GLOBAL BIOGEOCHEMICAL CYCLES, VOL. 15, NO. 4, PAGES 831-844, December, 2001
An examination of the ‘‘continental shelf pump’’ in an open
ocean general circulation model
Andrew Yool and Michael J. R. Fasham
Southampton Oceanography Centre, Southampton, England, UK
Abstract. In a recent study of the shelf region of the East China Sea, Tsunogai et al. [1999]
estimated that a combination of air-sea exchange and biological and physical transport
processes could transfer carbon from the shelf region into the open ocean at a rate of 35 g C m2
yr1. Contrasting with the solubility and biological pumps of the open ocean, they described this
collective activity as the ‘‘continental shelf pump’’ and suggested that if this pump operated
throughout the world’s shelf regions, it could be responsible for ocean uptake of 1 Gt C yr1
(50% current ocean uptake of anthropogenic CO2). In this work a general circulation model
(GCM) is used to explore the potential strength of this pump across the world’s shelves. Since the
GCM does not represent the continental shelf regions explicitly, a parameterization of the pump
has been used. Results of simulations find modeled pump activity very variable between shelf
regions, with the East China Sea shelf behaving very similarly to the global average. Storage of
pump carbon is particularly high in the Atlantic Ocean and other regions where deep water is
formed. A considerable reservoir of pump carbon becomes trapped under the Arctic ice sheet.
Simple extrapolations from the results suggest that should shelf regions absorb CO2 at the rate of
the East China Sea, the pump would account for a net oceanic uptake of 0.6 Gt C yr1.
1. Introduction
ocean biological and solubility pumps traditionally recognized as
routes by which carbon reaches the deep ocean. For an overview of
these latter pumps, refer to Sarmiento [1993] and Raven and
Falkowski [1999].
Using observations made in the surface waters of the East China
Sea, Tsunogai et al. [1999] calculated a mean CO2 fugacity deficit
of 55 ± 5 ppm. This calculated deficit was found to agree well with
an earlier estimate of the export flux of carbon derived in a budget
for the East China Sea [Tsunogai et al., 1997]. Studying the
distributions of total dissolved carbonate and alkalinity, Tsunogai
et al. [1999] concluded that this large deficit could be explained as
follows. Since the shallow seafloor restricts the convection of
cooling water, cooling is greater for waters on the continental shelf
than for waters in neighboring open ocean areas. This leads to the
production of relatively cold and dense water, which in combination with biological production, results in a greater absorption of
CO2 in the continental shelf zone. The absorbed CO2, both in the
form of dissolved inorganic carbon (DIC) and dissolved/particulate
organic carbon (DOC/POC), is transported in this denser water into
the subsurface layer of the open ocean via isopycnal mixing
processes. Owing to a well-developed thermocline the hydrography of the East China Sea permits this transport to occur even
during the warm season. Scaling up from the East China Sea,
Tsunogai et al. [1999] suggest that if the processes they have
observed there operated over the entire global continental shelf
zone, this pump would account for a net oceanic uptake of CO2 of
1 Gt C yr1 (83.3 Tmol C yr1). Given that observational
[Takahashi et al., 1999] and model [Orr, 1997] estimates of the
world ocean’s current (1990 – 2000) anthropogenic CO2 uptake are
2 Gt C yr1 and that these exclude or poorly represent the
continental shelf zone, Tsunogai et al. [1999] work may have
important implications for studies of the natural and contemporary
carbon cycle. For instance, should the shelf regions be net
exporters of carbon to the open ocean, observational studies that
focus only on the open ocean will overestimate ocean outgassing
by including outgassing CSP carbon but excluding CSP uptake.
In many regions of the world ocean the margin between the
continental and oceanic tectonic plates is demarcated by the socalled continental shelf zones. These zones are commonly gently
sloping regions of the continental plates which because of the
contemporary sea level, lie beneath the sea surface down to a depth
of 200 m. They reach their greatest size off the coasts of Northern
Eurasia, Eastern Asia, Oceania, and the eastern Americas. Figure 1
shows the global distribution of the Earth’s surface which lies
between 0 and 200 m below contemporary sea level.
The relative shallowness of these zones (compared to the open
ocean) and the input of nutrients to them from the terrestrial
environment (both naturally and, recently, anthropogenically),
make them considerably more biologically productive than the
open ocean. Despite the fact that they make up only a relatively
small fraction (8%) of the world ocean’s surface the continental
shelf zones contribute an estimated 19 – 28% of total global ocean
production [Longhurst et al., 1995]. Some of this production is
exported to the open ocean, and recently, a debate [Duarte and
Agusti, 1998; Williams and Bowers, 1999; Duarte et al., 1999]
has focused on whether or not the open oceans are globally net
autotrophic (i.e., fix more inorganic carbon than they regenerate)
or net heterotrophic (i.e., regenerate more organic carbon than
they fix).
In this context and on the basis of observations collected from
the East China Sea (see Figure 1), Tsunogai et al. [1999] propose a
new term, the ‘‘continental shelf pump’’ (henceforth CSP), to
collectively describe a series of processes that transport carbon
from the atmosphere via the continental shelf to the deep ocean.
They identify this pump as being a separate mechanism to the open
Copyright 2001 by the American Geophysical Union.
Paper number 2000GB001359.
0886-6236/01/2000GB001359$12.00
1
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
2
90
Latitude [°N]
60
*
30
0
–30
–60
– 90
0
45
90
135
180
225
Longitude [°E]
270
315
360
Figure 1. The global distribution of continental shelf zones. The asterisk indicates the East China Sea shelf.
The model used in this study consists of two parts: the carbon
cycle model itself and the world ocean GCM which it is embedded
into.
surface) to 615 m at level 20 (ocean floor; 5500 m). Within the
confines of this relatively low resolution the GCM has a realistic
representation of land distribution and ocean floor depths. The
GCM is run with a time step of 24 hours.
The GCM is of Bryan-Cox primitive equation form, with
ocean temperature and salinity modeled as active tracers.
Monthly fields of heat flux [Esbensen and Kushnir, 1981],
precipitation/evaporation [Jaeger, 1976; Esbensen and Kushnir,
1981], and wind stress [Hellerman and Rosenstein, 1983] force
the GCM at its surface boundary. The dynamics of the GCM’s
surface are also driven by a Kraus-Turner bulk mixed layer
[Kraus and Turner, 1967]. This homogenizes the surface layers
down to the base of the calculated mixed layer. To prevent
excessive drift away from observations, temperature and salinity
in the surface layer are relaxed back toward the monthly
climatologies of Levitus and Boyer [1994] and Levitus et al.
[1994], respectively. A relaxation timescale of 60 days (for a
50 m mixed layer) is used.
The Gent and McWilliams [1990] eddy parameterization scheme
is implemented, and this permits the model to exclude explicit
horizontal diffusion. Tracers are also subject to Redi’s [1982]
along-isopycnal diffusion scheme. Vertical diffusion between
model layers is depth dependent, with the background coefficient
increasing from 0.1 cm2 s1 at the ocean’s surface to 1.5 cm2 s1
at 5000 m, from the observationally based estimates of Kraus
[1990].
The North Atlantic overturning circulation of the GCM for the
section at 24N is 24 Sv. This compares with an observational
estimate for the same section of 19 Sv by Hall and Bryden [1982].
Because of the narrowness of its representation in the model grid,
flow through the Bering Straits is negligible in the GCM. Flow from
the Pacific Ocean to the Indian Ocean via the Indonesian Throughflow is very strong in the GCM at 23 Sv. This is considerably
greater than the 7 Sv estimate of Wijffels et al. [1996] from
hydrographic sections of the region (see Banks [2000] for a
thorough study of this region in a variant of the GCM). For further
details regarding the GCM, refer to Gordon et al. [2000] and
Palmer and Totterdell [2001].
2.1. Ocean GCM
This work makes use of the U.K. Meteorological Office’s
HadCM3L ocean GCM. The world ocean is represented by a grid
of cells each measuring 2.5 latitudinally by 3.75 longitudinally. In
the vertical dimension the ocean is divided into 20 fixed level
layers. These increase in thickness from 10 m at level 1 (ocean
2.2. Carbon Cycle
Although carbon in the ocean occurs in many inorganic and
organic forms, for simplicity, this study uses the single DIC tracer
model [Orr et al., 1999] produced for phase 2 of the international
Ocean Carbon Cycle Model Intercomparison Project (OCMIP-2).
The protocol describing this model and the data fields that force it
As mentioned in the last paragraph, models are now commonly used to provide estimates of the world ocean’s uptake of
CO2 for studies of both the natural and anthropogenically forced
carbon cycle. Increasingly, full ocean general circulation models
(GCMs) are being used, with carbon represented simply by a
single DIC tracer [Taylor, 1995] or, in more complex models, by
a series of chemically or biologically based carbon tracers
[Fasham et al., 1993; Six and Maier-Reimer, 1996; Palmer
and Totterdell, 2001]. However, GCMs rarely represent the
continental shelf zone or the processes that occur on it. This
is partially related to the relatively coarse resolution of most
GCMs. Grid cells are typically far greater in extent than the
continental shelf. Furthermore, the physical processes that occur
where the open ocean meets the shelf (e.g., shelf breaks and
fronts) are not characterized well by coarse-resolution models.
For these and other reasons the continental shelf zones are rarely
explicitly included in model simulations. Modelers usually
assume that processes occurring on the shelves have very little
impact on the much larger open ocean areas of their models.
Consequently, most GCMs neither resolve nor represent the
continental shelf zone.
However, while GCMs may be incapable of explicitly representing the processes that Tsunogai et al. [1999] attribute to their CSP,
they could represent the processes that transport carbon off the
shelves and into the deep ocean (or, at least, the processes that
transport carbon from the regions adjacent to the shelves into the
open ocean). This study attempts to examine these latter processes
by parameterizing the shelf processes that transfer carbon to denser
shelf bottom water. The GCM used then follows the progress of this
carbon to attempt to quantify the strength of the CSP. Additionally,
the GCM’s results can suggest where CSP carbon outgasses in the
open ocean, an issue difficult to resolve observationally.
2. Model Description
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
3
90
Latitude [°N]
60
30
0
–30
–60
–90
0
0
45
1
90
2
135
3
180
225
Longitude [°E]
4
5
6
270
7
315
8
9
360
10
x 10
4
Figure 2. The distribution of continental shelf area onto the GCM grid. Area is in km2.
are freely available from the OCMIP-2 website (see http://
www.ipsl.jussieu.fr/OCMIP/). For detail beyond that described
below, refer to this site.
DIC is a passive tracer in the GCM, interacting solely with the
model ‘‘atmosphere’’ at the ocean’s surface. Its conservation
equation is
d½DIC
¼ q½DIC þ Fv þ F þ CSP;
ð1Þ
dt
where [DIC] is the model’s concentration (mol m3) of total
dissolved inorganic carbon; q is the three-dimensional (3-D)
transport operator representing changes in DIC concentration due
to advection, convection, diffusion, and mixing; Fv is the ‘‘virtual’’
flux term for DIC representing the changes in its surface
concentration due to evaporation and precipitation; F is the flux
term for DIC representing air-sea exchange of CO2; and CSP is the
flux term for DIC representing the addition of DIC to the GCM via
the parameterized CSP (see section 2.3).
The use of a virtual flux is required as the GCM has a ‘‘rigid
lid.’’ In the real ocean, evaporation and precipitation (EP) remove
and add water to the ocean, respectively, changing the volume of
water above a given point of the seafloor. In the GCM the fixed
vertical levels effectively prevent any change in volume, so a
virtual flux is introduced to account for changes to model tracers
brought about by the concentrating/diluting effects of EP. In this
GCM, evaporation and precipitation are additionally modified by
the relaxation term.
Air-sea exchange of CO2 is a function of the atmospheric and
surface ocean concentrations of CO2, air pressure, piston velocity,
and ice cover. In the model these latter three variables are specified
as OCMIP-2 monthly data fields. Atmospheric [CO2] is held at a
globally constant value of 278 ppm (preindustrial level; see below).
Surface ocean [CO2] is calculated from surface [DIC], temperature,
salinity, and [Alk] using standard OCMIP-2 carbon chemistry
routines, where [Alk] is alkalinity (eq m3) and is determined as
a normalized linear function of model salinity.
Although this model excludes many important processes, including the biological effects on both [DIC] and [Alk] and the riverine
input of carbon, it captures several aspects of the natural carbon
cycle and has proved a very useful baseline model for OCMIP-2
experiments.
A preindustrial (278 ppm) level of atmospheric CO2 was chosen
over a contemporary one (365 ppm) for several reasons. In order
to facilitate simulation analysis it is preferable to minimize model
drift by initiating simulations from an ‘‘equilibrium’’ state, where
carbon fluxes into and out of the ocean are in balance. The
contemporary atmosphere-ocean system is considerably out of
equilibrium because of the industrial CO2 perturbation, with a
net flux into the ocean at present. In contrast, prior to the industrial
period, atmospheric CO2 remained broadly constant (260 – 280
ppm) during the current interglacial period (10000 years), suggesting a natural carbon cycle close to equilibrium or, at least,
converging slowly onto one. Furthermore, the pump proposed by
Tsunogai et al. [1999] should operate independently of an anthropogenic perturbation. It should play a role in both the natural and
anthropogenically forced carbon cycle (although its strength in the
natural state is not calculated by Tsunogai et al. [1999]).
2.3. Parameterizing the CSP
As the processes which drive the CSP cannot be properly
represented within the GCM, a more direct approach was used to
simulate its effects. The estimate of the pump’s strength provided
by Tsunogai et al.’s [1999] study (35 g C m2 yr1) was assumed
to be constant across all of the world’s continental shelves. The
shelf areas were then regridded (see Appendix A) onto the GCM’s
grid, and the total shelf area of any given GCM grid cell was
calculated. Figure 2 shows the redistribution of shelf areas onto the
GCM grid.
During simulations using the CSP, DIC was added to GCM grid
cells according to the following formula:
of shelf in cell
ðpump rateÞð1 ice fractionÞ areaarea
of cell
ðcell thicknessÞ
ð2Þ
DIC was added to layer 8 of the GCM (113 – 165 m) where
possible. Layer 8 was chosen as the deepest model layer that did
not cross the 200 m threshold (the base of layer 9 is 242 m). Where
the GCM grid was shallower than layer 8, DIC was added to the
deepest layer. Since a large region of the continental shelf zone lies
in the Arctic Ocean (and a smaller region around the coastline of
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
4
90
25
9
27
26
1
60
23
Latitude [°N]
15
30
2
16
29
13
10
12
0
14
6
18
3
21
30
19
28
5
32
–60
31
24
–90
7
4
20
17
11
–30
8
22
0
45
90
135
180
225
Longitude [°E]
270
315
360
Figure 3. Annotated breakdown of the global shelf into regions. 1, Labrador Sea; 2, Grand Banks and Eastern
Seaboard; 3, Carribean Sea; 4, Amazonian shelf; 5, Patagonian shelf; 6, Southwest African shelf; 7, Northwest
African shelf; 8, North Sea; 9, Iceland and Greenland shelf; 10, East African shelf; 11, Madagascar shelf; 12,
Seychelles; 13, Indian shelf; 14, Indonesian archipelago; 15, Okhotsk Sea; 16, East China Sea; 17, Australian shelf;
18, New Guinea and nearby islands; 19, New Zealand shelf; 20, Western South America; 21, Western Central
America; 22, Western North America, 23, Bering Sea; 24, Antarctic shelf; 25, Russian Arctic; 26, Siberian Arctic; 27,
North American Arctic; 28, Kerguelen Islands; 29, Hawaii; 30, Galapagos Islands; 31, South Orkney; 32, South
Georgia.
Antarctica), the pump rate was affected by ice cover in the same
manner as the air-sea flux.
2.4. DOC/DIC Model Variant
The carbon model outlined above forms the basis for the bulk of
the work in this study. As a secondary experiment, a model was
constructed in which the carbon added to the ocean via the CSP is
added as a DOC-like tracer. Unlike DIC, this tracer does not
interact with the atmosphere, although it is remineralized at a fixed
rate back into DIC. The conservation equations for this model are
d½DOC
¼ q½DOC þ CSP l½DOC
dt
ð3Þ
d½DIC
¼ q½DIC þ Fv þ F þ l½DOC;
ð4Þ
dt
3
where [DOC] is the model’s concentration (mol m ) of dissolved
organic carbon, CSP is the flux term for DOC representing the
addition of DIC to the GCM via the parameterized CSP (it is the
sole input term for DOC), and l is the remineralization rate of
DOC into DIC. The motivation for this model is the suggestion by
Tsunogai et al. [1999] that biological production acts as a
mechanism in their proposed CSP. Carbon fixed by biological
activity cannot escape to the atmosphere until it has been
remineralized back into DIC. This model assumes that all of the
carbon sequestered by the pump is of biological origin and aims to
examine the significance of the delay a biological loop can
introduce and its consequences on the efficiency of the CSP.
However, the choice of duration for this delay is not simple. In
the real ocean, the pool of DOC is composed of a wide spectrum of
organic molecules, pragmatically categorized into labile (short
lifespan, days), semilabile (intermediate lifespan, months to years),
and refractive (long lifespan, centuries to millenia) forms [Kirchman et al., 1993; Carlson and Ducklow, 1995]. DOC lifespan is a
function of many factors, including chemical composition and
bonding, oxidation by UV radiation near the ocean surface, microbiological decomposition, adsorption onto particles, and even
interactions between DOC molecules themselves [Anderson and
Williams, 1999]. The DOC-like tracer in this model most closely
resembles semilabile DOC, given that this is produced at relatively
high rates (unlike refractive DOC) and survives long enough to be
exported deeper into the water column (unlike labile DOC). To
simplify the processes involved, a linear remineralization function
is assumed, and two simulations are performed, one with a monthly
decay rate (l = 0.0333 days1) and the other with an annual rate (l
= 0.0028 days1).
As a side note, in reality, biological activity produces POC as
well as DOC. Chen and Wang [1999] estimate that the East China
Sea exports 8.34 ± 4.20 Mt C yr1 of POC, 40% of total
organic carbon exported. Should the production and export of
POC be particularly high and have a size spectrum shifted toward
larger particles, the sinking of this material could significantly
enhance the action of the CSP. However, to simplify the model,
only the effect of a delaying loop on the pump’s activity is
examined here.
3. Simulations
3.1. Equilibrium Spin-up
Before any simulations involving the CSP were performed, the
carbon cycle model was spun up to approach its baseline equilibrium and to minimize model drift. The OCMIP-2 protocol was
followed, and the model was simulated until the globally integrated
annual air-sea flux of carbon (i.e., drift) was <0.01 Pg C yr1. This
took a period of 8000 simulated model years. The end state of
this spin-up period was used as the initial condition for all of the
simulations performed in this study.
3.2. Experiments
Four simulations were initialized from the equilibrium state: (1)
control, DIC only model and no CSP; (2) DIC only, DIC only
model and CSP active; (3) DOC monthly, DOC/DIC model, CSP
active, and monthly decay of DOC; (4) DOC annual, DOC/DIC
model, CSP active, and annual decay of DOC. Each simulation
was run for 5000 years. This period was sufficient for the outgassing flux of pump carbon to balance the influx parameterized by
the CSP to four significant figures. The remaining drift in the
simulations was considerably smaller than the equilibrium drift
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
Table 1. Breakdown of the Global Shelf Into Regions of Broadly
Continuous Shelf
Shelf Region
Area, 106
km2
Fraction,
%
Total world ocean shelf
Labrador Sea
Grand Banks and Eastern Seaboard
Carribean shelf
Amazonian shelf
Patagonian shelf
Southwest African shelf
Northwest African shelf
North Sea
Iceland and Greenland
East African shelf
Madagascar shelf
Seychelles
Indian subcontinent shelf
Indonesian archipelago
Okhotsk Sea
East China Sea
Australian shelf
New Guinea and nearby islands
New Zealand shelf
Western South America
Western Central America
Western North America
Bering Sea
Antarctic shelf
Russian Arctic shelf
Siberian Arctic shelf
North American Arctic shelf
Kerguelen Islands
Hawaii
Galapagos Islands
South Orkney
South Georgia
20.397
0.382
0.887
0.744
0.661
1.539
0.335
0.264
1.015
0.270
0.243
0.087
0.073
0.443
2.431
0.611
1.577
2.445
0.120
0.236
0.167
0.048
0.293
1.123
0.487
1.503
1.825
0.498
0.057
0.006
0.009
0.009
0.007
100
1.87
4.35
3.65
3.24
7.55
1.64
1.29
4.98
1.32
1.19
0.43
0.36
2.17
11.92
3.00
7.73
11.99
0.59
1.16
0.82
0.24
1.44
5.51
2.39
7.37
8.95
2.44
0.28
0.03
0.05
0.04
0.03
criterion. The Control simulation was used to correct the three CSP
simulations for drift.
To assess the regional strength of the CSP, the global shelf was
divided into 32 geographical areas. Figure 3 shows the location and
extent of these regions. Table 1 lists their areas and the fractions of
global shelf that they make up. They aim to characterize continuous areas of continental shelf and range in size from 2.445 106
km2 (Australian shelf) to 0.006 106 km2 (Hawaii).
4. Results
Plate 1 shows the actual rate of carbon addition of the parameterized CSP as per (2). Because of ice cover in the Arctic and
Antarctic the pump rate is reduced in these areas below that which
would be expected given the shelf areas there. This field was
applied identically in all three of the simulations that included the
CSP.
4.1. DIC Simulation
Figure 4 shows three global diagnostics collected from the four
simulations for the full 5000 years (the results of the DOC
simulations are discussed more fully below). For all three diagnostics the control simulation shows only slight, if any, drift. The
three CSP simulations behave broadly similarly. As the middle
panel shows, ocean surface pCO2 increases as pump carbon returns
to the surface. The lower panel shows how this is reflected in the
air-sea flux. In all three cases, there is a substantial negative flux as
5
pump carbon escapes back into the atmosphere. However, as the
top panel shows, the rate of carbon escape through the simulations
is insufficient to prevent a buildup of pump carbon in the oceans of
the three pump simulations. At the end of 5000 years, the DIC
simulation has retained 43.68 Gt of pump carbon. Given the pump
rate of 0.582 Gt C yr1, this represents 73.3 years worth of pump
carbon. As can be seen from Figure 4, after 5000 years, the
simulations have broadly reached, or are close to, equilibrium.
Following on from the bottom panel of Figure 4, Plate 2 shows
the global distribution of outgassing pump carbon from the DIC
simulation. As can clearly be seen, some pump carbon has reached
areas of the ocean far removed from the continental shelves. Of
particular note is the outgassing along the northeasterly flowing
Gulf Stream. This is primarily carbon entering from the Caribbean
Sea and Eastern Seaboard. Pump carbon from the Patagonian shelf
also outgasses along a major plume through the action of the Brazil
Current and the Antarctic Circumpolar Current (ACC). In several
regions though (e.g., Indonesian archipelago, North Sea, and East
China Sea), there is a considerable amount of outgassing occurring
on the parameterized continental shelf.
Globally, of the 0.582 Gt C annually added by the parameterized
CSP, 0.276 Gt C (47.4%) outgasses within the global shelf region.
To simplify analysis, this region is defined as all of those GCM
cells that overlie a region of the parameterized shelf. As Plate 2
shows, the remaining 52.6% is exported to neighboring or more
distant regions of the world ocean where it outgasses.
Table 2 shows a regional breakdown of outgassing and export.
Table 2 is ranked by an estimate of the efficiency of export (i.e., the
fraction of pump carbon added to a given shelf which is not
outgassed on that shelf ), and efficiencies range from 23.7%
(Iceland and Greenland) to 91.6% (Hawaii). Negative efficiencies
occur where owing to import from another shelf region, more
pump carbon is outgassed than added to a particular region. Since
the carbon added to a given shelf region is indistinguishable from
that added to another region, it is not possible to separate the
fraction of outgassing pump carbon of local origin from that
imported from another shelf region. This complicates the issue of
export efficiency, so the figures calculated should be treated as
simple estimates.
There are few obvious patterns to the export efficiencies of the
various shelf regions. Although the top two regions (Hawaii and
the Galapagos) are both small island shelves, other ostensibly
similar regions (Seychelles and South Georgia) have considerably
lower efficiencies. The four Western Pacific regions all have
relatively high export efficiencies, as do those of the subpolar
eastern Atlantic and Western Africa. One pattern which does
appear to hold is that the least efficient regions (
30%) are all
polar (or subpolar) regions in the Northern Hemisphere (with the
sole exception of South Georgia). However, in contrast, highlatitude regions in the Southern Hemisphere, like the Antarctic
shelf and the Kerguelen Islands, have efficiencies >60%. By
coincidence, the East China Sea, the focus of Tsunogai et al.’s
[1999] study, has an export efficiency extremely close to (and the
closest to) that of the global average.
Pump carbon enters the modeled ocean near its surface and, as
Plate 2 has shown, exits across a relatively large area of its surface.
Table 3 and Plates 3 and 4 give a fuller picture of the distribution of
pump carbon through the ocean interior at year 5000. Table 3
summarizes the vertical distribution of pump carbon in the world
ocean. More than 94% of the total inventory is located deeper than
200 m, and >72% is deeper than 1 km. Note, however, that the
actual concentration of stored pump carbon generally falls with
depth.
Plate 3 shows the zonal average of pump carbon from the DIC
simulation both down the Atlantic basin and back up the Pacific
basin. There are considerable differences in the patterns of
6
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
Plate 1. Rate of carbon addition due to parameterized ‘‘continental shelf pump.’’ Pump rate is in g C m2 yr1.
storage between the basins. Because of the Atlantic’s role in deep
water formation, relatively high concentrations of pump carbon
have made their way into midwater (2 km depth), southward
moving North Atlantic Deep Water (NADW). As this water mass
reaches the ACC, its contents are distributed around the Southern
Ocean. Pump carbon from the Patagonian and, particularly,
Indian Ocean shelves also helps to raise pump carbon concentrations in the Southern Ocean. Because of the relatively low
pump inputs directly to the waters of the Southern Ocean (the
Antarctic shelf contributes only 6.7 Mt C yr1) its surface waters
have considerably less pump carbon than the waters of the
Atlantic Ocean.
Concentrations of pump carbon through the Pacific basin are
considerably (1.5 mmol C m3) lower than those in the Atlantic
basin. This is at least partially related to the fact that per unit
volume the Pacific receives less pump carbon than the Atlantic (20
versus 57 mmol m3 yr1) and that Pacific shelves are generally
less efficient exporters than Atlantic ones (42 versus 59%). The
absence of deep water formation traps most pump carbon in the
surface kilometer of the ocean. Pump carbon reaches the deep
Pacific partially through vertical diffusion but mostly via the ocean
conveyor belt. In the real world this is perhaps most clearly
illustrated by the distribution of 14C, the carbon radioisotope.
Because this isotope is replenished from the atmosphere and
decays on timescales comparable to that of the conveyor, it exists
at much depleted concentrations in the deep Pacific [Toggweiler et
al., 1989].
Plate 4 complements Plate 3 by showing the vertical inventory of
pump carbon from the DIC simulation. The carbon-rich waters of
the Atlantic and carbon-poor waters of the Pacific are clearly
Plate 2. Rate of ‘‘continental shelf pump’’ carbon outgassing from DIC simulation. Control-corrected to remove
background carbon air-sea fluxes. Outgassing rate in g C m2 y1.
Retained DIC [Gt C]
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
7
40
20
0.2
0
–0.2
–0.4
–0.6
2
pCO [ppm]
290
289
288
287
Air-sea flux [Gt C y -1]
0
0
0
0
1000
2000
3000
4000
5000
1000
2000
3000
4000
5000
1000
2000
3000
Time [years]
4000
5000
Figure 4. (top) Retained dissolved organic carbon (DIC) inventory (Gt C), (middle) ocean surface pCO2 (ppm), and
(bottom) mean annual air-sea flux (Gt C yr1). Dotted line is the control simulation; solid line is DIC simulation; dotdashed line is DOC monthly simulation; dashed line is DOC annual simulation.
discernable. The Indian Ocean is also now visible as a major store
of pump carbon. Interestingly, the presence of ice cover over the
Arctic Ocean traps some of the highest inventories of pump carbon.
However, the relatively smaller area of this sea reduces the global
importance of this store.
Finally, Table 4 breaks down the properties of the CSP for the
global ocean and five ocean basins. The total area portion of the
table details the areas of these regions, the fraction of the world
ocean that they make up, and the fraction of each of them that is
designated ‘‘open ocean’’ (i.e., GCM cells containing no continental shelf ). The Indian Ocean and, especially, the Arctic Ocean
have relatively greater proportions of their total area containing
continental shelf, while the Pacific and Southern Oceans consist of
mostly open ocean.
The total pump portion of Table 4 summarizes the CSP inputs to
the various basins. Perhaps surprisingly, shelf area (and thus CSP
inputs) is not evenly distributed around the globe. The Atlantic,
Indian, and Arctic Oceans are comparatively shelf-rich for their
sizes, while the Pacific and Southern Oceans have relatively
smaller shelf areas.
The total outgassing portion of Table 4 presents summaries of
pump carbon outgassing. While the Atlantic and Indian Oceans
outgas at lower rates than they receive pump carbon (particularly
the Indian Ocean), all of the other basins, especially the Southern
Ocean, outgas more pump carbon than they receive. These latter
basins are net importers of pump carbon from the Atlantic and
Indian basins (which act as net exporters). In the case of the
Southern Ocean, the majority of the carbon it receives comes from
the Indian basin, and the majority of its imports are outgassed,
primarily in open ocean areas (79.3%). A smaller fraction of the
carbon reaching the Southern Ocean is transported into the Pacific
basin where it outgasses there.
Note that although the value for the open ocean fraction of
outgassing for the global ocean matches the global pump efficiency
presented in Table 2, the corresponding values presented for the
individual ocean basins are not comparable because of transport of
pump carbon between ocean basins. Net transport for the global
ocean is by necessity zero, but the same is not true for ocean
basins.
Finally, the total inventory portion of Table 4 summarizes where
stored pump carbon in the ocean is located as well. Although the
largest share of pump carbon is found in the Pacific basin (33.8%),
when the basin’s large volume is taken into consideration, this
translates to an average concentration of pump carbon of 2.160
mmol m3. In contrast, while the Atlantic and Indian basins have
relatively lower shares of the total inventory of pump carbon (26.6
and 20.2%, respectively), when considering their lower volumes
both have average concentrations of >3.8 mmol m3. Similar to the
Pacific Ocean, the Southern Ocean’s share of the inventory
(15.4%) is diluted through the basin’s volume to 2.081 mmol
m3. Interestingly, and in contrast to the other basins which import
pump carbon, the Arctic Ocean stores pump carbon at concentrations much greater than any of the other basins (9.301 mmol
m3). Around 4% of the total inventory is stored within 1.2% of
the global ocean’s volume. As hinted at in Plate 4, the Arctic
Ocean’s large covering of permanent sea ice enables this storage.
Pump carbon enters the Arctic Ocean from its seasonally ice-free
fringes (where the majority also outgasses) but is able to remain
trapped for the longer term under the ice cap which prevents
communication with the atmosphere.
4.2. DOC Simulations
In both of the simulations in which a DOC-like precursor was
added instead of DIC the quantity of DIC stored at year 5000 is
increased. However, the difference in DIC storage between the
DOC monthly and DOC annual simulations is marked. Figure 4
shows three global diagnostics for both simulations alongside those
from the DIC and control simulations. In all three cases, only the
DOC annual simulation can readily be distinguished. At year 5000
the DIC simulation has a global inventory of pump carbon of 43.68
Gt C, while the DOC monthly simulation has 43.95 Gt C (+0.6%)
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
8
Table 2. Summary of Regional Continental Shelf Pump, On-Shelf Outgassing, Off-Shelf Export, and Export Efficiency of
Continental Shelf Pumpa
Pump, Gt C yr1
Shelf Region
Hawaii
0.0002
Galapagos Islands
0.0003
Amazonian shelf
0.0231
Western Central America
0.0017
Western North America
0.0103
Patagonian shelf
0.0539
Madagascar shelf
0.0031
Northwest African shelf
0.0092
New Guinea and nearby islands
0.0042
Western South America
0.0059
New Zealand shelf
0.0083
Southwest African shelf
0.0117
Kerguelen Islands
0.0020
Antarctic shelf
0.0067
Australian shelf
0.0856
Indian subcontinent shelf
0.0155
Carribean shelf
0.0260
Grand Banks and Eastern Seaboard
0.0310
East African shelf
0.0085
South Orkney
0.0002
Total world ocean shelf
0.5823
East China Sea
0.0552
Indonesian archipelago
0.0851
Seychelles
0.0026
Labrador Sea
0.0079
North Sea
0.0355
South Georgia
0.0002
Russian Arctic shelf
0.0231
Bering Sea
0.0314
Siberian Arctic shelf
0.0078
North American Arctic shelf
0.0017
Okhotsk Sea
0.0212
Iceland and Greenland
0.0031
a
Table entries are ranked by export efficiency.
and the DOC annual simulation has 48.99 Gt C (+12.2%). Figure 5
shows the fractional gain delivered by the DOC phase for both
simulations across the first 1000 years of the simulations. In both
cases they rapidly (within 200 – 300 years) equilibriate to constant
fractional gains over the DIC simulation. This occurs despite carbon
storage increasing throughout the duration of the simulations.
As Plate 5 and Table 5 show, however, this global constancy is
made up of considerable regional variation. Plate 5 shows the
percentage increase in pump carbon inventory resulting from the
DOC annual simulation. Note that while increases in pump
carbon storage in the shelf regions in Plate 5 often exceed
30%, these apparently extreme increases are usually in areas
where carbon storage in the DIC only simulation is low (see Plate
4). At the large scale, while the Atlantic and, to a lesser extent,
Pacific and Southern Oceans show relatively constant increases in
carbon storage of 10 – 15%, the Arctic and Indian Oceans lie
Table 3. Summary of the Vertical Distribution of Retained Pump
Carbon From the DIC Simulation
Depth Interval, m
0 – 200
200 – 500
500 – 1000
1000 – 2000
2000 – 5500
Total
Retained Carbon
Gt C
Fraction, %
mmol m3
2.5648
3.7101
5.7768
11.0125
20.6151
43.6794
5.87
8.49
13.23
25.21
47.20
100
3.0125
2.9694
2.8477
2.8808
2.7052
2.8049
Outgas, Gt C yr1
Export, Gt C yr1
Efficiency, %
0.0000
0.0000
0.0037
0.0003
0.0025
0.0139
0.0008
0.0030
0.0014
0.0019
0.0028
0.0040
0.0007
0.0024
0.0314
0.0057
0.0100
0.0120
0.0037
0.0001
0.2759
0.0265
0.0462
0.0015
0.0055
0.0253
0.0002
0.0171
0.0234
0.0059
0.0013
0.0185
0.0039
0.0002
0.0003
0.0194
0.0014
0.0078
0.0400
0.0022
0.0063
0.0028
0.0039
0.0055
0.0077
0.0013
0.0043
0.0541
0.0098
0.0160
0.0190
0.0048
0.0001
0.3064
0.0287
0.0389
0.0010
0.0024
0.0102
0.0001
0.0060
0.0081
0.0019
0.0004
0.0027
0.0007
91.60
86.66
83.85
81.92
75.79
74.23
72.39
67.94
66.82
66.75
66.45
65.64
64.44
63.67
63.25
3.21
61.65
61.15
56.89
53.49
52.62
52.04
45.66
40.76
30.14
28.68
28.53
26.11
25.63
23.90
21.16
12.91
23.65
outside this range. The Arctic ranges from 15% to values
around 22%, and these are particularly significant given that
the Arctic already stores pump carbon at high concentrations. In
contrast, the Indian Ocean ranges 5 – 10% with a strong northsouth gradient. In the Arctic Ocean it is almost certainly the case
that the lifespan of DOC allows more of it to get under the icecap
before it decays to DIC. The variation in the Indian Ocean is
more difficult to explain without a detailed analysis of pump
carbon fluxes within and between basins. However, some of the
variability is presumably due to regions where pump carbon is
very quickly subducted or otherwise isolated from the atmosphere
(i.e., pump carbon is stored whether DIC or DOC) or where
pump carbon is trapped near the surface for periods greater the
lifespan of model DOC (i.e., pump carbon is outgassed whether
DIC or DOC).
Table 5 shows the breakdown of export efficiency (as in Table 2)
by region for the DOC monthly and DOC annual simulations. For
each region the export efficiency and its percentage difference to
the DIC simulation are shown. Since the DIC and DOC tracers are
transported identically, where a region shows a lower efficiency
indicates that the region concerned has imported (or reimported
previously exported carbon) more pump carbon (DIC or DOC)
than in the DIC simulation and that this has increased outgassing
and decreased export efficiency. In the case of monthly DOC, 12 of
the 32 regions have decreased in efficiency, compared to only 5
with annual DOC. Generally, swings in efficiency, both positive
and negative, are more pronounced in the DOC annual simulation
(25.9 to +55.6%) than in the DOC monthly simulation (2.1 to
+10.8%), although efficiency changes usually match one another in
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
9
Plate 3. Zonal average distribution of continental shelf pump carbon (left) down the Atlantic and (right) up the
Pacific basins. The dashed lines indicate the boundary of the Southern Ocean. Concentration is in mmol m3.
direction between the simulations. Similar to Table 2, there are few
patterns to the changes. As a general rule, previously efficient
exporting regions tend to have become less efficient, while those
less efficient have become more efficient, although there are
notable exceptions to this (the Patagonian shelf in the former case
and the North American Arctic shelf in the latter).
Finally, for comparison with Plate 2, Plate 6 shows the difference
in patterns of outgassing between the DIC only and the DOC
annual simulations. Blue regions are where outgassing is reduced
in the DOC simulation, while red regions are where outgassing
increases. The general distribution of blue coastal areas and red
open ocean areas is consistent with the increase to CSP efficiency
by the addition of a DOC phase. Increases are particularly noticeable on the shelves of the eastern Americas, Australia, the East
China Sea, and the Bering Strait. In most regions, a decrease in
outgassing on shelf is matched by a neighboring off-shelf region of
increased outgassing. In the case of the Patagonian shelf the region
of decreased outgassing extends markedly into the open ocean.
5. Discussion
This study has aimed to provide a global estimate of the CSP by
extrapolating up from the East China Sea study of Tsunogai et al.
[1999] and using a simple carbon model and a relatively coarse
Plate 4. Vertical inventory of ‘‘continental shelf pump’’ carbon from DIC simulation. Inventory in mol m2.
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
10
Table 4. A Breakdown of Major Pump Carbon Statistics by Ocean Basinsa
Total area, 106 km2
Basin fraction, %
Ocean fraction, %
Total pump, Gt C yr1
Basin fraction, %
Ocean fraction, %
Total outgassing, Gt C yr1
Basin fraction, %
Ocean fraction, %
Total inventory, Gt C
Basin fraction, %
Ocean fraction, %
World
Atlantic
Indian
Pacific
Southern
Arctic
355.95
...
82.07
0.5823
...
0
0.5823
...
52.62
43.68
...
87.98
70.00
19.67
77.69
0.1698
29.16
0
0.1563
26.84
57.12
11.62
26.61
85.87
58.86
16.54
69.63
0.1967
33.78
0
0.1613
27.70
42.56
8.84
20.23
77.60
141.63
39.79
89.42
0.1364
23.42
0
0.1493
25.64
52.02
14.76
33.78
94.06
72.49
20.36
89.64
0.0440
7.56
0
0.0780
13.40
79.25
6.73
15.41
95.88
12.96
3.64
39.65
0.0354
6.08
0
0.0374
6.42
24.08
1.73
3.97
72.56
a
In addition to the four major ocean basins, the Artic Ocean is also distinguished because of its relatively large continental shelf area.
Each of the four statistics (total area, total pump, total outgassing, and total inventory) of the table is broken into three parts: the statistic
itself, the fraction of the world ocean each of the basins contributes, and the fraction for each basin that occurs within that basin’s open
ocean area. See text for further details.
ocean GCM. It is important to note that because the parameterized
pump uses the shelf export rate calculated by Tsunogai et al.
[1999], rather than some estimate of the quantity of carbon transferred to the vicinity of the shelf floor, the GCM can only predict a
maximum export rate equal to that found by Tsunogai et al. [1999].
GCM-predicted export efficiency gives a more useful measure of
the pump strength. If it is very low, the majority of carbon added to
the shelf will remain there and outgas locally because it creates a
disequilibrium. If high, the majority will be transported and outgassed elsewhere in the GCM.
In the case of the East China Sea itself the GCM predicts that
52.0% of the carbon added near the seafloor of the shelf is exported
to either the open ocean or to other shelf regions (neighboring or
otherwise). This corresponds to an export of 18.2 g C m2 yr1.
Assuming that the GCM correctly models the hydrography around
the East China Sea, the argument presented in the preceding
paragraph would suggest that to correct for this lower absolute
quantity of export (i.e., to bring it up to the same level as found by
Tsunogai et al. [1999]), the quantity added to the seafloor should
be approximately doubled (ignoring nonlinearities in air-sea
exchange and transport).
Scaling up this doubling in the base rate of pump carbon addition
to the whole world and assuming that all other things remain equal,
this would take the total export of the world ocean’s shelves to
0.589 Gt C yr1 (from an addition to the shelf floor of 1.120 Gt C
yr1). This figure is 27% of Takahashi et al.’s [1999] estimate of
the world ocean’s uptake of anthropogenic CO2 during 1995, and
such a flux would be significant if ignored in estimates based upon
either observations or modeling studies.
Several caveats should be noted however. First, Tsunogai et al.’s
[1999] estimate of the strength of the CSP is based upon contemporary measurements. Currently the world ocean is out of equilibrium with the atmosphere because of the anthropogenic
perturbation (although the surface ocean is considerably closer to
1.18
DIC only
Monthly
Annual
1.16
Fractional gain [dimensionless]
1.14
1.12
1.1
1.08
1.06
1.04
1.02
1
0.98
0
100
200
300
400
500
600
Time [years]
700
800
900
1000
Figure 5. Fractional gain in carbon storage due to the DOC phase. Dotted line is baseline (DIC simulation); dotdashed line is DOC monthly simulation; dashed line is DOC annual simulation. Gain is dimensionless.
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
11
Plate 5. Percentage difference in the vertical inventories of continental shelf pump carbon between the DOC annual
and DIC simulations.
equilibrium than the deep ocean). At equilibrium the CSP may be
less effective, and the premises of this study less secure. In
particular, the above global estimate of CSP activity is strongly
tied to the pre-industrial equilibrium approach of this study. However, note that Tsunogai et al.’s [1999] observations of the CSP’s
current role in the East China Sea render it important to contemporary studies irrespective of its role in the natural carbon cycle.
Second, a major assumption in this study has been that the CSP
operates equally in all regions of the world shelf. This assumption
is critical, and its validity is dependent on patterns of seasonal
forcing (physical and biological) and hydrography. In particular,
polar shelf regions (which, by area, make up 21% of the modeled
shelf ) have very different seasonal regimes of temperature and
biological activity to those of the East China Sea. However, how
common these patterns are across the world shelf cannot be
ascertained until a global survey of shelf regions is undertaken
with particular regard to the properties that drive the CSP. Synoptic
observations such as satellite measurements of SST or surface
chlorophyll may provide a qualitative estimate of the occurrence of
conditions suitable for strong CSP activity.
For example, in the case of the East China Sea the formation of
cool, dense water which both sinks and absorbs CO 2 is crucial to
the activity of the CSP. If one restricts sites of potential CSP
activity to those regions with a net negative annual heat flux (as
determined from the Josey et al. [1999] climatology), this reduces
the available shelf area to 13.2%. Calculating the CSP efficiency of
Plate 6. Change in the rate of continental shelf pump carbon outgassing between the DIC only and DOC annual
simulations. Blue areas indicate regions of reduced outgassing in the DOC annual simulation, while red areas indicate
increased outgassing. Change in outgassing rate is in g C m2 yr1.
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
12
Table 5. Export Efficiencies of the Continental Shelf Pump for the DOC Monthly and DOC Annual simulationsa
DOC Monthly
Shelf Region
Efficiency
Percent Gain
DOC Annual
Efficiency
Percent Gain
Hawaii
91.57%
0.0
90.70%
1.0
Galapagos Islands
86.62%
0.0
86.06%
0.7
Amazonian shelf
83.84%
0.0
84.86%
+1.2
Western Central America
81.90%
0.0
81.94%
+0.0
Western North America
75.78%
0.0
76.69%
+1.2
Patagonian shelf
74.78%
+0.8
81.87%
+10.3
Madagascar shelf
72.37%
0.0
72.46%
+0.1
Northwest African shelf
67.96%
+0.0
69.03%
+1.6
New Guinea and nearby islands
66.58%
0.4
63.89%
4.4
Western South America
67.58%
+1.2
72.22%
+8.2
New Zealand shelf
67.33%
+1.3
72.23%
+8.7
Southwest African shelf
65.80%
+0.2
68.50%
+4.3
Kerguelen Islands
68.23%
+5.9
74.28%
+15.3
Antarctic shelf
63.90%
+0.4
66.93%
+5.1
Australian shelf
63.56%
+0.5
67.56%
+6.8
Indian subcontinent shelf
63.20%
0.0
64.03%
+1.3
Carribean shelf
61.88%
+0.4
64.58%
+4.8
Grand Banks and Eastern Seaboard
61.33%
+0.3
66.73%
+9.1
East African shelf
56.89%
0.0
57.38%
+0.9
South Orkney
54.79%
+2.4
60.18%
+12.5
Total world ocean shelf
52.93%
+0.6
56.74%
+7.8
East China Sea
52.48%
+0.9
58.05%
+11.6
Indonesian archipelago
45.76%
+0.2
47.86%
+4.8
Seychelles
40.67%
0.2
37.98%
6.8
Labrador Sea
30.53%
+1.3
32.70%
+8.5
North Sea
29.22%
+1.9
33.23%
+15.9
South Georgia
31.60%
+10.8
38.23%
+34.0
Russian Arctic shelf
26.37%
+1.0
30.27%
+16.0
Bering Sea
25.91%
+1.1
32.88%
+28.3
Siberian Arctic shelf
24.31%
+1.7
25.18%
+5.3
North American Arctic shelf
20.70%
2.1
15.67%
25.9
Okhotsk Sea
13.66%
+5.8
20.08%
+55.6
Iceland and Greenland
23.32%
1.4
27.44%
+16.0
a
The percentage changes from the DIC simulation are also shown. Entries are ranked by export efficiency of the DIC simulation.
this smaller shelf area reduces it to 49.8% (from 52.6%). Interestingly, the efficiency of the East China Sea under this regime rises
to 71.6% (from 52.0%), although when considering the global
statistics, the CSP would only export 0.065 Gt C yr1 (from 0.589
Gt C yr1). However, aside from the cooling process described by
Tsunogai et al. [1999], it is possible that other density-based
mechanisms could exist to drive the CSP. In tropical regions,
evaporation of the shallow shelf waters would increase salinity and
thus water density, while in polar regions the extrusion of salt by
the formation of sea ice could also increase water density. Future
work could examine whether other processes are feasible as
alternate mechanisms to drive the CSP.
Third, export from the shelf regions is expressly a function of the
GCM’s modeled circulation and hydrography. Grid resolution,
submarine topography, the representation of vertical structure
(including isopycnals), subgrid-scale processes, etc., all affect the
GCM’s description of world ocean circulation. At the simplest
level, grid resolution restricts the size of oceanographic features
that can be represented. The coarse resolution of HadCM3L
undoubtedly affects its representation of the near-shore processes
which transport pump carbon into the ocean interior. Models which
better capture reality will undoubtedly produce better estimates of
CSP activity, particularly those models which explicitly represent
shelves (and potentially the processes which drive the CSP).
Finally, the carbon cycle model used in this study is extremely
simple. By representing carbon solely as DIC the model misses out
the significant role biology plays both in the natural carbon cycle
and in the CSP scheme outlined by Tsunogai et al. [1999].
Biological processes will affect the CSP in many different ways,
including seasonal patterns of production, production and packaging of different forms of carbon, variation in the remineralization
rate of organic carbon to DIC, and shifts in surface alkalinity
(which affect the buffering capacity of seawater and, consequently,
air-sea exchange). The carbon cycle model also excludes riverine
fluxes of carbon to the ocean. Chen and Wang’s [1999] budget for
the East China Sea estimates a riverine flux of DIC that is as great
as that they estimate for carbon uptake from the atmosphere (when
including other forms of carbon, the riverine flux is more than
double that of the air-sea flux). Furthermore, the model also
ignores the role the shelf plays as a substrate where organic carbon
can be slowly remineralized. Even when treated as DOC, modeled
pump carbon is always available to transport processes.
However, as a side note to these caveats, it is important to stress
that this study is merely a first step toward quantifying the global
role of the proposed CSP mechanism. These preliminary results
find that the modeled East China Sea is fairly average in its
potential to export carbon and suggest that the CSP may play an
important role in the carbon cycle. Future work improving the
CSP’s representation in GCMs and that of shelves in general
should better establish its strength and improve estimates of its
influence on open ocean carbon budgets.
6. Conclusions
1. Globally, given a carbon addition rate of 35 g C m2 yr1, the
GCM finds that 52.6% (0.306 Gt C yr1) is exported from shelf
regions of the model.
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
2. The East China Sea shelf has a export efficiency of 52.0%,
very similar to the global average.
3. Export efficiency from other regions of the world varies
widely, from 91.6% in Hawaii to 23.7% in the Iceland and
Greenland shelf region.
4. Assuming all other things are equal, scaling pump carbon
addition to reproduce Tsunogai et al.’s [1999] East China Sea
export would predict a global CSP of 0.589 Gt C yr1.
5. Despite being added all over the world, pump carbon mostly
enters the deep ocean (>1 km) via sites of deep water formation.
6. Because of its permanent ice cover, a large reservoir of pump
carbon is located in the Arctic Ocean.
7. Modeling pump carbon as a decaying DOC-like tracer (i.e.,
temporarily unable to interact with the atmosphere) increases its
storage in the ocean (by 12.2% in the case of annually decaying
DOC).
Appendix A: Regridding the Continental Shelf
Data used to determine the location of the continental shelf was
provided by the National Geophysical Data Center (Boulder,
Colorado). The TerrainBase Global DTM version 1.0 database
contains a matrix of land topography and ocean bathymetry for the
entire world, gridded at 5 min intervals.
For ease of data handling, this matrix was first averaged down to
one gridded at 30 min intervals. Then for each GCM cell a list of
the overlapping database cells was identified. Weighting the cells
on these lists for both area and the fraction they overlapped a given
GCM cell, the total area of continental shelf within each GCM cell
was calculated. Since values of topography in the database are
integers, only cells with values <0 m and >201 m were counted
as continental shelf.
Because of the coarse resolution of the GCM grid, large areas of
continental shelf found themselves regridded into GCM cells that
are classified as land in the GCM. This is unsurprising given the
necessary proximity of areas of continental shelf to land. To
remedy this problem, ‘‘land-locked’’ areas of continental shelf
were redistributed evenly to ocean GCM cells within the four cell
neighborhood of land GCM cells. In principle, this permits ocean
GCM cells to contain a greater area of continental shelf than the
area of the cells themselves. In practice, only a minority of GCM
cells were affected, and given that this study was a parameterization of the CSP, this was not considered a problem. These cells
effectively ‘‘service’’ fringe regions of continental shelf inaccessible to the coarse resolution GCM.
Finally, areas of continental shelf occurring within land-locked
regions of the GCM (e.g., Baltic Sea, Black Sea, Hudson Bay,
Persian Gulf, and Red Sea) were ignored since CSP activity cannot
reach the open ocean from these regions. Although a small amount
of communication between the Mediterranean Sea and the North
Atlantic is parameterized in the GCM, areas of continental shelf
within the Mediterranean Sea were ignored because the GCM’s
low resolution prevents a reasonable circulation from developing
within this basin.
Table A1. Total Shelf Area From the Topographic Database and
How it Changes as the Shelf is Regridded and Redistributed onto
the GCM Matrix
Process
Area, 106 km2
Fraction, %
Area calculated from topology
Regridding to GCM matrix
Masking with GCM land
Four cell neighborhood
Remove inland seas
24.405
24.395
18.896
23.157
20.397
100
99.9
77.4
94.9
83.6
13
The result of this procedure was a 2-D field giving the total
continental shelf area in each open ocean GCM cell (see Figure 2).
Table A1 summarizes the changes to ocean shelf area produced by
the regridding procedure. Note that although the final shelf area is
only 20.4 106 km2, this area is contained within GCM grid cells
with a total sea area of 68.3 106 km2.
Acknowledgments. A.Y. is funded by the European Union as part of
the Global Ocean Storage of Anthropogenic Carbon (GOSAC) project, a
regional component of phase 2 of the international Ocean Carbon Cycle
Model Intercomparison Project (OCMIP-2). The HadCM3L ocean GCM is
maintained and developed by the ocean applications branch of the U.K.
Meteorological Office, and the authors are grateful to S. Spall and H. Banks
in particular for their help. A.Y. would like to thank M. Baird, A. Edwards,
and S. Emsley for critical comments and J. House, C. Le Quéré, C. Prentice,
and D. Wallace for stimulating discussions regarding the continental
shelves. The authors are very grateful to two anonymous referees for their
invaluable comments and suggestions.
References
Anderson, T. R., and P. J. le B. Williams, A one-dimensional model of
dissolved organic carbon cycling in the water column incorporating combined biological-photochemical decomposition, Global Biogeochem. Cycles, 13, 337 – 349, 1999.
Banks, H. T., Indonesian Throughflow in a coupled climate model and the
sensitivity of the heat budget and deep overturning, J. Geophys. Res.,
105, 26,135 – 26,150, 2000.
Carlson, C. A., and H. W. Ducklow, Dissolved organic carbon in the upper
ocean of the central equatorial Pacific Ocean, 1992: Daily and finescale
vertical variation, Deep Sea Res. Part II, 42, 639 – 656, 1995.
Chen, C.-T. A., and S.-L. Wang, Carbon, alkalinity, and nutrient budgets on
the East China Sea continental shelf, J. Geophys. Res., 104, 20,675 –
20,686, 1999.
Duarte, C. M., and S. Agusti, The CO2 balance of unproductive aquatic
ecosystems, Science, 281, 234 – 236, 1998.
Duarte, C. M., S. Agusti, P. A. del Giorgio, and J. J. Cole, Regional carbon
imbalances in the oceans: Response, Science, 284, 1735b, 1999.
Esbensen, S. K., and Y. Kushnir, The heat budget of the global ocean: An
atlas based on estimates from surface marine observations, Rep. 19, Clim.
Res. Inst., Oreg. State Univ., Corvallis, 1981.
Fasham, M. J. R., J. L. Sarmiento, R. D. Slater, H. W. Ducklow, and R.
Williams, Ecosystem behavior at Bermuda Station ‘‘S’’ and Ocean
Weather Station ‘‘India’’: A general circulation model and observational
analysis, Global Biogeochem. Cycles, 7, 379 – 415, 1993.
Gent, P. R., and J. C. McWilliams, Isopycnal mixing in ocean circulation
models, J. Phys. Oceanogr., 20, 150 – 155, 1990.
Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns,
J. F. B. Mitchell, and R. A. Wood, The simulation of SST, sea ice extents
and ocean heat transport in a version of the Hadley Centre coupled model
without flux adjustments, Clim. Dyn., 16, 147 – 168, 2000.
Hall, M. M., and H. L. Bryden, Direct estimates and mechanisms of ocean
heat transport, Deep Sea Res., 29, 339 – 359, 1982.
Hellerman, S., and M. Rosenstein, Normal monthly windstress over the world
ocean with error estimates, J. Phys. Oceanogr., 13, 1093 – 1104, 1983.
Jaeger, L., Monatskarten des Niederschlags fur die ganze Erde, Tech.
Rep.139, Ber. Deutsche Wetterdienstes, Offenbach, Germany, 1976.
Josey, S. A., E. C. Kent, and P. K. Taylor, New insights into the ocean heat
budget closure problem from analysis of the SOC air-sea flux climatology, J. Clim., 12, 2856 – 2880, 1999.
Kirchman, D. L., C. Lancelot, M. Fasham, L. Legrende, G. Radach, and
M. Scott, Dissolved organic matter in biogeochemical models of the
ocean, in Toward a Model of Ocean Biogeochemical Processes, edited
by G. T. Evans and M. J. R. Fasham, pp. 209 – 226, Springer-Verlag,
New York, 1993
Kraus, E. B., Diapycnal mixing, in Climate-Ocean Interaction, edited by
M. E. Schlesinger, pp. 269 – 293, Kluwer Acad., Norwell, Mass., 1990.
Kraus, E. B., and J. S. Turner, A one-dimensional model of the seasonal
thermocline, II, The general theory and its consequences, Tellus, 19, 98 –
105, 1967.
Levitus, S., and T. P. Boyer, World Ocean Atlas 1994, vol. 4, Temperature,
NOAA Atlas NESDIS 4, 117 pp.Natl. Oceanic and Atmos. Admin., Silver
Spring, Md., 1994.
Levitus, S., R. Burgett, and T. P. Boyer, World Ocean Atlas 1994, vol. 3,
Salinity, NOAA Atlas NESDIS 3, 99 pp., Natl. Oceanic and Atmos.
Admin., Silver Spring, Md., 1994.
14
YOOL AND FASHAM: CONTINENTAL SHELF PUMP
Longhurst, A., S. Sathyendranath, T. Platt, and C. Caverhill, An estimate of
global primary production in the ocean from satellite radiometer data, J.
Plankton Res., 17, 1245 – 1271, 1995.
Orr, J. C., Ocean Carbon-Cycle Model Intercomparison Project (OCMIP):
Phase 1 (1993 – 1997), IGBP/GAIM Rep. 7, GAIM Office, EOS, Univ. of
N. H., Durham, 1997.
Orr, J. C., R. Najjar, C. L. Sabine, and F. Joos, Abiotic-HOWTO, Internal
OCMIP report, 25 pp., LSCE/CEA Saclay, Gif-sur-Yvette, France, 1999.
Palmer, J. R., and I. J. Totterdell, Production and export in a global ocean
ecosystem model, Deep Sea Res., Part I, 48, 1169 – 1198, 2001.
Raven, J. A., and P. G. Falkowski, Oceanic sinks for atmospheric CO2,
Plant Cell Environ., 22, 741 – 755, 1999.
Redi, M. H., Oceanic isopycnal mixing by coordinate rotation, J. Phys.
Oceanogr., 11, 1443 – 1451, 1982.
Sarmiento, J. L., Ocean carbon cycle, Chem. Eng. News, 30 – 43, May 31,
1993.
Six, K. D., and E. Maier-Reimer, Effect of plankton dynamics on seasonal
carbon fluxes in an ocean general circulation model, Global Biogeochem.
Cycles, 10, 559 – 583, 1996.
Takahashi, T., R. Wanninkhof, R. A. Feely, R. F. Weiss, D. W. Chipman, N.
Bates, J. Olafsson, C. Sabine, and S. C. Sutherland, Net sea-air CO2 flux
over the global oceans: An improved estimate based on the sea-air pCO2
difference, paper presented at 2nd International Symposium on CO2 in
the Oceans Cent. for Global Environ. Res., Tsukuba, Japan, Jan. 18 – 22,
1999.
Taylor, N. K., Seasonal uptake of anthropogenic CO2 in an ocean general
circulation model, Tellus, Ser. B, 47, 145 – 169, 1995.
Toggweiler, J. R., 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, 1989.
Tsunogai, S., S. Watanabe, J. Nakamura, T. Ono, and T. Sato, A preliminary study of carbon system in the East China Sea, J. Oceanogr., 53,
9 – 17, 1997.
Tsunogai, S., S. Watanabe, and T. Sato, Is there a ‘‘continental shelf pump’’
for the absorption of atmospheric CO2, Tellus, Ser. B, 51, 701 – 712, 1999.
Wijffels, S. E., N. Bray, S. Hautala, G. Meyers, and W. Morawitz, The
WOCE Indonesian Throughflow repeat hydrography sections: I10 and
IR6, Int. WOCE Newsl., 24, 25 – 28, 1996.
Williams, P. J. le B., and D. G. Bowers, Regional carbon imbalances in the
oceans, Science, 284, 1735b, 1999.
M. J. R. Fasham and A. Yool, Southampton Oceanography Centre,
European Way, Southampton SO14 3ZH, England, UK. ([email protected].
ac.uk)
(Received October 12, 2000; revised July 18, 2001;
accepted August 2, 2001.)