Southern Hemisphere orbital forcing and its e ﬀ ects on CO 2 and tropical Paciﬁc climate

The western Paciﬁc warm pool (WPWP) is an important heat source for the atmospheric circulation and inﬂuences climate conditions worldwide. Understanding its sensitivity to past radiative perturbations may help better contextualize the magnitudes and patterns of current and projected tropical climate change. Here we present a new 5 Mg/Ca-based sea surface temperature (SST) reconstruction over the past 400 kyr from the Bismarck Sea, o ﬀ Papua New Guinea, along with results from a transient earth system model simulation. Our results document the primary inﬂuence of CO 2 forcing on glacial/interglacial WPWP SSTs and secondary e ﬀ ects due to changes in wind-driven tropical boundary currents. In addition to the SST, deep ocean temperature reconstruc- 10 tions from this core are linked with Southern Ocean temperature and sea-ice variations on timescales of ∼ 23 kyr. It is proposed that Southern Hemisphere insolation changes serve as pacemaker for sea-ice variations in the Southern Ocean, which in turn modulate windstress curl-driven upwelling of carbon-rich waters, hence controlling atmospheric CO 2 and tropical WPWP temperatures.


Introduction
Climate-model based projections of future climate change in the Pacific show an enhanced tropical warming response to increased greenhouse gas concentrations, in particular along equator (Xie et al., 2010). However, the simulated magnitude of this response over the next century is comparable to common tropical Sea Surface Tem-20 perature (SST) errors of state-of-the art climate models run under present-day conditions (Widlansky et al., 2012). This raises the issue of robustness of regional climate change projections in the tropical Pacific. Further independent evidence for an amplified sensitivity of equatorial SSTs to radiative perturbation changes could be obtained from past climate archives. Introduction The aim of our study is to identify the links between variability in Western Pacific Warm Pool (WPWP), where the current annual mean temperatures exceed 28 • C ( Fig. 1), radiative forcing changes and large-scale climate responses in extratropical regions on orbital timescales. Utilisation of tropical Pacific SST proxy records requires careful consideration of 5 potential seasonal biases of proxy data (Schneider et al., 2010), orbital effects (Timmermann et al., 2007) and of remote signals from the Northern Hemisphere icesheets (Timmermann et al., 2004). Extracting these effects to derive the greenhouse gas contribution in past SST records is challenging, but can be best accomplished by the analysis of long-term SST proxy datasets spanning at least several orbital cycles.

10
Hitherto only few such datasets are available to date (Fig. S1).
Here we set out to generate and analyse a ∼ 400 kyr-long high-resolution SST reconstruction using Mg/Ca ratios of the surface-dwelling planktonic foraminifera Globigerinoides ruber from a sediment core (MD05-2920) in the WPWP, near Papua New Guinea (Fig. 1). This location has a relatively high sedimentation rate of about 15 10 cm kyr −1 because of the enhanced input of river particles associated with precipitation over northern Papua New Guinea (Tachikawa et al., 2011), and contains exceptionally well-preserved foraminiferal tests (Fig. S2). We will also derive from the same core the deep ocean temperature history by using δ 18

O values of benthic foraminifera
Cibicidoides wuellerstorfi and Uvigerina peregrina. 20 To provide further insights into physical mechanisms of glacial variability we analyse output from a transient ∼ 400 kyr-long numerical modelling experiment (abbreviated as TR400) conducted with a coupled 3-dimensional ice-sheet climate model (gLOVE). The numerical experiments are free of the specific chronological uncertainties of the MD05-2920 core and will be used to cross-validate the paleo climate reconstructions. Using 25 TR400, we will further estimate the CO 2 -induced SST signal and explain the potential origin of residual SST variability in the WPWP record. The combined paleo-data and earth system model analysis will provide information on the mechanisms that may have caused the glacial cycle in atmospheric CO 2 and hence WPWP SST variability. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Regional setting
The studied core MD05-2920 (2 • 51.48 S, 144 • 51.48 E, water depth 1843 m, core length of 36 m, Fig. 1a) was collected during the MD148 IMAGES XIII PECTEN cruise and is located in an area with weak present-day seasonal SST variability (28.5 to 29.3 • C in the upper 50 m, Fig. 1b) (Locarnini et al., 2010). To document the relation 5 between climatological SST variability and regional and large-scale ocean currents we show in Fig. 2a, b the observed climatological January and July SST (shading) using AVHRR SST data in 9 km resolution (Vazquez-Cuervo et al., 1998) along with simulated 10 yr climatology of surface zonal currents (contours) from a 50 yr hindcast simulation conducted with the NCEP-wind forced eddy-resolving OfES/MOM3 model 10 (Masumoto et al., 2004). The annual cycle in SST is related to meridional shifts of the Intertropical Convergence Zone (ITCZ) (Tachikawa et al., 2011) and associated changes of Sverdrup transport and the strength of the New Guinea Coastal Current (NGCC), a western boundary current branch of the South Equatorial Current (SEC) (Fig. 2a, b). During austral summer (January, Fig. 2a) the northwestern winds push 15 the ITCZ to its southern position and wind-driven NGCC flows eastward, transporting cooler water to the core site by advection and a subtle coastal upwelling along the New Guinea (Kuroda, 2000). During austral winter (July, Fig. 2b), the southeastern winds shift the ITCZ northward and the NGCC flows westward along the coast of Papua New Guinea. The Island of New Britain serves as a topographic barrier and prevents intru-20 sion of cooler water to the Bismarck Sea. On inter-annual timescales, the El Niño Southern Oscillation shows a moderate influence on SST (Fig. S3a). Sea surface salinity in the upper 50 m ranges between 34.2 and 34.8 practical salinity units (psu), with a mean annual value of 34.5 psu (Fig. S3b). The studied area is oligotrophic with a mean chlorophyll a concentration of 0.29 mg m Introduction  et al., 2006). Currently, the bottom water at the core location is made up of a mixture of upper Circumpolar Deep Water (Kawabe and Fujio, 2010;Sokolov and Rintoul, 2000;Tsimplis et al., 1998) formed at southern high latitudes and old Pacific Deep Water, which is 5 linked to Antarctic Bottom Water. As Antarctic Bottom Water moves northward, it gains buoyancy by geothermal heating (Emile-Geay and Madec, 2009). Further density transformations occur as a result of topographic mixing, leading to the formation of southward flowing Pacific Deep Water.

Samples and age model
The WPWP core MD05-2920 ( Fig. 1) is composed of non-laminated greyish olive clay with dispersed bioclasts and foraminiferal tests. The two major components of the sediments are represented by a terrigenous fraction and calcium carbonates (Tachikawa et al., 2011). Individual tests of the surface-dwelling planktonic foraminifera G. ruber 15 (white), as well as the benthic foraminifera C. wuellerstorfi and U. peregrine, were picked at every 4 to 10 cm interval from the 250-355 µm size-fraction. The two G. ruber morphotypes are considered to represent slightly different calcification depths: G. ruber sensu stricto (s.s.) is calcifying in the uppermost 30 m of the water column, while G. ruber sensu lato (s.l.) lives at depths below 30 m (Wang, 2000). We compared the Introduction  (Lisiecki and Raymo, 2005). The deglacial change in benthic 5 foraminiferal δ 18 O of core MD05-2920 might slightly lag the continental ice volume because of diachronous deepwater temperature variations (Lisiecki and Raymo, 2009;Skinner and Shackleton, 2005) and a delayed deep ocean propagation of the glacioeustatic δ 18 O signal during glacial terminations (Friedrich and Timmermann, 2012).
For the given age model, the sedimentation rate varied from 6 cm kyr −1 to 10 13 cm kyr −1 , with a mean value of 10 cm kyr −1 (Tachikawa et al., 2011). The mean temporal resolution of the Mg/Ca analyses is about 450 yr over the last 140 kyr. For the older part of the core, the mean resolution is 1100 yr, but with a better resolution of about 600 yr around the terminations. 15 The benthic foraminiferal δ 18 O measurements were carried out at CEREGE using a mass spectrometer (Finnigan Delta Advantage) equipped with a carbonate preparation device. The measured isotopic values are normalized against NBS19. Mean external reproducibility is better than 0.05 ‰.

Isotopic and Mg/Ca analyses
Prior to the multi-step cleaning for Mg/Ca analysis, thirty tests of G. ruber (250-20 355 µm) were weighed to assess the potential influence of test dissolution on Mg/Ca ratios (Fig. S2). All the weighed tests were gently crushed and subjected to the Mg cleaning procedure (Barker et al., 2003). After dissolution of tests, the solutions were analysed by an ICP-OES instrument (Jobin Yvon Ultima C). The accuracy of our measurements was checked by international calibration (Greaves et al., 2008;Rosenthal 25 et al., 2004). The mean external reproducibility and accuracy of Mg/Ca are better than 0.5 %. Mg/Ca is converted into SST using Mg/Ca (mmol mol −1 ) = 0.  (Anand et al., 2003). The overall uncertainty of the temperature estimate is ± 1 • C based on the scatter of calibrations (Anand et al., 2003).
To assess the potential bias of Mg/Ca-SST in relation to partial test dissolution, individual foraminiferal test weight and test loss during cleaning were quantified. The test weight varies from 8.2 to 14.5 µg for G. ruber, with a mean value of 5 11.5 ± 1.1 µg individual −1 (1σ, N = 562, Fig. S2). Even though a subtle decreasing trend is observed for the older part of the core, the correlation between G. ruber test weight and age is weak (R 2 = 0.26, N = 562), and there is no systematic change on glacial/interglacial timescales (Fig. S2). It has been already shown that the moderate dissolution does not significantly modify foraminiferal Mg/Ca (Tachikawa et al., 2008).

10
The proportion of test loss during cleaning is estimated at 18 % to 75 %, with a mean value of 34 ± 8 % (1σ, N = 562). For the whole core, the loss is clearly smaller than the critical value of 80 % (Tachikawa et al., 2008). The present calcite saturation depth in the Western Equatorial Pacific is approximately 2500 m (Feely et al., 2012), deeper than the water depth of the core location (1843 m).

Transient climate model experiment
To further study the effects of time-varying greenhouse gas concentrations and orbitally-induced insolation changes on tropical Pacific and Southern Ocean climate, we conducted a transient model simulation with a 3-dimensional coupled ice sheet climate model (hereafter referred to as gLOVE) that covers the forcing history of the past 20 406 ka. The model has been developed from the previous version of LOVECLIM(V1.1) (Goosse et al., 2010) and the community ice sheet model GLIMMER(V1.7.1) (Rutt et al., 2009). The spectral atmospheric component uses T21 horizontal resolution (approx. grid resolution 5.625 • × 5.625 • ) and 3 vertical layers. Atmospheric dynamics are governed by the quasi-geostrophic potential vorticity equation and ageostrophic forcing terms that are diagnosed from the vertical motion field (Opsteegh et al., 1998) The sea ice-ocean component "Coupled Large-scale Ice-Ocean" (CLIO) (Goosse and Fichefet, 1999) consists of a primitive equation ocean general circulation model with 3 • × 3 • resolution on a partly rotated grid in the North Atlantic. Vertical resolution consists of 20 unevenly spaced levels. Mixing along isopycnals, the effect of mesoscale eddies on diapycnal transports and mixing (Gent and McWilliams, 1990), 5 and downsloping currents at the bottom of continental shelves are parameterized. The terrestrial vegetation model "VEgetation COntinuous DEscription" (VECODE) (Brovkin et al., 1997) consists of two plant functional types and non-vegetated deserts zones. Each grid cell assumes a partial coverage by these three land cover types depending on the annual mean temperature and rainfall amount and variability.

10
The land-ice model GLIMMER is a three-dimensional thermomechanical ice-sheet model for grounded ice. The ice dynamics are simulated using the shallow-ice approximation (Rutt et al., 2009). In the current configuration we simulate the entire Northern Hemisphere domain (approx. 30 • N-90 • N) in a polar stereographic projection with 100 × 100 km grid resolution. Eleven sigma-levels are used to represent the ice sheet 15 dynamics and thermodynamics. At the bottom interface between bedrock and ice noslip conditions are applied. In this simulation we adopted a daily positive-degree-day (PDD) scheme using daily-mean climatological temperature and precipitation data. We developed an asynchronous Northern Hemisphere coupling strategy between LOVECLIM and GLIMMER similar to the one applied in CLIMBER (Ganopolski 20 et al., 2010). To account for the longer timescales of ice sheet dynamics relative to atmosphere-ocean-sea ice-vegetation dynamics, the coupling intervals for the ice sheet model are typically longer (1000 model years) than the coupling intervals for LOVECLIM (50 model years). Each time a 50 yr block of LOVECLIM integration is finished, the updated climatologies of precipitation, surface temperature are passed to 25 GLIMMER. In GLIMMER these surface temperatures are used to calculate PDDs after horizontal interpolation to the finer ice-sheet grid and vertical interpolation using a lapse rate 0.5 K 1000 m −1 . The surface mass balance is obtained from PDD and the simulated LOVECLIM precipitation field. After completion of a 1000 yr-long GLIMMER simulation segment the resulting ice sheet height and extent is used to update the orography and surface albedo in LOVECLIM for the subsequent 50 yr-long model simulation. Slowly time-varying boundary conditions due to astronomical and greenhouse gas forcing are implemented as accelerated forcings into LOVECLIM.
Choosing an acceleration factor of for instance 20, 50 yr of LOVECLIM simulation 5 are subject to the external forcing history of a 1000 yr interval. With this method, only LOVECLIM experiences accelerated orbital and greenhouse gas forcing, whereas the ice-sheet model GLIMMER experiences unaccelerated forcings. A simulation covering the forcing history of the last 406 ka will then result in 20 300 LOVECLIM model years and 406 000 GLIMMER model years. More details on the advantages and disadvan-10 tages of the orbital acceleration technique can be found in (Timm and Timmermann, 2007;Timm et al., 2008). The initialization of the transient coupled simulation starts in interglacial conditions at 406 000 BP. Without proper knowledge of the exact ice sheet conditions at this time, we chose the initial conditions of the Eemian period at 125 000 BP, which was obtained 15 from previous uncoupled interglacial simulations with LOVECLIM and GLIMMER. The gLOVE present-day climate sensitivity amounts to 3.2 K for a CO 2 doubling.
The overall performance of gLOVE in terms of simulating the Northern Hemisphere ice volume over the last 406 ka is illustrated in Fig. S4. The simulated ice volume correlates well with the reconstructed sea level evolution (Siddall et al., 2003;Waelbroeck 20 et al., 2002). However, ice-volume changes during Marine Isotope Stage (MIS) 6, 8 and 10 are underestimated and the last glacial period does not terminate completely. Note, that no major attempts were made to tune the model towards the evolution of past ice-sheet variations. The effect of Northern Hemisphere ice-sheets on Southern Hemisphere climate is quite limited in LOVECLIM and hence we expect biases of North-25 ern Hemisphere ice-sheet evolution to play only a small role for our assessment of Southern Hemisphere climate change and its response to CO 2 variations and Southern Hemisphere orbital forcing. This will be demonstrated further below by comparing the model simulation with Southern Hemisphere proxy data.

WPWP SST reconstructions
Over the past 400 kyr, Mg/Ca values in G. ruber varied between 3.43 and 5.68 mmol mol −1 (Fig. S2), which corresponds to an SST range between 24.4 and 30.1 • C, according to a commonly-used calibration (Anand et al., 2003) (Fig. 3). The 5 core-top value of 28.7 • C agrees well with the present-day annual mean temperature of 28.9 • C ( Figs. 1 and 3). The glacial/interglacial SST amplitude attains values of up to 3 • C during Terminations I and III, and 4 • C for Terminations II and IV (Fig. 3).
Our MD05-2920 Mg/Ca record exhibits very pronounced glacial/interglacial variability over the past 400 kyr, with characteristic orbital periodicities of 100 kyr (eccen-10 tricity), 41 kyr (obliquity) and 23 kyr (precession) (see Fig. S5). It shares the general glacial/interglacial features with other records from the central WPWP but with more detailed variability for Terminations III and IV (Fig. S1).
Comparison between the MD05-2920 SST reconstruction and the simulated SSTs in experiment TR400 at the core location reveals a high degree of correspondence 15 (Fig. 3) and a maximum cross-correlation of 0.78 at lag 0 kyr. The observed in-phase relationship on orbital timescales also supports the age model of MD05-2920, which was derived from a benthic foraminiferal δ 18 O stack. However, we find considerably smaller amplitudes of glacial/interglacial SST change in the model solution of only ∼ 1.7 • C (Last Glacial Maximum, LGM, to present day 19-23 kyr), compared to the 20 reconstructed range of 3.4 ± 0.5 • C (1σ, N = 31). It is worth putting these estimates into the context of the LGM multi-model uncertainties. Using output from the Paleomodel Intercomparison Project 2 (PMIP2) multi-model ensemble (Braconnot et al., 2007) (see Fig. 3, colored bars) we find glacial/interglacial SST differences in the WPWP ranging from 1.7 • C (IAP-FGOALS model) to 4.0 • C (GFDL CM2.1 model). This discrep- 25 ancy is reminiscent of the fact that the warm pool heat budget is delicately balanced by rather large individual contributions and that uncertainties, for instance in cloud or Introduction  (Monnin et al., 2001;Pepin et al., 2001;Petit et al., 1999;Raynaud et al., 2005), which suggests that greenhouse gas forcing is 5 one of the primary drivers for the reconstructed and observed temperature variability (Fig. 4). To identify any local/seasonal effects, we subtracted the CO 2 radiative forcing effect (proportional to the logarithm of CO 2 concentrations) from the Mg/Ca-SST record using the regressions shown in Fig. 4. Comparing the resulting residual Mg/Ca-SST variability with the simulated strength of the wind-driven surface tropical boundary 10 current off Papua New Guinea in TR400 (Fig. 5a), we find a high correlation (0.88) at precessional timescales of 23 kyr. Wind variability in the western tropical Pacific is to a large-extend determined by the meridional movements of the ITCZ. At present, when the ITCZ is at its southernmost position in austral summer, the SST at the core location decreases because the weaker northward boundary current allows cooler sur- 15 face water to enter the study site (Fig. 2a). Using this present-day analogy we suggest that the residual SST precessional cycle is modulated by past changes in ITCZ position (Fig. 5a). Independent support for this idea is provided by the good agreement between the reconstructed river particle inputs, derived from Titanium to Calcium ratio in core MD05-2920 that is an indicator of Sepik River particle discharge (Tachikawa 20 et al., 2011), and simulated rainfall over Papua New Guinea in TR400 (Fig. 5b).
To rule out that seasonal biases (Schneider et al., 2010) affect the 23 kyr variability in residual SSTs we calculate the annual mean of the product between simulated seasonal SST variations in TR400 and the present-day climatology of euphotic zone chlorophyll concentrations (Fig. S3c), scaled by the annual mean chlorophyll concen-25 tration at the site and then subtract the simulated annual mean SST. The resulting annual mean chlorophyll-weighted SST residual is anti-correlated with the reconstructed residual SST (Fig. S6), thus suggesting that seasonal biases are not the primary reason for the precessional-scale variability in reconstructed SSTs. With CO 2 as one of the primary drivers of climate change in both the WPWP and the Southern Hemisphere (Clark et al., 2012a;Kohler et al., 2010;Timmermann et al., 2009), it is also understandable that a high correlation (significant above the 95 % level) exists between the WPWP Mg/Ca SST and the deuterium-based temperature reconstructions from Antarctica (Jouzel et al., 2007) (slope 3, R 2 = 0.7, Fig. S7). Al-5 though both regions respond locally to the radiative forcings, they differ in their respective feedback mechanisms that amplify this response. Whereas the water-vapor feedback (Manabe and Stouffer, 1980) is the key amplifier for WPWP surface temperature changes, low continental heat capacity and the lack of evaporation on land are key elements in amplifying the Antarctic land temperature response to CO 2 changes.

Linking WPWP deepwater temperature history with Southern Hemisphere processes
The deglacial range of benthic foraminiferal δ 18 O records of core MD05-2920 amounts to about 1.3 to 1.6 ‰ (Fig. 3). We also observe that surface conditions in the WPWP (Mg/Ca-SST) and deep ocean conditions (benthic δ 18 O) varied in unison over the last 15 400 kyr (Table S1), including the major glacial terminations (Fig. 3) Considering that deep waters at the WPWP core site are a mixture of different water masses including a strong southern ocean component (see Sect. 2), we further compare the estimated deep temperature signal with an independent bottom water temperature reconstruction from benthic foraminiferal Mg/Ca on Chatham Rise in the Southern Ocean (Elderfield et al., 2012). Since the benthic Mg/Ca record is free from 5 the local salinity effect (Elderfield et al., 2012), the common feature of the two records reflects southern ocean temperature variability on orbital and sub-orbital timescales (Fig. 6d). Interestingly, both deep water temperature estimates have a strong precessional component (Fig. 6d) which correlates well with the sea-salt flux-based sea-ice reconstructions (Wolff et al., 2006) from the EPICA ice core and the simulated sea-ice 10 area in TR400 (Fig. 6b, see Table S2 for results of cross spectral analysis). Moreover, deep-ocean temperature records are highly coherent with fixed season austral spring insolation (Stott et al., 2007;Timmermann et al., 2009)  The sea-ice response to the shortwave radiation changes will affect the strength of the atmosphere-ocean momentum transfer. For instance, even for constant westerly 20 winds, an increase of austral spring insolation will reduce the Southern Ocean sea-ice area in TR400 and its compactness thus increasing vertical ocean velocities due to changes in Ekman pumping.
Ekman pumping velocities for the open ocean are usually calculated as where ρ represents the water density, τ AO x,y the zonal and 25 meridional stress components between atmosphere and ocean, f the Coriolis parameter and ∂ x,y the partial derivatives in x and y direction. The atmosphere-ocean stress vector is denoted here as τ AO = (τ AO x , τ AO x ). In partially sea-ice covered areas, however, with a sea-ice fraction of α, such as the Southern Ocean, one has to take into account also the stress between sea-ice and ocean, which can be calculated from τ IO = C IO ρ|u i − u o |(u i − u o ), with C IO = 0.0055 representing a drag coefficient and u i , u o the ice and ocean velocities, respectively. Here we calculate the Ekman pumping below partially covered sea-ice areas from the output provided by gLOVE using w We define an Ekman pump- 5 ing efficacy E e as E e = w N e /w AO e which characterizes the ratio of the wind-stress curl that is available for Ekman pumping in the ocean in sea-ice covered regions.
The Ekman pumping efficacy is smaller than 1 (E e < 1) in regions of sea-ice convergence or divergence, such as the Ross and Weddell Seas (Goosse and Fichefet, 1999). As already proposed in Goosse and Fichefet (1999), extended Southern Hemisphere sea-ice coverage under glacial conditions may lead to a sea-ice-induced damping of the Southern Hemisphere upwelling, even when the mean winds remain the same. Here we pursue this idea further to identify potential causes for orbital-scale variations in ocean ventilation and CO 2 . Under present-day conditions most of the carbon-rich waters upwell south of the Antarctic Polar Front. To study the effect of sea-ice varia- 15 tions on Ekman suction in the Southern Ocean in TR400, we calculate E e at 60 • S in each of the 50 yr segments of ocean model output, that represent 1000 yr of orbital forcing history. As shown in Fig. 6c, over the last 350 kyr the efficacy of Ekman pumping calculated from TR400 varied by a factor of 2.5, in unison with austral spring insolation changes, 20 simulated sea-ice coverage and EPICA sea-salt fluxes (Wolff et al., 2006). Phases of high (low) eccentricity and high (low) amplitude precessional cycles in austral spring insolation correspond to interglacial (glacial) conditions and high (low) atmospheric CO 2 , thus suggesting the presence of a nonlinear climate-carbon cycle rectification mechanism in the Southern Hemisphere. Here we propose that the orbital-modulation 25 of sea-ice coverage and related Southern Ocean Ekman pumping control the upwelling of carbon-rich waters and in turn the degassing of CO 2 (Fig. 6c) into the atmosphere. This process may be further amplified by the sea-ice control on air-sea gas exchange Introduction  , 2000). Resulting CO 2 variations feed back onto the sea-ice coverage and the Ekman pumping efficacy, thus providing a positive feedback.

Discussion
Our proposed mechanism of Southern Hemisphere orbital control of sea ice, Ekman pumping efficacy and CO 2 release can account for the apparent orbital pacemaking of 5 glacial CO 2 changes over at least the past 400 ka. In contrast to other Southern Hemisphere CO 2 mechanisms which oftentimes invoke remote physical connections from the Northern Hemisphere (Anderson et al., 2009;Gildor et al., 2002) (Fig. 7) such as the bipolar seesaw and latitudinal shifts of the Southern Hemisphere Westerlies (Denton et al., 2010;Clark et al., 2012b), our 10 mechanism relies primarily on regional Southern Hemisphere forcing (Fig. 7). Furthermore, it may provide a simple framework to explain the synchronization of northern and Southern Hemisphere climate change during the Late Pleistocene: Changes in austral spring insolation or length of austral summer, which affect Southern Hemisphere seaice and CO 2 , are highly correlated with boreal summer insolation anomalies (Huybers 15 and Denton, 2008) that cause variations in Northern Hemisphere ice-volume (Fig. 7). CO 2 changes induced by Southern Hemisphere processes then contribute partly to the waxing and waning of the Laurentide and Eurasian ice-sheets via direct forcing of positive-degree-days or through changes in atmospheric moisture and heat supply from the tropical Ocean (Rodgers et al., 2003) (Fig. 7). The CO 2 anomalies fur-20 ther modulate the sea-ice and climate response in the Southern Hemisphere (Figs. 6 and 7). Note in particular that the Ekman pumping efficacy mechanism does not depend on the controversial hypothesized shifts of Southern Hemisphere Westerlies (see Anderson et al., 2009;Denton et al., 2010;Menviel et al., 2008Menviel et al., , 2011Sime et al., 2013 for a discussion of model results and proxy data). Introduction The proposed mechanism works only in cold climate states such as the Middle to Late Pleistocene, when enough Southern Ocean sea ice is available year-round with large-enough seasonality. In warmer states, such as the Pliocene we hence expect 15 a much weaker orbital modulation of CO 2 and thus also a weaker CO 2 contribution to Northern Hemisphere ice-sheet dynamics on eccentricity timescales.

Conclusions
This study combined evidences from high-resolution WPWP SST, benthic δ 18 O records and a transient coupled ice-sheet climate model simulation to elucidate the links be-20 tween high latitude Southern Hemisphere processes and tropical climate responses over the past ∼ 400 kyr.
We found that the tight relationship between the WPWP SST and atmospheric CO 2 identified in both paleo data and model simulation points to a dominant influence of CO 2 radiative forcing in Western Pacific climate variations. In accordance with paleo-Introduction Both records exhibit precessional variability, which is explained in terms of orbitallymodulated Southern Hemisphere sea ice variations and the presence of an oceanic tunnel which links Southern Ocean surface temperature changes to the deep Pacific Ocean. We argued that precessional-scale sea-ice variations in the Southern Ocean were caused by changes in austral spring insolation. Evidence was presented, that 10 further documents the influence of orbital-scale variations of sea ice extent on the efficacy of Ekman pumping in the Southern Ocean. Glacial/interglacial variations can result in a modulation of the Ekman pumping efficacy by a factor of 2-3, even in the presence of only weakly changing westerly winds. We proposed that these variations in Ekman pumping might have played a key role in the upwelling of carbon-rich waters, 15 thus controlling atmospheric CO 2 concentrations and in turn tropical climate.
Acknowledgements. 20 Competing interests statement: The authors declare that they have no competing financial interests.