Deep-Sea Research I 48 (2001) 1169}1198

Production and export in a global ocean ecosystem model

J.R. Palmer , I.J. Totterdell
*
Hadley Centre for Climate Prediction and Research, The Meteorological Ozce, London Road, Bracknell, Berkshire,
RG12 2SY, UK

George Deacon Division, Southampton Oceanography Centre, European Way, Southampton, Hampshire, SO14 3ZH, UK
Received 28 January 1999; received in revised form 23 August 1999; accepted 2 August 2000

Abstract

The Hadley Centre Ocean Carbon Cycle (HadOCC) model is a coupled physical}biogeochemical model of
the ocean carbon cycle. It features an explicit representation of the marine ecosystem, which is assumed to be
limited by nitrogen availability. The biogeochemical compartments are dissolved nutrient, total CO , total

alkalinity, phytoplankton, zooplankton and detritus. The results of the standard simulation are presented.
The annual primary production predicted by the model (47.7 Gt C yr\) compares well to the estimates
made by Longhurst et al. (1995, J. Plankton Res., 17, 1245) and Antoine et al. (1996, Global Biogeochem.
Cycles, 10, 57). The HadOCC model "nds high production in the sub-polar North Paci"c and North Atlantic
Oceans, and around the Antarctic convergence, and low production in the sub-tropical gyres. However in
disagreement with the observations of Longhurst et al. and Antoine et al., the model predicts very high
production in the eastern equatorial Paci"c Ocean. The export #ux of carbon in the model agrees well with
data from deep-water sediment traps. In order to examine the factors controlling production in the ocean,
additional simulations have been run. A nutrient-restoring simulation con"rms that the areas with the
highest primary production are those with the greatest nutrient supply. A reduced wind-stress experiment
demonstrates that the high production found in the equatorial Paci"c is driven by excessive upwelling of
nutrient-rich water. Three further simulations show that nutrient supply at high latitudes, and hence
production there, is sensitive to the parameters and climatological forcings of the mixed layer sub-
model. Crown Copyright  2001 Published by Elsevier Science Ltd. All rights reserved.

Keywords: Global ocean; Carbon cycle; Ecosystem modelling; Primary production; Nitrogen cycle; Particulate #ux

1. Introduction

The Hadley Centre Ocean Carbon Cycle (HadOCC) model has been developed for use in global
ocean carbon cycle modelling. Our knowledge of both the natural marine carbon cycle and its

* Corresponding author. Tel.: #44-23-80596342; fax: #44-23-80596247.

E-mail address: Ian.J.Totterdell@soc.soton.ac.uk (I.J. Totterdell).

0967-0637/01/$ - see front matter Crown Copyright  2001 Published by Elsevier Science Ltd. All rights reserved.
PII: S 0 9 6 7 - 0 6 3 7 ( 0 0 ) 0 0 0 8 0 - 7
1170 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

response to perturbation is still far from complete, but it is widely accepted that the oceans are
taking up around 2 Gigatonnes (Gt) of the anthropogenic carbon; see Schimel et al. (1995) for an
overview. It is necessary to use models in order to better understand the processes which transport
the carbon into the deep ocean and to assess the possible changes in the natural cycle due to global
warming. The earliest models were box models (e.g. Siegenthaler and Joos, 1992) but with
increasing computing power and better parametrisations of physical and biogeochemical processes
a number of studies were made using ocean general circulation models (e.g. Sarmiento et al., 1992;
Maier-Reimer, 1993). The "rst such models had at best a simple parametrisation of the biologically
mediated processes: basic estimations of the uptake of anthropogenic CO could be made but it

was not possible to accurately assess the strengths of any carbon cycle feedbacks which involve
climate change and the ocean biology. Sarmiento and Le QueH reH (1996) found in a simulation of the
global carbon cycle for the next 100 years that changes in the amount and geographical pattern of
marine primary production, the amount of carbon "xed by phytoplankton during photosynthetic
growth, could have signi"cant e!ects. Six and Maier-Reimer (1996) presented the "rst ocean
carbon cycle model that featured an explicit representation of the marine ecosystem. This paper
describes a new model which has been developed for both ocean-only and coupled ocean}
atmosphere}vegetation carbon cycle studies.
The HadOCC model is based on the Cox (1984) type ocean general circulation model used at the
UK Meteorological O$ce (including the Hadley Centre). The model simulates the essential aspects
of carbonate chemistry and biological production and export. Several extra tracers are carried to
model the carbon cycle. These are dissolved inorganic carbon (DIC), total alkalinity, a nitrogenous
nutrient, phytoplankton, zooplankton and sinking detritus. The model can be divided conceptually
into separate inorganic and biological parts. Inorganic carbonate chemistry and partial pressure
physics are well understood and can be reproduced with fair accuracy even in a simple carbon cycle
model. The export of biologically generated soft tissue (organic matter) and hard tissue (carbonate)
to the deep ocean, collectively known as the biological pump, is much more di$cult to model. There
are many direct measurements of total primary productivity made using C incubation methods,
as well as indirect estimates based on remotely sensed surface chlorophyll concentrations, to which
model predictions can be compared. But it is not total primary production that is the key biological
driving "eld for an ocean carbon cycle model. Nutrients can be cycled through the food web many
times before being lost to the deep ocean, and it is the "nal export of carbon and nutrients to deeper
waters, largely as a sinking particle #ux, that determines the circulation of carbon and nutrients on
timescales of decades to centuries and that must be modelled.
The requirement for the HadOCC model to run century-scale carbon cycle simulations, some-
times coupled to an atmospheric model, means that it has to be extremely simple in order not to use
excessive computer resources. A simple model can also be more easily analysed, which is a useful
feature if the model shows unexpected behaviour as part of a coupled system. We also believe that
a simple model, properly constructed, is less likely to respond in an unreasonable way to changes in
climatological forcings (for example). We made the deliberate choice during the development of the
model to include only certain major processes, as described in the following sections and in the
appendix. We have not included processes such as nitrogen "xation since, having only an
incomplete understanding of the controlling factors, we cannot guarantee that any parametrisation
we use will not lead to an unreasonable response of the model to climate change in the scenario
runs. The model is not intended to reproduce the patterns of primary production in particular
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1171

regions in particular years, as it seems unlikely that the long-term operation of the global ocean
carbon cycle is sensitive to processes at such time and space scales. However, in a future study we
intend to force the model with climatological "elds featuring realistic interannual variability and
examine the resulting variability in the primary productivity "eld. Planned future extensions to the
basic HadOCC model are considered in Section 7.
This paper describes the physical, chemical and ecosystem components of HadOCC ocean
carbon cycle model before going on to look at the modelled "eld of primary production in the
annual mean. This is compared to various global estimates and the CZCS-derived "elds described
in Longhurst et al. (1995), and kindly supplied in electronic form by the authors. Export production
is discussed and compared to sediment trap results. In this study we do not examine the long
timescale changes to the concentrations of nutrient (nitrate) or DIC in the deep ocean caused by the
export #ux, but are concerned only with the near-surface, short timescale processes. The physical
ocean controls of productivity are investigated by comparing HadOCC to a simple nutrient-
resetting biology model. The sensitivities of primary production to windstress and the mixed layer
model are then studied.

2. Model description

2.1. Ocean GCM

The UK Met. O$ce implementation of the primitive equation model described by Bryan (1969)
and Cox (1984) is global with a realistic approximation to continental coastlines and bottom
topography. The ocean model con"guration used for this study is similar to that for the Hadley
Centre coupled model described in Johns et al. (1997) (however it should be noted that in the
current study the separate and interactive sea-ice sub-model is not present). The horizontal grid is
regular with a resolution of 2.53 latitude and 3.753 longitude. There are 20 levels in the vertical with
thicknesses ranging from about 10 m for each of the top 4 levels to 615.3 m for the bottom 7 levels.
The model is run with a tracer timestep of 24 h. Tracers are di!used along isopycnals with
a di!usion coe$cient of 2000 m s\. Use of the eddy parametrisation scheme of Gent and
McWilliams (1990) with a thickness di!usion coe$cient of 2000 m s\ allows us to use no explicit
horizontal di!usion. Vertical di!usion and eddy viscosity coe$cients are set using the Richardson
number scheme of Pacanowski and Philander (1981), with the background vertical di!usion
coe$cient ranging from 0.1 cm s\ at the surface to 1.5 cm s\ at 5000 m, from the observation-
ally based estimates of Kraus (1990). In the mixed layer the vertical di!usion coe$cient is set using
a simpli"ed version of the scheme of Large et al. (1994). Tracers are also mixed near the surface
using the bulk mixed layer scheme of Kraus and Turner (1967).
Water columns that become gravitationally unstable, for example because of wintertime surface
cooling, are fully convected at the end of each timestep using the scheme of Rahmstorf (1993). The
ocean surface is forced with the monthly mean climatological heat #uxes of Esbensen and Kushnir
(1981), windstresses of Hellerman and Rosenstein (1983) and the freshwater #uxes of Jaeger (1976)
and Esbensen and Kushnir for precipitation and evaporation, respectively. In addition to the
prescribed heat and salt #uxes, surface values of temperature and salinity are relaxed back towards
the monthly mean climatological "elds from the atlases of Levitus and Boyer (1994) and Levitus
1172 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

et al. (1994), respectively, with a relaxation timescale for a 50 m mixed layer of about 60 days. It
should be noted that the relaxation freshwater #uxes are about an order of magnitude larger than
the imposed precipitation and evaporation #uxes. In the experiments described in this paper
a simple scheme was used over sea-ice areas in which surface temperatures are not allowed to fall
below !1.83C. Ice cover is assumed at this temperature and air}sea #uxes of heat, salt and CO

are blocked.

2.2. Carbon cycle model

The carbon cycle model embedded in the ocean GCM includes a nutrient}phytoplankton}
zooplankton}detritus (NPZD) model. The NPZD model uses nitrogen as the limiting nutrient, and
the #ows of carbon are calculated using "xed stoichiometric ratios. This section gives a brief
overview of the NPZD model and the carbon chemistry; a more detailed description is given in the
appendix. The model compartments or state variables are nutrient, phytoplankton, zooplankton,
detritus, dissolved inorganic carbon (DIC) and total alkalinity. The "rst four compartments are
considered in terms of their nitrogen contents. The carbon contents of the phytoplankton,
zooplankton and detritus are related to the nitrogen contents by "xed carbon:nitrogen ratios. All
the model compartments are advected, di!used and mixed as oceanic tracers, and exist at all the
model levels, although the concentrations of phytoplankton and zooplankton decrease rapidly
with depth below the euphotic zone. The biological processes modelled are primary production,
natural mortality, grazing, egestion, respiration and the sinking and remineralisation of detritus.
The photosynthetically available radiation (PAR) needed for primary production is a "xed fraction
of the Esbensen and Kushnir (1981) short-wave #ux. Nutrient is the sum of nitrate and ammonium,
and its only sources and sinks are due to biological processes, i.e., there is no riverine input or
sedimentation, and nitri"cation and denitri"cation processes are not considered. Phytoplankton
are responsible for primary production, the biomass-speci"c rate of which depends on the
availability of nutrient and light and on the temperature. Detritus is the only model compartment
which sinks, and so is the mechanism for most of the downward transport of carbon. As well as the
organic carbon `soft tissuea sinking #ux, which is modelled explicitly, the export of biogenic
calcium carbonate (the `hard tissuea #ux) is also represented as an instantaneous redistribution of
alkalinity and carbon at depth.

3. Experiments

Several model experiments are described in this paper. In all cases the ocean temperature,
salinity and nutrient (nitrate) "elds were initialised from the atlases of Levitus and Boyer (1994),
Levitus et al. (1994) and Conkright et al. (1994), respectively. Phytoplankton, zooplankton and
detritus "elds were initialised with values of 1;10\
mol l\. In the standard experiment (S) the
model was driven for 21 yr using monthly mean climatological forcings.
The phytoplankton, zooplankton and detritus "elds settled down to stable repeating cycles
within a few years with the global phytoplankton biomass averaging 0.98, 0.99 and 1.00 Gt C in
years 5, 10 and 20, respectively, responding to slow multidecadal drifts in near-surface nutrient
concentrations in some regions. In the standard experiment S the gross primary production (total
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1173

Table 1
Experiments

Code Experiment

S Standard. Climatological forcing, HadOCC biology

W Global windstress;0.5
M1 Global wind mixing energy;1.5
M2 Buoyancy decay depth increased to 300 m
M3 Buoyancy decay depth increased to 1000 m
R Nutrient-restore biology

plankton growth measured in mass of "xed carbon, without subtracting respiration) rose from
41.8 Gt C yr\ in the "rst year of spinup to 47.6 Gt C yr\ in year 20. The drift from year 20 to 21
was a 0.1% increase globally, though locally there were changes of up to 2%. The areas with
increasing production were between 20 and 403S in the Indian and western Atlantic Ocean, while
those with decreasing production were the eastern Atlantic between 20 and 403N and the eastern
Paci"c between 10 and 403N. Long-term re-distributions of the nutrient "eld by the ocean
circulation or biological activity are not discussed here, because the aim is to validate the primary
production that is driven by a nutrient distribution close to the observations.
The results discussed in this paper are annual means from year 21. In addition to the standard
run S, experiments were performed to test the sensitivity of biological production to various aspects
of the ocean model. The ocean surface windstress was halved over the entire globe in experiment W.
The mixed layer depth was increased in experiment M1 by globally increasing the wind mixing
energy and in experiments M2 and M3 by deepening the penetration depth scale of the surface
buoyancy #ux. A much simpli"ed biology scheme was used in experiment R in which the nutrient
"eld was restored towards a reference "eld. Table 1 summarises the experiments.

4. Validation of biological production in HadOCC

4.1. Global annual mean primary production

The earliest estimates of global ocean annual total primary production were based on oxygen
evolution measurements from incubations (for example Riley, 1944). However it was the develop-
ment of the C incubation technique by Steemann Nielsen (1952) that triggered a host of global
production estimates throughout the 1960s and 1970s. In the same study Steemann Nielsen himself
used data from the Galathea expedition to obtain a value of 15 Gt C yr\ for global primary
production (net of respiration). The compilation by Berger et al. (1987) contained about 8000 data
points and yielded a global total of 26.9 Gt C yr\. Sundquist (1985) lists a collection of global
primary production estimates based on C data, which are well spread between 20 and
45 Gt C yr\, with one estimate of 60 to 80 Gt C yr\.
The ship-based sample approach to estimating global primary production is limited by sparse
geographical and seasonal coverage. Large areas, such as the Southern Ocean, remain relatively
1174 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

unsampled, especially in wintertime. The availability of satellite measured ocean colour data for the
years 1978}1986 (Feldman et al., 1989) has allowed researchers to extrapolate in situ measurements
of chlorophyll, light levels, and their relationships with primary production to all areas of the ocean
and all seasons. Although seasonal illumination at high latitudes and regions of persistent
cloudiness restrict the data set so that it is not fully comprehensive, the CZCS instrument has
provided the "rst global view of plant production in the oceans.
Based on sea-surface chlorophyll concentrations derived from CZCS data, Antoine et al. (1996)
combined a photosynthesis light model and illumination data to derive a seasonal global distribu-
tion of primary production. Their computed global total production ranges from 36.5 to
45.6 Gt C yr\, depending on the assumptions made. A similar study of primary production was
made by Longhurst et al. (1995) who characterised regions of the ocean using 57 biogeochemical
provinces, de"ned according to their physical oceanography and chlorophyll "elds. Their estimate
of total primary production in all oceans is 44.7}50.2 Gt C yr\. More recently, Field et al. (1998)
estimated total oceanic primary production to be 48.5 Gt C yr\ using CZCS data and the
Vertically Generalised Production Model of Behrenfeld and Falkowski (1997). These estimates are
all on the high side of those summarised by Sundquist (1985).
Ocean biogeochemistry models have also been used to derive information about global produc-
tion rates. Using a global ocean circulation model in which a phosphate tracer was continually
forced towards an observed surface distribution, Najjar et al. (1992) estimated the new production
(i.e. fuelled by nitrogen not recycled within the euphotic zone) to be 12}15 Gt C yr\. They used
various depth pro"les for the remineralisation rates of sinking particulate organic matter and
dissolved organic matter. Translating new production into total primary production is not easy,
but assuming the typical export ratio from the HadOCC model of 15% a global total primary
productivity of 80}100 Gt C yr\ is found; alternatively the Six and Maier-Reimer (1996) export
ratio of around 25% leads to a value of 50}60 Gt C yr\.
The more complex global model of Six and Maier-Reimer (1996) includes explicit representa-
tions of phosphate, phytoplankton, zooplankton, particulate and dissolved organic matter, but
resolves only two layers in the euphotic zone, down to 50 and 100 m depth. This model gives
a global total primary production, net of respiration, of 43.6 Gt C yr\, of which 11.1 Gt C yr\
is new production. The con"dence for this type of model in estimating global new production is
based partly on its ability to reproduce observed distributions of biologically active tracers (oxygen,
nutrients, carbon) in the oceans. For example, faithful reproduction of the global mean phosphate
depth pro"le indicates a good balance between export production and physical transport pro-
cesses. Con"dence in primary production estimates from a model validated principally using
biologically active tracers may be low however because the ratio new to total primary production,
which is undoubtedly highly variable both spatially and temporally, is not known accurately.
The global total primary production calculated in the HadOCC model standard simulation run
S is 47.7 Gt C yr\. The sensitivity of the primary production to various physical parameters will
be considered later, but the values calculated in the di!erent experiments range from
36.1 Gt C yr\ (reduced windstress experiment W) to 50.7 Gt C yr\ (mixed layer experiment
M3). These values agree well with the observations and the results of the other models discussed.
The HadOCC model values are for gross production however and if the global #ux due to
phytoplankton respiration is subtracted the resulting net primary production is found to be
40.5 Gt C yr\. This is slightly lower than the estimates based on CZCS data. A signi"cant
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1175

Fig. 1. Annual mean primary production net of respiration is shown for the HadOCC model (top) and the Longhurst et
al. (1995) dataset derived from CZCS ocean colour data.
1176 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Fig. 2. Annual mean primary production for the HadOCC model is shown as zonal means globally, for the Atlantic and
for the Paci"c. The LSPC95 dataset is shown as a dashed line. Atlantic and Paci"c HadOCC curves are shown as "ne
lines in the global plot.

proportion of the production calculated by considering remotely sensed chlorophyll "elds is found
adjacent to the continents, however. The ability of the HadOCC model to represent such high
productivities depends on the processes causing them, and is discussed in the following section.
It is di$cult to separate the HadOCC model's value for the total primary production into
separate new and regenerated parts. HadOCC features a single compartment for nutrient, so it is
not possible to make the distinction between production fuelled by new and by recycled nitrogen. If
the sinking #ux of detritus is equated to the part of the production that is exported and that is in
turn equated to new production over an annual cycle (see Section 4.3) then it is possible to estimate
the annual new production in the top seven levels (i.e. down to 113 m) to be 9.4 Gt C yr\.

4.2. Regional primary production

The HadOCC model predicts a global annual primary production that is consistent with other
estimates, but how well does it reproduce the geographical patterns of production? Fig. 1 compares
the annual primary production produced by HadOCC with that determined by Longhurst et al.
(1995; hereafter referred to as LSPC95) from CZCS data. A visual inspection of the two plots
indicates that many of the general features apparent in the satellite-derived plot are also present in
the HadOCC plot. In both plots there are zonal bands of high production around 453S and 453N,
with another at the equator. The subtropical gyres are regions of low productivity. These features are
also clear in Fig. 2, which shows the zonally integrated primary production calculated by HadOCC
and LSPC95. Note that all productivity values given in this section are for total production.
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1177

However there are signi"cant di!erences between the two plots in Fig. 1. Much of the production
in LSPC95 is located very close to coastlines or islands, whereas that in HadOCC is smoothly
spread over a larger area. The high coastal production seen in LSPC95 is mainly on the continental
shelves where the sea #oor is at most a few hundred metres deep and often much shallower.
Nutrients which are released from the sediments can be quickly returned to the euphotic zone and
can fuel a signi"cant proportion of total production (for example, see Davies, 1975; Rowe et al.,
1975). Also, the maximum depth of mixing is limited by the shallow sea #oor. HadOCC, with
a coarse resolution, features only the larger continental shelves, and along most coastlines land
points are directly adjacent to water 1000 m deep or more. Therefore, for the same input of
nutrients from the deep ocean HadOCC will show a lower total production. A second reason is
that HadOCC does not have any inputs of nutrient to the ocean from the land via rivers or other
run-o! processes. This potentially signi"cant source of nutrient for coastal ecosystems is therefore
absent from HadOCC, and in such cases the model will under-predict the production. Lastly, much
of the nutrient supplied to the coastal zone in the real ocean is upwelled. HadOCC cannot
reproduce this localised process accurately. Instead the upwelling is spread over several grid points,
in turn causing the production to be spread out geographically.
For example, in the north-west Indian Ocean LSPC95 shows the production to be concentrated
along the coast of the Arabian Peninsula, where there is a strong supply of nutrient seasonally
upwelled by the monsoonal circulation. HadOCC shows productivity evenly spread over the whole
northwest Indian Ocean, though the rate is less than the highest values found by LSPC95. The
total model annual production in a box west of Sri Lanka and down to 153S is 2.7 Gt C, close to the
3.1 Gt C in the dataset of LSPC95. However, it is also possible that HadOCC will over estimate the
production in such cases, and the Benguela upwelling system is an example: again the model
spreads the high productivity over a larger area than the satellite observations indicate, but as Fig.
2 shows the zonal integral of total production across the Atlantic predicted by HadOCC is up to
50% higher than the "gure from LSPC95 between 30 and 103S. The "nding is therefore that in
most circumstances the model underestimates coastal primary production, but in some examples of
systems driven by coastal upwelling the model will overestimate the total production over a larger area.
There can be quantitative di!erences even where the geographical distribution is comparable.
The most noticeable is that HadOCC predicts a rate of production of more than 0.4 g C m\
day\ over a large area of the eastern and central equatorial Paci"c, with a peak of over 0.8 g C m\
day\. In contrast, LSPC95 shows the highest rates of production to be below 0.4 g C m\ day\
(except adjacent to the coast and around the Galapagos Islands), with much of the region below
0.2 g C m\ day\. The annual production integrated between 203S and 203N in the Paci"c is
given by HadOCC as 11.7 Gt C and by LSPC95 as 7.5 Gt C. It will be shown later that the primary
production in the equatorial Paci"c is fuelled by upwelled nutrient in HadOCC. This suggests that
one cause of the unrealistically high productivity is excessive upwelling. This hypothesis will be
examined in Section 6.1.
HadOCC and LSPC95 also disagree north of 503N in both the Paci"c and Atlantic Oceans.
HadOCC shows that the production decreases from between 0.4 and 0.5 g C m\ day\ around
453N to between 0.3 and 0.4 g C m\ day\ north of 603N, but LSPC95 shows an increase from
the roughly the same "gure at 453N to around 0.8 g C m\ day\. Yoder et al. (1993) analysed the
CZCS data and suspected that the chlorophyll values were overestimated polewards of 403. They
developed a `correctiona procedure using in situ chlorophyll concentrations. It is notable that the
1178 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

production study of Antoine et al. (1996), also based on the CZCS data but employing a correction
procedure similar to that of Yoder et al. (1993), shows productivity values of less than
0.3 g C m\ day\ in the Atlantic north of 603N. This "gure is obviously closer to the HadOCC
prediction.
Another factor that may be responsible for the disagreement in both the North Paci"c and the
North Atlantic is the simple ice model in HadOCC which has no impact on the salinity distribution
through ice formation and melting. Therefore the stable strati"cation of low salinity layers of melt
water which assist plankton growth in many high latitude areas of the real ocean are not simulated
in HadOCC, and the model's productivity will be lower in those areas.
HadOCC predicts the band of productivity at latitude 453S to be greater than
0.4 g C m\ day\ all around the world, but in LSPC95 the production is substantially lower.
There are abundant nutrients in the surface waters at that latitude both in HadOCC and in the real
ocean (Conkright et al., 1994), so it is unlikely that the physical supply of nutrients in the model is
causing the error; since the half-saturation constant for nutrient uptake has the value
0.1 mmol Nutrient m\ the rate of uptake and production should not be signi"cantly di!erent for
nutrient concentrations of 10 or 20 mmol Nutrient m\. It is possible that in the Southern Ocean
the simpli"cations made in the biological model, that the primary producers are homogeneous and
limited only by light and nitrogenous nutrient, are not fully valid. It is observed that north of the
Antarctic Polar Front the growth of diatoms, which contribute much of the primary production in
other nutrient-replete areas, seasonally depletes silicate in the surface waters, preventing their
further growth (QueH guiner et al., 1997). There is also increasing evidence that the low concentra-
tions of dissolved iron in the surface waters of the Southern Ocean (and the equatorial Paci"c) limit
the growth-rates of at least some of the phytoplankton species present (Martin et al., 1990). Neither
of these factors are represented in HadOCC, and may be reasons why the productivity observed is
lower than that predicted by HadOCC in this region.

4.3. Export production

Export production, that part of the total primary production which is transported to below the
euphotic zone, is quantitatively equal to new production (that fuelled by nitrogen new to the
euphotic zone), at least if considered over several annual cycles. The nitrogen newly introduced to
the euphotic zone must balance that lost to below it if the concentration is not to change
continuously.
It is thought that the sinking of particles is the principal export mechanism; it is certainly the
easiest to measure, and many estimates of the export production consider it exclusively. Only a few
studies have attempted to quantify other export mechanisms. Bacastow and Maier-Reimer (1991)
and Najjar et al. (1992) examined the transport of dissolved organic matter (DOM) out of the
euphotic zone using GCMs. The former study found that the dissolved export was twice that of the
sinking particles, while the latter needed the dissolved transport to dominate in order to match
nutrient #ux data and avoid nutrient trapping. It should be noted however that both models aimed
to simulate the high DOM concentrations observed by Sugimura and Suzuki (1988), which were
around 300
mol C (kg seawater)\ in surface waters. More recent measurements show that the
concentration is approximately 70}80
mol C (kg seawater)\ and it is reasonable to expect that
models which simulate this lower value would "nd DOM transport to be less important in export
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1179

production. Yamanaka and Tajika (1997) used a GCM to diagnose the #uxes which would "t the
recent DOC observations, and found that at 100 m depth the annual downwards #ux of carbon
was 8 Gt in particles and only 3 Gt in the dissolved fraction, while below 400 m the export was
mediated exclusively by sinking particles. Another export mechanism that has been studied is
transport by zooplankton which practise diel vertical migration (Angel, 1989). Organic carbon
consumed at the surface may be carried to a depth of several hundred metres before it is egested or
respired. Alternatively the organism itself may be eaten below the euphotic zone. Longhurst et al.
(1990) found the #ux of respiratory carbon to be 13}58% of the particulate sinking #ux.
The HadOCC model does not currently feature DOM, nor do the zooplankton migrate, so the
only signi"cant export mechanism is the sinking of detritus (downwelling and downward mixing of
phytoplankton, zooplankton and detritus are negligible by comparison). We calculate the export
ratio in HadOCC as the detrital sinking #ux through the model level centred at 139 m depth
divided by the total surface primary production. Fig. 3 shows the annual mean ratios for all grid
points. Most values lie in the range 13}17% and the global average is 14.9%.
There have been only a few experimental determinations of the export ratio. Pace et al. (1987)
used shallow sediment traps during the VERTEX programme in the northeast Paci"c and reported
new production to be 13}25% of total production, implying an identical range for the export ratio.
Lohrenz et al. (1992) found the export ratio to vary from 4 to 14% at the US JGOFS Bermuda
Atlantic Time series Study (BATS) site; again, this was determined using shallow sediment traps.
Our calculation of the ratio agrees with both these ranges. There have also been attempts to
determine the f-ratio (the ratio of new to total production, Eppley, 1981) by measuring nitrate and
ammonium uptake using N. Harrison et al. (1987) analysed eight such studies and reported
f-ratios varying over the whole range from zero to one. All the data sets considered were coastal
however and most of the values reported from oceanic waters (see also Lewis et al., 1986) were
below 0.4. Assuming the f-ratio can be equated to the export ratio in the circumstances of the
experiments, the values are somewhat higher than our calculations and the estimates using
sediment traps. It is possible that the a proportion of the production in the real ocean is being
exported from the euphotic zone by the other mechanisms previously discussed, which HadOCC is
unable to represent. Six and Maier-Reimer (1996) calculated new production and total production
in their model to be 11.1 and 43.6 Gt C yr\ respectively, implying an export ratio of 25.5%. Of
the production exported, 8.8 Gt C yr\ was in the form of sinking particulates so if the export
ratio was calculated only in terms of that it would be 20.2%. These "gures, calculated at 100
m depth, are in approximate agreement with those from the HadOCC model.
Deeper in the water column, Betzer et al. (1984) found that the dry weight #ux at 900 m depth
was 8.8}10.4% of the mass of carbon "xed in primary production. The global sinking #ux in level
13 of the HadOCC model (centred at 995.55 m) is 2.8% of the global annual production. For 2000
m depth, it can be inferred from Fig. 1 of Suess (1980) that the #ux/production ratio is 4}5%, while
Pace et al. (1987) "nd the ratio to be 1}1.5% and the compilation of Lampitt and Antia (1997)
indicates a value between 0.5 and 2.5%. The #ux in level 15 of HadOCC (centred at 2116.15
m depth) is 0.6% of the production, in good agreement with the data.
Fig. 3 shows the variation of the export ratio with total production. The "gures refer to annual
means calculated in the HadOCC model. For annual production less than about 50 g C m\ yr\
the ratio increases with increasing production, but above that the ratio slowly decreases. This result
disagrees with most other studies. Eppley and Peterson (1979) found that the percentage of
1180 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Fig. 3. The annual mean particulate carbon #ux across 139 m depth is divided by the primary production to yield the
export ratio at each grid point, which is plotted against the primary production.

production that was new (i.e. the export ratio) increased approximately in proportion to produc-
tion up to almost 50% when the production was 200 g C m\ yr\, and then remained around
that "gure for higher rates of production. This study was based on N uptake experiments, similar
to those analysed by Harrison et al. (1987) which gave very high f-ratios. Several authors have
analysed sediment trap data and "tted functions of the form Flux"Az?P@ where z is the depth and
P is the total production. The #ux is proportional to the production if b"1, in which case the
export ratio (equal to Flux/P) is independent of production. Betzer et al. (1984) considered dry
weight (rather than organic carbon) #uxes from deep sediment traps and found b"1, but their
re-analysis of the data from Suess (1980) showed b"1.41$0.15. The VERTEX trap data reported
by Pace et al. (1987) showed a high degree of scatter, but was consistent with b"1. However, if the
four sites where the euphotic zone was only 50 m deep are excluded, the remaining "ve sites give
b"0 over the range 200}800 mg C m\ day\. This implies an export ratio that decreases with
increasing production, in agreement with the HadOCC results. Berger and Wefer (1990) found that
b"1.5 "tted their trap data. Lohrenz et al. (1992) examined 150 m sediment trap data from the
BATS site and found b"0.88, so the export ratio decreased as production increased, in agreement
with HadOCC. However, the BATS data considered are not annual totals from di!erent sites, but
instead consecutive monthly totals from the same site, and some of the export measured in any
given month may relate to primary production in the previous month. It is not obvious that the
value of b found in this study where the data are separated in time can be compared to other values
obtained from data which are spatially separated. We consider in Section 7 reasons why the range
of export ratios in the HadOCC model is so low, and why the response of the ratio to increased
production disagrees with many experimental studies.

5. Physical controls of export production

To what extent does the physical ocean model control primary production and export pro-
duction, and which mechanisms are important? Clearly the physical and biological aspects of
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1181

Fig. 4. Processes acting on the euphotic zone nutrient in the HadOCC model are plotted (annual means) as zonal means.
The biology curve shows uptake by phytoplankton for growth minus the sum of the nutrient recycling processes
(including remineralisation) so a positive value shows a net export of nutrient, mainly as sinking detritus. The biological
export is balanced by advection, di!usion and mixing of dissolved nutrient.

HadOCC are closely coupled. Biological uptake generates the vertical nutrient gradients that
control di!usive and mixed layer model nutrient supply. Export of nutrient as detritus impacts on
the nutrient concentration in upwelling waters. Here we demonstrate that changes to upwelling
and the mixed layer model have particular consequences for the export production. We begin by
analysing the mechanisms responsible for nutrient supply in the HadOCC model.

5.1. Nutrient supply processes in the HadOCC model

In order to understand the balance between the physical and biological models in controlling
export production, we need to assess which processes in the model dominate nutrient supply to the
euphotic zone. Fig. 4 shows zonal totals of nitrate #uxes to the euphotic zone due to various
processes. We have de"ned the euphotic zone to be the top 7 model levels, down to a depth of 113
m; this de"nition yields the maximum global export production. Changes due to advection,
di!usion and mixing processes are shown as positive where they supply nitrate to the euphotic
zone. The advection and di!usion curves include both horizontal and vertical processes. Changes
due to the Gent and McWilliams (1990) eddy parametrisation scheme are included in the di!usion
curve. The mixing curve is dominated by the bulk mixed layer model, but includes a small
contribution from convective adjustment. The curve labelled `biologya shows the net removal of
nutrient from the water by biological processes, i.e. uptake by phytoplankton for growth minus the
sum of remineralisation, excretion, etc. Net removal is therefore a positive quantity. As the #ux due
to the sinking of detritus is much larger than the other transport #uxes for the biological
compartments, the #ux shown by this curve is approximately equal to the export production. The
sum of the advection, di!usion and mixed layer curves minus the biology curve gives the net
accumulation of nutrient in the euphotic zone at each latitude over the relevant year of the model
simulation. The average and maximum values of the net accumulation are, respectively, !0.001
1182 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

and !0.066 Gmol NO\ 3latitude\ day\, much smaller than any of the process curves, and so

the net accumulation rate is not shown on the "gure.
It is clear that the nutrient utilised by the phytoplankton is supplied to the euphotic zone by
advection at low latitudes and by entrainment into the mixed layer at high latitudes. The high
primary production apparent in the equatorial Paci"c in Fig. 2 is driven by nutrient advection.
Analysis of the component vertical and horizontal terms show this supply to be made up of strong
upwelling along the equator with subsequent horizontal #ow out from the equator at the surface.
The data of Longhurst et al. (1995) do not show the production in this region to be as high as the
model predicts (see Fig. 1), and in the following section we investigate if the di!erence is due to
excessive equatorial upwelling in HadOCC. Since the upwelling advective #ux of nitrate depends
on the nitrate concentration below the euphotic zone and is independent of the surface depletion
rate, one might expect the equatorial production rate to be more dependent on the physical ocean
model than on the biological model. This is, of course, a simpli"cation. In the real ocean, plankton
growth at the equator is observed to be weak despite relatively high nutrient concentrations, and
nutrient may be lost from the euphotic zone by Ekman pumping and subduction before it has been
used up in growth. Furthermore, high rates of export production result in high rates of re-
mineralisation in the water masses that are upwelling, so that growth rates and upwelling #uxes are
not uncoupled.
The supply of nitrate to the biology at high latitudes is dominated by mixing processes. This is in
agreement with the classical understanding of nutrient supply in the North Atlantic and Paci"c,
where deeper water with a high nitrate concentration is entrained into the mixed layer, predomi-
nantly on a seasonal cycle. This conceptual scheme is represented in many models, e.g. Fasham et
al. (1990) and Evans and Parslow (1985). In contrast to advective processes, supply of nitrate to the
euphotic zone is sensitive to the concentrations in the surface waters. The faster the biology uses up
nitrate, the faster mixing processes can replenish it because the vertical gradient of nitrate
concentration is enhanced. Thus we expect the export production rates at high latitudes to be much
more sensitive to the biology model than those at lower latitudes where upwelling processes
dominate supply. Polewards of the subtropical gyres, however, nutrients are never, or only
seasonally, limiting so that it is in the oligotrophic subtropics that the mixed layer model is likely to
exert most control.

5.2. Nutrient supply processes in a nitrate-restore model

In order to assess more clearly the role of the physical ocean in determining rates of export
production we reduced the representation of the biology so that it responds linearly to the nutrient
supplied by the ocean circulation. This simpli"ed model is hereafter referred to as the nitrate-
restore model. The ocean biogeochemistry model of Najjar et al. (1992) parametrised the biological
pump by a simple restoring of the ocean model phosphate distribution to that observed in the top
119.3 m of the ocean. A timescale of 30 days was used to de"ne the strength of the restoring term.
They used this biological model to study parametrisations of the downward transport of export
production inferred from the restoring term.
We use a similar method to diagnose the export production implied by combining the observed
global nitrate distribution (Conkright et al., 1994) and our model ocean circulation. We concep-
tually divide nitrate transport within the ocean into two components; tracer transport by physical
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1183

Fig. 5. As for Fig. 4, processes acting on the euphotic zone nutrient are plotted as annual zonal means. But in this
experiment, a simple nutrient-restoring biology (expt R) has been used in place of the HadOCC biology.

oceanic processes, and all biologically mediated transport including primary and subsequent
export production by sinking particles. Biological processes which have been suggested as impor-
tant but which are not included in the HadOCC model (e.g. the limitation of growth rate by the
lack of iron, the production and export of dissolved organic matter, denitri"cation etc.) are
implicitly present in the biological pump inferred by the nitrate-restore model. The physical
transport and biological pump act together to maintain the nitrate distribution in the model close
to the observed, initial distribution.
Whereas Najjar et al. (1992) used only a surface restoring term, in order to drive a parametrised
sinking #ux, we use a restoring term throughout the water column, and at all surface points.
Biological depletion of nitrate occurs very quickly in the euphotic zone, on timescales of days to
weeks, whereas the same nitrate exported to the deep ocean and remineralising there requires
decades to centuries to a!ect the tracer distributions. To account for these di!ering timescales we
use a restoring term with a timescale of 60 days in the top 164 m (top 8 model levels), increasing to
100 yr below 3654 m (bottom 3 levels). The restoring term at any gridpoint is then proportional
to the deviation of the local nitrate concentration from the climotology and inversely proportional
to the timescale at that depth. The restoring of nitrate at depth gives information about the implied
pro"le of remineralisation, but in this study it simply serves to keep the deep ocean nitrate close to
that of the initial distribution. We recognise that the explicit NPZD biology model can respond to
changes in the physical transport of nutrient on much shorter timescales than the restoring model
(a few days and 60 days, respectively); however because the export ratio is very insensitive to the
rate of production in the HadOCC model the annually averaged "gures which we consider here
will not be signi"cantly a!ected.
The annual mean nitrate distribution in the euphotic zone after the nitrate-restore model has
been run for 20 yr is shown in Fig. 6. The results have been averaged zonally, vertically over the top
113 m of the ocean and in time over year 21 of the experiment. The restored nitrate, shown as
a dashed line, is slightly higher everywhere than the initial distribution which is shown as a solid
1184 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Fig. 6. The zonal, annual mean euphotic zone nutrient is shown for the observed distribution (Conkright et al., 1994),
HadOCC experiment S and the nutrient-restoring experiment R.

line. The export production in the nitrate-restore model is assessed by looking at the nitrate source
and sink terms in the top 7 model levels (top 113 m). Fig. 5 shows the source and sink terms in the
same way as the HadOCC model terms were shown in Fig. 4, so a net removal of nutrient by the
inferred biology is a positive quantity. The similarities between Figs. 4 and 5 are remarkable. Both
show an equatorial peak in the biology term, which is the export production from the top 7 model
levels, of about 5.5 Gmol of nitrate per degree of latitude per day, and a lower peak of about 2.5 at
mid-latitudes in the northern hemisphere. The peak corresponding to production in the Southern
Ocean is around 6 Gmol NO\ 3lat\ day\ and is slightly higher in the nitrate-restore model than

in the HadOCC model.
The mechanisms of nutrient supply are very similar in both experiments, with the equatorial
production peak almost entirely supplied by advection, and the high latitudes supplied largely by
mixing except for a substantial advective component at 453S. The surface nitrate concentrations
are a useful additional diagnostic in understanding the processes at work here. Fig. 6 shows the
nitrate concentrations from the atlas of Conkright et al. (1994), the HadOCC model and the
nutrient-restore model, integrated zonally and to a depth of 113 m. The nutrient restoring term acts
to reduce the surface nitrate towards the observed distribution over most of the ocean, so that the
zonal mean nitrate in the nitrate-restoring experiment is everywhere higher than the observations.
The HadOCC model yields a zonal mean nutrient curve that is rather close to that observed, but
deviates signi"cantly in the Southern Ocean. Here the HadOCC surface nutrient is somewhat
lower than the observed distribution. This is not the result of excessive export production, because
the nutrient-restore experiment yields a similar export production in this region. Further analysis
of the nutrient source and sink terms reveals a strong subsurface depletion in the Southern Ocean
driven by circulation. In the restore experiment this depletion is balanced by the nutrient-restore
term, but at depths below the euphotic zone so that no enhancement is seen in Fig. 5, and may be
related to the strong production peak seen at 453S.
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1185

6. Sensitivity of export production to changes in the models

6.1. Changing the windstress

We have shown above that advection is the principle mechanism supplying nutrient to the model
equatorial euphotic zone. Close to the equator, where the Coriolis force is small, wind stress on the
ocean surface drives upwelling of subsurface waters, bringing nutrients into the euphotic zone. We
investigated the sensitivity of the model by running HadOCC with the zonal and meridional
windstresses halved over the whole of the ocean. The windstresses used in the HadOCC model are
those of Hellerman and Rosenstein (1983), abbreviated hereafter as HR. These have been the
subject of much discussion, e.g. BoK ning et al. (1991), with many authors "nding HR too strong. For
example, in a tropical Paci"c ocean con"guration of the ocean model used here, Davey et al. (1994)
multiply HR by 0.75 everywhere. Equatorial upwelling may be poorly modelled for a variety of
reasons, including coarse horizontal and vertical resolution and inadequate representation of
equatorial wave processes. While we use the windstress to control the amount of upwelling, we do
not wish to argue that windstress errors are the principal cause of error in the model results.
As an independent indicator of the e!ects of changing the wind stress, we looked at the forcing of
heat in the equatorial Paci"c. An additional heat #ux is used in the HadOCC ocean model
consistent with that needed to restore sea-surface temperatures seasonally towards the Levitus and
Boyer (1994) climatology. Since this restoring term is used in addition to climatological forcings of
surface and penetrative heat #uxes, it may be used to diagnose inconsistencies between the heat
forcings and ocean circulation. In an ideal ocean model the restoring term would be zero. In the
ocean model used for this work the restoring heat term is high in areas where ocean currents are
poorly modelled, in particular in the regions of the Gulf Stream, the Kuroshio and the Benguela
currents. We averaged the heat #uxes in the upwelling region of the eastern equatorial Paci"c
between 53S and 53N, 180 and 803W. In the original full windstress experiment, the annual mean
restoring heat #ux in this box was 30.3 W m\, strongly into the ocean, suggesting too much
equatorial upwelling of cooler water. In the half windstress model, the equatorial box restoring
heat #ux was 10.7 W m\ out of the ocean, suggesting that we have slightly over compensated for
the excessive upwelling.
The global, Atlantic and Paci"c zonal mean primary productivity for the half windstress
experiment are shown in Fig. 7. The productivities for the full windstress experiment are shown in
grey for comparison. Reducing the windstress has clearly had a huge impact in the equatorial
Paci"c, bringing the productivity in line with the LSPC95 estimate, although the restoring heat #ux
discussed above suggests that the upwelling is now too small. In the equatorial Atlantic the e!ects
are mixed: productivity is lowered south of the equator, but there is little change to the north. In the
global plot the production in the Paci"c dominates. Clearly, the wind stress has a signi"cant
in#uence at the equator, with little e!ect poleward of 403S and 403N. This is in good agreement
with the degree to which advection contributes to export production as shown in Fig. 4.

6.2. Changing the mixed layer model

The winter deepening and summer shallowing of the mixed layer is the principle control of
nutrient supply in many ecosystem models of the boreal oceans, such as that of Fasham et al.
1186 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Fig. 7. Annual mean primary production is shown for the HadOCC model experiment W in which the windstresses have
been halved. The results for the full windstress experiment S (see Fig. 2) are shown in grey. For comparison, the dotted
lines show the zonal means of the Longhurst et al. (1995) data. The three panels show the zonal mean primary production
globally, for the Atlantic and for the Paci"c.

(1990). To investigate the sensitivity of the HadOCC model to changes in the mixed layer model we
tried three experiments. In the "rst the wind mixing energy was increased globally by a factor of 1.5
(experiment M1). In experiments M2 and M3 we increased the depth scale of the decay of the
surface buoyancy energy from 100 to 200 and 1000 m, respectively. Expressed as annual and glo-
bal zonal means, the mixed layer depths increased at all latitudes, as expected. The top panel in
Fig. 8 shows the fractional changes from standard run S. The increases were largest in experiment
M3, increasing by between about 20 and 30 m between the 30 and 603 latitudes in each hemisphere,
and by less than 10 m in the equatorial and polar regions. Primary productivities showed an
interesting response to the increased mixed layer depths. The fractional changes in primary
production (from the standard run S) are shown in the lower panel of Fig. 8. The response to
changes in the mixed layer depth is weakly negative at the equator, high northern latitudes and in
the Southern Ocean, and very strongly positive in the subtropics.
In the equatorial ocean, the slightly deeper mixed layer does not extend down to high-nutrient
waters and the nutrient supply (which we have shown is primarily the result of upwelling) is not
increased. Phytoplankton are mixed down to more poorly lit waters, reducing growth rates
slightly. The subtropical gyres are regions of weak downwelling and rely on vertical mixing
processes for nutrient supply. The deeper mixed layer reaches down into higher nutrient waters
bringing signi"cantly more nutrient into the euphotic zone. Because the HadOCC ecosystem here
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1187

Fig. 8. Fractional changes in mixed layer depths (upper panel) and primary production (lower panel) in the deeper mixed
layer HadOCC experiments (M1, M2, M3) relative to the standard run (S). The zonal means of the ratios have been
plotted. Primary production shows a strong positive response to mixed layer depth increases in the oligotrophic
subtropics.

is strongly nutrient limited the primary production response is very strong. In the Southern Ocean
nutrients are never limiting so any increase in nutrient concentration will have little e!ect on
growth rates, but the deeper mixed layer carries phytoplankton to more poorly lit depths and so
reduces production. In high northern latitudes the ecosystem can be nutrient limited in the summer
in some areas, when a deeper mixed layer will increase the nutrient supply. A deeper mixing of the
phytoplankton will partially o!set this e!ect.

7. Summary

The Hadley Centre ocean carbon cycle model (HadOCC) embeds a four-component NPZD
ecosystem and a carbon chemistry model within an ocean GCM. In this paper we have examined
aspects of the annual mean modelled primary and export production. The global primary
production in the model is about 48 Gt C yr\, comparing well with recent estimates based on in
situ and satellite measurements. The fraction of primary production that is exported to the deeper
ocean agrees well with sediment trap data. The geographical distribution of production is in
reasonable agreement with that revealed by studies based on satellite measurements of sea-surface
chlorophyll, indicating that the mechanisms which exert the strongest control on primary
1188 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

production in the real ocean are adequately represented in the model. From this we conclude that
the HadOCC model is suitable for studying the possible e!ects of climate change on ocean
ecosystems and the ocean carbon cycle.
We have described some of the di!erences between the primary production in the HadOCC
model and that derived from satellite chlorophyll data by Longhurst et al. (1995), which we refer to
as LSPC95. It is important to note that there is considerable uncertainty in this interpretation of
satellite data, as can be seen by comparing LSPC95 with the results of Antoine et al. (1996). The
model and data agree broadly in the distribution of low production in the subtropical gyres with
elevated production in equatorial and high latitudes. However, certain regions of the ocean stand
out as poorly modelled in HadOCC. Coastal regions of high production are not well modelled by
the low resolution of the horizontal grid. The equatorial Paci"c is a region of very high production
in the model, a feature not supported by the data. The mid-latitude southern oceans are also
regions of rather high production in the model. The high northern latitudes are regions of very high
production according to LSPC95, but of much more modest production according to HadOCC
and Antoine et al.
Is the excessive modelled production in areas such as the equatorial Paci"c a result of errors in
the biology model or the ocean model? By using a simple nutrient-restoring biology model we have
shown that high equatorial production is consistent with the modelled ocean circulation and the
observed nitrate distribution (Conkright et al., 1994), from which we deduce that the modelled
circulation is at least partly at fault here. Reducing the model surface windstress, which itself may
be too high in the Hellerman and Rosenstein (1983) dataset, reduces the upwelling at the equator
and brings the equatorial Paci"c production into line with LSPC95. We have also demonstrated
the sensitivity of the modelled primary production to the mixed layer model, showing a strong
positive response to increased mixed layer depth in the oligotropic subtropical gyres and a weak
negative response at other latitudes.
We recognise however that the NPZD model incorporated in the HadOCC model is a very
simple representation of the ocean ecosystem. This simplicity has been forced by computer memory
storage and run-time constraints, but is consistent with our desire to develop a model of the marine
carbon cycle which is capable of being understood in detail. We must consider whether the model's
ability to represent the marine carbon cycle has been signi"cantly compromised by its simplicity.
The model produces a very small range of export ratios (most values lie between 13 and 17%),
whereas the sediment trap compilation of Lampitt and Antia (1997) shows a seven-fold range of
values. Also, for production rates above 50 g C m\ yr\ the model shows a slight decrease in the
export ratio with increasing production. Lampitt and Antia only see such a decrease above
200 g C m\ yr\, and an increase in the ratio with production might be expected: areas of low
annual production tend to be dominated by picoplankton which do not contribute to the export
because the microzooplankton which eat them do not produce fecal pellets, whereas areas of high
annual production often additionally feature larger phytoplankton which contribute directly to the
sinking #ux or are grazed directly by fecal pellet-producing mesozooplankton (Michaels and Silver,
1988). An example of a change in community structure a!ecting export was described by Boyd and
Newton (1995) who found that the sinking #ux of POC from the surface layer of the North Atlantic
Ocean in the spring of 1989 was twice that measured in 1990, despite the primary production being
approximately the same. They showed that the bloom of 1990 consisted of smaller cells than
the previous year, and the e$ciency with which the biomass was exported was reduced because of
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1189

the increased number of trophic steps to the consumers that produced the sinking particles. The
HadOCC model has just one phytoplankton and one zooplankton, and has no representation of
the `microbial loopa, so it can only have one community structure and operate in only one mode in
relation to export.
We showed in Section 4.3 that the HadOCC model's global export ratio at 2000 m is within the
range of observations, but the lack of variation in the ratio between individual grid boxes indicates
that at some points at least the model is predicting the export #ux incorrectly. The model will
e!ectively be exporting the globally correct amount of carbon with an partially incorrect geo-
graphical distribution. However since the timescale for the return of the carbon to the surface
(several hundreds of years) will be comparable with the ocean stirring time the errors should be
smoothed out in the deep ocean by mixing and so the operation of the carbon cycle on a global
scale should not be signi"cantly a!ected. The amount of CO returned to solution in the surface

waters will also be in error since that is the di!erence between production and export but as there
are other processes which also a!ect the surface concentration of DIC the overall e!ect is not easy
to predict. We plan to describe and validate the deep ocean concentration of DIC and the air}sea
#uxes of CO produced by HadOCC in a future paper.

Although the version of the HadOCC model described in this paper is currently being used for
carbon cycle experiments in coupled mode and for scenario runs in ocean-only mode, it is planned
to develop the model. The NPZD model will be incorporated into a 1.25;1.253 version of the
physical ocean model to examine how the improved resolution changes the nutrient supply
processes and hence the distribution of primary production. At standard resolution, a model of the
production and consumption of dissolved organic matter is being coupled to the NPZD model.
The biological model is also being extended to feature diatoms and dissolved silicate as additional
variables, in order to better simulate the processes which link production to export. We also intend
to use the extended model to examine the e!ect of iron limitation on primary production.

Acknowledgements

This work was supported by the Department of the Environment, Transport and the Regions
under grant PECD 7/12/37. We also thank the three anonymous reviewers for their constructive
criticisms.

Appendix A. The carbon cycle model

The HadOCC ecosystem model calculates the #ows of nitrogen between four model compart-
ments; nutrient, phytoplankton, zooplankton and detritus (N, P, Z, D). The #ows of carbon within
the system are coupled to the #ows of nitrogen by "xed carbon:nitrogen ratios C , C and C , but
  
these #ows have no e!ect on the ecosystem because growth is not carbon limited.
The Eulerian equation for the rate of change of any of the ecosytem components ¹ can be
G
written
d¹
G "advection#di!usion#mixing#sinking#biology.
dt
1190 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Table 2
Ecosystem model parameters

Parameter Symbol Value

C : N ratio of phytoplankton C 6.625


C : N ratio of zooplankton C 5.625

C : N ratio of detritus C 7.5

Phyto.: max. photosynthetic rate, 103C P 0.8 day\

Phyto.: initial slope of the P}I curve
0.055 (W m\)\ day\
Phyto.: Q temp.-dep. growth mantissa Q 1.0
 
Phyto.: nutrient-lim. half-saturation K 0.1 mMol m\
,
Phyto.: respiration, speci"c rate  0.02 day\
Phyto.-dep. speci"c mortality m 0.05 day\ (mMol m\)\

Zoopl.: grazing, maximum rate g 1.0 day\

Zoopl.: grazing, half-saturation K 0.75 mMol m\
%
Zoopl.: grazing threshold F 0.1 mMol m\ day\
 
Zoopl.: assimilation e$ciency of phyto.  0.7

Zoopl.: assimilation e$ciency of detr.  0.5

Zoopl.: constant speci"c mortality  0.05 day\

Zoopl.: zoopl.-dep. speci"c mortality  0.2 day\ (mMol m\)\

Detritus: remin. rate, shallow  0.1 day\
Detritus: remin. rate, deep  0.02 day\
Detritus: sinking velocity 10 m day\
Carbonate precipitated per unit PP  0.01


The advection, di!usion and mixing terms include all physical transport processes of a tracer in the
ocean model. The sinking term is zero except for the detritus which sinks at a speed of 10 m day\.
The biology terms for each of the four ecosystem compartments are described below. Parameters
for which no source is given were determined by "tting the biological model to data using a simple
1-D column version of the model. Many of the zooplankton parameters have been obtained this
way, because the zooplankton compartment represents both micro- and meso-zooplankton.
Parameter values are given in Table 2.

a. The phytoplankton equation:

P
"RP!G !mP!P,
t 

 
where
N
R"¸ ,
N#K
,
I
¸"P  ,
 (P /
)#I
 
P "P Q2\,
  


0, P)0.01 mmol m\
m"
m otherwise.

 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1191

The terms in the main equation represent growth, grazing by zooplankton, natural mortality and
respiration, respectively. The phytoplankton-speci"c growth rate of phytoplankton, R, is given by
combining light and nutrient limitation terms. The Michaelis}Menten forms are used for these
terms to model the saturation of nutrient uptake (at high nutrient concentration) and photosyn-
thesis (at high irradiance). The value of the half-saturation parameter K in the nutrient-limitation
,
term (0.1 mmol Nitrogen m\) is from Taylor et al. (1991). The use of the Michaelis}Menten form
for photosynthesis follows Baly (1935), although strictly here the term represents growth and it has
been assumed that there are no internal stores of "xed carbon, i.e. growth is balanced. In the
equation for ¸, the light limitation term, P is the temperature-dependent maximum growth rate,


is the initial slope of the photosynthesis-irradiance (P}I) curve and I is the average solar

irradiance within the ocean layer. Note that the irradiance pro"le in the ocean is determined using
the two-band scheme of Paulson and Simpson (1977), and the light attenuation coe$cients are
constant with depth, i.e. there is no variation related to chlorophyll concentration. The temperature
dependence of P allows the growth rate (which is P at 103C) to increase in response to
 
increased local water temperature ¹ using a Q parameter (Eppley, 1972) raised to a power that

increases by 1.0 for every rise of 103C. For the model experiments reported in this paper, Q was

set equal to 1.0, so there was no temperature variation of the growth rate. The P value,

0.8 day\, is consistent with the growth rates measured by Strom and Welschmeyer (1991) during
the SUPER study in the sub-arctic Paci"c. The value of
, 0.055 (W m\)\ day\, corresponds
to 0.02 mg C (mg Chl a)\ h\ (
Einsteins m\ s\)\ if a carbon : chlorophyll ratio of 40 is
assumed. It is approximately double the values found in the North Atlantic by Harrison and Platt
(1986), but around half the values implied by the results obtained in the subarctic Paci"c by Booth
et al. (1988). Note that ¸ is set to zero below level 9 of the model (i.e. below 242.8 m depth) to
prevent numerical under#ow problems. There are three processes which cause a loss of phytoplan-
kton biomass. Respiration has a constant phytoplankton-speci"c rate  (technically only carbon
should be released, but because of the "xed C : N ratio there has to be a release of nitrogen as well).
Grazing by zooplankton G is described below. Mortality has a phytoplankton-speci"c rate mP

that varies linearly with P and represents the stresses due to overpopulation. Mortality is reduced
to zero if the phytoplankton population falls below a threshold to simulate the formation of
dormant cysts in extremely inhospitable conditions. The (atomic) carbon-to-nitrogen ratio for
phytoplankton C is that found by Red"eld et al. (1963), 106 : 16 or 6.625.

b. The zooplankton equation:
Z
"G !( Z# Z).
t   

 
The terms represent grazing (by zooplankton) and mortality, respectively. Zooplankton graze on
phytoplankton and detritus according to a Holling Type III formulation Holling, 1965. This
`Sa-shaped response allows a `refugea for the prey from grazing when the prey population is low,
but when it is high the grazing rate saturates. The zooplankton are assumed to be non-discrimina-
tory feeders, ingesting phytoplankton and detritus in the proportion that each is available.
However this proportion must be considered in terms of the biomass, rather than the nitrogen
content which has a di!erent relation to biomass in phytoplankton and detritus. The biomass is
taken to be proportional to the sum of the masses of carbon and nitrogen present, and is expressed
1192 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

in terms of a "ctional prey with a C:N ratio of 106 : 16, i.e. that found by Red"eld et al. (1963)
¹ 14.01#12.01C

 "B " ,
¹  14.01#12.01C
0
where C "106/16 and ¹ is P, Z or D. Note that the biomass-equivalent quantities ¹ are used
0

only to determine the ingestion rate; the actual ingestion yuxes are given in terms of the nitrogen
content as usual. The total ingestion rate I is
 
F
I "B Zg ,
    F#K
%
where K is the half-saturation constant for grazing and F is the e!ective prey, i.e. the total food
%
available above a threshold F :
 
F"max(0, (B P#B D!F )).
   
The loss-rates of phytoplankton (G ) and detritus (G ) due to grazing are then (in nitrogen units)
 
I P
G "   ,
 B P#B D
 
I D
G "   .
 B P#B D
 
Of the phytoplankton and detritus ingested only fractions  and  are available for assimilation
 
into zooplankton biomass. These fractions are used for both carbon and nitrogen. The value used
for phytoplankton is consistent with the results of Hassett and Landry (1988), while that used for
detritus has been "tted. The carbon and nitrogen available for assimilation are given by
a " C G # C G ,
!      
a " G # G .
,    
Because new zooplankton biomass must have a constant C:N ratio of C , not all the available

material can be assimilated. The rate of assimilation is


a
G "min ! , a .
 C ,

The ingested material that is not assimilated is released as described in the detritus and nutrient
equations. Note that with the C : N ratios used in this paper all the available nitrogen is assimilated,
but not all the carbon. The value of C is set to be 5.625, consistent with the results of Verity (1985),

who found the ratio to be around 5.5 for tintinids, and Ba mstedt (1986) who found the ratio for
mid-latitude shallow copepods to be between 5 and 5.5. Zooplankton mortality includes grazing by
predators from higher trophic levels which are not modelled explicitly (the closure term), as well as
death by natural causes. The density-dependent term, with coe$cient  , represents an increase in

the local population of the predators, either because of growth or because of aggregation in
a region of more abundant food (Steele and Henderson, 1981, 1992). Material lost by mortality
enters the detritus and nutrient compartments, as described in the detritus equation. The model
zooplankton display no `behavioura: that is, their growth and feeding are controlled only by the
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1193

current local concentrations of the prey and the zooplankton themselves, and no movement, such
as diel migration, is imposed (other than physical transports).

c. The detritus equation:

D
"m P#( Z# Z)#E !D!G ,
t "    " 

 
where


C
m "m min 1,  ,
" C



(C G #C G !C G )
E "min (G #G !G ),       .
"    C

The terms represent #ows to/from the detritus compartment due to the following processes:
phytoplankton natural mortality, zooplankton mortality, egestion by zooplankton, remineralisa-
tion and grazing by zooplankton. The detritus compartment comprises a combination of dead
phytoplankton and zooplankton and egested faecal pellets. It is recycled through zooplankton
grazing and through remineralisation. As much of the #ux due to phytoplankton mortality as
possible becomes detritus. If C is less than C , as in this study, the associated #ux of carbon limits
 
the #ow to detritus; the remaining nitrogen returns to dissolved nutrient (see below). One third of
the zooplankton biomass lost through mortality becomes detritus (Fasham et al., 1990). Again, the
remainder returns to dissolved nutrient. The egestion #ux E is the material that has been ingested
"
by zooplankton but not assimilated. As much as possible, consistent with balancing the carbon
#ows, becomes detritus. The loss processes from the detritus compartment are remineralisation and
grazing. In the real ocean the former is largely mediated by bacteria, but these are not represented
in the HadOCC model which converts a fraction of the detritus to dissolved nutrient each day.
A larger proportion is remineralised in the top eight layers (i.e. to a depth of 164.8 m) than in deeper
levels to simulate the more active biological processes there. For example, the warmer temperatures
allow bacterial decomposition to proceed more quickly in the surface waters. Combined with the
sinking rate the deep remineralisation rate of 0.02 day\ gives a depth scale of detrital breakdown
of 500 m, comparable with the "gure in Volk and Ho!ert (1985). The grazing process has
been described above. The C : N ratio of detritus has been set at 7.5, which is the average ratio
found by Martin et al. (1987). The sinking rate of 10 m day\ (Fasham et al., 1990) represents the
full size spectrum of sinking particles, not just the large fast-sinking ones. The (very small) sinking
#ux of detritus reaching the ocean #oor is returned to the mixed layer; this is similar to the method
used by Bacastow and Maier-Reimer (1990) for detritus that was not remineralised in the bottom
level.

d. The nutrient equation:

N
"!RP#(m!m )P#P
t "

 
#( Z# Z)#(G #G !G !E )#D.
      "
1194 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

The terms represent the following processes: phytoplankton growth, phytoplankton natural
mortality, phytoplankton respiration, zooplankton mortality, release of nutrient during zoo-
plankton grazing and remineralisation of detritus.

e. The DIC and total alkalinity equations:

For completeness we include the equations for dissolved inorganic carbon (DIC) and total
alkalinity A :
2


(DIC) C
"(mC !m C )P# C !  ( Z# Z)
t  "   3  

 
# (C G #C G !C G !C E )#C P
       " 
# C D!RC P! RC P#D ,
    !!
A N
2 "! !2 RC P#2D ,
t t   !!

 
 


0 layer-depth )1500 m,
D "
!!  C
  I(RP) otherwise,
D !1500
 
where total alkalinity is de"ned as
A "[HCO\]#2[CO\]#[B(OH)\]#[OH\]![H>]
2   
and where D is the local depth of the ocean and I(RP) is the depth integral of phytoplankton
 
growth RP. All the terms in the DIC equation, except the last two, correspond directly to terms in
the nutrient equation. The penultimate term represents the precipitation of calcium carbonate from
solution in sea water by phytoplankton to form shells and other hard body parts, which sub-
sequently sink below the euphotic zone. This sinking #ux of carbon, often known as the `hard
tissuea #ux, is comparable in size to the organic carbon (or `soft tissuea) #ux, but has no associated
#ux of nitrogen. The #ux is modelled in a simple way in HadOCC, as being proportional (and
additional) to the uptake of carbon for primary production;  is the constant of proportionality.

Because it is the ionic form CO\ that is precipitated there is a decrease in total alkalinity, two

molar equivalents for each mole of carbonate precipitated. This is represented by the penultimate
term of the alkalinity equation. Calcium carbonate is super-saturated in the upper 2000 m or so of
the ocean, so the shells, etc., have no tendancy to dissolve until they have sunk to below the depth of
saturation, the lysocline. However, it has been observed (Rios et al., 1995) that some dissolution
occurs above this depth, so in HadOCC the lysocline is set at 1500 m depth. Carbonate that is
precipitated in the euphotic zone is returned to solution in the model instantly and evenly at all
depths between the lysocline and the base of the water column. This dissolution is represented by
the term D . The alkalinity is changed as well. Note that the C:N ratio for detritus must be less
!!
than three times that for the zooplankton, or the zooplankton mortality term in the DIC equation
becomes negative (and so unphysical), but this restriction is not serious in practice. Biological
processes other than carbonate precipitation alter the alkalinity. It has been observed (Goldman
and Brewer, 1980) that uptake of nitrate causes a release of OH\ ions and so there will be an
increase in the total alkalinity as de"ned above. The equation for alkalinity follows from identifying
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1195

nutrient as nitrate, which is reasonable as the ammonium concentration is generally low. Not
shown in the above equations, because the process is not biological, are the adjustments to the DIC
and total alkalinity concentrations caused by freshwater #uxes. Because the ocean GCM uses
a surface boundary condition of zero vertical motion, freshwater #uxes are modelled using surface
salinity #uxes. Addition or removal of DIC and alkalinity in the surface layer are required for
consistency. Failure to include these #uxes leads to large and spurious air}sea #uxes of CO . The

partial pressure of CO in the surface waters is determined by solving equations representing the

sea water acid}base system. The expressions for the dissociation constants of carbonic acid,
hydrogen carbonate, boric acid and water and for the solubility of CO in seawater are taken from

DOE (1994). Using the salinity-dependent boron concentration of Peng (1987), the acid}base
system is solved using the method of Bacastow (1981) to yield the concentration of carbonic acid
and hence the partial pressure of CO . The wind-dependent gas exchange formulation of Wannin-

khof (1992) is used to calculate the air}sea #ux of CO from the wind climatology used to drive the

ocean model. The partial pressure of CO in the atmosphere is a prescribed constant.


References

Angel, M.V., 1989. Does mesopelagic biology a!ect the vertical #ux?. In: Berger, W.H., Smetacek, V.S., Wefer, G. (Eds.),
Productivity of the Oceans: Present and Past. Wiley, New York, pp. 155}173.
Antoine, D., AndreH , J.-M., Morel, A., 1996. Oceanic primary production 2. Estimation at global scale from satellite
(coastal zone color scanner) chlorophyll. Global Biogeochemical Cycles 10, 57}69.
Bacastow, R.B., 1981. Numerical evaluation of the evasion factor. In: Bolin, B. (Ed.), Carbon Cycle Modelling. Wiley,
New York, pp. 95}101.
Bacastow, R., Maier-Reimer, E., 1990. Ocean-circulation model of the carbon cycle. Climate Dynamics 4, 95}125.
Bacastow, R., Maier-Reimer, E., 1991. Dissolved organic carbon in modeling oceanic new production. Global Bi-
ogeochemical Cycles 5, 71}85.
Baly, E.C.C., 1935. The kinetics of photosynthesis. Proceedings of the Royal Society of London Series B 117, 218}239.
Ba mstedt, U., 1986. Chemical composition and energy content. In: Corner, E.D.S., O'Hara, S.C.M. (Eds.), The Biological
Chemistry of Marine Copepods. Clarendon Press, Oxford, pp. 1}58.
Behrenfeld, M.J., Falkowski, P., 1997. Photosynthetic rates derived from satellite-based chlorophyll concentration.
Limnology and Oceanography 42, 1479.
Berger, W.H., Fischer, K., Lai, C., Wu, G., 1987. Oceanic productivity and organic carbon #ux. Part 1, Overview and
maps of primary production and export production. Scripps Institution of Oceanography, SIO Reference 87-30, pp.
1}67.
Berger, W.H., Wefer, G., 1990. Export production: seasonality and intermittency, and paleoceanographic implications.
Palaeogeography, Palaeoclimatology, Palaeoecology (Global and Planetary Change Section) 89, 245}254.
Betzer, P.R., Showers, W.J., Laws, E.A., Winn, C.D., DiTullio, G.R., Kroopnick, P.M., 1984. Primary productivity and
particle #uxes on a transect of the equator at 1533W in the Paci"c Ocean. Deep-Sea Research 31, 1}11.
BoK ning, C.W., DoK scher, R., Isemer, H.-J., 1991. Monthly mean wind stresses and Sverdrup transports in the North
Atlantic: a comparison of the Hellerman-Rosenstein and Isemer-Hasse climatologies. Journal of Physical Oceanogra-
phy 21, 221}235.
Booth, B.C., Lewin, J., Lorenzen, C.J., 1988. Spring and summer growth rates of subarctic Paci"c phytoplankton
assemblages determined from carbon uptake and cell volumes estimated using epi#uorescence microscopy. Marine
Biology 98, 287}298.
Boyd, P., Newton, P., 1995. Evidence of the potential in#uence of planktonic community structure on the interannual
variability of particulate organic carbon #ux. Deep-Sea Research I 42, 619}639.
Bryan, K., 1969. A numerical method for the study of the circulation of the world ocean. Journal of Computational
Physics 4, 347}379.
1196 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Conkright, M.E., Levitus, S., Boyer, T.P., 1994. NOAA Atlas NESDIS 2, World Ocean Atlas 1994, Vol. 1: Nutrients, U.S.
Government Printing O$ce, Washington, DC, 150pp.
Cox, M.D., 1984. A primitive equation, 3-dimensional model of the ocean. GFDL Ocean Group Technical Report No. 1,
143 pp.
Davey, M.K., Ineson, S., Balmaseda, M.A., 1994. Simulation and hindcasts of tropical Paci"c Ocean interannual
variability. Tellus 46A, 433}447.
Davies, J.M., 1975. Energy #ow through the benthos in a Scottish sea loch. Marine Biology 31, 353}362.
DOE, 1994. In: Dickson, A.G., Goyet, C. (Ed.), Handbook of Methods for the Analysis of the Various Parameters of the
Carbon Dioxide System in Sea Water; Version 2. ORNL/CDIAC-74.
Eppley, R.W., 1972. Temperature and phytoplankton growth in the sea. Fisheries Bulletin 70, 1063}1085.
Eppley, R.W., 1981. Autotrophic production of particulate matter. In: Longhurst, A.R. (Ed.), Analysis of Marine
Ecosystems. Academic Press, London, pp. 342}361.
Eppley, R.W., Peterson, P.J., 1979. Particulate organic matter #ux and planktonic new production in the deep ocean.
Nature 282, 677}680.
Esbensen, S.K., Kushnir, Y., 1981. The heat budget of the global ocean: an atlas based on estimates from surface marine
observations. Climate Research Institute, Report No. 19. Oregon State University, Corvallis, USA.
Evans, G.T., Parslow, J.S., 1985. A model of annual plankton cycles. Biological Oceanography 3, 327}347.
Fasham, M.J.R, Ducklow, H.W., McKelvie, S.M., 1990. A nitrogen-based model of plankton dynamics in the oceanic
mixed layer. Journal of Marine Research 48, 591}639.
Feldman, G., Kuring, N., Ng, C., Esaias, W., McClain, C., Elrod, J., Maynard, N., Endres, D., Evans, R., Brown, J., Walsh,
S., Carle, M., Podesta, G., 1989. Ocean color: Availability of the global data set. Eos Transactions, AGU 70, 634}641.
Field, C.B., Behrenfeld, M.J., Randerson, J.T., Falkowski, P., 1998. Primary production of the biosphere: integrating
terrestrial and oceanic components. Science 281, 237}240.
Gent, P.R., McWilliams, J.C., 1990. Isopycnal mixing in ocean circulation models. Journal of Physical Oceanography 20,
150}155.
Goldman, J.C., Brewer, P.G., 1980. E!ect of nitrogen source and growth rate on phytoplankton-mediated changes in
alkalinity. Limnology and Oceanography 25, 352}357.
Harrison, W.G., Platt, T., 1986. Photosynthesis-irradiance relationships in polar and temperate phytoplankton popula-
tions. Polar Biology 5, 153}164.
Harrison, W.G., Platt, T., Lewis, M.R., 1987. f-Ratio and its relationship to ambient nitrate concentration in coastal
waters. Journal of Plankton Research 9, 235}248.
Hassett, R.P., Landry, M.R., 1988. Short-term changes in feeding and digestion by the copepod Calanus Paci"cus.
Marine Biology 99, 63}74.
Hellerman, S., Rosenstein, M., 1983. Normal monthly windstress over the world ocean with error estimates. Journal of
Physical Oceanography 13, 1093}1104.
Holling, C.S., 1965. The functional response of predators to prey density and its role in mimicry and population
regulation. Memoirs of the Entomological Society of Canada No. 45, 61pp.
Jaeger, L., 1976. Monatskarten des Niederschlags fur die ganze Erde. Bet Deutsche Wetterdienstes 18 (139) 38.
Johns, T.C., Carnell, R.E., Crossley, J.F., Gregory, J.M., Mitchell, J.F.B., Senior, C.A., Tett, S., Wood, R.A., 1997. The
second Hadley Centre coupled ocean-atmosphere GCM: Model description, spinup and validation. Climate Dynam-
ics 13, 103}134.
Kraus, E.B., Turner, J.S., 1967. A one dimensional model of the seasonal thermocline, II. The general theory and its
consequences, Tellus 19, 98}105.
Kraus, E.B., 1990. Diapycnal mixing. In: Schlesinger, M.E. (Ed.), Climate-Ocean Interaction.. Kluwer, Amsterdam, pp.
269}293.
Lampitt, R.S., Antia, A.N., 1997. Particle #ux in the deep seas: regional characteristics and temporal variability. Deep-Sea
Research 44, 1377}1403.
Large, W.G., McWilliams, J.C., Doney, S.C., 1994. Oceanic vertical mixing: a review and a model with a nonlocal
boundary layer parametrisation. Reviews of Geophysics 32, 363}403.
Levitus, S., Boyer, T.P., 1994. NOAA Atlas NESDIS 2, World Ocean Atlas 1994, Vol. 4: Temperature. U.S. Government
Printing O$ce, Washington, DC, 117pp.
 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198 1197

Levitus, S., Burgett, R., Boyer, T.P., 1994. NOAA Atlas NESDIS 2, World Ocean Atlas 1994, Vol. 3: Salinity. U.S.
Government Printing O$ce, Washington, DC, 99pp.
Lewis, M.R., Harrison, W.G., Oakey, N.S., Herbet, D., Platt, T., 1986. Vertical nitrate #uxes in the oligotrophic ocean.
Science 234, 870}873.
Lohrenz, S.E., Knauer, G.A., Asper, V.L., Tuel, M., Michaels, A.F., Knap, A.H., 1992. Seasonal variability in primary
production and particle #ux in the northwestern Sargasso Sea: US JGOFS Bermuda Atlantic Time-series study.
Deep-Sea Research 39, 1373}1391.
Longhurst, A.R., Bedo, A.W., Harrison, W.G., Head, E.J.H., Sameoto, D.D., 1990. Vertical #ux of respiratory carbon by
oceanic diel migrant biota. Deep-Sea Research 37, 685}694.
Longhurst, A., Sathyendranath, S., Platt, T., Caverhill, C., 1995. An estimate of global primary production in the ocean
from satellite radiometer data. Journal of Plankton Research 17, 1245}1271.
Maier-Reimer, E., 1993. Geochemical cycles in an ocean general circulation model. Preindustrial tracer distributions..
Global Biogeochemical Cycles 7, 645}677.
Martin, J.H., Fitzwater, S.E., Gordon, R.M., 1990. Iron de"ciency limits phytoplankton growth in Antarctic waters.
Global Biogeochemical Cycles 4, 5}12.
Martin, J.H., Knauer, G.A., Karl, D.M., Broenkow, W.W., 1987. VERTEX: carbon cycling in the northeast Paci"c.
Deep-Sea Research 34, 267}285.
Michaels, A.F., Silver, M.W., 1988. Primary production, sinking #uxes and the microbial food web. Deep-Sea Research
35, 473}490.
Najjar, R.G., Sarmiento, J.L., Toggweiler, J.R., 1992. Downwards transport and fate of organic matter in the ocean:
simulations with a general circulation model. Global Biogeochemical Cycles 6, 45}76.
Pacanowski, R.C., Philander, S.G.H., 1981. Parametrisation of vertical mixing in numerical models of tropical oceans.
Journal of Physical Oceanography 11, 1443}1451.
Pace, M.L., Knauer, G.A., Karl, D.M., Martin, J.H., 1987. Primary production, new production and vertical #ux in the
eastern Paci"c Ocean. Nature 325, 803}804.
Paulson, C.A., Simpson, J.J., 1977. Irradiance measurements in the upper ocean. Journal of Physical Oceanography 7, 952}956.
Peng, T.-H., 1987. Seasonal variability of carbon dioxide, nutrients and oxygen in the northern North Atlantic surface
water: observations and a model. Tellus 39B, 439}458.
QueH guiner, B., TreH guer, P., Peekin, I., Scharek, R., 1997. Biogeochemical dynamics and the silicon cycle in the Atlantic
sector of the Southern Ocean during austral spring 1992. Deep-Sea Research II 44, 69}89.
Rahmstorf, S., 1993. A fast and complete convection scheme for ocean models. Ocean Modelling 101, 9}11.
Red"eld, A.C., Ketchum, B.H., Richards, F.A., 1963. The in#uence of organisms on the composition of sea-water. In: Hill,
M.N. (Ed.), The Sea. Vol. 2. Wiley-Interscience, New York, pp. 26}77.
Riley, G.A., 1944. The carbon metabolism and photosynthetic e$ciency of the Earth as a whole. American Scientist,
Spring Issue 32, 129.
Rios, A.F., Anderson, T.R., Perez, F.F., 1995. The carbonic system distribution and #uxes in the NE Atlantic in spring
1991. Progress in Oceanography 35, 295}314.
Rowe, G.T., Cli!ord, C.H., Smith Jr., K.L., Hamilton, P.L., 1975. Benthic nutrient regeneration and its coupling to
primary productivity in coastal waters. Nature 255, 215}217.
Sarmiento, J.L., Le QueH reH , C., 1996. Oceanic carbon dioxide uptake in a model of century-scale global warming. Science
274, 1346}1350.
Sarmiento, J.L., Orr, J.C., Siegenthaler, U., 1992. A perturbation simulation of CO uptake in an ocean general

circulation model. Journal of Geophysical Research 97, 3621}3645.
Schimel, D., Enting, I.G., Heimann, M., Wigley, T.M.L., Raynaud, D., Alves, D., Siegenthaler, U., 1995. CO and the

carbon cycle. In: Houghton, J.T., Meira Filho, L.G., Bruce, J., Lee, H., Callander, B.A., Haites, E., Harris, N., Maskell,
K. (Eds.), Climate Change 1994: Radiative Forcing of Climate Change and an Evaluation of the IPCC IS92 Emission
Scenarios. Cambridge University Press, Cambridge, pp. 35}71.
Siegenthaler, U., Joos, F., 1992. Use of a simple model for studying oceanic tracer distributions and the global carbon
cycle. Tellus 44B, 186}207.
Six, K.D., Maier-Reimer, E., 1996. E!ects of plankton dynamics on seasonal carbon #uxes in an ocean general circulation
model. Global Biogeochemical Cycles 10, 559}583.
1198 J.R. Palmer, I.J. Totterdell / Deep-Sea Research I 48 (2001) 1169}1198

Steele, J.H., Henderson, E.W., 1981. A simple plankton model. American Naturalist 117, 676}691.
Steele, J.H., Henderson, E.W., 1992. The role of predation in plankton models. Journal of Plankton Research 14,
157}172.
Steemann Nielsen, E., 1952. The use of radioactive carbon (C) for measuring organic production in the sea. Journal du
Conseil International pour l'Exploration du Mer 18, 117}140.
Strom, S.L., Welschmeyer, N.A., 1991. Pigment-speci"c rates of phytoplankton growth and microzooplankton grazing in
the open subarctic Paci"c Ocean. Limnology and Oceanography 36, 50}63.
Suess, E., 1980. Particulate organic carbon #ux in the oceans-surface productivity and oxygen utilization. Nature 288,
260}263.
Sugimura, Y., Suzuki, Y., 1988. A high-temperature catalytic oxidation method for the determination of non-volatile
dissolved organic carbon in seawater by direct injection of a liquid sample. Marine Chemistry 24, 105}131.
Sundquist, E.T., 1985. Geological perspectives on carbon dioxide and the carbon cycle. Geophysical Monographs 32,
5}59.
Taylor, A.H., Watson, A.J., Ainsworth, M., Robertson, J.E., Turner, D.R., 1991. A modelling investigation of the role of
phytoplankton in the balance of carbon at the surface of the North Atlantic. Global Biogeochemical Cycles 5,
151}171.
Verity, P.G., 1985. Grazing, respiration, excretion and growth rates of tintinnids. Limnology and Oceanography 30,
1268}1282.
Volk, T., Ho!ert, M.I., 1985. Ocean carbon pumps: Analysis of relative strengths and e$ciencies in ocean-driven pCO

changes. In: Sundquist, E.T., Broecker, W.S. (Eds.), Carbon Dioxide and the Carbon Cycle, Archean to Present,
Geophysical Monograph Series, Vol. 32. AGU, Washington DC., pp. 99}110.
Wanninkhof, R., 1992. Relationship between wind speed and gas exchange over the ocean. Journal of Geophysical
Research 97, 7373}7382.
Yamanaka, Y., Tajika, E., 1997. The role of dissolved organic matter in the marine biogeochemical cycle: studies using an
ocean biogeochemical general circulation model. Global Biogeochemical Cycles 11, 599}612.
Yoder, J.A., McClain, C.R., Feldman, G.C., 1993. Annual cycles of phytoplankton chlorophyll concentrations in the
global ocean: a satellite view. Global Biogeochemical Cycles 7, 181}193.