Introduction
(a) Ensemble mean and (b) ensemble standard
deviation
of CMIP5 AOGCMs for the projected change
Δζ in ocean dynamic sea level for 2081–2100 with respect to
1986–2005 under the mid-range scenario RCP4.5, expressed as percentages of
ensemble-mean global-mean sea-level rise hθ due to thermal
expansion for the same scenario.
![]()
Atmosphere–ocean general circulation models (AOGCMs) are widely used for
projections of future sea-level change
e.g.. On the basis of AOGCM
results contributed to the Coupled Model Intercomparison Project Phase 5
(CMIP5), global-mean sea-level rise (GMSLR) of 0.32–0.63 m (5–95 %,
median 0.47 m) is projected by 2081–2100 under the mid-range RCP4.5
scenario considered in the Fifth Assessment Report of the Intergovernmental
Panel on Climate Change (IPCC) . Of this,
0.14–0.23 m (median 0.19 m) is the thermosteric contribution, due to
expansion of seawater as the ocean takes up heat, representing 30–50 %
of the total. Other contributions to GMSLR are due mostly to loss of land
ice. Glaciers worldwide give 15–40 % of the total. The median projected
contributions from the Greenland and Antarctic ice sheets are smaller,
although the latter is the largest source of uncertainty.
The range of the thermosteric contribution (hereafter denoted hθ)
also represents a substantial uncertainty in projections of GMSLR. It arises
partly from differences among models in climate sensitivity, determined by
surface and atmospheric responses to radiative forcing, and partly from
differences in the ocean processes which transport heat from the surface and
redistribute it in the interior of the ocean
. The
three-dimensional distribution of additional heat within the ocean affects
hθ because of the dependence of thermal expansivity on temperature
and pressure, quantified by the “expansion efficiency of heat”
, the ratio of hθ to the global ocean increase
in heat content. From CMIP5 results, this ratio is 0.12 m YJ-1
(1 YJ ≡ 1024 J), with a 90 % confidence interval of
0.10–0.14 m YJ-1 , indicating that
there is an uncertainty in the hθ that results from a given
increase in ocean heat content. By contrast, redistribution of the salt
content of the ocean makes a negligible contribution to GMSLR or its
uncertainty.
Sea-level change is not expected to be globally uniform. Changes in ocean
circulation, temperature and salinity (and hence density) alter dynamic
sea level ζ(x,t), where x is location and t time. This
quantity is defined as
ζ(x,t)≡η(x,t)-η(x,t)‾,
where η is sea-surface height relative to a
surface on which the geopotential has a uniform and constant value,
and the overline indicates the mean over the ocean area, so
ζ‾=0 by construction.
Hence
Δη=Δζ+Δη‾,
in which the last term is GMSLR. That is, the local change in sea level
Δη has contributions from GMSLR and from change in dynamic
sea level Δζ. The spatial standard deviation of the CMIP5
model mean Δζ is about 30 % of the model-mean global-mean
thermosteric contribution hθ (Fig. a).
There is a substantial model spread in Δζ, although in some
regions, notably the Arctic, the model spread is smaller than in the previous
phases of CMIP considered by earlier IPCC reports
.
Nonetheless, the CMIP5 RCP4.5 local spread in the pattern, measured by the
ensemble standard deviation of ζ(x), is 30 % on average of
the model mean hθ, and for example it exceeds 100 % in the
North Atlantic (Fig. b).
There are three features of Δζ that the models have in common
(Fig. a)
: (i) a meridional contrast between a band of positive change to
the north of the Antarctic Circumpolar Current (ACC) and a band of negative
change to the south, (ii) a meridional dipole in the North Atlantic, also positive
to the north and negative to the south and (iii) positive Δζ in the
Arctic. Although these qualitative features are robustly predicted, the
affected regions have the largest model spread in Δζ.
The Southern Ocean feature results both from changes to the surface heat flux
and from an intensification and southward shift of the westerly wind stress,
which strengthens the Ekman drift and tends to tilt the isopycnals
.
Eddies tend to oppose the latter effect by removing available potential
energy, thus partly compensating for the effect of wind-stress change in
Δζ, and limiting the sensitivity of the circumpolar circulation to
wind-stress change
. Most AOGCMs used for multidecadal simulations do not
resolve ocean eddies at high latitudes, so their results will depend on their
parametrisations of eddy advection on isoneutral surfaces
e.g..
The North Atlantic feature in Δζ is caused by increased ocean
buoyancy at high latitudes under CO2 forcing .
The buoyancy increase is due to reduced heat loss and increased
precipitation. As well as tending to raise sea level, it leads to a reduction
of the Atlantic meridional overturning circulation (AMOC), by 0–50 % by
2100 in CMIP5 AOGCMs, depending on model and scenario .
The circulation change causes a redistribution of properties, giving a
negative Δζ in the subtropical North Atlantic gyre. The
enhanced sea-level rise in the Arctic has been attributed to increased
buoyancy from reduction of salinity ,
consistent with greater precipitation and river inflow.
investigated how much of the model spread in CMIP5
Δζ was caused by the AOGCMs' different projections of surface
momentum flux change in response to increasing CO2. They did so by
computing the field of surface wind-stress change simulated for doubled
CO2 by each CMIP5 AOGCM, and imposed these fields as perturbations in a
set of experiments (one for each CMIP5 model) with the FAMOUS AOGCM
, which is a low-resolution and consequently relatively
inexpensive version of HadCM3. carried out a
corresponding study for surface heat flux and freshwater flux changes. These
studies show that part of the model spread in Δζ arises from the
spread of surface flux changes predicted by AOGCMs ,
especially regarding the amplitude of the changes.
However, the FAMOUS experiments tend to be similar in their patterns of
change; they do not reproduce the diversity of patterns of Δζ in
the AOGCMs supplying the surface flux perturbations. The unexplained model
spread in patterns and amplitude of Δζ must arise from dependence
on the ocean model formulation and unperturbed state. These aspects are so
far largely unexplored and need further constraint, but comparisons of the
ocean response in AOGCMs are complicated by their different predictions of
changes to surface fluxes experienced by the ocean.
Consequently, the Flux-Anomaly-Forced Model Intercomparison Project (FAFMIP)
was proposed to isolate the ocean uncertainty, by comparing results from
AOGCM experiments in which model-independent surface flux perturbations are
imposed on the ocean. FAFMIP is a component of CMIP6, the phase of CMIP which
is now beginning . At the time of writing there were 10
modelling groups planning to run FAFMIP experiments as part of their
contributions to CMIP6, namely ACCESS (Australia), CCCma/CanESM (Canada),
CNRM/CERFACS (France), GFDL (USA), GISS (USA), IPSL (France), MIROC (Japan),
MPI-ESM (Germany), MRI (Japan) and UKESM (UK). FAFMIP is an element of the
science plan for the World Climate Research Programme (WCRP) Grand Challenge
on regional sea-level change and coastal impacts.
The AOGCMs participating in FAFMIP will include new three-dimensional ocean
diagnostics of the rates of change of temperature and salinity due to the
individual processes which transport heat and salt within the ocean (resolved
advection, dianeutral mixing, etc.). Such ocean process-based diagnostics
have previously been included in only a small number of models
e.g., and cannot be estimated accurately from
other archived data. The FAFMIP experiments and diagnostics will for the
first time permit us to attribute differences in the ocean among a wide range
of models in the unperturbed state and in CO2-forced climate change to
particular processes and aspects of model formulation.
The FAFMIP experiments will provide information on the sensitivity of the
AMOC to buoyancy forcing of the magnitude and pattern of that predicted for
CO2 forcing, and will support investigation of the correlation between
ocean heat uptake efficiency and the magnitude of the AMOC
. The
application of common perturbations to surface fluxes in FAFMIP will provide
information about the ocean's role in determining patterns of sea-surface
temperature change worldwide (of relevance to the Grand Challenge on clouds,
circulation and climate sensitivity). Similarly the results will be of
relevance to studies of subsurface ocean temperature change in the vicinity
of Greenland and Antarctic ice shelves
, where warming may
promote basal melting of ice shelves and consequent sea-level rise through
the effect on ice-sheet dynamics (of relevance to the Grand Challenge on
melting ice and global consequences, as well as sea level).
FAFMIP will thus help with understanding and accounting for the spread in
simulated ocean responses in general to changes in surface fluxes resulting
from CO2 forcing. In the next section we describe the design of FAFMIP,
and in the following section we present preliminary results from experiments
that have been carried out in a small number of existing AOGCMs to test the
design.
Design
The aim of the FAFMIP tier-1 experiments is to study the response of the
ocean to the changes in its surface fluxes caused by CO2-forced climate
change, in particular regarding sea level, ocean heat uptake and ocean
circulation. The design allows the effects of changes in surface fluxes of
momentum, heat and freshwater to be separated, and aims to simplify the
analysis of the diversity of ocean response by imposing the same changes in
surface fluxes in every AOGCM. This is by contrast with CO2-forced
climate-change experiments, such as 1pctCO2, in which the surface fluxes are
model-dependent.
AOGCMs and surface flux perturbations
The atmosphere and ocean are a tightly coupled system, especially through the
interaction of surface heat flux and sea surface temperature (SST). It typically requires millennia of
“spin-up” integration of an AOGCM with constant atmospheric composition
(the pre-industrial control experiment, denoted “piControl”) to reach an
approximately steady state in the deep ocean, owing to its large heat
capacity and weak thermal connection to the surface
. Even
then, a small “climate drift” may persist. Experiments have been done
successfully in which surface fluxes from one climate state of an AOGCM are
transplanted into a simulation of another climate state of the same AOGCM
. However, if one replaces AOGCM ocean
surface fluxes with real-world estimates or with fluxes diagnosed from
another model, a large climate drift will result, because they will
not be consistent with the AOGCM's own surface climate. Therefore the FAFMIP
experiments instead impose perturbations, added to the surface
fluxes that are computed within the AOGCM from the state of the system
, technically like the flux adjustment
that was formerly used in AOGCMs but with a different
purpose.
The principle of the FAFMIP experiments is that the ocean should respond as
it does during an AOGCM climate-change experiment, as nearly as possible,
including interactively simulated atmosphere–ocean feedbacks and unforced
variability. The FAFMIP design contrasts with that of studies using
(uncoupled) ocean GCMs, such as the CORE project
e.g., in which bulk formulae are used to
compute fluxes from prescribed observationally derived surface climate
variables, and the experiments of , with a
prescribed geographically uniform surface heat flux perturbation and feedback
parameter. Those approaches yield valuable and complementary information
about the response of the ocean to perturbations, but are less like the AOGCM
projections whose uncertainty we aim to investigate.
Deriving the surface flux perturbations
Annual-mean FAFMIP surface flux perturbations of
(a) momentum, (b) heat, (c) water;
(d) shows the model-mean change in the surface heat flux Q into
the seawater in the time mean of the final decade of the faf-heat experiment
relative to the control, not including the imposed heat flux perturbation
F. The ocean area average of (b) is 1.86 W m-2, of
(c) 0.072×10-6 kg m-2 s-1 and of
(d) 0.07 W m-2. The grey box in (b) is the North
Atlantic region to which we refer in Sect. .
![]()
Climate-change projection is concerned mostly with scenarios of radiative
forcing increasing on decadal timescales. The idealised scenario called
“1pctCO2” in CMIP6 (and CMIP5), beginning from a piControl state and with
atmospheric CO2 concentration increasing at 1 % year-1, is
commonly taken to be indicative of anthropogenic climate change expected
during this century. It is a useful benchmark because it has been studied
since the first AOGCM experiments in the early 1990s, while the more
policy-specific scenarios, involving emissions of many species and
complicated time profiles of forcing, have been revised several times. The
transient climate response (TCR) is likewise used for convenient comparison
of the magnitude of climate change, and is defined as
the difference from piControl of the time-mean global-mean surface air
temperature during years 61–80 of 1pctCO2, centred around the time
(70 years) at which CO2 reaches double its piControl concentration.
For consistency with this conventional choice, we obtain the surface flux
perturbations for FAFMIP from years 61–80 of CMIP5 1pctCO2 experiments. In
experiments with time-dependent forcing scenarios, the geographical pattern
of sea-level change is fairly constant in time, but has increasing amplitude
, as is often assumed for surface
air temperature and other surface quantities
. To investigate the causes of
the patterns we therefore do not need to include interannual variation in the
surface flux perturbations. In the FAFMIP experiments, the perturbations are
imposed from the start and held constant (apart from their seasonal cycle).
Tests with the FAMOUS AOGCM indicated that similar geographical patterns of
sea-level change result from time-dependent flux perturbations from 1pctCO2
experiments as from the time-independent FAFMIP flux perturbations.
The surface flux perturbations are derived from a set of 13 CMIP5 AOGCMs for
which all the required diagnostics are available, namely CNRM-CM5,
CSIRO-Mk3-6-0, CanESM2, GFDL-ESM2G, HadGEM2-ES, MIROC-ESM, MIROC5,
MPI-ESM-LR, MPI-ESM-MR, MPI-ESM-P, MRI-CGCM3, NorESM1-ME and NorESM1-M. More
AOGCMs could have been included for some types of perturbation, but it was
decided to use this restricted but consistent set of AOGCMs for all
perturbations, in order to permit comparison with the model-mean change in
sea level and other quantities from the same set of AOGCMs. The diversity in
the surface flux changes from the individual CMIP5 AOGCMs is illustrated by
(, their Fig. 2). For each
of four types of surface flux (zonal and meridional momentum, heat and
freshwater, Sect. ), a difference field for each model is
computed between the climatological monthly time means of years 61–80 of
1pctCO2, using the first member in cases of an ensemble, and of the
corresponding 20 years of piControl, then interpolated to a common 1∘
latitude–longitude grid. Finally, the mean of the models is calculated.
The resulting model-mean fields for use in the FAFMIP experiments are stored
in CF-netCDF files at http://www.fafmip.org. They are
monthly means, which can be regarded as applying at the middle of the month,
and it is recommended to interpolate linearly between them in time to obtain
updates at the atmosphere–ocean coupling interval. Horizontal interpolation
to the required ocean model grid may not exactly preserve the global integral
but the differences are not likely to be important.
Experiments
The FAFMIP experiments (Table ) branch from piControl and
have piControl boundary conditions (atmospheric composition, solar
irradiance, land surface, etc.). The best point to branch would be the same
point as the 1pctCO2 experiment, with which FAFMIP results may be compared.
The experiments are proposed as 70 years long, but because a large
perturbation is switched on instantaneously at the start, useful results
could be obtained from shorter integrations of computationally expensive
models. During the first several decades of a typical AOGCM 1pctCO2
integration, the global-mean surface air temperature and net heat flux into
the ocean rise roughly linearly in time e.g..
The flux perturbations in FAFMIP integrations are typical of year 70 of a
1pctCO2 experiment. Therefore 70 years of a FAFMIP integration will apply
roughly the same time-integral forcing to the ocean as 702≃100
years of a 1pctCO2 integration.
FAFMIP experiments.
Name
Ocean surface flux perturbation
Tier 1
faf-stress
Zonal and meridional momentum
faf-heat
Heat
faf-water
Freshwater
Tier 2
faf-all
All from faf-stress, faf-heat and faf-water
faf-passiveheat
Heat as in faf-heat, but added as a passive tracer
The process-based tendency diagnostics (Sect. )
should be included in the FAFMIP experiments and in the DECK abrupt4xCO2 and
1pctCO2 and the corresponding section of piControl. The faf-passiveheat
experiment is identical to piControl except for the inclusion of the added
heat tracer, so a separate integration may not be needed.
Three experiments are required for participation in FAFMIP (tier 1):
In faf-stress we impose a perturbation in surface zonal and
meridional momentum flux, i.e. wind stress (Fig. a), created
from the CMIP5 diagnostics of surface downward fluxes of eastward
(tauu) and northward (tauv) momentum. Its dominant feature
is the increase in westerly wind stress in the Southern Ocean. The stress
perturbation is added to the momentum balance of the ocean water surface. For
instance, the modified equation of motion of the top layer of a hydrostatic
ocean model is
∂uh∂t=-(u⋅∇)uh-1ρ∇hp-f×uh+1ρR+1mt(τw+τi+S),
where u is velocity, subscript h indicates the horizontal part, t
is time, p hydrostatic pressure, ρ density, f the product of
the Coriolis parameter and the vertical unit vector, R the
vertical and horizontal convergence of horizontal momentum (in N m-3)
due to subgridscale processes (including the shear stress which conveys the
surface momentum fluxes into the subsurface), τw
the wind stress, τi the stress exerted by sea ice,
S the faf-stress momentum flux perturbation (in Pa, like
τw,i), and mt is the mass per unit area
of the top layer of the model, to which the surface momentum fluxes are
applied. No perturbation should be made directly to any turbulent mixing
scheme that depends on the wind stress, nor to the sea-ice momentum balance,
although both of these could be indirectly influenced since
τw and τi may be affected
by changes in the surface uh.
In faf-heat we impose a perturbation on the heat flux into the
seawater surface (Fig. b), created from the CMIP5
diagnostic of surface downward heat flux in seawater (hfds), i.e.
the sum of net downward radiative fluxes, sensible and latent heat fluxes to
the atmosphere, and heat fluxes between sea ice and seawater. The heat flux
perturbation is strongly positive in the North Atlantic and in the Southern
Ocean. Imposing a heat flux perturbation in an AOGCM by adding it to the
ocean surface layer alters the SST and thus modifies the surface heat flux so
as to oppose the perturbation. Such a strong negative feedback does not occur
with the momentum flux, which is only fairly weakly affected by the seawater
surface velocity, nor with the freshwater flux, which does not depend on the
surface salinity. The method for implementing faf-heat is the one used by
, described below and compared with alternatives
(Sect. ); it is intended to avoid this negative feedback, but
permits feedbacks due to ocean circulation change. The method allows us to
partition ocean temperature change between the effects of local addition of
heat and changing heat transports, using three-dimensional ocean tracer
fields of “added heat” and “redistributed heat” (Sect. ).
In faf-water we impose a perturbation on the freshwater flux into
the seawater surface (Fig. c), created from the CMIP5
diagnostic of water flux into seawater (wfo), i.e. the sum of
precipitation, evaporation, river inflow and water fluxes between floating
ice (sea ice and icebergs) and seawater. Its pattern is dominated by that of
precipitation change, being positive near the Equator and at mid- to
high latitudes, and negative in the subtropics. In the Arctic there is also
increased water input from river inflow, and a pronounced band of reduced
water input from melting along the sea-ice margin, which retreats to higher
latitude in the 2×CO2 climate.
Two further experiments are recommended (tier 2):
In faf-all the surface flux perturbations of momentum, heat and
freshwater are simultaneously applied, using the same method for heat as in
the faf-heat experiment. By comparison with the tier-1 experiments, faf-all
will be used to quantify non-linearities in the combination of the effects of
the perturbations. If the combination is linear, the ocean response to
CO2 forcing may be interpreted as the sum of the effects.
In faf-passiveheat a surface flux equal to the surface heat flux
perturbation of the faf-heat experiment is applied instead to a passive
“added heat” tracer (Sect. ), initialised to zero. This tracer
does not affect the model evolution, so the experiment is equivalent to
piControl, with an extra diagnostic tracer. Comparison of faf-passiveheat
with faf-heat will allow the effect on the distribution of the added heat
from changes in ocean heat transport to be assessed, because these changes do
not occur in faf-passiveheat.
The three methods for treating the surface heat flux in faf-heat and
faf-all, described in Sect. . The methods differ regarding the SST
which is used to calculate the net surface heat flux Q from the atmosphere
and sea ice into the ocean water. In method A it is obtained from the
top-layer ocean temperature θ as usual in an AOGCM, in method B from
the redistributed heat tracer TR, and in method C the SST and sea ice are
prescribed from the climatology of the AOGCM control experiment.
![]()
Apart from the partial suppression of changes in surface heat flux in
faf-heat (discussed above and in the next section), the surface fluxes of
momentum, heat and freshwater are computed as usual in the AOGCM. In general
they will all differ from the piControl state because of climate change
caused by applying the perturbation fluxes to the ocean. In models where the
sensible heat content of ocean surface water fluxes (precipitation,
evaporation and runoff) is considered, faf-water will in effect also impose a
small heat flux perturbation. Further technical notes on the implementation
of each of the experiments can be found at
http://www.fafmip.org.
Treatment of the surface heat flux
(a, c, e) The change ΔQ (W m-2) in the surface heat flux
into seawater in the time mean of the 70 years of the HadCM3 faf-heat
experiment relative to the control, not including the imposed heat flux
perturbation, in methods A, B and C; (b, d, f) annual time series of the change
relative to control in global-mean surface air temperature ΔT, ocean
volume-mean temperature and the maximum of AMOC streamfunction, with the three
methods. In method A the heat flux perturbation F causes ΔT to
increase (b), giving a negative feedback on ocean heat uptake (d); this
effect is prevented in the other methods, as discussed in Sect. .
![]()
AOGCMs used for FAFMIP preliminary experiments.
Name
L
Ocean horizontal grid
References
F
ΔQ
ΔAMOC
G
NA
FAF
1pct
HadCM3
20
1.25∘ longitude–latitude
1.80
-0.04
0.45
-5.6
-2.3
CanESM2
40
1.4∘ longitude × ∼ 0.93∘ latitude
with small updates*
1.90
0.05
0.53
-7.3
-2.5
GFDL-ESM2M
50
1∘ tripolar, refined at low latitude to 13∘ in tropics
1.86
0.10
0.45
-12.0
-6.8
MPI-ESM-LR
40
0.13–1.65∘ curvilinear
with small updates
1.97
0.15
0.51
-9.5
-3.8
GISS-E2-R-CC
32
1.25∘ longitude × 1.0∘ latitude
* The most important update is
the use of a baroclinicity-dependent formulation for the eddy transfer
coefficient in the scheme of .
The column marked “L” indicates the number of ocean model levels.
The column marked “F” is the ocean area-mean surface heat flux
perturbation (W m-2) in faf-heat.
The columns
marked “ΔQ” indicate the time-mean area-mean difference
in the surface heat flux (W m-2)
computed by the AOGCM between faf-heat and the control,
“G” for the global ocean area, “NA” for the North Atlantic
area marked in Fig. b.
The columns marked “ΔAMOC” indicate the change in the AMOC (Sv),
“FAF” for the time mean of the
last decade of faf-heat compared with its control,
“1pctCO2” for the time mean of years 61–80 in 1pctCO2 compared with
piControl in CMIP5 results with the same model.
In this section we consider methods for treating the surface heat flux in
faf-heat and faf-all. The methods differ regarding the calculation of the net
surface heat flux Q from the atmosphere and sea ice into the ocean water
computed by the AOGCM from its prognostic state. We refer to the method used
by as “B”, and compare it with two alternatives,
referred to as “A” and “C”, with experiments using the HadCM3 AOGCM
(Table ). The three methods are summarised and compared in
Fig. .
In all methods, the net surface heat flux applied to the top-layer θ
in faf-heat is Q+F, where F is the FAFMIP prescribed heat flux
perturbation, and θ stands for the ocean model temperature field –
either potential or conservative, whichever is used in the equation of state
to compute density. Let us write Q=Qc for the piControl experiment
and Q=Qp for faf-heat. In both experiments there is unforced
interannual variation in Q, while the prescribed F has no interannual
variation (although it does have a seasonal cycle). The climatological mean
difference in net surface heat flux between the experiments is
Q+=〈Qp〉+F-〈Qc〉=〈ΔQ〉+F,
where ΔQ=Qp-Qc and 〈〉 indicates a
climatological time mean. The aim is that Q+, the difference in surface
heat flux between faf-heat and piControl, should equal F, the CMIP5
model-mean difference in surface heat flux between the 2×CO2 climate (in 1pctCO2) and piControl.
In method A, the heat flux perturbation F is added to the top
layer in the prognostic equation for θ, and the heat fluxes between
atmosphere, sea ice and ocean are calculated as usual in the AOGCM. Since
F>0 in large regions and in the global mean (Fig. b),
surface air temperature generally rises, causing a negative change ΔQ
in the net surface heat flux into the ocean. This change opposes F, so
Q+‾<F‾ (Eq. ; the overline indicates the
mean over the ocean area as before). In the HadCM3 experiment with method A,
global-mean surface air temperature rises by 0.8 K
(Fig. b), and the ocean area mean 〈ΔQ‾〉=-0.81 W m-2, while F‾=1.86 W m-2. Thus
only 1+〈ΔQ‾〉/F‾=56 % of the heat
flux perturbation is added to the ocean. Locally 〈ΔQ〉 is
generally of opposite sign to F (compare Figs. b
and a), as expected, and it is of particularly large
magnitude in the North Atlantic.
In method B (further discussed in Sect. ), we
introduce a passive tracer TR, i.e. one which does not affect density. It
is initialised to θ at the start of the experiment, and subsequently
transported by all the same processes as θ. The model's surface heat
flux Q is applied to TR as well as to θ, but TR does not
feel the heat flux perturbation F. The critical difference from method A is
that the SST for computing Q is supplied by TR instead of θ,
and is therefore not directly affected by F. This mitigates the feedback in
which ΔQ opposes F. Similarly TR is used instead of θ
in calculations of the heat fluxes between the ocean and sea ice, so that F
does not directly affect the sea-ice heat budget. If F=0, TR evolves
like θ and the climate will be the same as in piControl. Method B is
more complicated than method A because of the need for TR and small
modifications to the coupling to atmosphere and sea-ice submodels. However,
extra tracers are a standard mechanism in many OGCMs because of the role in
ocean biogeochemistry and for diagnostics such as idealised age and chemical
species.
In the HadCM3 experiment with method B, the change in global-mean surface air
temperature is prevented (Fig. b), and ocean area mean
〈ΔQ‾〉=+0.037 W m-2, so 102 % of the
heat flux perturbation is added to the ocean, causing a greater increase in
ocean heat content than in method A (Fig. d). Locally
ΔQ is no longer markedly anticorrelated with F (compare
Figs. b and c). Whereas method A puts
less heat than the intended F into the North Atlantic, method B puts more
than intended, as a result of the weakening of the AMOC caused by the heat
flux perturbation. This change in ocean circulation reduces the advective
heat convergence to the North Atlantic. In consequence, in the unmodified
AOGCM, the regional SST tends to cool, and the surface heat flux into the
ocean tends to increase (although there is still a net heat flux out of the
ocean in the majority of the region). show that
about one-third of the reduction in heat convergence may thus be offset by a
further increase in surface heat flux. This feedback mechanism is presumably
at work in the CMIP5 1pctCO2 experiments from which the faf-heat F has been
calculated, and F therefore includes an enhancement due to reduction of
advective heat convergence in those models. But the mechanism operates in
faf-heat as well, because weakening of the AMOC will reduce the convergence
of TR, from which Q is calculated. Hence this phenomenon is
exaggerated in method B, making Q+ larger than intended. The change in
advection also means that Q+ does not have the intended geographical
distribution.
In method C for faf-heat, the AOGCM uses climatological monthly
time means of SST and sea ice from piControl, instead of the prognostic state
of the system, to compute the ocean surface heat flux Q. The sea-surface
conditions evolve in response to F in the ocean submodel, but these changes
do not affect the atmosphere submodel. Because this method suppresses the
interaction between surface climate and atmosphere, an ocean climate drift
results even if F=0.
In our HadCM3 test of method C, the surface climate for F=0 stabilises
within about 100 years; this timescale is no doubt model-dependent. The ocean
area-mean SST is 1.4 K warmer than in the HadCM3 control, with cooling of
more than 2 K in the North Atlantic, although the AMOC is unaffected in
strength, and warming of more than 2 K in low latitudes. The sea-surface
conditions applying to the ocean and atmosphere are therefore markedly
different, since the latter is prescribed unchanged from the control. In the
global mean the ocean warming penetrates to about 500 m depth, with both
cooling and warming of more than 1 K in magnitude at greater depths in high
northern latitudes. The ocean area-mean surface salinity increases by about
0.5 PSU. These changes are comparable in magnitude with those that result
from 2×CO2 forcing, meaning that for method C, unlike
method B, a new control experiment with F=0 is required in parallel to
faf-heat to evaluate the response to the perturbative F.
In method C, the effect of F on ΔQ via SST is eliminated, and
Q+ is close to F. In HadCM3 with method C 97 % of the global mean
F is added to the ocean, whose heat content therefore increases slightly
less than in method B (Fig. d). However, ΔQ is not
zero everywhere (Fig. e), because the faf-heat and
corresponding piControl integrations have different unforced variability in
the atmosphere, and because changes in ocean surface velocity are seen by the
atmosphere.
Since method C applies the heat flux perturbation accurately to the North
Atlantic, without allowing the strong local feedback on Q, its simulation
of the AMOC decline may give the best estimate of the response to the
intended F. Compared with method C, the AMOC weakening is too small in
method A and too large in method B (Fig. f). We note that
when (, their
Fig. 5) applied surface heat flux perturbations from CMIP5 AOGCMs to the
FAMOUS AOGCM using method B, the weakening of the AMOC in FAMOUS was not
systematically stronger than in the CMIP5 AOGCMs. This indicates that
strength of advection feedback on Q is model-dependent, as we also see
later (Sect. ).
Although method C is arguably most accurate, it has the disadvantages that it
is more computationally expensive (because of the need for a new control
integration), the ocean climate state is different from the unmodified AOGCM
whose response we wish to investigate and the physical interaction between
atmosphere and ocean is unrealistically suppressed, including feedbacks which
could be of interest. We therefore adopt method B for faf-heat.
Added and redistributed heat
Changes in θ in method B can be partitioned into those due to modified
tracer transport processes (due to change in circulation, diffusion, etc.)
and those due to added heat following. We
are interested in the evolution of the climatological state, so all terms
should be interpreted as climatological time means (and we omit
〈〉 for the sake of legibility). As in the previous section, we
use subscripts c and p to denote variables in the piControl and perturbed
(faf-heat) experiments respectively. By Φ(θ) we denote the net heat
convergence due to all heat fluxes in the interior of the ocean,
both resolved and parametrised subgridscale. The function Φ depends on
diffusivities and other attributes of the model state which affect heat
transport, as well as the velocity.
In the piControl experiment,
∂θc∂t=Qc+Φc(θc),
setting the volumetric heat capacity to unity for convenience.
The θ field is three-dimensional but Q applies only at the surface.
In faf-heat θ=θp is affected by the imposed heat flux
perturbation F as well as by the atmosphere–ocean heat flux Q=Qp
simulated by the AOGCM, so
∂θp∂t=Qp+F+Φp(θp),
where Φp is a different function from Φc because of
changed velocities, diffusivities, etc.
The redistributed heat tracer TR, which we described for method B in
Sect. , is initialised to θc and has
Qp=Qc+ΔQ as its surface flux, so its evolution
equation is
∂TR∂t=Qp+Φp(TR).
Since TR is initialised to θc, we write
TR=θc+ΔTR, i.e. ΔTR=0 initially. Let us
also split Φp into Φc+ΔΦ. (For example,
this splits the advective heat convergence into the part
-∇⋅(TRvc) due to the piControl
velocity field vc and the part
-∇⋅(TRΔv) due to the change in the
velocity field with respect to the piControl.) These decompositions assume
that the heat convergence function depends linearly on the relevant variables
of the climate state and acts linearly on the tracers. In that case
Eq. () becomes
∂TR∂t=Qc+ΔQ+Φc(θc)+ΔΦ(θc)+Φp(ΔTR)=ΔQ+ΔΦ(θc)+Φp(ΔTR)
if the piControl is a steady state, so that Qc+Φc(θc)=0 (Eq. ). TR is called the
“redistributed heat” tracer by , because it diagnoses
the effect of changes in tracer transport processes (changes in circulation,
diffusivities, etc., giving rise to ΔΦ) on the unperturbed
θc. If ΔΦ vanishes, and assuming ΔQ=0 as
well, ΔTR=0 always, meaning that TR evolves identically to
θc and thus they remain equal. Changes in ocean heat transport
may induce a non-zero ΔQ (Sect. ), which will affect
TR as well.
Ocean model diagnostics of particular interest to FAFMIP analyses
(as well as the process-based diagnostics of Table ).
CMIP short name
Unit
CF standard name
zos
m
sea_surface_height_above_geoid
zostoga
m
global_average_thermosteric_sea_level_change
thetao
∘C
sea_water_potential_temperature
*bigthetao
∘C
sea_water_conservative_temperature
thetaoga
∘C
(volume mean of thetao)
*bigthetaoga
∘C
(volume mean of bigthetao)
*opottempmint
∘C kg m-2
integral_wrt_depth_of_product_of_sea_water_density_and_potential_temperature
*ocontempmint
∘C kg m-2
integral_wrt_depth_of_product_of_sea_water_density_and_conservative_temperature
so
1 × 10-3
sea_water_salinity
*somint
1 × 10-3 kg m-2
integral_wrt_depth_of_product_of_sea_water_density_and_salinity
msftmyz
kg s-1
ocean_meridional_overturning_mass_streamfunction
msftyyz
kg s-1
ocean_y_overturning_mass_streamfunction
hfds
W m-2
surface_downward_heat_flux_in_sea_water
wfo
kg m-2 s-1
water_flux_into_sea_water
*rsdoabsorb
W m-2
net_rate_of_absorption_of_shortwave_energy_in_ocean_layer
*pathetao
∘C
sea_water_additional_potential_temperature
*prthetao
∘C
sea_water_redistributed_potential_temperature
*pabigthetao
∘C
sea_water_additional_conservative_temperature
*prbigthetao
∘C
sea_water_redistributed_conservative_temperature
* Indicates diagnostics which are newly introduced in
CMIP6.
The CMIP short names are used by the Climate Model Output Rewriter (CMOR)
software and in naming datasets to be submitted to CMIP6.
The CF standard names are defined by the CF metadata convention
(http://www.cfconventions.org).
In order to reveal where the extra heat from the heat flux perturbation is
stored in the ocean, we include an “added heat” tracer TA in faf-heat.
This tracer is initialised to zero (so ΔTA≡TA) and it has
F as its surface flux (we note that heat is added in the global mean,
although F is not positive everywhere, as seen in
Fig. b). Its evolution equation is
∂TA∂t=F+Φp(TA).
so TA=0 always if F=0. This tracer is similar to the “passive
anomalous temperature” of , whose experimental design was
different. The added heat tracer is also included in the faf-passiveheat
experiment, where its surface source is the same but its evolution is
different, because it is subject to the same circulation and subgridscale
processes as in the control state, and Φc replaces Φp in
Eq. ().
Considering Eqs. (), () and (), we see that
∂θp∂t=∂TR∂t+∂TA∂t.
Thus we can interpret changes in ocean heat content in faf-heat as the sum of
redistribution (including the effect of ΔQ) and addition. In
practice, achieving exact equality may not be possible due to non-linearities in the
implementation of tracer transport operators.
Careful formulation is required to ensure that Q is applied in the same way
to θ and TR, and some differences may be unavoidable, depending
on model formulation. In particular, absorption of solar radiation should
occur with the same vertical profile for both (assuming that some of it
penetrates the top layer), and the same heat flux should be applied to both
of them for evaporation and precipitation (if the sensible heat content of
these water fluxes is considered in the model). If the same amount of heat is
extracted from both tracers for frazil sea-ice formation, θ may
sometimes fall below freezing point, requiring special treatment of the
equation of state; on the other hand if θ and TR are separately
kept above freezing, there will be a difference in the heat fluxes implied.
Further technical notes can be found at
http://www.fafmip.org. It may be useful to check the
implementation of TR in the model with an experiment in which F=0,
which should reproduce the piControl experiment.
Diagnostics
FAFMIP experiments should include standard CMIP6 monthly mean and other
diagnostics of atmosphere, ocean and cryosphere, as in the CMIP6 DECK, which
is a small set of experiments (including piControl and 1pctCO2) used to
evaluate model characteristics of climate and climate change
. These standard diagnostics provide a large amount of
information which will support many kinds of analysis that cannot be
anticipated in detail. The standard ocean diagnostics are described in detail
for the Ocean Model Intercomparison Project (OMIP) by
in this issue, and in Table we list a subset of particular
importance to FAFMIP, for which they are priority 1 as monthly means. We
refer to them here by their CMIP “short names”.
Analysis of sea-level change and ocean heat uptake will use diagnostics of
sea level, ocean temperature and salinity (zos, zostoga,
thetao or bigthetao, thetaoga or
bigthetaoga, opottempmint or ocontempmint,
so and somint, where the choice of alternatives depends
on whether the prognostic ocean temperature is potential or conservative).
Analyses of the AMOC will use the overturning streamfunction
(msftmyz or msftyyz). The faf-heat and faf-passiveheat
experiments should include monthly means of the added heat tracer TA
(pathetao or pabigthetao), and faf-heat should include
monthly means of the redistributed heat tracer TR (prthetao or
prbigthetao).
New diagnostics for process-based ocean temperature
and salinity tendencies, required by FAFMIP and
described in detail in Sect. 9 of .
CMIP short name
CF standard name
opottemptend
tendency_of_sea_water_potential_temperature_expressed_as_heat_content
opottemprmadvect
tendency_of_sea_water_potential_temperature_expressed_as_heat_content_due_to_residual_mean_advection
opottemppadvect
tendency_of_sea_water_potential_temperature_expressed_as_heat_content_due_to_parameterized_eddy_advection
opottemppmdiff
tendency_of_sea_water_potential_temperature_expressed_as_heat_content_due_to_parameterized_mesoscale_diffusion
opottemppsmadvect
tendency_of_sea_water_potential_temperature_expressed_as_heat_content_due_to_parameterized_submesoscale_advection
opottempdiff
tendency_of_sea_water_potential_temperature_expressed_as_heat_content_due_to_parameterized_dianeutral_mixing
ocontemptend
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content
ocontemprmadvect
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content_due_to_residual_mean_advection
ocontemppadvect
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content_due_to_parameterized_eddy_advection
ocontemppmdiff
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content_due_to_parameterized_mesoscale_diffusion
ocontemppsmadvect
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content_due_to_parameterized_submesoscale_advection
ocontempdiff
tendency_of_sea_water_conservative_temperature_expressed_as_heat_content_due_to_parameterized_dianeutral_mixing
osalttend
tendency_of_sea_water_salinity_expressed_as_salt_content
osaltrmadvect
tendency_of_sea_water_salinity_expressed_as_salt_content_due_to_residual_mean_advection
osaltppadvect
tendency_of_sea_water_salinity_expressed_as_salt_content_due_to_parameterized_eddy_advection
osaltpmdiff
tendency_of_sea_water_salinity_expressed_as_salt_content_due_to_parameterized_mesoscale_diffusion
osaltpsmadvect
tendency_of_sea_water_salinity_expressed_as_salt_content_due_to_parameterized_submesoscale_advection
osaltdiff
tendency_of_sea_water_salinity_expressed_as_salt_content_due_to_parameterized_dianeutral_mixing
The units of the temperature tendency diagnostics are W m-2,
and of the salinity tendency diagnostics kg m-2 s-1.
Either the potential temperature or the conservative temperature diagnostics
should be included, depending on which is the prognostic of the model.
The effect of advection by the model (resolved) velocity field
can be calculated as
the difference between the effects of residual mean advection
and parametrised eddy advection.
The latter should include both mesoscale and submesoscale effects.
If these are not distinguished in the model,
the diagnostics for parametrised submesoscale advection should be omitted.
Analysis of ocean tracer budgets will use the ocean surface heat and water
fluxes requested as standard CMIP monthly diagnostics. Surface fluxes affect
only the top layer of the ocean, except for shortwave (solar) radiation,
which penetrates more deeply (diagnosed by rsdoabsorb). The net
surface heat and water fluxes into seawater (hfds and wfo)
are particularly useful, because model-dependent details of implementation,
especially regarding sea ice, can make it an intricate or impossible task to
compute the net fluxes from other CMIP diagnostics. The net surface flux
diagnostics of heat and water are defined somewhat inconsistently by CMIP6
and CF, in that hfds should contain Q computed by the model,
not including the FAFMIP heat flux perturbation, but wfo
should include the FAFMIP water flux perturbation. The FAFMIP
steering committee will request each participating group to supply files of
the flux perturbations (of momentum, heat and water) as actually applied by
them, on the grid of their ocean model, to be made available on the project
website for use in inter-comparative analysis.
Inter-comparative analysis of ocean interior change is a priority for FAFMIP,
motivating the introduction of the three-dimensional process-based tendency
diagnostics for prognostic temperature and salinity (Table ). These
diagnostics are described in detail in Appendix L of .
Different models parametrise interior transports in many ways, so for the
purpose of intercomparison it is necessary to aggregate them into broad
classes. We distinguish advection by the model velocity field, parametrised
eddy advection (mesoscale and submesoscale if treated separately), mesoscale
diffusion (by eddies along neutral or isopycnal surfaces), and dianeutral
mixing (including diapycnal diffusion, convection and boundary-layer mixing).
In addition there is a net tendency diagnostic, whose time-integral over any
period should equal the change in the prognostic tracer between the start and
end of that period. The difference between the net tendency and the sum of
the individual process diagnostics will yield a residual that accounts for
any other schemes not separately identified, including the effect of surface
fluxes. The tendency diagnostics are expressed as rates of change of heat and
salt content in grid cells, i.e. ∂(mCpθ)/∂t and
∂(mS)/∂t, where S is salinity, m is the mass per unit
area of the grid cell and Cp the specific heat capacity. In Boussinesq
models with fixed cell thicknesses, m is a constant for each grid cell, but
otherwise it is variable.
The tendency diagnostics are requested at priority 1 as annual means, and at
priority 2 as monthly means for analysis of high-frequency variability,
recognising that this implies a substantial amount of storage. Diagnostics of
vertical and lateral tracer diffusivity detailed in Appendices M and N of
are also requested at priority 1 as annual means. The
tendency and diffusivity diagnostics should be included in the DECK 1pctCO2
and abrupt4xCO2 experiments, and in the piControl experiment, at least in the
portion which is parallel to the FAFMIP experiments as well as in the FAFMIP
experiments. These diagnostics will give information which has never
previously been available for AOGCMs in general, concerning the roles of the
various interior processes in the maintenance of the steady state, unforced
variability and the response to climate change.
Summary and plans
The purpose of the Flux-Anomaly-Forced Model Intercomparison Project is to analyse the simulated response of the ocean to changes in surface
fluxes resulting from CO2 forcing in AOGCMs. The specific interests
which motivated the proposal of FAFMIP are
The magnitude of ocean heat uptake in response to climate
change, which determines global-mean sea-level rise due to thermal expansion
and influences the transient climate response.
The geographical patterns of sea-level
change due to ocean density and circulation change simulated by the models.
The weakening of the Atlantic meridional overturning circulation, which
affects regional sea-level rise and climate change.
The ocean's role in determining the patterns of sea-surface temperature
change, which influences climate sensitivity to CO2.
Subsurface warming of the ocean near to the Greenland and Antarctic
ice sheets, where it might enhance basal melting of ice shelves and hence
sea-level rise through the dynamical response of the ice-sheets.
These topics are all aspects of the Earth system response to forcing, and
they are of particular relevance to the WCRP Grand Challenges on regional
sea-level rise, melting ice, and climate sensitivity. The motivation for
FAFMIP is to find ways of reducing the uncertainty in projections in
policy-relevant scenarios, by applying observational constraints and improved
physical understanding to refine the models.
In the FAFMIP tier-1 experiments faf-stress, faf-heat and faf-water,
prescribed perturbations are applied to the ocean surface in the fluxes of
momentum, heat and freshwater respectively. The flux perturbations have a
seasonal cycle but no interannual variation, and are obtained from a model
mean of changes simulated in CMIP5 AOGCM experiments at year 70 in 1pctCO2
experiments (with CO2 increasing at 1 % year-1). They are thus
typical of simulated CO2-forced climate change in magnitude and
geographical pattern. The intention of applying the same surface flux
perturbations in all AOGCMs in FAFMIP is to reveal the dependence of the
response on the ocean model. The FAFMIP tier-1 experiments amount to
210 years of integration, which is a modest requirement compared with many
CMIP6 subprojects. There are two tier-2 experiments of 70 years each, one of
which can be achieved by adding a diagnostic to the control experiment,
thereby avoiding the need for a separate integration. We have carried out
preliminary tier-1 experiments with pre-CMIP6 AOGCMs to test and demonstrate
the experimental design. Our models exhibit diversity in the pattern and
magnitude of simulated changes, with some common qualitative features.
We find that momentum and water flux perturbation do not affect the AMOC
significantly, but the AMOC weakens in faf-heat, by 6–12 Sv depending on
the model, in response to the heat added to the North Atlantic. The AMOC
weakening is reinforced by a feedback on the surface heat flux whereby, as
the AMOC declines, the SST in the North Atlantic tends to cool, so the heat
flux from the atmosphere to the ocean increases
. This effectively doubles the heat
flux perturbation in that region (although it is a small effect in the global
mean). Consequently the AMOC weakening in faf-heat is larger than the
expected response for 1pctCO2. However, the net extra heat input to the North
Atlantic, including the feedback, is similar in all the models, indicating
that the model spread in AMOC weakening in faf-heat is mainly due to
differences in ocean model response, rather than to a spread in the buoyancy
forcing.
Despite its exaggerated magnitude, this coupled feedback is a physical effect
which must also occur in the CMIP5 1pctCO2 experiments from which the heat
flux perturbation was derived , and presumably in
general in climate change simulated by AOGCMs. Our results therefore strongly
suggest that it is an important effect on the weakening of the AMOC in
response to CO2 forcing, and may not have been sufficiently appreciated.
found a similar large positive feedback from
increased heat input on the weakening of the AMOC in response to addition of
freshwater around Greenland. We note that the positive feedback on AMOC
weakening is a distinct effect from the negative feedback on AMOC weakening
in which the cooling in the North Atlantic promotes convection and deep water
formation and tends to strengthen the circulation e.g..
The negative feedback is an oceanic phenomenon, not a coupled one.
Global-mean surface air temperature cools over a large fraction of the world
in faf-water, by 0.3 K in the global model mean. The global-mean input by
the water flux perturbation is very small compared with its local values (its
ocean area mean is 7.2×10-8 kg m-2 s-1, 2 orders
of magnitude smaller than its spatial standard deviation of 5.4×10-6 kg m-2 s-1), so the phenomenon is probably a response
to its geographical pattern. Global-mean surface temperature change is small
in faf-stress and faf-heat (note that the heat added to the ocean in faf-heat
is prevented from directly affecting the surface air temperature), but there
is substantial warming near to Antarctica and in the Arctic, and strong
cooling in the North Atlantic and northern mid-latitude land areas in
faf-heat associated with the AMOC weakening. Heat is added in faf-heat mainly
at high latitude, and is transported equatorward and downward in a
model-dependent way. This implies a spread in ocean heat uptake efficiency
and global-mean sea-level rise due to thermal expansion.
As in many previous studies, the main geographical features of dynamic
sea-level change are an increase in the gradient across the ACC (small
sea-level rise to the south, large to the north), a dipole in the North
Atlantic (small sea-level rise in the subtropical gyre, large sea-level rise
to the north), and enhanced sea-level rise in the Arctic. We find that the
Southern Ocean feature is caused in roughly equal measure by momentum and
heat flux perturbations, and somewhat counteracted by the water flux
perturbation. In the Southern Ocean, where there is the greatest increase in
ocean heat content in faf-heat, heat uptake is largely passive, while in
faf-stress there is wind-driven redistribution of heat from high to low
latitude. The Arctic feature is mainly due to the water flux perturbation.
The North Atlantic feature results from the heat and water flux
perturbations, which both give a meridional contrast in buoyancy flux
(greater to the north, causing more sea-level rise). In faf-heat this effect
is opposed by reduced heat transport due to the weakening of the AMOC, which
redistributes heat from high to low northern latitude. Redistribution is also
responsible for strongly enhanced sea-level rise along the Atlantic coast of
North America in faf-heat.
The results from the pre-CMIP6 trial experiments show that there will be
many qualitative and quantitative features to be analysed in CMIP6. The CMIP6
FAFMIP experiments and the piControl and idealised CO2 experiments with
FAFMIP models will contain diagnostics for rates of change of temperature
and salinity due to separate ocean interior transport processes (advection,
diffusion, etc.). Such diagnostics have been available in only a few models
previously, and were not included in the preliminary experiments that we have
carried out for this paper. They will yield a great deal of new information.
In the piControl the diagnostics will enable us to study the balance of ocean
processes in the mean state and unforced variability of the coupled
atmosphere–ocean system. In the FAFMIP experiments and the idealised
CO2 climate-change experiments they will allow us to identify the
mechanistic explanations both for the common features of the model responses
to surface flux forcing and for the differences among models.
The FAFMIP steering committee will promote the analysis of the experiments,
bearing in mind the scientific questions which motivated the project.
Comparison of the results from different AOGCMs will aim to identify the
causes of the spread in their simulated climate change, in terms of model
formulation and emergent behaviour. We envisage that in the light of further
analysis we may devise additional tier-2 experiments, for instance to study
the effect of surface heat flux feedbacks. It may also be useful to carry out
ensemble experiments to quantify the influence of unforced variability,
although the major features of the forced response are expected to be robust
in view of the large size of the perturbations. Using the faf-all experiment,
the effect of combining the flux perturbations will be studied.
The application of common surface flux perturbations is a technique which has
not been widely used up to now as a means to study ocean climate change
simulated by AOGCMs in response to CO2 forcing. We therefore hope that
the FAFMIP experiments will offer new insight into the reasons for model
spread in the ocean response, without the confounding influence of diversity
in atmospheric response. Where the patterns of ocean climate change differ
among the models in FAFMIP experiments, we expect that these differences will
correspond to those which the models exhibit in the AOGCM scenario-forced
projections. On the other hand, when the ocean models agree in FAFMIP
experiments, it will give us greater confidence in the results, and we will
be able to infer that the atmosphere models are the source of uncertainty in
projections of ocean climate change.