Rapid Coupling of Antarctic Temperature and Co 2 Printer-friendly Version Interactive Discussion Climate of the past Discussions Rapid Coupling of Antarctic Temperature and Atmospheric Co 2 during Deglaciation Rapid Coupling of Antarctic Temperature and Co 2 Printer-friendly Version Interactive Disc

This discussion paper is/has been under review for the journal Climate of the Past (CP). Please refer to the corresponding final paper in CP if available. Abstract Antarctic ice cores provide clear evidence of a close coupling between variations in Antarctic temperature and the atmospheric concentration of CO 2 during the glacial/interglacial cycles of the past 800 thousand years. Precise information on the relative timing of the temperature and CO 2 changes can assist in refining our under-5 standing of the physical processes involved in this coupling. Here, we focus on the last deglaciation, 19 000 to 11 000 years before present, during which CO 2 concentrations increased by ∼80 parts per million by volume and Antarctic temperature increased by ∼10 • C. Utilising a recently developed proxy for regional Antarctic temperature, derived from five near-coastal ice cores, and two ice core CO 2 records with high dating pre-10 cision, we show that the increase in CO 2 lagged the increase in regional Antarctic temperature by only 0–400 years. This new value for the lag, consistent for both CO 2 records, implies a faster feedback between temperature and CO 2 than the centennial to millennial-scale lags suggested by previous studies.


Introduction
The last deglaciation is the largest naturally-forced global climate change in Earth's recent climate history.Its initial impetus, as with the sequence of glacial/interglacial transitions that came before it, is most commonly attributed to orbitally induced variations in insolation at high northern latitudes (Hays et al., 1976).Amplification through climate feedback processes of the relatively weak orbital signal is then required to explain the full magnitude of the glacial-interglacial climate change (Lorius et al., 1990).Changes in the carbon cycle play a central role among these feedbacks (Lorius et al., 1990;Shackleton, 2000).Mounting evidence attributes a large component of the deglacial CO 2 increase to release of old CO 2 from the deep southern ocean through changes in its biogeochemistry and physical circulation (Anderson et al., 2009;Sigman et al., 2010;Skinner et al., 2010) exact mechanisms involved and their relative contributions (Fischer et al., 2010;Sigman et al., 2010;Toggweiler and Lea, 2010).Determining the timing of the increase in atmospheric CO 2 with respect to the increase in temperature (hereafter "the lag"), is crucial for the models seeking to discriminate between mechanisms (Schmittner and Galbraith, 2008;Ganopolski and Roche, 2009;Lee et al., 2011).
Antarctic ice cores are unique in preserving a record of both temperature variation and atmospheric CO 2 concentration.Water stable isotope ratios (δ 18 O ice and δD ice ) from the ice are proxies for temperature above the inversion layer at the time of snow formation (Jouzel et al., 1997), while the CO 2 is preserved in air bubbles in the ice.
The transformation of snow to glacial ice isolates these bubbles at a depth of 50-100 m, leaving them younger than the surrounding ice by an amount ∆ age.The ∆ age must therefore be known for the lag between temperature and CO 2 to be accurately determined.
The most commonly cited studies constraining the lag for the last deglaciation have used ice core records from the East Antarctic Plateau (EAP) and report lags of 400-1000 years from Vostok (Fischer et al., 1999) and 200-1400 years from EPICA Dome C (Monnin et al., 2001).Studies of previous deglaciations also suggest millennial-scale lags (Fischer et al., 1999;Mudelsee, 2001;Siegenthaler et al., 2005;Caillon et al., 2003).However, the low accumulation rates at these sites (2-3 gcm −2 yr −1 ) lead to lengthy intervals between deposition and bubble close off, and consequent high values for ∆ age: 2400-3300 years for recent times and 4500-5200 years at the last glacial maximum (Fischer et al., 1999;Monnin et al., 2001).These large values are associated with uncertainties of typically 10-20 %, which are comparable to the reported lag values (Loulergue et al., 2007).
The contribution to the lag uncertainty from ∆ age can be minimised by considering ice core records from higher accumulation sites.Of the currently available CO 2 records for the deglaciation, those with the lowest ∆ ages are the Siple Dome (Ahn et al., 2004) and Byrd (Neftel et al., 1988;Staffelbach et al., 1991) ice cores.The ∆ ages at these sites are 200-300 years for recent times and 500-800 years at the last glacial Introduction

Conclusions References
Tables Figures

Back Close
Full maximum, an order of magnitude smaller than records from the EAP.Analysis of the Siple Dome stable isotope and CO 2 records suggests lags in the 150-400 year range (Ahn et al., 2004), but this result suffers from the fact that the Siple Dome deglacial isotope record contains abrupt changes not observed in other Antarctic records (Taylor et al., 2004;Brook et al., 2005), suggesting a local climate signal that would not be expected to correlate with CO 2 evolution.
Our approach is to compare the Siple Dome and Byrd CO 2 records to an index of regional Antarctic temperature, T proxy , derived (Pedro et al., 2011) from a composite of high-resolution stable isotope records from the Law Dome, Siple Dome, Byrd, EPICA Dronning Maud Land and Talos Dome ice cores (Fig. 1a, inset shows core locations).
Since T proxy contains ice cores representing the Indian, Atlantic, and Pacific coastal Antarctic regions, it is expected to provide a better representation of high-latitude Southern Ocean processes than records from single sites.This approach requires that T proxy and the two CO 2 records share a common timescale.This is achieved by synchronising all records to the Greenland Ice Core Chronology 2005 (GICC05) using the rapid and effectively globally synchronous variations in CH 4 concentrations found in both Antarctic and Greenland ice cores.

Record synchronisation
The T proxy record is already available on GICC05 (Pedro et al., 2011).We synchronise the timescales of the Siple Dome and Byrd CO 2 records to GICC05 using previously published "gas-age" depth ties from CH 4 -based synchronisations of the Byrd (Blunier and Brook, 2001) and Siple Dome (Brook et al., 2005) records with Greenland records.The Byrd CO 2 data on the GRIP SS09 timescale (Johnsen et al., 1997) are transferred via GRIP depth (Blunier and Brook, 2001) to GICC05 ages by linear interpolation using stratigraphical markers (Rasmussen et al., 2008).Similarly, the Siple Dome CO 2 Figures data, on a GISP2 timescale (Ahn et al., 2004) are transferred via GISP2 depth (Meese et al., 1997) to GICC05 ages again by linear interpolation using stratigraphical markers (Rasmussen et al., 2008).

Lag calculation
We determine lag quantitatively, following previous studies (Fischer et al., 1999;Mudelsee, 2001;Siegenthaler et al., 2005;Ahn et al., 2004), by maximising the timelagged correlation between the deglacial temperature and CO 2 curves.Two methods are used: first by direct correlation between T proxy and CO 2 (direct method) and second, as used in a prior Siple Dome study (Ahn et al., 2004), by correlation between the corresponding derivative curves, ∂T proxy /∂t and ∂CO 2 /∂t (derivative method).The derivative method has smaller sensitivity to misidentification of the pre-and post-transition levels at the expense of increased sensitivity to measurement noise, especially in the sections of sparse data.Since it is not clear a priori which approach is superior, we apply both methods in parallel.Prior to the analyses, the CO 2 data are first linearly interpolated to 20-year resolution over the 9-21 ka interval to match the 20-year resolution of T proxy .Derivatives, ∂T proxy /∂t and ∂CO 2 /∂t, are calculated on the same grid.
For both methods the data are then smoothed and the time-lagged correlation determined for lags in the −400 to +1000 year range.The robustness of the lag calculation to CO 2 data uncertainty is evaluated by repeating the lag determination 100 times using random realisations of the CO 2 measurements, each sampled using the published uncertainties (represented by the error bars in Fig. 1a).
As the lag shows some sensitivity to the degree of smoothing applied and the overall time interval chosen for assessment we repeated the analysis using 19 different degrees of smoothing and 35 different choices of time interval.The different degrees of smoothing were applied by convolution with a Gaussian filter with a standard deviation ranging from 105-375 years in steps of 15 years; the minimum width of the smoothing corresponds to the approximate sampling rate of CO 2 data in the sparsely sampled parts of the data set, and the maximum was chosen to preserve a clear Antarctic Cold Introduction

Conclusions References
Tables Figures

Back Close
Full Reversal signature.Similar results were obtained by replacing the Gaussian filter with a simple running mean filter of equivalent half-width.The choices of the time interval start and end points for the lag calculation were varied within reasonable limits that encompass the overall deglaciation; specifically, the start and end ages were varied in 200 year steps with the start spanning 19.4-18.2ka (7 options) and end spanning 12.4-11.6ka (5 options), as illustrated by the black triangles of Fig. 1a.This covers the start and end of the deglaciation as defined from the perspective of the onset and end of the warming trend in T proxy (Pedro et al., 2011); note that younger than around 11.5 ka both CO 2 curves have high early Holocene values that have no counterpart in the T proxy record.This suggests different processes are responsible and we exclude this period from the correlation interval.The full sensitivity analysis generates 66 500 optimal lag values for each of the direct and derivative methods.

Results and discussion
Close correspondence between T proxy and both CO 2 records is observed (Fig. 1a), supporting the hypothesis that marine processes at high southern latitudes are linked to the deglacial CO 2 increase.Visual inspection of the relative timing of the deglacial CO 2 and T proxy increase suggests little or no lag of CO 2 after temperature.Quantitative results on the relative timing are summarised by the four histograms in Fig. 2b.The widths of the individual distributions reflect the sensitivity of the lag values to CO 2 measurement uncertainties, lag determination method (direct or derivative), CO 2 measurement uncertainties, differences between the two CO 2 data sets, smoothing filter widths and the start and end points used in the analysis, as determined by Monte Carlo-style sensitivity analysis.The distributions show larger spread for the derivative method, but overall consistency between the two sites and for the two methods.Using the means and standard deviations quoted in Fig. 2b We must also consider relative dating uncertainty between T proxy and the two CO 2 records.This uncertainty comprises a component associated with synchronising CH 4 records between multiple cores, σ sync , and a component associated with the ∆ age uncertainties of the individual cores, σ ∆age .The dating uncertainty in T proxy ,σ T proxy , ranges from 199-384 years over the time interval considered, with a mean of 269 years (Pedro et al., 2011).The value of σ lag is smaller than σ T proxy because σ sync applies to both T proxy and the CO 2 data and thus partially cancels (as seen in the case where data from only one core is used and σ sync cancels completely).The similarity of the results from the independent lag calculation using Siple Dome and Byrd CO 2 data supports that the records are well-synchronised on centennial scales or better.We estimate the contribution to σ lag from σ ∆age and σ sync to be no more than 200 years.This relative dating uncertainty is independent of the abovementioned 34-year methodological uncertainty.Combining the two figures, we arrive at our final estimate that the deglacial CO 2 increase lagged Antarctic temperature by only 0-400 years.
This result is substantially lower and more tightly constrained than the previous Vostok and EPICA Dome C estimates of 200-1400 years and 400-1000 years, albeit consistent within their reported uncertainties.This short lag is supported by more recent empirical constraints indicating that EPICA Dome C ∆ age (and thus the CO 2temperature lag) was significantly overestimated during the last deglaciation (Loulergue et al., 2007).It is important to emphasise that this lag represents an average for the entire deglaciation.Therefore, in a coupled system where multiple processes are likely to be involved in the CO 2 increase (Fischer et al., 2010;Sigman et al., 2010), this value reflects the contributions of all processes weighted by their contribution to the CO 2 increase.This result shows that the effective combination of these mechanisms operated with little or no lag.Additional observational constraints on the mechanisms are obtained by considering in more detail the millennial and sub-millennial trends in CO 2 throughout the deglaciation.The increase in CO 2 , as noted previously (Fischer et al., 1999;Monnin et al., 2001;Ahn et al., 2004), occurs in two main steps.These steps coincide with Figures

Back Close
Full the two periods of significant warming in T proxy (pink band, Fig. 2) and are separated by a step down in CO 2 concentrations as T proxy exhibits significant cooling within the core of the Antarctic Cold Reversal (dark blue band, Fig. 2).Evidence from Southern Ocean marine sediment cores directly links each of these two warming steps with release of CO 2 accumulated in the deep Southern Ocean during the last glacial period: cores from south of the Antarctic Polar Front show pulses in upwelling (represented by opal fluxes) coinciding with each warming step (Anderson et al., 2009), while cores from the Atlantic sector of the Southern Ocean identify a source of old ( 14 C-depleted) carbon-rich deep water that dissipated over two corresponding intervals (Skinner et al., 2010).An explanation consistent with the reduced lag values identified here, proposes that the increases in upwelling were responsible for the simultaneous delivery of both sequestered heat and CO 2 to the atmosphere around Antarctica (Anderson et al., 2009;Skinner et al., 2010).However, this raises the question of what stimulated the increases in Southern Ocean upwelling.
A clue is provided by examining the time relationship between T proxy and the major millennial and sub-millennial climate events recorded by the North GRIP δ 18 O ice record (Fig. 2); i.e. the timing of the well-known bipolar seesaw (Broecker, 1998;Stocker and Johnsen, 2003).Since T proxy and the CO 2 records are synchronised to GICC05, the records from both hemispheres can be directly compared.There is also little or no time lag separating the major millennial and sub millennial climate transitions in Greenland (Lowe et al., 2008) from the onsets and ends of the warming and cooling trends in Antarctica (Pedro et al., 2011).A picture thus emerges from the ice cores of rapid communication between Antarctic temperatures, Greenland temperatures and atmospheric CO 2 .We give attention here to two potentially complimentary mechanisms which appear consistent with these observations.One mechanism supports a greater role for atmospheric pathways (Anderson et al., 2009;Toggweiler and Lea, 2010;Lee et al., 2011) while the other supports a greater role for the ocean (Skinner et al., 2010;Schmittner and Galbraith, 2008).Both mechanisms begin with the discharge of meltwater Introduction

Conclusions References
Tables Figures

Back Close
Full and ice into the North Atlantic during Greenland Stadial 2 (Heinrich Stadial 1), possibly due to orbitally driven increases in summer insolation.This causes a collapse of the Atlantic Meridional Overturning Circulation (AMOC) (Ganopolski and Rahmstorf, 2001;McManus et al., 2004).The atmospheric pathway proposes that rapid cooling of the North Atlantic region and sea ice expansion initiates an atmospheric reorganisation in which both the Inter Tropical Convergence Zone and the Southern Hemisphere mid-latitude westerlies are displaced south and/or intensified (Anderson et al., 2009;Toggweiler and Lea, 2010;Lee et al., 2011).The changed westerly regime over the Southern mid-and high-latitude oceans then dissipates sea ice, displaces cold fresh surface waters, drawing up warmer deeper waters which release heat and CO 2 .A recent modeling study lends support to this sequence of events and also incorporates an important role for ocean biogeochemical feedbacks (Lee et al., 2011).In line with the constraints identified here, the model supports essentially simultaneous North Atlantic cooling and atmospheric CO 2 release from the high-latitude Southern Ocean, corresponding to an atmospheric CO 2 rise of 20-60 ppm, where the range is dependent on the biological response.However Lee et al. (2011) do not simulate the observed response of Antarctic temperatures, possibly due to fixed Southern Hemisphere sea ice.
The ocean pathway invokes the bipolar seesaw: freshwater-induced collapse of the AMOC reduces northward ocean heat transport, thereby allowing heat to accumulate in the south (Broecker, 1998;Stocker and Johnsen, 2003).This melts back sea ice, removing a physical barrier to the release of CO 2 and exposing the Southern Ocean surface waters to the action of the westerlies.As above, the strengthened westerlies promote upwelling and ventilation of CO 2 from the deep waters (Skinner et al., 2010).The bipolar seesaw, as originally conceived (Broecker, 1998), operates on the millennial timescales of ocean heat transport and hence fails to meet the timing criteria identified here.However, an ocean wave-mediated seesaw (Stocker and Johnsen, 2003) could be capable of sufficiently rapid signal transmission.In one model study, the bipolar-seesaw-induced changes in the circulation modes couple with changes in

Conclusions References
Tables Figures

Back Close
Full ocean biogeochemistry and carbon storage in the terrestrial biosphere to produce little or no lag between Greenland and Antarctic temperatures and CO 2 changes (Schmittner and Galbraith, 2008).Unlike, Lee et al. (2011), this model does simulate the observed seesaw-relationship between North Atlantic and Greenland temperatures.Notably, the initial (within the first 250 years) rapid response in CO 2 in Schmittner and Galbraith (2008) is attributed to CO 2 release from the terrestrial biosphere (triggered by cooling associated with the freshwater-induced AMOC collapse) and not to physical or biogeochemical components of the Southern Ocean carbon cycle.However, this does not dismiss the possibility of a rapid ocean CO 2 response to bipolar seesawrelated changes in ocean circulation.Recent studies emphasise the sensitivity of the timescales of ocean ventilation to model resolution; at least under modern boundary conditions, meso-scale eddy-resolving ocean models suggest much faster ventilation times than the coarser resolution models that are currently used for palaeo-simulations (Maltrud et al., 2010).Current palaeo-modeling efforts appear to be accurately simulating key components of the deglacial sequence of events.However, there is not yet a simulation which fully captures all of the observed trends and phase relationships between Antarctic and North Atlantic temperatures and atmospheric CO 2 .The results presented here, with their improved timing constraints over previous studies, provide a more stringent target for future modeling efforts.

Conclusions
The ice core observations point to a tightly-coupled system operating with little or no time delay between the onsets/terminations of North-Atlantic climate stages and nearsimultaneous trend changes in both Antarctic temperature and atmospheric CO 2 .As it stands, the observed timing of events lends support to the current concept of an atmospheric teleconnection between the northern and southern high-latitudes, which forces wind-driven CO 2 release from the Southern Ocean.However, sorting out the relative Introduction

Conclusions References
Tables Figures

Back Close
Full roles of atmospheric and oceanic coupling mechanisms and the interactions between them remains a major challenge.More densely sampled and higher precision CO 2 measurements from high-accumulation ice core sites may assist with constraining the time evolution of the temperature/CO 2 lag during different stages of the deglaciation.However, if the response in the Southern Hemisphere is instantaneous, then higher resolution records may not be sufficient.Further progress in understanding these mechanisms requires more work at the interface between palaeoclimate observation and Earth system modeling, in particular modeling efforts to more tightly constrain the potential timescales of bipolar seesaw-induced changes in CO 2 ventilation.
The T proxy series is archived at the Australian Antarctic Data Center (http://data.aad. gov.au/) and at the World Data Center for Paleoclimatology (http://www.ncdc.noaa.gov/paleo/).The CO 2 records on the GICC05 timescale will be similarly archived following publication.Introduction Full Fig. 2. Atmospheric CO 2 and the bipolar seesaw on the GICC05 timescale.Significant warming and cooling trends in T proxy are represented by coloured vertical bands, adopted from a previous study (Pedro et al., 2011).Climate in the North Atlantic region is represented by the NorthGRIP ice core δ 18 O record (Lowe et al., 2008).Changes in the slope of T proxy are synchronous with climate transitions in the North Atlantic (vertical orange lines) within relative dating uncertainties (horizontal error bars).The deglacial increase in CO 2 occurs in two steps, corresponding to the significant warming trends in T proxy (pink bands).A pause in the CO 2 rise is aligned with a break in the warming trend (central grey band) during the Antarctic Cold Reversal (ACR).Within the core of the Antarctic Cold Reversal significant cooling in T proxy (dark blue band), coincides with an apparent decrease in CO 2 .Fast-acting inter-hemispheric coupling mechanisms linking Antarctica, Greenland and the Southern Ocean are required to satisfy these timing constraints.
. However, there are open questions about the Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , we arrive at a weighted value for the lag of 163 years and a methodological uncertainty of 34 years.Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Fig. 1 .
Fig. 1.The phase relationship between regional Antarctic temperature and atmospheric CO 2 .A) Antarctic temperature proxy T proxy (green) and CO 2 data (1 σ error bars) from Byrd (red) and Siple Dome (blue) on the GICC05 timescale.In the example shown, the CO 2 curves have been smoothed with a Gaussian filter of width 105 years.(B) Lag histograms for the two lag determination techniques, direct correlation and correlation of derivatives, using each of the two CO 2 data sets (red, blue) and the best fit of Gaussian distributions (black curves and µ and σ values).The histogram widths reflect each lag determination's sensitivity to the degree of data smoothing, CO 2 measurement uncertainties and different choices of lag calculation data interval start and end points -black triangles in part (A).The map shows the location of the Antarctic ice core sites from which data have been used (source: NASA).