Ocean Carbon-Cycle Model Intercomparison Project (OCMIP):
Phase I (1995-1997)

By James C. Orr

Annual mean air-sea flux of anthropogenic CO2 in 1990

Click here for the PDF (Adobe Acrobat) version of this report

Edited by Dork Sahagian
Layout by Karin Tierney


IPSL (Institute Pierre Simon Laplace):

J. C. Orr, P. Monfray, and O. Aumont
LSCE/CEA Saclay, Unité Mixte CEA-CNRS, Bat. 709 L'Orme, F-91191 Gif-sur-Yvette, Cedex, FRANCE

MPI (Max Planck Institut fur Meteorologie):

E. Maier-Reimer and U. Mikolajewicz
Max Planck Institute, Bundesstrasse 55, D-2000 Hamburg 13, GERMANY


N. K. Taylor and J. Palmer
Hadley Center, UK Met. Office, London Rd., Bracknell, Berkshire RG12 2SY, ENGLAND

Princeton University/GFDL (Geophysical Fluid Dynamics Laboratory):

J. L. Sarmiento, R. Murnane, C. Le Quéré, T. Hughes, N. Gruber
AOS Program, Princeton University, P.O. Box CN710, Princeton, NJ 08544-0710 USA

R. M. Key, C. L. Sabine
Dept. Of Geosciences, Princeton University, Guyot Hall, Princeton, NJ 08544-1003 USA

J. R. Toggweiler, B. Samuels
GFDL/NOAA, P.O. Box 308, Princeton, NJ 08542, USA


The Ocean Carbon-cycle Model Intercomparison Project (OCMIP) was made possible by financial support from:

European Union (EU) Environment and Climate Programme (Contract ENV4-CT95-0132)
US National Science Foundation (NSF)
US National Aeronautical and Space Administration (NASA)

Operating funds for the GAIM office have been provided by:

US National Science Foundation (NSF)
US National Oceanographic and Atmospheric Administration (NOAA)
US Department of Energy (DOE)
US Environmental Protection Agency (EPA)

1. Introduction

Ocean carbon-cycle models offer a means to help synthesize our understanding about the redistribution of carbon in the ocean and the resulting effect on the global carbon cycle. The need to improve these models and to ultimately incorporate them into a composite description of the earth's complex biogeochemical and physical climate system motivated members of the IGBP/GAIM Task Force to launch a project to compare ocean carbon-cycle models in 1994. As a result, four groups began collaborating in January 1995 within the framework of the Ocean Carbon-Cycle Model Intercomparison Project (OCMIP). Participants included groups from (1) Hamburg (Max Planck Institute of Meteorology), (2) Princeton (Geophysical Fluid Dynamics Laboratory and Princeton University), (3) Bracknell (Hadley Center, UK Met. Office), and (4) Paris (Institute Pierre Simon Laplace). Prior to OCMIP, all four groups had already used global-scale, 3-D models to study the ocean carbon cycle, but without coordination or the common protocols necessary for comparison of model results.

Modelers also saw the need for OCMIP as being fundamental for two reasons: (1) the basic need to compare results between models, and (2) the awareness that models improve more rapidly when resources are pooled and understanding is shared between modelling groups. For consistency, standard boundary conditions and protocols were developed for OCMIP and analysis was centralized (at IPSL/LSCE). This report presents the results from the first phase of this project (OCMIP-1), from 1995 to 1997. The second phase (OCMIP-2), from 1998 to 2000, will further explore the causes for principal model differences, some of which were revealed during OCMIP-1.

As the name implies, OCMIP's main interest is the carbon cycle. Hence the primary concern has been to focus on the abilities of models to predict ocean carbon distributions and air-sea fluxes of CO2. The OCMIP-1 strategy was to study (1) natural CO2, with simulations which were allowed to reach equilibrium with pre-industrial atmospheric CO2 (at 278 ppm), and (2) anthropogenic CO2, with simulations forced by observed atmospheric CO2 from pre-industrial time to present.

In addition, to evaluate model behavior, OCMIP-1 compared simulated vs. observed 14C. A global network of 14C samples was taken during GEOSECS in the 1970's and more recent sections from WOCE are now beginning to become available. During OCMIP-1, we exploited existing GEOSECS 14C measurements, taking advantage of recently developed techniques to distinguish between bomb and natural 14C components. Natural 14C offers a powerful test of an ocean model's deep ocean circulation; bomb 14C helps constrain the modeled circulation of surface and intermediate waters. Bomb 14C also appears to exhibit similar behavior to anthropogenic CO2, under certain conditions. Exploiting the 14C- CO2 relationship, when appropriate, would offer one way to circumvent the difficulty of directly measuring the small anthropogenic change in dissolved inorganic carbon (DIC) in the ocean, relative to the large DIC pool which is naturally present. Thus in OCMIP-1, we explored model relationships between bomb 14C and anthropogenic CO2 to lend insight into indirect 14C evaluation of model-predicted uptake of anthropogenic CO2. Moreover, we directly evaluated model-predicted anthropogenic CO2 uptake during OCMIP-1 by comparing model results to new data-based estimates of anthropogenic CO2 [Gruber, 1998; Gruber et al., 1996; Sabine and Key, in press].

2. Models and Methods

The four OCMIP-1 models are diverse. Details about each can be found in the original references for the models from Princeton [Sarmiento et al., 1995; Stephens et al., 1998], MPI [Maier-Reimer, 1993], Hadley [Taylor, 1995], and IPSL [Aumont et al., 1998, Aumont et al., in press; Madec and Imbard, 1996; Orr, 1996]. Table 1 provides a condensed list of some of the principal differences between the OCMIP-1 models. The models all use 3-D grids, using from tens of thousands to nearly a million grid cells to discretize the global ocean. Horizontal grid size varies most near the equator, by up to a factor of 10 between models. Vertically, models use from 2 to 10 layers to discretize the surface 100 m, and from 12 to 30 layers for the entire water column. Many other factors differ between models as well. Differences include coordinate systems, numerics, advection schemes, and specified air-sea fluxes of heat and water. The models also differ in their descriptions of turbulence, isopycnal mixing, turbulent mixing in the surface boundary layer, as well as their basic equations describing ocean circulation.

Table 1: Principal differences between models participating in OCMIP-1. Table 1

Model predictions are known to be sensitive to some of these differences between models. For example, changing the vertical turbulence specified a priori in the GFDL model has dramatic results on ocean circulation and the distribution of natural 14C [Toggweiler et al., 1989a]. Horizontal turbulence as well as parameterizations of isopycnal mixing and sub-grid scale eddies are crucial to modelling passive tracers [England, 1995]. The choice of advection scheme, always a tradeoff between numerical precision and cost, is important even with passive tracers such as anthropogenic CO2 and bomb 14C [Orr, 1996]. Other seemingly small technical details concerning model grids and numerics may likewise cause important differences [Yin and Fung, 1991].

OCMIP-1 has focused on identifying the major differences between predictions from four different ocean carbon-cycle models. But model comparison alone cannot identify, with certainty, the causes of model differences. As shown by Table 1, the models have just too many differences. Still, by identifying problems, model comparison can incite sensitivity tests with individual models to explore their causes.

3. Carbon Simulations

3.1 Natural CO2

For the natural carbon cycle in OCMIP-1, we separately ran simulations to distinguish effects due to two fundamental processes, which along with ocean circulation control the distribution of natural CO2. The first process relates to the temperature-dependent solubility of CO2. The cold waters which fill the deep ocean from the high latitudes are rich in CO2. Secondly, ocean biota act to reduce surface ocean CO2 through the combined action of planktonic uptake, rapid transport to depth of resulting particulate organic carbon, and subsequent bacterial degradation. We denote these two processes as the solubility and biological pumps, respectively [Volk and Hoffert, 1985].

Although temperature is a major control on the solubility of CO2, heat fluxes between the atmosphere and ocean are not perfectly analogous to those for solubility driven fluxes of CO2. While the surface ocean-atmosphere chemical equilibration rate for CO2 is finite (roughly 1 year,) [Broecker and Peng, 1974], the thermal equilibration rate is effectively instantaneous. To illustrate the difference between heat and CO2 (i.e., the effect of gas exchange), OCMIP included an additional simulation where air-sea CO2 fluxes were forced to be infinitely rapid.

We distinguished four effects by making three equilibrium simulations:

  1. the potential solubility pump-which includes carbon chemistry, but air-sea carbon fluxes are driven entirely by thermal fluxes via forced (non-realistic) instantaneous equilibrium between pre-industrial air (278 ppmv) and surface waters;
  2. the solubility pump which is identical to the potential pump except that it incorporates a finite (realistic) air-sea resistance to CO2 gas exchange; and
  3. the combined pump (solubility plus biological pumps, together). We determined the effects due to the solubility pump, gas exchange {(2)-(1)}, the biological pump {(3)-(2)}, as well as the total (combined pump).

Resulting sea-to-air fluxes from the three simulations as well as the biological pump are displayed as zonal integrals for the global ocean (Fig. 1). The main patterns of sources and sinks of natural CO2 are driven by thermodynamics (heat fluxes) and to a lesser extent, changes in salinity. Thus all solubility simulations exhibit ocean out-gassing in the tropics and uptake in the high latitudes. However, potential fluxes are strongly attenuated by the finite CO2 gas exchange coefficient. In reality, surface ocean CO2 requires about a year to adjust to a CO2 change in the atmosphere [Broecker and Peng, 1974], so potential and solubility simulations show major differences where surface waters are replenished rapidly in zones of upwelling and convection. Additionally, the biological pump counteracts the solubility pump either by bringing respired CO2 (produced by bacterial degradation of organic matter) to the surface via upwelling and deep convection (mostly in the high latitudes), or by consuming CO2 at the surface via photosynthesis (mostly in the subtropical gyres and the tropics).

Fig. 1: The zonally integrated air-sea flux of natural CO2 for the global ocean as predicted by 3 ocean carbon-cycle models (MPI, dotted red line; IPSL, solid blue line; and Princeton, dashed green line). Components of the flux are identified as (a) the potential solubility pump (solubility effects + instantaneous air-sea exchange), (b) the solubility pump (with realistic finite gas exchange), (c) the biological pump, and (d) the combined natural effects (solubility and biological pumps). The biological pump is determined by difference between the combined pump run and the solubility pump run.

It is difficult to evaluate model predictions for the natural carbon cycle because available data are scattered and contaminated by anthropogenic CO2. However, OCMIP-1 separately added the natural component (combined pump) to the anthropogenic component (section 3.2) in order to compare that total to the present-day measurements. The results for 1990 from the OCMIP models exhibit roughly similar patterns to those observed in the North-Atlantic [Takahashi et al., 1995]. As shown in section 3.2, anthropogenic fluxes in 1990 rival the magnitude of natural fluxes, especially in the Austral ocean where the CO2 sink has increased most since the onset of the industrial revolution.

With results from these same natural CO2 simulations, OCMIP investigated the north-south transport of carbon in the pre-industrial ocean (Fig. 2). All models suggest that there was not a large pre-industrial inter-hemispheric transport of oceanic carbon. In contrast, based on an atmospheric model and atmospheric measurements of CO2, Keeling et al. [1989a] extrapolate changes in the N-S difference between surface measurements at Mauna Loa and South Pole to determine a pre-industrial difference of -0.82 ppm. They estimated that this pre-industrial difference implies a interhemispheric northward transport in the pre-industrial atmosphere of roughly 1 Pg C yr-1. Keeling et al. [1989a] further reasoned that during pre-industrial time, the global ocean should have transported an equivalent flux southward. All OCMIP simulations reveal a global-ocean carbon loop, transporting 0.6 Pg C yr-1 from Bering Strait, across the Arctic Ocean, southwards through the Atlantic, across the Southern Ocean, and northwards again through the Indian and Pacific Oceans, back to Bering Strait. However, this loop is purely oceanic. It does not affect the inter-hemispheric budget for the atmosphere. Interhemispheric transport by the different ocean models is at most only 0.1 Pg C yr-1, southward.

All four experiments failed to produce the previously suggested, large disequilibrium between the North and the South [Keeling et al., 1989a; Keeling et al., 1989b]. That is, northward carbon transport in the Indian and Pacific Oceans essentially compensates southward transport in the Atlantic in all OCMIP-1 model runs. Even in the extreme case, if there were instantaneous equilibrium between air and sea (potential pump), which would enhance southward carbon transport across the equator by 50%, only a small amount of carbon in that run is exported to the other oceans (i.e., carried across 35oS). Most of the CO2 transported across the equator in the potential pump simulation is later lost to the atmosphere in the tropics, so the impact on the Mauna Loa - South Pole difference in atmospheric CO2 is minor. And the effect of gas exchange is to reduce interhemispheric transport of carbon to essentially a negligible amount (0.1 pg c yr-1 southward). More details about the OCMIP-1 results for the natural carbon cycle are provided in Sarmiento et al. [1998]

Fig. 2: Northward tracer transport of natural CO2, integrated for the global ocean. Subfigures and line patterns are as described in Fig. 1.

In contrast with the OCMIP-1 model results, studies which exploit ocean measurements in the Atlantic Ocean. Broecker and Peng [1992] and Keeling and Peng [1995] suggest that pre-industrial interhemispheric ocean transport was substantially higher, from 0.3 to 0.5 Pg C yr-1 southward. A recent study with the IPSL ocean model appears to resolve the discrepancy between the ocean measurement and modelling communities [Aumont et al., in press]. That study finds that the modeled interhemispheric transport increases from 0.1 to 0.25-0.45 Pg C yr-1, after including effects due to continental erosion, transport of carbon by rivers, and subsequent transport of riverine carbon within the ocean. Thus by taking the role of rivers into account, ocean model estimates now seem to agree with ocean data-based estimates of interhemispheric transport.

On the other hand, there remains a discrepancy between the Aumont et al. [in press] results and the Keeling [1989] estimate of the pre-industrial difference in atmospheric CO2 at the surface for Mauna Loa minus the South Pole. The latter study estimated the pre-industrial difference to be -0.82 ppm based on extrapolation of the trend in the difference of atmospheric CO2 between both stations vs. fossil emissions. To compare results, Aumont et al. [in press] use their CO2 fluxes as boundary conditions in a 3-D atmospheric model (TM2 from Heimann, [1995]). They find that the pre-industrial ocean can explain at most -0.3 ppm of the -0.82 ppm estimate from Keeling et al. [1989a]. If the ocean model predictions and the Keeling et al. interhemispheric difference in atmospheric pCO2 are both correct, then the remaining -0.5 ppm must be due to the rectification effect (i.e., the covariance between seasonal changes in atmospheric transport and the seasonal variability of terrestrial NPP and soil respiration [Keeling et al., 1989a].

3.2 Anthropogenic CO2

Since the beginning of the industrial revolution, the rise in atmospheric CO2 has caused the air-to-sea CO2 flux to increase everywhere. This drift from the natural system, termed the anthropogenic CO2 perturbation, is difficult to measure directly in the ocean. On the other hand, the four OCMIP-1 models all provided estimates of oceanic uptake of anthropogenic CO2.

The range of estimates of global uptake of anthropogenic CO2 during the 1980's from the four OCMIP-1 models is 1.5 to 2.2 Pg C yr-1 (Table 2). This range falls well within the spread of flux estimates from previous compilations of 1-, 2-, and 3-D ocean model results: 2.0+/-0.5 Pg C yr-1 [Orr, 1993], 2.0+/-0.6 Pg C yr-1 [Siegenthaler and Sarmiento, 1993], and 2.0+/-0.8 Pg C yr-1 [Schimel et al., 1995]. The three latter studies give wider ranges mainly because they include additional uncertainties due to our imperfect understanding of the global distribution of bomb 14C, which is used to calibrate ocean models.

Table 2: Mean uptake of anthropogenic CO2 by the ocean during 1980-1989
Model Ocean Uptake [Pg C yr-1]
Princeton/GFDL 2.2
Hadley 2.1
IPSL 1.5
MPI 1.6

The general patterns of regional uptake are grossly similar between models (Fig. 3). Ocean uptake is highest in the high latitudes and at the equator, i.e., in zones where deep waters uncontaminated with anthropogenic CO2 communicate readily with the surface via upwelling and convection. Low fluxes are evident in the subtropics, where surface waters have had longer to equilibrate with the atmosphere.

Yet, models disagree substantially about local patterns of anthropogenic CO2 uptake. The largest differences between models are found in the vast Southern Ocean (south of 30oS), which occupies about one third the surface area of the entire ocean. The model-derived position of maximum CO2 uptake varies by roughly 15 degrees. Estimates of the quantity of anthropogenic CO2 absorbed by the ocean differ by nearly a factor of two between models. The four models all absorb from one third to one half of their uptake south of 30oS. Predicted uptake is tied to the modeled intensity of mixing between the surface and deep ocean, the limiting factor on oceanic absorption of anthropogenic CO2. Most of the difference in global uptake between models can be explained by discrepancies south of 30oS. In the 30oS-30oN equatorial band (53% of the ocean's surface) all four models absorb 0.8 Pg C yr-1. North of 30oN, differences between models are large, but the small ocean area causes relatively little difference in the total absorbed by each model. The Southern Ocean is clearly the dominant sink where models differ most.

Fig. 3: Zonally integrated, annual mean, air-sea flux of anthropogenic CO2 in 1990. Note that the sign convention for the flux is opposite to that in Figure 1.

Fig. 4: Maps of the annual mean air-sea flux of anthropogenic CO2 in 1990 [mol m-2 yr-1].

Fig. 5: Maps of the vertical integral (inventory) of anthropogenic CO2 in 1990 in [mol m-2].

Differences are even more striking in maps of the air-sea flux of anthropogenic CO2 (Fig. 4). Although the zonal means in the Princeton and Hadley models are essentially identical in the Southern Ocean, maps from the two models do not resemble one another. Three of the models, including Hadley, show rather homogenous distributions of the air-sea flux in the Southern Ocean. On the other hand, the Princeton model uptake is quite patchy, centered on large convective cells that persist year round in that model. All the models exhibit very different uptake patterns, both in the Southern Ocean and elsewhere. Due to ocean circulation and mixing, the inventory (vertical integral of concentration) of anthropogenic CO2 exhibits a much smoother distribution than does the flux field of CO2 (Fig. 5). The models agree that most of the anthropogenic CO2 is stored in the subtropical gyres, particularly in the southern hemisphere. However, predicted inventories of anthropogenic CO2 still differ a great deal, both concerning the magnitude and position of anthropogenic CO2 storage.

In addition to present day estimates, one can also compare model results for future scenarios, such as IPCC stabilization case S450 (Fig. 6a). Models agree to within ±20% for the historical period, but diverge subsequently, following the sharp peak in the growth rate of atmospheric CO2 in the 1990's (Fig 6b). Numerous reasons might cause such divergence, but sensitivity tests made offline in the IPSL model suggest that at least some of the divergence may be explained by the choice of the advection scheme.

Advection dominates CO2 transport throughout much of the ocean. Three different advection schemes were used by the OCMIP-1 models. That may explain in part why resulting CO2 uptake estimates differ. As a sensitivity study, we ran the IPSL model with the three OCMIP-1 advection. All three schemes yield the same global air-sea CO2 flux (to within about 20%) throughout the historical record (Fig. 6c). However, Upstream minus MPDATA differences reach 60% after the peak growth rate (Fig 6d); CTCS minus MPDATA differences are more than an order of magnitude smaller, but also follow the change in the growth rate of atmospheric CO2.

Temporal change is not spatially uniform for all three schemes. In 1995, use of the Upstream scheme results in large differences, mostly in the vast Southern Ocean and tropics, where CO2-impoverished waters from the deep ocean contact the surface. By 2165, differences have grown nearly everywhere, but mostly in the same two regions. Future uptake with the Upstream scheme is about twice as much in the tropics and is 40% more in the Southern Ocean, relative to the reference scheme MPDATA.

Unfortunately, discrepancies found here cannot be applied directly to explain disparities between the different OCMIP models. Many other factors come into play. For instance, the IPSL model is quite different from the LSG model, which uses the Upstream advection scheme. Differences in the grid resolution, time step, advection fields, and explicit (IPSL) vs. implicit (MPI) numerical schemes mean that numerical diffusion will differ. Furthermore, for these tests only the advection scheme was altered. If other explicit and prognostic diffusion were eliminated, the Upstream approach with only numerical diffusion would produce less uptake. Moreover, if simulations had been run in an online model (instead of the IPSL offline model), active tracers, temperature and salinity, would also have been affected. That would in turn affect the modeled circulation, which would feedback on the passive tracers.

In addition to model comparison, we directly evaluated OCMIP-1 model results for anthropogenic CO2 on the basis of newly available data-based estimates of anthropogenic CO2 in the Atlantic [Gruber, 1998; Gruber et al., 1996] and Indian Oceans [Sabine and Key, in press, Sabine et al., in press] In the Atlantic Ocean, two north-south sections (Figs. 7 and 8) and one section along roughly 60oS (Fig. 9) indicate that most of the four OCMIP models over-predict storage of anthropogenic CO2 in the deep Southern Ocean (south of 50oS). At 60oS, penetration of anthropogenic CO2 appears to exhibit substantial longitudinal variability which is not captured by the models (Fig. 9, west of the prime meridian). In the Indian Ocean sector of the Southern Ocean, most of the models appear to over-predict storage of anthropogenic CO2 (Fig. 10). Overestimates of storage in the Southern Ocean is particularly important because that area is also where all models predict that much of the anthropogenic CO2 is absorbed.

Fig. 6: (a) Atmospheric CO2 from 1765 to 1990 followed by IPCC scenario S450 (solid) and its growth rate (dots), (b) History of the flux of anthropogenic CO2 from atmosphere to ocean as predicted by the four OCMIP models, all forced by observed atmospheric CO2 from 1765 to 1995, followed by IPCC future scenario S450. Before 1990, models fall into 2 categories. Models with explicit mixed layer dynamics (Hadley and IPSL models) respond more rapidly to the sharp maximum in the S450 scenario's rate of change of atmospheric CO2 with time, which occurs in the 1990's; air-sea fluxes lag this peak due to the slow response time of the ocean. A larger effect is the later divergence between MPI and IPSL models. Sensitivity tests in the IPSL model show that large differences in carbon uptake are caused by changing the advection scheme as shown in (c) for the IPSL model with the MPDATA (solid), CTCS (dots), and Upstream (dashes) advection schemes, and (d) the relative differences using MPDATA as the reference. The advection scheme used in each of the OCMIP models is indicated by consistency in line signatures between subplots.

There are two important questions which must be addressed regarding the data-based methodology: (1) what are the uncertainties? and (2) what is the detection limit? It appears that the random and systematic errors result in error bars of about ±10 µmol kg-1 [N. Gruber and C. Sabine, pers. comm.]; however, there is substantial spatial variability in the error, which is largely a function of Apparent Oxygen Utilization (AOU). The question of the detection limit is more difficult to answer, but preliminary efforts suggest that it may be about 5 µmol kg-1. Such a high detection limit makes it more difficult to distinguish if models are performing well in regards to the distribution of deep-ocean anthropogenic CO2. Yet differences between models are so large that the data-based estimates should be used as a reference, unless their associated uncertainties are substantially larger than estimated above. For instance, in the Southern Ocean between 50oS and 60oS, models bracket the data-based estimates for the top-to-bottom inventory of anthropogenic CO2. Comparison of model and data-based estimates suggests that three OCMIP-1 models over-predict storage of anthropogenic CO2 in the Southern Ocean. The fourth model (IPSL) appears to underpredict storage of anthropogenic CO2 in the Southern Ocean by about 20%. However, predictions from this version of the IPSL model should be considered a lower bound because it is known to underpredict uptake of 14C and CFC's in the same region.

Fig. 7: Data-based and model estimates of anthropogenic CO2 [µmol kg-1] in the Western Atlantic along a transect constructed by combining stations from the Transient Tracers in the Ocean (TTO) North Atlantic Study (NAS) in 1981-1982), the TTO Tropical Atlantic Study (TAS) in 1982-1983, and the South Atlantic Ventilation Experiment (SAVE), in 1989. Data-based estimates are from [Gruber, 1998] Model estimates were constructed by sampling, the annual mean OCMIP-1 model distributions at the same station positions, in 1982 for TTO (North Atlantic) and in 1989 for SAVE (South Atlantic).

Fig. 8: Data-based and model estimates of anthropogenic CO2 [µmol kg-1] in the Eastern Atlantic along a transect constructed by combining stations from TTO (in 1981-1982), SAVE (in 1989), and the Meteor 11/5 cruise (in 1991). Data-based estimates are from [Gruber, 1998] Model estimates were constructed as described in the previous figure.

Fig. 9: Data-based and model estimates of anthropogenic CO2 [µmol kg-1] in the Southern Ocean along a transect from Meteor 11/5 (1991). Data-based estimates are from [Gruber, 1998] The annual mean model distributions were sampled at the Meteor 11/5 locations in 1991.

Fig. 10: Data-based and model estimates of anthropogenic CO2 [µmol kg-1] in the East Indian Ocean along a transect constructed by combining stations from the World Ocean Circulation Experiment (WOCE) sections I8S and I9N, in 1995 . Data-based estimates are from [Sabine and Key, in press], who used the [Gruber et al., 1996] method modified for the Indian Ocean. Model estimates were constructed by sampling, the annual mean OCMIP-1 model distributions at the same station positions, in 1995.

4. Tracer Validation

In OCMIP-1, participating models were evaluated by making standard simulations for 14C and comparing those results to available measurements. Deep-ocean circulation patterns were evaluated with natural 14C, whose input function is effectively in steady-state relative to the circulation time of the deep ocean (roughly 1000 yr). Such a timescale is clearly important for the natural carbon cycle which is also in steady state. Furthermore, having a realistic deep-ocean circulation becomes more important for future anthropogenic CO2 simulations as that tracer penetrates further into deeper waters. Near-surface ocean circulation has been validated with bomb 14C, a transient tracer, whose atmospheric boundary condition has varied considerably in recent years.

4.1 Natural 14C

Natural 14C is useful as a clock for deep ocean circulation. Atoms of 14C are produced naturally when nitrogen is bombarded with cosmic radiation in the upper atmosphere. This 14C rapidly attaches to molecules of CO2, enters the ocean through air-sea gas exchange, and subsequently decays (half-life of 5730 yr), mostly in the deep ocean. Gradients of 14C result, both vertically and horizontally, depending upon how long waters have been isolated from the ocean surface and upon ocean circulation patterns. To validate ocean models, we have compared simulated natural 14C with the observed natural component [Broecker et al., 1995] from measurements taken during GEOSECS, a global-scale campaign during the 1970's.

Along the Western Pacific GEOSECS section (Fig. 11), all simulations reproduce the gross features of the observations. In all models, younger waters appear to invade relatively rapidly from the south and more slowly from above, thereby maintaining a minimum in the deep North Pacific. The absolute minima (the oldest water) from the four models (from -239 to -264 ä) are close to the observed minimum (-248 ä). But as shown in Table 3, all models over-predict the implied age of deep North Pacific minimum waters, as determined by the difference between their 14C content vs. the level of natural 14C in the surface Antarctic, the major source region (see Toggweiler and Samuels [1993]).

Table 3: Difference in natural delta14C (in ä) between Antarctic surface waters and the deep North Pacific minimum (along the W. Pacific GEOSECS track).
Model Deep North Pacific Surface Antarctic Difference Age (yr)
-140 to -160
-88 to -108
914 to 1109

* Stratification at the surface is particularly notable in the IPSL model, an artifact which appears mostly due to the semi-diagnostic forcing (where T and S are restored toward the observations throughout the water column over much of the ocean, not only at the surface (prognostically) as in the other 3 models). More recent simulations in the IPSL model have eliminated this problem. Note that the IPSL difference (and thus age) fall to within the observed range when instead the "surface" Antarctic C-14 level is taken at 200 m.

Fig. 11: Section of natural delta14C (in ä) along (a) the Western Pacific GEOSECS cruise track (b) as estimated from observations by [Broecker et al., 1995], and in the (c) GFDL, (d) Hadley, (e) MPIM, and (f) IPSL models.

A consistent problem in all models is that along-bottom transport of Antarctic Bottom Water (AABW) appears much too slow (Table 4). Simplistically assuming infiltration only by advection from the south, i.e., neglecting transport by other processes (namely vertical turbulent diffusion), one can calculate the rate at which AABW is advected northward. The models over-predict the implied age difference by 50 to 130%. Although such age calculations are overly simplistic and it would be better to have additional information from simulations with an ideal age tracer, natural 14C shows clearly that AABW renewal rates in all four models are much too slow.

Table 4: Along bottom natural delta14C (in ä) difference (between 45oS and 45oN) in the Pacific
Model Deep North Pacific Deep South Pacific Difference Age (yr)

As rates differ, so do the pathways by which waters invade the deep ocean, to the extent that the shape of the minimum itself is affected. For instance, at 45oN, high natural 14C values extend too deep from the surface in all models except that from Hadley. In the IPSL model, the minimum is pushed all the way to the bottom. This problem appears related to the IPSL model's semi-diagnostic forcing, since the modeled distribution is similar to that from the "robust-diagnostic" version of Toggweiler et al. [1989a].

In the Atlantic, differences between models are even more evident. Unlike the Pacific which has only one source of deep water, the Atlantic has two, making it more difficult to simulate circulation patterns correctly. To do so, a model must obtain the proper balance between filling of its deep North Atlantic with North Atlantic Deep Water (NADW) from the north, and infiltration of AABW from the south.

Fig. 12: Section of natural delta14C (in ä) along (a) the Western Atlantic GEOSECS cruise track (b) as estimated from observations by [Broecker et al., 1995], and in the (c) Princeton-GFDL, (d) Hadley, (e) MPIM, and (f) IPSL models.

Along the Western Atlantic GEOSECS section (Fig. 12), the observations reveal that the basin north of the equator is filled with >500 year-old NADW waters penetrating from the north. The youngest water penetrating furthest south is a tongue at mid-depth, centered at 2500 m. Antarctic waters have an older signature, despite their recent contact with the atmosphere, due to mixing with the ocean's oldest waters in the Pacific (see Fig. 11). Antarctic Waters penetrate northward both at intermediate depths (AAIW) and along the bottom (AABW).

In the GFDL model, younger waters from the north penetrate no deeper than 2500 m. This is a classic problem first identified by Toggweiler et al. [1989a]. The GFDL model's oldest deep waters are found at depth in the north, unlike the observed distribution. More realistically, the MPI, Hadley, and IPSL models simulate younger waters which fill the North Atlantic basin. All three of those models simulate a tongue of younger water penetrating from the north (NADW) confined by "older" water below (AABW) and above (AAIW). However the skill varies between these three models in their ability to reproduce the observed distribution. For example, the general distribution in the MPI model appears somewhat too smooth, the age minima in the tongue of NADW in the Hadley model appear too shallow, and the along-bottom gradient north of 40oN in the IPSL model has a vertical instead of a horizontal structure. In contrast to the GFDL model's excessive penetration of AABW from the south (due to lack of deep NADW), the MPI, Hadley, and IPSL models all exhibit inadequate northward penetration of AABW. Such behavior for the latter three models in the Atlantic is analogous to the behavior of all four OCMIP-1 models in the Pacific, where along-bottom, northward penetration of AABW is inadequate.

A related area of concern is the near-surface Southern Ocean. The few pre-nuclear measurements of 14C in this area suggest that surface values near Antarctica should be between -140 and -160 ä (see Toggweiler and Samuels [1993]). But no ocean model completely succeeds in reproducing such low values. Simulated natural 14C in the GFDL model appears a little high, between -120 to -140; agreement is slightly worse in the models from Hadley Centre (between -100 and -140 ä) and from MPI (around -100 ä). The IPSL model exhibits the worst agreement with surface natural 14C ranging between -40 to -80 ä. In the latter model, the problem is clearly related to inadequate communication between the near surface layers and the deep, most notably in the Southern Ocean (see footnote to Table 3).

4.2 Bomb 14C

Beginning with the atmospheric nuclear weapons tests during the 1950's, the 14C content of the atmosphere increased sharply. At its peak in 1963, atmospheric 14C was nearly double that found during pre-industrial times. Subsequently atmospheric 14C has declined, following implementation of international treaties banning atmospheric weapons tests. Today atmospheric 14C is only about 10% greater than during pre-industrial times. By carefully separating out the nuclear component through correlations with other ocean tracers [Broecker et al., 1985; Broecker et al., 1995], oceanographers have been able to use the bomb 14C signal in the ocean as a data set with which to validate ocean models.

Figure 13 shows simulated distributions of bomb 14C, along with measurements during GEOSECS, represented as the vertical integral of the concentration, which is referred to as the inventory. General patterns between models and data are similar, with lows in equatorial and high latitude regions and highs in the subtropical gyres. Despite general agreement though, local variability is substantial.

Inventory maximums observed in the western portion of subtropical gyres are displaced to the east in simulations from the GFDL and MPI models. This artifact is most noticeable in the Northern Pacific gyre but is also found to some degree in the southern hemisphere and in the Atlantic. Toggweiler et al. [1989b] attributed this problem to the "Veronis effect", an artificial upwelling between western boundary currents and the adjacent coast found in models where turbulent mixing is oriented only along horizontal and vertical surfaces; in the real ocean mixing is thought to occur along surfaces of constant density (isopycnals) which are often inclined.

Fig. 13: The vertical integral (inventory) of bomb 14C during GEOSECS (1973 for the Atlantic, 1974 for the Pacific, and 1978 for the Indian Ocean). Observed values are indicated by filled circles, colored according to the same scheme as used for the model results. Units are in 109 atoms 14C cm-2.

The IPSL model exhibits a nearly constant east-west distribution, but it still does not capture the western maximum as observed. On the other hand, in the Hadley model, the inventory maximums remain in the west as observed. We hypothesize that two enhancements to the Hadley model explain its better agreement with the observations. First, the Hadley model has included an explicit formulation for mixing along isopycnal surfaces. Thus it better captures the steeply sloping isopycnal surfaces along western margins. Hence anomalous upwelling is reduced and less model 14C is displaced to the east. Secondly, vertical inventories are higher in the west simply because the mixed layer is deeper. This shoaling of the mixed layer from west to east is evident in the North Pacific GEOSECS section for bomb 14C (Fig. 14). All the OCMIP-1 models have a difficult time to reproduce this shoaling, but the Hadley and IPSL models seem to perform slightly better. Both models have explicit formulations for mixing within the mixed layer and below.

Fig. 14: Data-based estimates of bomb delta14C (in ä) along the North Pacific GEOSECS section in 1973 as estimated by [Broecker et al., 1995] vs. the same section in the four OCMIP-1 models. Observed values are indicated by filled circles, colored according to the same scheme as used for the model results Thus model bias is indicated by color contrast between the data and the modeled distribution. Problems are particularly apparent in all models in the western basin, where penetration of bomb 14C is too shallow.

Fig. 15: Modeled inventory (vertical integral of the concentration) at each grid point between 30oS and 70oN in the Atlantic Ocean, for anthropogenic CO2 in 1990 vs. bomb 14C. That relationship is given for bomb 14C in 1975 (GEOSECS era) and in 1990 (WOCE era). Over this time period, patterns of the oceanic distribution of bomb 14C change substantially, whereas patterns of the distribution of anthropogenic CO2 change very little.

Subsequent sensitivity tests with the IPSL model have shown that improving formulations for both isopycnal mixing and the mixed layer help resolve the anomalous symptoms revealed by the bomb 14C inventory. However, these two fixes do not offer a complete solution. For example, part of the improved east-west distribution in the Hadley model is due to surface levels of 14C which are higher than observed in the west (Fig. 14). These higher surface levels of 14C offset the opposite problem (14C levels which are lower than observed) deeper down. Hence the inventory in the western basin is roughly correct, but the vertical distribution is not. Other studies [Follows and Marshall, 1996; Williams et al., 1995] suggest that eddy transport, which is not explicitly included in such coarse resolution models, plays an important role.

Bomb 14C is of particular interest because of its similarities to anthropogenic CO2, which is 99% 12C. Both species are isotopes of carbon. Both are transferred from the atmosphere to ocean by air-sea gas exchange. Concentrations of both isotopes in the atmosphere have changed substantially since pre-industrial times. But the form of the atmospheric records (concentration vs. time) of these two tracers are quite different. This is the main reason why their distributions differ in the ocean. However, comparison of these two tracers in the four OCMIP models for grid boxes in the Atlantic Ocean between 30oS and 70oN (Fig. 15) reveal that differences are large for oceanic bomb 14C at times close to the 1963 peak in atmospheric 14C. That is, 14C samples collected during the global GEOSECS campaign during the 1970's are distributed much differently than is anthropogenic CO2. During 1990, the model relationships between the two tracers are much improved (Fig. 15). That improved relationship suggests that more recent bomb 14C samples, such as those now being taken as part of the World Ocean Circulation Experiment (WOCE), will better serve to help indirectly estimate the anthropogenic CO2 component in the ocean. In other regions of the ocean correlations are different but the same general improvement of the correlation is generally observed. The exception is the Southern Ocean, where the predicted time-dependent correlation varies between models. For example, even for WOCE-era 14C (1990), the GFDL model finds essentially no correlation in the Southern Ocean; conversely, the IPSL model exhibits a tight, nearly linear correlation.

The classical albeit indirect means of evaluating global ocean carbon-cycle model performance is through comparison of simulated vs. measured radiocarbon. As discussed previously, OCMIP has studied both natural and bomb 14C. The uncertainties that are associated with observational-based methods used to distinguish the bomb 14C component from the natural 14C background are particularly large in the Southern Ocean. To reduce uncertainties, one can compare models to the measured WOCE (1990's) minus GEOSECS (1970's) 14C difference [Key, 1997]. Key [1997] compared the observed WOCE - GEOSECS delta14C increment along a zonally compressed section in the Pacific Ocean to analogous results from the Princeton-GFDL model. Similar comparisons are planned for models participating in OCMIP-2

Fig. 16: The historical increase in the number of 3-D ocean models used to study the global carbon cycle, as indicated by the number of models represented at the 2nd, 3rd, and 4th International CO2 conferences (filled circles, held in Kandersteg in 1985, Hinterzarten in 1989, and Carqueiranne in 1991) and those participating in phase 1 and 2 of OCMIP. (crosses).

5. Summary and Conclusions

Ocean carbon-cycle modelling groups have begun to take advantage of their recent increase in numbers (Fig. 16). They have begun working closely together, pooling resources and developing standards, to make model validation and comparison easier, more rapid, and more thorough. The initial four modelling groups established in 1995 began to make standard simulations for natural and anthropogenic components of CO2 and 14C. Results have been compared both to available data sets (model validation) and between models. That first phase (OCMIP-1) lasted three years (1995-1997) and the principal results are reported here.

For CO2, pre-industrial or "natural" simulations of CO2 revealed relatively large differences in predicted patterns of air-sea gas exchange, particularly in the Southern Ocean. Yet, all models agreed that global interhemispheric transport of oceanic carbon was small, with southward transport in the Atlantic being compensated by northward transport in the Pacific and Indian Oceans. For anthropogenic CO2, the 4 model estimates for global ocean uptake fell within 1.85 ± 0.25 Pg C yr-1 (±19%), for the 1980-1989 average. All models predicted that 1/3 to 1/2 of their uptake occurred south of 30¡S, due to the Southern Ocean's extensive surface area and its active exchange between surface and deep waters. Yet in the Southern Ocean, the magnitude of anthropogenic CO2 uptake still differed between models by nearly a factor of two, and local patterns of uptake were very different. Global model estimates diverge with time under the IPCC future stabilization scenario, as the Southern Ocean uptake becomes relatively more dominant, and as differences in deep water circulation come into play. Sensitivity tests with the IPSL model suggest that these differences are least in part due to the use of different advection schemes. Comparison of model and data-based estimates suggests that three of the models may overpredict storage in the Southern Ocean by about factor of two, whereas the IPSL model underpredicts storage by about 20%. However, other tracer simulations indicate that the IPSL prediction should be considered a lower bound.

We compared model predictions of 14C to data-based estimates of the natural component of 14C. We found that northward penetration of AABW occurred much too slowly along the Western Pacific GEOSECS track. In the Atlantic Ocean, three of the four models are able to simulate young water penetrating from the north (NADW) which fills the deep northern basin, as observed; only the Princeton-GFDL model produced NADW that did not penetrate below 2500 m.. All models were able to simulate a northward penetration of older intermediate water (AAIW), but with varying success. Northward penetration of older AABW was excessive in the GFDL model, due to its lack of deep NADW; conversely, northward penetration of AABW appeared inadequate in the other three models. For bomb 14C, all models showed some difficulty in reproducing the observed west-to-east gradient in the inventory along the Northern Pacific GEOSECS section. However, the Hadley model reproduces the direction of the observed gradient. Although the cause of the problem is still apparent, it is less severe in the Hadley model because it includes specific formulations for mixing along isopycnal surfaces and for the mixed layer. We also explored the ability of bomb 14C to act as a proxy for anthropogenic CO2, which is not ideal because of dramatically different atmospheric boundary conditions: atmospheric levels of bomb 14C rose and fell rapidly after atmospheirc weapons testing; atmospheric levels anthropogenic CO2 have risen gradually. Relative patterns of the oceanic distribution of bomb 14C have changed substantially in recent decades, whereas those for anthropogenic CO2 have changed little. In all four models, inventories of GEOSECS-era bomb 14C show poor correlations with inventories of anthropogenic CO2 for the Atlantic Ocean between 30oS and 70oN. On the other hand, for WOCE-era 14C, all models show a much improved relationship with CO2.

Through OCMIP-1, we have demonstrated that predictions from ocean carbon-cycle models differ regionally by a substantial amount, particularly in the Southern Ocean, where modeled air-sea fluxes of anthropogenic CO2 are also largest. Within OCMIP, such differences have motivated plans for OCMIP-2. This second phase of OCMIP (1998-2000) involves 13 models and additional simulations [Orr et al., 1997]. The focus remains on CO2, but OCMIP-2 also includes emphasis on new circulation tracers, such as CFC-11 and CFC-12, and new biogeochemical tracers, such as O2. OCMIP-2 will also include simulations with a common biogeochemical model so that participants can better study effects due to differences in modeled ocean circulation. Furthermore, OCMIP-2 includes data specialists who are leading the JGOFS and WOCE synthesis for CO2, 14C, and CFC's, which will strengthen model validation efforts.

Studies during the first two phases of OCMIP have relied on ocean models run under present climatological conditions, where circulation patterns do not evolve with time. Beyond OCMIP-2, future work will probably focus on the impact of changing climate on marine biogeochemistry as well as the feedback of changes in marine biogeochemistry on climate. To validate such simulations, it will be crucial to focus on how well models are able to reproduce observed interannual variability. More information concerning this project can be found on the OCMIP Web page http://www.ipsl.jussieu.fr/OCMIP


Aumont, O., J. C. Orr, D. Jamous, O. Marti, G. Madec, and P. Monfray. A degradation approach to accelerate simulations to steady state in a 3-D tracer-transport model of the global ocean, Climate Dynamics, 14, 101-116, 1998.

Aumont, O., J. Orr, P. Monfray, W. Ludwig, and J. Probst, What role do rivers play in the interhemispheric transport of carbon? in 2nd International Symposium on CO2 in the Oceans, Tskuba, Japan, in press.

Broecker, W., T. Peng, G. Ostlund, and M. Stuiver, The distribution of bomb radiocarbon in the ocean, J. Geophys. Res., 90, 6953-6970, 1985.

Broecker, W.S., and T.H. Peng, Gas exchange rates between air and sea, Tellus, 26, 1974.

Broecker, W.S., and T.H. Peng, Interhemispheric transport of carbon dioxide by ocean circulation, Nature, 356, 587-589, 1992.

Broecker, W.S., S. Sutherland, W. Smethis, T.H. Peng, and G. Ostlund, Oceanic radiocarbon: Separation of the natural and bomb components, Global Biogeochem. Cycles, 263-288, 1995.

England, M.H., Using Chlorofluorocarbons to Assess Ocean Climate Models, Geophys Res. Lett. 22, 3051-3054, 1995.

Follows, M.J., and J.C. Marshall, On models of bomb 14C in the ocean, J. Geophys. Res., in press.

Gruber, N., Anthropogenic CO2 in the Atlantic Ocean, Global Biogeochem. Cycles, 12, 165-192, 1998.

Gruber, N., J.L. Sarmiento, and T.F. Stocker, An improved method for detecting anthropogenic CO2 in the ocean, Glob. Biogeochem. Cycles, 10, 809-837, 1996.

Heimann, M., The TM2 tracer model, model description and user manual, Deutsches Klimarechenzentrum, 10, 47, 1995.

Keeling, C., S. Piper, and M. Heimann, A three-dimensional model of atmospheric CO2 transport based on observed winds: 4. 1. Analysis of observations data,, in Aspects of climate Variability in the Pacific and Western Americas, AGU Monograph 55, 165-236, 1989a.

Keeling, C., S. Piper, and M. Heimann, A three-dimensional model of atmospheric CO2 transport based on observed winds: 4 Mean annual gradients and interannual variations, in Aspects of climate Variability in the Pacific and Western Americas, AGU Monograph 55, 305-363, 1989b.

Keeling, R., and T. Peng, Transport of heat, O2, and CO2, by the Atlantic's thermohaline circulation, Phil. Transl Roy. Soc. London B, 348, 133-142, 1995.

Key, R.M., U.S. WOCE Report: Changes in Pacific Ocean Distribution of Radiocarbon, pp. 5-8, Texas A&M University, WOCE Office, College Station, TX, 1997.

Madec, G., and M. Imbard, A global ocean mesh to overcome the North Pole singularity, Clim. Dynam., 12, 381-388, 1996.

Maier-Reimer, E., Geochemical cycles in an ocean general circulation model, Preindustrial tracer distributions, Glob. Biogeochem. Cycles, 7, 645-677, 1993.

Orr, J., Accord between ocean models predicting uptake of anthropogenic CO2, Water, Air, and Soil Pollution, 70, 465-481, 1993.

Orr, J.C., The Ocean Carbon-Cycle Intercomparison Project of IGBP/GAIM, in Ocean Storage of Carbon Dioxide, B. Ormerod, pp. 33-52, CRE Group Ltd., Cheltenham, UK, 1996.

Orr, J.C., J.L. Sarmiento, E. Maier-Reimer, U. Mikolajewicz, P. Monfray, J. R. Palmer, J. R. Toggweiler, R. M. Key, C. L. Sabine, and N. Gruber, Anthropogenic CO2 in the ocean: 3-D models vs. reality, in 5th International CO2 Conference, Cairns, Australia, 1997.

Sabine, C. L., and R. M. Key, Looking for fossil CO2 in the oceans, 1998 U.S. WOCE Report, No. 10, 26-29, U.S. WOCE Office, College Station, TX, in press, 1998.

Sabine, C. L., R. M. Key, C. Goyet, K. M. Johnson, F. J. Millero, A. Poisson, J. L. Sarmiento, D. W. R. Wallace, and C. D. Winn. Anthropogenic CO2 inventory of the Indian Ocean. Global Biogeochem. Cycles, in press. Sarmiento, J., R. Murnane, and C. Lequere, Air-Sea CO2 Transfer and the Carbon Budget of the North Atlantic, Phil. Trans. Roy. Soc. London S B., 348, 211-219, 1995.

Sarmiento, J.L., T.M.C. Hughes, R.J. Stouffer, and S. Manabe, Simulated response of the ocean carbon cycle to anthropogenic climate warming, Nature, 393, 245-249, 1998.

Schimel, D., I. Enting, M. Heimann, T. Wigley, D. Raynaud, and e. al., The Carbon Cycle, in Climate Change 1994, Radiative Forcing of Climate Change and an Evaluation of the IPCC IS92 Emission Scenarios, Press. Synd of Univ. Of Cambridge, 35-71, 1995.

Siegenthaler, U., and J. Sarmiento, Atmospheric Carbon Dioxide and the Ocean, Nature, 365, 119-125, 1993.

Stephens, B., R. Keeling, M. Heimann, K. Six, and e. al, Testing global ocean carbon cycle models using measurements of atmospheric O2 and CO2 concentration, Global Biogeochem. Cycles, 12, 213-230, 1998.

Takahashi, T., Takahashi, T. T., and S. Sutherland, An assessment of the role of the North Atlantic as a CO2 sink, Phil. Trans. R. Soc. Lond., B348, 143-152, 1995.

Taylor, N.K., Seasonal uptake of anthropogenic CO2 in an ocean general circulation model, Tellus Series B Chemical and Physical Meteorology, 47, 145-169, 1995.

Toggweiler, J., K. Dixon, and K. Bryan, Simulations of radiocarbon in a coarse-resolution world ocean model: 1, Steady-state prebomb distributions, J. Geophys. Res, 94, 8217-8242, 1989a.

Toggweiler, J.R., K. Dixon, and K. Bryan, Simulations in a course-resolution world ocean model. 1: Steady state pre-bomb distributions 2: Distributions of bomb produced carbon, J. Geophys. Res., 94, 8217-8264, 1989b.

Toggweiler, J.R., and B. Samuels, New Radiocarbon Constraints on the Upwelling of Abyssal Water to the Ocean's Surface., in The Global Carbon Cycle, NATO ASI Series edited by M. Heimann, pp. 585-586, Springer-Verlag, Berlin, 1993.

Volk, T., and M. Hoffert, Ocean carbon pumps: analysis of relative strengths and efficiencies in ocean driven atmospheric CO2 changes, AGU Monograph 32, 99-110, 1985.

Williams, R.G., M.A. Spall, and J.C. Marshall, Does Stommel's Mixed Layer Demon Work?, J. of Phys. Oceanogr., 25, 3089-3102, 1995.

Yin, F.L., and I.Y. Fung, Net diffusivity in ocean general circulation models with nonuniform grids, J. Geophys. Res., 96, 10,773-10,776, 1991.