Introduction
The tropical overturning circulation driven by tropical convection is an
important driver of the hydrological cycle of the tropics and subtropics.
Additionally, variability in tropical convection can cause changes to
midlatitude weather patterns on a wide range of timescales
e.g.,. The tropical overturning
circulation can be broken down into zonal (Walker circulation; henceforth WC)
and meridional (Hadley circulation; henceforth HC) components. Both the WC
and HC are susceptible to external forcing, and numerous studies have shown –
both theoretically and through modeling – that the WC and HC (at least in
the Northern Hemisphere) will weaken in response to climate change
e.g.,.
This is associated with an overall weakening of the global hydrological cycle
and weaker convective updrafts over the tropical oceans
e.g.,.
Observation-based analysis for the historical period, and modeling studies
for historical and future periods, however, are inconclusive about important
aspects of the tropical circulation response to climate change. An
alternative line of investigation is to use a paleoclimate interval, such as
the mid-Pliocene warm period (henceforth mPWP; ca. 3.3–3 Ma), as an
imperfect analogue of present and near-future climate
e.g.,. The mPWP was the most recent period
where CO2 concentrations were similar to those levels that are projected
to be reached this century and the
continental configuration was highly similar (relative to other paleoclimate
intervals further in the past) to present . Since a
wealth of paleoclimate data is available for the mPWP, and its climate
conditions are expected to be similar to the climate of the near future
, this interval is considered useful for validation of
future climate simulations. Previous modeling studies examining changes to
the tropical circulation in the mPWP found a weakening of the WC and
HC similar to studies examining future climate
, but these are single-model
studies only, where the results could be skewed by individual model biases.
One of the key features of the mPWP derived from paleoclimate records was
significantly warmer high-latitude regions compared to present, with
relatively similar tropical sea surface temperatures (SSTs) to modern
.
There are also strong indications that the upwelling regions off the coast
of western South America were much warmer than present, causing a weakening
of the west–east SST gradient in the tropical Pacific and indicating a
flattening of the thermocline
. These changes in zonal
and meridional SST gradients could have had important effects on the tropical
overturning circulation of the mPWP, and studying these effects could provide
insight into the usefulness of the mPWP as an analogue for future climate
change since they differ significantly from present day.
Multiple effects due to increased greenhouse gases (GHGs) can slow down the
tropical overturning circulation . Initially, GHG
forcing alone warms the free troposphere, which tends to increase atmospheric
static stability over the tropical oceans and reduce convective
precipitation. This effect, sometimes referred to as the “direct” GHG
forcing, occurs almost immediately after GHG is increased, as the mid-upper
troposphere warms in response to decreased radiative cooling to space
. An indirect effect of GHG forcing is warming of the
surface, with the land warming almost instantaneously, and ocean warming
lagging by several decades, causing increased land–sea temperature contrasts
. This tends to lower surface pressures over
land, relative to the oceans, which leads to an increased land–sea pressure
gradient, moisture convergence and precipitation over land, and a
strengthening of the monsoonal circulations
e.g.,. The delay in surface warming over the
oceans increases static stability there as the free troposphere warms, while
stability is either unchanged or decreases over land where surface
temperatures can warm at a similar rate to the free troposphere
.
Eventually the SSTs also warm, increasing boundary layer specific humidity,
which – assuming fairly constant relative humidity under climate change
– increases nonlinearly with increasing
temperature by the Clausius–Clapeyron relation. In order for precipitation to
balance evaporation on a global scale, and since most precipitation occurs
through convection in the tropics, the tropical convective mass flux must
decrease . In other words, tropical convective
updrafts do not have to be as strong as before to maintain the balance between
evaporation and precipitation due to the increased water vapor
content due to climate change.
A factor that can alter local (although not global mean) tropical convection
from increasing GHGs is the spatial pattern of the SST warming. Regions where
SSTs warm faster (slower) than the tropical mean are likely to experience
smaller (larger) increases in static stability than the tropical mean. This
is because the warming of the tropical troposphere is fairly uniform since
temperature gradients are difficult to maintain there
. This change to the regional pattern of tropical
convection, caused by the pattern of SST warming and increased land–sea
contrast, could lead to a redistribution of tropical and subtropical
precipitation e.g.,.
The weakening of the tropical zonal overturning circulation (WC) is among the
most robust responses of the tropical circulation to future climate change
. The primary driver of the
projected slowdown of the WC is thought to be from warming SSTs and
associated reduction in convective mass flux
.
Additionally, it is well known that the WC strength can be modulated by the
zonal gradient in tropical SST ; for example,
on interannual timescales the warm phase of ENSO (El Niño–Southern Oscillation) reduces the
zonal SST gradient and weakens the WC. Thus, any change to zonal SST
gradients resulting from increased GHGs is likely to be important for
modulating the WC response to climate change. During the mPWP, the zonal SST
gradient in the tropical Pacific is thought to have been much weaker than
present – primarily due to warmer SSTs in the upwelling regions off the west
coast of South America – causing the WC to be weaker than present
. Observations and reanalysis data show
contradictory trends in recent WC strength over the satellite era, with some
indicating weakening
e.g., and others
indicating strengthening
e.g.,. Nonetheless,
future projections from the Coupled Model Intercomparison Project Phase 5
(CMIP5) indicate a robust weakening of the WC
and overall tropical overturning circulation over the 21st Century, which is
associated with reductions in the zonal gradient of tropical SSTs
.
The weakening of the meridional overturning circulation (HC) in response to
climate change is less robust than the weakening of the WC. A poleward
expansion of the descending branch of the HC in the wintertime Northern
Hemisphere is considered the most robust projection for the future response
. Additionally, as the
troposphere warms, the HC is expected to expand vertically as the tropopause
height increases in the tropics, an effect which may already be apparent in
observations . The CMIP5 models show fairly
good agreement for a weakening of the northern hemispheric HC, with
substantial disagreement over the response of the southern hemispheric cell
. However,
satellite observations and reanalysis data suggest that the HC has, in fact,
strengthened rather than weakened since 1979 . This apparent contradiction may be the result of poor
model parameterization of clouds or convection
e.g., or due to natural
variability. For example, internal fluctuations in the tropical and
extratropical oceans have effects on the tropical circulation in the short
term that could be masking a longer-term trend .
Part of the variability in HC strength could be explained by changes in
meridional SST gradients, which have been shown to weaken (strengthen) the HC
if these gradients weaken (strengthen)
e.g.,.
The meridional SST gradient from the tropics to the poles is greatly reduced in
the mPWP
e.g.,,
which to leading order weakens the HC as shown in .
This reduced-meridional-SST-gradient world can provide a test bed for the
climate models to better gauge the sensitivity of the HC to these boundary
conditions.
Our study aims to examine the tropical circulation of the mPWP using multiple
general circulation models (GCMs). Our paper attempts to show the
similarities and differences of the tropical circulation of the mid-Pliocene
to the tropical circulation under anthropogenic climate change. As such, the
mPWP provides a unique opportunity to gain a better understanding of large-scale climate dynamics in a warmer world. The structure of this article will
be as follows: Sect. 2 will explain the model dataset and diagnostics that
will be used for our analysis in the data and methods section. Then Sect. 3
will examine the modeled tropical circulation of the mPWP with comparisons to
CMIP5, broken down into one section for the WC and one for the HC. Section 4
will present further discussion and our primary conclusions.
Data and methods
Climate model simulations
To examine the tropical circulation of the mPWP, we used the Pliocene Model
Intercomparison Project (PlioMIP), which aims to simulate the climate of the
mPWP . Two different
groups of models were run; atmosphere-only general circulation models (AGCMs)
and coupled ocean–atmosphere general circulation models (AOGCMs). The AGCM
models use a prescribed monthly SST climatology that does not change through
the integration, while the AOGCM SSTs are allowed to evolve freely.
Hereafter, we refer to the AGCM model group as PRES (for prescribed SSTs)
and the AOGCM group as CPLD (for coupled ocean–atmosphere models). The
mid-Pliocene boundary conditions are based on the Pliocene Research,
Interpretation and Synoptic Mapping Phase 3 (PRISM3) paleoclimate
reconstructions . A list of the PlioMIP models
examined in this study is included in Table 1, and further details on the
experiments and model configuration can be found in
.
The list of models available from PlioMIP for the prescribed SST
(PRES) and coupled ocean–atmosphere (CPLD) experiments. The
“X” (dash) indicates that the model data
are available (unavailable). Two models in CPLD (GISS and NCAR CCSM4) do not
have equivalents in PRES (we elected not to use data from NCAR CAM3.1
because it was not available for pressure levels). In the left column, the
letter in parentheses after the model name shows its index in Fig. 6. The
right column shows the tropical (30∘ S–30∘ N) mean
near-surface air temperature response (ΔTs, units K) for each model
in CPLD.
PlioMIP model list
Model (letter)
PRES
CPLD
CPLD ΔTs
(K)
GISS (A)
–
X
1.36
HadCM3 (B)
X
X
2.56
MIROC4m (C)
X
X
2.58
MRI-CGCM2.3 (D)
X
X
1.48
NCAR CCSM4 (E)
–
X
1.24
NorESM-L (F)
X
X
1.89
The experimental setup for each group of models is as follows. Atmospheric
CO2 levels are set at 405 ppm for both groups of models, while all other
GHGs are kept the same as the preindustrial (PI) control runs. Additionally,
the solar constant, orbital parameters and aerosol concentrations are kept
the same as PI as well. Sea surface temperature, coastlines, vegetation type
and ice sheet boundary conditions are specified from PRISM3
. The models in PRES are run with an integration
length of 50 years, with 20 years for spin-up, giving 30 years of
mid-Pliocene climatology. The models in CPLD are run with an integration
length of 500 years or longer to allow the ocean to fully equilibrate to the
increase in GHGs. Note that the PRISM3 boundary conditions employ a technique
described as “warm peak averaging” by to
represent the average conditions of the warm periods over the approximately
300 ky duration of the mPWP. It has been suggested that using paleoclimate
reconstructions over such a long period may not be appropriate because
shorter-term fluctuations in climate forced by changes in orbital parameters
mean that proxy data may not be representative of conditions from the time
period consistent with the boundary conditions
e.g.,. Indeed, over the ∼ 300 ky
mPWP, it has been shown that solar insolation exhibited very large
fluctuations . The next phase of the PlioMIP
project, PlioMIP2, will mitigate this issue by focusing on a narrower
time period at 3.2 Ma, with consistent orbital and SST forcing.
Climate diagnostics
To compute diagnostics to measure tropical circulation change, some
post-processing was necessary on the data from PlioMIP. The data were
interpolated to a 2.5 × 2.5∘ grid for all models to create a
common resolution for computing multi-model mean statistics. We define the
response as the difference, for any given quantity, between the mPWP
simulation and the PI control simulation for each model. The response is
evaluated using monthly mean climatologies computed over all available years
of simulation (typically 150 years in most models). Meridional wind data were
available on 10 common vertical levels between 1000 and 100 hPa for all
models listed in Table 1. Lastly, if a model did not contribute output to the
CPLD experiment, it was not included in our analysis.
We compute multi-model means for PRES and CPLD using the models listed in
Table 1. Each model's response is first scaled by its tropical mean
(30∘ S–30∘ N) surface temperature response in CPLD
(Table 1), consistent with the procedure followed for future climate
simulations by . In figures showing multi-model
means, stippling indicates regions of agreement across the model ensemble
and is added where all but one member in the model group agree on the sign
of the response. The small number of models in each PRES and CPLD prevents a
more robust statistical analysis of the model agreement. JJA refers to a
3-month June–July–August mean, while DJF refers to a 3-month
December–January–February mean.
To diagnose changes to the HC, zonally averaged meridional mass
stream function (Ψ) was calculated , then
latitude-height cross sections of this quantity were created. This field
highlights the vertical and meridional circulation associated with the HC,
with positive (negative) values indicating clockwise (counterclockwise)
overturning. To diagnose changes to the WC, we use upper-tropospheric
(200 hPa) velocity potential (χ), which is anticorrelated with regions
of divergence (convergence) that are strong indicators of upwelling
(downwelling) associated with the WC. We also use sea level pressure (SLP) to
diagnose changes to the WC and overall tropical overturning circulation (SLP
data were not available from the MIROC4m model).
Simulated changes to the tropical circulation during the mid-Pliocene
Response of the tropical climate
We begin with a brief overview of the simulated changes to the multi-model
mean, annual mean tropical climate in the mPWP, relative to the
PI. In CPLD the tropical oceans warm by ∼ 0.5–1.5 K
(Fig. 1a), and tropical land regions generally warm even more than the oceans
(upwards of 5 K in some arid regions). The 850 hPa wind response indicates a
strengthening of the easterly trade winds across much of the tropics and
subtropics, with indications of strengthened southwesterly monsoonal flow
over the northwestern Indian Ocean.
CPLD (a) and PRES (b) multi-model mean, annual
mean surface temperature response (shading, units K) and 850 hPa wind
response (reference vector denotes 3 m s-1 K-1). Prior to
computing the multi-model mean, the response in each model is scaled by its
tropical mean surface temperature response in CPLD (see Sect. 2.2 and
Table 1). (c) CPLD multi-model annual mean sea surface temperature
(SST) minus PRISM3 reconstructed SST.
Over a large region of the tropical ocean, the imposed SST perturbation in
PRES (Fig. 1b) shows weak warming compared to PI. The only region that sees
significant warming in the tropics in PRES is the eastern Pacific, especially
near the coast of South America, with some regions warmer by up to 5 K,
which drastically weakens the west–east SST gradient in the tropical Pacific.
The PRES simulations do warm in the subtropics, and they show even greater warming
(by a factor of 1.2 in the Northern Hemisphere (NH) and 1.1 in the Southern Hemisphere (SH)) than CPLD in the
extratropics, especially at high latitudes (not shown). Additionally, the
tropics (30∘ S–30∘ N) warm more in CPLD (1.75 K) than in
PRES (0.95). This acts to reduce the meridional equator–pole surface
temperature gradient more in PRES than in CPLD.
As Fig. 1, except showing the precipitation response as a percentage
of the preindustrial control climatology (shading). The precipitation rate
from the preindustrial control climatology is shown by thick black contours,
interval 2 mm d-1 and starting at 4 mm d-1. Stippling indicates
regions where at least all but one of the models in the model group agree on the sign
of the response.
Tropical precipitation generally increases in CPLD, particularly over the
oceans at the outer edges of the climatologically “wet” zones (Fig. 2a). This
pattern indicates an expansion of the wet regions on both sides of the
equator, but with a significantly stronger response in the NH. Land precipitation increases substantially in a band
covering the Sahel, the Middle East and southern Asia. By contrast, precipitation
decreases over the South Atlantic, the eastern South Pacific Convergence Zone
(SPCZ) and over continental southern Africa and Amazonia.
The precipitation response in PRES (Fig. 2b) is rather different to that in
CPLD, with moderate decreases over the climatological wet regions and
increases over much of the southern tropical ocean including over the eastern
SPCZ. While precipitation increases poleward of the northern Intertropical Convergence Zone (ITCZ) in CPLD and
PRES, in PRES it decreases on the southern flank along the equator. Land
precipitation shows similar or even stronger increases than CPLD in the
band between northern Africa and southern Asia, and over Australia, regions where
the simulated response agrees closely with available paleoclimate records
. The enhanced rainfall
over relatively arid land regions is consistent with enhanced monsoonal
activity . We note that the largest differences
in precipitation response between CPLD and PRES (Fig. 2) occur where there
are the largest differences in SST warming (Fig. 1).
Response of the Walker circulation
The ascending branch of the annual mean WC is located over the Maritime
Continent, where the region of strongest climatological upper-tropospheric
divergence (negative 200 hPa χ) indicates large-scale ascent and
convective outflow (Fig. 3a). The compensating descending branch of the
climatological WC is located over the eastern Pacific in a region of strong
convergence. The response of the WC to mPWP climate change in CPLD indicates
a slight (∼ 3–4 %) weakening of the χ field over the Maritime
Continent (based on the multi-model mean of the percentage change of the
absolute value of the minima in the 200 hPa χ field for each model in
that region), a strengthening over the eastern Pacific and a westward shift
of the ascending branch over the western Indian Ocean (Fig. 3a). However, the
weakening of the ascending branch is not robust, as indicated by a lack of
model consensus on the sign of the response, with two models (NCAR CCSM4 and
NorESM-L) showing a slight strengthening.
The WC response in PRES shows an unambiguous weakening, with decreased
divergence over the western Pacific and Indian oceans, and increased
divergence over Africa (Fig. 3b). The pattern of χ response projects
negatively onto the background climatological χ: weakening of the
divergence is focused in the regions of strongest climatological ascent,
extending from the Indian Ocean into the western and central Pacific, and
increased divergence is found over the eastern Pacific near the
climatologically descending limb of the WC. The overall weakening response of
the WC in PRES is highly robust among the different models – more so than
for CPLD – with unanimous agreement on the sign, which suggests a role for
tropical SST changes. However, there is a large spread among the four models in
PRES for the WC weakening of ∼ 1–29 % using the same WC strength-change metric used for CPLD.
As Fig. 1, except showing the 200 hPa annual mean velocity
potential (χ) response (shading) and 200 hPa divergent wind annual mean
response vectors (reference vector denotes 1 m s-1 K-1).
Contours show 200 hPa χ from the preindustrial control climatology,
interval 2 × 106 m2 s-1 with dashed contours
negative and the zero line thickened. Stippling as in Fig. 2.
Another way to examine the WC response is through the surface response, by
using SLP to measure changes in the mass distribution
of the atmosphere. In Fig. 4a, b we show, for both groups of models, that the
SLP response displays a similar pattern of WC response to that obtained using
χ. In CPLD, the tropics show relatively minor changes in SLP, with an
increase over the Maritime Continent and western Pacific and a decrease over
the Indian Ocean associated with increased ascent and upper-level divergence
there (see Figs. 4a, 3a). This pattern represents a minor weakening of the
climatological zonal gradient in tropical SLP (not shown), consistent with
weakening of the WC in CPLD. By contrast, the PRES experiment shows an
increase in SLP across the Maritime Continent and into the western–central
tropical Pacific, with a smaller increase in the eastern Pacific. This
pattern represents a more pronounced weakening of the zonal gradient in
tropical SLP, consistent with a more substantial weakening of the WC in PRES
than in CPLD.
The patterns of χ and SLP response in the tropics are driven, to leading
order, by changes to tropical convection, which drives the mass circulation
and hence the WC e.g.,.
Since the tropical troposphere warms uniformly in response to climate change,
local changes in static stability must be controlled by changes in local
surface temperature . To investigate
the role of local surface warming we define the quantity dSST* as the
difference between the local SST response and the tropical
(20∘ S–20∘ N) mean SST response in each model for both
PRES and CPLD (Fig. 4a, b, shading). There is a very strong association
between the spatial pattern of dSST* and that of the SLP response (anomaly
correlation of -0.65 for the multi-model mean and a range of -0.56 to
-0.66 for the 10th–90th percentile of CPLD models), and also with 200 hPa
χ (anomaly correlation of -0.57 for the multi-model mean, with a range
of -0.43 to -0.6 for the 10th–90th percentile of CPLD models). The SLP
(χ) response tends to show high pressure (upper-tropospheric
convergence) over regions where dSST* is negative; that is, there is
large-scale descent where the local warming is less than the tropical mean
warming. This may explain why the WC weakens less in CPLD than in PRES
because the pattern of surface warming in CPLD produces a maximum located in
regions of climatological divergence and ascent, particularly in the eastern
Indian Ocean and around the Maritime Continent.
To further investigate the relation between the pattern of SST warming in
CPLD and the WC, a correlation between dSST* averaged over
90–150∘ E, 20∘ S–20∘ N and WC strength change
(using the method defined earlier in this section using 200 hPa χ) was
performed. A positive correlation of 0.61 was found among the models (i.e.,
higher dSST* tends to have less WC weakening). This shows that the rate of
SST warming in and around the Maritime Continent (eastern Indian Ocean and
western Pacific) is important for modulating the change in WC strength due to
climate change (i.e., faster warming may lead to less weakening or in fact
strengthening).
As Fig. 1, except showing the dSST* response (shading) and SLP
response (contours, interval 0.2 hPa K-1 with blue as negative, red
as positive, and black denoting the zero line). Stippling as in Fig. 2.
Response of the Hadley circulation
In CPLD the annual mean climatology of mass stream function (Ψ) consists
of two overturning cells on either side of the equator, with the southern
cell being slightly stronger (Fig. 5a). These cells describe the HC: ascent
occurs in the equatorial regions, with descent generally occurring in the
subtropics. The response in CPLD is largely negative and is centered just
north of
the equator but encompasses the ascending region of both cells. This implies
a weakening of the ascending branch in the NH and a
strengthening in the SH. This general pattern of HC
strength change is consistent with the expected changes due to future climate
change as simulated by the CMIP5 coupled models
e.g.,.
There is also strengthening apparent in the descending branch of the HC in
the NH, but model agreement is low.
CPLD (a, c, e) and PRES (b, d, f) meridional mass
stream function response (shading) and preindustrial control climatology
(contours, interval 2 × 1010 kg s-1, with dashed
lines as negative and the zero line thickened): (a, b) annual mean, (c, d) DJF mean, (e, f) JJA mean. Stippling as in Fig. 2.
The HC response in CPLD displays a distinct seasonal cycle because the
appearance of two distinct cells is an artifact of taking the annual mean. In fact, in
DJF and JJA there exists one dominant cell straddling the equator,
ascending in the summer hemisphere and descending in the winter hemisphere
(Fig. 5c, e). Henceforth, we refer to the seasonal cells with respect to the
hemisphere containing the descending limb. The weakening of the DJF (i.e.,
NH) cell is very evident in CPLD, with a tendency for greater weakening in
the ascending limb (Fig. 5c). One feature that was not apparent in the annual
mean is the increase in cell height, as evidenced by the positive Ψ
response around 150 hPa. This is associated with a well-documented increase
in tropopause height consistent with warming tropical SSTs, which through
moist adiabatic adjustment, leads to a higher tropopause
e.g.,.
There is a clear poleward shift of the descending branch of the DJF cell in
the mid-lower troposphere, which would imply a poleward shift of the
midlatitude eddy-driven jet.
The response of the JJA (i.e., SH) cell also shows weakening that is stronger
in the ascending than descending region (Fig. 5e). However, in contrast to
DJF, the weakening shows little model agreement (indicated by the lack of
stippling), and indeed the JJA cell actually strengthens in two-sixths of models
(Fig. 6). This increased uncertainty in the response of the southern HC has
been documented for future climate change
, but to our
knowledge this is the first time a similar result has been shown for the
mPWP. The increase in tropopause height is stronger in JJA than DJF, and
there is a small – but robust – expansion poleward of the descending branch
around 30∘ N that mirrors the result in the NH for DJF (Fig. 5c, e).
Given the apparent non-robustness of the HC weakening response, the
robustness of the poleward shift suggests that the strength change and
meridional expansion are controlled by different processes. Interestingly,
the spatial pattern of the weakening of the JJA cell, and its nonrobustness,
are very similar to the response seen among the CMIP5 models for future
climate change e.g.,. This has been linked to
the slower rate of tropical warming in the SH compared to the NH
e.g.,.
Overall, the response of the annual mean HC is quite different in PRES than
in CPLD. In PRES both cells weaken, with the largest weakening in the
ascending branch of the southern cell just south of the equator (Fig. 5b).
The weakening of the northern HC occurs in the center of the cell, indicating
an overall weakening of cell strength, as opposed to a weakening of the
ascending branch in CPLD. The southern HC, however, weakens mainly in the
ascending region, which is opposite to the strengthening seen there in CPLD.
There is a slight poleward expansion of the descending region of the northern
HC in PRES, which is more robust in terms of model agreement than in CPLD.
Similar amounts of HC weakening are found in DJF and JJA in PRES (Fig. 5d,
f). However, there are differences in the response around the edges of the
cell in each season. There is strong model agreement that in DJF the
ascending portion of the cell moves slightly poleward (Fig. 5d), while in JJA
both the ascending and descending portions of the cell expand poleward
(Fig. 5f). This robust weakening and expansion of the southern HC in the
ascending region is consistent with an expansion of the tropical wet zones
described in Sect. 3.1 (Fig. 2b) and with the single AGCM study of
. While this expansion of the HC is also somewhat
evident in CPLD, it is much weaker and only robust for the northern cell
(i.e., DJF) in the lower troposphere. An increase in tropopause height is not
evident in any season for PRES, which is likely related to the lack of
tropical warming in the imposed SSTs in PRES (Fig. 1b). Although PlioMIP
temperature data were not archived from model layers in the mid-upper
troposphere, we speculate that the warming in that region in PRES would be
substantially weaker than in CPLD because of the reduced latent heating
driven by muted tropical surface warming
.
Causes of seasonal asymmetry in HC change
Next we attempt to provide an explanation for why the largest non-robustness
in the HC response – and the largest differences between PRES and CPLD –
occur in JJA. Meridional gradients in SST and lower-tropospheric temperature
between the tropics and subtropics have been shown to affect HC strength in
both mPWP and future climates
e.g.,.
We further investigate this mechanism using the quantity dTs, which we define
as the change in the meridional surface temperature gradient between the
equatorial region (5∘ S–5∘ N) and the northern
(15–25∘ N) or southern (15–25∘ S) subtropics. We compute
dTs separately for the two winter hemispheres in each model, then compare the
values of dTs with a common measure of HC strength in each model: the
absolute change in the maximum (minimum) value of Ψ
(Δ|Ψ|max) for the Northern (Southern) Hemisphere e.g.,.
Figure 6 shows that, to leading order, the seasonal asymmetry in the
multi-model mean HC response in CPLD is explained by differences in dTs. The
relatively small change in HC strength during JJA
(Δ|Ψ|max of
-0.01 × 10-11 kg s-1 K-1 in the multi-model
mean) is accompanied by a small ΔTs in that season (0.03 K K-1),
compared to the larger values seen in DJF
(Δ|Ψ|max of
-0.13 × 10-11 kg s-1 K-1, 0.33 K K-1). In
other words, the stronger weakening of the NH cell than the SH cell in CPLD
could be explained by the subtropics warming faster than the tropics – and
the associated weakening of the tropical–subtropical surface temperature
gradient – in DJF, but not in JJA (Fig. 6). This is reminiscent of the
result from , who found a significant positive
linear relationship between Δ|Ψ|max and dTs among
a suite of 30 CMIP5 models. However, in our sample of six CPLD models from
PlioMIP, we find only a modest relationship (r=-0.46) between dTs and
Δ|Ψ|max, and this only emerges when DJF and JJA
data are pooled (not shown). The correlation values for DJF and JJA
separately are not significant and show the opposite relationship (r=0.39
and r=0.14, respectively).
CPLD and PRES multi-model mean DJF (blue) and JJA (red)
Δ|Ψ|max (left bar of pair) and ΔTs (right
bar of pair), with individual models indicated by letters (see Table 1 and
text for details).
The reason for the much weaker dTs in the Southern Hemisphere in CPLD is that
the southern tropical oceans warm more slowly than the northern tropical
oceans (shown in Fig. 4a for the annual mean). This explains the hemispheric
asymmetry in dTs, and this pattern is very similar for DJF and JJA (not
shown). Therefore, we propose that the seasonal asymmetry in HC response in
CPLD is, at least partially, explained by a corresponding asymmetry in the
tropical SST warming between the hemispheres. A further demonstration of this
idea comes from PRES, where dTs and Δ|Ψ|max are
very similar in both seasons and hemispheres (Figs. 4b, 6). We note that the
hemispheric asymmetry in surface warming found in CPLD, and the associated
changes in the HC, are similar to those projected for future climate change
from CMIP5 (e.g., He and Soden, 2015).
Discussion and conclusions
The tropical circulation response in two groups of models in PlioMIP show an
overall slowdown of the zonal and meridional components of the tropical
overturning circulation. However, we found that there are significant
differences in the Walker circulation (WC) and Hadley circulation (HC)
response between the two model groups. The ascending branch of the WC
weakens, but does so much less in CPLD than in PRES and less than reported
in CMIP5 models by. Across the two experiments
we find that the sign and magnitude of the WC response is strongly related to
pattern of SST warming, particularly around the Maritime Continent and
eastern Indian Ocean. We also find that the change in the winter hemispheric
meridional SST gradient between equatorial regions and the subtropics effects
the HC response. We find that the southern subtropics warm more slowly than
the northern subtropics in CPLD and there is thus an asymmetry in the HC
response between JJA and DJF.
The dependence of the WC response on the pattern of SST warming is contrary
to studies of future climate, where the weakening of the WC is found to be a
robust response to global warming, and is not as strongly related to the
spatial pattern of SST warming e.g.,. However,
there exists a two-way coupling between the WC and SSTs through the Bjerknes
feedback , making it possible that the models
in CPLD exhibit greater Indian Ocean warming precisely because the WC weakens
less. In PRES, where SSTs are held fixed, we see a significant weakening of
the WC compared to PI, which indicates that the atmospheric changes from
global warming weaken the WC. This provides further evidence that the pattern
of SST warming in CPLD is working against the overall tendency for WC
weakening driven by an increasing atmospheric stability. Only a new set of
experiments designed to separate the SST and WC changes, beyond the scope of
this study, can fully resolve this problem of cause and effect.
What we have found in PlioMIP for the HC is generally consistent with
projections of future climate. Thus, the PlioMIP models indicate that
the Pliocene tropical circulation could be similar to near-future (i.e., this
century) HC changes caused by anthropogenic climate change. The HC response
of both groups of models is similar to projections of future climate, with
CPLD having the most similar response to CMIP5. In fact, the spatial pattern
of the CPLD JJA HC response looks incredibly similar to that of the RCP 8.5
CMIP5 response in Fig. 1b in . As we have shown,
this similarity is likely explained by the similar meridional asymmetry in
tropical SST response in CPLD and CMIP5, with regions in the southern tropical
Pacific warming slower than those along and north of the equator. It would
seem that the meridional pattern of tropical SST response is a robust
response to increased GHGs since it is seen in both the PlioMIP coupled
simulations and in CMIP5.
The spatial pattern of tropical circulation response in PlioMIP shows that
regions with greater warming of the tropical SSTs tend to see increases in
precipitation and upper-level divergence, i.e., the “warmer-get-wetter”
pattern discussed in . Our results are also in
agreement with the single AGCM mid-Pliocene experiments of
using prescribed SSTs and
using an interactive ocean model, providing
support that the details are robust to model configuration. However, it is
striking that the pattern of tropical SST warming produced in CPLD is largely
inconsistent with the PRISM3 paleoclimate SST reconstructions used as
boundary conditions in PRES
(Fig. 1c). We have
shown that this has important implications for the tropical circulation, and
so this requires further urgent investigation if the mPWP is to be considered
as a reliable analogue for future climate.