bomb C-14 and anthropogenic CO2 are similarly affected by gas exchange and vertical diffusion. In other words, the ratio of the penetration depths of the two tracers is fairly constant with changes in transfer velocity and diffusivity. The stability in this relationship was exploited in order to compute the inventory of anthropogenic CO2 in the ocean. The procedure is as follows: (1) The observed bomb C-14 penetration depth is computed. (2) The model ratio of the penetration depths of bomb C-14 to anthropogenic CO2 is computed for various regions in the model. (3) An estimate of the actual anthropogenic CO2 penetration depth is made by combining 1 and 2 above. (4) Using the model surface anthropogenic CO2 (which is close to equilibrium with the atmosphere) and 3 above gives an estimate of the anthropogenic CO2 inventory, region by region. The total inventory estimated this way is 124 Gt C, whereas the GCM of Murnane (the GFDL model) gives 95 Gt C.
The main point of Jorge's presentation was that we need to creatively combine model output and data to get at global and regional estimates of anthropogenic CO2 in the ocean.
Jorge then showed scatterplots of the bomb C-14 inventory as a function of the anthropogenic CO2 inventory for the four OCMIP-1 models. The differences were very large among the models. In particular, the IPSL and Hadley models showed much tighter relationships than the GFDL and MPI models. He suspects the differences are due to the differences in mixing among the models.
Jorge then discussed some recent work on the so-called Harvardton-Bear Index, which is a model diagnostic that reflects the relative importance of low latitudes with respect to high latitudes in determining atmospheric CO2. Summarizing work of David Archer, it was shown that 3-D models give a significantly higher value than box models. The bottom line is that in box models, cold waters (~5 deg C) exert the most influence on the atmosphere and in GCMs warm waters (~12 deg C) exert the most influence. Another measure is the effect of depleting the high latitudes of surface nutrients; this has a big effect on atmospheric CO2 (~100 ppm) in box models and only a modest effect in the MPI model. Archer claims the difference is due to mixing, which is high in the GCMs, giving greater importance to low latitudes.
Chris Sabine then extended the discussion of tracer-tracer distributions. He showed that the anthropogenic tracers (CFCs, bomb C-14 and anthropogenic CO2), although grossly similar, have significant differences. He then showed some scatterplots of anthropogenic CO2 inventory versus bomb C-14 inventory in the Indian and Pacific Oceans. The NCAR and GFDL models were both quite different from the "observations," with the GFDL model showing much more scatter than the NCAR model.
Bob Key then showed similar plots, but this time for C-14 versus tritium, and only for observations. He noted that plots like these can help distinguish between northern and southern component waters, because of the asymmetry in the tritium input.
There was some discussion of the next OCMIP-2 meeting. It will occur in about one year, in a location to be determined. There was also discussion about holding an OCMIP special session at an international meeting, but no firm plans were made.