ArticlePDF Available

The Simulation of SST, Sea Ice Extents and Ocean Heat Transports in a Version of the Hadley Centre Coupled Model Without Flux Adjustments


Abstract and Figures

Results are presented from a new version of the Hadley Centre coupled model (HadCM3) that does not require flux adjustments to prevent large climate drifts in the simulation. The model has both an improved atmosphere and ocean component. In particular, the ocean has a 1.25° × 1.25° degree horizontal resolution and leads to a considerably improved simulation of ocean heat transports compared to earlier versions with a coarser resolution ocean component. The model does not have any spin up procedure prior to coupling and the simulation has been run for over 400 years starting from observed initial conditions. The sea surface temperature (SST) and sea ice simulation are shown to be stable and realistic. The trend in global mean SST is less than 0.009 °C per century. In part, the improved simulation is a consequence of a greater compatibility of the atmosphere and ocean model heat budgets. The atmospheric model surface heat and momentum budget are evaluated by comparing with climatological ship-based estimates. Similarly the ocean model simulation of poleward heat transports is compared with direct ship-based observations for a number of sections across the globe. Despite the limitations of the observed datasets, it is shown that the coupled model is able to reproduce many aspects of the observed heat budget.
Content may be subject to copyright.
C. Gordon áC. Cooper áC. A. Senior áH. Banks
J. M. Gregory áT. C. Johns áJ. F. B. Mitchell
R. A. Wood
The simulation of SST, sea ice extents and ocean heat transports
in a version of the Hadley Centre coupled model
without ¯ux adjustments
Received: 1 October 1998 / Accepted: 20 July 1999
Abstract Results are presented from a new version of
the Hadley Centre coupled model (HadCM3) that does
not require ¯ux adjustments to prevent large climate
drifts in the simulation. The model has both an im-
proved atmosphere and ocean component. In particular,
the ocean has a 1.25°´1.25°degree horizontal resolu-
tion and leads to a considerably improved simulation of
ocean heat transports compared to earlier versions with
a coarser resolution ocean component. The model does
not have any spin up procedure prior to coupling and
the simulation has been run for over 400 years starting
from observed initial conditions. The sea surface tem-
perature (SST) and sea ice simulation are shown to be
stable and realistic. The trend in global mean SST is less
than 0.009 °C per century. In part, the improved simu-
lation is a consequence of a greater compatibility of the
atmosphere and ocean model heat budgets. The atmo-
spheric model surface heat and momentum budget are
evaluated by comparing with climatological ship-based
estimates. Similarly the ocean model simulation of
poleward heat transports is compared with direct ship-
based observations for a number of sections across the
globe. Despite the limitations of the observed datasets, it
is shown that the coupled model is able to reproduce
many aspects of the observed heat budget.
1 Introduction
Coupled ocean-atmosphere general circulations models
(OAGCMs) have become a valuable tool in attempting
to understand and predict climate change (Houghton
et al. 1996). One of the major drawbacks of these models
has been the large climate drifts that occur when they are
used to simulate the current climate. In many models
these drifts have been alleviated by the use of ¯ux ad-
justments (Sausen et al. 1988; Manabe et al. 1991; Johns
et al. 1997). Over the past few years a number of de-
velopments have been made to the Hadley Centre global
coupled model that is used in the investigation of an-
thropogenic climate change. Recent improvements to
both the atmosphere and ocean models have gone a long
way towards reducing the model climate drift, and
thereby obviated the need for ¯ux adjustment. Other
models have also recently demonstrated the ability to
produce stable climate simulations without ¯ux adjust-
ments (Barthelet et al. 1998; Boville and Gent 1998).
This work describes results from a simulation of pre-
industrial climate using the latest version of the Hadley
Centre coupled model (HadCM3), which does not use
¯ux adjustments. We concentrate on the analysis of the
mean simulation of sea surface temperature (SST), sea
ice extents, surface heat and momentum ¯uxes and
ocean heat transports, as these are some of the key cli-
mate variables for which, in the past, it has been dicult
to obtain a realistic and stable simulation.
It is important to establish that the model realises a
stable climate for sound physical reasons. In particular,
we consider the balance of heating terms and therefore
concentrate on the comparison with available observa-
tions of both the atmospheric model simulation of sur-
face heat ¯uxes and the ocean model simulation of heat
transports. The lack of salinity drifts are equally im-
portant in maintaining a stable climate simulation and,
as discussed brie¯y in a later section, the salinity drifts in
the model over 400 years do not lead to major changes in
the ocean circulation. The fresh water budget and the
ocean salinity simulation will be discussed in detail in a
separate paper.
Section 2 discusses climate drift in coupled models
and provides the framework for the structure of the rest
of the study. In Sect. 3 the various components of the
coupled model are described. Section 4 outlines the
initialisation of the model and the convergence of
Climate Dynamics (2000) 16:147±168 ÓSpringer-Verlag 2000
C. Gordon (&)áC. Cooper áC. A. Senior áH. Banks
J. M. Gregory áT. C. Johns áJ. F. B. Mitchell áR. A. Wood
Hadley Centre for Climate Prediction and Research,
Bracknell, UK
the coupled model state. The simulation of SST and sea
ice extents in the coupled model are discussed in Sect. 5.
Section 6 considers the surface ¯uxes of heat and mo-
mentum from a simulation of the atmospheric model
using observed SSTs. These are compared with estimates
of these ¯uxes from climatologies based primarily on
ship observations. The ocean heat transports in the
model are discussed in Sect. 7 and, ®nally, the results are
summarised in Sect. 8.
2 Climate drift in coupled models
The atmosphere and ocean interact via the interface at the
sea surface and the SST and sea ice extents are therefore
crucial predictands in a coupled model simulation. The
simulation of ocean temperatures and sea ice extents also
depends critically on ocean heat transports. In any ocean
region, in the long term annual mean, the net surface heat
exchange must be balanced by the ocean advection of
heat into (or out of) the region. In the zonal average this is
expressed by the simpli®ed equation for the depth and
zonally averaged ocean potential temperature:
acos /
cos /
where hidenotes a zonal and depth average, Q
is the
zonal mean heating at the surface, F
is the ¯ux of heat
across the latitude /and Hthe ocean depth. Other
variables are standard notation. Diusion eects have
been ignored in Eq. (1) since in most regions they are not
a major contribution to the model zonal mean heat
budget (see Fig. 17). It follows from Eq. (1) that there
will be a net drift in the depth averaged zonal mean
potential temperature (i.e. @hhi=@t6 0) if the ocean
meridional advection and net surface heating terms do
not balance. It is not a simple matter to relate the ver-
tically integrated temperature changes to SST changes,
Eq. (1) being only applicable to the vertical mean tem-
perature, and within the vertical column there may be
compensating temperature drifts. Equation (1) does,
however, provide a very basic framework for discussing
climate drift of ocean temperatures in a coupled model.
It is apparent that if the ocean heat transports
implied by the atmospheric model surface ¯uxes are
signi®cantly dierent to the actual ocean model heat
transports, then the coupled model ocean temperatures
must drift to enable a new balance to be established. If a
stable equilibrium exists then the SST and sea ice extents
must drift in such a way as to bring the model implied
heat transports to be in balance with the ocean heat
transports. (Here the terminology `implied' heat trans-
ports will be used for transports determined from the
integration of air sea ¯uxes and `actual' heat transports
those determined either from direct ocean observations
or from the calculation of the temperature advection
terms in the ocean model.) Weaver and Hughes (1996)
demonstrated that in a model with climate drift,
a `minimum' ¯ux adjustment could be applied which
simply ensured the ¯ux balance discussed in a zonal
mean sense. Recently Boville and Gent (1998) have, in
part, attributed their success in obtaining a realistic
un-¯ux adjusted coupled simulation to the compatibility
of the actual and implied heat ¯uxes (Bryan 1998).
The heat transports in the ocean model depend upon
the full three dimensional temperature, salinity and
current structure. In particular, salinity is important
because of its eects, via the density, on the general
circulation. It can also be important locally at the
surface in determining the mixed layer structure and
thereby aecting the mixed layer, and SST, response to
surface heat forcing. Even if the global-average salinity
is constant, a poor simulation of the precipitation,
evaporation, runo and ice melt and freezing can all lead
to local salinity drifts. The long term ocean response to a
poor simulation of surface water ¯uxes can be poten-
tially more serious than that for heat ¯uxes, because
there is no feedback on the ¯uxes from the surface sa-
linity as there is from surface temperature. In extreme
cases, over-freshening of the surface at high latitudes
could lead to a signi®cant reduction in the thermohaline
circulation. In HadCM3 systematic errors in the fresh
water budget do not lead to large changes in circulation.
We concentrate on the analysis of the heat budget.
Earlier versions of the Hadley Centre coupled model
(`UKTR' and `HadCM2'), which required ¯ux adjust-
ment, certainly showed an imbalance between the im-
plied and actual ocean heat transports. In these earlier
versions, SSTs were constrained to be close to reality by
application of a ¯ux adjustment in the surface heating
term. Figure 1 shows the ocean heat transport from two
coupled integrations with the earlier versions of the
Hadley Centre model. Details of HadCM2 can be found
in Johns et al. (1997) and UKTR in Murphy (1995).
Both the implied heat transports calculated from the
atmospheric model surface ¯uxes and the actual ocean
heat transports from the ocean model are shown in
Fig. 1. For comparison, ocean heat transport estimates
using the Da Silva et al. (1994) ¯ux climatology (here-
after referred to as DS) are also included and, following
DS, we assume a transport of 0.1 PW at 65°N. The
calculations based on the AGCM ¯uxes for both models
show the same overall pattern as the climatology in the
Northern Hemisphere. The HadCM2 ¯uxes are clearly
closer to the DS values. The dierences in the Southern
Hemisphere will be discussed further in Sect. 6. Figure 1
also clearly shows that the ocean models used in Had-
CM2 and UKTR, when integrated in ¯ux adjusted
mode, underestimate the ocean heat transports by nearly
a factor of two. The ¯ux adjustment term essentially
makes up the dierence. When the HadCM2 model
is integrated without ¯ux adjustments (Gregory and
Mitchell 1997) there is a large SST drift as the coupled
model attempts to reach an unrealistic balanced state.
These factors point to the fact that a realistic simu-
lation of both the surface heat ¯uxes and the ocean heat
148 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
transports is necessary to reduce climate drift in the
surface temperature and sea ice simulation. It is there-
fore important to evaluate the heat budget in both
component models rather than simply to assess the
coupled model SST and sea ice simulation. An initial
comparison of HadCM3 with available observations will
be presented here. Experience with the un-¯ux adjusted
HadCM3 model has also shown that, away from the
equator and strong current regions, the seasonal simu-
lation of SST is strongly dependent on the surface ¯uxes.
This dependence is in agreement with the observations
(Gill and Niiler 1973).
3 Model description
3.1 The atmospheric component
The atmospheric model component in HadCM3 is a version of the
UKMO uni®ed forecast and climate model run with a horizontal
grid spacing of 2.5°´3.75°and 19 vertical levels using a hybrid
vertical co-ordinate. The time step is 30 min. The performance of
this model version in a simulation forced by observed SSTs is
described in Pope et al. (submitted 2000).
Some of the major changes in this model over the previous
version used in coupled simulations (Johns et al. 1997) are now
1. A new radiation scheme has been included which has 6 (8)
spectral bands in the shortwave (longwave) and represents the
eects of minor trace gases as well as CO
O and O
. (Edwards
and Slingo 1996). A parametrisation of a simple background
aerosol climatology (Cusack et al. 1998) is also included.
2. The convection scheme has been improved by adding a pa-
rametrisation of the direct impact of convection on momentum
(Gregory et al. 1997).
3. A new land surface scheme (Cox et al. 1998) includes the
representation of the freezing and melting of soil moisture and the
formulation of evaporation includes the dependence of stomatal
resistance on temperature, vapour pressure and CO
4. A parametrisation of orographic drag (Milton and Wilson
1996) and a new gravity wave drag scheme including anisotropy of
orography, high drag states and ¯ow blocking, and trapped lee
waves have been included (Gregory et al. 1998).
5. The partitioning of mixed phase clouds into ice and water has
been changed from 0 to )15 °Cto0to)9°C (Gregory and Morris
1996) based on evidence from observational data (Moss and
Johnson 1994). A parametrisation of the eective radius of cloud
droplets as a function of cloud water content and droplet number
concentration is also included (Martin et al. 1994). Several pa-
rameters in the layer cloud scheme (Smith 1990) have been altered.
Cloud cover forms when the standard deviation of the distribution
of total water content in a grid box goes above a critical relative
humidity (RHcrit). RHcrit is a constant value at each level of the
model. In HadCM3 this has been changed from 0.85 to 0.7 to
improve the top-of-the-atmosphere (TOA) radiation balance. The
equation for liquid precipitation includes a threshold value of the
total water content, C
, below which water does not precipitate.
The value of this threshold is dierent over land and ocean to
represent the dierent amounts of cloud condensation nuclei. In
HadCM3, these values were reduced from 8 ´10
to 2 ´10
land and from 2 ´10
to 0.5 ´10
over sea. In combination,
these changes improved the net TOA radiative ¯uxes when com-
pared with Earth Radiation Budget Experiment (ERBE) data,
especially over northern mid-latitude oceans.
6. In HadCM3 there were some signi®cant revisions to the
boundary layer mixing used in HadCM2; most importantly the
removal of a non-local mixing scheme (Smith 1993), the inclusion
of which was found to degrade the transport and sink of aerosols.
In addition to these major model improvements, there have
been a number of minor changes. These are documented in Pope
et al. (1999).
3.2 The ocean component
The ocean model has been speci®cally developed over a number of
years for use in coupled climate simulations. The version used in
the simulation described is substantially dierent to that previously
employed in simulations with the Hadley Centre coupled climate
model. The basic model, which previously had a 2.5°´3.75°hor-
izontal resolution ocean component, is described in Johns et al.
(1997). Here the modi®cations made in the current version will be
focused upon.
The ocean component is a 20 level version of the Cox (1984)
model on a 1.25°´1.25°latitude-longitude grid. There are six
ocean grid boxes to each atmosphere model grid box and each high
latitude ocean grid box can have partial sea ice cover. The vertical
levels are distributed to provide enhanced resolution near to the
ocean surface and are the same as those in the previous coarser
horizontal resolution version of the model (Johns et al. 1997). The
topography was taken from the ETOPO5 (1988) 1/12°resolution
dataset and interpolated onto the model grid. A simple smoother
was applied to remove gridscale noise. Roberts and Wood (1997)
showed that Bryan-Cox type models are highly sensitive to the
depth of the various channels along the Greenland±Iceland±Scot-
land ridge. Many of these channels are sub-gridscale and so three
routes (one grid point wide on the velocity grid) through the ridge
were excavated. The Denmark Strait and Iceland-Faeroes ridge
were reduced to 797.9 m (bottom of level 12), while the Faeroe-
Scotland ridge was set to 534.7 m (bottom of model level 11). This
leads to a long-term mean out¯ow of approximately 8.5 Sv in the
coupled simulation, compared with the observed out¯ow of around
5±6 Sv (Dickson and Brown 1994). An island is also placed
at the North Pole to avoid the polar singularity in the spherical
co-ordinate system.
Horizontal eddy mixing of tracers is parameterised using the
Visbeck et al. (1997) version of the Gent and McWilliams (GM)
(1990) adiabatic thickness diusion and the Redi (1982) along is-
opycnal diusion scheme. The near surface vertical mixing is pa-
rameterised by a hybrid approach in which the mixing of tracers is
carried out via a Kraus-Turner mixed layer sub-model (Kraus and
Turner 1967) and a K-Theory scheme. The convective adjustment
is modi®ed in the region of the Denmark Straits and Iceland±
Fig. 1 The zonally averaged annual mean northward heat transport
as a function of latitude calculated from the atmosphere and ocean
components of previous ¯ux adjusted versions of the Hadley Centre
model (units PW). Climatological estimates based upon the Da Silva
et al. (1994) compilation of surface ¯uxes are also included
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 149
Scotland Ridge to better represent the down slope mixing of the
over¯ow water. The adjustment is modi®ed in this region since
previous studies have shown the simulation of the sub-polar gyre to
be sensitive to the details of the mixing of the over¯ow water
(Roberts et al. 1996). There is also a parametrisation of the out¯ow
of water from the Mediterranean in which water is partially mixed
across the Strait of Gibraltar. A detailed description of these
schemes can be found in Appendix A.
Shortwave solar radiation is selectively absorbed with depth
using a double exponential decay. Attenuation coecients appli-
cable to a global uniform Jerlov Type IB are used (Paulson and
Simpson 1977).
The sea ice model, which is the same as that used in HadCM2,
uses a simple thermodynamic scheme and contains parametrisa-
tions of ice drift and leads (Cattle and Crossley 1995). The model
description is repeated here for completeness. The thermodynamics
of the model is based on the zero-layer model of Semtner (1976). A
parametrisation of ice concentration based on that of Hibler (1979)
is included. Ice concentration is not allowed to exceed 0.995 in the
Arctic and 0.980 in the Antarctic since completely unbroken ice
cover is rarely observed in reality even in pack ice. Ice forms pre-
dominantly by freezing in the leads; it melts at the surface during
summer and at the base throughout the year. Ice depth can be
increased by the formation of `white ice' (Ledley 1985) where the
weight of snow forces the ice-snow interface below the water line.
The eect of sea ice formation and melt on ocean salinity is
accounted for within the model, assuming a constant salinity of
0.6&for sea ice. Sublimation increases ocean salinity, as the salt is
assumed to blow into leads, and white ice formation reduces it to
account for the salt added in converting snow to ice. Snowfall
reduces ocean salinity in leads, and accumulates onto ice, and all
rainfall is assumed to reach the ocean through leads.
Surface ¯uxes over the ice and leads fractions of each grid box,
and surface temperatures, are calculated separately within the at-
mosphere component of the model, assuming a linear temperature
pro®le in the ice. The surface albedo is 0.8 at )10 °C and below, and
between )10 and 0 °C it falls linearly to 0.5. This parametrisation
aims to reproduce some of the eect of the ageing of snow, the
formation of melt ponds, and the relatively low albedo of bare ice.
A windmixing energy (see Appendix A), used in the mixed layer
model, is calculated using the drag coecient appropriate for leads
and weighted by the leads fraction of the grid square. Oceanic heat
¯ux into the base of the ice is related to the temperature dierence
between the ocean top level and the base of the ice (assumed to
be at freezing point of )1.8 °C) with a coupling coecient of
20 Wm
A simple parametrisation of sea ice dynamics based on Bryan
(1969) is also included. The windstress is applied to the ocean be-
neath the ice. The ice thickness, concentration and snow depth are
advected using the top layer ocean current, using an upstream
advection scheme. Ice rheology is crudely represented by prevent-
ing convergence of ice once the ice depth reaches 4 m (Steele et al.
1997). The ice may become deeper than 4 m due to further freezing.
Del-squared horizontal diusion of ice depth is also applied, with a
coecient 2000 m
3.3 Coastlines and Indonesian through-¯ow
Although the ocean model has a 1.25°´1.25°horizontal grid, for
ease of coupling the coastline used in the coupled model is that of
the atmospheric model 2.5°´3.75°grid. In particular, this means
that there is only a very coarse representation of the channels be-
tween the Paci®c and Indian Oceans in the Indonesian through-
¯ow region. In reality the major gap for barotropic ¯ow from the
Paci®c to the Indian ocean is between Papua New Guinea (PNG)
and the Indonesian islands, with ¯ow between Australia and PNG
blocked due to shallow topography. Because of the coarse grid
coastline, ¯ow occurred through an unrealistic gap between Indo-
nesia and Asia in a preliminary setup of the model. In addition, the
minimum water depth of 140 m assumed in the model permitted
too much through-¯ow. To compensate for the coarse grid, baro-
tropic ¯ow is therefore only allowed in the model described here
between PNG and Indonesia.
3.4 Atmosphere-ocean coupling
The models are coupled once per day. The atmospheric model is
run with ®xed SSTs through the day and the various forcing ¯uxes
are accumulated each atmospheric model time step. At the end of
the day these ¯uxes are passed to the ocean model which is then
integrated forwards in time. The updated SSTs and sea ice extents
are then passed back to the atmospheric model. As there are six
ocean grid points to every atmospheric grid point interpolation
and/or averaging is used to transfer ®elds between the two grids
(see Appendix B for details).
River out¯ow is also included allowing ocean salinity feedbacks
via changes over land. Runo is converted into river out¯ow using
river catchments over land and associated coastal out¯ow points are
de®ned relative to the model grid. River transport is not modelled
explicitly, so runo is transported instantaneously to the coast.
3.5 Water conservation
The rigid lid in the ocean model means that the fresh water ¯ux at
the ocean surface does not lead to changes in the volume of water
in the column. In order to close the global salinity budget, the
conversion of the surface water ¯ux W, into an equivalent salinity
¯ux F
, is achieved by FsÿWS0=q0, where q0is a constant
reference density and S
a constant salinity of 35&.
Using the constant S
, instead of the spatially and temporally
varying S, means that there is no drift in global mean salinity when
there is no net water ¯ux, but the local eect of surface water ¯uxes
will be exaggerated in regions where S<S
, and under-repre-
sented when S>S
. The impact of this boundary condition on the
model simulation is generally small (Lowe and Gregory 1998).
A small amount of model rainfall runs into internal drainage
basins and a small term, equivalent to 0.01 mm day
, is applied
uniformly over the ocean to allow for this. Even with river runo
included, the area-average water ¯ux in HadCM3 coupled model is
not zero. Most of the imbalance, equivalent to a loss of
7.3 mm year
of water averaged over the ocean and a salinity drift
of 0.1&in 1000 years, comes from the accumulation of snow on
the Antarctic and Greenland ice-sheets, 80% on the former. (These
rates compare fairly well with those summarised in Houghton et al.
1996 and with other coupled models such as Russell et al. 1995.) In
the real world, fresh water is returned to the ocean by the calving of
icebergs, a process not represented in the model. The mass balance
of the ice sheets is not well known so we therefore assume that
accumulation should balance calving for each ice sheet, and apply
an appropriate water ¯ux uniformly over the areas of the adjacent
oceans where icebergs occur (Harvey 1976; Bigg et al. 1996). Of
course, the meltwater of the icebergs should not really be uniformly
distributed, but the model is not noticeably sensitive to the distri-
bution, furthermore the current state of observations and model-
ling does not justify any more elaborate treatment of this relatively
small term. The largest local values of the iceberg term are about
0.15 mm day
Over the North Atlantic iceberg region, these additional terms
are about 5% of the size of P-E. In Ban Bay and much of the
Arctic, they are between 10% and 20% and are also above 10% in
limited areas around Antarctica, and in small regions near the coast
above 20%. Over the majority of the world ocean, they are less
than 1% of the size of P-E.
A problem arises with the salinity simulation for small isolated
basins, such as the Caspian Sea, and others in the model, such as
the Baltic, where there is no mixing across the straits which connect
them to the world ocean. In these latter cases, the model would be
improved by including exchange across the straits (as done, e.g., by
Russell et al. 1995). In HadCM3, the salinity at every point is
constrained to remain within the limits of 0 and 40&. Some iso-
lated basins reach these limits; keeping them there implies a small
150 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
non-conservation of water amounting to 0.2 mm year
over the ocean.
These issues associated with water conservation and its eects
on the salinity drifts in the model were masked in earlier model
versions by the use of salinity ¯ux adjustment terms.
4 Coupled model initialisation and convergence
The coupled model was initialised with an atmospheric model state
appropriate to mid-September and ocean temperatures and salini-
ties were speci®ed from the Levitus and Boyer (1994) and Levitus
et al. (1995) mean climatology for the month of September. Ocean
currents were zero at the initial time. Initial sea ice extent, thickness
and concentration were taken from a previous model simulation.
The coupled model simulation was integrated for over 400 years
and most of the results that follow in the subsequent sections show
®elds averaged over the years 81±120 and years 361±400. It is
noteworthy that no prior stand-alone ocean spin-up of the atmo-
sphere and ocean/sea ice models is needed to achieve a balanced
initialised state for use in the coupled model.
Comparison of the coupled model simulation with observa-
tions is complicated by the pre-industrial trace gases that are used
in this simulation of a `control' climate. The model is simulating
the pre-industrial period and, as such, data from this time are
ideally needed for model evaluation. Such data is not available and
more recent data must be used. The model in also initialised with
the Levitus and Boyer (1994) and Levitus et al. (1995) temperature
and salinity climatology since a pre-industrial initial dataset is
also not available. These complications need to be considered in
assessing the comparisons that follow. In general, however, com-
parison of the GISST SSTs for the 1870s and the 1980s shows that
the local model drifts discussed in later sections are much larger
than the dierences expected because of the mis-matching of time
The time series of global mean SST anomaly and top of the
atmosphere (TOA) radiative heat ¯ux are shown in Fig. 2. The SST
anomaly is de®ned as the dierence between the model SST and the
GISST2.2 values (Rayner et al. 1996). The GISST temperatures are
used for comparison as they are based on surface ship and satellite
observations and therefore have a much better data coverage than
the Levitus XBT based climatology. The annual mean GISST and
Levitus and Boyer (1994) and Levitus et al. (1995) SST climatol-
ogies are in agreement over most regions of the globe to within a
few tenths of a degree centigrade. Larger dierences of around one
degree occur in the Nordic, Barents and Kara Seas and in the
Southern Ocean. The dierences in the climatology do not aect
the conclusions drawn about the major model systematic SST
errors highlighted in later sections.
The pre-industrial trace gases used will lead to the model being
approximately 0.3 °C colder than the present day GISST temper-
atures (Houghton et al. 1996). Figure 2 shows that the global SST
in the coupled model remains within 0.3 °C of climatology over the
400 years of the simulation and there is little sign of a long term
systematic drift. Certainly the drifts are much smaller than the
observed rate of global warming. Over the 400 years, the global
mean model SST drift is )0.009 °C per century.
There is considerable interannual and interdecadal variability
evident in the time series. The major characteristics of the geo-
graphical pattern of simulated SST are also very stable after the
®rst few decades (to be discussed in the next section). The TOA ¯ux
is also generally less than 0.5 Wm
. Over most of the latter part of
the run it is negative, indicating a net cooling of the climate system.
The temperature and salinity drifts below the ocean surface are
illustrated in Fig. 3a, b. In the upper 1000 m for the ®rst 50 years of
the simulation there is a drift in temperatures away from the initial
Levitus and Boyer (1994) and Levitus et al. (1995). After this time
the drift at these depths becomes considerably smaller with maxi-
mum temperature dierences 0.5 °C too warm at 200 m in the
zonal mean. The source of this warming is not clear and is being
further investigated. In the deep ocean there is a slow cooling which
will have a very long equilibration time scale. The time series of
global volume weighted temperature in Fig. 3c shows there is a net
cooling trend of 0.02 °C per century. This cooling trend is consis-
tent with the net heat loss indicated by the TOA ¯ux in Fig. 2.
The corresponding drift in salinity is shown in Fig. 3b. After
400 years the depth distribution of the salinity drift shows an in-
crease below 1000 m and a freshening above this depth. Over the
400 years of the integration the global average surface salinity
freshens by nearly 0.8&, although 50% of this drift occurs in the
®rst 100 years. North Atlantic salinity drifts are particularly im-
portant because of their potential impact on the thermohaline
overturning circulation. In the North Atlantic the surface salinity
freshens by approximately 0.2&over the 400 years of the inte-
gration, whereas at 500 m there is an increase in salinity of 0.6&
over a shorter period of 50 years. It will be shown later that
the overturning cell in the North Atlantic is stable throughout the
integration, demonstrating the relatively small impact of these
salinity drifts on the circulation.
The time series of global mean sea ice area and volume are
shown in Fig. 4a, b. Monthly and annual mean values are shown.
The ice area (Fig. 4a) reaches a stable extent in the ®rst few years of
the simulation. The ice volume (Fig. 4b) is also stable but it takes a
longer time to reach its long term value. In the ®rst 30 years of the
simulation there is a net increase in ice volume indicating an
increase in the ice thickness away from its initial value (which was
speci®ed from a previous coupled simulation). A comparison with
observations of the geographical distribution of sea ice in the model
simulation is made in Sect. 5.
Overall we see that after 400 years, the global mean upper ocean
and sea ice ®elds have reached a stage of relatively small drift. This
capability to start from the observed ocean state (Levitus and
Boyer 1994; Levitus et al. 1995) and to maintain this without large
drifts over 400 years is a notable success of the model, particularly
as no spin-up is required to precondition the coupled model. How
much of the cooling in the deep ocean is a consequence of the pre-
industrial trace gas concentrations used has not been established.
Since a climatology of ocean conditions in pre-industrial times
is not available, proxy data and modelling experiments will be
necessary to determine this.
5 Simulation of the sea surface temperature
and sea ice extents
The SST and sea ice extents, both in reality and in the coupled
model, are the central variables through which the atmosphere/
ocean/ice coupling takes place. A realistic simulation of these
Fig. 2 Time series of the annual global mean SST dierence (°C)
from the GISST climatology and the annual global mean top of
atmosphere ¯ux imbalance (Wm
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 151
Fig. 3 a Time/depth series of annual global mean temperature drift
(°C). Dierences are calculated from the Levitus and Boyer (1994 and
1995) climatology and the upper 1000 m has an expanded depth scale.
bAs abut for salinity drift (
), ctime series of ocean global volume
weighted temperature (°C)
Fig. 4 a Time series of global
monthly (thin line) and annual
(thick line) mean sea ice area
). bTime series of
global monthly (thin line) and
annual (thick line) mean sea ice
volume (10
152 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
variables is of prime importance in determining the usefulness of
the coupled model for climate simulations.
5.1 Sea surface temperature
The simulation of annual mean SST is shown in Fig. 5a averaged
over the years 361±400. For comparison the same ®eld from the
GISST2.2 observed climatology is shown in Fig. 5b. All of the
major observed features in the SST ®eld are reproduced. In par-
ticular, the model is able to maintain the sharp horizontal gradients
in SST associated with the major ocean currents such as the North
Atlantic Current (NAC), Kuroshio and Antarctic Circumpolar
Current (ACC). These gradients are maintained in the HadCM3
simulation because of the 1.25°resolution in the ocean model
whereas, in earlier non-¯ux adjusted versions of the coupled model,
with a low resolution ocean, these gradients were very diuse.
Of course, in ¯ux adjusted models the gradients are arti®cially
maintained by the ¯ux adjustment terms.
Dierences between the model SSTs and the GISST climatol-
ogy of SST are shown in Fig. 6a, b for the time periods covered by
the years 81±120 and 361±400. Over much of the ocean the coupled
model after 400 years is simulating the SSTs to within 1 °C of their
observed values. There are, however, some notable regions where
the dierences are considerably larger than 1 °C. A comparison of
the SST dierence ®elds for the two time periods shows the pattern
is very stable, with the ®elds being similar in Fig. 6a, b. One notable
exception is the greater cooling in the North Atlantic in the later
period. The NAC is too zonal in this later period (Fig. 5a) com-
pared to the earlier time, and this leads to the cold SSTs in the
region where the NAC in reality turns north. This SST drift does
not lead to large change in the Atlantic thermohaline overturning
circulation or the poleward heat transports between the two time
periods (see later sections and Figs. 19 and 20).
The major systematic error throughout the simulation is the
cooling in the North Paci®c, where the SSTs are too cold by more
than 3 °C over a large region. This cooling establishes itself in
the ®rst decade of the coupled simulation and its pattern and
magnitude remain essentially constant throughout the integration.
Analysis of the cooling has pointed most strongly to the poor
simulation of surface heat ¯uxes in this region, combined with the
very shallow summer-time mixed layer depths occurring in the
North Paci®c (in the model and in reality). There is also a local
maxima in the cooling in the region of the Kuroshio, which sepa-
rates from the coast too far south in the model. As the Kuroshio
and its extension are associated with high horizontal SST gradients,
shifts in their position lead to large local SST errors.
The North Atlantic is reasonably well represented by the cou-
pled model, other than some large local errors associated with shifts
in high gradient regions,. The tracking of the North Atlantic cur-
rent shows to some extent the northward turn in the isotherms to
the east of the Grand Banks but this is considerably underestimated
compared to reality, especially in the later time period.
The model equatorial central east Paci®c is too cold, whereas
south of the equator the model is too warm in the eastern tropical
Paci®c and Atlantic. This is to be expected from the lack of sim-
ulated stratocumulus cloud in the atmospheric model which will be
discussed in Sect. 6. The region o the Californian coast is also too
warm, probably for the same reason. The too cold equatorial SSTs
are consistent with the excessive equatorial easterlies simulated in
the atmosphere-only simulation described later.
The ®nal major error is the warming in the Southern Ocean.
There are a number of complex processes that determine the SST in
this region. In addition to the surface heat ¯uxes and advection by
the ACC, the seasonal mixing of surface waters with the warmer
underlying Circumpolar Deep Water (CDW) also eect the SST. In
the south, the ocean is partly covered by ice in winter and the heat
used in melting the ice in spring and summer also contributes to the
mixed layer heat budget. The model has excessive shortwave ¯uxes
in this region, associated with too little cloud, and this will con-
tribute to the warming. However, sensitivity experiments have
shown that the SST warming is also aected by the details of the
vertical mixing and the heat lost from the ocean in melting ice.
The peak SST errors in the Southern Ocean are in regions of
tight SST gradient and are associated with the incorrect positioning
of these gradients (compare the SST ®elds in Fig. 5a, b).
Fig. 5 a The coupled model annual mean SSTs (°C) averaged over
the years 361±400. bThe annual mean SST (°C) from the GISST2.2
climatology. In both ®gures the contour interval is 4 °C
Fig. 6 a The model minus climatology annual mean SST (°C)
dierence averaged over the years 81±120. The contour interval is
1°C. bAs abut for the model years 361±400
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 153
These are also recognised as regions of strong topographic steering
of the ACC, and the peak SST errors may well be associated with
the incorrect steering of the current in the model. A parallel sim-
ulation using an ocean model with 40 vertical levels (the version
described here has 20 levels) indeed led to some reduction in the
peak errors in the Southern Ocean. In the 40 level model the near
bottom resolution was doubled leading to a better representation of
topographic steering.
One region of particular note is immediately to the south of the
Cape of Good Hope in Southern Africa where there is a signi®cant
SST error in Fig. 6a, b. This region is known to be high in eddy
activity and it has been suggested that there is a substantial eddy
heat ¯ux through this area (Thompson et al. 1997). The lack of
eddy resolution may be particularly important in this region.
The overall SST error pattern described above is similar in
many aspects to that found in both the NCAR CSM1 (Boville and
Gent 1998) and ARPEGE T42/OPAICE (Barthelet et al. 1998)
non-¯ux adjusted coupled models. The similarity between the
models suggests that the regions described above may be dicult to
simulate because of the physical processes governing the SST in
these areas, or that the models described have common errors in
the simulation. Overall there is a vast improvement in the SST
simulation compared to the earlier non-¯ux adjusted version of the
Hadley Centre model (HadCM2) with a low resolution ocean
component (Gregory and Mitchell 1997).
5.2 Sea ice
Because of the importance in climate change of the insulation of the
ocean surface by ice cover and the ice-albedo feedback, a realistic
simulation of sea ice distributions is of critical importance in the
coupled model. Time series of sea ice area and volume for both
hemispheres throughout the coupled simulation are shown in
Figs. 7 and 8. The observed SSM/I (Special Sensor Microwave
Imager, NSIDC 1989) maximum and minimum extents are also
marked on the ®gures. In the Arctic the sea ice extents are over-
estimated at the maxima when compared to the SSM/I data
(Fig. 7a). The seasonal minimum is simulated very well. The reason
for this discrepancy in maximum extent can be understood by
looking at the modelled geographical distribution of sea ice. Fig-
ure 9a shows the simulated annual mean ice concentrations for the
Northern Hemisphere for years 361±400. The satellite estimates
from SSM/I are also shown in Fig. 9b. It can be seen by comparing
Fig. 9a and b, and the seasonal cycle of Northern Hemisphere ice
area (Fig. 10a), that the winter ice extends too far into the North
Paci®c and that the Barents Sea remains ice covered throughout the
year. These contribute to the overestimation of the ice extent
maxima. The Northern Hemisphere ice volume (Fig. 7b) initially
increases and then stabilises after about 30 years.
The Northern Hemisphere sea ice simulation is much improved
over that obtained with un-¯ux adjusted earlier versions of the
Hadley Centre model which employed a low resolution ocean
component. In these earlier versions the ice extents were too exten-
sive, reaching south of Iceland in winter. One of the major impacts of
using the 1.25°´1.25°ocean model is a much improved simulation
of the North Atlantic Current which advects warmer water into the
Norwegian Sea and therefore reduces ice cover in this region. Once
this warmer water has entered the basin it should, in reality, ¯ow
northwards in the Norwegian Current and then split, one compo-
nent heading northwards as the West Spitsbergen current, another
¯owing eastwards into the Barents Sea (P®rman et al. 1994). In the
model almost the entire ¯ow is northwards, with very little of this
water getting onto the continental shelf. This failure is partly due to
the model's lack of Svalbard as an island. Sensitivity studies have
shown that including the island improves the simulation in the re-
gion. As already described, without the warm North Atlantic water
reaching the Barents Sea, sea ice is too extensive in this region.
The Southern Hemisphere ice extents are simulated well com-
pared to climatology, both in the seasonal variation (Figs. 8a and
10b) and in the mean distribution (Fig. 11a, b). The mean seasonal
cycle depicted in Fig. 10b shows the southern hemisphere winter-
time sea ice extents to be too large in the model. There is an
adjustment time of around 100 years over which the Antarctic sea
ice volume increases towards its long term stable value (Fig. 8b).
The general features of the ice motion (not shown) are rea-
sonably well simulated compared with Emery et al. (1997).
Advection on the east Greenland coast has reasonable speeds
(5±10 cms
) but it is restricted to a very narrow current; this may
be why export through the Fram Strait is only 0.03 Sv compared
to the 0.09 Sv suggested by Aagaard and Carmack (1989). The
Beaufort Gyre is too strong which may well be associated with the
treatment of the surface stress in the very simple representation of
ice dynamics. The Transpolar Drift has to go round the polar is-
land. The simulation correctly shows westward motion around the
Antarctic coast but, because of the lack of Coriolis turning in the
Fig. 7 a Time series of
Northern Hemisphere
monthly (thin line) and an-
nual (thick line) mean sea ice
area (10
). The observed
seasonal extents (dashed
lines) are from SSM/I data
(NSIDC 1989). bTime series
of Northern Hemisphere
monthly (thin line) and
annual (thick line) mean sea
ice volume (10
154 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
simple ice model, the ice velocity tends to have an onshore com-
ponent and hence divergence and ice production around the coast
are too weak. Further simulations, with a revised free drift ice
formulation in which the wind and water stress on freely moving
ice are correctly represented, show an improved simulation of the
ice velocity ®eld in this region.
It is important to establish that, in addition to the good simu-
lation of SST and sea ice extents, that the simulated surface heat
¯uxes and ocean heat transports are within the bounds set by
the available observations. The following two sections therefore
consider the evaluation of the momentum and heat budget in the
coupled model.
6 Comparison of observed and modelled surface ¯uxes
In an atmospheric model, even when the SSTs are
speci®ed at their observed values, the calculation of the
surface ¯uxes of momentum (i.e. wind stress), heat and
fresh water depends upon a range of simulated model
variables including low level winds, air temperatures and
humidities, as well as the cloud cover throughout the
atmospheric column. These surface ¯uxes have been
dicult to simulate realistically in climate and numerical
weather prediction models (White 1996). It is these
¯uxes which constitute the forcing at the ocean surface
in the coupled model and it is therefore important that
they are accurately represented. The degree of accuracy
required depends upon the sensitivity of the ocean sim-
ulation to the various forcings. Considerable work has
been done to investigate this ocean sensitivity to wind
stress forcing in the tropics (e.g. Harrison et al. 1989)
but little has been done over the global oceans.
We consider the mean surface ¯uxes from a `stand
alone' integration of the atmospheric component of the
coupled model for the 10 year period 1979 to 1988. The
SSTs for this experiment were those used in the ®rst
Atmospheric Model Intercomparison Project (AMIP-
Gates 1992) and in the remainder of the work this sim-
ulation will be referred to as `the AMIP simulation'. We
Fig. 8a, b As Fig. 7 but for
the Southern Hemisphere
sea ice
Fig. 9 a Modelled annual
mean Northern Hemisphere
sea ice concentrations for the
years 361±400. Concentra-
tion is plotted as a percent-
age and the contour interval
is 20%. bAs abut from
SSM/I data
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 155
concentrate on the ¯uxes from this atmosphere simula-
tion, rather than those of the coupled simulation, in
order to help isolate systematic errors in the component
models of the coupled system. Brief comparisons will
also be made with coupled model surface ¯uxes.
This atmospheric model version is identical to that
used in the coupled simulation apart from the speci®-
cation of atmospheric trace gases (i.e. no additional
`tuning' was carried out). The AMIP simulation uses
trace gas concentrations appropriate to the 1979 to 1988
period, whereas the coupled model uses pre-industrial
values. The simulation of the atmospheric ®elds in the
AMIP simulation is described in Pope et al. (1999).
Although our primary interest concerns the surface
heat budget we will also attempt to evaluate the surface
wind stress as this is clearly an important forcing func-
tion for the ocean. The freshwater budget will be eval-
uated in a separate paper. The surface wind stress ®elds
from the model are compared with the climatologies of
Hellerman and Rosenstein (HR) (1983) and Josey et al.
(1996). The surface heat ¯uxes are compared with the
climatological estimates of DS and Josey et al. (1996).
The evaluation of the surface ¯uxes simulated by the
atmospheric component of the coupled model is limited
by the large uncertainties in the climatological estimates
of these ¯uxes. Throughout the discussion that follows
we will concentrate on estimates of the annual mean
¯uxes simply for clarity and brevity of presentation.
These are also the ®elds that are of most relevance to
the long term climate drift in the model. However, we
note that the seasonal evolution of SST is also crucially
dependent on the surface ¯ux simulation (Gordon and
Bottomley 1985).
6.1 The surface wind stress
The annual mean surface wind stress from the AMIP
simulation is shown in Fig. 12a and the climatology of
HR is shown for comparison in Fig. 12b. Overall the
patterns of wind stress are well simulated. The trade
wind regions have higher stresses in HR but this dier-
ence is probably within the uncertainty in the climato-
logical estimates. Figure 12c shows the equivalent stress
®elds from the more recent estimates of Josey et al.
(1996), and the model is in better agreement with this
climatology in the trade wind regions of both hemi-
Major dierences occur in the Southern Ocean where
the model stresses are considerably larger than either
climatology although very little data is available in this
region. Comparison of low level model winds with
equivalent ®elds from the ECMWF re-analysis clima-
tology (ERA, Gibson et al. 1997) show a good agree-
ment between the climate model and analysis ®elds in
the Southern Ocean although, of course, the ERA will
be largely unconstrained in this region and the analysis
Fig. 10 a The mean monthly seasonal cycle of Northern Hemisphere
sea ice area for the model years 361±400 and from SSM/I data. Units
are 10
.bAs abut for the Southern Hemisphere
Fig. 11 a Modelled annual
mean Southern Hemisphere
sea ice concentrations for the
years 361±400. Concentra-
tion is plotted as a percent-
age and the contour interval
is 20%. bAs abut from
SSM/I data
156 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
may re¯ect systematic errors in the ECMWF model.
Further data studies are required to be able properly to
evaluate the model in this region.
There are also signi®cant dierences between the
model and climatology in the North Atlantic where the
simulated stresses in the westerlies are too weak. This
is associated with a general high pressure bias at high
latitudes, leading to too weak surface pressure gradients
in mid-latitudes (see Pope et al. 1999 for details).
In the equatorial regions the Hellerman stresses are
known to be too large (Josey et al. 1996). The model
equatorial stresses appear to be too strong in the Paci®c
when compared with the ECMWF analysis and ship
winds (not shown). This can, in part, explain the too
cold SSTs in the modelled equatorial Paci®c (Fig. 6).
The annual mean wind stress curl from the AMIP
model and HR is shown in Fig. 12d, e. The overall
pattern of the curl is well reproduced by the model and
the positioning of the zero curl is represented realisti-
cally. In the north Paci®c the zero windstress curl in part
de®nes the separation point of the Kuroshio western
boundary current (Hulbert et al. 1996). In reality, and in
the coupled model (see Sect. 5), the Kuroshio extension
is approximately coincident with the region of high
horizontal SST gradient. The overall curl magnitudes
are generally smaller in the model, compared to HR,
which is consistent with the dierences already noted in
the stress magnitudes. The annual mean wind stress curl
from the coupled model, meaned over the years 81±120,
is shown in Fig. 12 f. The wind stress curl zero line has
moved south in the western North Paci®c when com-
pared to the atmosphere only AMIP simulation de-
scribed already with prescribed SSTs. This movement is
consistent with the coupled model SST errors in the
Kuroshio extension region of the North Paci®c (Fig. 6a,
b). Additional sensitivity experiments in which the ocean
model was forced with HR climatological stresses sug-
gest that this southward movement of the Kuroshio
Fig. 12 a Ten year annual mean surface stress (Nm
AMIP simulation with the HadAM3 model. bAnnual mean surface
stress from the Hellerman and Rosenstein (1983) climatology (Nm
cAnnual mean surface stress from the Josey et al. (1996) climatology
). dTen year annual mean wind stress curl from the AMIP
simulation with the HadAM3 model. (The contour interval is
0.5 ´10
and negative regions are shaded.) eAnnual mean
wind stress curl from the Hellerman and Rosenstein (1983)
climatology. (The contour interval is 0.5 ´10
and negative
regions are shaded.) fAnnual mean wind stress curl from years
81±120 of the HadCM3 coupled simulation (The contour interval
is 0.5 ´10
and negative regions are shaded.)
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 157
separation is a consequence of the model wind stress
distribution. With the HR stresses the separation point
was located further to the north and therefore closer to
its observed position. In both the atmosphere only and
coupled model there are considerably higher meridional
gradients of wind stress curl in the Southern Ocean,
compared to HR, and these are associated with the
strong model stresses in this region.
6.2 The net surface heating
The net heat ¯ux at the surface has four components; the
net shortwave and longwave ¯uxes, the latent heat ¯ux
associated with surface evaporation, and the sensible
heat ¯ux. The shortwave and latent heat ¯uxes are the
dominant contributions both in terms of magnitude and
geographical variation. The net annual mean surface
heating from the AMIP simulation and the corrected DS
climatology are shown in Fig. 13a, b respectively. The
corrections to the DS climatology are made to ensure a
global balance of the surface heat budget (see Da Silva
1994 for details). A contour interval of 50 Wm
been used in order to highlight the main features. The
dierence plot in Fig. 13c has a smaller contour interval
(20 Wm
) to show greater spatial detail.
From Fig. 13a, b it can be seen that the major pattern
features are well simulated by the model. In detail there
are, however, some notable dierences evident in the
dierence ®eld in Fig. 13c. For comparison, Fig. 13d
shows the dierence between the Josey et al. (1996) un-
corrected total heat ¯ux and the same quantity from the
DS climatology. The uncorrected climatologies are used
so that the dierences in Fig. 13d are only due to the
detailed calculation methods and not the correction pro-
cedures. There is a dierence between the climatologies of
around 10±20 Wm
over much of the ocean which gives
some measure of the level of uncertainty in the observa-
tional estimates. Note also, particularly in the Southern
Ocean and the sub-tropical North Paci®c, that the model
is in better agreement with the SOC climatology.
The dierence ®eld in Fig. 13c shows that in the
tropical east Paci®c and Atlantic of the Southern
Hemisphere the model has too much heat into the ocean
surface. This is particularly marked on the Equator in
the east Paci®c, where the climatology shows the maxi-
mum heating some distance from the coast (Fig. 13b),
whereas the model shows a large heating all the way to
the coast (Fig. 13a). This feature is associated with the
poor simulation of marine stratocumulus in these areas.
A similar feature is also evident in the tropical Atlantic
stratocumulus region. Figure 14a shows the dierence in
the net shortwave ¯ux between the model and DS. Both
the Atlantic and Paci®c regions referred to stand out as
having an excessive shortwave heat ¯ux at the surface.
Note that the Californian stratocumulus region also
shows up clearly in the shortwave dierence map. As
shown earlier, when this atmospheric model is coupled
to the ocean, these stratocumulus regions show SSTs
that are too warm (see Fig. 6a, b).
Some other notable dierences in Fig. 13c are the
tropical and sub-tropical regions in the North Paci®c
Fig. 13 a Ten year annual mean of the net surface heat ¯ux from the
AMIP simulation with the HadAM3 model (contour interval is
50 Wm
). bAnnual mean surface heat ¯ux from the Da Silva et al.
(1994) climatology (contour interval is 50 Wm
). cDierence of
model minus Da Silva et al. (1994) annual mean net surface heat ¯ux
(contour interval is 20 Wm
). dDierence between the Josey et al.
(1996) and Da Silva et al. (1994) annual mean uncorrected
climatologies of net surface heat ¯ux (contour interval is 10 Wm
158 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
where there is more heat entering the ocean in the model
than suggested by DS. Immediately under the peak of the
trades (i.e. the region of maximum trade wind stress in
Fig. 12, centred on 20°N, 135°W) this bias is reversed
and the model exhibits a greater cooling than DS. In this
region the overall increased heating over a broad area is a
result of the larger surface short wave ¯ux in the model
(Fig. 14a), which is compensated for in the peak of the
trades by enhanced surface evaporation (Fig. 14b). In the
Indonesia/east Indian Ocean region Fig. 13c shows that
the model gains too much heat, mainly as a consequence
of the enhanced short wave ¯ux, by more than 50 Wm
in some areas.
The dierence between the mean net surface heating
in the coupled model for the years 361±400 and the
AMIP simulation of the same quantity (Fig. 13a) is
shown in Fig. 15. This ®gure shows the impact of the
coupled model drift in SST on the net surface heat ¯ux
simulation. The region of strong equatorial Paci®c
heating extends further west in the coupled model and is
more intense due to the excessive upwelling of cold water
on the equator, which leads to a reduced surface evap-
oration. The excessive upwelling is consistent with too
strong low level equatorial winds. Note also the large
¯ux dierences associated with the changed position of
sharp horizontal SST gradient in the model. These re-
gions are associated with the extension of the northern
oceans' boundary currents and the ACC. In the North
Paci®c, where there is the largest SST error in the model,
there is less heat loss in the coupled model than in the
AMIP simulation due to a reduction in surface evapo-
ration as the modelled SST cools. Note also the greater
cooling in the Paci®c and Atlantic stratocumulus regions
associated with the increased SST, and therefore surface
latent heat loss, in the coupled model.
The large uncertainties in the climatological estimates
need to be considered when assessing the comparisons
made above. In particular, from a comparison of
Fig. 14a, b, showing the shortwave and latent heat ¯ux
dierence ®elds between the model and corrected DS, it
is apparent that there is a net bias between the model
and DS. The model shortwave ¯uxes are everywhere
larger than those from DS, whereas the latent heat ¯uxes
are in many regions larger in the model, leading to some
cancellation. The net longwave heat losses (not shown)
are also generally larger in the model than in corrected
DS, globally by about 7 Wm
Both the climatological and modelled surface heat
¯uxes can easily be integrated to give the implied ocean
heat transports. These transports are plotted in Fig. 16
(coupled model transports are also shown on the ®gure
and will be discussed in the next section). In both cases
for simplicity of comparison the heat transport at 65°N
was assumed to be 0.1 PW (Da Silva et al. 1994). The
un-corrected DS climatology has a global imbalance of
30 Wm
and the heat transports from the un-corrected
®elds are therefore meaningless. The DS heat transport
curve plotted in Fig. 16 is from the corrected climatol-
ogy which has a constraint of a zero global mean net
heating imposed. The implied heat transport curve
is therefore closed. The AMIP simulation, which used
present day trace gas amounts and SSTs, has a net TOA
imbalance of 3.5 Wm
, much of which will be re¯ected
in the imbalance of the global mean surface heating
(Pope et al. 1999) and the implied ocean heat transport
is therefore not closed. For comparison with DS the
model curve included in Fig. 16 uses the AMIP model
transports that have been adjusted to ensure a zero
global mean net surface heating.
There is considerable similarity between the corrected
transports from the model and climatology. The largest
dierences occur in the Southern Hemisphere, in par-
ticular, in the sub-tropics. The role of the excessive heat
Fig. 14 a Dierence between the ten year annual mean HadAM3
AMIP simulation of surface short wave ¯ux and the Da Silva et al.
(1994) climatology of the same quantity (the contour interval is
10 Wm
and values dierences greater than 20 Wm
are shaded).
bAs abut for the surface latent heat ¯ux
Fig. 15 Dierence between the HadCM3 coupled model (year 361±
400) and AMIP simulation of net surface annual mean heat ¯ux
(contour interval is 40 Wm
and dierences greater than 20 Wm
are shaded)
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 159
input in the stratocumulus (Sc) regions in the AMIP
simulation in determining the discrepancies in the
Southern Hemisphere sub-tropical transport can be as-
sessed by arti®cially correcting the heat input in these Sc
regions back to DS. This calculation suggests that be-
tween the equator and 15°S an additional ¯ux of ap-
proximately 0.4 PW of heat is entering the ocean in the
AMIP simulation compared to DS. At 15°S this would
account for 80% of the discrepancy in the implied ocean
heat transport shown in Fig. 16. In part, the dierent sign
of the heat transport in the Southern Ocean can be ac-
counted for by the lower sub-tropical values. The correct
slope of the transport curve between approximately 15°S
and 40°S indicates an agreement between the model and
DS ¯uxes. In addition, between 40°S and 60°S the net
surface heating dierence ®eld between the model and DS
(Fig. 13c) shows the model losing more heat to the at-
mosphere than the climatology, and this is re¯ected in the
implied ocean heat transports. In fact, the DS climatol-
ogy shows a net heat gain in this region. The observed
¯uxes in the Southern Ocean should be treated with
considerable caution as the data coverage is very sparse.
7 Comparison of observed and modelled
ocean heat transports
To complement the discussion of the atmospheric model
surface heat budget, in this section we consider the
ability of the ocean model to reproduce the geographical
distribution of global ocean heat transports when com-
pared with direct ocean observations.
7.1 Global distribution of poleward heat transports
In the AMIP simulation present day SSTs and atmo-
spheric greenhouse gas concentrations were used. In the
coupled model pre-industrial gas concentrations are
employed which reduce the TOA ¯ux imbalance by
approximately 1.5±2.0 Wm
. Therefore the global
mean imbalance at the start of the coupled simulation
is approximately half the imbalance in the AMIP
experiment. As the coupled simulation proceeds, the
TOA imbalance quickly reduces to less than 0.5 Wm
(Fig. 2). This happens due to the rapid adjustment of the
SSTs in the coupled model.
Shown on Fig. 16 are a number of estimates of model
poleward ocean heat transport using net ¯uxes from the
atmosphere component of the coupled model from near
the beginning of the coupled HadCM3 simulation, and
after 100 and 400 years. Note also that in Fig. 16 the
assumption of 0.1 PW heat transport at 65°N has not
been made in plotting the curves from the coupled
model, in fact, the model suggests a transport of ap-
proximately 0.25 PW at this latitude. For comparison
the ocean heat transports as calculated within the ocean
component of the coupled model are also included in
Fig. 16 for the same time periods. As noted earlier in
Sect. 2, the long term mean estimates of heat transports
from the surface ¯uxes and from the ocean model must
agree as the model reaches equilibrium. A critical issue is
whether this state can be achieved whilst maintaining a
realistic climate simulation in other respects.
Although the actual and implied heat transports
dier by up to 0.2 PW in the sub-tropics, comparison of
Fig. 16 with the equivalent plot for the earlier ¯ux ad-
justed versions (HadCM2 and UKTR) in Fig. 1, shows
that there is a much greater consistency in the new
model between the atmospheric model surface ¯uxes and
the ocean heat transports.
The modelled mean global heat transports for the
years 361±400 are shown again in Fig. 17 and are bro-
ken down into component parts due to the zonal mean
meridional overturning circulation, the gyre circulation
(de®ned as the deviation from the zonal mean), the
along isopycnal diusive component and the eddy
transports due to the Gent and McWilliams (1990)
The meridional overturning ¯ux contains the contri-
bution from the near surface Ekman transport, and this
dominates in the tropics. In the model the contribution
from the along isopycnal diusive term and from the
bolus velocity advection is small at most latitudes, and
only makes a signi®cant contribution to the zonal mean
in the region of the Deacon Cell in the Southern Ocean.
The smallness of the bolus advection term should not be
interpreted as meaning the Gent and McWilliams (1990)
parametrisation does not in¯uence the heat transports.
What it shows, at least in this model, is that the most
important aspect of the Gent and McWilliams (1990)
parametrisation is its ability to better maintain water
mass properties (which comes about because of the
absence of any background horizontal diusion in the
scheme), rather than it producing a signi®cant heat ¯ux
directly due to eddies. The along isopycnal diusive ¯ux
also makes a small contribution to the depth averaged
meridional transport.
Fig. 16 The HadCM3 annual zonal mean poleward heat transport
(PW) as a function of latitude at dierent times during the coupled
simulation. The AMIP and Da Silva et al. (1994) values are also
plotted. A correction was applied to the AMIP and Da Silva et al.
(1994) ¯uxes to ensure a global mean balance when plotting the
implied ocean heat transports. See text for details
160 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
7.2 Comparison with ocean observations
To evaluate the modelled ocean heat transports the
model results are compared with those obtained from
direct ocean observations. Figure 18 shows recent direct
estimates of meridional ocean heat transports from ob-
servations across the marked latitudes, with the authors
indicated by their initials. The stars indicate that the
volume ¯ux across the section is non-zero and the
numbers indicate transports of temperature normalised
to PW. The estimates are calculated from the instanta-
neous pro®les of velocity and temperature measured on
a single hydrographic section. For comparison, Fig. 19a,
b shows the advective heat transport across the marked
latitudes for the mean of years 81±120 and 361±400 from
the coupled model. In addition to the observed sections
shown in Fig. 18, some extra model sections have been
included in Fig. 19a, b. The model estimates are calcu-
lated from the mean velocity and temperature pro®les
for the period indicated. Observational estimates make
the assumption that the velocity and temperature pro-
®les measured on a single hydrographic section are
representative of the long-term mean velocity and tem-
perature pro®les. The eddy heat ¯ux cannot be estimated
accurately from observations. Here we consider only the
large-scale advective heat transport.
A comparison of the heat transport for the two time
periods (years 81±120 and 361±400) shows there to be a
reduction in the Atlantic heat transport by O(0.1 PW) in
the later period, while transports in the other basins are
generally unchanged. Comparing Figs. 18 and 19a sug-
gests that in the North Atlantic and North Paci®c the
ocean heat transports produced by the coupled model
are in reasonable agreement with the observational es-
timates. In the South Atlantic the model shows less di-
vergence in the heat transport than the observational
estimates suggest, and the magnitude of the northward
heat transport near 30°S is larger than all the published
estimates with the exception of that by Fu (1981) of 0.8
PW (not shown). For a fuller discussion of the model
heat transports in the South Atlantic see Banks (To
appear 1999).
In the Indian and South Paci®c oceans the Indone-
sian through-¯ow (ITP) gives a non-zero mass transport
across the sections and the numbers indicated on
Figs. 18 and 19 across these oceans are temperature
transports (multiplied by q0Cp). (Note the transport
q0CpRsection vT dx dz is not invariant under change of
temperature scale when q0Rsection vdxdz6 0, the mass
transport across the section is non-zero.)
Fig. 18 Direct ocean observational estimates of the meridional heat
transport at marked latitudes. The authors for particular sections are
indicated by initials; MW, Macdonald and Wunsch (1996), BRC,
Bryden et al. (1991), HB, Hall and Bryden (1982), TBB,Tsimplis
et al. (1996), R, Rintoul (1991), TW, Toole and Warren (1993) and
SK, Saunders and King (1995)
Fig. 19 a As Fig. 18 but for the HadCM3 coupled model for years
81±120. bAs Fig. 18 but for the HadCM3 coupled model for years
361±400. In both aand bsome additional model sections are included
Fig. 17 HadCM3 component contributions of the annual zonal mean
poleward heat transport (years 361±400) as a function of latitude
(PW). Component contributions from the gyre, meridional overturn-
ing, along isopycnal diusion and eddy bolus velocities are shown
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 161
In the Indian Ocean the temperature transport is
southwards as seen in the observations but is larger in
magnitude, while in the South Paci®c the temperature
transport is in the opposite direction to most of the
observations. These dierences are associated with the
magnitude of the ITF in the model, 24 Sv, which is
larger than the observational range of 0±20 Sv (Wijels
et al. 1996), and dominates the temperature transports
in the basins. The combined Indo-Paci®c heat transports
are in reasonable agreement with the observations.
7.3 Comparison with the Atlantic
and Paci®c 24°N sections
To further validate the model a more detailed compar-
ison of the components contributing to the total heat
transports has been carried out for the sections at 24°N
in the Atlantic and the Paci®c. These 24°N sections are
considered to provide robust observational estimates of
the ocean heat transport because they are located in the
centre of the sub-tropical gyre and the eddy heat trans-
port is considered small compared to the total heat
transport. Table 1 gives the breakdown of the total heat
transport into its western boundary component (WBC),
Ekman (wind driven) and interior components. In the
model the WBC is de®ned by maximising the volume
transport from the western boundary.
Comparing the components from HadCM3 at 24°N
with the observational estimates of Hall and Bryden
(1982) for the Atlantic, and Bryden et al. (1991) for the
Paci®c, we can see similar dierences for both sections;
an overestimate of the heat transported northward by
the WBC and of the heat transported southward by the
interior and an underestimate of the heat transported
northward by the Ekman layer. The underestimate of
the Ekman layer heat transport is related to the reduced
Ekman transport across the section; 3.8 Sv (Atlantic)
and 7.5 Sv (Paci®c), compared with 5.0 Sv (Atlantic)
and 12.0 Sv (Paci®c) in the observational estimates. The
observational estimates are based on the climatological
windstresses from Hellerman and Rosenstein (1983)
which may bias the Ekman transport to larger values.
For example, across 24°N in the Paci®c, calculating the
Ekman transport from the Josey et al. (1996) climatol-
ogy rather than HR leads to a reduction of 3 Sv.
The remaining 1.5 Sv discrepancy between model and
observations is relatively small.
The overturning component of the heat transport in
HadCM3 is 1.07 PW for the Atlantic, accounting for
93% of the total heat transport (compared with the
observational estimate of 1.28 PW accounting for 105%
of the total). In the Paci®c, the overturning component is
0.14 PW accounting for 28% of the total (compared
with the observational estimate of 0.38 PW which ac-
counts for 50% of the total). In both cases, not only is
the magnitude of the overturning component smaller
than the observed estimate but it suggests a greater role
for the gyre component in the model than is indicated by
the observations.
Boning et al. (1996) compared the strength of the
North Atlantic Deep Water (NADW) cell at 25°Nwith
the poleward heat transport at the same latitude, from a
variety of models with dierent resolutions and forcings.
Although there is a considerable range amongst the
models in the value of the heat transport at 25°N, there
is a simple linear relationship between the strength of the
NADW cell and the poleward heat transport that ®ts all
the model results. The overturning meridional/depth
stream function for the North Atlantic in the coupled
model for the periods years 81±120 and 361±400, is
shown in Fig. 20a, b. The strength of the NADW cell is
stable, whereas a deeper cell clearly `spins up' over the
400 years of the integration. The strength of the NADW
cell at 24°N is between 16±18 Sv, with a corresponding
heat transport of 1.1 PW. These values also ®t the simple
relationship between the overturning and heat transport
derived in Boning et al. (1996). Hall and Bryden (1982)
estimate the magnitude of the NADW cell to be 19 Sv
(integrating their Table 5). The model transport of
NADW agrees well with observations at three key lo-
cations where there are robust observational estimates
(Wood et al. 1999).
The volume transports at 24°N in the Atlantic can
also be compared with observations for dierent tem-
perature classes. Table 2 shows the transports in tem-
perature classes for the model during the years 81±120
and 361±400 compared with the observational estimate
of Hall and Bryden (1982).
Compared with Hall and Bryden (1982), HadCM3
transports more water northwards in the warmest tem-
perature class and less in the two classes below this. By
the later period, about 2 Sv of warmest water has moved
to the cooler temperature classes.
For the earlier period, there is some indication of
southward ¯ow of NADW moving to the temperature
class 4 < h<7°C. This trend appears to continue, so
Table 1 Values for the heat transport components (PW) for the
40 year mean of years 81±120 from HadCM3 and the range (based
on the decadal means for years 81±120). The total heat transport is
split into components in two dierent ways: (a) Western Boundary
Current (WBC)/Ekman/Interior and (b) Overturning/Gyre
Observed 40 y mean Range
Atlantic 24°N
WBC 1.73 2.00 1.95±2.02
Ekman 0.42 0.28 0.27±0.29
Interior )0.93 )1.14 )1.15±)1.10
Total 1.22 1.14 1.12±1.16
Overturning 1.28 1.07 1.04±1.09
Gyre )0.06 0.07 0.08±0.07
Paci®c 24°N
WBC 1.73 2.32 2.23±2.37
Ekman 0.93 0.67 0.66±0.68
Interior )1.91 )2.49 )2.41±)2.55
Total 0.76 0.50 0.49±0.51
Overturning 0.38 0.14 0.13±0.16
Gyre 0.38 0.36 0.36±0.35
162 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
that during the later period all NADW transport is in
the warmer class. This is consistent with the reduction in
northward heat transport between the two periods and
with the shallower depth of the NADW cell in the later
period. Overall, the adjustment of the transport in
temperature classes is consistent with a reduction in the
eective temperature dierence of the overturning. The
detailed vertical structure of the temperature drift in the
coupled model is complex and has yet to be analysed in
detail. The global drifts shown in Fig. 3a show the near
surface cooling, with a larger warming below this that
extends down to 1000 m. The equivalent diagnostic for
the North Atlantic (not shown), shows a similar cooling
at the surface (see Fig. 6b) but with an even larger
subsurface warming, peaking at 2 °C around 1000 m.
The Atlantic warming extends to all deeper levels.
In summary, the overturning cell in the Atlantic is
slightly too weak in HadCM3 when compared to Hall
and Bryden (1982). This is consistent with the dierence
in total heat transport of O(0.1 PW) between model and
observed as predicted by the simple linear relationship
in Boning et al. (1996). The larger magnitude in the
observations may be partly attributable to the Ekman
transport calculated from HR which is larger than the
model Ekman transport. Macdonald (1995) found that
reducing the Ekman transport from 5.4 Sv (from HR
stresses) to 4.2 Sv (ECMWF stresses) reduced the total
heat transport by 0.13 PW. Finally, the percentage of
the heat transport explained by the overturning and
gyre components is probably related to the structure
of the western boundary current in the model. With a
horizontal resolution of 1.25°the boundary currents
cannot be simulated realistically.
8 Summary
The HadCM3 Hadley Centre coupled model has a much
improved SST and sea ice climatology when compared
to earlier Hadley Centre models without ¯ux adjust-
ments (Gregory and Mitchell 1997). One of the key
features in this advance has been the improved simula-
tion of both the surface heat ¯uxes and the ocean
poleward heat transports, which in the HadCM3 model
are in broad agreement with the observed estimates. This
was not the case in earlier versions of the model. In
HadCM3 the heat transports needed by the atmospheric
component to maintain its present day climate agree
reasonably well with the heat transports simulated by
the ocean model. As suggested by Weaver and Hughes
(1996) this leads to a stable model. Boville and Gent
(1998) and Bryan (1998) also conclude that this consis-
tency of ¯uxes is, in part, responsible for the stability of
their 300 year simulation with the non-¯ux adjusted
CSM1 model. They also attribute the stability to the
initialisation procedure used. In HadCM3 we have no
pre-coupled initialisation and this is not a pre-requisite
for a stable simulation with this model.
We conclude the following:
a. The atmospheric model, run in atmosphere-only
mode, is capable of a realistic simulation of the surface
heat ¯ux and surface stress ®elds. This in itself is a sig-
ni®cant achievement in atmospheric model performance.
One major limitation here is the quality of the ¯ux
climatologies used to evaluate the model.
b. When coupled, the ocean model approximately
maintains the poleward heat transports as implied by the
atmosphere only case. That is, the system is self-con-
sistent and there is no need for the SST to undergo large
drifts in order to rebalance the heat budget. Heat ¯ux
adjustments are therefore not required and the coupled
model can be initialised directly from an observed ocean
state without the need for separate equilibration of the
ocean model. Although there are salinity drifts in the
model these are not large enough to signi®cantly dete-
Fig. 20 a Annual mean North Atlantic overturning stream function
(Sv) for years 81±120 of the HadCM3 coupled model. bAs abut for
years 361±400
Table 2 Northward volume transport in temperature classes (Sv)
Hall and
h>17°C 8.9 13.8 11.4
12 < h<17°C 2.5 0.6 0.9
7<h<12°C 6.6 1.3 3.1
4<h<7°C)2.3 )6.8 )16.4
h<4°C)15.6 )8.8 2.0
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 163
riorate the ocean circulation and thereby the atmosphere
climate simulation.
c. The ocean component of the coupled model sim-
ulates poleward heat transports in broad agreement with
estimates from oceanographic sections.
The SST pattern in the model stabilises after only a few
decades. The sea ice simulation is encouraging and the
improvement in ocean heat transports in the model has
an important consequence in improving the position of
the ice edge. The strength of the thermohaline circula-
tion in the model is stable although there is a drift in its
vertical structure through the integration. There is also
signi®cant multi-decadal variability, with an amplitude
of 1±2 Sv in the maximum North Atlantic overturning
streamfunction (Wood et al. 1999).
A major emphasis of this study has been the evalu-
ation of the surface heat exchange and ocean heat
transport processes in the coupled model. It is important
in coupled models that this kind of analysis is carried
out to ensure that the mechanisms in the model, at least
those in determining the mean balanced state, are real-
istic. This is necessary if we are to have con®dence in
numerical experiments to understand climate and pre-
dict climate change using such models.
Appendix A
The ocean model used in HadCM3 has a number of sub-gridscale
parametrisations and details of these are given in this appendix.
A1 Vertical mixing of momentum
Momentum is mixed vertically using a K-Theory ¯ux-gradient
parametrisation. In the upper ocean mixed layer a simpli®ed ver-
sion of the Large et al. (1994) non-local parametrisation is used.
Based on results from large eddy simulations, K
is simply repre-
sented as a quadratic function of the depth z(see Large et al. 1994
for details). A boundary layer scaling depth his determined as the
depth where the gradient Richardson number Ri exceeds a critical
value of 0.3, and the diusion coecient K
is speci®ed by:
The coecients a
and a
are determined by the requirement
that near to the surface as z®0, K
(z)»ku*z(kis von Karman's
constant, u* is the ocean friction velocity) and that K
is contin-
uous at the depth h. To prevent large K
values in deep convective
layers a maximum value of 80 m is imposed on h. The use of this
formulation leads to a realistic simulation of the surface current
direction. This is particularly important in areas of sea ice since in
the model the direction of the ice drift is determined by the ocean
In the region z>h, below the surface layer, the Pacanowski and
Philander (1981) K-Theory parametrisation is used. The mixing
coecient K
is the sum of a term dependent on the local gradient
Richardson number and a depth independent background value of
1.0 ´10
. The Richardson number dependent part only
makes signi®cant contributions in the equatorial zone because, at
1.25°resolution, in all other regions current shears below the mixed
layer are very small. (Note. In the earlier HadCM2 version of this
model, Johns et al. 1997, the Pacanowski and Philander (1981)
scheme was also used to mix momentum in the surface boundary
layer. In the upper layer it led to small mixing coecients and
implied Ekman depths of less than the top model level at 10 m.
Surface current directions were therefore poorly simulated. This
has been improved in HadCM3 by the inclusion of the simpli®ed
Large et al. 1994 scheme described.)
A2 Vertical mixing of tracers
The mixing is carried out by the K-Theory diusion described
using the simpli®ed Large et al. (1994) scheme in the mixed layer
(see later) and below this the Pacanowski and Philander (1981)
scheme. The background tracer diusion coecient is now a
function of depth. This depth dependent diusivity is the same as
that used in HadCM2 and is shown in Table A for completeness.
This pro®le is a close ®t to the theoretical/observational estimate of
Kraus (1990), and the near surface values are in good agreement
with recent direct measurements in the upper 1000 m of the ocean
(Ledwell et al. 1993).
In the upper ocean boundary layer a hybrid tracer mixing
scheme is used. In order to better simulate the upper ocean mixed
layer structure, in addition to the diusive mixing, a Kraus-Turner
(1967) energetics parametrisation is also used (Gordon and Bot-
tomley 1985). The input of energy from the wind that is available for
mixing is parameterised by kq
u* (the wind mixing energy), q
is a
reference sea water density and ka dimensionless constant of order
unity. For a water column of uniform density, the turbulent energy
available for mixing at a depth zis also decayed exponentially via
exp()z/d). This is included to limit the eects of surface induced
mixing as the mixed layer deepens (see Kraus 1977 for discussion).
The constants kand dare determined to give the same dissipation as
that observed during the MILE study (Davis et al. 1985).
This turbulent energy is then used to mix the vertical temper-
ature and salinity pro®les so that the work done against gravity in
mixing down lighter water balances the energy input. The partial
mixing scheme of Thompson (1976) is used rather than carrying an
explicit mixed layer depth. A simple convective adjustment is im-
plemented when there is a negative surface buoyancy forcing. The
convection is partially penetrative in that 15% of the energy re-
leased by convection is used in further deepening (Kraus 1977). In
the layer which is well mixed by the KT scheme the diusive mixing
of tracers is trivial as the water column is already fully mixed.
A3 Eddy mixing
For eddies to be resolved explicitly would require a resolution
much ®ner that that currently possible in climate models. The eddy
Table A Depth and thickness of model levels and the tracer
diusivities. See text for explanation
Level Depth
Vertical diusivity
1 5.0 10.0 ±
2 15.0 10.0 1.03
3 25.0 10.0 1.06
4 35.1 10.2 1.08
5 47.9 15.3 1.11
6 67.0 23.0 1.16
7 95.8 34.5 1.22
8 138.9 51.8 1.32
9 203.7 77.8 1.46
10 301.0 116.8 1.68
11 447.1 175.3 2.01
12 666.3 263.2 2.50
13 995.6 395.3 3.23
14 1501.0 615.0 4.34
15 2116.0 615.0 6.06
16 2731.0 615.0 7.79
17 3347.0 615.0 9.51
18 3962.0 615.0 11.23
19 4577.0 616.0 12.95
20 5193.0 616.0 14.68
164 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
eects must therefore be parameterised. The Gent and McWilliams
(1990) (GM) scheme parameterises sub-gridscale eddy activity by
removing potential energy in an adiabatic manner by diusion of
isopycnal layer thickness. Earlier experiments with the 1.25°ocean
model showed that it is not appropriate to set a globally constant
thickness diusion coecient. This was because in order to main-
tain both the sharp horizontal density gradients associated with the
North Atlantic Current and Kuroshio extension and, at the same
time, simulate signi®cant eddy ¯uxes in the Antarctic Circumpolar
Current (ACC), dierent thickness diusion coecients are re-
quired (see Wright 1997 for details). In the model described here,
the formulation of Visbeck et al. (1997) is therefore used in which
the thickness diusion coecient is determined locally.
Visbeck et al. (1997) highlight three cases where dierent values
of jare appropriate. They suggest values of 2000 m
in a wind
driven channel (analogous to the ACC), 500 m
in a frontal
zone (analogous to the boundary currents) and 300 m
in a
convective chimney. These values are in broad agreement with
those required by the 1.25°ocean model to maintain the climato-
logical temperature and salinity structure of the ACC and western
boundary currents (Wright 1997).
Visbeck et al. (1997) suggest a time scale for the growth of
baroclinic instability of Tÿ2f Riÿ1where Tis the time scale for
the growth of eddies, fis the Coriolis parameter and Riÿ1is the
inverse depth averaged Richardson number. Following Treguier
et al. (1997), the vertical average is taken over the upper ocean
between 100 m and 2000 m. Following Visbeck et al. (1997) the
eddy transfer coecient is then de®ned as jaTÿ1l2where ais a
constant equal to 0.015 and lis a length scale chosen to be the
width of the baroclinic zone. Following Visbeck et al. (1997) this
length scale is determined from the horizontal distribution of the
inverse time scale T
. A baroclinic zone was de®ned as a region
where the inverse timescale is greater than 1.4 ´10
, this
number being determined by experiment. A minimum jof
350 m
and a maximum of 2000 m
are allowed.
The thickness diusivity values simulated by the model (Wright
1997) are in general agreement, both in magnitude and position,
with those suggested by both Visbeck et al. (1997) and Treguier
et al. (1997). With the spatially and time varying thickness diusion
coecient, the GM scheme enhances mixing in regions of baro-
clinic instability. This also allows the model to maintain the tight
density gradients in the northern gyres, whilst simulating a signi-
®cant eddy ¯ux across the ACC.
The Gries et al. (1998) numerical implementation of the Redi
(1982) isopycnal mixing scheme is also employed, as this leads to a
considerable reduction in grid point noise compared to the original
Cox (1987) implementation of the Redi (1982) scheme. The along
isopycnal diusion coecient is 1000 m
and does not vary
with depth. There is no explicit horizontal tracer diusion in
the model. The coecient of horizontal momentum viscosity is
formulated as AMA1A2cos (latitude) with A
3000 m
. The latitude dependence of A
is included to prevent
numerical instability at high latitudes as the grid lines converge,
while still resolving mid-latitude western boundary currents.
A4 The simple over¯ow scheme
One of the major systematic errors in an earlier version of the
Hadley Centre 1.25°´1.25°global ocean model was a too-zonal
simulation of the North Atlantic Current (NAC) (Wright and
Gordon 1997). In the coupled model this would clearly have im-
plications for the SST and therefore for the simulation of climate
over Western Europe.
A similar problem was observed in the comparison of a one
degree North Atlantic version of the Hadley Centre model with an
equivalent version of the MICOM isopycnal model (Roberts et al.
1996). In the Hadley Centre model the over¯ow water across the
Greenland±Iceland±Scotland ridge tended to mix with the water to
the south of the sills. There were two sources of this erroneous
mixing. The ®rst was caused by a numerical restriction in the
isopycnal mixing scheme. In the Cox (1987) implementation of the
scheme used in this earlier model version, a maximum value was
imposed on the isopycnal slope used in calculating the diusivity.
The imposition of this maximum slope can lead to a large diapycnal
component to the diusion in regions of large horizontal changes in
density (such as the over¯ow regions).
The second problem occurred due to the parametrisation of
convective overturning in the model. In the convective mixing
scheme, if two boxes were unstable then they were simply mixed
together and so on down the column to end up with a stable pro®le.
This led to any dense water entering near the top of the column
being mixed out with the underlying water instead of running down
the slope as a boundary current.
In order to test the hypothesis that the southward shift of the
NAC was associated with the wrong water mass properties of the
over¯ow water to the south of the sills, a number of sensitivity
studies were carried out. These concentrated on the imposition of
buoyancy anomalies in the sub-polar gyre, essentially to force the
water to sink. The main result from these sensitivity experiments
was to show that if the over¯ow water can be caused to sink to
the right depths then the NAC problem is alleviated (Wright and
Gordon 1997).
Two schemes have been developed in recent years to alleviate
some of the erroneous over¯ow mixing problems described. The
®rst is the Gerdes et al. (1991) (hereafter GKW) modi®cation to the
isopycnal mixing used in the model. In the GKW formulation the
restriction of a maximum slope is relaxed, but the diusivity along
the isopycnals is reduced as the slope is increased. This avoids the
generation of a diapycnal mixing component to the diusion which
occurs in the standard version of the scheme.
The scheme of Roether et al. (1994) aims to address the other
aspect of the erroneous mixing in the model. As noted above the
standard convective mixing leads to any dense water entering near
the top of the column being mixed out with the underlying water.
The Roether et al. (1994) scheme instead ®nds the level of neutral
buoyancy for the over¯ow water and mixes the water into this
model level. All the intervening water is then moved up in the water
column to replace the dense water which has been mixed in. This
way there is less mixing with intervening levels. This scheme was
implemented only in the over¯ow region (approximately 0±45°W
and 58±70°N). To avoid excessive mixing as the over¯ow water
`steps down' the model bottom slope, the topography to the south
of the over¯ow sills was modi®ed so that the drop on the southern
side of the sills takes place over a single gridpoint.
A 10 year simulation with the ocean only model showed that
the inclusion of these schemes led to a greater sinking of the
over¯ow water. However, the results were also very sensitive to the
surface forcing used in the model. A short 20 years coupled sim-
ulation in which the over¯ow parametrisation was removed was
run parallel to the HadCM3 simulation described in this paper.
After 10 years the inclusion of the over¯ow scheme leads to a
greater volume of dense bottom water in the sub-polar gyre and the
vertical density gradients in the upper 1000 m are better main-
tained. In the simulation without the over¯ow scheme the over¯ow
water mixes with the upper layers and diuses the density gradient.
After 20 years it is dicult to assess the impact of the scheme on
watermasses in the sub-polar gyre because of the high level of
natural variability in the coupled model; (e.g. after 20 years the
maximum Atlantic overturning streamfunction changes by less
than 0.5 Sv between the simulations with and without the over¯ow
scheme and this is well within the decadal variability of the model).
A longer century time scale simulation and signi®cance tests would
be necessary to de®nitively assess the impact of this and the other
parametrisations used in the coupled model.
A5 The Mediterranean out¯ow scheme
The Mediterranean out¯ow parametrisation attempts to simulate
the exchange of water between the Mediterranean and the Atlantic.
At each level between 1 and 13 in the model (approximately the
upper 1000 m), an average is formed of tracer values for two points
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 165
in the Mediterranean and two in the Atlantic adjacent to the Straits
of Gibraltar (denoted by Tbelow). For the temperature (T
) at each
of the four points a rate of change is calculated as:
The constant lre¯ects how much of the gridbox should be mixed in
a time step. Hence, near the surface, Mediterranean points become
fresher due to the fresher Atlantic water, and at depth Atlantic
points become more saline. The constant lis chosen to simulate a
transport of 1 Sv through the Straits of Gibraltar.
Appendix B: data transfer between grids
Particular problems in the techniques for exchanging information
between the ocean and the atmosphere arise from the models being
on dierent horizontal grids. The information passed from the
ocean to the atmosphere describes the state of the surface (tem-
perature and sea ice parameters). Because the sea ice ®elds can have
sharp discontinuities, bilinear interpolation from the higher-reso-
lution ocean grid is not considered satisfactory. Instead, we use an
area-weighted average of the six ocean gridboxes, which cover the
area of each atmosphere gridbox, making full use of all the infor-
mation available.
The information passed from the atmosphere to the ocean de-
scribes the surface ¯uxes of momentum, heat and fresh water. It is
essential that the coupling should be conservative i.e. the area-av-
erage of each quantity on the two grids must be equal. We could set
the six ocean points in an atmosphere box all equal to the single
atmosphere value for the box, but we prefer to have ocean ®elds
which vary continuously in space. To achieve both these aims, we
®rst use bilinear interpolation from the atmosphere grid to obtain a
®eld which varies on the higher-resolution ocean grid. Then we cal-
culate the area-average of the six ocean boxes in each atmosphere
box, and compare it with the atmosphere value. Since interpolation is
not conservative, the values will not generally be equal. We adjust the
six ocean values by dividing by the ratio of the average to the at-
mosphere value, or subtracting the dierence, to regain equality in
each atmosphere box. Division is used for positive-de®nite quanti-
ties, such as downward penetrative shortwave radiation (subtraction
might give unphysical negative values), and addition for others, such
as net surface water ¯ux (where there might be six nearly balancing
values of opposite signs, and division by their small sum could pro-
duce unreasonably large values). This adjustment procedure depends
on the coincidence of the ocean and atmosphere land-sea masks, so
that both grids have the same ocean area in each atmosphere box.
Acknowledgements We acknowledge the valuable input from our
many colleagues at the Hadley Centre in the development of
the coupled model. We would also like to acknowledge the input
from Ron Stouer, one of the referees, whose thorough review
has helped to signi®cantly improve the original manuscript. The
research was funded by the Department of the Environment,
Transport and Regions Climate Prediction Programme and the
Public Meteorological Service programme.
Aagaard K, Carmack EC (1989) The role of sea ice and other fresh
water in the Arctic circulation. J Geophys Res 94: 14485±14498
Banks HT (2000) Ocean heat transports in the South Atlantic in a
coupled climate model J Geophys Res (to appear)
Barthelet P, Terray L, Valcke S (1998) Transient CO
using the ARPEGE/OPAICE non ¯ux corrected coupled
model. Geophys Res Lett 25: 2277±2280
Bigg GR, Wadley MR, Stevens DP, Johnson JA (1996) Prediction
of iceberg trajectories in the North Atlantic and Arctic Oceans.
Geophys Res Lett 23: 3587±3590
Boning CW, O.Bryan F, Holland WR, Doscher R (1996) Deep
water formation and meridional overturning in a high resolu-
tion model of the north Atlantic. J Phys Oceanogr 26: 1142±
Boville BA, Gent PR (1998) The NCAR Climate Systems Model,
Version One. J Clim 11: 1115±1130
Bryan FO (1998) Climate drift in a multi-century integration of the
NCAR Climate System Model. J Clim 11: 1455±1471
Bryan K (1969) Climate and the ocean circulation III The ocean
model. Mon Weather Rev 97: 806±827
Bryden HL, Roemmich DH, Church JA (1991) Ocean heat trans-
port across 24°N in the Paci®c. Deep Sea Res 38: 297±324
Cattle H, Crossley J (1995) Modelling Arctic climate change. Philos
Trans R Soc London A352: 201±213
Cox P, Betts R, Bunton C, Essery R, Rowntree PR, Smith J (1998)
The impact of new land surface physics on the GCM simulation
of climate and climate sensitivity. Clim Dyn 15: 183±203
Cox MD (1984) A primitive equation, 3 dimensional model of the
ocean. GDFL Ocean Group Technical Rep 1, Princeton NJ,
USA, 143 pp
Cox MD (1987) Isopycnal diusion in a z-co-ordinate model.
Ocean Modell 74: 1±5 (unpublished manuscript)
Cusack S, Slingo A, Edwards JM, Wild M (1998) The radiative
impact of a simple aerosol climatology on the Hadley Centre
climate model. QJR Meteorol Soc 124: 2517±2526
Da Silva AM, Young C, Levitus S (1994) Atlas of surface marine
data, NOAA Atlas NESDIS 6 (Available from US Department
of Commerce)
Davis RE, de Szoeke R, Niiler P (1981) Variability in the upper
ocean during MILE. Part II: Modelling the mixed layer re-
sponse. Deep Sea Res 28 A: 1453±1475
Dickson RR, Brown J (1994) The production of North Atlantic
deep water: sources, rates and pathways. J Geophys Res 12: 319±
Edwards JM, Slingo A (1996) Studies with a ¯exible new radiation
code. I: choosing a con®guration for a large-scale model. QJR
Meteorol Soc 122: 689±719
Emery WJ, Fowler CW, Maslanik JA (1997) Satellite-derived maps
of Arctic and Antarctic sea ice motion: 1988 to 1994. Geophys
Res Lett 24: 897±900
ETOP05 (1988) Global 5¢´5¢depth and elevation. Technical re-
port, National Geophysical Data Centre, NOAA, US Depart-
ment of Commerce, Boulder, USA
Fu L-L (1981) The general circulation and meridional heat trans-
port of the subtropical South Atlantic determined by inverse
methods. J Phys Oceanogr 11: 1171±1193
Gates WL (1992) AMIP: the Atmospheric Model Intercomparison
Project. Bull Am Meteorol Soc 124: 1613±1646
Gent P, McWilliams JC (1990) Isopycnal mixing in ocean circula-
tion models. J Phys Oceanogr 20: 150±155
Gerdes R, Koberle C, Willebrand J (1991) The in¯uence of
numerical advection schemes on the results of ocean general
circulation models. Clim Dyn 5: 211±226
Gibson JK, Kallberg P, Uppala S, Hernandez A, Nomura A,
Serrano E (1997) ERA Description. ECMWF Re-Analysis
Project Series 1
Gill AE, Niiler PP (1973) The theory of the seasonal variability in
the ocean. Deep Sea Res 20: 141±178
Gregory D, Morris D (1996) The sensitivity of climate simulations to
the speci®cation of mixed phase clouds. Clim Dyn 12: 641±651
Gordon C, Bottomley M (1985) The parametrisation of the upper
ocean mixed layer in coupled ocean/atmosphere models. In: JCJ
Nihoul (ed) Proc 16th Liege Colloquium on Ocean Hydrody-
namics, Elsevier, Amsterdam
Gregory D, Kershaw R, Inness PM (1997) Parametrisation of
momentum transport by convection II: tests in single column
and general circulation models. QJR Meteorol Soc 123: 1153±
166 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
Gregory D, Shutts GJ, Mitchell JR (1998) A new gravity wave drag
scheme incorporating anisotropic orography and low level wave
breaking: impact upon the climate of the UK Meteorological
Oce Uni®ed Model. QJR Meteorol Soc 124: 463±493
Gregory JM, Mitchell JFB (1997) The climate response to CO
the Hadley Centre coupled AOGCM with and without ¯ux
adjustment. Geophys Res Lett 15: 1943±1946
Gries SM, Gnanadesikan A, Pacanowski RC, Larichev VD,
Dukowicz JK, Smith RD (1998) Isoneutral diusion in a
z-coordinate ocean model. J Phys Oceanogr 28: 805±830
Hall MM, Bryden HL (1982) Direct estimates and mechanisms of
ocean heat transport. Deep Sea Res 29: 339±359
Harrison DE, Kessler WS, Giese BS (1989) Ocean circulation
model hindcasts of the 1981±83 El Nino: thermal variability
along ship-of-opportunity tracks. J Phys Oceanogr 19: 397±418
Harvey JG (1976) Atmosphere and ocean: our ¯uid environment
(Harsham, Artemis press, pp 143)
Hellerman S, Rosenstein M (1983) Normal monthly wind stress
over the world ocean with error estimates. J Phys Oceanogr 13:
Hibler WD (1979) A dynamic-thermodynamic sea ice model. J Phys
Oceanogr 9: 815±846
Houghton JT, Meira Filho LG, Callander BA, Harris N, Katten-
berg A, Maskell K (1996) Climate change 1995. The science of
climate change. Cambridge University Press, Cambridge, UK
Hulbert HE, Walleraft AJ, Schnitz WJ Jr, Hogan PJ, Metzyer EJ
(1996) Dynamics of the Kuroshio/Oyashio current system using
eddy-resolving models of the North Paci®c Ocean. J Geophys
Res 101: 941±976
Johns TC, Carnell RE, Crossley JF, Gregory JM, Mitchell JFB,
Senior CA, Tett SFB, Wood RA (1997) The second Hadley
Centre coupled ocean-atmosphere GCM: Model description,
spinup and validation. Clim Dyn 13: 103±134
Josey SA, Kent EC, Oakley D, Taylor PK (1996) A new global air-
sea heat and momentum climatology. Int WOCE Newsl 24: 3±5
Kraus EB (1990) Diapycnal mixing. In: ME Schlesinger (ed)
Climate-ocean interactions. Kluwer, Amsterdam, Netherlands,
pp 269±293
Kraus EB (ed) (1977) Modelling and prediction of the upper layers
of the ocean. Pergamon, Oxford
Kraus EB, Turner JS (1967) A one dimensional model of the sea-
sonal thermocline II. The general theory and its consequences.
Tellus 19: 98±105
Large WG, McWilliams JC, Doney SC (1994) Oceanic vertical
mixing: A review and a model with a non-local boundary layer
parametrization. Rev Geophys 32: 363±403
Ledwell JR, Watson AJ, Law CS (1993) Evidence for slow mixing
across the pycnocline from an open-ocean tracer release
experiment. Nature 364: 701±703
Ledley TS (1985) Sea ice: multiyear cycles and white ice. J Geophys
Res 90: 5676±5686
Levitus S, Boyer TP (1994) World ocean atlas 1994, vol 4: tem-
perature. NOAA/NESDIS E/OC21, US Department of Com-
merce, Washington, DC, p 117
Levitus S, Burgett R, Boyer TP (1995) World ocean atlas 1994, Vol
3: salinity. NOAA/NESDIS E/OC21, US Department of
Commerce, Washington, DC, p 99
Lowe JA, Gregory JM (1998) The salinity drift of HadCM3: de-
scription, analysis and a scheme to correct the global average.
Hadley Centre Techn Note 3, Hadley Centre, Meteorological
Oce, Bracknell, UK
Macdonald AM (1995) Oceanic ¯uxes of mass, heat and freshwa-
ter: a global estimate and perspective. PhD thesis. 326 pp,
Massachusetts Institute Of Technology Woods Hole Oceanogr
Inst Joint Programme, Woods Hole, Mass
Macdonald AM, Wunsch C (1996) Oceanic estimates of global
ocean heat transport. US WOCE Rep, pp 28±30
Manabe S, Stouer RJ, Spelman MJ, Bryan K (1991) Transient
responses of a coupled ocean-atmosphere model to gradual
changes of atmospheric CO
. Part I: annual mean response.
J Clim 4: 785±818
Martin GM, Johnson DW, Spice A (1994) The measurement and
parametrisation of eective radius of droplets in warm strato-
cumulus clouds. J Atmos Sci 51: 1823±1842
Milton SF, Wilson CA (1996) The impact of parameterised sub-
grid scale orographic forcing on systematic errors in a global
NWP model. Mon Weather Rev 124: 2023±2045
Moss SJ, Johnson DW (1994) Aircraft measurements to validate
and improve numerical model parametrisations of ice to water
ratios in clouds. Atmos Res 34: 1±25
Murphy JM (1995) Transient response of the Hadley Centre cou-
pled ocean-atmosphere model to increasing carbon dioxide.
Part I: control climate and ¯ux adjustment. J Clim 8: 36±56
NSIDC (1989) DMSP SSM/I brightness temperatures and sea
ice concentration grids for the polar regions. NSIDC Dis-
tributed Active Archive Centre, University of Colorado,
Boulder, USA
Pacanowski RC, Philander SG (1981) Parametrisation of vertical
mixing in numerical models of tropical oceans. J Phys Oceanogr
11: 1443±1451
Paulson CA, Simpson JJ (1977) Irradiance measurements in the
upper ocean. J Phys Oceanogr 7: 952±956
P®rman SJ, Bauch D, Gammelsrod T (1994) The northern Barents
Sea: water mass distribution and modi®cation. In: Johannessen
OM, Muench RD, Overland JE (eds) The Polar Oceans and
their role is shaping the global environment, the Nansen Cen-
tennial Volume, Geophys Monogr Ser vol 85, pp 77±94, AGU,
Washington DC
Pope VD, Gallani ML, Rountree PR, Stratton RA (1999) The
impact of new physical parameterisations in the Hadley Centre
Climate Model. Clim Dyn 16: 123±146
Rayner NA, Horton EB, Parker DE, Folland CK, Hackett PB
(1996) Version 2.2 of the Global sea Ice and Sea Surface
Temperature (GISST) data set, 1903±1994. Climate Research
Tech Note (CRTN) 74, Hadley Centre, Meteorological Oce,
Bracknell, UK
Redi MH (1982) Oceanic isopycnal mixing by coordinate rotation.
J Phys Oceanogr 12: 1154±1158
Rintoul SR (1991) South Atlantic inter-basin exchange. J Geophys
Res 96: 2675±2692
Roberts MJ, Wood RA (1997) Topography sensitivity studies with
a Bryan-Cox type ocean model. J Phys Oceanogr 27: 823±836
Roberts MJ, Marsh R, New AL, Wood RA (1996) An intercom-
parison of a Bryan-Cox type ocean model and an isopycnic
ocean model. Part I: the subpolar gyre and high latitude
processes. J Phys Oceanogr 26: 1495±1527
Roether W, Roussenov VM, Well R (1994) A tracer study of the
thermohaline circulation of the eastern Mediterranean. In:
Malanotte-Rizzoli P, Robinson AR (eds) Ocean processes in
climate dynamics: global and Mediterranean example. Kluwer
Academic Press, Dordrecht, pp 371±394
Russell GL, Miller JR, Rind D (1995) A coupled atmosphere-ocean
model for transient climate change studies. Atmosphere-Ocean
33: 683±730
Sausen R, Barthel K, Hasselmann K (1988) Coupled ocean-at-
mosphere models with ¯ux adjustment. Clim Dyn 2: 145±163
Saunders PM, King BA (1995) Oceanic ¯uxes on the WOCE A11
section. J Phys Oceanogr 25: 1942±1958
Semtner AJ (1976) A model for the thermodynamic growth of sea
ice in numerical investigations of climate. J Phys Oceanogr 6:
Smith RNB (1990) A scheme for predicting layer clouds and their
water content in a general circulation model. QJR Meteorol Soc
116: 435±460
Smith RNB (1993) Experience and developments with the layer
cloud and boundary layer mixing schemes in the UK Meteo-
rological Oce Uni®ed Model. In: Proc ECMWF/GCSS
workshop on parametrisation of the cloud-topped boundary
layer, 8±11 June, 1993. ECMWF, Reading, England
Steele M, Zhang J, Rothrock D, Stern H (1997) The force balance
of sea ice in a numerical model of the Arctic Ocean. J Geophys
Res 102: 21 061±21 079
Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports 167
Thompson SR, Stevens DP, Doos K (1997) The importance of
inter-ocean exchange south of Africa in a numerical model.
J Geophys Res 102 (C2): 3303±3315
Thompson RORY (1976) Climatological numerical models of the
surface mixed layer of the ocean. J Phys Oceanogr 6: 496±503
Toole JM, Warren BA (1993) A hydrographic section across the
subtropical south Indian Ocean. Deep Sea Res 40: 1973±2019
Tsimplis M, Bacon S, Bryden H (1996) Heat ¯uxes of the south
Paci®c estimated through inverse models. WOCE Int Newsl 24:
Treguir AM, Held IM, Larichiv VD (1997) On the parametrisation
of quasi-geostrophic eddies in primitive equation ocean models.
J Phys Oceanogr 27: 567±580
Visbeck M, Marshall J, Haine T, Spall M (1997) On the speci®-
cation of eddy transfer coecients in coarse resolution ocean
circulation models. J Phys Oceanogr 27: 381±402
Weaver JW, Hughes TMC (1996) On the incompatibility of ocean
and atmosphere models and the need for ¯ux adjustments. Clim
Dyn 12: 141±170
White G (ed) (1996) WCRP workshop on air-sea ¯ux ®elds for
forcing ocean models and validating GCMs. WMO/TD-762
Wijels SE, Bray N, Hautala S, Meyers G, Morawitz W (1996) The
WOCE Indonesian through¯ow repeat hydrography sections:
I10 and IR6. Int WOCE Newsl 24: 25±28
Wood RA, Keen AB, Mitchell JFB, Gregory JM (1999) Localised
collapse of the thermohaline circulation in a climate model with
increasing atmospheric CO
. To appear in Nature
Wright DK (1997) A new eddy mixing parametrisation and ocean
general circulation model. Int WOCE Newsl 26: 27±29
Wright DK, Gordon C (1997) The positioning of the North
Atlantic Current. Ocean Applications Technical Note 14,
Meteorological Oce, Bracknell, UK
168 Gordon et al.: The simulation of SST, sea ice extents and ocean heat transports
... Here, we contrast unforced and naturally forced simulations under LGM and PI conditions in an ensemble using the Hadley Centre Coupled Model Version 3.4 (HadCM3 (Gordon et al., 2000;Pope et al., 2000;Reichler & Kim, 2008;Stott et al., 2000);). We examine the mean local response of the surface climate to volcanism in the two climate states (Section 3.1). ...
... We performed them using HadCM3, a three-dimensional, coupled atmosphere-ocean general circulation model (AOGCM) that is widely used for paleoclimate study (Armstrong et al., 2022;Bühler et al., 2021;Collins et al., 2001;Flato et al., 2014;Reichler & Kim, 2008;Tindall et al., 2009;Valdes et al., 2017). Computational efficiency allows for long-term integrations, with a simulated climate comparable to other AOGCMs and observations (Gordon et al., 2000;Jackson & Vellinga, 2013). ...
Full-text available
Plain Language Summary Climate variability describes the spatial and temporal variations of Earth's climate. Understanding these variations is important for estimating the occurrence of extreme climate events such as droughts. Yet, it is unclear whether climate variability depends on the mean surface temperature of the Earth or not. Here, we investigate the effects of natural forcing from volcanic eruptions and solar activity changes on climate variability. We compare simulations of a past (cold) and present (warm) climate with and without volcanism and solar changes. We find that overall, the climate system responds similarly to natural forcing in the cold and warm state. Small local differences mainly occur where ice can form. To evaluate the simulated variability, we use data from paleoclimate archives, including trees, ice‐cores, and marine sediments. Climate variability from forced simulations agrees better with the temperature variability obtained from data. Natural forcing is therefore critical for reliable simulation of variability in past and future climates.
... Mexico, Aceves-Rangel et al., 2018, investigated the potential of 20 species of Pines, to establish the microclimatic profile of these species, the potential areas of establishment for these species were determined, for their economic and social potential, since they are specific sources of timber, firewood, resin, and seeds, widely used by local communities for their microeconomic development. Analysis of forest species under future scenarios, the case of Beilschmiedia miners or northern acorn, which was modeled under the climate change scenario proposed by the IPCC (Intergovernmental Panel on Climate Change), using the descriptor variables according to the HADCM3 climate change model (Gordon et al., 2000), as a result, they determined that unfortunately, there is a reduction of surface area for this species in 2050, with a reduction of almost 600,000 hectares (Pliscoff and Fuentes-Castillo, 2011). Regarding biodiversity studies of forest species, there are studies of the presence and probability of distribution of species not yet described, or species with a strong potential for endemism (Avila et al., 2014). ...
Full-text available
Species distribution models (SDM), in many disciplines, have become an important tool in the environmental management of the territory, this includes several lines of knowledge, such as health, epidemiology, biodiversity, conservation, and forestry studies. This article is oriented to describe the importance, procedures, analysis, applications, and controversies that have arisen around this useful tool. This article focuses on highlighting research applications in forest management and conservation, as well as its applications and possible challenges for the future of forestry systems.
... The same experiments and emulator approach are used here to quantitatively address different sensitivities of temperature, precipitation, southerly wind (V-wind) changes to orbital parameters, CO 2 , and ice volume. HadCM3 is an atmosphere-ocean fully coupled general circulation model (Gordon et al., 2000). The assessment of multi-model simulation on East Asian climate shows that HadCM3 can well simulate the annual and seasonal surface air temperature and precipitation climatology in East Asia (Jiang et al., 2005). ...
Quaternary monsoon changes in East Asia have been extensively investigated by proxy records from continental and marine archives. However, these proxy indicators often show controversial characteristics in terms of trends and rhythms, leading to perplexedunderstanding of orbital-scale monsoon dynamics. Here we review the orbital-scale monsoon variability and dynamics in East Asia by comparing multiple proxies from loess, lake, speleothem, and marine records with the HadCM3 modeling result. Evolutionary power spectra of loess grain size and sea surface temperature exhibit a remarkable shift from 41- to 100-kyr cycles across the mid-Pleistocene transition (MPT), whereas other proxy records (e.g. δ¹³C of loess carbonate, pollen concentration in lake sediments, and magnetic mineral compositions in marine sediments) display distinct and persistent precession cycles through the Quaternary, along with strong 100-kyr cycles after the MPT. Simulations with the HadCM3 climate model reveal that the effects of orbital parameters, ice volume, and CO2 concentration on the temperature, precipitation, and southerly winds are seasonally and spatially different in East Asia. In the summer season, orbitally induced insolation plays a dominant role in driving changes in these three climate variables except for summer precipitation in south China (20-30°N), whilst annual changes in precipitation and temperature are jointly affected by insolation, ice volume, and CO2. Proxy-model comparison suggests that several land-based proxies are sensitive to changes in summer precipitation, annual precipitation, and annual temperature, though their responses to astronomical, ice, and CO2 forcing being quite different between north and south China. Our proxy-model comparison reveals that diverse expression of Quaternary climate periodicities was provoked by different sensitivities of marine and terrestrial proxies to seasonal and/or annual changes in precipitation and temperature, and by different responses of temperature and precipitation to insolation and ice/CO2 forcing. We suggest that understanding of Quaternary climate change can be deepened by further comparison of quantitatively reconstructed paleotemperature and precipitation data with high-resolution regional climate modeling results.
... x Gordon et al. (2000) Herold et al. ...
The early Eocene greenhouse climate maintained by high atmospheric CO2 concentrations serves as a testbed for future climate changes dominated by increasing CO2 forcing. In particular, the early Eocene Arctic region is important in the context of future CO2 driven climate warming in the northern polar region and associated shrinking Arctic sea ice. Here, we present early Eocene Arctic sea ice simulations carried out by six coupled climate models within the framework of the Deep-Time Model Intercomparison Project (DeepMIP). We find differences in sea ice responses to CO2 changes across the ensemble and compare the results with available proxy-based sea ice reconstructions from the Arctic Ocean. Most of the models simulate seasonal sea ice presence at high CO2 levels (≥ 840 ppmv = 3× pre-industrial (PI) level of 280 ppmv). However, the threshold when sea ice permanently disappears from the ocean varies considerably between the models (from <840 ppmv to >1680 ppmv). Based on a one-dimensional energy balance model analysis we find that the greenhouse effect likely caused by increased atmospheric water vapor concentration plays an important role in the inter-model spread in Arctic winter surface temperature changes in response to a CO2 rise from 1× to 3× the PI level. Furthermore, differences in simulated surface salinity in the Arctic Ocean play an important role in the control of local sea ice formation. These differences result from different implementations of river run-off between the models, but also from differences in the exchange of waters between a brackish Arctic and a more saline North Atlantic Ocean that are controlled by the width of the gateway between both basins. As there is no geological evidence for Arctic sea ice in the early Eocene, its presence in most of the simulations with 3× PI CO2 level indicates either a higher CO2 level and/or an overly weak polar sensitivity in these models.
... The PlioMIP2 experimental design did not include orbital sensitivity experiments. We therefore assess orbital uncertainty by including a number of sensitivity experiments run with a single model, HadCM3 (Gordon et al., 2000). Table 2 shows the top-of-the-atmosphere (TOA) insolation for specified times within the period 2.9 to 3.3 Ma. ...
Full-text available
Reconciling palaeodata with model simulations of the Pliocene climate is essential for understanding a world with atmospheric CO2 concentration near 400 ppmv (parts per million by volume). Both models and data indicate an amplified warming of the high latitudes during the Pliocene; however, terrestrial data suggest that Pliocene northern high-latitude temperatures were much higher than can be simulated by models. We focus on the mid-Pliocene warm period (mPWP) and show that understanding the northern high-latitude terrestrial temperatures is particularly difficult for the coldest months. Here the temperatures obtained from models and different proxies can vary by more than 20 ∘C. We refer to this mismatch as the “warm winter paradox”. Analysis suggests the warm winter paradox could be due to a number of factors including model structural uncertainty, proxy data not being strongly constrained by winter temperatures, uncertainties in data reconstruction methods, and the fact that the Pliocene northern high-latitude climate does not have a modern analogue. Refinements to model boundary conditions or proxy dating are unlikely to contribute significantly to the resolution of the warm winter paradox. For the Pliocene high-latitude terrestrial summer temperatures, models and different proxies are in good agreement. Those factors which cause uncertainty in winter temperatures are shown to be much less important for the summer. Until some of the uncertainties in winter high-latitude Pliocene temperatures can be reduced, we suggest a data–model comparison should focus on the summer. This is expected to give more meaningful and accurate results than a data–model comparison which focuses on the annual mean.
... While the majority of the current generation of climate models project a decrease in annual mean precipitation with global warming (see Chapter 22), the rate of the Amazon precipitation decrease in relation to global warming varies widely between the models. A family of climate models notable for their projection of severe Amazon drying, HadCM3 (Gordon et al. 2000), project annual precipitation in the eastern Amazon ...
Full-text available
This Report provides a comprehensive, objective, open, transparent, systematic, and rigorous scientific assessment of the state of the Amazon’s ecosystems, current trends, and their implications for the long-term well-being of the region, as well as opportunities and policy relevant options for conservation and sustainable development.
The Holocene, which started approximately 11.5 ka, is the latest interglacial period with several rapid climate changes with timescales, from decades to centuries, superimposed on the millennium-scale mean climate trend. Climate models provide useful tools to investigate the underlying dynamic mechanisms for the climate change during this well-studied time period. Thanks to the improvements in the climate model and computational power, transient simulation of the Holocene offers an opportunity to investigate the climate evolution in response to time-varying external forcings and feedbacks. Here, we present the design of a new set of transient experiments for the whole Holocene from 11.5 ka to the preindustrial period (1850; HT-11.5 ka) to investigate both the combined and separated effects of the main external forcing of orbital insolation, atmospheric greenhouse gas (GHG) concentrations, and ice sheets on the climate evolution over the Holocene. The HT-11.5 ka simulations are performed with a relatively high-resolution version of the comprehensive Earth system model CESM1.2.1 without acceleration, both fully and singly forced by time-varying boundary conditions of orbital configurations, atmospheric GHGs, and ice sheets. Preliminary simulation results show a slight decrease in the global annual mean surface air temperature from 11.5 to 7.5 ka due to both changes in orbital insolation and GHG concentrations, with an abrupt cooling at approximately 7.5 ka, which is followed by a continuous warming until the preindustrial period, mainly due to increased GHG concentrations. Both at global and zonal mean scales, the simulated annual and seasonal temperature changes at 6 ka lie within the range of the 14 PMIP4 model results and are overall stronger than their arithmetic mean results for the Middle Holocene simulations. Further analyses on the HT-11.5 ka transient simulation results will be covered by follow-up studies.
Full-text available
Decadal prediction experiments with six coupled climate models initialized yearly from 1960 to 2009 are used to evaluate the prediction skill for sea surface temperature (SST) over the North Atlantic. The stepwise pattern projection method (SPPM) and the multi-model super-ensemble (MMSE) method are adopted to improve the SST prediction skill averaged over forecast lead years 2–5 in an independent training period over the North Atlantic. After removing the linear trend, the prediction skill is obviously reduced, indicating that the prediction skill largely derives from the correct prediction of the long-term trend. The MMSE and SPPM calibrations, except for the SPPM-corrected CanCM4, are proven to be capable of reproducing the observed SST climatology. The prediction skill of single models is enhanced in terms of the anomaly correlation coefficient (ACC) and mean-square-error skill score (MSESS) after SPPM correction. Then, the prediction skill of SPPM–MIROC5, which is the optimal SPPM-corrected model, is further compared with that of the MMSE forecast. The ACC of SPPM–MIROC5 is slightly higher than that of the MMSE forecast. Unlike the homogeneous and slight reduction in MSE over the North Atlantic Ocean for MMSE, the MSE of SPPM–MIROC5 increases remarkably in middle to high latitudes, which is caused by large MSEs in certain years.
Full-text available
This Report provides a comprehensive, objective, open, transparent, systematic, and rigorous scientific assessment of the state of the Amazon’s ecosystems, current trends, and their implications for the long-term well-being of the region, as well as opportunities and policy relevant options for conservation and sustainable development.
Full-text available
A subgrid-scale form for mesoscale eddy mixing on isopycnal surfaces is proposed for use in non-eddy-resolving ocean circulation models. The mixing is applied in isopycnal coordinates to isopycnal layer thickness, or inverse density gradient, as well as to passive scalars, temperature and salinity. The transformation of these mixing forms to physical coordinates is also presented.
A theory of the layer formation due to surface processes is presented, which is more general than that used in the preceding paper I. Convection due to heating at depth and cooling at the surface is included, as well as the mechanical stirring due to wind action. The theory is applicable to arbitrary forms of heating, including intermittent or continuous processes, and could be used to investigate diurnal as well as seasonal effects. A detailed application is made to the case treated approximately in I, for which a solution is now obtained in analytic form. The results obtained allow a quantitative, as well as qualitative, comparison with the ocean. It is found that reasonable layer depths are predicted using measured heating rates, and a value of the turbulent kinetic energy input to the water deduced from the mean surface stress. The effects of heating at depth can be comparable with wind stirring, even when the temperature of the upper layer is increasing. During the winter, convection due to surface cooling dominates the processes which deepen the layer. DOI: 10.1111/j.2153-3490.1967.tb01462.x
Current numerical models of oceanic circulation differentiate between the eddy diffusion and viscosity transport along the geopotential horizontal and vertical directions only. In order to model the effect of anisotropic turbulence as diffusive transport along and across density surfaces, the isopycnal mixing tensor has been transformed from a diagonal second-rank tensor in the isopycnal coordinate system to a tensor containing off-diagonal elements in the geopotential coordinate system.
The available information and existing theories of ocean mixing are reviewed. Evidence is produced to show that the eddy diffusion coefficient is inversely proportional to the buoyancy frequency N. The vertical flux of buoyancy is therefore a linear function of N. The observed mean distribution of N can hence be used to estimate the mean meridional transport of buoyancy and heat. The implications of this distribution for climate theory and modelling are discussed.