Development of a Reduced-Complexity Climate Model and Applications on Glacial-Interglacial Timescales

Transcription

1 Development of a Reduced-Complexity Climate Model and Applications on Glacial-Interglacial Timescales Inauguraldissertation der Philosophisch naturwissenschaftlichen Fakultät der Universität Bern vorgelegt von Stefan Ritz von Ferenbalm (BE) Leiter der Arbeit: Prof. Dr. Thomas F. Stocker Abteilung für Klima und Umweltphysik Physikalisches Institut der Universität Bern

2

3 Development of a Reduced-Complexity Climate Model and Applications on Glacial-Interglacial Timescales Inauguraldissertation der Philosophisch naturwissenschaftlichen Fakultät der Universität Bern vorgelegt von Stefan Ritz von Ferenbalm (BE) Leiter der Arbeit: Prof. Dr. Thomas F. Stocker Abteilung für Klima und Umweltphysik Physikalisches Institut der Universität Bern Von der Philosophisch naturwissenschaftlichen Fakultät angenommen. Bern, 31. Mai 211 Der Dekan: Prof. Dr. Silvio Decurtins

4

5 Contents Thesis Summary 5 1 Introduction From the Past to the Future The Energy Balance of the Atmosphere Glacial-Interglacial Cycles Earth System Models of Intermediate Complexity The Bern3D EMIC Bibliography The Bern3D Energy and Moisture Balance Atmosphere A Dynamical Ocean Energy Balance Atmosphere Model Extended Model Description Discretizations Model Setup and Options Bibliography AMOC Reconstruction Introduction Method Ocean Temperature Reconstructions kyr AMOC Reconstruction AMOC Reconstruction of the Last Deglaciation Conclusions Bibliography Noble Gases as Proxies of Mean Ocean Temperature Introduction Model Formulation of Noble Gases and N Sensitivity of δkr atm to Various Model Parameters Calibration Uncertainties Sensitivity of δkr atm to Ocean Mixing Conclusions Acknowledgments Bibliography Atmospheric Radiocarbon during the Last Deglaciation Introduction Model Setup Radiocarbon Production Rate

6 4 CONTENTS 5.4 Radiocarbon Distribution of the LGM Ocean Abrupt Deglacial AMOC Changes Conclusions Bibliography Ocean Circulation Changes and Marine Reservoir Age Outlook The Bern3D Model Model Applications Bibliography A Abbreviations 155 Acknowledgments 157 Publications 159 Erklärung gemäss RSL Curriculum vitæ 163

7 Thesis Summary Global climate models are very valuable tools in climate sciences and are therefore used in many areas. The applications differ depending on the complexity of the components represented in the model. Three-dimensional physical models of the ocean and the atmosphere permit the analysis of the present-day and past dynamics within each component and to project future anthropogenically induced changes. Model results can be compared to observations of temperature, salinity, and various tracers such as chlorofluorocarbons and radiocarbon. For the analysis of past states and transient changes of the climate system where observations are not available, information must be reconstructed from proxy data from climate archives such as tree-rings, marine and lake sediments, ice sheets and glaciers, corals, and speleothems. With climate models, the gained information can be interpreted by simulating these proxies. In contrast to comprehensive climate models, climate models of reduced complexity have the benefit that simulations of a certain time interval require much less CPU time. This permits a larger number of simulations as well as simulations over longer time intervals. The high computational efficiency of this category of climate models is very favorable for studies where novel tracers are implemented, sensitivities are quantified, or new hypotheses are tested, because these studies involve many simulations. Reduced complexity climate models are however generally of greatly reduced complexity compared to comprehensive climate models, processes are simplified or ignored, and many important forcings are prescribed. The aim of this thesis is to extend the cost-efficient Bern3D three-dimensional reducedcomplexity ocean model by a two-dimensional energy and moisture balance atmosphere in order to make the model suitable for transient simulations in the past and into the future. Strong emphasis is put on the cost-efficiency of the coupled ocean-atmosphere model, as the model should remain fit for the types of studies mentioned above. This is documented in several paleoclimatic applications performed with the coupled model. Chapter 1 discusses the energy balance of the atmosphere and gives an overview on the current state of knowledge concerning the occurrence of glacial-interglacial cycles and the processes involved in the last deglaciation. In a third part, Earth System Models of Intermediate Complexity (EMICs) are introduced and positioned within the suite of coupled ocean-atmosphere models. Then, the current state of the Bern3D EMIC is summarized. Chapter 2 describes in detail the energy and moisture balance model that is coupled to the Bern3D ocean model, and the modern and last glacial maximum (LGM) state of the coupled ocean-atmosphere model is presented. Two first applications of the coupled model are discussed. In the first, several simulations of the past 8, years are performed and the contributions of the different forcing factors to the glacial-interglacial atmospheric temperature change are analyzed. From these simulations, the sensitivity of ocean temperature to atmospheric temperature, Atlantic meridional overturning circulation (AMOC), and Antarctic Bottom Water strength is analyzed at 23 locations. In a second application the equilibrium

8 6 THESIS SUMMARY climate sensitivities of the modern and of the LGM state are compared. The temperature rise for a doubling of the CO 2 concentration from LGM conditions is with 4.3 C notably larger than in the modern case (3 C). The relaxation time scale of atmospheric temperature is strongly dependent on the response of AABW to the CO 2 change. Chapter 3 presents a new method to qualitatively and quantitatively reconstruct the evolution of the AMOC strength on glacial-interglacial timescales by the combination of paleoclimate records and climate model simulations. Two reconstructions of the past 33 kyr are performed, one based on proxies of deep ocean temperature, the other on time series of sea-surface temperature proxies. Both AMOC reconstructions suggest an increase at glacial inceptions followed by a decrease throughout glacial intervals. Because of the age-scale uncertainties of the temperature reconstructions it is concluded that the method is not suitable to detect abrupt ocean circulation changes. However, the method may be useful to quantitatively estimate the magnitude and long-term trends in the AMOC strength on glacial-interglacial timescales. In Chapter 4, a proposition is tested where past global mean ocean temperatures can be reconstructed by measuring noble gas concentrations in ice core bubbles. For this, krypton, xenon, argon, and molecular nitrogen are implemented into the Bern3D model and the characteristics of these novel paleoclimatic proxies are explored. It is concluded that atmospheric noble gas concentrations are suitable proxies of global mean ocean temperature. Changes in ocean volume need to be considered when reconstructing ocean temperatures from noble gases. Calibration curves are provided to translate ice-core measurements of krypton, xenon, and argon into a global mean ocean temperature change. Chapter 5 analyzes the decline of atmospheric 14 C during the last deglaciation that has not yet been successfully explained. The contributions of changes in the radiocarbon production rate, rapid changes of the AMOC, and of the difference between the LGM and modern state of the ocean are quantified. Depending on the carbon distribution within the ocean during the LGM, the LGM-to-modern atmospheric 14 C difference ranges between 4 and 219. Modeled LGM benthic-planktonic radiocarbon age differences are compared to reconstructions from marine sediments. Chapter 6 discusses the effect of abrupt ocean circulation changes on marine radiocarbon reservoir age. Model results are compared to observations of the Younger Dryas, a cold interval during the last deglaciation. It is found that reservoir ages decrease by about 1 yr in the Atlantic as a consequence of a shutdown of the AMOC. The effect of changes in sea-ice cover also play a major role. Therefore, changes in marine reservoir age need to be considered when dating marine organisms during the deglaciation. In this study the ocean-only model is used as the study was carried out prior to the development of the atmospheric component of the model. Finally, an outlook is given in Chapter 7.

9 Chapter 1 Introduction 1.1 From the Past to the Future By the massive burning of fossil fuels, changes in land-use and other anthropogenic activities, mankind is currently steadily increasing the amount of carbon dioxide (CO 2 ), methane (CH 4 ) and other greenhouse gases in the atmosphere (IPCC, 27). While greenhouse gases are almost transparent regarding the shortwave radiation from the Sun, they partially absorb the Earth s longwave radiation and re-emit it in all directions. Because a part of the longwave radiation is re-emitted back towards the Earth s surface, the outgoing radiation into space is reduced. The higher the greenhouse gas concentration in the atmosphere, the more outgoing radiation is absorbed which in turn warms the atmosphere (see Section 1.2 for details). One of the most important topics of climate research is to assess naturally and anthropogenically induced future changes in the climate system on a global and regional scale with respect to future warming, sea-level rise, the likelihood of extreme weather events, and their dependence on emission scenarios. This requires an understanding of the internal feedbacks of the climate system. How will the Greenland and Antarctic ice sheets respond to an initial warming? How sensitive are the climate system and global temperatures in particular to a given change in atmospheric CO 2? How do clouds and aerosols feed back? What is the impact on the largescale ocean circulation, a very important contributor to the Earth s heat distribution? To tackle these and many other related questions, climate models are used to understand presentday climate dynamics and to calculate climate projections into the future by prescribing fossil fuel and aerosol emissions, land-use changes, and other forcing factors that are based on social, economic, and technological assumptions (IPCC, 27). However, in order to make robust predictions of future climate change, it is very important to understand the natural variations of climate in the past. This involves the detection of past climatic changes from paleoclimatic records such as tree-rings, ice, lake and marine sediment cores, speleothems, etc., and their attribution to particular processes. This can only be done by the combination of careful analysis of paleoclimatic data and climate model simulations. In this thesis, a reduced-complexity ocean model is extended to include an energy-balance atmosphere to make the model suitable for paleoclimatic model studies on the time-scale of glacial-interglacial cycles or for sensitivity studies where many simulations are required. In the studies presented in this thesis, the model is applied to a range of paleoclimatic topics that focus on the glacial-interglacial cycles of the last 8, years and on the last deglaciation, i.e., the last 2, years. This introductory chapter describes the atmospheric energy balance, provides a short review of the research done on the characteristics of glacialinterglacial cycles and on the mechanisms involved in the last deglaciation. Then, Earth System Models of Intermediate Complexity (EMICs) are introduced and positioned within the suite of coupled ocean-atmosphere models. Finally, the current state of the Bern3D EMIC

10 8 1. INTRODUCTION is summarized. 1.2 The Energy Balance of the Atmosphere Solar Radiation The energy available at the Earth s surface that enables life comes almost exclusively from the Sun. The total solar irradiance S (referred to as the solar constant) at the Earth s distance from the Sun is about 1361 Wm 2 (Kopp & Lean, 211). However, the power of the Sun is not constant but varies at all wavelengths and over multiple timescales. Various observations have detected a 11-year cycle where S varies by approximately 1.6 Wm 2 between recent minima and maxima (Kopp & Lean, 211, and references therein). The mean daily incoming solar radiation is with 34 W m 2 only a fourth of S, because the surface of the Earth A E = 4πr 2 E π catches πr2 E S of the solar power, where r E is the radius of the Earth. Parts of the incoming solar radiation is scattered in the atmosphere by clouds, aerosols and the air and is partly reflected back to space. The reflected amount depends on the cloud cover, on the aerosol concentration, and on the atmospheric optical depth that in turn depends on the atmospheric humidity. The solar radiation that reaches the surface of the Earth is partly reflected at the surface. The surface albedo, the relative amount of the reflected radiation, depends on the surface and varies between 4 % and 8 %. Dark surfaces like the ocean or forests have a rather low albedo, while the albedo is high for ice and snow. In total, about one third of the incoming solar radiation is reflected back to space (Fig. 1.1). This fraction is often referred to as the planetary albedo. The Sun radiates most of its energy at wavelengths between.25 µm and 2.5 µm. The spectrum is close to that of a black body with a temperature of about 6 K (Peixoto & Oort, 1992). The difference between the spectral solar irradiance at the top of the atmosphere (TOA) and solar radiation at the Earth s surface reveals absorption bands of the various species of molecules of the air (Fig. 1.2). The molecules in the air absorb parts of the solar radiation while being pushed into a higher energetic state. Every molecule absorbs at different wavelengths. The most important absorbing gas at the range of wavelengths of the solar radiation is water vapor. According to Kirchhoff s law of thermal radiation, the Earth in equilibrium must radiate as much energy as it absorbs. The radiation spectrum of the Earth is similar to that of a blackbody. Because the Earth is much cooler than the Sun, the Earth emits at longer infrared wavelengths (Wien s law; Fig. 1.3). Therefore, the outgoing radiation of the Earth s surface is termed longwave radiation as opposed to the shortwave radiation of the Sun. Next to the longwave radiation, energy is transferred from the surface to the atmosphere by thermals that occur when the boundary air layer is heated by the surface (also referred to as sensible heat flux), and by evapotranspiration (referred to as latent heat flux) Greenhouse Effect As with the solar radiation, certain air molecules partially absorb the Earth s longwave radiation while passing into a higher rotational or vibrational state. Atmospheric gases with absorption bands in the wavelength interval of the surface radiation are referred to as greenhouse gases. The most important natural greenhouse gases are water vapor, CO 2, ozone (O 3 ), CH 4, and nitrous oxide (N 2 O) (Fig. 1.4). The combined absorption of all greenhouse gases is also shown in Fig The absorbed energy is eventually re-emitted in all directions, thus partially back towards the surface. Therefore, energy is transferred to the lower atmosphere

11 1.2. THE ENERGY BALANCE OF THE ATMOSPHERE 9 Figure 1.1: Estimate of the Earth s annual and global mean energy balance (Kiehl & Trenberth, 1997; IPCC, 27). In equilibrium, the net radiation at the top of the atmosphere is zero. About one third of the incoming solar radiation is reflected back to space. Due to the greenhouse gases in the atmosphere, the longwave radiation of the Earth s surface is absorbed and partially reflected back towards the surface. and to the surface leading to warmer temperatures than when the direct heating of the solar radiation were the only warming mechanism. This mechanism is referred to as the natural greenhouse effect. Without this natural greenhouse effect, the surface of the Earth were only 22 C. This value is easily calculated by equating the solar insolation F in = (1 α p )F TOA with the blackbody outgoing radiation of the Earth s surface F out = σt 4 (Stefan-Boltzmann law). α p.33 is the planetary albedo (Kiehl & Trenberth, 1997), F TOA = 342 Wm 2 the solar irradiance at TOA, σ = W m 2 K 4 the Stefan-Boltzmann constant, and T the surface temperature of the Earth in Kelvin Anthropogenic Greenhouse Effect Next to water vapor that absorbs the Earth s surface radiation at broad wavelength intervals, CO 2 has an important absorption band at around 15 µm (Fig. 1.4). Because absorption goes towards 1 % already in the lower atmosphere, it has been argued that an anthropogenic increase in atmospheric CO 2 would not increase the greenhouse effect and thus not further contribute to the global warming. However, this is not true for the following reason. First, it must be noted that the absorbed radiation by the air molecules is always re-emitted back to the atmosphere. One half is emitted towards the surface, but the other half is emitted upwards. The re-emitted upward radiation is re-absorbed at a higher atmospheric level, re-emitted, etc. At levels near the top of the troposphere, typically between 5 and 1 km altitude, the radiation is emitted back to space. Note that TOA cannot be attributed to a particular altitude, because the atmospheric pressure decreases approximately exponentially with height. The bottom 1 km of the atmosphere contain around three quarters of the total atmospheric mass. As the radiation at TOA occurs at a lower temperature than the radiation at the Earth s surface, the thermal radiation is weaker (Fig. 1.3). When atmospheric CO 2

12 1 1. INTRODUCTION Figure 1.2: Spectrum of incoming solar radiation at the top of the atmosphere (upper curve) and at sea level (lower curve). The shaded area represents the absorption bands of the air molecules. The unshaded area between the curves represents the portion of the solar radiation that is reflected back to space by clouds or backscattered by the air and aerosols (Peixoto & Oort, 1992). N Figure 1.3: Spectral outgoing longwave radiation in the infrared at the top of the atmosphere measured over the Sahara (Hanel et al., 1972). The absorption bands of the atmospheric greenhouse gases are clearly visible. The dashed lines indicate the blackbody radiation of an object at various temperatures. The blackbody radiation at 32 K approximates the surface radiation of the Earth (at the bottom of the atmosphere). The atmosphere is almost opaque at the wavelength of around 15 µm because of CO 2. Thus, the radiation at a wavelength of 15 µm is radiated to space at upper-most CO 2 layer, approximately at the Tropopause where the temperature is about 22 K.

13 1.3. GLACIAL-INTERGLACIAL CYCLES 11 Figure 1.4: Spectral transmission and absorption of infrared radiation in the troposphere of the each molecule of the air that contributes to the total atmospheric absorption (other minor species and anthropogenically induced species are not shown). These are carbon monoxide CO, CH 4, N 2O, O 3, CO 2, and water vapor H 2O and HDO, where HDO is semi-heavy water that is enriched by a deuterium isotope. The total atmospheric absorption is shown in the bottom panel (Hartmann, 1994). increases, the thermal radiation into space occurs at higher and colder atmospheric levels than before. Hence, the outgoing radiation decreases. This leads to a warming of the atmosphere until the incoming and the outgoing radiation at TOA are in balance. The CO 2 -induced warming of the atmosphere leads to several important feedbacks such as the water vapor feedback, where the warmer atmosphere can take up more water vapor. This positive feedback leads to an additional warming of the atmosphere because H 2 O is a powerful greenhouse gas. 1.3 Glacial-Interglacial Cycles During the past million years, warm climatic phases referred to as interglacial periods succeeded cold, glacial intervals (ice ages) and vice versa with a cyclicity of approximately 1, years (1 kyr). During glacial times, atmospheric and ocean temperatures were lower than during warm phases, and land ice (also referred to as ice sheets) covered large parts of North America and Europe. Consequently, sea level was lower. At the last glacial maximum (LGM) 2 kyr ago, global mean atmospheric temperatures were 4 C to 7 C cooler than today (before anthropogenically induced climate change; Jansen et al., 27). The temperature difference was with 2 C to 3 C smaller in the tropics than at high latitudes (Farrera et al., 1999), and the LGM sea level was about 12 m lower than today (Peltier, 22). In the following,

14 12 1. INTRODUCTION the current understanding of the mechanisms involved in the transitions from warm to cold phases and vice versa is summarized Orbital Parameters Solar radiation varies in time and is therefore an external forcing factor of the climate system. The temporal variations of the solar radiation are due to the precession of the Earth s axis, changes of the tilt of the Earth s axis (also referred to as obliquity), and changes of the eccentricity of the Earth s orbit around the Sun (Fig. 1.5). The periodicities of these parameters are approximately 23 kyr for precession, 41 kyr for obliquity, and 1 kyr for eccentricity (e.g., Berger, 1978, Figs. 1.6a-c). Changes in precession and obliquity only affect the regional distribution but not the annually integrated insolation of the Earth, since the distance between the Earth and the Sun is not affected. Milankovitch (1941) was the first to propose a link between the orbital cycles and climate. He hypothesized that summer changes in insolation (Fig. 1.6d) caused by the precession and obliquity cycles modulated the Northern Hemisphere ice-sheet volume in a direct way by ablation leading alternately to warmer and colder climatic conditions globally, i.e. to warm climatic phases and ice ages. He also hypothesized that the ice sheets should lag the insolation forcing by approximately 5 kyr due to their slow response. Hays et al. (1976) partially confirmed this theory with an oxygen isotope ratio (δ 18 O) record of the past 3 kyr from a marine sediment core. The record, a proxy for global ice volume, clearly shows coherent variations of the ice sheets and the high latitude Northern Hemisphere insolation in the precession and obliquity frequencies. Also, the proposed phase lag between the insolation and the ice sheets was found by comparing the δ 18 O ratios to summer sea-surface temperature proxies of the same sediment cores. However, the largest ice-sheet changes were found to vary with the 1-kyr cyclicity of eccentricity, today known as the glacial-interglacial cycles. Extended ice volume proxy records find that this 1-kyr cyclicity is abundant back to 1.2 million years (Myr) ago (Fig. 1.6e; Shackleton, 1995; Lisiecki & Raymo, 25). In addition, other records, for example air temperature reconstructed from an Antarctic ice core back to 8 kyr before present (BP, before year 195 AD) (Petit et al., 1999; EPICA Community Members, 24; Jouzel et al., 27), show this 1-kyr cyclicity (Fig. 1.6f) Glacial-Interglacial Cycles of the Last Million Years The radiative forcing caused by insolation changes is far too small to account alone for the reconstructed glacial-interglacial atmospheric temperature changes (see for example Section Eccentricity (~1 kyr) Obliquity (~41 kyr) Precession (~23 kyr) Figure 1.5: Visualization of the orbital parameters eccentricity, obliquity, and precession.

15 1.3. GLACIAL-INTERGLACIAL CYCLES 13 Precession e sin(ω) ~ Eccentricity e Marine δ 18 O ( ) Atm. CO 2 (ppm) Atm. N 2 O (ppb) a) b) c) d) e) f) g) h) i) Obliquity (deg) ǫ July insolation at 65ºN (W m 2 ) Antarctic δd ( ) Atm. CH 4 (ppb) Time (kyr BP) Figure 1.6: Variations over the last 8 kyr of the orbital parameters, solar radiation, ice volume and atmospheric temperature proxies, and various greenhouse gases (BP: before present). a) Precession (Berger, 1978). Is a function of the eccentricity e and the longitude of the perihelion relative to the moving vernal equinox ω (see Section and Fig. 2.2 for details). b) The obliquity of the Earth s axis in degrees (Berger, 1978). c) The eccentricity (see Fig. 2.2 for details; Berger, 1978). d) Mid July solar radiation at 65 N (Berger, 1978). e) δ 18 O stack from benthic foraminifera of marine sediment cores (Lisiecki & Raymo, 25). The oxygen isotope ratio is a proxy for global ice volume. f) Deuterium-to-hydrogen ratio (δd) from the Antarctic EDC ice core is an air temperature proxy (Jouzel et al., 27). g) Atmospheric CO 2 concentrations (in parts per million) from air enclosed in Antarctic ice cores (Petit et al., 1999; Lüthi et al., 28). h) Atmospheric CH 4 concentrations in parts per billion from the Antarctic EDC ice core (Loulergue et al., 28). i) Atmospheric N 2O concentrations from the Antarctic EDC ice core (Schilt et al., 21a) (in black) and from the Talos Dome ice core (Schilt et al., 21b) (in gray). Intervals without measurements arise where no robust N 2O measurements were possible. The shaded vertical bars indicate interglacials based on the criterion of EPICA Community Members (24) (intervals where Antarctic δd 43 ).

16 14 1. INTRODUCTION 2.1). Hence, internal feedbacks of the climate system must have amplified the radiative forcing caused by the changes of the orbital parameters. The two most important amplifiers are the ice-albedo feedback, i.e. changes in the reflectivity of the Earth s surface through changes in ice-sheet and sea-ice cover, and changes in the atmospheric greenhouse gas concentrations, where water vapor is the most important, followed by CO 2, CH 4, and N 2 O. The atmospheric mineral dust load is another important factor that impacts the radiative forcing. The forcing imposed by the dust aerosol is however much more complex compared to the greenhouse gases, because dust partly absorbs and partly scatters incoming solar radiation, but it also absorbs and emits outgoing longwave radiation (Tegen, 23). The magnitude and even the sign of the radiative forcing depends on the optical properties of the dust, i.e. on the size, on the vertical distribution, on the presence of clouds, etc. The global annual radiative forcing inferred by the atmospheric mineral dust load is most likely negative (IPCC, 27). Reconstructions of the atmospheric concentrations of CO 2, CH 4, and N 2 O of the past 8 kyr from air bubbles trapped in Antarctic ice cores also show large glacial-interglacial variations (Lüthi et al., 28; Loulergue et al., 28; Schilt et al., 21a,b, Figs. 1.6g-i). The same is true for the dust load. Antarctic dust deposition was about 25 times higher during glacials resulting from a reduced hydrological cycle during ice ages (Lambert et al., 28). In Section 2.1 it is shown that changes in the radiative forcing caused by parametrized changes of the ice-sheet extent and the atmospheric greenhouse gas concentrations are sufficient to explain the global glacial-interglacial temperature variations when the feedback associated with atmospheric water vapor is taken into account. Even though the magnitude of the global glacial-interglacial temperature change can be explained and the trigger for glacial inceptions and glacial terminations is believed to be the external solar forcing, the reason for the observed 1-kyr cyclicity is still far from understood, because changes in insolation due to the eccentricity are very small compared to the precessional and obliquity caused changes. Another interesting feature of the past glacialinterglacial cycles is that from about 2.7 to 1.2 Myr ago, ice-sheet changes followed the 41-kyr cyclicity of obliquity (Fig. 1.7; Shackleton, 1995; Lisiecki & Raymo, 25). This raises questions of the difference between these climatic phases and why the transition from the 41-kyr to the 1-kyr cyclicity occurred. Solving these questions will give important hints to the mechanisms of glacial inceptions and glacial terminations Mid-Pleistocene Transition The glacial-interglacial cycles of the early-pleistocene, the period from about 2.7 to 1 Myr ago that was dominated by the 41-kyr cyclicity, have been attributed to the changes in Earth s obliquity (Raymo & Nisancioglu, 23; Huybers, 26), while the glacial-interglacial cycles of the late-pleistocene (the last 1 Myr) have been thought to be the result of changes in eccentricity (Hays et al., 1976; Imbrie et al., 1993; Raymo, 1997). However, these studies cannot explain the transition from the 41-kyr to the 1-kyr glacial-interglacial cycles that is referred to as the mid-pleistocene Transition (Fig. 1.7). Because of the lack of a substantial change of the orbital parameters at the mid-pleistocene Transition, several modeling studies have associated a long-term cooling with the origin of the transition. Proposed cooling mechanisms are a possible long-term drawdown of atmospheric CO 2 concentrations due to tectonic processes (Raymo, 1997) or increased ice-sheet thickness due to a decreased ice flow. It is argued that low-friction regolith underlying the ice sheets is removed by multiple cycles of ice-sheet erosion exposing high-friction, unweathered crystalline bedrock which leads to thicker ice sheets (Clark & Pollard, 1998; Clark et al., 26). However, recent atmospheric

17 1.3. GLACIAL-INTERGLACIAL CYCLES kyr cycles Mid-Pleistocene Transition 1-kyr cycles Marine δ 18 O ( ) Time (kyr BP) 5 Figure 1.7: The evolution of global ice volume over the Pleistocene epoch (Lisiecki & Raymo, 25). The Pleistocene is divided into the early-pleistocene from about 2 to 1 Myr ago that is characterized by glacialinterglacial cycles with a 41-kyr periodicity, the late-pleistocene that is dominated by glacial-interglacial cycles with a 1-kyr periodicity, and the mid-pleistocene Transition in between. CO 2 reconstructions from ice cores back to 8 kyr BP (Lüthi et al., 28) and from boron isotopes back to 2 Myr BP (Hönisch et al., 29) lack a gradual decrease in interglacial CO 2 concentrations and therefore do not support the suggestion that a long-term CO 2 drawdown was the main cause for the mid-pleistocene Transition. Paillard (1998) developed a simple model that can reproduce the 41-kyr cycles of the early- Pleistocene, that positions the mid-pleistocene Transition correctly, and that reproduces all 1-kyr cycles of the late-pleistocene with respect to global ice volume. The model assumes three possible climate states: an interglacial state, a mild glacial state, and a full glacial state. The model also incorporates a linear ice-sheet volume formulation that is forced by the summer insolation at 65 N. Transitions between the climate states occur as soon as the ice volume reaches prescribed thresholds. With constant ice volume thresholds and without taking into account long-timescale changes in atmospheric CO 2, the model successfully reproduces the glacial-interglacial cycles of the late-pleistocene. The model is also successful in simulating the glacial-interglacial cycles of the entire Pleistocene when the glacial ice-volume threshold is linearly increased in time and a linear trend is added to the radiative forcing to account for the possible gradual decrease of atmospheric CO 2 during the early-pleistocene. Although the conceptual model presented by Paillard (1998) clearly indicates a strong link between high northern latitude summer insolation and global ice volume, the model is highly tuned and therefore too simple to reveal the mechanisms behind the coupling of insolation and ice-sheet extent. A similar result is obtained in a modeling study by Berger et al. (1999). They simulate the transient behavior of the Northern Hemisphere ice-sheet volume to changes in the insolation and in atmospheric CO 2 over the past 3 Myr using the LLN-2D climate model. The model incorporates a two-dimensional atmosphere of the Northern Hemisphere resolved in latitude and altitude, a mixed-layer ocean, dynamic sea-ice, and a model of the three major Northern Hemisphere ice sheets. In their simulations the ice-sheet volume is dominated by the 1-kyr cycle when CO 2 is fixed to 22 ppm and by the 41-kyr cycle when CO 2 is fixed to 28 ppm. In a simulation where CO 2 is linearly decreased from 32 ppm at 3 Myr BP to 2 ppm at 1

18 16 1. INTRODUCTION Myr BP, the dominating period switches from 41-kyr to 1-kyr at 1 Myr BP as found in the ice-volume reconstructions. They also note that forcing the model with the glacial-interglacial variations of atmospheric CO 2 as reconstructed from ice cores does not significantly change the outcome but that it increases the amplitude of the glacial-interglacial cycles. Even though the recent progress in atmospheric CO 2 reconstructions does not suggest a long-term decrease over the last 3 Myr, the model studies remain valid if the long-term cooling observed in several Atlantic and Pacific sea-surface and deep-ocean temperature time series reconstructions (e.g., Lawrence et al., 21; Liu & Herbert, 24; Liu et al., 25) is caused by another internal mechanism. Raymo et al. (26) propose that before the mid- Pleistocene Transition, the Northern and Southern Hemisphere ice sheets waxed and waned, each controlled by local summer insolation. Because precession is out of phase between hemispheres, they argue that the precessional signal is canceled out in globally integrated proxies such as the δ 18 O global ice-volume proxy leaving the in-phase obliquity component of insolation to dominate those records. They further argue that after the mid-pleistocene Transition, climate has cooled to an extent that the grounding line of the Antarctic ice sheet fell below sea-level. With this, the Antarctic ice sheet is no longer paced by local insolation but by sea-level changes that are caused by changes in the Northern Hemisphere ice sheets. Hence, ice sheets of the late-pleistocene in both hemispheres varied in phase at both obliquity and precession frequencies leading to the 1-kyr cyclicity. An alternative hypothesis by Huybers (29) suggests that the mid-pleistocene Transition was not a consequence of long-term cooling, but that the Pleistocene variability was chaotic and that transitions from the 41-kyr cycle to the 1-kyr cycle and vice versa have occurred spontaneously Importance of Southern Hemisphere Insolation Although the studies discussed in the previous section have contributed to the understanding of the pacing of glacial-interglacial cycles, they do not focus on how solar insolation triggers glacial-interglacial cycles. One question that is still under debate is to what extent Northern and Southern Hemisphere solar insolation contributes to the glacial-interglacial climate variability. Milankovitch (1941) and Hays et al. (1976) proposed that glacial-interglacial cycles are driven by Northern Hemisphere summer insolation, because the ice sheets covary with northern high latitude summer insolation intensity. Further evidence in favor of this hypothesis has been put forward by Kawamura et al. (27). They suggest that Antarctic climate responds to Northern Hemisphere insolation because they find that Antarctic air temperatures lag Northern Hemisphere summer insolation intensity at the precession and obliquity timescales by a few millennia. Huybers & Denton (28) on the other hand propose that Southern Hemisphere temperatures did not covary with the Northern Hemisphere insolation intensity, but rather with the duration of the Southern Hemisphere summer and that therefore a link between Northern and Southern Hemispheres is not necessary. Schulz & Zeebe (26) postulate that increasing insolation in both the Northern and Southern Hemispheres are necessary to trigger glacial terminations. They analyzed midsummer insolation at 65 N and 65 S that are generally antiphased. However, at seven glacial terminations of the late-pleistocene, Northern and Southern Hemisphere midsummer insolation increase in concert for more than 1 years. They propose that the total energy supplied during these phases are large enough to trigger deglaciations.

19 1.3. GLACIAL-INTERGLACIAL CYCLES 17 Stott et al. (27) locate Southern Hemisphere summer insolation to trigger deglaciations. In a marine sediment core from the western tropical Pacific that covers the last deglaciation, they find a significant deep-sea temperature increase by about 2 C. This temperature rise leads the increase of atmospheric CO 2 and tropical surface temperatures by approximately 1 kyr. Because the signal must originate from the Southern Ocean, they conclude that Southern Hemisphere insolation triggers deglaciations. More pieces of evidence are needed to pinpoint the role of each hemisphere to the glacialinterglacial climate variability. After the work by Huybers & Denton (28) that show a significant correlation between 65 N midsummer insolation and Southern Hemisphere summer duration (and hence average annual mean atmospheric temperatures), earlier conclusions of the Northern Hemisphere insolation as the driver of glacial-interglacial cycles must be reconsidered Late-Pleistocene Glacial Inceptions and Terminations To study the internal feedbacks of the climate system involved in glacial inceptions, coupled climate models that incorporate atmospheric, ocean, ice-sheet, and vegetation components are used. Modeling studies of Wang & Mysak (22), Wang et al. (25), Calov et al. (25), Kubatzki et al. (26), and Calov et al. (29) have simulated glacial inceptions using EMICs (see Section 1.4; Claussen et al., 22) to investigate the role of the orbital and CO 2 forcing, ocean thermohaline circulation, mineral dust, and vegetation feedbacks on glacial inceptions. Solar and atmospheric CO 2 radiative forcing are prescribed in all simulations. In the sensitivity study of Calov et al. (25), the solar forcing alone is sufficient for a glacial inception with the ice-albedo feedback as the major mechanism. The CO 2 forcing, as well as the feedbacks of the ocean and of the vegetation only accelerate glacial inception, while dust deposition on ice acts as a negative feedback. Wang et al. (25) note the importance of the vegetation-albedo feedback for ice-sheet buildup. They conclude that the slower icesheet growth over Eurasia compared to the ice-sheet growth over North America was due to the vast forests in Eurasia and their effect in reducing surface-albedo. In their simulations, constant vegetation leads to a considerably slower ice-sheet growth in North America and prevents the Eurasian ice sheet from appearing. Kubatzki et al. (26) find that changes the ocean and the vegetation amplify insolation induced glacial inception, but that reductions in atmospheric CO 2 have only a minor impact on ice-sheet buildup. Studies with more complex climate models were so far restricted to time-slice simulations (Yoshimori et al., 22; Vettoretti & Peltier, 24; Otieno & Bromwich, 29; Vavrus et al., 211). In these models, perennial snow cover gives indication of glacial inceptions instead of dynamic ice-sheet models as in the EMIC studies. In the model simulations of Vettoretti & Peltier (24), strong obliquity forcing alone or a strong eccentricity modulated precessional forcing combined with reduced atmospheric CO 2 are sufficient to induce glacial inception in the Canadian Arctic Islands. Otieno & Bromwich (29) discuss the importance of atmospheric conditions on glacial inceptions. They find that an increase in the frequency of extremely wet winters and cold spring and summer seasons are not sufficient for the growth of ice sheets in areas that are initially not glaciated unless summer temperatures cool by at least 4 C. Because present climate models are not yet able to realistically reproduce the transient changes of atmospheric CO 2 on glacial-interglacial timescales, the radiative forcing of CO 2 must be prescribed. It is however generally accepted that the difference between the glacial and interglacial atmospheric CO 2 inventory was taken up by the oceans. Many processes

20 18 1. INTRODUCTION involved in the ocean uptake of the additional carbon have been identified, but the efficiency of each mechanism is still unknown. Reviews on this topic are given by Sigman & Boyle (2), Archer et al. (2), Fischer et al. (21), and Sigman et al. (21). The terrestrial biosphere contained 3 7 Pg less carbon during glacials than today (Köhler & Fischer, 24, and references therein). This corresponds to a 1 to 3 % reduction of the carbon stock compared to the pre-industrial pool of about 23 Pg carbon (Sarmiento & Gruber, 26). This carbon anomaly is emitted to the atmosphere and then taken up by the ocean in addition to the atmospheric CO 2 anomaly. An evident feature of the glacial-interglacial cycles of the last million years is the saw-tooth shaped behavior of the ice volume, atmospheric temperature, etc. (Fig. 1.6; Broecker & van Donk, 197), where a slow glaciation phase of 7 to 1 kyr is followed by a rapid deglaciation phase of only about 1 kyr. As found in the ice-sheet proxy records and mentioned above, the ice sheets vary coherently with the precessional and obliquity driven insolation during glaciation phases. However, perhaps due to temporary reductions of the Atlantic meridional overturning circulation during insolation-driven warming phases every 2 kyr (Timmermann et al., 21), the ice-sheet volume does not simply follow insolation changes, but ice sheets continue to grow over several precession and obliquity cycles. At the end of the glaciation, at the glacial maxima, it is hypothesized that the ice sheets have become unstable because of their large size. Non-linear ice-sheet responses to initial insolation-triggered melting that might lead to the rapid deglaciation might be ice-berg calving to the ocean (Denton & Hughes, 1983), the formation of lakes on top of the ice sheets or at ice sheet margins (so called proglacial lakes, Andrews, 1973), dust accumulation on top of the ice sheets that reduces albedo (Berger et al., 199), and delayed bedrock rebound that keeps the ice at lower altitudes during phases of melting (Oerlemans, 198; DeBlonde & Peltier, 1991). The atmosphere warms due to the ice-albedo feedback of the melting ice sheets. Possibly due to the invoked climate changes, the ocean releases its stored carbon back to the atmosphere, again amplifying the warming by the greenhouse gas feedback. With the expansive ice-sheet melting and the increasing atmospheric greenhouse gas concentrations, the climate rapidly passes into an interglacial Termination 1 Many studies have been focusing on the last deglaciation from approximately 18 to 11 kyr BP, in order to describe and explain the transition from the LGM to the Holocene, the current warm interval. This transition is also referred to as Termination 1. High resolution reconstructions of Greenland and Antarctic air temperatures derived from ice cores demonstrate that Termination 1 was not gradual, but that it was interrupted by several abrupt climate changes (Figs. 1.8a,b). The deglaciation is separated into Heinrich event 1 from approximately 16.8 kyr to 14.6 kyr BP (Hemming, 24; Rasmussen et al., 26), the Bølling-Allerød from approximately 14.6 to 12.8 kyr BP (Rasmussen et al., 26) and the Younger Dryas from approximately 12.8 to 11.6 kyr BP (Rasmussen et al., 26). During Heinrich event 1, the Atlantic meridional overturning circulation (AMOC) that transports warm surface waters from the tropics to the high latitudes is believed to have shut down as a consequence of massive surges and melting of icebergs from the Laurentide ice sheet into the North Atlantic Ocean (Fig. 1.8e; Bond et al., 1992; Maier-Reimer & Mikolajewicz, 1989; Stocker et al., 1992; Rahmstorf, 1994; Stocker, 2; Stocker & Marchal, 2; McManus et al., 24). During this time, Southern Hemisphere temperatures rose. Also, atmospheric CO 2 increased considerably (Fig. 1.8c), possibly released from the Southern Ocean due to increased ventilation of the deep waters (Anderson et al., 29; Lee et al., 211). The Bølling-Allerød warm phase

21 1.3. GLACIAL-INTERGLACIAL CYCLES 19 Greenland δ18o ( ) Atm. CO2 (ppm) a) b) c) d) H1 B-A YD Antarctic δd ( ) Relative sea level (m) 231Pa/23Th e) Time (kyr BP) Pa/Th 232 based Pa/Th 238 based 5 Figure 1.8: The evolution of various reconstructed quantities during the last deglaciation. a) The Greenland oxygen isotope record from the NGRIP ice core (NGRIP members, 24) is a proxy for Greenland air temperature. b) Antarctic deuterium-to-hydrogen ratio from the EDC ice core (Jouzel et al., 27) is a proxy of Antarctic air temperature. c) Atmospheric CO 2 concentration (Monnin et al., 21). d) Barbados upliftcorrected eustatic sea level (Peltier & Fairbanks, 26) relative to the modern sea level. The data points are interpolated by a spline with a cutoff period of 1.5 kyr. Two splines are calculated due to the uncertainty of the measurements at around 13 kyr BP. e) Qualitative Atlantic meridional overturning circulation strength reconstruction based on Pa/Th activity ratios of an Atlantic sediment core (McManus et al., 24). The different phases of the deglaciation are separated by dashed lines: Heinrich event 1 (H1), the Bølling-Allerød warm phase (B-A), and the Younger Dryas (YD).

22 2 1. INTRODUCTION followed beginning with an abrupt resumption of the AMOC (Ganopolski & Rahmstorf, 21; Weaver et al., 23; Knorr & Lohmann, 23; Liu et al., 29) linked with warmer temperatures in the Northern Hemisphere and cooling in the Southern Hemisphere referred to as the Antarctic Cold Reversal. Before the end of the deglaciation, Northern Hemisphere climate dropped back to glacial temperatures during the Younger Dryas whereas Southern Hemisphere temperatures increased. Again, the AMOC is believed to have weakened by the drainage of proglacial lakes (Alley, 2; Broecker, 26). Atmospheric CO 2 increased throughout the Younger Dryas. This anti-phasing between the northern and the Southern Hemispheres is referred to as the bipolar seesaw and is related to the changes in the latitudinal heat transport by the AMOC (Crowley, 1992; Broecker, 1998; Stocker & Johnson, 23). Sea level rose as a consequence of ice-sheet melting and thermal expansion. Several relative sea-level reconstructions have been made for the last deglaciation from various sites around the world. While the fossil-coral based relative sea-level (RSL) record from Barbados covers the full deglaciation from about 22 kyr to 7 kyr BP (Peltier & Fairbanks, 26), other records cover parts of the last deglaciation: RSL reconstructions from the Sunda Shelf (Hanebuth et al., 2) and from the Bonaparte Gulf (Yokoyama et al., 2) cover the interval from the onset of Termination 1 to 14 kyr BP, while the reconstructions from Tahiti (Bard et al., 1996) and the Huon Peninsula (Papua New Guinea) (Chappell & Polach, 1991) provide information on the deglaciation from 15 kyr onwards. From regional relative sea-level reconstructions the changes in global ice volume mostly cannot be directly inferred because of regional isostatic rebound. Luckily, the region of the Barbados sea-level record of Peltier & Fairbanks (26) was only subject to small isostatic rebound. Therefore, this data set approximates global mean (eustatic) sea level changes (Fig. 1.8d). Unfortunately, the data points are sparse. There is for instance no data during Heinrich event 1, and the data is uncertain during the Bølling-Allerød and the Younger Dryas. The reconstructed regional sea-level variations have been compared to results from models of glacial isostatic adjustment to gain information on deglacial extent and changes of ice sheets (Peltier, 24; Bassett et al., 25). A more detailed review on the last glacial termination is given by Denton et al. (21). 1.4 Earth System Models of Intermediate Complexity Climate models are invaluable tools for climate research and are used in a large variety of studies. In addition to observations of the present-day climate system and to reconstructions of past climate from paleoclimatic records, climate models give the possibility to understand the dynamics of the current and the past climate system on regional to global scales, to analyze hypotheses regarding mechanisms of past climate changes, to project future climate change, and to test new paleoclimatic proxies. Today, a wide range of different types of climate models are available. At the beginning of every modeling study, the appropriate climate model must be chosen. Here, the suite of currently available climate models is introduced and the advantages and disadvantages of the different types are reviewed. Finally, the Bern3D coupled ocean-atmosphere model is positioned within the suite of models. The focus is put on atmosphere and ocean models. Although every climate model is different, atmosphere-ocean models can be separated into different classes depending on the complexity of their atmospheric and ocean component (Fig. 1.9). When setting up a study that for example involves the dynamics of either the ocean or the atmosphere system, the first approach might be to choose a model as complex and as highly resolved as possible in order to most realistically simulate the climate system.

23 1.4. EARTH SYSTEM MODELS OF INTERMEDIATE COMPLEXITY 21 Ocean Dimension OGCM Atmosphere 1 2 Ocean (lat/z) + EBM (lat) Ocean (lat/z) + EBM (lat/lon) / stat. dyn. atm. (lat/z) OGCM + EBM (lat/lon) / stat. dyn. atm. (lat/z) 3 AGCM + SST AGCM + slab ocean AOGCM Figure 1.9: Overview of the different complexities of ocean, atmosphere, and coupled ocean-atmosphere models (Stocker, 29). Here, the model suite is categorized according to the spatial dimensions of the model. The resolved dimensions are given for the one- and two-dimensional models, where lat stands for latitude, lon for longitude, and z for depth. GCM stands for general circulation model, OGCM for Ocean-GCM, AGCM for Atmosphere-GCM, and EBM for energy and moisture balance model. Shaded in gray are the categories of the reduced-complexity models that are also called Earth System Models of Intermediate Complexity (EMICs) when coupled to other model components such as a land vegetation model. Boxes marked with a dash are combinations that are rarely used. Simpler models that are located in the upper triangle of the matrix are not discussed here. Many conceptual models of the early days of climate research belong to these categories. Some of these so called simple climate models are still used today, for example to calculate atmospheric greenhouse gas concentrations following given future emissions, the global mean surface temperature response to a computed radiative forcing, and the global mean sea-level rise (IPCC, 27; Meinshausen et al., 29). In this case, a coupled atmosphere-ocean general circulation model (AOGCM) is chosen, such as for example the NCAR CCSM model (Collins et al., 26) or the HadGEM2 model (Collins et al., 28). AOGCMs consist of three-dimensional dynamical components of the atmosphere and the ocean. Because of the complexity and the high spatial resolution of these models, they are limited to integration times of the order of ten to hundreds of years due to limited computer power capacities. Simulations are usually calculated simultaneously on hundreds of processors in order to reduce the duration of a simulation. As a consequence, often only very few simulations are performed. In a study that focuses on the atmosphere and where the ocean is not relevant, the necessary computer power can be substantially reduced by simulating only the atmosphere by using an Atmosphere-GCM (AGCM). In this case, boundary conditions must be provided. This can be done by prescribing sea-surface temperatures (SSTs) to present-day observations. The time saved per model year can in turn be spent on better resolution, longer simulations or a larger number of simulations. If the model study requires some interaction with the surface ocean, then a slab-ocean can be coupled to the AGCM. The slab-ocean is a two-dimensional model of the surface ocean. It allows air-sea exchange of heat as well as evaporation and precipitation. Horizontal heat transport within the ocean (referred to as Q-flux) is prescribed. The Q- flux of every model cell is determined from an atmosphere-only simulation with prescribed SSTs. It is the residual between the air-sea heat flux and the prescribed temporal SST change. Because the model is only two-dimensional, ocean currents cannot be realistically

24 22 1. INTRODUCTION represented. However, the model is much more computationally efficient compared to an Ocean-GCM (OGCM). In a study that focuses on the ocean and assumes that the atmosphere is playing a minor role, an OGCM is used to reduce computer power requirements. In this case, SSTs are restored to observational data. While this setup is appropriate for simulations of modern climate states, simulations of past climate states are difficult because of the lack of past SST reconstructions. Transient simulations are an even bigger challenge. This limitation can be overcome by introducing a simple two-dimensional energy and moisture balance atmosphere model (EBM). The EBM calculates the radiative transfer of incoming solar radiation through the atmosphere, the effect of greenhouse gases, etc. Also, evaporation and precipitation are parametrized. However, horizontal transport of heat and moisture is limited to diffusion and advection by prescribed winds. Many reduced complexity climate models fall into this category with a two-dimensional atmosphere coupled to a three-dimensional ocean. These models are typically of lower spatial and temporal resolution than AOGCMs and physical processes within the ocean are also simplified as discussed below. The Bern3D coupled ocean-atmosphere model introduced in this thesis (Chapter 2) belongs to this category. Other model examples of this category are the UVic model (Weaver et al., 21), GENIE (Edwards & Marsh, 25), and Climber-3α (Montoya et al., 25). As mentioned above, not all models can be strictly placed within these categories. The ECBilt-CLIO model (referred to as LOVECLIM when other components such as a land vegetation model are included, Goosse et al., 21), for instance, is also a reduced-complexity model but incorporates a simple three-dimensional atmosphere. Even simpler models contain zonally averaged ocean components such as the Bern2.5D model (Stocker et al., 1992) or Climber-2 (Petoukhov et al., 2). To be more precise, note that Climber-2 and Climber-3α consist of a two-dimensional statistical dynamical atmosphere model resolved in latitude and depth instead of an EBM. When reduced complexity models represent the marine carbon cycle and incorporate other components such as a land vegetation model, they are referred to as Earth system models of intermediate complexity (EMICs; Claussen et al., 22). Because of the cost-efficiency of EMICs, many simulations can be performed. Therefore, they are typically applied for process studies, the testing of new hypotheses, the testing of new paleoceanographic tracers, and even probabilistic assessments can be performed using Monte Carlo based methods. Also, simulations on glacial-interglacial cycles, i.e. on of the order of 1, years are possible. To summarize, the complexity of a climate model can be lowered by reducing the dimension of a climate component such as the atmosphere, using a coarser spatial and temporal resolution, or by simplifying equations and parametrizations of the model. In the example of the Bern3D ocean component, frictional geostrophic equations of motion (Winton & Sarachik, 1993) are used instead of primitive equations as typically used in OGCMs (e.g., Smith & Gent, 22). In the frictional geostrophic balance the momentum advection and eddy-viscosity terms are substituted by a simple frictional term which is proportional to the velocity. With this, the system of equations is free of spatial derivatives of the velocity. Therefore, computational costs are considerably reduced when calculating the discretized solutions. Also, the equation of state, i.e. the determination of the ocean density, is approximated from the UNESCO formula (Gill, 1982) using a third order polynomial approximation of Winton & Sarachik (1993). The measures applied to reduce the complexity of the Bern3D ocean component are

25 1.5. THE BERN3D EMIC 23 thoroughly discussed by Müller (27). The user of a reduced complexity climate model always has to be aware of the limitations of the model. For example, with the 5 1 horizontal resolution of the Bern3D model, zonal gradients within ocean basins and the large scale ocean circulation are in principle resolved, but smaller scale phenomena such as locations of deep water formation may deviate from the real world leading to misrepresentations of heat transport and exchange in the northern North Atlantic, etc. As another example, as a consequence of changes in the atmospheric winds, the intertropical convergence zone tends to shift in AOGCM simulations when the AMOC is shut down by a freshwater discharge into the North Atlantic (e.g., Wu et al., 28). In a model where the atmosphere is represented by an EBM, winds are fixed and therefore cannot vary. Thus, changes in precipitation patterns and ocean upwelling will not occur and hence bias the model result. 1.5 The Bern3D EMIC In the past years, the Bern3D model has developed from an ocean-only model (Edwards & Marsh, 25; Müller et al., 26; Müller, 27) to an EMIC. The most important step in this transition is the incorporation of an EBM. The EBM and the present-day state of the coupled ocean-atmosphere model is thoroughly discussed in Chapter 2. However, next to the physical core of the model, various other components have been developed. The marine carbon cycle according to the OCMIP-2 protocol has been implemented (Müller, 27; Müller et al., 28) and extended by adding an iron cycle and prognostic formulations of the production of organic matter, calcium carbonate (CaCO 3 ), and opal shells (Tschumi et al., 28; Parekh et al., 28). A much more complete representation of the marine carbon cycle is given when running the Bern3D model with a marine ecosystem component (Gangstø, 29; Gangstø et al., 211). Instead of the 11 biogeochemical tracers simulated by the standard carbon cycle module (these are dissolved inorganic and organic carbon including their isotopes DIC, DIC- 13, DIC-14, DOC, DOC-13, DOC-14, alkalinity, oxygen, phosphate, silicate, and iron), the ecosystem model consists of 26 tracers including explicit formulations of nanophytoplankton, micro- and mesozooplankton, diatoms, calcite, and aragonite. Because the current version of the marine ecosystem model is very costly compared to the physical core of the model, it is not used as the standard marine carbon cycle component. The Bern3D model further has been coupled to a sediment diagenesis model that permits the representation of long timescale carbon fluxes to the sediment pool (Tschumi, 29; Tschumi et al., 21). Furthermore, the Bern3D model is currently being expanded by coupling the Bern version of the LPJ dynamic land vegetation model of Sitch et al. (23), Strassmann et al. (28), Stocker et al. (211), Wania et al. (29a), and Wania et al. (29b). The LPJ model has a standard spatial resolution of and ten plant functional types. It contains a land use component, i.e. croplands, pasture, and urban areas are distinguished, as well as peatlands and methane modules that allow the representation of permafrost and their effect on climate. Future versions of this land vegetation model will be referred to as the LPX model (LPX stands for Land processes and exchanges, R. Spahni and B. Stocker, personal communication 211). Several EMICs also include a dynamic ice sheet model (e.g., Goosse et al., 21; Ganopolski et al., 21). In the Bern3D model, ice sheets are not explicitly calculated but prescribed as presented in Chapter 2. The Bern3D model includes a number of physical tracers besides the biogeochemical tracers mentioned above. These are the radiocarbon-to-carbon ratio 14 R (Müller et al., 26, Chapter

26 24 1. INTRODUCTION 6), protactinium (Pa) and thorium (Th) (Siddall et al., 27), neodymium isotopes 143 Nd and 144 Nd (Rempfer et al., 211), chlorofluorocarbons CFC-11 and CFC-12 (Müller et al., 26), and noble gases argon (Ar), krypton (Kr), xenon (Xe), as well as the radioactive argon isotope 39 Ar and molecular N 2 (Chapter 4, Müller et al., 26). Finally, an Ensemble Kalman Filter accompanies the Bern3D model to enable inverse modeling studies (Gerber et al., 29; Gerber, 29; Gerber & Joos, 21). Even for an EMIC, the Bern3D model performance is attractive. When running the model on one core of an Intel 2.8 GHz i7 central processing unit (CPU), 1 model years are run in 12 minutes when only the physical core of the coupled ocean-atmosphere model is used, in 5 min when the marine carbon cycle is included, in 6 min when the sediment diagenesis module is included, and in 7 min for the fully coupled version with the land vegetation module. This permits 8-kyr simulations to be run in one week time, or simulations of the last deglaciation of the fully coupled version in one day. For probabilistic assessments of future climate projections, approximately 24 simulations from year 18 to year 24 AD can be run per day on the 72 CPUs of the cluster currently used at the division of Climate and Environmental Physics of the University of Bern. Compared to this, a 1 model year simulation of the NCAR CCSM-3 AOGCM ( atmosphere, 1 1 ocean) processed simultaneously on 252 CPUs requires approximately one month to finish (F. Lehner, personal communication 21). It again needs to be emphasized, however, that the reduced complexity comes at a price: resolution is greatly reduced, processes are simplified or ignored, and many important forcings are prescribed. This simplification must be borne in mind when model results are being interpreted and applied to address problems in climate dynamics.

27 BIBLIOGRAPHY 25 Bibliography Alley, R. B., 2. The Younger Dryas cold interval as viewed from central Greenland, Quaternary Science Reviews, 19, Anderson, R. F., Ali, S., Bradtmiller, L. I., Nielsen, S. H. H., Fleisher, M. Q., Anderson, B. E., & Burckle, L. H., 29. Wind-driven upwelling in the Southern Ocean and the deglacial rise in atmospheric CO 2, Science, 323, Andrews, J. T., The Wisconsin Laurentide ice sheet: dispersal, problem of rate of retreat, and climatic implications, Arctic Alpine Research, 5, Archer, D., Winguth, A., D., L., & Mahowald, N., 2. What caused th glacial/interglacial pco 2 cycles?, Reviews of Geophysics, 38, Bard, E., Hamelin, B., Arnold, M., Montaggioni, L., Cabioch, G., Faure, G., & Rougerie, F., Deglacial sea-level record from Tahiti corals and the timing of global meltwater discharge, Nature, 382, Bassett, S. E., Milne, G. A., Mitrovica, J. X., & Clark, P. U., 25. Ice sheet and solid earth influences on far-field sea-level histories, Science, 39, Berger, A., Li, X. S., & Loutre, M. F., Modelling northern hemisphere ice volume over the last 3 Ma, Quaternary Science Reviews, 18, Berger, A., Gallee, H., Fichefet, T., Marsiat, I., & Tricot, C., 199. Testing the astronomical theory with a coupled climate ice-sheet model, Global and Planetary Change, 3, Berger, A. L., Long-term variations of daily insolation and quaternary climatic changes, Journal of the Atmospheric Sciences, 35, Bond, G., Heinrich, H., Broecker, W., Labeyrie, L., McManus, J., Andrews, J., Huon, S., Jantschik, R., Clasen, S., Simet, C., Tedesco, K., Klas, M., Bonani, G., & Ivy, S., Evidence for massive discharges of icebergs into the North-Atlantic Ocean during the last glacial period, Nature, 36, Broecker, W. S., Paleocean circulation during the Last Deglaciation: A bipolar seesaw?, Paleoceanography, 13, Broecker, W. S., 26. Was the Younger Dryas triggered by a flood?, Science, 312, Broecker, W. S. & van Donk, J., 197. Insolation changes, ice volume, and the δ 18 O record in deep-sea cores, Reviews of Geophysics and Space Physics, 8, Calov, R., Ganopolski, A., Claussen, M., Petoukhov, V., & Greve, R., 25. Transient simulation of the last glacial inception. Part I: glacial inception as a bifurcation in the climate system, Climate Dynamics, 24, Calov, R., Ganopolski, A., Kubatzki, C., & Claussen, M., 29. Mechanisms and time scales of glacial inception simulated with an Earth system model of intermediate complexity, Climate of the Past, 5, Chappell, J. & Polach, H., Post-glacial sea-level rise from a coral record at Huon Peninsula, Papua New Guinea, Nature, 349, Clark, P. U. & Pollard, D., Origin of the Middlee Pleistocene transition by ice sheet erosion of regolith, Paleoceanography, 13, 1 9. Clark, P. U., Archer, D., Pollard, D., Blum, J. D., Rial, J. A., Brovkin, V., Mix, A. C., Pisias, N. G., & Roy, M., 26. The middle Pleistocene transition: characteristics, mechanisms, and implications for long-term changes in atmospheric pco2, Quaternary Science Reviews, 25, Claussen, M., Mysak, L. A., Weaver, A. J., Crucifix, M., Fichefet, T., Loutre, M. F., Weber, S. L., Alcamo, J., Alexeev, V. A., Berger, A., Calov, R., Ganopolski, A., Goosse, H., Lohmann, G., Lunkeit, F., Mokhov, I. I., Petoukhov, V., Stone, P., & Wang, Z., 22. Earth system models of intermediate complexity: closing the gap in the spectrum of climate system models, Climate Dynamics, 18,

28 26 BIBLIOGRAPHY Collins, W. D., Bitz, C. M., Blackmon, M. L., Bonan, G. B., Bretherton, C. S., Carton, J. A., Chang, P., Doney, S. C., Hack, J. J., Henderson, T. B., Kiehl, J. T., Large, W. G., McKenna, D. S., Santer, B. D., & Smith, R. D., 26. The Community Climate System Model version 3 (CCSM3), Journal of Climate, 19, Collins, W. J., Bellouin, N., Doutriaux-Boucher, M., Gedney, N., Hinton, T., Jones, C. D., Liddicoat, S., Martin, G., O Connor, F., Rae, J., Senior, C., Totterdell, I., Woodward, S., Reichler, T., & Kim, J., 28. Evaluation of the HadGEM2 model, Tech. Rep. 74, Met Office Hadley Centre, Exeter, U. K., Available at Crowley, T., North Atlantic Deep Water cools the Southern Hemisphere, Paleoceanography, 7, DeBlonde, G. & Peltier, W. R., A one-dimensional model of continental ice volume fluctuations through the Pleistocene - implications for the origin of the Mid-Pleistocene Climate Transition, Journal of Climate, 4, Denton, G. H. & Hughes, T. J., Milankovitch theory of ice-sheet linkage between regional insolation and global climate, Quaternary Research, 2, Denton, G. H., Anderson, R. F., Toggweiler, J. R., Edwards, R. L., Schaefer, J. M., & Putnam, A. E., 21. The Last Glacial Termination, Science, 328, Edwards, N. R. & Marsh, R., 25. Uncertainties due to transport-parameter sensitivity in an efficient 3-D ocean-climate model, Climate Dynamics, 24, EPICA Community Members, 24. Eight glacial cycles from an Antarctic ice core, Nature, 429, Farrera, I., Harrison, S., Prentice, I., Ramstein, G., Guiot, J., Bartlein, P., Bonnefille, R., Bush, M., Cramer, W., von Grafenstein, U., Holmgren, K., Hooghiemstra, H., Hope, G., Jolly, D., Lauritzen, S., Ono, Y., Pinot, S., Stute, M., & Yu, G., Tropical climates at the Last Glacial Maximum: a new synthesis of terrestrial palaeoclimate data. I. Vegetation, lake levels and geochemistry, Climate Dynamics, 15, Fischer, H., Schmitt, J., Lüthi, D., Stocker, T. F., Tschumi, T., Parekh, P., Joos, F., Köhler, P., Völker, C., Gersonde, R., Barbante, C., Le Floch, M., Raynaud, D., & Wolff, E. W., 21. The role of Southern Ocean processes on orbital and millenial CO 2 variations - A synthesis, Quaternary Science Reviews, 29, Gangstø, R., 29. Increasing atmospheric CO 2, ocean acidification and pelagic ecosystems, Ph.D. thesis, University of Bern. Gangstø, R., Joos, F., & Gehlen, M., 211. Sensitivity of pelagic calcification to ocean acidification, Biogeosciences, 8, Ganopolski, A. & Rahmstorf, S., 21. Rapid changes of glacial climate simulated in a coupled climate model, Nature, 49, Ganopolski, A., Calov, R., & Claussen, M., 21. Simulation of the last glacial cycle with a coupled climate ice-sheet model of intermediate complexity, Climate of the Past, 6, Gerber, M., 29. Assimilating ocean tracer data into the Bern3D model using an ensemble Kalman filter, Ph.D. thesis, University of Bern. Gerber, M. & Joos, F., 21. Carbon sources and sinks from an Ensemble Kalman Filter ocean data assimilation, Global Biogeochemical Cycles, 24, GB34. Gerber, M., Joos, F., Vázquez-Rodriguez, M., Touratier, F., & Goyet, C., 29. Regional air-sea fluxes of anthropogenic carbon inferred with an Ensemble Kalman Filter, Global Biogeochemical Cycles, 23, GB113. Gill, A. E., Atmosphere-Ocean Dynamics, Academic Press, London. Goosse, H., Brovkin, V., Fichefet, T., Haarsma, R., Huybrechts, P., Jongma, J., Mouchet, A., Selten, F., Barriat, P.-Y., Campin, J.-M., Deleersnijder, E., Driesschaert, E., Goelzer, H., Janssens, I., Loutre, M.-F., Morales Maqueda, M. A., Opsteegh, T., Mathieu, P.-P., Munhoven, G., Pettersson, E. J., Renssen, H., Roche, D. M., Schaeffer, M., Tartinville, B., Timmermann, A., & Weber, S. L., 21. Description of the Earth system model of intermediate complexity LOVECLIM version 1.2, Geoscientific Model Development, 3,

29 BIBLIOGRAPHY 27 Hanebuth, T., Stattegger, K., & Grootes, P. M., 2. Rapid flooding of the Sunda Shelf: A late-glacial sea-level record, Science, 288, Hanel, R. A., Conrath, B. J., Kunde, V. G., Prabhakara, C., Revah, I., Salomonson, V. V., & Wolford, G., The Numbus 4 infrared spectroscopy experiment 1. Calibrated thermal emission spectra, Journal of Geophysical Research, 77, Hartmann, D. L., Global Physical Climatology, Academic Press, University of Washington. Hays, J. D., Imbrie, J., & Shackleton, N. J., Variations in the Earth s orbit: Pacemaker of the ice ages, Science, 194, Hemming, S. R., 24. Heinrich events: Massive late Pleistocene detritus layers of the North Atlantic and their global climate imprint, Reviews of Geophysics, 42, RG15. Hönisch, B., Hemming, N. G., Archer, D., Siddall, M., & McManus, J. F., 29. Atmospheric carbon dioxide concentration across the mid-pleistocene Transition, Science, 324, Huybers, P., 26. Early Pleistocene glacial cycles and the integrated summer insolation forcing, Science, 313, Huybers, P., 29. Pleistocene glacial variability as a chaotic response to obliquity forcing, Climate of the Past, 5, Huybers, P. & Denton, G., 28. Antarctic temperature at orbital timescales controlled by local summer duration, Nature Geoscience, 1, Imbrie, J., Berger, A., Boyle, E. A., Clemens, S. C., Duffy, A., Howard, W. R., Kukla, G., Kutzbach, J., Martinson, D. G., McIntyre, A., Mix, A. C., Molfino, B., Morley, J. J., Peterson, L. C., Pisias, N. G., Prell, W. L., Raymo, M. E., Shackleton, N. J., & Toggweiler, J. R., On the structure and origin of major glaciation cycles. 2. The 1,-year cycle, Paleoceanography, 8, IPCC, 27. Climate Change 27: the physical science basis. Contribution of working group I to the fourth assessment report of the intergovernmental panel on climate change, Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. Jansen, E., Overpeck, J., Briffa, K. R., Duplessy, J.-C., Joos, F., Masson-Delmotte, V., Olago, D., Otto- Bliesner, B., Peltier, W. R., Rahmstorf, S., Ramesh, R., Raynaud, D., Rind, D., Solomina, O., Villalba, R., & Zhang, D., 27. Palaeoclimate, in Climate Change 27: The Physical Science Basis. Working Group I Contribution to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pp , eds Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tignor, M., & Miller, H. L., Cambridge University Press. Jouzel, J., Masson-Delmotte, V., Cattani, O., Dreyfus, G., Falourd, S., Hoffmann, G., Minster, B., Nouet, J., Barnola, J. M., Chappellaz, J., Fischer, H., Gallet, J. C., Johnsen, S., Leuenberger, M., Loulergue, L., Luethi, D., Oerter, H., Parrenin, F., Raisbeck, G., Raynaud, D., Schilt, A., Schwander, J., Selmo, E., Souchez, R., Spahni, R., Stauffer, B., Steffensen, J. P., Stenni, B., Stocker, T. F., Tison, J. L., Werner, M., & Wolff, E. W., 27. Orbital and millennial Antarctic climate variability over the past 8, years, Science, 317, Kawamura, K., Parrenin, F., Lisiecki, L., Uemura, R., Vimeux, F., Severinghaus, J. P., Hutterli, M. A., Nakazawa, T., Aoki, S., Jouzel, J., Raymo, M. E., Matsumoto, K., Nakata, H., Motoyama, H., Fujita, S., Goto-Azuma, K., Fujii, Y., & Watanabe, O., 27. Northern Hemisphere forcing of climatic cycles in Antarctica over the past 36, years, Nature, 448, Kiehl, J. T. & Trenberth, K., Earth s annual global mean energy budget, Bulletin of the American Meteorological Society, 78, Knorr, G. & Lohmann, G., 23. Southern Ocean origin for the resumption of Atlantic thermohaline circulation during deglaciation, Nature, 424, Köhler, P. & Fischer, H., 24. Simulating changes in the terrestrial biosphere during the last glacial/interglacial transition, Global and Planetary Change, 43,

30 28 BIBLIOGRAPHY Kopp, G. & Lean, J. L., 211. A new, lower value of total solar irradiance: Evidence and climate significance, Geophysical Research Letters, 38. Kubatzki, C., Claussen, M., Calov, R., & Ganopolski, A., 26. Sensitivity of the last glacial inception to initial and surface conditions, Climate Dynamics, 27, Lambert, F., Delmonte, B., Petit, J. R., Bigler, M., Kaufmann, P. R., Hutterli, M. A., Stocker, T. F., Ruth, U., Steffensen, J. P., & Maggi, V., 28. Dust-climate couplings over the past 8, years from the EPICA Dome C ice core, Nature, 452, Lawrence, K. T., Sosdian, S., White, H. E., & Rosenthal, Y., 21. North Atlantic climate evolution through the Plio-Pleistocene climate transitions, Earth and Planetary Science Letters, 3, Lee, S.-Y., Chiang, J. C. H., Matsumoto, K., & Tokos, K. S., 211. Southern Ocean wind response to North Atlantic cooling and the rise in atmospheric CO 2: Modeling perspective and paleoceanographic implications, Paleoceanography, 26, PA1214. Lisiecki, L. E. & Raymo, M. E., 25. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ 18 O records, Paleoceanography, 2, PA13. Liu, Z., Otto-Bliesner, B. L., He, F., Brady, E. C., Tomas, R., Clark, P. U., Carlson, A. E., Lynch-Stieglitz, J., Curry, W., Brook, E., Erickson, D., Jacob, R., Kutzbach, J., & Cheng, J., 29. Transient simulation of last deglaciation with a new mechanism for Bølling-Allerød warming, Science, 325, Liu, Z. H. & Herbert, T. D., 24. High-latitude influence on the eastern equatorial Pacific climate in the early Pleistocene epoch, Nature, 427, Liu, Z. H., Altabet, M. A., & Herbert, T. D., 25. Glacial-interglacial modulation of eastern tropical North Pacific denitrification over the last 1.8-Myr, Geophysical Research Letters, 32, L2367. Loulergue, L., Schilt, A., Spahni, R., Masson-Delmotte, V., Blunier, T., Lemieux, B., Barnola, J.-M., Raynaud, D., Stocker, T. F., & Chappellaz, J., 28. Orbital and millennial-scale features of atmospheric CH 4 over the past 8, years, Nature, 453, Lüthi, D., Le Floch, M., Bereiter, B., Blunier, T., Barnola, J. M., Siegenthaler, U., Raynaud, D., Jouzel, J., Fischer, H., Kawamura, K., & Stocker, T. F., 28. High resolution carbon dioxide concentration record 65, 8, years before present, Nature, 453, Maier-Reimer, E. & Mikolajewicz, U., Experiments with an OGCM on the cause of the Younger Dryas, Tech. Rep. 39, Max-Planck-Institut für Meteorologie, Hamburg, Germany. McManus, J. F., Francois, R., Gherardi, J.-M., & Keigwin, L. D., 24. Collapse and rapid resumption of Atlantic meridional circulation linked to deglacial climate changes, Nature, 428, Meinshausen, M., Meinshausen, N., Hare, W., Raper, S. C. B., Frieler, K., Knutti, R., Frame, D. J., & Allen, M. R., 29. Greenhouse-gas emission targets for limiting global warming to 2 C, Nature, 458, Milankovitch, M., Kanon der Erdbestrahlung und seine Anwendung auf das Eiszeitenproblem, Royal Serbian Academy Special Publication 133, (English translation published in 1969 by Israel Program for Scientific Translations, US Dept. Comm.). Monnin, E., Indermühle, A., Dällenbach, E., Flückiger, J., Stauffer, B., Stocker, T. F., Raynaud, D., & Barnola, J.-M., 21. Atmospheric CO 2 concentrations over the Last Glacial Termination, Science, 291, Montoya, M., Griesel, A., A., L., Mignot, J., Hofmann, M., Ganopolski, A., & Rahmstorf, S., 25. The earth system model of intermediate complexity CLIMBER-3α: Part I: description and performance for present-day conditions, Climate Dynamics, 25, Müller, S. A., 27. Large-scale ocean circulation, air-sea gas exchange, and carbon isotopes in a threedimensional, computationally efficient ocean model, Ph.D. thesis, University of Bern, Bern, Switzerland. Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 26. Water mass distribution and ventilation time scales in a cost-efficient, three-dimensional ocean model, Journal of Climate, 19,

31 BIBLIOGRAPHY 29 Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 28. Modeled natural and excess radiocarbon - sensitivities to the gas exchange formulation and ocean transport strength, Global Biogeochemical Cycles, 22, GB311. NGRIP members, 24. High-resolution record of Northern Hemisphere climate extending into the last interglacial period, Nature, 431, Oerlemans, J., 198. Model experiments on the 1,-year glacial cycle, Nature, 287, Otieno, F. O. & Bromwich, D. H., 29. Contribution of atmospheric circulation to inception of the Laurentide Ice Sheet at 116 kyr BP, Journal of Climate, 22, Paillard, D., The timing of Pleistocene glaciations from a simple multiple-state climate model, Nature, 391, Parekh, P., Joos, F., & Müller, S. A., 28. A modeling assessment of the interplay between aeolian iron fluxes and iron-binding ligands in controlling carbon dioxide fluctuations during Antarctic warm events, Paleoceanography, 23, PA422. Peixoto, J. P. & Oort, A. H., Physics of Climate, Springer, New York. Peltier, W. R., 22. On eustatic sea level history: Last Glacial Maximum to Holocene, Quaternary Science Reviews, 21, Peltier, W. R., 24. Global glacial isostasy and the surface of the ice-age earth: The ICE-5G (VM2) model and GRACE, Annual Review of Earth and Planetary Sciences, 32, Peltier, W. R. & Fairbanks, R. G., 26. Global glacial ice volume and Last Glacial Maximum duration from an extended Barbados sea level record, Quaternary Science Reviews, 25, Petit, J. R., Jouzel, J., Raynaud, D., Barkov, N. I., Barnola, J. M., Basile, I., Bender, M., Chappellaz, J., Davis, M., Delaygue, G., Delmotte, M., Kotlyakov, V. M., Legrand, M., Lipenkov, V. Y., Lorius, C., Pepin, L., Ritz, C., Saltzman, E., & Stievenard, M., Climate and atmospheric history of the past 42, years from the Vostok ice core, Antarctica, Nature, 399, Petoukhov, V., Ganopolski, A., Brovkin, V., Claussen, M., Eliseev, A., Kubatzki, C., & Rahmstorf, S., 2. CLIMBER-2: a climate system model of intermediate complexity. Part I: model description and performance for present climate, Climate Dynamics, 16, Rahmstorf, S., Rapid climate transitions in a coupled ocean-atmosphere model, Nature, 372, Rasmussen, S. O., Andersen, K. K., Svensson, A. M., Steffensen, J. P., Vinther, B. M., Clausen, H. B., Siggaard-Andersen, M. L., Johnsen, S. J., Larsen, L. B., Dahl-Jensen, D., Bigler, M., Röthlisberger, R., Fischer, H., Goto-Azuma, K., Hansson, M. E., & Ruth, U., 26. A new Greenland ice core chronology for the last glacial termination, Journal of Geophysical Research, 111, D612. Raymo, M. E., The timing of major climate transitions, Paleoceanography, 12, Raymo, M. E. & Nisancioglu, K., 23. The 41 kyr world: Milankovitch s other unsolved mystery, Paleoceanography, 18, 111. Raymo, M. E., Lisiecki, L. E., & Nisancioglu, K. H., 26. Plio-Pleistocene ice volume, antarctic climate, and the global δ 18 o record, Science, 313, Rempfer, J., Stocker, T. F., Joos, F., Dutay, J.-C., & Siddall, M., 211. Modelling isotopes of neodymium with a coarse resolution ocean circulation model: Sensitivities to model parameters and source/sink distributions, Geochimica et Cosmochimica Acta, submitted. Sarmiento, J. L. & Gruber, N., 26. Ocean biogeochemical dynamics, Princeton University Press, Princeton and Oxford. Schilt, A., Baumgartner, M., Blunier, T., Schwander, J., Spahni, R., Fischer, H., & Stocker, T. F., 21a. Glacial-interglacial and millennial-scale variations in the atmospheric nitrous oxide concentration during the last 8, years, Quaternary Science Reviews, 29,

32 3 BIBLIOGRAPHY Schilt, A., Baumgartner, M., Schwander, J., Buiron, D., Capron, E., Chappellaz, J., Loulergue, L., Schüpbach, S., Spahni, R., Fischer, H., & Stocker, T. F., 21b. Atmospheric nitrous oxide during the last 14, years, Earth and Planetary Science Letters, 3, Schulz, K. G. & Zeebe, R. E., 26. Pleistocene glacial terminations triggered by synchronous changes in Southern and Northern Hemisphere insolation: The insolation canon hypothesis, Earth and Planetary Science Letters, 249, Shackleton, N. J., New data on the evolution of Pliocene climate variability, in Paleoclimate and evolution, with emphasis on human origins, pp , eds Vrba, E. S., Denton, G. H., Partridge, T. C., & Burckle, L. H., Yale University Press, New Haven, Ct. Siddall, M., Stocker, T. F., Henderson, G. M., Joos, F., Frank, M., Edwards, N. R., Ritz, S. P., & Müller, S. A., 27. Modeling the relationship between 231 Pa/ 23 Th distribution in North Atlantic sediment and Atlantic meridional overturning circulation, Paleoceanography, 22, PA2214. Sigman, D. M. & Boyle, E. A., 2. Glacial/interglacial variations in atmospheric carbon dioxide, Nature, 47, Sigman, D. M., Hain, M. P., & Haug, G. H., 21. The polar ocean and glacial cycles in atmospheric CO 2 concentration, Nature, 466, Sitch, S., Smith, B., Prentice, I. C., Arneth, A., Bondeau, A., Cramer, W., Kaplan, J. O., Levis, S., Lucht, W., Sykes, M. T., Thonicke, K., & Venevsky, S., 23. Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model, Global Change Biology, 9, Smith, R. D. & Gent, P. R., 22. Reference manual for the Parallel Ocean Program (POP), ocean component of the Community Climate System Model (CCSM2. and 3.), Tech. rep., LA-UR , Los Alamos National Labratory, Los Alamos, NM, Stocker, B. D., Strassmann, K., & Joos, F., 211. Sensitivity of Holocene atmospheric CO 2 and the modern carbon budget to early human land use: analyses with a process-based model, Biogeosciences, 8, Stocker, T. F., 2. Past and future reorganizations in the climate system, Quaternary Science Reviews, 19, Stocker, T. F., 29. Introduction to climate modelling, Physics Institute, University of Bern, 148 pp. Stocker, T. F. & Johnson, S. J., 23. A minimum thermodynamic model for the bipolar seesaw, Paleoceanography, 18, 187. Stocker, T. F. & Marchal, O., 2. Abrupt climate change in the computer: Is it real?, PNAS, 97, Stocker, T. F., Wright, D. G., & Mysak, L. A., A zonally averaged, coupled ocean-atmosphere model for paleoclimate studies, Journal of Climate, 5, Stott, L., Timmermann, A., & Thunell, R., 27. Southern Hemisphere and deep-sea warming led deglacial atmospheric CO 2 rise and tropical warming, Science, 318, Strassmann, K. M., Joos, F., & Fischer, G., 28. Simulating effects of land use changes on carbon fluxes: past contributions to atmospheric CO 2 increases and future commitments due to losses of terrestrial sink capacity, Tellus B, 6, Tegen, I., 23. Modeling the mineral dust aerosol cycle in the climate system, Quaternary Science Reviews, 22, Timmermann, A., Knies, J., Timm, O. E., Abe-Ouchi, A., & Friedrich, T., 21. Promotion of glacial ice sheet buildup kyr B.P. by precessionally paced Northern Hemispheric meltwater pulses, Paleoceanography, 25, PA428. Tschumi, T., 29. Modeling the Ocean s Contribution to Past and Future Changes in Global Carbon Cycling, Ph.D. thesis, University of Bern. Tschumi, T., Joos, F., & Parekh, P., 28. How important are Southern Hemisphere wind changes for low glacial carbon dioxide? A model study, Paleoceanography, 23, PA428.

33 BIBLIOGRAPHY 31 Tschumi, T., Joos, F., Gehlen, M., & Heinze, C., 21. Deep ocean ventilation, carbon isotopes, marine sedimentation and the deglacial CO 2 rise, Climate of the Past Discussion, 6, Vavrus, S., Philippon-Berthier, G., Kutzbach, J. E., & Ruddiman, W. F., 211. The role of GCM resolution in simulating glacial inception, The Holocene,, in press, doi:1.1177/ Vettoretti, G. & Peltier, W. R., 24. Sensitivity of glacial inception to orbital and greenhouse gas climate forcing, Quaternary Science Reviews, 23, Wang, Z. M. & Mysak, L. A., 22. Simulation of the last glacial inception and rapid ice sheet growth in the McGill Paleoclimate Model, Geophysical Research Letters, 29, 212. Wang, Z. M., Cochelin, A. S. B., Mysak, L. A., & Wang, Y., 25. Simulation of the last glacial inception with the green McGill Paleoclimate Model, Geophysical Research Letters, 32, L1275. Wania, R., Ross, I., & Prentice, I. C., 29a. Integrating peatlands and permafrost into a dynamic global vegetation model: 1. Evaluation and sensitivity of physical land surface processes, Global Biogeochemical Cycles, 23. Wania, R., Ross, I., & Prentice, I. C., 29b. Integrating peatlands and permafrost into a dynamic global vegetation model: 2. Evaluation and sensitivity of vegetation and carbon cycle processes, Global Biogeochemical Cycles, 23. Weaver, A. J., Eby, M., Wiebe, E. C., Bitz, C. M., Duffy, P. B., Ewen, T. L., Fanning, A. F., Holland, M. M., MacFadyen, A., Matthews, H. D., Meissner, K. J., Saenko, O., Schmittner, A., Wang, H., & Yoshimori, M., 21. The UVic Earth System Climate Model: Model description, climatology, and application to past, present and future climates, Atmosphere-Ocean, 39, Weaver, A. J., Saenko, O. A., Clark, P. U., & Mitrovica, J. X., 23. Meltwater pulse 1A from Antarctica as a trigger of the Bølling-Allerød warm interval, Science, 299, Winton, M. & Sarachik, E. S., Thermohaline oscillations induced by strong steady salinity forcing of ocean general-circulation models, Journal of Physical Oceanography, 23(7), Wu, L., Li, C., Yang, C., & Xie, S.-P., 28. Global teleconnections in response to a shutdown of the Atlantic meridional overturning circulation, Journal of Climate, 21, Yokoyama, Y., Lambeck, K., De Deckker, P., Johnston, P., & Fifield, L. K., 2. Timing of the Last Glacial Maximum from observed sea-level minima, Nature, 46, Yoshimori, M., Reader, M. C., Weaver, A. J., & McFarlane, N. A., 22. On the causes of glacial inception at 116 kabp, Climate Dynamics, 18,

34

35 Chapter 2 The Bern3D Energy and Moisture Balance Atmosphere 2.1 A Coupled Dynamical Ocean Energy Balance Atmosphere Model for Paleoclimate Studies Stefan P. Ritz, Thomas F. Stocker, and Fortunat Joos Published in Journal of Climate, Volume 24, pp , 211.

36 34 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 349 A Coupled Dynamical Ocean Energy Balance Atmosphere Model for Paleoclimate Studies STEFAN P. RITZ, THOMAS F. STOCKER, AND FORTUNAT JOOS Climate and Environmental Physics, Physics Institute, and Oeschger Centre for Climate Change Research, University of Bern, Bern, Switzerland (Manuscript received 21 July 29, in final form 13 September 21) ABSTRACT The Bern3D coupled three-dimensional dynamical ocean energy balance atmosphere model is introduced and the atmospheric component is discussed in detail. The model is of reduced complexity, developed to perform extensive sensitivity studies and ensemble simulations extending over several glacial interglacial cycles. On large space scales, the modern steady state of the model compares well with observations. In a first application, several 8 -yr simulations with prescribed orbital, greenhouse gas, and ice sheet forcings are performed. The model shows an increase of Atlantic meridional overturning circulation strength at glacial inceptions followed by a decrease throughout the glaciation and ending in a circulation at glacial maxima that is weaker than at present. The sensitivity of ocean temperature to atmospheric temperature, Atlantic meridional overturning circulation (AMOC), and Antarctic bottom water (AABW) strength is analyzed at 23 locations. In a second application the climate sensitivities of the modern and of the Last Glacial Maximum (LGM) state are compared. The temperature rise for a doubling of the CO 2 concentration from LGM conditions is 4.38C and thus notably larger than in the modern case (38C). The relaxation time scale is strongly dependent on the response of AABW to the CO 2 change, since it determines the ventilation of the deep Pacific and Indian Ocean. 1. Introduction With increasing computer power, the realism of climate models could be increased and thus models have become more complex. Spatial resolution has been refined to better resolve small-scale phenomena, and more processes have been included. Nonetheless, computer power still limits the most complex type of models, the coupled atmosphere ocean general circulation models (AOGCMs), from permitting simulations exceeding a few hundred years [e.g., the National Center for Atmospheric Research (NCAR) Community Climate System Model (CCSM; Collins et al. 26) and third climate configuration of the Met Office Unified Model (HadCM3; Gordon et al. 2)]. At the time of writing, Liu et al. (29) have presented the longest simulations with an AOGCM of almost 8 kyr (1 kyr 5 1 yr) using CCSM3. Otherwise, for simulations on a multimillenial time scale, so-called earth system models of intermediate Corresponding author address: Stefan P. Ritz, University of Bern, Physics Institute, Climate and Environmental Physics, Sidlerstr. 5, Bern 312, Switzerland. ritz@climate.unibe.ch complexity (EMICs) have been built. They are typically based on simplified physics and parameterizations of a larger number of processes [e.g., the Bern2.5D global model (Stocker et al. 1992), the University of Victoria (UVic) earth system climate model (Weaver et al. 21), ECBilt-CLIO (Goosse et al. 25), the Climate and Bisphere Model version 3a (Climber3a) (Montoya et al. 25), and the Grid Enabled Integrated Earth system model (GENIE) (Edwards and Marsh 25)]. To the present day, only few model simulations have been done spanning more than one glacial cycle of approximately 1 years. Those studies used models with zonally averaged ocean basins (Tuenter et al. 25) or with a three-dimensional atmosphere but a slab ocean component and accelerated variations of the orbital configuration (Jackson and Broccoli 23). Here we present a very cost-efficient coupled three-dimensional dynamical ocean energy balance atmosphere intermediate complexity model, which permits simulations on glacial-to-interglacial time scales. Currently 5 model years per day can be run on a single personal computer CPU. With this, the model is considerably more efficient than most three-dimensional EMICs. Thus, extensive long time-scale parameter sensitivity studies or ensemble

37 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 simulations become feasible. Because the model also includes a prognostic formulation of the carbon cycle and a palette of other tracers, it is a powerful tool for comprehensive paleoceanographic model studies. In this paper, we introduce the atmospheric component of the coupled model in detail, present the modern steady state of the coupled ocean atmosphere model, and perform several coupled 8 -yr simulations with prescribed orbital, greenhouse gas, and ice sheet forcings. 2. Model description a. The ocean model component The Bern3D ocean component is a seasonally forced three-dimensional frictional geostrophic global ocean model with coarse spatial resolution. It consists of grid boxes in the horizontal direction and 32 vertical layers. The year is divided into 48 time steps, corresponding to about a week per time step. It is based on the ocean model of Edwards et al. (1998) and described in detail by Müller et al. (26). A new feature is the possibility of barotropic flow around the American continent and Australia. In the modern control state, there is.5 Sv (1 Sv m 3 s 21 ) northward flow through the Bering Strait and 23 Sv Indonesian Throughflow from the Pacific to the Indian Ocean. In ocean-only simulations, the model is run under restoring surface boundary conditions for temperature and salinity. Temperature fields are taken from Levitus and Boyer (1994) and salinity fields from Levitus et al. (1994). The model also contains a prognostic carbon cycle (Parekh et al. 28; Tschumi et al. 28). The Bern3D ocean component has been used for a range of applications (Gerber and Joos 21; Gerber et al. 29; Ritz et al. 28; Parekh et al. 28; Tschumi et al. 28; Müller et al. 28; Siddall et al. 27; Muscheler et al. 27). b. The energy balance model of the atmosphere The single-layer energy balance is described in spherical coordinates using u 2 f; 2pg for the longitude and q 2 f2p/2; p/2g for the latitude and is similar to the model described by Weaver et al. (21). The spatial and temporal resolutions are equal to the resolution of the ocean model. Notation and values of the model parameters are given in Table 1. Depth-integrated horizontal heat fluxes are parameterized in terms of eddy-diffusive fluxes with uniform zonal, K u, and meridional, K q, diffusivities. The vertical energy fluxes consist of shortwave (sw) and longwave (lw) fluxes at the top of the atmosphere (TOA) and across the atmosphere ocean (AO), atmosphere sea ice (AI), and atmosphere land (AL) boundaries, respectively: h a r a c p,a t T a 5 h a r a c p,a 1 r 2 cos 2 q u K u u T a 1 (sinq) cos2 qk q (sinq) T a 8 >< F AO total 1 L y r o (P EAO ); over ocean 1 F AI total 1 L s r o (P EAI ); over sea ice >: F AL total 1 L y r P; over land, o 1 F TOA sw F up sw FTOA lw (1) where T a is the atmospheric temperature. For model stability reasons, horizontal advective transport is not taken into account. The fluxes F AO total, FAI total, and FAL total at the bottom of the atmosphere (BOA) are the heat gains of the ice-free ocean, sea ice, and land surface, respectively. They are separated into a shortwave and longwave radiation term, a sensible heat flux, and a latent heat flux term: F AO total 5 FBOA sw F AI total 5 FBOA sw s«o T 4 o 1 s«a T4 a 1 FAO sh L y r o E AO, (2) s«o T 4 i 1 s«a T4 a 1 FAI sh L s r o EAI (3) (the emissivities of water and ice are very similar for infrared wavelengths), and F AL total 5 FBOA sw s«l T 4 l 1 s«a T4 a 1 FAL sh, (4) where T o is the surface ocean temperature, T i the surface sea ice temperature, and T l the surface temperature over land. Note that since evaporation is included in and in FAI total, it needs to be subtracted in Eq. (1). In this version, water is not stored on land, and therefore evaporation is zero on land boxes. Land temperatures are calculated by solving F AO total r l c p,l h l T l t 5 F AL total. (5) The land surface scale height h l is chosen to be 2 m. This corresponds to the depth to which temperature is affected by seasonality (Hartmann 1994). For reasons

38 36 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 351 TABLE 1. Parameter values for energy and moisture transport in the atmosphere and for the sea ice model. Parameter Value Description h a 8194 m Atmospheric scale height for temperature h q 18 m Moisture scale height h l 2 m Land surface scale height r a 1.25 kg m 23 Reference density of air r o 1 kg m 23 Reference density of water r l 2 kg m 23 Reference land density r i 913 kg m 23 Reference density of sea ice c p,a 14 J kg 21 K 21 Specific heat of air c p,l 148 J kg 21 K 21 Land reference specific heat capacity c p,o 444 J kg 21 K 21 Specific heat of seawater under ice r m Radius of the earth s W m 22 K 24 Stefan Boltzmann constant «a cos 2 q Atmospheric emissivity «o.96 Ocean emissivity «l.95 Reference land emissivity (sandy, saturated soil) S 1353 W m 22 Solar constant r h,max.85 Max. relative humidity r h,precip.7 Relative humidity after precipitation K u m 2 s 21 Zonal eddy diffusivity K q (1 1 q 1 p p/2 ) Meridional eddy diffusivity cos 2 q m 2 s 21 q K u m 2 s 21 Zonal eddy diffusivity for moisture q K q m 2 s 21 Meridional eddy diffusivity for moisture l 1 W m 22 K 21 Water vapor feedback parameter DT ct max 88C Temperature reduction at cloud top when j 5 1 K i 1 4 m 2 s 21 Sea ice diffusion coefficient D l 3 W m 22 K 21 Bulk coefficient for sensible heat on land L y J kg 21 Latent heat of evaporation L s J kg 21 Latent heat of sublimation L f J kg 21 Latent heat of fusion of ice c h.58 Empirical constant u t.15 m s 21 Skin friction velocity at ice ocean boundary I cond W m 21 K 21 Thermal conductivity of ice H.1 m Minimal ice thickness x Continental values for fractional runoff; for Africa x 5.16, for Antarctica x 5.83 of simplicity, except for Antarctica (AA), the parameters for the land surface are chosen to be global and correspond to sandy, saturated soil (Martin 22). For Antarctica, c AA p,l 5 21 J kg 1 K 21. These choices help to decrease atmospheric temperature seasonality in this region. Incoming solar radiation F TOA sw is calculated following the algorithm of Berger (1978). A parameterization of Bintanja (1996) determines how much of the incoming radiation is transmitted through the atmosphere to the bottom of the atmosphere (F BOA sw ), how much of it is reflected back into space (F up sw ), and how much is absorbed by the atmosphere: F BOA sw 5 [1 a(u, q, n)] [1 j(q, n)]f down cl 1 jf down ov, (6) where F down cl and F down ov are the radiation fluxes transmitted through the atmosphere for clear-sky and overcast conditions, respectively; j(q, n) denotes the zonally averaged fractional cloud amount climatology taken from the 4-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-4), with n being the model time step within a year; and a(u, q, n) denotes the surface albedo. Analogously, F up sw 5 [1 j(q, n)]fup cl 1 jf up ov. (7) Here F down cl, F down ov, F up cl, and Fup ov include approximations for the absorptive and reflective properties of the atmospheric constituents, the solar zenith angle and the surface elevation [taken from 5-minute gridded elevations/bathymetry for the world (ETOPO5); see global/etopo5.html] and are calculated as described by Bintanja (1996). However, the following change has been made: In contrast to Bintanja (1996), cloud optical depth t, a measure of cloud transparency, is not set to a constant value but rather depends on the liquid water path,

39 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 W 5 q a r a h q, the integrated amount of water in the atmospheric column: log 1 (t) log e [log 1 (W/W )], (8) with W 5 1 kg m 22 (Stephens 1978); q a is the surface specific humidity. It satisfies a balance equation (see below). Following Weaver et al. (21), the parameterization for outgoing planetary infrared irradiance for clear-sky conditions at TOA of Thompson and Warren (1982) is used and extended by a parameterization for the radiative forcing owing to deviations of atmospheric CO 2 concentrations from a reference value. Additionally, a simplified term for CH 4 greenhouse gas forcing is added, as well as a term representing the water vapor feedback: F TOA lw 5 a 1 1 a 2 T a 1 a 3 T 2 a 1 a 4 T3 a DF ln pco 2 (t) 23CO 2 pco 2, q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi y pch 4 (t)ppb 1 pch 4, ppb 1 ldt a, (9) where pco 2, ppm is the preindustrial atmospheric carbon dioxide concentration, DF 23CO W m 22 (Myrhe et al. 1998), pch 4, 5 7 ppb is the preindustrial atmospheric methane concentration, and y 5.36 W m 22 (Shi 1992). The coefficients a i depend on the relative humidity r h according to a i 5 b 1, i 1 b 2,i r h 1 b 3,i r 2 h. (1) The coefficients b j,i are taken from Thompson and Warren (1982); r h 5 q a /q s (T a ), where q s (T a ) is the saturation specific humidity (explained below). The last term of Eq. (9) is a very simple approximation for the water vapor feedback. We define the seasonally dependent temperature deviation from the modern control state DT a 5 T a (n, y) T CTRL a (n), where T a (n, y) is the global mean atmospheric temperature at model time step n of year y; T CTRL a is the temperature average of a 5-kyr control run. The feedback parameter l is tuned (l 5 1 W m 22 K 21 ) to produce an equilibrium climate sensitivity (global temperature rise for a doubling of the atmospheric CO 2 content) of 38C for a modern steady state. It is found that the effect of clouds on the longwave radiation needs to be accounted for, especially at high latitudes, where cloud cover is high compared to the global mean value. Thus, the following simple parameterization is implemented into the model: Since for the presence of clouds the location of emitted longwave radiation is the cloud-top level instead of the earth s surface, the graybody radiation temperature is lower and thus also the outgoing radiation (Hartmann 1994). Therefore, we reduce the graybody radiation temperature of Eq. (9) to T ct a 5 T j DTct a max, (11) depending on fractional cloud cover j. Here T ct a expresses the temperature at cloud top and DT ct max the temperature reduction when j 5 1; DT ct max 5 88C is chosen such that a reasonable global atmospheric temperature is obtained. Following Weaver et al. (21), evaporation at the ocean surface E AO is calculated according to E AO 5 r a C E juj r o [q s (T o ) q a ], (12) where juj is the surface wind speed at 1-m height from ERA-4 reanalysis, and C E 5 C E (u, q, n) is the Dalton number. It is diagnosed during a 5-yr initialization run by solving Eq. (12) for C E and using monthly evaporation fields from ERA-4 reanalysis as well as a T a climatology from ERA-4 and ocean temperatures from the ocean-only simulation. As proposed by Isemer et al. (1989), # C E # Also, q s (T o ) is the saturation specific humidity at the ocean surface (g water per kg air). It is calculated using the parameterization of Bolton (198): q s (T s ) 5 c 1 exp c 2 (T 273:15 K) s, (13) c 3 1 T s 273:15 K where c g kg 21, c , and c K; T s is either ocean temperature T o or the atmospheric temperature T a. Sublimation over sea ice E AI is parameterized as evaporation [Eq. (12)], except for replacing surface ocean saturation specific humidity q s (T o ) by sea ice surface saturation specific humidity q s (T i ) 5 c 1 exp c 4 (T 273:15 K) i, (14) c 5 1 T i 273:15 K where c g kg 21, c , and c K. Precipitation occurs when relative humidity rises above a maximum value r h,max. Precipitation stops when r h is equal to r h,precip : 8 < r a h a P 5 r o Dt [q a r h,precip q s (T a )], r h. r h,max :, otherwise. (15)

40 38 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 353 A fraction of the precipitation over land is instantly transported to the ocean as runoff. Its pathway is defined by a pseudoelevation map: Each land box contains information about the direction of runoff (north, south, west, or east). This fraction x is taken from observations (Hartmann 1994) and is different for every continent. It ranges from.33 to.43, except for Africa, where reevaporation is high (x 5.16), and Antarctica, where the ice sheet is in broad equilibrium and the mass increase by snowfall is compensated by snowmelt and iceberg calving at the ocean margin (x 5.83). The rest, which in the real world would be stored on land or reevaporated to the atmosphere, is distributed uniformly into every surface ocean box. Note that it would physically make more sense to redirect (1 2 x) P back to the atmosphere, since this is the fraction that is taken up by the soils and eventually reevaporated. However, because of the coarse resolution and the parameterization of the model, a large part of the reevaporated moisture would be precipitated again in the following time step. This would strongly increase the atmosphere ocean moisture turnover and thus the amount of runoff. Evaporation and precipitation in m s 21 are converted into a latent heat flux by multiplying the reference density of water r o and the latent heat of evaporation L y. At the atmosphere ocean and atmosphere sea ice interface the parameterization of Weaver et al. (21) for the sensible heat flux is used: Fsh AO/AI 5 r a C H c p,a juj(t a T s ), (16) where C H 5.94C E is the Stanton number and T s is either surface ocean or sea ice temperature. Sensible heat fluxes over land surface are parameterized as F AL sh 5 D l (T a T ), (17) l using a constant bulk coefficient D l (Martin 22). In contrast to the energy balance Eq. (1), moisture is transported by diffusion and advection. Meridional advection is important in the tropical regions where moisture is transported equator ward by intertropical convergence (ITC), where precipitation occurs. In a diffusive-only scheme, moisture would not converge but diverge in this region. Zonally averaged monthly wind velocity fields from ERA-4 reanalysis are applied (note that zonally resolved winds would lead to convergence in various boxes in the mid and high latitudes and thus, as a direct effect, to an unrealistically high amount of precipitation. These boxes negatively affect the state of the model. In the three-dimensional real world, convergence only leads to precipitation when the rising air masses cool sufficiently). The wind fields are vertically density weighted and averaged up to the moisture scale height. The vertical fluxes are given by evaporation and precipitation. The moisture balance equation is formulated as follows: t q a 5 1 r cosq u uq a 1 1 r 2 cos 2 q u Kq u u q a 1 r (sinq) cos qyq a 1 1 r 2 (sinq) Kq q cos2 q 3 (sinq) q a 1 r o (E P). (18) r a h q Eddy diffusivities K q u and K q q are assumed to be globally constant. The sea ice model component is based on work by Semtner (1976) and Hibler (1979) and is similar to the sea ice model of Edwards and Marsh (25). The model determines three variables: fractional sea ice area A i, ice thickness H i, and surface sea ice temperature T i. Note that H i is averaged over the ice and the open-ocean fractions. Ice dynamics are kept very simple: Ice flows with the surface ocean currents, and processes induced by horizontal gradients are parameterized by diffusion with a diffusivity K i. Vertical heat fluxes are separated into the atmosphere ocean heat flux across ice-free areas, F AO total [Eq. (2)], the atmosphere ice heat flux across ice-covered areas, F AI total [Eq. (3)], and the ice-ocean heat flux F IO 5 c h u t (T f T o )r o c p,o, (19) which brings T o back to the freezing temperature T f by either melting or growing ice. Note that T f is salinity dependent and is parameterized as T f (S) K ( S 1.4S 2 1.4S 3 ) K (2) (Doronin and Kheisin 1977). Also, u t is the skin friction velocity at the ice ocean boundary, and c h is an empirical constant after McPhee (1992). The parameter values are given in Table 1. The total heat flux from the atmosphere is F BOA total 5 (1 A i ) F AO total 1 A i FAI total, where A i is the ice-cover fraction. Fractional ice area satisfies a balance equation that consists of a horizontal flow term, an ice area production term, and an ice area destruction term. The ice produced in the open-ocean area is uniformly spread using a minimal thickness H. In the ice area destruction term, the ice

41 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 temperature of the ice is calculated by equating the radiative incoming heat flux with the conductive heat flux through the ice, thus assuming a linear temperature profile within the ice: F AI total 1 I cond H i (T f T i ) 5, (25) FIG. 1. Outline of how sea ice is distributed for the ice-melting term in Eq. (23); A i is the ice-covered fraction of the box, H i the height of the ice, dh i the melted ice layer, and da i the ice fraction that is melted away. is assumed to be distributed uniformly between height and 2H i /A i over the ice-covered fraction A i (Fig. 1). Following simple intercept theorems, the formed openocean fraction da i can be derived from 2H i /A i A i 5 dh i da i, (21) where dh i is the thickness of the melted ice layer. Hence, da i dt 5 A2 i dh i. (22) 2H i dt Thus, fractional sea ice area is calculated by solving the following equation: Similarly, H i t A i 1 $ (ua t i ) K i = 2 A i 5 max, 1 A! i F IO F AO total H r i L f "!# A 2 i 1 min, 2H i F IO F AI r i L f total E r o r i. (23)! 1$ (uh i ) K i = 2 H i 5 (1 A i ) max, FIO F AO total r i L f! total F IO F AI 1 A i r i L f E r o r i. (24) If the new ice thickness is smaller than a minimal ice thickness H, then H i and A i are set to zero. The surface where I cond is the thermal conductivity of ice. In the model, T i 5 min(t i, T f ). A very simple parameterization for ocean and sea ice albedo is used: a A i. (26) The coefficients are chosen to match values of ice-free and fully ice-covered ocean areas of the Kukla and Robinson (198) ocean sea ice albedo climatology. Over land, the zonally averaged land albedo climatology of Kukla and Robinson (198) is used. Note that the presence of sea ice requires the sensible heat flux F sh and evaporation E to be separated into an ice-covered and an open-ocean fraction. Finally, the heat flux into the ocean is calculated as F O Heat 5 (1 A i ) max(fio, F AO total ) 1 A i FIO 1 Q m. (27) As in Eq. (23), ice is formed over the open-ocean fraction when F IO. F AO total. In this case the released heat of fusion (F IO F AO total ) is considered in the first term of Eq. (27). Also, Q m is the heat of fusion of the additional amount of meltwater that is added to the ocean when the ice thickness falls below the minimal thickness (H i, H but in the previous time step H i,t2dt. ). The freshwater flux is calculated as " F O Fw 5 P 1 R (1 A )E i A F IO F AI total i r i L f!# 1 (1 A i ) max, FIO F AO total ri 1 Q m, r i L f r o L f r o (28) where R is runoff and Q m /(L f r o ) is the additional amount of freshwater added to the ocean when H i, H but H i,t2dt.. In the model, not freshwater but salt is added and taken out of the ocean, respectively: F O Salt 5 S ref FO Fw, with S ref psu being a reference salinity for the surface ocean. Because of the absence of dynamics in the atmosphere, an Atlantic-to-Pacific freshwater flux (Zaucker et al. 1994) must be prescribed. We apply.17 Sv to

42 4 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 355 increase the strength of the Atlantic meridional overturning circulation (AMOC). The freshwater is taken from the North Atlantic from the 38 to 718N basin and distributed into the North Pacific from 468 to 718N. An additional freshwater flux out of the ocean is applied to the Ross Sea and the Weddell Sea (two boxes each) in order to stimulate deep water formation in these regions. This correction is due to the fact that the small-scale processes in these regions are not resolved by the model. We apply a flux of.2 Sv divided into these four boxes. This flux correction is compensated by adding the same freshwater amount to the remaining ocean boxes around Antarctica ( S). The horizontal transport term in the energy and moisture balance Eqs. (1) and (18) is solved implicitly. Thus, the time step for the energy balance model (EBM) can be chosen equal to the ocean time step. Since the velocities and diffusivities in Eqs. (1) and (18) do not vary interannually, the linear systems of equations need to be inverted only during the first year of every simulation. This makes the EBM very efficient. Since the sea ice has low flow speeds and diffusion, it is not solved implicitly. We halved the time step for the land temperature [Eq. (5)] for numerical stability reasons. Equation (25) cannot be solved analytically. Note that T i 4 is linearized and discretized using a first-order Taylor approximation so that T 4 i,t 5 3T 4 i,t Dt 1 4T3 i,t Dt T i,t, where T i,t2dt is the temperature at the previous time step. Atmospheric temperature values of the boxes closest to the poles (poleward of 718) are averaged longitudinally after every time step to increase numerical stability. The discretization schemes Euler forward, centered differences, and variable upwind (for zonal advection) are used. 3. Present-day simulations a. Ocean The parameters described in the model section particularly the relative humidity after precipitation r h, precip, zonal and meridional eddy diffusivities K u and K q, the temperature reduction at cloud top for overcast conditions DT ct max, the Dalton number C E, and the freshwater correction fluxes from the North Atlantic to the Pacific and in the Ross and Weddell Seas have been tuned such that a good representation of the modern climate is achieved. The model tuning was done on the basis of observational fields and Taylor diagrams. Special emphasis was placed on atmospheric and surface ocean temperature, sea surface salinity, sea ice cover, Atlantic and Pacific zonal mean temperature, salinity, and radiocarbon concentration. Global relative standard deviations and correlations between the mentioned quantities and observations were calculated and optimized. A steady state of the coupled atmosphere ocean model is obtained by spinning up the ocean-only model for 1 kyr, followed by a 5-yr EBM initialization run, where evaporation patterns of ERA-4 reanalysis are approached by diagnosing the Dalton number C E seasonally at every grid point. Finally, ocean and atmosphere are coupled and a follow-on 1-kyr spinup is performed. The result is a stable and steady model state: Global mean net TOA radiation fluxes converge to zero. The 1-yr average of globally integrated radiation fluxes at TOA is.5 PW, which corresponds to an average flux of.1 W m 22. The modern steady-state annual mean overturning circulation of the Atlantic and Pacific basin and of the global ocean are shown in Fig. 2. North Atlantic Deep Water (NADW) reaches down to 3 4-km depth before flowing southward. Because of the coarse resolution of the model, NADW is formed south of Greenland, one box row south of the Greenland Iceland Norwegian (GIN) Seas, where deep water should be formed. The AMOC strength with a maximum of approximately 14 Sv is low compared to other models (Randall et al. 27). Observations of the Atlantic radiocarbon content (Key et al. 24) are, however, consistent with the AMOC strength (Fig. 3). The global overturning shows the Southern Ocean overturning cell with a strength of approximately 18 Sv. Again, this strength leads to a deep Pacific radiocarbon concentration that compares well with the observations (Fig. 4). Zonally and annually averaged latitude depth plots of ocean temperature, salinity, radiocarbon, and phosphate distributions are shown for the Atlantic (Fig. 3) and for the Pacific (Fig. 4). Atlantic temperatures agree well with observations of Levitus and Boyer (1994). The Pacific below 1-km depth is about 18C too cold. The salinity fields in both Atlantic and Pacific are too fresh at the surface and too salty at depth. This deficiency is possibly linked to the distribution and thus to the parameterization of evaporation and precipitation in the atmosphere. Besides radiocarbon, constraints on the state of the overturning circulation can be inferred from distributions of nutrients and other biogeochemical tracers. Therefore, the model is run with the prognostic carbon cycle, allowing us to compare the model output to observations from the World Ocean Atlas 21 (WOA1; Conkright et al. 22) for phosphate and silicate, and the Global Data Analysis Project (GLODAP) data (Key et al. 24) for dissolved inorganic carbon, alkalinity, and chlorofluorocarbon (CFC-11). The modeled phosphate distribution is in fair agreement with the observations. The largest deficiencies are found in the Pacific,

43 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 2. Modern annual mean Atlantic, Pacific, and global overturning circulation (in Sv). The Indonesian Throughflow at 18S in the Pacific (gray bar) produces a discontinuity in the contour lines. where surface concentrations are too high and the surfaceto-deep gradient is too small. In the Atlantic, the highest observational values are found in the equatorial upwelling region of the African Margin (Gulf of Angola). In the model, the highest values are also found at the African Margin, but the region extends to the Gulf of Guinea. Therefore, a sharp change to lower values is found in the zonally averaged field (Fig. 3). The phosphate accumulation in the deep ocean of the African Margin is a consequence of underestimated ventilation of the deepest boxes. For a quantitative comparison between the model and observations, a Taylor diagram is constructed providing information about the relative standard deviation and the correlation between annually averaged model and data fields (Fig. 5a). For CFC-11, a transient simulation was performed prescribing atmospheric concentrations starting at year 1931 A.D. and ending at year 2. Because the observational data from GLODAP (Key et al. 24) were collected during multiple years of the 199s, the data of every cruise station were interpolated and regridded on to the model depth grid and then compared to the corresponding year of the model output. Additionally, the model results for temperature, salinity, and radiocarbon are compared to the results of the ocean-only simulation described by Müller et al. (26). It is expected that the ocean-only simulation performs better than the coupled simulation because the surface ocean boundary is restored to observations, while in the coupled run here the surface ocean conditions follow from the energy balance model. Thus, it is remarkable that the temperatures of the coupled run agree nearly as well with the observations as those of the ocean-only simulation. Correlation and relative standard deviation of the salinity, however, is quite poor. As described earlier, this is primarily due to the poorly represented surface-to-deep gradient of salinity. On the other hand, radiocarbon of the coupled run compares better with observations, indicating that the time scales of surface-to-deep transport are realistic in the model. b. Surface ocean and atmosphere Annual mean atmospheric and sea surface temperature (SST), sea surface salinity (SSS), evaporation, and precipitation are shown in Fig. 6 and compared to observations from Levitus and Boyer (1994) for SST and Levitus et al. (1994) for SSS and to ERA-4 reanalysis data for atmospheric temperature, evaporation, and precipitation [for atmospheric temperature, values at standard sea level pressure ( hpa) are used for the comparison]. Atmospheric temperatures in the tropics and the latitudinal gradient are well represented by the model. The largest deficiency is found in the GIN Seas, where temperatures are too low. Because deep water formation occurs too far south in the model, the warm ocean currents do not contribute to the atmospheric temperature in this region, leading to too cold temperatures. Because in zonal direction advection is not considered in the model and diffusion is constant, zonal gradients are smaller than in ERA-4 data. Sea surface temperatures are in good agreement with the observations. In the tropics, temperatures are too low. This deficit is associated with the precipitation pattern. As described before, sea surface salinity is generally too low compared to the data of Levitus et al. (1994). In the initialization phase of the energy balance model, the Dalton number C E is diagnosed while prescribing ERA-4 evaporation. Thus, the simulated evaporation pattern matches the data well. The total amount of evaporation (and precipitation) is 2% lower than in ERA-4. Because winds are zonally averaged, model precipitation has large differences to the data of ERA-4. When using zonally resolved winds, the global precipitation pattern improves, but unrealistically high precipitation occurs in various

44 42 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 357 FIG. 3. Atlantic zonally and annually averaged latitude depth fields of (top) ocean temperature, (second row) salinity, (third row) natural radiocarbon (bomb-produced radiocarbon is filtered out in the observations), and (bottom) phosphate concentrations. (left) Model results are compared to (right) observations from Levitus and Boyer (1994) for temperature, Levitus et al. (1994) for salinity, GLODAP (Key et al. 24) for radiocarbon, and the World Ocean Atlas 21 (WOA1; Conkright et al. 22) for phosphate.

45 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 4. As in Fig. 3, but for the Pacific. boxes (as a direct consequence of convergence), which negatively affects the state of the model. Because NADW is formed too far south, modeled sea ice extent is too large in the Arctic Ocean east of Greenland as compared to the dataset of Rayner et al. (23). On the other hand, Southern Ocean sea ice extent is slightly too low throughout the year (Fig. 7).

46 44 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 359 The seasonal variability of temperature in the atmosphere reproduces the larger July-to-January temperature difference over continents compared to the ocean (Fig. 8). The amplitude is too large over the GIN Seas because deep-water formation, which attracts the warm surface currents and which occurs too far south in the model, is a winter-only phenomenon. The amplitude is also too large over South America and the Southern Ocean. Evaporation and precipitation vary strongly during the season. The model reproduces the seasonal shift of precipitation in the equatorial zone (Fig. 9). Correlation and relative standard deviation of modeled annual mean, January, and July fields of atmospheric temperature, evaporation, and precipitation with ERA-4 data are displayed in Fig. 5b. Meridional heat transport in the model is shown in Fig. 1. The ocean heat fluxes are compared to observationally based estimates from Lumpkin and Speer (27), Fasullo and Trenberth (28), and Trenberth and Caron (21). Modeled Atlantic meridional transport is.2 to.6 PW lower than found in the observations. In the Indo- Pacific, they agree well. Atmospheric heat flux is divided into sensible and latent heat flux. Compared to the databased estimates of Fasullo and Trenberth (28), atmospheric heat transport is by up to 1.5 PW too low in the Northern Hemisphere and by up to 2 PW too low in the Southern Hemisphere. The total heat flux, the sum of atmospheric and ocean heat flux, is also shown. c. Sensitivities FIG. 5. (a) Taylor diagram showing correlation and relative standard deviation of modeled annual mean ocean temperature, salinity, and various tracer fields compared to observations. The filled symbols and the plus sign (1) represent the coupled model; the outlined symbols indicate the ocean-only model, as in Fig. 8 of Müller et al. (26). Observations are taken from Levitus and Boyer (1994) and Levitus et al. (1994) for temperature and salinity, respectively; GLODAP (Key et al. 24) for radiocarbon, chlorofluorocarbon CFC-11, dissolved inorganic carbon (DIC) and alkalinity; and WOA1 (Conkright et al. 22) for phosphate and silicic acid. Model results for CFC-11 were determined from a transient simulation, in which atmospheric concentrations were prescribed from years 1931 to 2 A.D. Perfect agreement with the data is at point (1, ) in the Taylor diagram. (b) Correlation and In a first sensitivity test, modeled atmospheric temperatures are compared to ERA-4 data at 2 m above ground by taking into account land topography based on ETOPO5 (data available online at gov/mgg/global/etopo5.html). Atmospheric temperature at altitude z is calculated by applying a constant lapse 5 T a Gz in the atmosphere land flux Eq. (4), the equation for sensible heat flux over land (17) and in Eq. (9) for the longwave radiation flux at TOA, thus updating Eq. (11) to rate G 5 5 K km 21. We replace T a by T alt a 5 T a j DTct max Gz. As in the standard case, the temperature at sea level is used for the horizontal transport. The value of G was chosen such that the root-mean-square T ct a relative standard deviation of modeled atmospheric temperature, evaporation and precipitation with ERA-4 reanalysis data. Black symbols indicate the annual mean; gray symbols, January fields; outlined symbols, July fields. Only evaporation over the ocean is compared. Symbols for annual mean and January precipitation are on top of each other.

47 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 6. Model results of annual mean atmospheric temperature, sea surface temperature, sea surface salinity, evaporation, and precipitation compared to ERA-4 reanalysis data [for atmospheric temperature, values at standard sea level pressure ( hpa) are used for the comparison]. In the third column the model observation differences are displayed.

48 46 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 361 FIG. 7. Seasonal variation of modeled sea ice cover compared to the dataset of Rayner et al. (23). The dataset is averaged over years , the time span of the ERA-4 data. (a),(b) Time series of Northern and Southern Hemisphere sea ice cover. The ice cover in (b) was calculated from the original dataset ( resolution, solid line) and after regridding to the Bern3D grid (dashed line). (c) (f) February and September fractional ice area, respectively. deviation between modeled temperature and ERA-4 is minimized. Global and annual mean atmospheric temperature at sea level is 16.38C when topography is taken into account (run henceforth called ALTI), which compares to 15.48C in the standard case (called CTRL); the mean ERA-4 temperature at sea level is 15.78C. Root-meansquare deviations between atmospheric temperature of ALTI and ERA-4 at 2 m above ground are 2.88C for the annual mean, 5.58C in January, and 4.38C in July. In comparison, root-mean-square deviations between atmospheric temperature of CTRL and ERA-4 at sea level are 3.8C for the annual mean, 5.48C in January, and 4.58C in July. So generally, the model compares better to ERA-4 when topography is taken into account. Also, temperatures over the Eurasian continent are closer to ERA-4. On the other hand, temperatures over Antarctica and the Southern Ocean are too warm. As a consequence, AABW is too weak. Modeled Northern Hemisphere summer temperatures deviate most from ERA-4 in the vicinity of the Hudson Bay (Fig. 8). Thus, in a second sensitivity test, the Hudson

49 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 8. Seasonal variations of atmospheric temperature. Model results of (left) July and (middle) January temperatures, and (right) July January differences, are compared to ERA-4 reanalysis data. In the bottom row the model observation differences are displayed. Bay is taken into account in the model by replacing two land boxes by ocean boxes in North America such that the Hudson Bay is connected to the Arctic Ocean. The presence of the Hudson Bay indeed leads to significantly cooler July air temperatures in this area by approximately 38C compared to the standard case. However, it cannot explain the deviation from ERA-4 of approximately 13.58C. Also, the Hudson Bay hardly affects the adjacent boxes and it does not affect the overall state of the model. 4. Application of the coupled Bern3D model: Paleosimulations a. Last Glacial Maximum To analyze the circulation state of the model under Last Glacial Maximum (LGM) conditions, a 1-kyr steadystate simulation is performed, setting orbital parameters to values at 2 kyr B.P. (before present; i.e., before 195 A.D.), atmospheric CO 2 to 18 ppm, atmospheric CH 4 to 35 ppb, and increasing surface albedos to account for the presence of Northern Hemispheric ice sheets during the LGM. Since ice sheets are not simulated explicitly by the model, their extent is taken from Peltier (1994). During the first 2 kyr of the simulation, the ice sheet is linearly scaled from the modern to the LGM state, taking into account the relocation of the freshwater from the ocean onto the land and the additional latent heat flux (see appendix for details). As described in section 2, the freshwater relocation is done by adding salt to the ocean instead of lowering the sea level. As a consequence, ocean gateways such as the Bering Strait remain open. The glacial state of the overturning circulation has been qualitatively reconstructed by analyzing estimates of LGM nutrient distributions inferred from Cd/Ca ratios of benthic foraminifera found in marine sediment cores (Lynch-Stieglitz et al. 27; Marchitto and Broecker 26). As in the present-day simulation, the model is run with the prognostic carbon cycle. Additionally, LGM

50 48 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 363 FIG. 9. Seasonal variations of evaporation and precipitation. Model results of January and July (top two rows) evaporation and (bottom two rows) precipitation are compared to ERA-4 reanalysis data. ERA-4 evaporation over land surface is not shown in this figure. In the third column the model observation differences are displayed. aeolian iron fluxes have been applied (Mahowald et al. 26). Modeled phosphate concentrations are converted to cadmium concentrations following the linear relationship described by Elderfield and Rickaby (2). In the modeled glacial state, AMOC is shallower and somewhat weaker compared to the modern state, forming the so-called Glacial North Atlantic Intermediate Water (GNAIW) in agreement with paleoclimatic reconstructions (Labeyrie et al. 1992; Gherardi et al. 29). On the other hand, Antarctic Bottom Water (AABW) becomes stronger and penetrates farther to the north (Fig. 11). This is also seen in the reconstruction of the cadmium distribution. Compared to the paleoceanographic estimates, model concentrations are by approximately.1 to.2 nmol kg 21 too high. Also, too strong nutrient trapping in the equatorial surface ocean creates too high cadmium values in this region. However, as discussed for the modern phosphate distribution, these high concentrations occur at the African Margin, where no paleoceanographic data are available (Marchitto and Broecker 26). Modeled equatorial concentrations in the central surface and deep Atlantic are by approximately.1 nmol kg 21 lower. Note that paleoceanographic data used for the reconstruction are very sparse.

51 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 1. (a) Northward transport of heat in the Atlantic, Pacific, and global ocean. The model (thick lines) is compared to data from Lumpkin and Speer (27) (squares) and to data of Trenberth and Caron (21) derived from ECMWF atmospheric fields (thin lines). The total ocean heat flux is also compared to newer estimates inferred from satellite retrievals from the Earth Radiation Budget Experiment (ERBE) and Clouds and Earth s Radiant Energy System (CERES) fields (Fasullo and Trenberth 28) (dashed line). (b) Modeled northward heat transport of the atmosphere, the ocean, and in total (thick lines). The atmospheric transport is separated into sensible and latent heat flux. The model is compared to the satellite-based data of Fasullo and Trenberth (28) (thin lines). LGM sea ice cover is larger in the North Pacific and the Southern Ocean compared to the modern model state, whereas the North Atlantic shows little difference (Fig. 12). The Southern Ocean sea ice margin extends farthest in the Atlantic sector to about 458S in winter and 68S in summer. This is in rough agreement with reconstructions from Gersonde et al. (25), who also find the largest sea ice extent in the Atlantic sector to be about 488S in winter and 528S in summer. In a sensitivity test the Bering Strait was closed. Compared to the standard setup, this results in a (2 Sv) weaker and shallower AMOC and stronger AABW, and the Arctic Ocean becomes fresher. When additionally closing the passage between North America and Greenland, the Arctic freshens more. Global mean atmospheric temperature is not significantly affected. Comprehensive climate models such as NCAR CCSM and HadCM3 also show a shallower and weaker AMOC in the LGM state as compared to the modern state. Others, such as the Model for Interdisciplinary Research on Climate (MIROC; Hasumi and Emori 24) and ECBilt- CLIO, show a stronger and deeper AMOC (Otto-Bliesner et al. 27). Weber et al. (27) developed a metric to analyze the reasons for the different behaviors of the models and applied it to various models of the Paleoclimate Modeling Intercomparison Project (PMIP2). Here, we apply the same metric to the Bern3D model. First, the components of the Atlantic freshwater budget according to the balance equation M surf 5 M ov 1 M az 1 M diff 1 M BS (in equilibrium) for the modern state and their difference to the LGM state are calculated. Here M surf is the basin integral of evaporation, precipitation, continental runoff, ice melt, brine rejection, and flux corrections; M ov is the meridional overturning component and M az the azonal component of the oceanic freshwater transport through the southern boundary. The Bering Strait and diffusive contributions are determined as a residual term R 5 M diff 1 M BS. Other quantities to determine Southern Ocean controls versus Atlantic processes are r Atl, the density difference between the northern and southern ends of the Atlantic basin (taken at 558N and 38S, respectively and averaged over the top 15 m), and r NS, the density contrast between NADW and AABW (taken at 558N and 558S, respectively and averaged over all depth levels). The results of the Bern3D model are summarized and compared to other models in Tables 2 and 3. As in most PMIP2 models, compared to observations (Pardaens et al. 23; Weijer et al. 1999) M surf is overestimated and M az underestimated in the modern state of the Bern3D model. The overturning component M ov is negative but too small compared to the observations. The sign of M ov simulated by NCAR CCSM, HadCM3, and the UVic model is inconsistent with the observations. The diffusive transport is overestimated, as can be expected from a coarse-resolution model. The differences between modern and LGM values for these quantities are small in the Bern3D model compared to the PMIP2 models, except for UVic. Although M surf remains the same, M ov switches to a positive value. This change is compensated by the azonal component. The differences in M ov and M az between glacial and interglacial

52 5 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 365 FIG. 11. (a) Last Glacial Maximum (LGM) Atlantic overturning circulation. (b) Zonally averaged modeled Atlantic cadmium concentrations for the LGM. Modeled cadmium concentrations are obtained by applying the following equation to the PO 4 model output: [Cd] nmol kg 21 /(2(33 nmol kg 21 /[PO 4 ] 2 1) 1 1) (Elderfield and Rickaby 2). (c) LGM cadmium concentration estimates from the ratio of Cd/Ca in shells of benthic foraminifera (Marchitto and Broecker 26). The data locations are indicated in the figure. (d) Modeled cadmium field, where model data are only taken at the data locations of Marchitto and Broecker (26). The interpolation is done as in (c). state are consistent with CCSM, HadCM3, MIROC, and ECBilt. Weber et al. (27) conclude that none of these quantities are good indicators for the changes in AMOC strength, basically because their LGM-to-modern differences are all in line and independent of the differences in AMOC strength. This is also true for the Bern3D model, especially because M surf exhibits negligible changes and DM ov is positive but DC AMOC negative. In a second step, Weber et al. (27) analyze the correlation of DC AMOC with Dr atl and Dr NS. The processes that correlate positively with DC AMOC indicate potential drivers for the AMOC strength. In the Bern3D model, DM surf, Dr NS, and DC MOC and Dr atl are anticorrelated. Therefore, as for the other EMICs (ECBILT-CLIO and UVic) analyzed by Weber et al. (27), the changes in the analyzed quantities are too small to attribute them as controlling processes for the change in AMOC strength (Table 3). b. Transient simulations over the past 8 years 1) VARIATIONS IN GREENHOUSE GAS CONCENTRATIONS AND ICE SHEET EXTENT Simulations over ice age cycles are possible with the Bern3D model. Here we present seven transient simulations over the past 8 years (Table 4), where orbital forcing, atmospheric greenhouse gases CO 2 and CH 4 and ice sheets are prescribed. In simulation ORBI, only orbital forcing is applied. In addition to the orbital forcing, simulation PCO2 varies CO 2, simulation PCH4 varies CH 4, and simulation ICE1 varies ice sheets. In ALL1, all forcings are taken into account. For CO 2 and CH 4, data of Lüthi et al. (28) and Loulergue et al. (28), respectively, are used (Fig. 13c). As before, ice sheets for the modern and LGM states are taken from Peltier (1994) and scaled using the benthic d 18 O stack of Lisiecki and Raymo (25), which is a proxy of global ice volume (Fig. 13d; see appendix for details). The simulations that involve ice sheet changes are computed twice: once taking into account albedo changes, the freshwater relocation from the ocean onto the land and vice versa, and the additional latent heat flux when ice sheets are formed and wasted (simulations ICE1 and ALL1) and once only accounting for albedo changes (ICE2 and ALL2). Snow albedo feedback is not included in the simulations. The changes in atmospheric global and annual mean temperature and AMOC strength are shown in Fig. 13. In the model, the effects of orbital forcing only or in combination with CH 4 forcing are small. The CO 2 and orbital forcing change global mean atmospheric

53 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 12. Seasonal variation of LGM sea ice cover. (a) Time series of Northern and Southern Hemisphere sea ice cover; (b) February and (c) September fractional ice area. temperatures by 1.58 to 28C, where the higher value stands for large glacial-to-interglacial changes such as the LGM-to-modern change. The lower value represents smaller glacial-to-interglacial changes (e.g., at 5 kyr B.P.). The AMOC is not significantly affected by this forcing. Ice sheet forcing provokes glacial-tointerglacial temperature changes of 18 to 1.58C. In ICE1, when freshwater relocation from the ocean to the continent is taken into account, the AMOC increases rapidly at glacial inceptions and decreases again at the terminations. The reason for this behavior is the densification of the surface ocean due to the ice sheet buildup. In the North Atlantic, this leads to stronger convection and hence to stronger AMOC. Analogously, the AMOC weakens during glacial terminations. This result is in agreement with an earlier study by Meissner and Gerdes (22), who also find an AMOC intensification during glaciation. In ALL1, when all forcings are combined, glacial-tointerglacial temperatures vary by 38 to 48C. This temperature difference is at the lower end of the 48 to 78C cooling estimated for the LGM (Jansen et al. 27). Modeled glacial-to-interglacial temperatures vary by 28 to 38C in the tropics, by 68 to 88C in the southern high latitudes, and by 3.58 to 68C in the northern high latitudes. Pollen records also indicate to a 28 to 38C cooler climate in the tropics during the LGM (Farrera et al. 1999). For the high latitudes, temperature reconstructions from polar ice cores indicate a LGM-to-modern temperature increase of about 98C in Antarctica (Stenni et al. 21) and about 228C in Greenland (Dahl-Jensen et al. 1998). We tentatively attribute the large difference between reconstructed and modeled northern highlatitude temperatures to the too low sea ice sensitivity and to the absence of topographic effects of the North American and Eurasian ice sheets, which changes the course of the jet stream. Besides the different magnitudes of the change, the evolution of temperature at the poles and in the tropics and of the global mean are similar, suggesting a fast glacial-to-interglacial transition and a more gradual interglacial-to-glacial transition, as also seen in the atmospheric CO 2 (Fig. 13c) and the d 18 O (Fig. 13d) records. The relatively fast interglacial-toglacial transitions found in the dd Antarctic ice core record (Jouzel et al. 27) (Fig. 13e), a proxy for atmospheric temperature in Antarctica, are not represented by the model. As in ICE1, AMOC in ALL1 increases at the glacial inception, but its strength decreases again with the cooling. At the glacial maxima, the AMOC is weaker than during the interglacial periods. The weakening of the AMOC is also observed when the ice sheet freshwater relocation is not taken into account. Global atmospheric temperatures are similar in ALL1 and ALL2. Note that the parameterization of the ice sheets does not allow ablation at the ice sheet margin in an ice sheet buildup phase. Also, the switch from an ice sheet melting to an ice sheet build-up phase and vice versa occurs globally at the same time. This might lead to an overestimation of the abrupt AMOC changes that are induced by the switch from ice sheet melting to ice sheet buildup.

54 52 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 367 TABLE 2. Values for the AMOC strength C AMOC, the Atlantic basin-integrated result of evaporation, precipitation, and other freshwater fluxes M surf, the meridional overturning component M ov and the zonal component M az of the oceanic freshwater transport through the South Atlantic boundary, and a rest term R that includes Bering Strait and diffusive contributions to close the Atlantic freshwater budget (in Sv). The values shown are for the modern model state (mod) and for the LGM to modern difference (D 5 LGM 2 mod). The Bern3D values are compared to observations (Pardaens et al. 23; Weijer et al. 1999) and to various PMIP2 models (Weber et al. 27). C AMOC M surf M ov M az R mod D mod D mod D mod D mod D Bern3D & Obs CCSM HadCM MIROC ECBilt UVic ) LOCAL OCEAN TEMPERATURE SENSITIVITY TO CHANGES IN GLOBAL MEAN ATMOSPHERIC TEMPERATURE, AMOC, AND AABW We focus on ALL1, the most realistic simulation of this study, and determine the sensitivity of ocean temperature in various regions to changes in global mean atmospheric temperature, AMOC, and AABW strength. Therefore, the assumption is made that the ocean temperature time series at any location in the ocean can be reconstructed by a linear combination of global mean atmospheric temperature T atm, AMOC, and AABW strength (C AMOC, Fig. 14c, and C AABW, Fig. 14d) time series. The purpose is to test whether the inversion of this approach would be feasible, that is, to determine important global-scale atmosphere (T atm ) and ocean (C) variables using a specific combination of reconstructions of local temperature from paleoceanographic records. First the quantities are normalized as follows: e T 5 (T T)/max(T T), where T is the average over 8 kyr, and likewise for C. Then the linear combination TABLE 3. Modern to LGM differences (LGM 2 modern) of C AMOC, M surf, the density difference between the northern and southern ends of the Atlantic r atl, and the density difference between NADW and AABW r NS in the Bern3D model and in various PMIP2 models (Weber et al. 27). The potential controlling process for the change in AMOC strength in each model is indicated in the last column. None of these processes could be attributed as a controlling factor in the Bern3D model as well as in ECBilt and UVic, since they are either anticorrelated to C AMOC or insignificant. DC AMOC (Sv) DM surf (Sv) Dr Atl (kg m 23 ) Dr NS (kg/m 23 ) Control Bern3D 21.4 &..4 &. CCSM r NS HadCM M surf MIROC r Atl 1 r NS ECBilt UVic ef 5 a Tatm e Tatm 1 a AMOC e CAMOC 1 a AABW e CAABW (29) is fitted to the local ocean temperature time series T e oc by minimizing g 5 å 8, t51 (e f t Toc, e t) 2. The calculation returns values for the coefficients a Tatm, a AMOC, and a AABW, which give information on the sensitivity of the local temperature to these quantities. Note that T atm, C AMOC, and C AABW are not orthogonal, but their correlations are small:.46 between T atm and C AMOC, 2.44 between T atm and C AABW, and.7 between C AMOC and C AABW. Modeled deep ocean temperatures can be validated by comparing the results to reconstructions from marine sediment cores. However, because of major difficulties in reconstructing deep-sea temperatures, only few time series exist currently. Martin et al. (22) provide reconstructions from benthic foraminiferal Mg/Ca records for the last 33 kyr in the eastern tropical Atlantic and for the past 23 kyr in the eastern tropical Pacific (Figs. 14a,b). Because of the large uncertainties of the reconstructions of about 1.58C, we cannot go into detail TABLE 4. Summary of the 8-kyr model simulations. Quantities not mentioned in the prescribed forcing column are kept constant at preindustrial values. These are land mask, winds, atmospheric and ocean diffusivities, terrestrial surface albedo (unless ice sheets are present), cloud cover, atmospheric CO 2 and CH 4, and ice sheets. Simulation Prescribed forcing Freshwater relocation for ice sheets ORBI Orbital PCO2 Orbital, CO 2 PCH4 Orbital, CH 4 ICE1 Orbital, ice sheets Yes ICE2 Orbital, ice sheets No ALL1 Orbital, CO 2, CH 4, ice sheets Yes ALL2 Orbital, CO 2, CH 4, ice sheets No

55 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL 368 JOURNAL OF CLIMATE 53 VOLUME 24 FIG. 13. A series of 8-kyr simulations. (a) Global and annual mean atmospheric temperatures for a simulation with varying solar insolation (green), insolation and CH4 (orange), insolation and CO2 (red), insolation and ice sheets (dark blue when freshwater relocation from the ocean onto the land and the additional latent heat flux is taken into account, light blue when it is not and only the surface albedo changes), and the combination of the three (black and gray, respectively, when the freshwater relocation is not taken into account). For the parameters held constant, preindustrial values are used. (b) Changes of AMOC strength for the runs described above. For ORBI, PCH4, PCO2, and ICE2, the 5-yr running average is plotted to disambiguate the panel. (c) (e) Atmospheric CO2 (Lu thi et al. 28), the benthic d18o stack of Lisiecki and Raymo (25) (which is a proxy for global ice volume), and the dd Antarctic deuterium record (a proxy for atmospheric temperature in Antarctica) (Jouzel et al. 27), respectively. The shaded vertical bars indicate interglacial periods (Augustin et al. 24).

56 54 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 369 FIG. 14. (a),(b) Atlantic and Pacific equatorial deep ocean temperature of simulation ALL1 compared to reconstructions of this region (Martin et al. 22). (c) AMOC strength. (d) AABW strength. (e) Normalized atmospheric temperature (gray) and deep equatorial Atlantic temperature (black). The normalization was done as follows: e T 5 (T T)/max(T T), where T is the temperature average. The red curve is a fit of the function ef 5 a Tatm e Tatm 1 a AMOC e CAMOC 1 a AABW e CAABW to the normalized local temperature; C AMOC and C AABW are the AMOC and AABW strengths, respectively, as in (c) and (d). The coefficients as well as the correlations between atmospheric temperature and local temperature R Tatm, and between the fit and the local temperature R fit, are listed. (f) As in (e), but for the deep equatorial Pacific.

57 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 when comparing the data with the model results. However, as already discussed for the modern ocean, the modeled deep Pacific is too cold by about 18C. One major feature of the reconstruction, the early warming of the Pacific during glacial times, is not reflected in the model. Model and reconstruction compare somewhat better in the Atlantic, where both show a gradual temperature decrease during glaciation. The fit for the deep equatorial Atlantic temperature time series correlates surprisingly well with the modeled temperature (R fit 5.99; Fig. 14e). For a comparison, the correlation between atmospheric and ocean temperature is substantially lower, with R Tatm 5.9. Thus, the circulation plays an important role in this region. The calculated coefficients are a Tatm 5.75, a NADW 5.53, and a AABW In other words, a 1% change in atmospheric temperature explains a 75% change in the deep equatorial Atlantic temperature. A 1% increase in AABW strength explains a local temperature decrease by 13%. At this location, all three components contribute to changes in the local temperature, although the role of AABW is minor. The same exercise for deep equatorial Pacific temperature results in a correlation of R fit 5.97, and the coefficients read a Tatm 5.7, a NADW 5.15, and a AABW (Fig. 14f). As expected, AABW is more important here than in the Atlantic, and correspondingly the influence of AMOC is less important. However, it is surprising that the AMOC is not negligible in this region: when AMOC is not taken into account in the fitting, the correlation R fit is reduced to.96. The described procedure has been applied to 23 locations in the ocean. The chosen regions are the northern, equatorial, and southern Atlantic and Pacific and the equatorial and southern Indian Ocean. In every region an average of several boxes is taken (light gray area in Fig. 15) and separated into upper ocean (4 65-m depth), intermediate ocean ( m depth), and deep ocean (.23 m, depending on the topography). In all locations, the sensitivity is highest to the atmospheric temperature. AMOC has a strong influence in the Southern Hemisphere intermediate and deep oceans and in the deep Atlantic and Indian Oceans. In most cases, an increase in AMOC leads to a warming. Exceptions are only the upper equatorial and southern Atlantic, where a stronger AMOC exports more heat to the North and thus leads to a local cooling. In the Pacific and Indian Ocean, stronger AABW leads to cooling. In the intermediate Atlantic, where the AABW cell returns south, a stronger AABW induces warming. Using the local sensitivity coefficients calculated above, the model can potentially be used to extract information of the ocean circulation from paleoceanographic temperature reconstructions. This requires ocean temperature time series at multiple locations. 5. Application of the coupled Bern3D model: Climate sensitivity Here we investigate the possibility of model statedependent equilibrium climate sensitivity by comparing the temperature response of the model to a doubling and quadrupling of atmospheric CO 2 concentration for the modern and the LGM states. Atmospheric pco 2 is doubled from 278 to 556 ppm and quadrupled to 1112 ppm in the modern case and from 18 to 36 ppm and to 72 ppm, respectively, in the LGM case. The model is run for 1 years to a new equilibrium. We compare the change in global mean atmospheric temperature and the associated relaxation time scales. As described earlier, the modern state equilibrium climate sensitivity to a doubling of atmospheric CO 2 is tuned to 38C. Modeled climate sensitivity of the LGM state is 4.38C. This higher sensitivity is probably due to the presence of more sea ice in the LGM. When sea ice melts, surface albedo decreases strongly and more radiation is absorbed. Note that we assume that the water vapor feedback parameter l remains constant. Also, land ice remains constant. A quadrupling of atmospheric CO 2 from the modern state results in a temperature increase of 5.58C and from the LGM state in an increase of 7.38C. To quantify the relaxation time scales, a least squares exponential fit of the form T(t) 5 T f1 [å 3 i51 a i exp( t/t i )]g with the constraint å 3 i 5 1 a i 5 1 is performed, where T is the equilibrium climate sensitivity and t i are characteristic time scales of the transient global mean response. The results for the four simulations are summarized in Table 5 and Fig. 16. In the 2 3 CO 2 modern case, t yr, t yr, and t yr with the coefficients a , a 2 5.3, and a How long it takes for the climate to equilibrate depends on the efficiency of ocean heat uptake. The empirical time scales cannot be assigned to individual processes, but in general t 1 covers short-term processes such as wind-driven mixing at the surface ocean and convection into the intermediate and deep ocean; t 2 and t 3 cover intermediate and long-term processes that transport the temperature signal to the deep ocean. To a large extent, the equilibration time scale of the deep ocean is determined by the AMOC and AABW strength. When comparing the modern 2 3 CO 2 and 4 3 CO 2 simulations, it is notable that a 3 and t 3 of the 4 3 CO 2 run are substantially reduced. While the ocean circulation is hardly affected by the CO 2 doubling, the AMOC collapses in the 4 3 CO 2 model world and AABW strength increases from 18 to 28 Sv. The relaxation time scales of the LGM 2 3 CO 2 and 4 3 CO 2 simulations are

58 56 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 371 FIG. 15. Sensitivity of ocean temperature at various regions to changes in atmospheric temperature, AMOC, and AABW strength in the ALL1 simulation. Every region is an average of several boxes (light gray area) and is separated into upper ocean (4 65-m depth), intermediate ocean ( m depth), and deep ocean (.23-m depth depending on the topography). For reasons of topography, the deep North Atlantic location does not represent bottom waters in this region. The bars depict the values of the coefficients in Eq. (29), given in %. They therefore give the percentage of a local temperature change per 1% change in atmospheric temperature, AMOC strength, and AABW strength, respectively. In brackets the local temperature change is given in 8C per 8C change in atmospheric temperature (and 8C per Sv change in AMOC and AABW, respectively). The R value is the correlation between modeled local temperature time series and the linear combination of atmospheric temperature, AMOC, and AABW time series according to the given coefficients. similar. However, the coefficient a 3 is reduced in the 4 3 CO 2 run, indicating a faster overall response in the 4 3 CO 2 run compared to the 2 3 CO 2 run. Compared to the modern 2 3 CO 2 simulation the time scales are longer, though. In particular, t 3 is twice as large. The steady-state LGM ocean circulation after the perturbation shows a strong AMOC of 17 Sv (2 3 CO 2 ) and 18 Sv (4 3 CO 2 ), respectively, filling the entire Atlantic. AABW strength decreases to 14 Sv in both runs. It appears that the response time of the climate system to a strong CO 2 perturbation strongly depends on the reaction of its components. Since in the Bern3D model winds are not affected by the warming, t 1 and a 1 are similar in all simulations; however, t 2 and t 3 anticorrelate TABLE 5. Equilibrium climate sensitivity and the relaxation time scales for a doubling and quadrupling of atmospheric CO 2 in the modern and LGM climate state. The 4 3 CO 2 case is compared to results from the HadCM3 model (Li and Jarvis 29) (2s confidence interval in brackets). DT (K) a 1 t 1 (yr) a 2 t 2 (yr) a 3 t 3 (yr) 2 3 CO 2 modern CO 2 modern CO 2 HadCM ( ) ( ) (78 191) ( ) 2 3 CO 2 LGM CO 2 LGM

59 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 FIG. 16. The 2 3 CO 2 and 4 3 CO 2 climate sensitivity simulations for the modern and LGM case as summarized in Table 5. The temperature responses are normalized to compare the rates of change. Also shown is a 4 3 CO 2 simulation from the HadCM3 model for the modern case (Li and Jarvis 29). Because the model was only run for 1 years, the uncertainties are large. with AABW strength, which determines how fast the temperature signal is transported to the remotest regions of the ocean. The role of the AMOC is minor. Li and Jarvis (29) have also calculated the relaxation time scales to a 4 3 CO 2 perturbation of the modern climate using the HadCM3 AOGCM (Table 5). While HadCM3 shows a similar response time on the short time scale (t yr), t yr and t yr are approximately 4 times larger than in the Bern3D model. We attribute this slower response in atmospheric temperature of HadCM3 to a slower surface-to-deep transport in the ocean than in the Bern3D model. Note that the uncertainties are quite large in HadCM3 because the model was run for only 1 years. 6. Conclusions We have developed an efficient coupled threedimensional dynamical ocean energy balance atmosphere model capable of performing sensitivity studies and ensemble simulations on glacial interglacial time scales. In this work, an energy balance component has been added to the previously developed physical ocean circulation model (Edwards et al. 1998; Müller et al. 26) and the modules describing the penetration of transient tracers (CFCs, anthropogenic carbon, bomb-produced and natural radiocarbon; Müller et al. 26, 28); the cycling of carbon, carbon isotopes, alkalinity, phosphate, iron, silica, calcite, and oxygen (Parekh et al. 28; Tschumi et al. 28); 231 Pa/ 23 Th (Siddall et al. 27); the ecosystem model Pelagic Interaction Scheme for Carbon and Ecosystem Studies (PISCES) and the cycle of aragonite (Aumont and Bopp 26); a sediment module (Tschumi 29); and an ensemble Kalman filter framework for inverse analyses (Gerber et al. 29; Gerber and Joos 21). Although the resolution of the model is coarse, zonal gradients within ocean basins are resolved. Thorough comparisons of the modern model state to observations and reanalysis data have been made to validate the model. Several 8 -yr simulations were performed where the model was forced with orbital parameters, greenhouse gases, and ice sheets. We find that changes in these parameters have direct effects on the global overturning circulation. Atlantic meridional overturning varies between approximately 8 and 19 Sv in these transient simulations, with the lowest values during glacial maxima and terminations and highest values during glacial inceptions. When ice sheets, CO 2, and CH 4 are prescribed, the model shows glacial interglacial global temperature variations of 38 to 48C. This is on the lower end of the 48 to 78C cooling estimated for the LGM (Jansen et al. 27). A low bias is expected as the cooling influence of forcing by dust and by vegetation changes is not considered here. The CO 2 -only forcing accounts for 1.58 to 28C and the ice sheet-only forcing for 18 to 1.58C. The model shows an increase of AMOC strength at the glacial inception due to the relocation of freshwater from the surface ocean onto the land for the ice sheet buildup. In the glacial state, NADW weakens and shallows as indicated by paleoceanographic reconstructions. We analyzed the sensitivity of the ocean temperature at 23 locations to changes in atmospheric temperature, AMOC, and AABW strength. It is surprising how well the local temperature correlates with a linear combination of these three components. Although atmospheric temperature has the greatest influence at all analyzed locations, depending on the region, AMOC and AABW also have considerable influence. Modeled LGM climate is, with a global mean temperature change of 4.38C, more sensitive to a doubling of CO 2 than the modern climate (38C climate sensitivity). How long it takes for the ocean to equilibrate strongly depends on the reaction of AABW to the CO 2 change. In the LGM case, the CO 2 doubling weakens AABW strength. Thus, the relaxation time scale is approximately twice as large as in the modern case. On the other hand, the AMOC collapses in a modern 4 3 CO 2 world, AABW strength increases, and equilibration is faster. With the recently completed coupling of a sediment model (Tschumi 29) and the incorporation of the PISCES ecosystem model, the Bern3D model is approaching a comprehensive EMIC. The next step will be to couple the Lund Potsdam Jena dynamic global vegetation model (Sitch et al. 23; Strassmann et al. 28) to the Bern3D model.

60 58 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 373 Future work could include fine tuning of the model using an ensemble Kalman filter (Gerber et al. 29). Applications of the coupled model could include comparing past long-time scale local climate reconstructions such as SST from marine sediment cores (Martrat et al. 27) with the model. This could further validate the model and provide a quantitative understanding of the proxy behavior and of the past global climate by bringing the variations found in the reconstruction into a global context. Also, the step can be made to estimate past AMOC and AABW changes using the model and bottom water temperature time series. However, the result will have large uncertainties due not only to the model but to the difficulties in reconstructing bottom water temperatures. We are now also in a position to tackle transient glacial interglacial changes in the atmospheric and oceanic carbon isotopes D 14 C and d 13 C. Again, the model results will be compared to observations (Reimer et al. 24; Marchitto et al. 27; Elsig et al. 29). One of the main goals of paleoclimate research is to quantitatively explain glacial interglacial variations of atmospheric CO 2. Here, the addition of the EBM to the ocean component will allow us to progress beyond earlier work with the Bern3D model (Parekh et al. 28; Tschumi et al. 28). The Bern3D model will serve as a tool with which ensemble simulations of the Earth System over many 1 -yr periods are possible. Acknowledgments. This study was funded by GRACCIE (CONSOLIDER-INGENIO 21) and by the Swiss National Science Foundation. F. Joos acknowledges support by the European Community s Seventh Framework Programme (FP7/27 213) through the European Project on Ocean Acidification (EPOCA) and the Project Past4Future and by the Swiss Staatssekretariat für Bildung und Forschung (Cost Action 735). We thank J. Lynch-Stieglitz for providing the LGM cadmium data, M. Gerber for sharing the regridded CFC data, and N. Edwards for valuable comments concerning the EBM. We also thank N. Edwards, S. A. Müller, T. Tschumi, P. Parekh, R.Gangstø, M. Gerber, and A. Bass for their contributions in the development of the Bern3D model and three anonymous reviewers, whose comments have considerably improved the paper. ERA-4 reanalysis fields used in the EBM are provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). APPENDIX Ice Sheet Scaling Ice sheets for the modern and LGM states are taken from Peltier (1994) and linearly scaled using the benthic d 18 O stack of Lisiecki and Raymo (25). High-frequency variability of the d 18 O dataset was removed by applying a spline fit according to Enting (1987) with a cutoff period of 1 kyr. The ice sheet is assumed to have a constant height h ice. At the ice sheet margin the height goes abruptly from h ice to. Here h ice is calculated by solving å (u,q) (A LGM ice A mod ice )h r i ice r o 5 n oc Dh oc, (A1) where A ice is the fractional ice sheet covered area per box for the LGM and modern case, respectively, n oc is the number of surface ocean boxes, and Dh oc 5 12 m is the sea level rise since the LGM. For the ice sheet interpolation we distinguish between the Northern and Southern Hemisphere. In the Northern Hemisphere, the sum of the fractional ice sheet amount at longitude u and time t is calculated as follows: å q. A ice (t) 5 h å q. A LGM ice 1 (1 h) å A mod ice, (A2) q. where h 5 (d 18 O(t) 2 d 18 O mod )/(d 18 O LGM 2 d 18 O mod ) and d 18 O mod and d 18 O LGM are the modern and LGM d 18 O values. When h is between and 1 then the ice sheet is interpolated between LGM and modern conditions; when h, or h. 1 the ice sheet is extrapolated. The ice is latitudinally distributed starting at the northernmost land box. If, for example, å q. A ice (u eg, t eg ) 5 2.3, then A ice of the two northernmost land boxes at u 5 u eg is set to 1 and A ice of the box to the south is set to.3. The method is analogously used in the Southern Hemisphere. In the simulations where freshwater relocation from the ocean onto the land (and vice versa) is taken into account, the freshwater required for the ice sheet buildup is taken out of the ocean globally, whereas melted ice flows along the precipitation runoff path. REFERENCES Augustin, L., and Coauthors, 24: Eight glacial cycles from an Antarctic ice core. Nature, 429, Aumont, O., and L. Bopp, 26: Globalizing results from ocean in situ iron fertilization studies. Global Biogeochem. Cycles, 2, GB217, doi:1.129/25gb2591. Berger, A. L., 1978: Long-term variations of daily insolation and quaternary climatic changes. J. Atmos. Sci., 35, Bintanja, R., 1996: The parameterization of shortwave and longwave radiative fluxes for use in zonally averaged climate models. J. Climate, 9, Bolton, D., 198: The computation of equivalent potential temperature. Mon. Wea. Rev., 18, Collins, W. D., and Coauthors, 26: The Community Climate System Model version 3 (CCSM3). J. Climate, 19,

61 2.1. A DYNAMICAL OCEAN ENERGY BALANCE ATMOSPHERE MODEL J O U R N A L O F C L I M A T E VOLUME 24 Conkright, M. E., R. A. Locarnini, H. E. Garcia, T. D. O Brien, T. P. Boyer, C. Stephens, and J. I. Antonov, 22: World Ocean Atlas 21: Objective analyses, data statistics, and figures CD-ROM Documentation. National Oceanographic Data Center, 17 pp. [Available online at ftp://ftp.nodc.noaa. gov/pub/data.nodc/woa/publications/readme.pdf.] Dahl-Jensen, D., K. Mosegaard, N. Gundestrup, G. D. Clow, S. J. Johnsen, A. W. Hansen, and N. Balling, 1998: Past temperatures directly from the Greenland ice sheet. Science, 282, Doronin, Y. P., and D. Kheisin, 1977: Sea Ice. Amerind, 323 pp. Edwards, N. R., and R. Marsh, 25: Uncertainties due to transportparameter sensitivity in an efficient 3-D ocean-climate model. Climate Dyn., 24, , A. J. Willmott, and P. D. Killworth, 1998: On the role of topography and wind stress on the stability of the thermohaline circulation. J. Phys. Oceanogr., 28, Elderfield, H., and R. E. M. Rickaby, 2: Oceanic Cd/P ratio and nutrient utilization in the glacial Southern Ocean. Nature, 45, Elsig, J., and Coauthors, 29: Stable isotope constraints on Holocene carbon cycle changes from an Antarctic ice core. Nature, 461, Enting, I. G., 1987: On the use of smoothing splines to filter CO 2 data. J. Geophys. Res., 92, Farrera, I., and Coauthors, 1999: Tropical climates at the Last Glacial Maximum: A new synthesis of terrestrial palaeoclimate data. I. Vegetation, lake levels and geochemistry. Climate Dyn., 15, Fasullo, J. T., and K. E. Trenberth, 28: The annual cycle of the energy budget. Part II: Meridional structures and poleward transports. J. Climate, 21, Gerber, M., and F. Joos, 21: Carbon sources and sinks from an ensemble Kalman filter ocean data assimilation. Global Biogeochem. Cycles, 24, GB34, doi:1.129/29gb3531.,, M. Vazquez-Rodriguez, F. Touratier, and C. Goyet, 29: Regional air-sea fluxes of anthropogenic carbon inferred with an ensemble Kalman filter. Global Biogeochem. Cycles, 23, GB113, doi:1.129/28gb3247. Gersonde, R., X. Crosta, A. Abelmann, and L. Armand, 25: Seasurface temperature and sea ice distribution of the Southern Ocean at the EPILOG Last Glacial Maximum A circum- Antarctic view based on siliceous microfossil records. Quat. Sci. Rev., 24, Gherardi, J. M., L. Labeyrie, S. Nave, R. Francois, J. F. McManus, and E. Cortijo, 29: Glacial interglacial circulation changes inferred from 231 Pa/ 23 Th sedimentary record in the North Atlantic region. Paleoceanography, 24, PA224, doi:1.129/ 28PA1696. Goosse, H., T. J. Crowley, E. Zorita, C. M. Ammann, H. Renssen, and E. Driesschaert, 25: Modelling the climate of the last millennium: What causes the differences between simulations? Geophys. Res. Lett., 32, L671, doi:1.129/25gl Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood, 2: The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 16, Hartmann, D. L., 1994: Global Physical Climatology. Academic Press, 411 pp. Hasumi, H., and S. Emori, 24: K-1 coupled GCM (MIROC) description. K-1Tech. Rep. 1, Center for Climate System Research, University of Tokyo, 34 pp. Hibler, W. D., 1979: A dynamic thermodynamic sea ice model. J. Phys. Oceanogr., 9, Isemer, H., J. Willebrand, and L. Hasse, 1989: Fine adjustment of large-scale air sea energy flux parameterizations by direct estimates of ocean heat transport. J. Climate, 2, Jackson, C. S., and A. J. Broccoli, 23: Orbital forcing of Arctic climate: Mechanisms of climate response and implications for continental glaciation. Climate Dyn., 21, Jansen, E., and Coauthors, 27: Palaeoclimate. Climate Change 27: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, Jouzel, J., and Coauthors, 27: Orbital and millennial Antarctic climate variability over the past 8, years. Science, 317, Key, R. M., and Coauthors, 24: A global ocean carbon climatology: Results from Global Data Analysis Project (GLODAP). Global Biogeochem. Cycles, 18, GB431, doi:1.129/24gb2247. Kukla, G., and D. Robinson, 198: Annual cycle of surfae albedo. Mon. Wea. Rev., 18, Labeyrie, L. D., J.-C. Duplessy, J. Duprat, A. Juillet-Leclerc, J. Moyes, E. Michel, N. Kallel, and N. J. Shackleton, 1992: Changes in the vertical structure of the North Atlantic Ocean between glacial and modern times. Quat. Sci. Rev., 11, Levitus, S., and T. Boyer, 1994: Temperature. Vol. 4, World Ocean Atlas 1994, NOAA Atlas NESDIS 4, 117 pp., R. Burgett, and T. Boyer, 1994: Salinity. Vol. 3, World Ocean Atlas 1994, NOAA Atlas NESDIS 3, 99 pp. Li, S., and A. Jarvis, 29: Long run surface temperature dynamics of an A-OGCM: The HadCM3 43CO 2 forcing experiment revisited. Climate Dyn., 33, Lisiecki, L. E., and M. E. Raymo, 25: A Pliocene Pleistocene stack of 57 globally distributed benthic d 18 O records. Paleoceanography, 2, PA13, doi:1.129/24pa171. Liu, Z., and Coauthors, 29: Transient simulation of last deglaciation with a new mechanism for Bølling-Allerød warming. Science, 325, Loulergue, L., and Coauthors, 28: Orbital and millennial-scale features of atmospheric CH 4 over the past 8, years. Nature, 453, Lumpkin, R., and K. Speer, 27: Global ocean meridional overturning. J. Phys. Oceanogr., 37, Lüthi, D., and Coauthors, 28: High-resolution carbon dioxide concentration record 65, 8, years before present. Nature, 453, Lynch-Stieglitz, J., and Coauthors, 27: Atlantic meridional overturning circulation during the Last Glacial Maximum. Science, 316, Mahowald, N. M., D. R. Muhs, S. Levis, P. J. Rasch, M. Yoshioka, C. S. Zender, and C. Luo, 26: Change in atmospheric mineral aerosols in response to climate: Last glacial period, preindustrial, modern, and doubled carbon dioxide climates. J. Geophys. Res., 111, D122, doi:1.129/25jd6653. Marchitto, T. M., and W. S. Broecker, 26: Deep water mass geometry in the glacial Atlantic Ocean: A review of constraints from the paleonutrient proxy Cd/Ca. Geochem. Geophys. Geosyst., 7, Q123, doi:1.129/26gc1323., S. J. Lehman, J. D. Ortiz, J. Flückiger, and A. van Geen, 27: Marine radiocarbon evidence for the mechanism of deglacial atmospheric CO 2 rise. Science, 316, Martin, F., 22: Klimamodellierung mit einem vereinfachten Modell höherer Auflösung. M.S. thesis, Abteilung für Klimaund Umweltphysik, Physikalisches Institut der Universität Bern, 89 pp.

62 6 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE 15 JANUARY 211 R I T Z E T A L. 375 Martin, P. A., D. W. Lea, Y. Rosenthal, N. J. Shackleton, M. Sarnthein, and T. Papenfuss, 22: Quaternary deep sea temperature histories derived from benthic foraminiferal Mg/ Ca. Earth Planet. Sci. Lett., 198, Martrat, B., J. O. Grimalt, N. J. Shackleton, L. de Abreu, M. A. Hutterli, and T. F. Stocker, 27: Four climate cycles of recurring deep and surface water destabilizations on the Iberian Margin. Science, 317, McPhee, M., 1992: Turbulent heat flux in the upper ocean under sea ice. J. Geophys. Res., 97, Meissner, K. J., and R. Gerdes, 22: Coupled climate modelling of ocean circulation changes during ice age inception. Climate Dyn., 18, Montoya, M., A. Griesel, A. Levermann, J. Mignot, M. Hofmann, A. Ganopolski, and S. Rahmstorf, 25: The earth system model of intermediate complexity CLIMBER-3a: Part I: Description and performance for present-day conditions. Climate Dyn., 25, Müller, S. A., F. Joos, N. R. Edwards, and T. F. Stocker, 26: Water mass distribution and ventilation time scales in a costefficient, three-dimensional ocean model. J. Climate, 19, ,, G.-K. Plattner, N. R. Edwards, and T. F. Stocker, 28: Modeled natural and excess radiocarbon: Sensitivies to the gas exchange formulation and ocean transport strength. Global Biogeochem. Cycles, 22, GB311, doi:1.129/ 27GB365. Muscheler, R., F. Joos, J. Beer, S. A. Müller, M. Vonmoos, and I. Snowball, 27: Solar activity during the last 1 yr inferred from radionuclide records. Quat. Sci. Rev., 26, Myrhe, G., E. J. Highwood, K. P. Shine, and F. Stordal, 1998: New estimates of radiative forcing due to well mixed greenhouse gases. Geophys. Res. Lett., 25, Otto-Bliesner, B. L., C. D. Hewitt, T. M. Marchitto, E. Brady, A. Abe-Ouchi, M. Crucifix, S. Murakami, and S. L. Weber, 27: Last Glacial Maximum ocean thermohaline circulation: PMIP2 model intercomparisons and data constraints. Geophys. Res. Lett., 34, L1276, doi:1.129/27gl Pardaens, A. K., H. T. Banks, J. M. Gregory, and P. R. Rowntree, 23: Freshwater transports in HadCM3. Climate Dyn., 21, Parekh, P., F. Joos, and S. A. Müller, 28: A modeling assessment of the interplay between aeolian iron fluxes and iron-binding ligands in controlling carbon dioxide fluctuations during Antarctic warm events. Paleoceanography, 23, PA422, doi:1.129/27pa1531. Peltier, W. R., 1994: Ice age paleotopography. Science, 265, Randall, D. A., and Coauthors, 27: Climate models and their evaluation. Climate Change 27: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 23: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 18, 447, doi:1.129/22jd267. Reimer, P. J., and Coauthors, 24: IntCal4 terrestrial radiocarbon age calibration, 26 cal kyr BP. Radiocarbon, 46, Ritz, S. P., T. F. Stocker, and S. A. Müller, 28: Modeling the effect of abrupt ocean circulation change on marine reservoir age. Earth Planet. Sci. Lett., 268, Semtner, A. J., 1976: A model for the thermodynamic growth of sea ice in numerical investigations of climate. J. Phys. Oceanogr., 6, Shi, G., 1992: Radiative forcing and greenhouse effect due to the atmospheric trace gases. Sci. China, 35B, Siddall, M., T. F. Stocker, G. M. Henderson, F. Joos, M. Frank, N. R. Edwards, S. P. Ritz, and S. A. Müller, 27: Modeling the relationship between 231 Pa/ 23 Th distribution in North Atlantic sediment and Atlantic meridional overturning circulation. Paleoceanography, 22, PA2214, doi:1.129/26pa1358. Sitch, S., and Coauthors, 23: Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model. Global Change Biol., 9, Stenni, B., V. Masson-Delmotte, S. Johnsen, J. Jouzel, A. Longinelli, E. Monnin, R. Rothlisberger, and E. Selmo, 21: An oceanic cold reversal during the last deglaciation. Science, 293, Stephens, G. L., 1978: Radiation profiles in extended water clouds. II: Parametrization schemes. J. Atmos. Sci., 35, Stocker, T. F., D. G. Wright, and L. A. Mysak, 1992: A zonally averaged, coupled ocean atmosphere model for paleoclimate studies. J. Climate, 5, Strassmann, K. M., F. Joos, and G. Fischer, 28: Simulating effects of land use changes on carbon fluxes: Past contributions to atmospheric CO 2 increases and future commitments due to losses of terrestrial sink capacity. Tellus, 6B, Thompson, S. L., and S. G. Warren, 1982: Parameterization of outgoing infrared radiation derived from detailed radiative calculations. J. Atmos. Sci., 39, Trenberth, K. E., and J. M. Caron, 21: Estimates of meridional atmosphere and ocean heat transports. J. Climate, 14, Tschumi, T., 29: Modeling the ocean s contribution to past and future changes in global carbon cycling. Ph.D. thesis, University of Bern, 15 pp., F. Joos, and P. Parekh, 28: How important are Southern Hemisphere wind changes for low glacial carbon dioxide? A model study. Paleoceanography, 23, PA428, doi:1.129/ 28PA1592. Tuenter, E., S. Weber, F. J. Hilgen, L. J. Lourens, and A. Ganopolski, 25: Simulation of climate phase lags in response to precession and obliquity forcing and the role of vegetation. Climate Dyn., 24, Weaver, A. J., and Coauthors, 21: The UVic Earth System Climate Model: Model description, climatology, and application to past, present, and future climates. Atmos. Ocean, 39, Weber, S. L., and Coauthors, 27: The modern and glacial overturning circulation in the Atlantic Ocean in PMIP coupled model simulations. Climate Past, 3, Weijer, W., W. P. M. De Ruijter, H. A. Dijkstra, and P. J. van Leeuwen, 1999: Impact of interbasin exchange on the Atlantic overturning circulation. J. Phys. Oceanogr., 29, Zaucker, F., T. F. Stocker, and W. S. Broecker, 1994: Atmospheric freshwater fluxes and their effect on the global thermohaline circulation. J. Geophys. Res., 99,

63 2.2. EXTENDED MODEL DESCRIPTION Extended Model Description The physics and the modern state of the atmospheric energy and moisture balance model (EBM) component of the Bern3D model are thoroughly described by Ritz et al. (211). In this section the shortwave radiation parametrization, the daily solar irradiation function, and the precipitation runoff scheme used in the model are presented in more detail. Also, additional model features that are not part of the standard model setup and therefore not described by Ritz et al. (211) are described. These are a simple snow albedo parametrization, and a bucket model of land hydrology Shortwave Radiation The shortwave radiation entering the atmosphere is determined using a parametrization of Bintanja (1996). The radiation transfer through the atmosphere is calculated separately for the clear-sky (cl) and the overcast (ov) case (Fcl down, Fov down ) and is equal to the incoming shortwave radiation at the top of the atmosphere (Fsw TOA, see Section 2.2.2) multiplied by the total transmissivity of the atmosphere: F down cl = [T oz T ra T Ta (T a )T α (α)t µ (µ) + βz]f TOA sw. (2.1) The total transmissivity is approximated by multiplying individual transmissivities T. The transmissivities are constants if no argument is given, and T(x) = a +a 1 x+a 2 x 2 +a 3 x 3 +a 4 x 4 otherwise. The subscript oz stands for ozone, ra for Rayleigh scattering, T a for the surface air temperature (the amount of water vapor in the entire atmospheric column is parametrized as a function of surface air temperature), α for the surface albedo and µ = cos ζ for the cosine of the solar zenith angle, where ζ = max ( π ( π )) 2,min 2,δ ϑ. (2.2) δ is the declination of the Sun, i.e., the latitude of the Sun on the celestial sphere, derived in Section 2.2.2, and ϑ { π/2;π/2} the latitude. Land surface albedo is taken from a zonally averaged climatology by Kukla & Robinson (198). For the ocean/sea-ice albedo a simple parametrization is used. A table containing all the coefficients of equation (2.1) can be found in Bintanja (1996). The transmissivity increases approximately linearly with the surface height z with proportionality factor β = km 1 (Bintanja, 1996). z is taken from the altitude map ETOPO5 (data available online at global/etopo5.html) regridded and smoothed onto the model grid. For the overcast case, reflection of shortwave radiation in both upward and downward direction by the clouds, as well as absorption through water vapor need to be taken into account. These processes are represented by the cloud optical depth τ and are parametrized by the relative transmissivity T τ (α) = T τ (.2) + α.2 [T τ (.8) T τ (.2)] (2.3).6 (Bintanja, 1996). As above, T τ is a polynomial of fourth order. F down ov = [T oz T ra T Ta (T a )T α (α)t µ (µ)t τ (α)λ + βz]f TOA sw, (2.4) where λ =.628 is a factor equal to the ratio between the total transmissivity in overcast and in clear-sky conditions for the standard cases (Bintanja, 1996). Again, the coefficients of the polynomials are listed in Bintanja (1996).

64 62 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE The fraction of the shortwave radiation which penetrates into the ocean and the land surface is calculated as follows: F BOA sw = (1 α)[(1 ξ)f down cl + ξf down ov ], (2.5) where ξ is the zonally averaged fractional cloud cover climatology taken from the 4-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-4). In opposition to Bintanja (1996), we do not distinguish between clear-sky and overcast surface albedo for reasons of simplicity. The upward shortwave radiation at the top of the atmosphere (TOA) is again different for the clear-sky and overcast case: and F up cl = G cl F down cl T Ta T α T µ (α) (2.6) F up ov = G ov F down ov T Ta T α T µ T τ (α). (2.7) G cl =.187 is the fraction of Fcl down that is reflected in the standard case (similar to the planetary albedo), and analogous for G ov =.324 (Bintanja, 1996). Bintanja (1996) found that the dependence of Fov up on µ itself depends on the surface albedo. T µ (α) is calculated as in equation (2.3). Note that the coefficients of the various transmissivities of equations (2.6) and (2.7) are not equal to the coefficients of the downward radiation case. Finally, the upward shortwave radiation at TOA is F up sw = (1 ξ)f up cl + ξf up ov. (2.8) The planetary albedo, the global averaged fraction between outgoing and incoming shortwave radiation, is calculated to be α p =.29 in the modern control simulation. This matches well with observations of Goode et al. (21), who measured α p =.297 ±.5. Cloud Optical Depth Cloud optical depth τ is a measure of cloud transparency and is defined as follows: I I = e τ, (2.9) where I is the radiation intensity before entering the cloud and I the intensity after exiting the cloud. In Bintanja (1996), the cloud optical depth τ is set to a constant value (τ = 5). Here, following a parametrization of Stephens (1978), τ is a function of the liquid water path, W = q a ρ a h q, the integrated amount of water in the atmospheric column (in kgm 2 ). q a is the surface specific humidity, ρ a = 1.25 kg m 3 the reference density of air, and h q = 18 m the moisture scale height. Stephens (1978) distinguishes visible and infrared radiation wavelength intervals (Fig. 2.1): i).3 µm λ.75µm (visible) ii).75 µm λ 4.µm (infrared) log 1 (τ) = log e [log 1 (W/W )]. (2.1) log 1 (τ) = log e [log 1 (W/W )], (2.11)

65 2.2. EXTENDED MODEL DESCRIPTION 63 cloud optical depth µm λ.75 µm.75 µm λ 4. µm liquid water path [g/m 2 ] Figure 2.1: Cloud optical depth as a function of liquid water path for two wavelength intervals (Stephens, 1978). with W = 1 kgm 2 and W 2 kgm 2. For reasons of simplicity and since the model does not distinguish wavelengths, only interval ii) is implemented into the model. However, using interval i) instead of ii) would affect model results only marginally. Using this cloud optical depth parametrization in the model instead of the constant value proposed by Bintanja (1996) results in an increase of outgoing shortwave radiation and thus to a reduction of temperatures in the atmosphere and the ocean Daily Solar Irradiation Solar irradiation as a function of latitude ϑ and time is computed using series expansions for the orbital parameters eccentricity e, obliquity ǫ and the longitude of the perihelion ω s measured from the moving vernal equinox (Fig. 2.2) (Berger, 1978). The following orbital parameters and solar irradiation values are valid for the time period from the present day back to 8 kyr BP (before present, before 195 AD). For time periods back to 1 million years BP, the orbital parameters and solar irradiation values from Berger & Loutre (1991) should be used. The eccentricity e and the longitude of the perihelion π are calculated from the series expansions 19 ( gi ) e(t)sin π(t) = M i sin 36 t + β i (2.12) and e(t)cos π(t) = i=1 19 i=1 ( gi ) M i cos 36 t + β i. (2.13) The time t = (in years) refers to year BP (year 195 AD) and is negative for the past. The perihelion longitude refers to a position relative to the fixed stars. The coefficients M i, g i and β i can be found in Table 4 of Berger (1978) or downloaded from

66 64 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE fixed stars WW P p ν π S ω s λ ψ a A γ E SS Figure 2.2: Elements of the Earth s orbit. E is the Earth with its obliquity ǫ, S the Sun, P the perihelion and A the aphelion, the locations where the distance between Earth and Sun are nearest and farthest, respectively, SS and WW the summer and winter solstice (the positions where the axis of the Earth is most oriented toward or away from the Sun) and γ the vernal equinox (where North and South Pole are of equal distance from the Sun). The eccentricity e is given by (a p)/(a+p). ω s is the longitude of the perihelion relative to the moving vernal equinox. ω s = mod (ψ +π,36 ), where ψ is the longitude of the vernal equinox relative to a position fixed by the stars counted clockwise, while π describes the motion of the perihelion relative to the fixed stars. λ is the longitude of the Earth relative to the vernal equinox and ν the longitude of the Earth relative to the perihelion. The thin arrows denote the motion of the Earth, the perihelion, and the vernal equinox. ac.be/index.php?page=astronomicalinsolationforcing. e(t) and π(t) are determined by combining Eqs. (2.12) and (2.13) using trigonometric identities. The obliquity ǫ(t) (in degrees) is computed using the expansion ǫ(t) = i=1 ( ) A i fi cos t + δ i. (2.14) The coefficients A i, f i and δ i can be found in Table 1 of Berger (1978). The general precession in longitude ψ(t) relative to the position fixed by the stars is given by ψ(t) = t i=1 ( ) F i f sin i t + δ i. (2.15) F i, f i, and δ i are listed in Table 5 of Berger (1978). The longitude of the perihelion relative to the moving vernal equinox ω is then given by ω(t) = mod (π(t) + ψ(t) + 18,36 ). (2.16) Note that ω differs from ω s (Fig. 2.2) by 18. This is due to a conversion from the heliocentric to the geocentric coordinate system (see Berger et al. (1993) for details). The next step is to calculate the longitude of the Earth relative to the vernal equinox λ = ν+ ω for a given date. ν is the positional angle of the Earth on its orbit, counted from the perihelion (Fig. 2.2). Because the orbit of the Earth is not a circle, the angular velocity of the Earth is not constant. In the following, the true Earth is transformed to a mean Earth which travels at a constant angular speed equal to 2π/ days 1. λ m for the mean Earth

67 2.2. EXTENDED MODEL DESCRIPTION 65 is calculated according to Berger (1978): [( ) e ( λ m = e3 1 + ) 1 e 8 2 sin ω e2 4 ( + e ) 1 e 2 ] sin(3 ω) + (n d 8 days) ( ) 1 e 2 sin(2 ω) days, (2.17) where n d is the number of days in a year of 365 days. The offset of 8 days represents March 21, the date of the vernal equinox. Subsequently, λ m is translated to the true longitude of the Earth: λ = λ m + (2e )sin e3 (λ m ω) + 5e2 4 4 sin[2(λ m ω)] + 13e3 12 sin[3(λ m ω)]. (2.18) Finally, the daily solar irradiation at latitude ϑ is computed using ρ = 1 e2 1 + ecos ν, sin δ = sinǫsin λ, cos H = tan ϑ tan δ, tan δ = sinδ cos δ, cos δ = 1 sin 2 δ, S = S ρ 2, S = S π, (2.19) where ρ is the Earth-Sun distance measured in units of the semi-major axis a (Fig. 2.2), δ the declination of the Sun, H the absolute value of the hour angle at sunrise and sunset, and S = 1353 W m 2 the solar constant (Berger, 1978). Note that the most recent estimate of the solar constant is 1361 Wm 2 (Kopp & Lean, 211). However, this value is not used in the standard version of the model. A small test quantity ε =.1 is used to increase computational efficiency in special situations: δ < ε, ϑ 9 < ε S cos ϑ δ < ε, ϑ 9 ε S cos δ ϑ > ε F TOA sw = ϑ > (9 δ ), ϑδ < S sin ϑ sin ( δ ) ϑ > (9 δ ), ϑδ > S [arccos sin δ cos H sin ϑ sin δ cos δ + cos ϑ cos δ ] 1 sin 2 H otherwise. (2.2) Runoff Precipitation P over land is instantly transported to the ocean as runoff. Its pathway is defined by a pseudo-elevation map: Every land cell contains information on the direction of the runoff (North, South, West or East) (Fig. 2.3). The freshwater input into the ocean by runoff is an essential quantity for the stability and strength of the overturning circulation. In the model, a stable Atlantic meridional overturning circulation is only obtained when a

68 66 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE Runoff Mask Ocean East South West North Figure 2.3: River runoff flow path. Each grid cell indicates in which direction the water is passed on. parameter χ is introduced that defines the relative amount of precipitation over land that is transported to the ocean as runoff. The rest is distributed uniformly into every surface ocean grid cell. χ is taken from observations (Table 2.1) (Hartmann, 1994). Note that it would physically make more sense to redirect (1 χ) P back to the atmosphere, since this is the fraction which is taken up by the soils and eventually re-evaporated to the atmosphere. However, due to the parametrization of the model, a large part of the re-evaporated moisture would be precipitated again in the following timestep strongly increasing the atmosphereocean moisture turnover and thus the amount of runoff Bucket Model of Land Hydrology The reason that the fractional runoff χ (see Section 2.2.3) is smaller than 1 is that in reality precipitation can fall as snow or it can be taken up by soils and later be re-evapotranspired to the atmosphere. Here, the possibility is introduced for land masses to store water by im- Table 2.1: Continental values for fractional runoff χ and precipitation from observations (Hartmann, 1994). Continent χ P obs [mm/yr] Europe Asia Africa Australia North America South America Antarctica

69 2.2. EXTENDED MODEL DESCRIPTION 67 plementing a simple bucket model of land hydrology, similar to those described by Hartmann (1994) and Williamson et al. (26). The bucket model would replace the fractional runoff parametrization. Although the bucket model is implemented into the model, it is not used in the version described in Ritz et al. (211). Similar to the problems described in Section 2.2.3, when using the bucket model, too much runoff enters the ocean which destabilizes the ocean circulation. The reason is a too high sensitivity of the deep water formation in the North Atlantic to the riverine input in the Arctic Ocean. This is possibly due to the very coarse resolution of the high latitudes in the model version of Ritz et al. (211). Perhaps the bucket model will become an option when the model is run with higher resolution at high latitudes. For every land surface grid cell, h w t = P r E + M R, (2.21) where h w is the soil moisture content expressed as an equivalent water depth in the bucket of the grid cell (in m), P r the precipitation when it falls as rain, M the melting rate of snow and R runoff. Precipitation falls as snow (P s ) whenever the land temperature T l C. The snow layer height equivalent h s (being snow with the density of liquid water) is determined by solving h s t = P s E s M, (2.22) with E s the sublimation rate of snow. Snow is melted when T l > C, thus bringing T l back down to C: c p,l ρ l h l Tl K T l > K M = L f ρ o t (2.23) otherwise (Hartmann, 1994). c p,l is the land reference specific heat capacity, ρ l the reference land density, h l the land-surface scale height, L f the latent heat of fusion of ice and ρ o the density of water. The values of the parameters are given in Ritz et al. (211). M may not exceed h s / t. Runoff is only greater than zero when the soil moisture content exceeds the soil moisture capacity h c = 15 cm (Hartmann, 1994). Then, R = 1 t (h w h c ) h w > h c otherwise. (2.24) Potential evaporation over land is calculated in a similar way as the evaporation over the ocean: E p = ρ ac W u ρ o [q s (T l ) q a ], (2.25) where C W is the transfer coefficient of moisture and u the surface wind speed at 1 m above ground. We assume C W = C H, the transfer coefficient of heat over land, which is determined using C H = D l (2.26) ρ a C a u

70 68 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE with D l the bulk coefficient for sensible heat on land and c p,a the specific heat of air given in Ritz et al. (211). The transpiration of vegetation is simply taken into account by including the parameter β E : E = β E E p, (2.27) with 1. h w h v β E = h w hv otherwise. (2.28) The transpiration rate of plants is dependent on the soil moisture content. Up to a certain moisture content h v, vegetation can transpire at the rate of potential evaporation E p. The sublimation of snow is calculated as in equation (2.25), but the saturation specific humidity q s (T l ) of ice is used: ( ) c4 (T i K) q s (T i ) = c 1 exp, (2.29) c 5 + T i K where c 1 = 3.8 g kg 1, c 4 = 21.87, and c 5 = K (Bolton, 198) Snow Albedo Parametrization As an option, a simple parametrization of snow albedo has been implemented into the model. It is not used in the version described in Ritz et al. (211). The parametrization relates land surface albedo to atmospheric temperature. When air temperature is too warm for snow, a minimum land albedo α min is prescribed. This value could be taken from the mask of Kukla & Robinson (198). At around C, land albedo becomes temperature dependent by assuming a relationship between snow cover and atmospheric temperature. Below a certain temperature, the area is assumed to be completely snow covered and the albedo reaches a maximum value α max. The parametrization also takes into account seasonality. In spring, warmer temperatures are required to melt the snow than in autumn to build up a snow layer, because in autumn the ground needs to be cooled first. The land surface albedo is parametrized as follows: α(t a,t) = α min + b(1 tanh{a[t a T a, + T sin(ωt + ϕ)]}), (2.3) where b = (α max α min )/2, ω = 2π/(365 days) and t is the time in days (Fig. 2.4). The parameters are explained in Table 2.2. The parameters have not yet been tuned, but preliminary values are given in Table 2.2. The parametrization is not applied over Antarctica, because the continent is covered by snow all year. 2.3 Discretizations This section describes in detail how the energy and moisture balance equations, the land temperature equation, and the sea-ice equations of the EBM were discretized in the model. Also, the method used in the model to transform freshwater fluxes into salt fluxes is discussed.

71 2.3. DISCRETIZATIONS 69 α max b α T Spring Fall α min T a, T a Figure 2.4: Snow albedo parametrization. The parameters are described in Table 2.2. Table 2.2: Parameter description and preliminary values of the snow albedo parametrization. Parameter Value Description α min.25 Land albedo, when there is no snow. Could also be taken from Kukla & Robinson (198). α max.6 Land albedo, when the ground is fully covered by snow. T a, C Temperature where the rate of albedo change is largest. T 1 C Seasonality effect. ϕ Phase lag of α. a 1 Measure of the temperature bandwidth between α min and α max Energy Balance Equation The horizontal transport of the energy balance equation h a ρ a c p,a t T a = h ( aρ a c p,a 1 r 2 cos 2 ϑ ϕ K ϕ ϕ T a + ) (sin ϑ) cos2 ϑk ϑ (sin ϑ) T a + Fsw TOA + F AO Fsw up F lw TOA total + L ( vρ o P E AO ) ; Ftotal AI + L ( sρ o P E AI ) over ocean ; over sea ice F AL total + L vρ o P ; over land (2.31) is solved implicitly. This permits an EBM timestep equal to the ocean timestep. For the vertical fluxes, T a of the previous timestep is used. Note that K ϕ is constant and K ϑ is a function of latitude (Ritz et al., 211). The derivatives are discretized as follows: t T a = 1 t ϕ T a = 1 ϕ ( T t i,j T t t i,j ), (2.32) ( T t i+1,j T t i,j). (2.33)

72 7 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE i and j are the grid-cell indices in zonal and meridional direction, respectively (Fig. 2.5). (K ϕ is constant) and 2 ϕ 2 T a = 1 ( ϕ) 2 (T t i+1,j 2T t i,j + T t i 1,j) (2.34) (sin ϑ) cos2 ϑk m (ϑ) (sin ϑ) T 1 a = [ (sin ϑ)] 2[cos2 ϑ e jkm(t j i,j+1 t T t cos 2 ϑ e j 1Km j 1 (Ti,j t Ti,j 1)]. t i,j) ϑ e stands for ϑ at the edge of the grid cell. ϕ = 1 and (sin ϑ) = 1/18 are constant. The discretized equation (2.31) can be written in the form T1,1 t b 1,1 T2,1 t b 2,1 M. Ti t = max,1 T1,2 t.. b imax,1 b 1,2. (2.35), (2.36) where i max is the number of grid cells in zonal direction, M a banded i max j max i max j max matrix with i max super- and i max subdiagonals and b a vector containing all the terms without Tm,n t such as the vertical fluxes. To solve this system of equations, matrix M is inverted using a special routine for banded matrices. Since M only depends on K ϕ and K ϑ, it is constant and must only be inverted once at the beginning of the model run Land Temperature Equation T l ρ l c p,l h l t = F total AL (2.37) for the land temperature is solved using the Euler forward discretization scheme as in equation (2.32). However, the timestep is halved in order for T l to converge. Thus, T t l = T t t l l t l ρ l c p,l h l F AL total (T t t l l ), (2.38) where t l = t/2 is the timestep of the land surface. Equation (2.38) is calculated twice per atmospheric timestep. In order for the energy budget to close, the land-atmosphere fluxes of both land timesteps are averaged for the determination of T a Moisture Balance Equation Moisture is transported by advection and diffusion. The vertical fluxes are given by evaporation and precipitation. In contrast to the energy transport, meridional advection is important in tropical regions. Moisture is transported equatorwards in the Intertropical Convergence (ITC), where precipitation occurs. The moisture balance equation is formulated as follows: t q a = 1 r cos ϑ ϕ uq 1 a + r 2 cos 2 ϑ ϕ Kq ϕ ϕ q a 1 r (sin ϑ) cos ϑvq a + 1 r 2 (sin ϑ) Kq ϑ cos2 ϑ (sin ϑ) q a + ρ o ρ a h q (E P). (2.39)

73 2.3. DISCRETIZATIONS 71 1 E 11 E 12 E 13 E j, Latitude i, Longitude 56 S 63 S 7 S 9 S Figure 2.5: How the indices i and j are used on the grid of the Bern3D model. Solid lines and bold numbers indicate the borders of the grid cells, while dashed lines and regular numbers denote the centers of the grid cells. Tracers such as temperature and salinity are calculated at the center of each grid cell (squares), the advective and diffusive fluxes are calculated at the borders (circles). u and v are zonally averaged wind velocity climatologies from ECMWF ERA-4 reanalysis (and hence only a function of latitude and time), and Kϕ q and K q ϑ are constant (see Ritz et al. (211) for details). The discretization of the moisture balance equation (2.39) is done using the same schemes as described in Section The additional zonal advection term is solved by using a simple weighted upwind scheme: where and ϕ (u q a) = 1 ( F u ϕ i,j Fi 1,j) u, (2.4) [( ) ( ) ] 1 1 Fi,j u = ut j 2 ft j qi+1,j t ft j qi,j t fj t = 1 2 t x j u t j 1 2 t x j u t j > 1 2 otherwise t x j u t j < 1 2. (2.41) (2.42) x j is the distance between the centers of the two grid cells bordering the velocity u j. Note that a more sophisticated weighted upwind scheme as described by Fiadeiro & Veronis (1977) might improve the result (Eq. 2.49). Since wind velocities are substantially lower in meridional direction, a centered-differences scheme is used to discretize this term: (sin ϑ) (cos ϑvq a) = 1 2 (sin ϑ) [cos ϑe jv j (q t i,j+1 + q t i,j) cos ϑ e j 1v j 1 (q t i,j + q t i,j 1)]. (2.43) In analogy to equation (2.36), the system of equations has the form M q q a t = ρ o 1 E + t t q a. (2.44) ρ a h q t Matrix M q depends on K q ϕ, K q ϑ, and on the wind velocities. The winds are prescribed and vary seasonally, but not interannually. Therefore, matrix M q also only needs to be calculated during the first year. In order to suppress oscillations in evaporation and precipitation

74 72 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE amount, precipitation is taken into account before solving equation (2.44) for q a : When precipitation ρ a h a P = ρ o t (q a r h,precip q s (T a )), r h > r h,max (2.45), otherwise is calculated and greater than zero, q a is set to r h,precip q s (T a ) according to equation q a = r h q s (T a ) Sea Ice Fractional sea-ice area and sea-ice height satisfy the balance equations ( A ice + t ( u A ice ) K ice 2 A ice = max, 1 A ice F IO F AO ) total H ρ ice L f ( A 2 ( ice F IO Ftotal AI + min, E ρ )) (2.46) o 2H ice ρ ice L f ρ ice and H ice t + ( u H ice ) K ice 2 H ice = (1 A ice )max (, F IO Ftotal AO ρ ice L f ( F IO Ftotal AI + A ice E ρ ) o, ρ ice L f ρ ice ) (2.47) where u = (u oc,v oc ) is the horizontal surface ocean velocity vector and K ice the constant sea-ice diffusivity (see Ritz et al. (211) for details). Note that the fractional sea-ice area, sea-ice height, ice density, and ice diffusivity are denoted A ice, H ice, ρ ice, and K ice, respectively instead of A i, H i, ρ i, and K i as in Ritz et al. (211) in order to prevent confusion with the grid-cell indices. A weighted upstream method as described by Fiadeiro & Veronis (1977) is used to discretize the advective term (here in the example of sea-ice height): H ice t + ( u H ice ) = H ice t = Ht i,j Ht t i,j t + 1 r cos ϑ ϕ (u och ice ) + 1 r r (sin ϑ) (cos ϑe jvi,j t t sin ϑ (cos ϑv och ice ) 1 r cos ϑ c j ϕ(ut t i,j H i,j t t ui 1,j t t t t H i 1,j ) Ĥ t t i,j cos ϑ e j 1v i,j 1Ĥt t t t ), (2.48) where H i,j = 1 2 with the Peclet number and Ĥ i,j = ) [(1 + Pect t i,j Pec t t Hi,j t t + i,j + 2 Pec t t i,j Pec t t i,j Pec t t H t t i,j + 2 ) ] (1 Pect t i,j Pec t t Hi+1,j t t, (2.49) i,j + 2 = ut t i,j x j, (2.5) K ice i,j + 1 Pec t t i,j Pec t t H t t i,j+1, (2.51) i,j + 2

75 2.3. DISCRETIZATIONS 73 with the Peclet number Pec t t i,j = vt t i,j y j. (2.52) K ice x j and y j are the distances between the centers of the two grid cells bordering the velocities u i,j and v i,j, respectively. ϑ c and ϑ e stand for ϑ at the center and at the edge of the grid cell, respectively. For the diffusive term, even though K ice is spatially constant, it is simpler to write it in terms of ϕ and ϑ, since the diffusivity is zero on land-ocean borders: K ice 2 H ice = K ice (ϕ,ϑ) H ice 1 = r 2 cos 2 ϑ ϕ K ice(ϕ,ϑ) ϕ H ice + 1 r 2 (sin ϑ) cos2 ϑk ice (ϕ,ϑ) (sin ϑ) H ice 1 = r 2 cos 2 ϑ c j ( ϕ)2[k i,j(hi+1,j t t Ht t i,j ) K i 1,j (Hi,j t t Hi 1,j t t )] + 1 r 2 ( sin ϑ) 2[cos2 ϑ e j K i,j (H t t i,j+1 Ht t i,j ) cos 2 ϑ e j 1 K i,j 1 (H t t i,j H t t i,j 1 )]. (2.53) Unlike for heat and moisture transport, equations (2.47) and (2.46) are not solved implicitly. The equation for ice temperature T ice (denoted T i in Ritz et al. (211)) with F AI total + I cond H ice (T f T ice ) =, (2.54) F AI total = F BOA sw σε o T 4 ice + σε at 4 a + F AI sh L sρ o E AI (2.55) cannot be solved analytically. The following Taylor-approximation for Tice 4 at time t is used: ) ( Tice,t 4 = (T ice,t t + T ice,t ) 4 = Tice,t t T ) 4 ( ice Tice,t t T ice,t t = Tice,t t 4 + 4Tice,t t(t 3 ice,t T ice,t t ) = 3Tice,t t 4 + 4Tice,t tt 3 ice,t, with T ice = T ice,t T ice,t t Freshwater Flux T ice T ice,t t (2.56) Because the Bern3D model is a rigid-lid model, the volume of the ocean remains constant. To simulate freshwater fluxes in the model, not freshwater, but salt is added and taken out of the ocean, respectively. The conversion between freshwater flux Ffw O and salt flux F salt O is done as follows: ( Fsalt O = (S ref + S) 1 V ) z x + V t, (2.57) with S ref = psu a reference salinity for the surface ocean, S = S S ref the salinity offset from the reference value, V and z the volume and height of the surface cell, respectively, and x the amount of freshwater added to the ocean. However, this method violates the salinity budget. This is avoided by setting S to zero. Thus, the salt flux is no longer dependent on the surface ocean salinity, which is different in every grid cell, but only on the reference salinity. This approximation closes the salinity budget.

76 74 2. THE BERN3D ENERGY AND MOISTURE BALANCE ATMOSPHERE Additionally, using a simple Taylor approximation, 1 V x + V = 1 1 ( x V x ) = x V V = η z, (2.58) where η is the fictional surface elevation owing to the freshwater input, 2.4 Model Setup and Options Fsalt O = S η ref t = S refffw O. (2.59) The steps described in Table 2.3 need to be carried out in order to use the coupled oceanatmosphere model. Tables 2.4 and 2.5 describe all possible preprocessor options of the energy balance model component. Important: In RUNNAME.ocn.parameter, the freshwater flux correction parameters must be set to: regionfrom: 1.1 regionto: 3.1 fwamount:.17 These settings are different from the settings of the mixed boundary conditions mode. Table 2.3: Steps required in order to use the coupled ocean-atmosphere model. The spinup of the biogeochemistry can be done after step 3 or during/after step 4 or 5. Note that the spinup of the biogeochemistry requires at least 5 model years. Model yrs Simulation Preprocessor options 1. 1, Ocean spinup under restoring boundary conditions -D atmtsres 2. 5 Diagnose EBM parameters -D atmtsres -D atmebmdiag 3. 5 EBM spinup with constant insolation -D atmebm -D ebm const insol 4. 1 EBM control simulation, diagnose steady state atmospheric temperatures (required for the water vapor feedback; Eq. 9 -D atmebm -D ebm ctrl diag -D ebm const insol in Ritz et al. (211)) 5. n Do whatever you want! -D atmebm -D ebm ctrl perturb

77 2.4. MODEL SETUP AND OPTIONS 75 Table 2.4: Preprocessor options of the energy balance model. Double horizontal lines divide between standard options, radiative forcing options and diagnostic options. -D atmebmdiag Use when running the model under restoring boundary conditions. Diagnoses the EBM parameters. -D atmebm Use to run the EBM. -D ebm const insol Keeps the solar irradiation constant over time. -D ebm ctrl diag Diagnose steady state atmospheric temperature. Required for -D ebm ctrl perturb. -D ebm ctrl perturb Enable water vapor radiative forcing feedback. -D ebm in rad co2 Requires in atmradco2=1 in forcing.parameter file; Read in CO 2 radiative forcing time series. -D ebm bgc rad co2 Use atm. CO 2 given by the biogeochemistry routine for the radiative forcing. -D ebm in rad ch4 Requires in atmch4=1 in forcing.parameter file; Read in CH 4 radiative forcing time series. -D ebm icesheet Use ice-sheet albedo forcing for paleo-simulations. Set in icesheet=1 in forcing.parameter file to read in Lisiecki & Raymo (25) δ 18 O for past transient ice sheet changes. -D ebm icesheet fw=1/ Account for freshwater relocation between ocean and ice sheet. if =1: Adjust freshwater anomaly to the modern state at the first timestep ( abrupt change). if =: Do not adjust (used e.g. after a restart). -D ebm icesheet heat Account for ice sheet latent heat release/take up. Note that too abrupt changes may lead to floating point exceptions. E.g., for LGM simulation: change δ 18 O from modern to LGM value slowly (within 2 kyr). -D ebm icesheet noradf Simulate ocean freshwater changes, but not surface albedo changes. -D ebm meantdiag Create global annual mean atmṫemperature time series output. -D ebm meanthemdiag Create hemispheric annual mean atmṫemperature time series output. -D ebm Tpolesdiag Create annual mean atm. temperature time series output at the poles. -D ebm runoffdiag Diagnose runoff. -D ebm toadiag Diagnose net radiative flux at top of atmosphere (TOA). Is a measure of how well the steady state of the model is. Table 2.5: Preprocessor options not used in the model version described in Ritz et al. (211). These options alter the state of the model. -D ebm topo Include land topography. Calculates atm. temperature above ground besides atm. temperature at sea-level pressure. Longwave outgoing radiation uses atm. temperature above ground instead of at sea-level pressure. -D ebm topo nocorr When using -D ebm topo: Still use atm. temperature at sea-level pressure for outgoing longwave radiation. -D ebm allowtbelow18k Allow atm. temperatures below 18 K. Some parametrizations are no longer valid, but the model will not explode. -D ebm land hydro Use the bucket model of land hydrology. -D ebm land snow Enable land snow cover in the bucket model. -D ebm snowalbedo Vary land albedo as a function of atm. temperature and season. Parameters have not yet been tuned.

78 76 BIBLIOGRAPHY Bibliography Berger, A. L., Long-term variations of daily insolation and quaternary climatic changes, Journal of the Atmospheric Sciences, 35, Berger, A. L. & Loutre, M. F., Insolation values for the climate of the last 1 million years, Quaternary Science Reviews, 1, Berger, A. L., Loutre, M.-F., & Tricot, C., Insolation and Earth s orbital periods, Journal of Geophysical Research, 98, 1,341 1,362. Bintanja, R., The parameterization of shortwave and longwave radiative fluxes for use in zonally averaged climate models, Journal of Climate, 9, Bolton, D., 198. The computation of equivalent potential temperature, Monthly Weather Reviews, 18, Fiadeiro, M. E. & Veronis, G., On weighted-mean schemes for the finite-difference approximation to the advection-diffusion equation, Tellus, 29, Goode, P. R., Qiu, J., Yurchyshyn, V., Hickey, J., Chu, M., Kolbe, E., Brown, C. T., & Koonin, S. E., 21. Earthshine observations of the Earth s reflectance, Geophysical Research Letters, 28(9), Hartmann, D. L., Global Physical Climatology, Academic Press, University of Washington. Kopp, G. & Lean, J. L., 211. A new, lower value of total solar irradiance: Evidence and climate significance, Geophysical Research Letters, 38. Kukla, G. & Robinson, D., 198. Annual cycle of surfae albedo, Monthly Weather Reviews, 18, Lisiecki, L. E. & Raymo, M. E., 25. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ 18 O records, Paleoceanography, 2, PA13. Ritz, S. P., Stocker, T. F., & Joos, F., 211. A coupled dynamical ocean - energy balance atmosphere model for paleoclimate studies, Journal of Climate, 24, Stephens, G. L., Radiation profiles in extended water clouds. II: Parametrization schemes, Journal of the Atmospheric Sciences, 35, Williamson, M. S., Lenton, T. M., Shepherd, J. G., & R., E. N., 26. An efficient numerical terrestrial scheme (ENTS) for Earth system modelling, Ecological Modelling, 198,

79 Chapter 3 A Quantitative Method to Reconstruct Atlantic Meridional Overturning Circulation Strength on Glacial-Interglacial Time Scales Abstract A new quantitative method to reconstruct past changes of Atlantic meridional overturning circulation (AMOC) and Antarctic bottom water (AABW) strength is presented using paleoclimatic reconstructions of atmospheric and ocean temperatures in combination with a climate model. It has been found that the ocean temperature history at any location and depth can be approximated by a linear combination of the global atmospheric temperature history and the evolution of the AMOC and the AABW. The regression coefficients of the linear combination are determined by a coupled ocean-atmosphere climate model. By providing information of deep ocean and sea surface temperatures obtained from marine sediments and of atmospheric temperatures from polar ice cores, the linear combination can be solved for the AMOC and AABW strength by a least mean squares approach. This reconstruction method has the advantage over Pa/Th- and δ 13 C-based reconstruction methods that the AMOC strength can be determined quantitatively. An AMOC reconstruction is performed for the last 33, years, and in higher resolution for the last deglaciation. A major problem of the method is that the uncertainties of the age scales of the temperature reconstructions are of several hundred years and can therefore significantly bias the result. This leads to the conclusion that the method is not suitable to reconstruct circulation during abrupt climatic changes such as Dansgaard-Oeschger events or during deglaciations. However, the method may provide quantitative insight on the magnitude and trends of the AMOC and AABW on glacial-interglacial timescales. 3.1 Introduction The Atlantic meridional overturning circulation (AMOC) is an important feature of the Earth s climate system today and in the past. It is characterized by northward flowing water masses at the surface ocean, deep water formation in the North and southward flowing water masses in the intermediate to deep ocean. Because of the large heat capacity of water, the northward flowing near-surface waters contribute effectively to the heat transport from the tropics to the mid- and high latitudes and therefore affect the regional climate in the north. Hence, changes of the AMOC strength can substantially influence climate in the

80 78 3. AMOC RECONSTRUCTION north, especially in Europe. A highly simplified image of the large scale circulation in the ocean is given in Fig In the modern ocean, the water masses sink in the Labrador Sea as well as in the Greenland-Iceland-Norwegian Seas to about 2 to 3 km depth forming North Atlantic Deep Water (NADW). However, deep water formation sites may have shifted in the past, and the depth of the southward flowing waters as well as the strength of the circulation were subject to changes. During the last glacial period, abrupt warming events of several degrees within a few decades followed by slower cooling were detected in Greenland ice cores (Dansgaard et al., 1984, 1993; NGRIP members, 24, Fig. 3.2). Effects of the so-called Dansgaard-Oeschger events were also found elsewhere in the Northern Hemisphere (Oeschger et al., 1984; Schulz et al., 1998; Peterson et al., 2; Wang et al., 21) and are believed to be the result of rapid changes of the AMOC strength (Stocker, 2). Other examples of a reduction or maybe even a complete shutdown of the AMOC are the Younger Dryas stadial from approximately 12.8 to 11.6 kyr before present (BP, before year 195 AD) (Alley, 2; Broecker, 26) and the 8.2 kyr event (Alley et al., 1997; Alley & Ágústsdóttir, 25). The reconstruction of the state and strength of the AMOC in the past is an important topic in paleoceanography. From stable carbon isotopic ratios (δ 13 C) and cadmium-to-calcium (Cd/Ca) ratios in shells of benthic foraminifera found in marine sediment cores, it has been proposed that the last glacial maximum (LGM) AMOC was shallower than it is today (Lynch- Stieglitz et al., 27, Fig. 3.3). This is corroborated by neodymium (Nd) isotope studies (Rutberg et al., 2; Piotrowski et al., 25). Other proxies that give information on the circulation state are benthic-planktonic age differences inferred from radiocarbon concentrations of surface and bottom dwelling foraminifera and corals (Robinson et al., 25; Thornalley et al., 211), and marine surface reservoir age differences inferred from surface ocean and atmospheric radiocarbon concentrations (Ritz et al., 28). Unfortunately, the proxies mentioned give no quantitative information on the strength of the AMOC. The protactinium-to-thorium (Pa/Th) activity ratio of ocean sediments is a recently discovered proxy that provides insight on the changes of the strength of the AMOC (McManus et al., 24; Gherardi et al., 29). However, the quality of the proxy is still under debate (Scholten et al., 28; Keigwin & Boyle, 28; Lippold et al., 29). The Atlantic-to-Pacific benthic δ 13 C gradient has also been used as a qualitative proxy of AMOC strength (Bard & Rickaby, 29). Ritz et al. (211) propose that ocean temperature at any location in the ocean can be approximated by a linear combination of global mean atmospheric temperature, AMOC, and Antarctic Bottom Water (AABW) strength. In a transient simulation over several glacialinterglacial cycles using a coupled ocean-atmosphere model, they test the assumption by fitting the linear combination using the time series of the simulated quantities to the simulated ocean temperature time series at 23 locations and depths spread over the world ocean. The regression coefficients determined by the fit give information on the dependency of local ocean temperature to changes in global mean atmospheric temperature, AMOC, and AABW strength. In this study, a quantitative method is developed to reconstruct the past evolution of AMOC and AABW strength on millennial and glacial-interglacial timescales based on the combination of the ocean temperature approximation presented by Ritz et al. (211), and ocean and atmospheric temperature time series reconstructions. AMOC strength reconstructions are presented for the period of the last deglaciation and in lower resolution for the past 33 kyr.

81 3.2. METHOD 79 Figure 3.1: The great ocean conveyor belt is a highly simplified image of the large scale ocean circulation (IPCC, 21). 3.2 Method Using the Bern3D reduced-complexity coupled ocean-atmosphere model, Ritz et al. (211) find that within the performed model simulations, ocean temperature at any location and depth x can be approximated by a linear combination of global mean atmospheric temperature (T model atm T fit ), AMOC strength (Ψmodel AMOC ) and AABW strength (Ψmodel AABW ): oc ( x,t) = a Tatm( x) Tatm model (t) + a AMOC( x) Ψ model AMOC (t) + a AABW( x) Ψ model AABW (t). (3.1) They quantify the contributions of changes of these three quantities on a change in ocean temperature at various locations and depths in the world ocean by determining the regression coefficients a Tatm, a AMOC, and a AABW by fitting the linear combination to the simulated ocean temperature using a least mean squares approach. The general idea of the AMOC reconstruction method is that Eq. (3.1) can be solved for Ψ model AMOC and Ψmodel AABW when ocean and atmospheric temperature time series from reconstructions (Toc rec,tatm) rec are provided: T rec oc ( x 1,t) = a Tatm ( x 1 ) T rec atm(t) + a AMOC ( x 1 ) Ψ AMOC (t) + a AABW ( x 1 ) Ψ AABW (t) T rec oc ( x 2,t) = a Tatm ( x 2 ) T rec atm(t) + a AMOC ( x 2 ) Ψ AMOC (t) + a AABW ( x 2 ) Ψ AABW (t). T rec oc ( x n,t) = a Tatm ( x n ) T rec atm(t) + a AMOC ( x n ) Ψ AMOC (t) + a AABW ( x n ) Ψ AABW (t), (3.2) where subscript n stands for the ocean temperature reconstruction at location x n. For the atmospheric temperature reconstructions, ice core proxy data from Antarctica (Jouzel et al., 27) and Greenland (NGRIP members, 24) are used. Because the atmosphere is much better mixed than the ocean, these data sets represent hemispheric to global temperature variations (depending on the time scale). The regression coefficients are determined by the Bern3D model (see below). Because there are two unknowns, at least two ocean temperature

82 8 3. AMOC RECONSTRUCTION δ 18 O ( ) YD Time (kyr BP) 4 2 Figure 3.2: The NGRIP stable oxygen isotope record is a proxy for Greenland air temperature (NGRIP members, 24). The abrupt Dansgaard-Oeschger events are numbered and the Younger Dryas (YD) stadial is indicated. time series reconstructions must be provided. If more than two ocean temperature reconstructions are given, the system of equations (3.2) is solved for Ψ AMOC (t) and Ψ AABW (t) by computing the least mean squares solution. As described above, the regression coefficients of the system of equations (3.2) are determined using the Bern3D reduced complexity coupled ocean-atmosphere model (Müller et al., 26; Ritz et al., 211). The ocean component of the model is solved using frictional geostrophic equations of motion. It consists of 36 by 36 grid cells horizontally and 32 depth layers. The atmosphere is described by an energy and moisture balance model with the same horizontal resolution as the ocean. The regression coefficients are determined by a model simulation, where during 18 kyr a broad range of climate states is produced by adding and removing freshwater from the North Atlantic and afterwards from the Southern Ocean to alter the AMOC and the AABW strength, and by adding and removing CO 2 from the atmosphere to change the atmospheric temperature (Fig. 3.4). At the location and depth of every ocean temperature reconstruction, the simulated ocean temperature evolution is fitted to the linear combination (Eq. 3.1) using the simulated global mean atmospheric temperature as well as the simulated AMOC and AABW strength, resulting in the regression coefficients a Tatm ( x i ), a AMOC ( x i ), and a AABW ( x i ), where i is the i th ocean temperature reconstruction. In order to get a more robust modeled ocean temperature, the ocean temperature is averaged over the model cell where the sediment core is located and the adjacent model cells closest to the sediment core location in zonal, meridional, vertical, and diagonal directions. The ocean and atmospheric temperature reconstructions must be interpolated to make the calculations of Eq. (3.2) possible. The data sets are low-pass filtered by applying a spline according to Enting (1987). For the reconstruction of the deglacial AMOC, a cut-off period of 1 kyr is used. The data sets used for the AMOC reconstruction on glacial-interglacial timescales are of lower resolution. For these reconstructions, a cut-off period of 1 kyr is applied.

83 3.2. METHOD 81 Figure 3.3: Modern and last glacial maximum (LGM) state of the AMOC as inferred from proxy data (Lynch- Stieglitz et al., 27). (a) The modern distribution of the dissolved biological nutrient phosphate (µmoll 1 ) in the western Atlantic (Conkright et al., 22). Also indicated is the southward flow of North Atlantic Deep Water (NADW) and Antarctic Bottom Water (AABW). (b) The LGM distribution of the carbon isotopic composition 13 C/ 12 C (expressed as δ 13 C) of shells of benthic foraminifera in the western and central Atlantic (Bickert & Mackensen, 24; Curry & Oppo, 25). The shallower NADW during the LGM is referred to as Glacial North Atlantic Intermediate Water (GNAIW). Data from different longitudes are collapsed in the same meridional plane. (c) Estimates of the cadmium (Cd) concentration (nmolkg 1 ) for the LGM from the ratio of Cd/Ca in the shells of benthic foraminifera (Marchitto & Broecker, 26). The modern δ 13 C composition and the dissolved Cd concentrations both show distributions similar to that of phosphate.

84 82 3. AMOC RECONSTRUCTION atm CO2 (ppm) Fw N Atl (Sv) AABW (Sv) a) b) c) d) e) f) Time (kyr) Fw SO (Sv) atm T (ºC) AMOC (Sv) Figure 3.4: Model simulation performed to determine the regression coefficients of Eq. (3.2). A broad range of climate states is created by a) adding and removing.3 Sv (1 Sv = 1 6 m 3 s 1 ) freshwater (Fw) from the North Atlantic from 5 N to 7 N, b) by adding and removing.2 Sv freshwater from the Southern Ocean (SO) from 6 S to 7 S, and c) by doubling and halving atmospheric CO 2 (maximum value: 556 ppm, minimum value: 153 ppm). The result is d) a complete shutdown of the AMOC followed by an AMOC state substantially stronger than in the standard case, e) an AABW strength (i.e. the maximum absolute SO overturning cell strength) substantially stronger and weaker, respectively, than in the standard case, and f) changes of atmospheric temperatures by several degrees. The AMOC and AABW strength can be reconstructed either qualitatively or quantitatively. In the qualitative reconstruction all quantities Tatm model, Ψ model AMOC, Ψmodel AABW, T oc rec, and Tatm rec are normalized according to X = (X X)/(max(X X), where X is the quantity and X the temporal mean value. Then, the regression coefficients a Tatm, a AMOC, and a AABW are dimensionless. In the quantitative reconstruction the absolute values of all quantities are only subtracted from their temporal mean value: X = X X. The dimensions of the regression coefficients are [a Tatm ] = K K 1, [a AMOC ] = K Sv 1, and [a AABW ] = K Sv 1. The quantitative approach has the benefit that past circulation changes can be quantified in Sverdrups (1 Sv = 1 6 m 3 s 1 ). However, uncertainties of the ocean and atmospheric temperature reconstructions with respect to the absolute values may be substantial and must be taken into account. Knowledge about absolute values is not required in the qualitative approach.

85 3.3. OCEAN TEMPERATURE RECONSTRUCTIONS Ocean Temperature Reconstructions The results of Ritz et al. (211) show that ocean temperatures at the surface ocean strongly correlate with atmospheric temperatures and depend only marginally on the state of the ocean circulation. In the intermediate and deep ocean, ocean temperatures are more sensitive to changes of the ocean circulation. Therefore, it makes sense to use deep ocean temperature reconstructions for the ocean circulation reconstruction. Unfortunately, only few deep ocean temperature reconstructions that span several glacial-interglacial cycles are published to this date. For the last deglaciation, the situation is even worse. Therefore, also sea surface temperature (SST) reconstructions are taken into account. However, possibly only abrupt circulation changes are revealed when using SSTs. We limit SST reconstructions to the Atlantic Ocean where changes of the AMOC are most likely to be monitored. Two reconstructions are distinguished. The first spans several glacial-interglacial cycles, the second spans the last deglaciation but with a higher resolution. This requires more highly resolved ocean temperature data sets. The deep ocean temperature and SST data sets used in the glacial-interglacial ocean circulation reconstruction are summarized in Table 3.1. The sediment core locations are depicted in Fig. 3.5a. Only data sets are used that span at least 25 kyr and have an average resolution of at least 3 kyr. The data sets used for the reconstruction of the circulation of the last deglaciation are summarized in Table 3.2 and Fig. 3.5b. Here, only data sets that span at least 19 kyr and with an average resolution of at least 5 yr are used. Since for this period no suitable deep ocean temperature records are available, only SST records are used kyr AMOC Reconstruction For the AMOC and AABW reconstruction of the last 33 kyr, first the regression coefficients at the locations of the sediment cores summarized in Table 3.1 are determined using the Bern3D model as described above for both qualitative and quantitative cases. The calculated regression coefficients and the correlations between the simulated ocean temperature and the least mean squares fit (R fit ) are given in Table 3.3 for the quantitative cases and in Table 3.4 for the qualitative cases. Note that the regression coefficients of data set #6 of Lawrence et al. (29) (Table 3.1) are not given in Tables 3.3 and 3.4, because the location of this data set corresponds to the deep water formation region of the model. Because the modeled deep water formation zone is too far south compared to the present day deep water formation regions of the Greenland-Iceland-Norwegian and Labrador Seas, the regression coefficients calculated by the model are very likely to be biased. Therefore, this data set is not used for the circulation reconstructions. As an example of how well the least mean squares ocean temperature fit compares to the simulated ocean temperatures, the simulated global mean atmospheric temperature, the simulated ocean temperature at the location of the sediment core of Martin et al. (22) and the least mean squares fit are shown in Fig. 3.6a. The correlation between simulated and fitted ocean temperatures (Tables 3.3 and 3.4 and Fig. 3.6a) demonstrates that this simple ocean temperature model is by no means perfect. However, at several locations, R fit is substantially higher than just the correlation between simulated ocean and atmospheric temperature (R Tatm ), indicating the importance of the ocean circulation at these locations. In a second step, all ocean temperature reconstructions as well as the Antarctic air temperature reconstruction of Jouzel et al. (27) are splined using a cut-off period of 1 kyr. This is illustrated in Fig. 3.7 for the ocean temperature reconstruction of Martin et al. (22) and the Antarctic air temperature as two examples. Because of the large cut-off period applied

86 84 3. AMOC RECONSTRUCTION Table 3.1: Deep and surface ocean temperature reconstructions from various sites around the world that are used for the glacial-interglacial overturning circulation strength reconstruction. The data sets span at least 25 kyr and have an average resolution better than 3 kyr. The time span is given in kyr before present (BP, before year 195 AD). Only surface ocean temperature records from the Atlantic Ocean are used (see also Fig. 3.5a). The proxies used for the temperature reconstruction are either the Mg to Ca or oxygen isotope (δ 18 O) ratios of planktonic or benthic foraminiferal species, or alkenones in coccolithophorids (U K 37). # Lat ( ) Lon ( ) Depth time span avg res Proxy Reference (m) (kyr BP) (kyr) Mg/Ca Martin et al. (22) Mg/Ca Elderfield et al. (21) δ 18 O Shackleton (2) δ 18 O Waelbroeck et al. (22) δ 18 O Waelbroeck et al. (22) U K 37 Lawrence et al. (29) U K 37 Martrat et al. (27) Mg/Ca Nürnberg et al. (2) Mg/Ca Nürnberg et al. (2) Mg/Ca Schmidt et al. (26) a) b) 4, ,8 9 13,19, , Figure 3.5: Locations of deep ocean (triangles) and surface ocean (circles) temperature time series reconstructions used for the overturning circulation strength reconstruction. (a) The data sets used for the glacialinterglacial reconstruction. The time span of the data sets is at least 25 kyr and the average resolution at least 3 kyr. Only surface ocean temperature records from the Atlantic Ocean are used. The index numbers correspond to the numbers in Table 3.1. (b) The data sets used for the last deglaciation reconstruction. Again, only Atlantic Ocean SST records are used. The index numbers correspond to the numbers in Table 3.2.

87 KYR AMOC RECONSTRUCTION 85 Table 3.2: SST reconstructions from various Atlantic sites that are used for the last deglaciation overturning circulation strength reconstruction. The data sets span at least 19 kyr and have an average resolution better than 5 yr. See also Fig. 3.5b. For datasets that span longer periods than the last 25 kyr, the average resolution was calculated for the last 25 kyr. # Lat ( ) Lon ( ) Depth time span avg res Proxy Reference (m) (kyr BP) (yr) Mg/Ca Waelbroeck et al. (21) Mg/Ca Waelbroeck et al. (21) Mg/Ca Waelbroeck et al. (21) Mg/Ca Weldeab et al. (27a) Mg/Ca Weldeab et al. (27b) Mg/Ca Weldeab et al. (26) Mg/Ca Farmer et al. (25) Mg/Ca Lea et al. (23) U K 37 Martrat et al. (27) U K 37 Pailler & Bard (22) U K 37 Kim et al. (22) U K 37 Zhao et al. (1995) Table 3.3: Regression coefficients at the deep ocean temperature and SST sediment core locations (Table 3.1) required the quantitative AMOC and AABW reconstructions of the last 33 kyr (Eq. 3.2). The regression coefficients are calculated by the model simulation described in Fig Also listed are the correlations R Tatm between ocean temperature at the particular location and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit. # a Tatm a AMOC a AABW R Tatm R fit (K K 1 ) (K Sv 1 ) (K Sv 1 )

88 86 3. AMOC RECONSTRUCTION 1 a) ~ ~ ~ fit T atm T oc T oc Normalized quantity.5.5 Normalized quantity b) c) #1 a Tatm =.31, a AMOC =.76, a AABW =.55 R Tatm =.23, R fit =.71 #7 a Tatm =.95, a AMOC =.9, a AABW =.13 R Tatm =.54, R fit =.92 #8 a Tatm =.93, a AMOC =.54, a AABW =.47 R Tatm =.77, R fit = Time (kyr BP) Normalized quantity Figure 3.6: Examples of the determination of the regression coefficients of Eq. (3.1) at the sediment core locations 1, 7, and 8 (see Table 3.4) by using the model simulation described in Fig Gray line: normalized global mean atmospheric temperature of the simulation, black line: normalized ocean temperature at the sediment core location, and red line: least mean squares fit of the ocean temperature according to Eq. (3.1). The normalization is done as follows: e X = (X X)/(max(X X), where X is either the atmospheric or the ocean temperature time series. The regression coefficients and the correlations R Tatm between ocean temperature and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit are given in the figure.

89 KYR AMOC RECONSTRUCTION 87 Table 3.4: Regression coefficients at the deep ocean temperature and SST sediment core locations (Table 3.1) required for the qualitative AMOC and AABW reconstructions of the last 33 kyr (Eq. 3.2). For the calculation of these regression coefficients, normalized quantities from the model simulation (Fig. 3.4) are used. Also listed are the correlations R Tatm between ocean temperature at the particular location and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit. # a Tatm a AMOC a AABW R Tatm R fit Atlantic deep ocean temperature ( C) a) b) Martin et al., 22 Spline (1 kyr cut-off period) Jouzel et al., 27 Spline (1 kyr cut-off period) Time (kyr BP) Antarctic air temperature ( C) Figure 3.7: Antarctic air temperature reconstruction and one example of an ocean temperature reconstruction (Martin et al., 22) in the original resolution (gray lines) and low-pass filtered with a cut-off period of 1 kyr (black lines). Because the mixing time scale within the atmosphere is considerably shorter than the chosen cut-off period, the spline of the atmospheric temperature reconstruction represents the global mean temperature evolution. In order to convert Antarctic air temperature to global mean atmospheric temperature, the Antarctic air temperature must be divided by a factor between 1.2 and 2 (Masson-Delmotte et al., 21).

90 88 3. AMOC RECONSTRUCTION compared to the mixing time scale of the atmosphere, the splined temperature record is assumed to represent qualitatively the global mean temperature evolution. This is not true for the amplitudes of the glacial-interglacial temperature variations, which are larger at high latitudes compared to the global mean. Masson-Delmotte et al. (26) estimate the glacialinterglacial temperature variations to be twice as large in Antarctica as in the global mean. Köhler et al. (21) and Masson-Delmotte et al. (21) argue that the polar amplification was not constant in time. Masson-Delmotte et al. (21) estimate a polar amplification factor of 1.2 for interglacial periods and a factor of 2 during glacial periods. Note that the temperature estimates are only required for the quantitative circulation reconstruction, whereas for the qualitative reconstruction, only the temperature patterns are used. Circulation reconstructions are separately done for deep ocean temperature reconstructions and SST reconstructions. As described in the methods section, the overdetermined system of equations (3.2) using all deep ocean temperature data sets of Table 3.1, respectively, is solved by singular value decomposition resulting in the least mean squares solution. For the quantitative case, the reconstructed AMOC and AABW strengths are shown in Figs. 3.8a-b. Note that a polar amplification factor of 2 was used to convert Antarctic air temperature to global mean air temperature. A second reconstruction is done using a polar amplification factor of 1.2. This hardly affects the AABW reconstruction, but the AMOC reconstruction shows larger amplitudes when using the lower amplification factor. If the ocean and atmospheric temperature reconstructions and the climate model were perfect, then any two ocean temperature reconstructions would be adequate to perfectly reconstruct the AMOC and AABW strengths. Because this is not the case, it makes sense to use ocean temperature reconstructions at locations that are very sensitive to changes in ocean circulation. At locations that are only weakly sensitive to circulation changes, a AMOC and a AABW are small. In these cases, according to Eq. (3.2), unrealistically large changes of the AMOC and AABW must be implied to correct for the imperfections of the atmospheric and ocean temperature reconstructions in order to reach equality between the ocean temperature reconstruction and the linear regression. The least mean squares solver of an overdetermined system automatically filters out such solutions, because they deviate strongly from the least mean squares solution. This is nicely shown in Figs. 3.8a-b, where a circulation reconstruction is done where only three of the five deep ocean temperature reconstructions are used, namely # 1, 4, and 5 (Table 3.3). The locations of these reconstructions are more sensitive to circulation changes than the two locations not used, because a AMOC and a AABW are higher and R fit is substantially higher than R Tatm. This AMOC reconstruction is very similar to the reconstruction where all data sets are used, because the other two records play only a minor role in the least mean squares solution. The AABW reconstructions are also similar, but here the two data sets (# 2 and 3) have more influence on the result. The AMOC reconstruction varies within a 45 Sv band. These variations are much larger than the AMOC changes of a simulation performed by Ritz et al. (211) (Fig. 3.8a). The reconstructed AABW variations are also larger compared to the simulation. These variations seem unrealistically high and are speculated to be the result of uncertainties in the temperature values of the ocean and atmospheric temperature reconstructions and of uncertainties in the values of the regression coefficients. The result is more robust when only normalized values are used and exact numbers become unimportant. The result of the qualitative AMOC and AABW reconstruction from deep ocean temperature reconstructions is displayed in Figs. 3.8c-d. The patterns of the qualitative and the quantitative reconstruction are similar. Again, choosing only ocean temperature data sets 1, 4, and 5 for the AMOC reconstruction results

91 KYR AMOC RECONSTRUCTION 89 model (Ritz et al., 211) standard reconstruction AMOC-only reconstruction only 3 ocean temp. data sets used polar amplification factor of 1.2 AMOC mean(amoc) (Sv) AMOC (normalized) a) b) c) d) Quantitative reconstruction based on deep ocean temperatures Qualitative reconstruction based on deep ocean temperatures AABW mean(aabw) (Sv) AABW (normalized) AMOC (normalized) e) f) Qualitative reconstruction based on sea surface temperatures AABW (normalized) Time (kyr BP) 1 5 Figure 3.8: Sensitivities of the AMOC and AABW reconstruction. (a) Quantitative reconstruction of the AMOC based on deep ocean temperature reconstructions (Table 3.1). Blue line: Standard reconstruction when all ocean temperature data sets are used and a polar amplification factor of 2 is used to convert Antarctic air temperature to global mean atmospheric temperature. Green line: only three out of five ocean temperature data sets are used. The result is very similar to the standard result, because the two ocean data sets not used play only a minor role in the least mean squares solution. Ocean temperatures at these locations are not sensitive to circulation changes. Red line: A polar amplification factor of 1.2 is used. Light blue line: AMOC-only reconstruction according to Eq. (3.3). Gray line: Model result of Ritz et al. (211). (b) Analogously for AABW. (c) Qualitative reconstruction of the AMOC based on normalized deep ocean temperature reconstructions (Table 3.2). (d) Analogously for AABW. (e) Qualitative AMOC reconstruction based on four normalized SST reconstructions (Table 3.2). (f) Analogously for AABW.

92 9 3. AMOC RECONSTRUCTION in a very similar reconstruction compared to when all 5 data sets are used. As expected, a AABW of the SST reconstructions are much smaller than the a AABW of the deep ocean temperature reconstructions, i.e. AABW has only very little influence on Atlantic SSTs. As an example, a AABW =.47 at location 2, which means that a 1 % change of AABW only leads to a 4.7 % SST change at this location. Because the surface Atlantic is not sensitive to changes of AABW, a realistic AABW reconstruction cannot be expected. Still, the result is shown in Fig. 3.8f for the qualitative reconstruction. The normalized AABW values vary from 1 to 1. A value of 1 suggests that AABW was 1 times stronger than the temporal mean value in order to account for the discrepancies between atmospheric and ocean temperatures. Because of these unrealistic results, it makes more sense to neglect AABW changes in the ocean temperature fit (Eq. 3.1) and to use the simplified linear combination T fit oc ( x,t) = a Tatm ( x) T model atm (t) + a AMOC ( x) Ψ model AMOC(t). (3.3) Henceforth, the resulting AMOC reconstruction is referred to as AMOC-only reconstruction. The circulation reconstruction described in Eq. (3.2) is referred to as full reconstruction. The regression coefficients of Eq. (3.3) are given in Table 3.5 for the qualitative circulation reconstruction case and in Table 3.6 for the quantitative reconstruction case. The AMOC-only reconstruction from the SST data shows major differences to the full reconstruction (Fig. 3.8e). Because AABW of the full reconstruction using deep ocean temperature data does not reach such unrealistic values as the SST-derived AABW, the deep ocean temperature derived AMOC-only reconstruction does not differ much from the full-reconstruction (Fig. 3.8a). The SST and deep ocean temperature derived AMOC reconstructions are compared for both qualitative and quantitative cases in Fig The reconstructions of the qualitative case are additionally compared to a reconstruction by Bard & Rickaby (29) who use the Atlantic-to- Pacific δ 13 C gradient of benthic foraminifera as a qualitative proxy of the AMOC strength. All three reconstructions as well as the model simulation of Ritz et al. (211) show a decline of the AMOC strength from 2 to 13 kyr BP followed by an increase until 1 kyr BP. From 1 kyr BP to 3 kyr BP, the ocean temperature derived reconstructions and the model show an AMOC decrease, while the δ 13 C derived reconstruction shows an increase. On shorter timescales, the reconstructions show large differences. One major factor of uncertainty is the fact that the atmospheric temperature reconstruction and all ocean temperature reconstructions are given on their own age scales which may differ substantially. The consequences of this uncertainty is illustrated in Fig. 3.1, where an AMOC-only reconstruction is shown using only the Martin et al. (22) ocean temperature data set. Also, the normalized atmospheric and ocean temperature time series are displayed. For instance, the AMOC shows a rapid increase at around 2 kyr BP. The reason for this increase is that the deglacial ocean temperature increase leads the atmospheric temperature rise which is probably due to the uncertainty of the age scales. Because a AMOC is positive at this particular location, an ocean temperature increase at this location without accompanying atmospheric temperature increase must lead to an increase in the AMOC to provide more warmer North Atlantic deep waters compared to the cold Antarctic Bottom Waters. However, if in reality the atmospheric and ocean temperature increase have occurred simultaneously, then the AMOC would have remained approximately constant. In the next section, an AMOC-only reconstruction is performed for the last deglaciation using more highly resolved SST data sets. The age scales of the temperature reconstructions are less uncertain during this period because radiocarbon dating is possible.

93 KYR AMOC RECONSTRUCTION 91 AMOC-only reconstruction AMOC mean(amoc) (Sv) a) based on deep ocean temp. records based on sea surface temp. records Bard & Rickaby, 29 model (Ritz et al., 211) AMOC (normalized) b) Time (kyr BP) 1 5 Figure 3.9: Comparison of the glacial-interglacial AMOC-only reconstructions. (a) Blue line: Quantitative reconstruction based on deep ocean temperature data sets. Red line: Quantitative reconstruction based on SST data sets. Gray line: Model simulation of Ritz et al. (211). (b) Blue line: Qualitative reconstruction based on deep ocean temperature data sets. Red line: Qualitative reconstruction based on SST data sets. Light blue line: Reconstruction of Bard & Rickaby (29) based on the Atlantic-to-Pacific δ 13 C gradient of benthic foraminifera. 1 ~ ~ ~ T oc (Martin et al., 22) T atm Ψ AMOC Normalized quantity Time (kyr BP) 1 5 Figure 3.1: Normalized atmospheric temperature (red line), ocean temperature at the sediment core location of Martin et al. (22), and the qualitative AMOC-only reconstruction (black line) when only using the Martin et al. (22) data set for the reconstruction. This figure illustrates the response of the AMOC with respect to the ocean and atmospheric temperature changes in order to fulfill Eq. (3.3).

94 92 3. AMOC RECONSTRUCTION Table 3.5: Regression coefficients at the deep ocean temperature and SST sediment core locations (Table 3.1) required for the qualitative AMOC-only reconstruction of the last 33 kyr (Eq. 3.3). Also listed are the correlations R Tatm between ocean temperature at the particular location and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit. # a Tatm a AMOC R Tatm R fit Table 3.6: Regression coefficients at the deep ocean temperature and SST sediment core locations (Table 3.1) required for the quantitative AMOC-only reconstruction of the last 33 kyr (Eq. 3.3). Also listed are the correlations R Tatm between ocean temperature at the particular location and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit. # a Tatm a AMOC R Tatm R fit (K K 1 ) (K Sv 1 ) AMOC Reconstruction of the Last Deglaciation For the ocean circulation reconstruction of the last deglaciation, ocean temperature data sets are used that span at least the last 19 kyr and that have an average resolution better than 5 yr (Table 3.2). The average resolution is chosen considerably higher than for the glacial-interglacial ocean circulation reconstruction in order to capture millennial-scale events during the deglaciation such as the Heinrich event 1 from about 16,8 to 14,6 years BP (Hemming, 24; Rasmussen et al., 26), the Bølling-Allerød period from 14,6 to 12,8 years BP (Rasmussen et al., 26, after Dansgaard-Oeschger event 1, Fig. 3.2) and the Younger Dryas from 12,8 to 11,6 years BP (Rasmussen et al., 26). Since to this date no deep ocean temperature reconstructions match these criteria, only SST data sets of the Atlantic Ocean are used. Because of the difficulties involved in the ocean circulation strength reconstruction using SST records as discussed in the previous section, the reconstruction is limited to the qualitative AMOC-only reconstruction. The ocean and atmospheric time series reconstructions are low-pass filtered using a spline with a cut-off period of 1 kyr. Due to this lower cut-off period compared the spline used in the glacial-interglacial ocean circulation reconstruction, the Antarctic air temperature record can no longer be used as a global atmospheric temperature signal. Therefore, the Antarctic air temperature record of Jouzel et al. (27) is only used for southern hemisphere ocean temperature data sets. For northern hemisphere records, the Greenland air temperature proxy record of NGRIP members (24) is used. For the regression coefficient calculation, two possibilities are distinguished. In the first, hemispheric mean air temperatures of the model are used, and in the second, the air temperatures at the ice core locations are used (Table 3.7). The ocean circulation reconstructions of both cases are very similar (Fig. 3.11). According to the reconstruction, the AMOC strength decreases from 19 to 16.5 kyr BP, then increases again until 12 kyr BP followed by a sharp drop. From 11 to 1 kyr BP, the AMOC variations are small. Apart from the large drop at the beginning of the deglaciation, the AMOC strength reconstruction does not follow the Pa/Th-based reconstruction of McManus et al. (24). Also, the reconstruction shows an AMOC increase to high values during the Younger Dryas.

95 3.5. AMOC RECONSTRUCTION OF THE LAST DEGLACIATION 93 Table 3.7: Regression coefficients at the SST sediment core locations (Table 3.2) required for the qualitative AMOC-only reconstruction (Eq. 3.3) of the last deglaciation. Two cases are distinguished. In one, hemispheric mean atmospheric temperatures of the model are used for the calculation of the regression coefficients (Hem). In the other case, the modeled air temperatures at the poles, i.e. at the ice core locations are used (Pol). Also listed are the correlations R Tatm between ocean temperature at the particular location and atmospheric temperature, and R fit between ocean temperature and the least mean squares fit. # a Tatm a AMOC R Tatm R AMOC Hem Pol Hem Pol Hem Pol Hem Pol AMOC (normalized) hemispheric regression coefficients polar regression coefficents H1 B-A YD Pa/Th 232 based Pa/Th 238 based Pa/ 23 Th Time (kyr BP) 5.1 Figure 3.11: AMOC-only reconstruction of the last deglaciation. Blue line: Hemispheric mean air temperatures are used for the calculation of the regression coefficients. Red line: Polar air temperatures are used for the regression coefficient calculation. Light blue lines: Qualitative AMOC strength reconstruction based on Pa/Th activity ratios of an Atlantic sediment core (McManus et al., 24).

96 94 3. AMOC RECONSTRUCTION This is in conflict to many studies that propose a reduction of the AMOC strength during the Younger Dryas due to freshwater discharges into the ocean (Broecker, 26, and references therein). In order to better understand the result of our AMOC reconstruction, i.e. of the least mean squares solution, an AMOC reconstruction is done separately for every ocean temperature data set where only this one ocean temperature record is used for the AMOC reconstruction (Fig. 3.12). The AMOC reconstructions from the SST reconstructions 11, 12, 13, 14, and 15 (see Table 3.2) show a result similar to the Pa/Th-based reconstruction with a weak AMOC during Heinrich event 1 and the Younger Dryas. The reconstruction from SST data set 3 suggests a period of strong AMOC during the Younger Dryas. AMOC reconstructions from the SST records 16 and 18 show very large normalized values because the ocean temperature is not sensitive to the AMOC at these sediment core locations (Table 3.7). Therefore, these two records do not have strong influence on the least mean squares solution of Fig The reconstructions from the SST data sets 17, 19, 2, 21, and 22 show a strong AMOC during the YD and partially during Heinrich event 1. The SST reconstructions 17, 19, and 2 show a pronounced temperature rise at the end of the YD. However, this temperature increase leads and lags, respectively, the atmospheric temperature increase by 6 to 1 years. These leads and lags might effectively be due to a change in the AMOC strength, but dating uncertainties may also play a significant role. Uncertainties of several hundred years arise from radiocarbon dating of ocean sediment cores because of uncertainties in the atmospheric calibration curve (Reimer et al., 29), possible variations of the marine reservoir age of several hundred years during ocean circulation changes (Ritz et al., 28), and especially because in sediment cores not every data point is dated. For example, SST record 17 of Farmer et al. (25) is dated at only 7 depths within the years 2, to 1, BP. The age scales are linearly interpolated in between. Also ice cores age scales have dating uncertainties of several hundred years during the last deglaciation. The uncertainties arise from uncertainties in layer counting, the ice age to gas age difference, and methane synchronization (Blunier et al., 27; Rasmussen et al., 26; Andersen et al., 26). 3.6 Conclusions A new method is presented to reconstruct past changes of AMOC and AABW strength using atmospheric and ocean temperature proxy records in combination with a climate model and results for the last 33 kyr and for the last deglaciation are calculated. This reconstruction method has the advantage over the Pa/Th- and the δ 13 C-based reconstruction methods that the AMOC strength can be determined quantitatively in Sverdrups. However, the AMOC and AABW reconstructions from this quantitative approach presented here show unrealistically large variations that are due to uncertainties in the values of the ocean and atmospheric temperature reconstructions and due to uncertainties of the modeled regression coefficients. Martin et al. (22) for example give a temperature uncertainty of ±1.4 C. This uncertainty is large compared to the total temperature range from 2 C to 3 C (Fig. 3.7). Unfortunately, the temperature uncertainty is not listed in most temperature reconstruction studies. The qualitative ocean circulation reconstruction approach leads to more realistic results. Several problems concerning this reconstruction method arise. The first major problem is the lack of deep ocean temperature reconstructions at locations sensitive to changes in ocean circulation and with suitable temporal resolution. Also, many sediment cores are located at ocean margins rather than in the open ocean and may therefore be influenced by factors other than AMOC or AABW such as small scale circulation changes or wind speed changes

97 3.6. CONCLUSIONS 95 normalized quantity normalized quantity normalized quantity #11 6 ~ ~ T oc T atm Ψ ~ AMOC #17 4 #12 #13 #14 #15 #16 H1 B-A YD 2 H1 B-A YD Time (kyr BP) Time (kyr BP) normalized quantity normalized quantity normalized quantity normalized quantity normalized quantity normalized quantity #18 #19 #2 #21 # normalized quantity normalized quantity normalized quantity Figure 3.12: AMOC-only reconstruction of the last deglaciation (black lines) when only using one ocean temperature data set for the reconstruction. A reconstruction is done for every SST record listed in Table 3.2. Blue lines: Normalized ocean temperature records. Red lines: Normalized air temperatures of either Greenland or Antarctica, depending on the hemispheric location of the ocean temperature sediment core.

98 96 3. AMOC RECONSTRUCTION that are not taken into account in our simple linear combination model (Eq. 3.1). Due to the few deep ocean temperature reconstructions available, SST reconstructions are also used for the ocean circulation reconstruction. However, because SSTs are more closely linked to atmospheric temperatures and therefore less sensitive to circulation changes than deep ocean temperatures, possibly only large AMOC changes can be monitored. Also, SSTs cannot be used to reconstruct AABW. Another major problem is the age-scale uncertainty. In order to obtain an independent AMOC reconstruction, the age-scales of the atmospheric and the ocean temperature records must be independent. On glacial-interglacial time scales, this is very difficult, as radiocarbon dating is not possible earlier than about 5 kyr BP (Reimer et al., 29). The age-scales of the ocean temperature data sets are correlated to stack records such as the benthic δ 18 O stack of Lisiecki & Raymo (25). Those stack records are in turn tuned to orbital parameters. Therefore, the age scales are no longer independent. Ice core age scales are derived by a combination of ice-sheet flow models, snow accumulation models and additional chronological age markers (Parrenin et al., 27). Both methods lead to large dating uncertainties. During the last deglaciation, radiocarbon dating of ocean sediment cores and annual layer counting within ice cores is possible. Still, dating uncertainties are of several hundred years and can therefore substantially influence the high resolution AMOC reconstruction. Due to the uncertainties of the age scales, we conclude that this ocean circulation strength reconstruction method is not suitable to detect abrupt ocean circulation changes. However, the method may be useful to quantitatively estimate the magnitude and long-term trends in the AMOC and AABW on glacial-interglacial time scales where accurate dating is not necessary. The reconstructions get more precise as the uncertainties in the atmospheric and ocean temperature proxy reconstructions are reduced and the number of reconstructions increase. Also, the circulation reconstruction can be performed by using more complex climate models such as atmosphere-ocean general circulation models in order to obtain more realistic regression coefficients. Both the deep ocean temperature and SST-based AMOC reconstructions of the past 33 kyr show a decline of the AMOC strength from 2 to 13 kyr BP followed by an increase until 1 kyr BP. From 1 kyr BP to 3 kyr BP, the ocean temperature derived reconstructions again show a decrease of the AMOC strength. Additional ocean temperature reconstructions, perhaps combined with less uncertainties in the temperature proxy reconstructions will permit a quantitative reconstruction of glacial-interglacial AMOC changes.

99 BIBLIOGRAPHY 97 Bibliography Alley, R. B., 2. The Younger Dryas cold interval as viewed from central Greenland, Quaternary Science Reviews, 19, Alley, R. B. & Ágústsdóttir, A. M., 25. The 8k event: cause and consequences of a major Holocene abrupt climate change, Quaternary Science Reviews, 24, Alley, R. B., Mayewski, P. A., Sowers, T., Stuiver, M., Taylor, K. C., & Clark, P. U., Holocene climatic instability: A prominent, widespread event 82 yr ago, Geology, 25, Andersen, K. K., Svensson, A., Johnsen, S. J., Rasmussen, S. O., Bigler, M., Röthlisberger, R., Ruth, U., Siggaard-Andersen, M.-L., Steffensen, J. P., Dahl-Jensen, D., Vinther, B. M., & Clausen, H. B., 26. The Greenland Ice Core Chronology 25, ka. Part 1: constructing the time scale, Quaternary Science Reviews, 25, Bard, E. & Rickaby, R. E. M., 29. Migration of the subtropical front as a modulator of glacial climate, Nature, 46, Bickert, T. & Mackensen, A., 24. Last Glacial to Holocene changes in South Atlantic deep water circulation, in The South Atlantic in the Late Quaternary: reconstruction of material budget and current systems, pp , eds Wefer, G., Mulitza, S., & Ratmeyer, V., Springer-Verlag, Berlin. Blunier, T., Spahni, R., Barnola, J.-M., Chappellaz, J., Loulergue, L., & Schwander, J., 27. Synchronization of ice core records via atmospheric gases, Climate of the Past, 3, Broecker, W. S., 26. Was the Younger Dryas triggered by a flood?, Science, 312, Conkright, M. E., Locarnini, R. A., Garcia, H. E., O Brien, T. D., Boyer, T. P., Stephens, C., & Antonov, J. I., 22. World Ocean Atlas 21, Objective Analyses, Data Statistics, and Figures, CD-ROM Documentation, National Oceanographic Data Center, Silver Spring, MD. Curry, W. B. & Oppo, D. W., 25. Glacial water mass geometry and the distribution of δ 13 C of ΣCO 2 in the western Atlantic Ocean, Paleoceanography, 2, PA117. Dansgaard, W., Johnsen, S. J., Clausen, H. B., Dahl-Jensen, D., Gundestrup, N., U., H. C., & Oeschger, H., Atlantic climate oscillations revealed by deep Greenland ice cores, in Climate Processes and Climate Sensitivity, Geophys. Mon. Ser., pp , eds Hansen, J. E. & Takahashi, T., American Geophysical Union, Washington, D. C. Dansgaard, W., Johnsen, S. J., Clausen, H. B., Dahl-Jensen, D., Gundstrup, N. S., Hammer, C. U., Hvidberg, C. S., Steffensen, J. P., Sveinbjörnsdottir, A. E., Jouzel, J., & Bond, G., Evidence for general instability of past climate from a 25-kyr ice-core record, Nature, 364, Elderfield, H., Greaves, M., Barker, S., Hall, I. R., Tripati, A., Ferretti, P., Crowhurst, S., Booth, L., & Daunt, C., 21. A record of bottom water temperature and seawater δ 18 O for the Southern Ocean over the past 44 kyr based on Mg/Ca of benthic foraminiferal Uvigerina spp., Quaternary Science Reviews, 29, Enting, I. G., On the use of smoothing splines to filter CO 2 data, Journal of Geophysical Research, 92, 1,977 1,984. Farmer, E. C., demenocal, P. B., & Marchitto, T. M., 25. Holocene and deglacial ocean temperature variability in the Benguela upwelling region: Implications for low-latitude atmospheric circulation, Paleoceanography, 2, PA218. Gherardi, J. M., Labeyrie, L., Nave, S., Francois, R., McManus, J. F., & Cortijo, E., 29. Glacial-interglacial circulation changes inferred from Pa-231/Th-23 sedimentary record in the North Atlantic region, Paleoceanography, 24, PA224. Hemming, S. R., 24. Heinrich events: Massive late Pleistocene detritus layers of the North Atlantic and their global climate imprint, Reviews of Geophysics, 42, RG15. IPCC, 21. Climate Change 21: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of of the Intergovernmental Panel on Climate Change, Intergovernmental Panel on Climate Change, Cambridge University Press.

100 98 BIBLIOGRAPHY Jouzel, J., Masson-Delmotte, V., Cattani, O., Dreyfus, G., Falourd, S., Hoffmann, G., Minster, B., Nouet, J., Barnola, J. M., Chappellaz, J., Fischer, H., Gallet, J. C., Johnsen, S., Leuenberger, M., Loulergue, L., Luethi, D., Oerter, H., Parrenin, F., Raisbeck, G., Raynaud, D., Schilt, A., Schwander, J., Selmo, E., Souchez, R., Spahni, R., Stauffer, B., Steffensen, J. P., Stenni, B., Stocker, T. F., Tison, J. L., Werner, M., & Wolff, E. W., 27. Orbital and millennial Antarctic climate variability over the past 8, years, Science, 317, Keigwin, L. D. & Boyle, E. A., 28. Did North Atlantic overturning halt 17, years ago?, Paleoceanography, 23, PA111. Kim, J.-H., Schneider, R. R., Müller, P. J., & Wefer, G., 22. Interhemispheric comparison of deglacial seasurface temperature patterns in atlantic eastern boundary currents, Earth and Planetary Science Letters, 194, Köhler, P., Bintanja, R., Fischer, H., Joos, F., Knutti, R., Lohmann, G., & Masson-Delmotte, V., 21. What caused Earth s temperature variations during the last 8, years? Data-based evidence on radiative forcing and constraints on climate sensitivity, Quaternary Science Reviews, 29, Lawrence, K. T., Herbert, T. D., Brown, C. M., Raymo, M. E., & Haywood, A. M., 29. High-amplitude variations in North Atlantic sea surface temperature during the early Pliocene warm period, Paleoceanography, 24, PA2218. Lea, D. W., Pak, D. K., Peterson, L. C., & Hughen, K. A., 23. Synchroneity of tropical and high-latitude Atlantic temperatures over the last glacial termination, Science, 31, Lippold, J., Grützner, J., Winter, D., Lahaye, Y., Mangini, A., & Christl, M., 29. Does sedimentary 231 Pa/ 23 Th from the Bermuda rise monitor past atlantic meridional overturning circulation?, Geophysical Research Letters, 36, L1261. Lisiecki, L. E. & Raymo, M. E., 25. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ 18 O records, Paleoceanography, 2, PA13. Lynch-Stieglitz, J., Adkins, J. F., Curry, W. B., Dokken, T., Hall, I. R., Herguera, J. C., Hirschi, J. J.-M., Ivanova, E. V., Kissel, C., Marchal, O., Marchitto, T. M., McCave, I. N., McManus, J. F., Mulitza, S., Ninnemann, U., Peeters, F., Yu, E.-F., & Zahn, R., 27. Atlantic meridional overturning circulation during the Last Glacial Maximum, Science, 316, Marchitto, T. M. & Broecker, W. S., 26. Deep water mass geometry in the glacial Atlantic Ocean: A review of constraints from the paleonutrient proxy Cd/Ca, Geochemistry, Geophysics, Geosystems, 7, Q123. Martin, P. A., Lea, D. W., Rosenthal, Y., Shackleton, N. J., Sarnthein, M., & Papenfuss, T., 22. Quaternary deep sea temperature histories derived from benthic foraminiferal Mg/Ca, Earth and Planetary Science Letters, 198, Martrat, B., Grimalt, J. O., Shackleton, N. J., de Abreu, L., Hutterli, M. A., & Stocker, T. F., 27. Four climate cycles of recurring deep and surface water destabilizations on the Iberian Margin, Science, 317, Masson-Delmotte, V., Stenni, B., Pol, K., Braconnot, P., Cattani, O., Falourd, S., Kageyama, M., Jouzel, J., Landais, A., Minster, B., Barnola, J. M., Chappellaz, J., Krinner, G., Johnsen, S., Röthlisberger, R., Hansen, J., Mikolajewicz, U., & Otto-Bliesner, B., 21. EPICA Dome C record of glacial and interglacial intensities, Quaternary Science Reviews, 29, Masson-Delmotte, V., Kageyama, M., Braconnot, P., Charbit, S., Krinner, G., Ritz, C., Guilyardi, E., Jouzel, J., Abe-Ouchi, A., Crucifix, M., Gladstone, R. M., Hewitt, C. D., Kitoh, A., LeGrande, A. N., Marti, O., Merkel, U., Motoi, T., Ohgaito, R., Otto-Bliesner, B., Peltier, W. R., Ross, I., Valdes, P. J., Vettoretti, G., Weber, S. L., Wolk, F., & Yu, Y., 26. Past and future polar amplification of climate change: climate model intercomparisons and ice-core constraints, Climate Dynamics, 26, McManus, J. F., Francois, R., Gherardi, J.-M., & Keigwin, L. D., 24. Collapse and rapid resumption of Atlantic meridional circulation linked to deglacial climate changes, Nature, 428, Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 26. Water mass distribution and ventilation time scales in a cost-efficient, three-dimensional ocean model, Journal of Climate, 19,

101 BIBLIOGRAPHY 99 NGRIP members, 24. High-resolution record of Northern Hemisphere climate extending into the last interglacial period, Nature, 431, Nürnberg, D., Müller, A., & Schneider, R. R., 2. Paleo-sea surface temperature calculations in the equatorial east Atlantic from Mg/Ca ratios in planktic foraminifera: A comparison to sea surface temperature estimates from U K 37, oxygen isotopes, and foraminiferal transfer function, Paleoceanography, 15, Oeschger, H., Beer, J., Siegenthaler, U., & Stauffer, B., Late glacial climate history from ice cores, in Climate Processes and Climate Sensitivity, Geophys. Mon. Ser., pp , eds Hansen, J. E. & Takahashi, T., American Geophysical Union, Washington, D. C. Pailler, D. & Bard, E., 22. High frequency palaeoceanographic changes during the past 14 yr recorded by the organic matter in sediments of the Iberian Margin, Palaeogeography Palaeoclimatology Palaeoecology, 181, Parrenin, F., Barnola, J.-M., Beer, J., Blunier, T., Castellano, E., Chappellaz, J., Dreyfus, G., Fischer, H., Fujita, S., Jouzel, J., Kawamura, K., Lemieux-Dudon, B., Loulergue, L., Masson-Delmotte, V., Narcisi, B., Petit, J.-R., Raisbeck, G., Raynaud, D., Ruth, U., Schwander, J., Severi, M., Spahni, R., Steffensen, J. P., Svensson, A., Udisti, R., Waelbroeck, C., & Wolff, E., 27. The EDC3 chronology for the EPICA Dome C ice core, Climate of the Past, 3, Peterson, L. C., Haug, G. H., Hughen, K. A., & Röhl, U., 2. Rapid changes in the hydrologic cycle of the tropical Atlantic during the last glacial, Science, 29, Piotrowski, A. M., Goldstein, S. L., Hemming, S. R., & Fairbanks, R. G., 25. Temporal Relationships of Carbon Cycling and Ocean Circulation at Glacial Boundaries, Science, 37, Rasmussen, S. O., Andersen, K. K., Svensson, A. M., Steffensen, J. P., Vinther, B. M., Clausen, H. B., Siggaard-Andersen, M. L., Johnsen, S. J., Larsen, L. B., Dahl-Jensen, D., Bigler, M., Röthlisberger, R., Fischer, H., Goto-Azuma, K., Hansson, M. E., & Ruth, U., 26. A new Greenland ice core chronology for the last glacial termination, Journal of Geophysical Research, 111, D612. Reimer, P. J., Baillie, M. G. L., Bard, E., Bayliss, A., Beck, J. W., Blackwell, P. G., Ramsey, C. B., Buck, C. E., Burr, G. S., Edwards, R. L., Friedrich, M., Grootes, P. M., Guilderson, T. P., Hajdas, I., Heaton, T. J., Hogg, A. G., Hughen, K. A., Kaiser, K. F., Kromer, B., McCormac, F. G., Manning, S. W., Reimer, R. W., Richards, D. A., Southon, J. R., Talamo, S., Turney, C. S. M., van der Plicht, J., & Weyhenmeyer, C. E., 29. INTCAL9 AND MARINE9 radiocarbon age calibration curves, -5, years Cal BP, Radiocarbon, 51, Ritz, S. P., Stocker, T. F., & Müller, S. A., 28. Modeling the effect of abrupt ocean circulation change on marine reservoir age, Earth and Planetary Science Letters, 268, Ritz, S. P., Stocker, T. F., & Joos, F., 211. A coupled dynamical ocean - energy balance atmosphere model for paleoclimate studies, Journal of Climate, 24, Robinson, L. F., Adkins, J. F., Keigwin, L. D., Southon, J., Fernandez, D. P., Wang, S. L., & Scheirer, D. S., 25. Radiocarbon variability in the western North Atlantic during the last deglaciation, Science, 31(5753), Rutberg, R. L., Hemming, S. R., & Goldstein, S. L., 2. Reduced North Atlantic Deep Water flux to the glacial Southern Ocean inferred from neodymium isotope ratios, Nature, 45, Schmidt, M. W., Vautravers, M. J., & Spero, H. J., 26. Western Caribbean sea surface temperatures during the late Quaternary, Geochemisty Geophysics Geosystems, 7, Q2P1. Scholten, J. C., Fietzke, J., Mangini, J., Garbe-Schönberg, C.-D., Eisenhauer, A., Schneider, R., & Stoffers, P., 28. Advection and scavenging: Effects on 23 Th and 231 Pa distribution off Southwest Africa, Earth and Planetary Science Letters, 271, Schulz, H., von Rad, U., & Erlenkeuser, H., Correlation between Arabian Sea and Greenland climate oscillations of the past 11, years, Nature, 393, Shackleton, N. J., 2. The 1,-year Ice-Age cycle identified and found to lag temperature, carbon dioxide, and orbital eccentricity, Science, 289,

102 1 BIBLIOGRAPHY Stocker, T. F., 2. Past and future reorganizations in the climate system, Quaternary Science Reviews, 19, Thornalley, D. J. R., Barker, S., Broecker, W. S., Elderfield, H., & McCave, I. N., 211. The deglacial evolution of North Atlantic deep convection, Science, 331, Waelbroeck, C., Labeyrie, L., Michel, E., Duplessy, J. C., McManus, J. F., Lambeck, K., Balbon, E., & Labracherie, M., 22. Sea-level and deep water temperature changes derived from benthic foraminifera isotopic records, Quaternary Science Reviews, 21, Waelbroeck, C., Duplessy, J. C., Michel, E., Labeyrie, L., Paillard, D., & Duprat, J., 21. The timing of the last deglaciation in North Atlantic climate records, Nature, 412, Wang, Y. J., Cheng, H., Edwards, R. L., An, Z. S., Wu, J. Y., Shen, C.-C., & Dorale, J. A., 21. A highresolution absolute-dated late pleistocene monsoon record from Hulu Cave, China, Science, 294, Weldeab, S., Schneider, R. R., & Kölling, M., 26. Deglacial sea surface temperature and salinity increase in the western tropical Atlantic in synchrony with high latitude climate instabilities, Earth and Planetary Science Letters, 241, Weldeab, S., Lea, D. W., Schneider, R. R., & Andersen, N., 27a. 155, years of West African Monsoon and Ocean Thermal Evolution, Science, 316, Weldeab, S., Schneider, R. R., & Müller, P., 27b. Comparison of Mg/Ca- and alkenone-based sea surface temperature estimates in the fresh water-influenced Gulf of Guinea, eastern equatorial Atlantic, Geochemistry Geophysics Geosystems, 8, Q5P22. Zhao, M., Beveridge, N. A. S., Shackleton, N. J., Sarnthein, M., & Eglinton, G., Molecular stratigraphy of cores off northwest Africa sea-surface temperature history over the last 8 ka, Paleoceanography, 1,

103 Chapter 4 Noble Gases as Proxies of Mean Ocean Temperature: Sensitivity Studies using a Climate Model of Reduced Complexity Stefan P. Ritz and Thomas F. Stocker Submitted to Quaternary Science Reviews, 211. Abstract Past global mean ocean temperature may be reconstructed from measurements of atmospheric noble gas concentrations in ice core bubbles. Assuming conservation of noble gases in the atmosphere-ocean system, the total concentration within the ocean only depends on solubility which itself is temperature dependent. Therefore, the colder the ocean, the more gas can be dissolved and the less remains in the atmosphere. Here, the characteristics of this novel paleoclimatic proxy are explored by implementing krypton, xenon, argon, and N 2 into a reduced-complexity climate model. The relationship between noble gas concentrations and global mean ocean temperature is investigated and their sensitivities to changes in ocean volume, sea surface salinity, sea-level pressure and geothermal heat flux are quantified. We conclude that atmospheric noble gas concentrations are suitable proxies of global mean ocean temperature. Changes in ocean volume need to be considered when reconstructing ocean temperatures from noble gases. Calibration curves are provided to translate ice-core measurements of krypton, xenon, and argon into a global mean ocean temperature change. Simulated noble gas to nitrogen ratios for the last glacial maximum are δkr atm =.77, δxe atm = 2.36, and δar atm =.2. The uncertainty of the krypton calibration curve due to uncertainties of the equilibrium climate sensitivity is estimated to be ±.35 C. An additional ±.3 C uncertainty must be added for the last deglaciation and up to ±.45 C for earlier transitions due to age-scale uncertainties in the sea-level reconstructions. Finally, the fingerprint of idealized Dansgaard-Oeschger events in the atmospheric krypton-to-nitrogen ratio is presented. A δkr atm change of up to.43 is simulated for a 2 kyr Dansgaard- Oeschger event, and a change of up to.65 is simulated for a 4 kyr event. The modeled equilibrium timescale of Kr and N 2 is approximately 75 years for the modern ocean.

104 12 4. NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE 4.1 Introduction A major focus of paleoclimate research is on the reconstruction of past air and sea temperatures. A large variety of proxies have been used to constrain these quantities on multiple time-scales. Some examples of air temperature proxies are fossil pollen records (Overpeck et al., 1985), oxygen isotopes in speleothems (McDermott, 24), or water isotopes in ice cores (Dansgaard et al., 1993; Johnsen et al., 21; Jouzel et al., 27). Lake and ocean temperatures can be determined by analyzing chironomids and diatoms (Battarbee, 2), alkenones (Brassell et al., 1986; Brassell, 1993; Müller et al., 1998), the composition of membrane lipids of marine Crenarchaeota (TEX 86 ) (Schouten et al., 22, 23), or the magnesium to calcium ratio of planktonic and benthic foraminferal species (Barker et al., 25; Elderfield et al., 26; Bryan & Marchitto, 28) in marine sediment cores. All the listed proxies have in common that they represent only local or regional climatic conditions and may be sensitive to complex and poorly understood biological processes. So far it has not been possible to obtain information of past changes in global-mean temperatures and hence of possible changes in the energy balance of the Earth. Headly & Severinghaus (27) describe a method to reconstruct past mean-ocean temperatures by measuring the krypton-to-nitrogen ratio (Kr/N 2 ) in air bubbles trapped in ice cores. The idea of the method is based on the assumption that noble gases and nitrogen exist in a closed atmosphere-ocean system and, due to their inertness, the total concentration within the ocean only depends on solubility which itself is temperature dependent. Although nitrogen is not completely inert, its source and sink processes can be neglected in the present considerations (Gruber, 24). The more gas is dissolved in the ocean, the less remains in the atmosphere. Past atmospheric gas concentrations are preserved in ice cores and can be measured. The xenon-to-nitrogen ratio (Xe/N 2 ) and the argon-to-nitrogen ratio (Ar/N 2 ) are also suitable for temperature reconstruction. However, routine Kr/N 2, Xe/N 2 and Ar/N 2 measurements from polar ice cores have not been published to this date. The purpose of this paper is to assess the potential of these novel proxies and to investigate their relationship to changes in ocean temperature by quantifying their sensitivities to changes in ocean volume, sea-surface salinity (SSS), sea-level pressure and geothermal heat flux. Therefore, we have implemented Kr, Xe, Ar and N 2 into the Bern3D model, an intermediate complexity coupled ocean-atmosphere climate model (Müller et al., 26; Ritz et al., 211). Extensive model simulations yield a calibration curve to translate ice-core measurements into a global mean ocean temperature change and estimates of the associated uncertainties. For reasons of simplicity, the paper focuses on krypton. However, xenon and argon behave analogously to krypton. Finally, we determine the equilibration time of Kr/N 2 and determine the fingerprint of Dansgaard-Oeschger events in the atmospheric krypton-to-nitrogen ratio as a function of event duration. 4.2 Model Formulation of Noble Gases and N 2 For this study we use the Bern3D coupled ocean-atmosphere climate model (Müller et al., 26; Ritz et al., 211). It consists of a frictional geostrophic ocean model with a horizontal resolution of grid boxes and 32 layers. The atmosphere is described by a twodimensional energy and moisture balance model with the same horizontal resolution as the ocean. Sea ice is dynamically calculated. The noble gases and nitrogen are implemented as follows: Because of the large abundance of the gases in the atmosphere, the surface ocean is assumed to approach saturation. The air-sea gas exchange in sea-ice free areas is calculated

105 4.2. MODEL FORMULATION OF NOBLE GASES AND N 2 13 as F as = k(c s C s), (4.1) where C s is the sea-surface concentration (mol m 3 ), C s the saturation concentration (in mol m 3 ) and k the gas transfer velocity for seawater (in m s 1 ). Gas exchange is suppressed in sea-ice covered areas. k is a function of the squared wind speed and of the Schmidt number Sc and is calculated as described by the OCMIP-2 protocol (Orr, 1999) but scaled down by 19 % for better agreement between modeled ocean inventories and observations, as suggested by Müller et al. (28). The wind speed fields are taken from (Orr, 1999) and Sc from Wanninkhof (1992). For xenon, Sc is calculated as described by Wanninkhof (1992) using coefficients provided by Jähne et al. (1987) (see appendix for details). The saturation concentration of krypton is calculated from the solubility in ml air at standard temperature and pressure (STP) per kg seawater taken from Weiss & Kyser (1978), and by using l air/mol Kr at STP (Dymond & Smith, 198), and the local ocean density from the model. The saturation concentrations of argon and nitrogen in µmol kg 1 seawater are taken from Hamme & Emerson (24), the saturation concentration of xenon in µmol kg 1 seawater is taken from Severinghaus (personal communication 21, see appendix for details). The saturation concentrations depend on sea surface temperature (SST), sea surface salinity (SSS) and sea-level pressure. Apart from the air-sea fluxes (4.1), noble gases and nitrogen are assumed to be conservative within the ocean, thus there are no source or sink processes. This is not entirely true for nitrogen, because denitrification and nitrogen fixation are source and sink processes for N 2, but they have a negligible effect on the total nitrogen inventory of approximately Tg N (calculated from values from Tables 4.1 and 4.2) since nitrogen fixation adds only 135±6 Tg N per year to the ocean whereas denitrification removes 245±7 Tg N per year from the ocean (Gruber, 24). The atmospheric inventory is calculated as described by Headly & Severinghaus (27). The atmosphere-ocean system is assumed to be a closed system. The terrestrial biosphere reservoir, including soil organic matter, is estimated to contain only.2 % of the total nitrogen mass (Schlesinger, 1997) and is therefore neglected. Thus, the total gas inventory of the atmosphere-ocean system I g tot = Ig atm + Ig ocn (4.2) is constant and does not change even for different climate states. The superscript g denotes the gas species Kr, Xe, Ar or N 2. I g tot can be calculated for present-day conditions because today s I g atm is known and Ig ocn is calculated by the model under present-day conditions (Ritz et al., 211). Global ocean volume is calculated from 5-minute gridded elevations/bathymetry data (ETOPO5; see The presentday I g atm is calculated by multiplying the known mole fraction of the gas in the present-day atmosphere by the total amount of air in moles. The total moles of air are determined by dividing the total mass of the atmosphere by its molar weight. The values used for the calculation are given in Table 4.2. Modeled mean ocean concentrations compare well with observations (Sarmiento & Gruber, 26) as shown in Table 4.1. The relative differences between modeled concentrations and observations are below 5 % for Kr, N 2 and Ar and 2 % for Xe. However, because of the low precision of the observations, these values need to be taken with caution. For climate states different from present-day, I g atm is calculated as a residual, i.e. I g atm = Ig tot Ig ocn. Changes in sea-level pressure are taken into account by scaling the saturation concentration Cs. According to Henry s law, the partial pressure and the dissolved concentration of a gas are directly proportional. Because the ocean component of the Bern3D model has a rigid-lid, changes in ocean volume are taken into account offline by scaling I g ocn.

106 14 4. NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE To test the assumption of noble gas abundance within the atmosphere, we have also implemented into the model the possibility of directly simulating atmospheric concentrations. Air-sea gas exchange is then calculated as F as = k(βc a C s ), (4.3) where C a the atmospheric concentration (mol m 3 ) and β the Bunsen solubility coefficient taken from Weiss & Kyser (1978) for krypton, and Weiss (197) for nitrogen, respectively. β is defined as the volume of gas at standard temperature and pressure absorbed per unit volume of liquid. In a steady state modern climate, this method leads to a 1.5 % lower ocean krypton inventory and to a 2.5 % lower nitrogen inventory. Simulating the entire atmosphere-ocean system has the disadvantage that changes in ocean volume and sea-level pressure cannot be taken into account because of the rigid-lid formulation of the ocean model. Therefore, and because the differences between both methods are small, we use air-sea gas exchange as calculated in Eq. (4.1) for all simulations. Because ice-core measurements provide the ratio of the atmospheric krypton to nitrogen and the variations of this quantity are small, the delta notation is used in the following to describe the Kr/N 2 deviations from the present-day atmosphere standard (Headly & Severinghaus, 27): ( ) (Kr/N2 ) sample δkr atm = 1 1. (4.4) (Kr/N 2 ) standard (Kr/N 2 ) standard = is the global and annual mean atmospheric value under modern conditions (Table 4.2). 4.3 Sensitivity of δkr atm to Various Model Parameters The Modern Ocean To determine the dependence of δkr atm on temperature and other factors, the sensitivity of δkr atm to changes in atmospheric CO 2, ocean diapycnal diffusivity K d, and wind stress is tested using our model. Changes in atmospheric CO 2 affect atmospheric temperatures and therefore ocean temperatures. CO 2 is varied within the range of 4 % to 2 % of the pre-industrial value of 278 ppm. Every simulation is run for 1, years into steady state. Changes in K d and wind stress affect mixing within the ocean which also results in an ocean temperature change. We vary K d in a range from 1 8 to 1 4 m 2 s 1 (standard value: 1 5 m 2 s 1 ). In the wind stress runs, wind stress is scaled by 18 % to 2 %. Apart from simulations with wind stress below 6 % of the modern forcing, the CO 2, diapycnal diffusivity, and wind stress sensitivities show a clear relationship between global mean ocean temperature and δkr atm (Figs. 4.1 and 4.2). The colder the ocean, the smaller the sensitivity of δkr atm to ocean temperature because more extended sea-ice cover reduces air-sea gas exchange leading to surface waters which no longer saturate in Kr and N 2. Wind stress sensitivity simulations with wind stress forcing below 6 % of the modern forcing show a different behavior than the rest of the simulations. In these cases the reduced wind stress substantially weakens the mixing within the ocean, especially in the Southern Ocean. For example, in the 5 % wind stress case, the Drake Passage throughflow weakens from 43 Sv (1 Sv = 1 6 m 3 s 1 ) in the standard case to 29 Sv, and the Southern Ocean overturning cell weakens from 18 Sv to 16 Sv. These circulation changes lead to colder SSTs at high latitudes in combination with an expansion of the sea-ice cover. Krypton concentrations in Southern Ocean surface waters near Antarctica drop down to 91 % of the saturation concentration.

107 4.3. SENSITIVITY OF δkr ATM TO VARIOUS MODEL PARAMETERS 15 δkr atm ( ) δxe atm ( ) δar atm ( ) a) atm. pressure effect salinity / volume effect temperature effect b) c) LGM rad. forcing only LGM rad. forcing +salinity +volume LGM rad. forcing +salinity LGM rad. forcing +salinity +volume +pressure atm. CO 2 sensitivity diapycnal diffusivity sensitivity wind stress sensitivity transient deglaciation T oc ( C) T oc ( C) T oc ( C) Figure 4.1: Atmospheric noble gas ratio as a function of global mean ocean temperature anomaly. a) Relationship between δkr atm and mean ocean temperature for various simulations. Black symbols: sensitivities of atmospheric CO 2 concentration, ocean diapycnal diffusivity, and wind stress on mean ocean temperature and δkr atm. CO 2 is varied within the range of 4 % to 2 % of the pre-industrial value of 278 ppm, diapycnal diffusivity from 1 8 to 1 4 m 2 s 1 (standard value: 1 5 m 2 s 1, and wind stress is scaled by 6 % to 2 %. The simulations show a clear relationship between δkr atm and mean ocean temperature. Red symbols: δkr atm and mean ocean temperature values for various LGM simulations relative to the modern steady state. The factors which control δkr atm (temperature, salinity, ocean volume and sea-level pressure) are separated in the simulations to quantify each contribution (arrows). Dark and light blue lines: transient deglaciation simulation using two scenarios of the Barbados sea level record (Peltier & Fairbanks, 26) (see Fig. 4.3a). Green line: transient simulation using the benthic δ 18 O stack ocean volume proxy of Lisiecki & Raymo (25). Except for the transient simulations, all simulations are run into steady state. b) Analogous for δxe atm. c) Analogous for δar atm.

108 16 4. NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE Table 4.1: Mean ocean concentrations of the simulated gases compared to observations (Gruber, 28; Sarmiento & Gruber, 26; Quinby-Hunt & Turekian, 1983). In brackets are the values used for the relative difference. Note that the relative differences are not very accurate because of the low precision of the observations. To calculate ocean inventories, a modern ocean volume of m 3 determined from the ETOPO5 bathymetry data is used. Gas Observation Model Relative (µmolm 3 ) (µmol m 3 ) difference (%) N 2 575, 578, 1 Ar 16, 15,7 2 Kr (3.8) Xe.5.59 (.6) 2 Table 4.2: Atmospheric composition and molecular weights. The total mass of dry air is g (Schlesinger, 1997; Sarmiento & Gruber, 26). Gas Molar weight Atmospheric relative (g mol 1 ) composition in dry air Dry air N Ar Kr Xe atm. CO 2 sensitivity wind stress sensitivity 15 % 2 % δkr atm ( ) % 18 % 2 % % % 25 % 3 % 4 % 8 % 7 % 6 % 9 % 1 % 12 % T oc ( C) Figure 4.2: Relationship between δkr atm and global mean ocean temperature for simulations where the wind stress was scaled by 18 % to 2 %. For comparison, the atmospheric CO 2 sensitivity simulations are also shown (Fig. 4.1a). In the simulations with wind stress below 6 % of the modern value, δkr atm decouples from ocean temperature because of large Southern Ocean sea-ice extent combined with low wind-driven ocean mixing.

109 4.3. SENSITIVITY OF δkr ATM TO VARIOUS MODEL PARAMETERS 17 In the 2 % wind stress case, krypton concentrations drop down to 86 % of the saturation concentration. We find that sea ice decouples the noble gas concentration from the temperature signal. However, these extreme situations only occur when ocean mixing is very low and the Southern Ocean is sea-ice covered to approximately 65 S throughout the year. Due to the low mixing the waters below sea ice are less quickly exchanged by saturated waters from ice-free regions. It must also be mentioned that the wind-driven gyre transport is too weak in the model (Müller et al., 26). The modeled Drake Passage throughflow for example is with 43 Sv much weaker than the data-based estimate of 14 Sv by Ganachaud & Wunsch (2). Therefore, Southern Westerlies and consequently the Antarctic Circumpolar Current must face an even larger change than the model suggests to decouple δkr atm from ocean temperature. The relationship between δxe atm and temperature (Fig. 4.1b) and δar atm and temperature (Fig. 4.1c) are analogous to the δkr atm results. Our simulations yield a linear relationship between the noble gases for constant sea-level conditions: δxe atm = 2.83 δkr atm (R 2 =.998) and δar atm =.29 δkr atm (R 2 =.9988). Because of this very tight relationship between the different noble gases, we limit our simulations to krypton Last Glacial Maximum To quantify the contribution of changes in sea surface salinity (SSS), ocean volume and sea-level pressure to δkr atm during glacial-interglacial cycles, several last glacial maximum (LGM) simulations are performed (Fig. 4.1). In these simulations we apply a set of LGM conditions and run the model for 1, years to steady state. In the first (referred to as LGM1), only the radiative forcing is adjusted to LGM conditions, i.e. atmospheric CO 2 is set to 18 ppm, atmospheric CH 4 to 35 ppb, and land-surface albedo is adjusted to take into account the presence of ice-sheets as described in Ritz et al. (211). The difference between modern and LGM δkr atm in LGM1 is the contribution of the global mean ocean temperature change. In the second simulation (LGM2), additionally the change in ocean salinity is taken into account. Mean ocean salinity increases by 1.2 psu to 36. psu, SSS also increases by 1.2 psu to 35.1 psu due to the ice-sheet buildup which lowers sea-level. The difference between LGM2 and LGM1 determines the contribution of the SSS change on δkr atm. The change in SSS also invokes small circulation changes: The AMOC increases by 1 Sv, Southern Ocean overturning circulation decreases by about 2 Sv and a 1 km deep North Pacific overturning cell is established. These changes cool the global ocean by.4 C and alter the temperature depth profile to slightly warmer temperatures in the top 1 m and to cooler temperatures below 15 m depth. Thus, temperature contributes indirectly to the δkr atm change. A third simulation (LGM3) additionally takes into account the smaller ocean volume. Again, ocean volume is calculated from the ETOPO5 bathymetry data. Changes in the bathymetry due to the pressure of the ice sheets on the Earth s surface are not taken into account. The ocean volume contribution to the δkr atm change follows from the δkr atm difference between LGM3 and LGM2. The presence of large ice-sheets and the lower sea-level during the last ice age also lead to an increased sea-level pressure (Mélières et al., 1991). This effect is additionally taken into account in simulation LGM4. Using the model of Mélières et al. (1991), we calculate an LGM-to-modern sea-level pressure difference of 13.8 hpa to 14.4 hpa. We use the average (14.1 hpa). Hence, the difference between LGM4 and LGM3 determines the sea-level pressure contribution. Because ocean volume, salinity and sea-level pressure are coupled, the contributions are combined and henceforth referred to as sea-level contribution.

110 18 4. NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE The temperature contribution is clearly the most important (Fig. 4.1a). Still, it is noteworthy that the sea-level contribution is substantial even though ocean volume itself only changes by about 3 % and sea-level pressure by about 1 %. The reason is that δkr atm is proportional to the I Kr atm /IN 2 atm fraction. Although the relative change of Ig ocn is the same for all gases g, the relative change of I g atm depends on the solubility of the particular gas. Nitrogen, for instance, is much less soluble than krypton. Thus, I N 2 atm /IN 2 ocn is larger than IKr atm /IKr ocn and hence the relative change of I N 2 atm is smaller than the change of IKr atm. Therefore, in order to generate a calibration curve for the noble gas proxy, the history of sea-level change must be known and taken into account. This is of particular importance as the temperature effect and the sea-level effect point in opposite directions Last Deglaciation In three transient deglaciation simulations from the LGM to the present-day (Fig. 4.1) all the contributions discussed above are combined. In the first, the model is forced as described by Ritz et al. (211) by atmospheric CO 2, CH 4, insolation and a simple ice-sheet/ocean volume parametrization which is tied to the Lisiecki & Raymo (25) benthic δ 18 O stack (LR4, Fig. 4.3a). High-frequency variability of LR4 was removed by applying a spline fit following Enting (1987) with a cutoff period of 1 kyr. In the other two transient deglaciation simulations, the more accurate Barbados uplift-corrected eustatic sea level record of Peltier & Fairbanks (26) is used. A spline with cutoff period of 1.5 kyr is applied to the sea-level data. Two different sea level scenarios are considered due to the uncertainty of the measurements at around 13 kyr B.P. (Fig. 4.3a). The differences between the three calibration curves underlines the importance of an accurate sea-level proxy for the δkr atm calibration curve. The two simulations forced by the Barbados sea-level record differ mainly by a rapid sea-level rise which takes place at about 14 kyr B.P. in scenario A (Fig. 4.3a), and at about 11.5 kyr B.P. in scenario B, respectively. The resulting calibration curves are shifted by as much as.3 C. For earlier deglaciations, where sea-level records such as the Barbados record are not available, less accurate sea-level data such as the LR4 stack or the Red Sea record (Siddall et al., 23) must be used. 4.4 Calibration Uncertainties Climate Sensitivity In the LGM simulations presented here, global mean ocean temperature is 2.7 C cooler than the modern ocean temperature. This value depends on the climate sensitivity of the model, i.e. the atmospheric temperature response to a given change in atmospheric CO 2. The model is tuned to produce an equilibrium climate sensitivity (ECS), a global temperature rise for a doubling of atmospheric CO 2 from 278 ppm to 556 ppm, of 3 C. The exact ECS, however, is uncertain. Hegerl et al. (27) give a 66 % probability for ECS to be between 2 C and 4.5 C with a best estimate of 3 C. Thus, an uncertainty must be added to the temperature contribution of the glacial-interglacial δkr atm change. To estimate this uncertainty, additional LGM simulations are performed, where the ECS is set to 2 C and 4 C, respectively (Fig. 4.4). Simulations with 4.5 C ECS are not possible with the current model setup, because certain model parametrizations are no longer valid under the cold glacial conditions. The relationship between δkr atm and mean ocean temperature is substantially different in the 2 C ECS case compared to the 3 C ECS case, because the temperature contribution to

111 4.4. CALIBRATION UNCERTAINTIES 19 a) relative sea level (m) b) δkr atm ( ) c) δkr atm ( ) atm CO Time (kyr B.P.) atm CO 2 sensitivity transient deglaciation scenario A transient deglaciation scenario B sea level A shifted by 1kyr sea level A shifted by +1kyr sea level B shifted by 1kyr sea level B shifted by +1kyr atm CO 2 (ppm) 18 Barbados sea level scenario A Barbados sea level scenario B sea level A shifted by +1/+5 kyr sea level A shifted by 1/ 5 kyr LR5 1kyr spline.6 without geothermal heat flux.7 doubled geothermal heat flux T oc ( C) sea level A shifted by 5kyr sea level A shifted by +5kyr sea level B shifted by 5kyr sea level B shifted by +5kyr T oc ( C) Figure 4.3: Various deglaciation forcing functions and transient simulations to determine δkr atm calibration uncertainties due to age-scale uncertainties and geothermal heat flux uncertainties. a) Deglaciation forcing functions. Because ocean volume and sea-level pressure are important factors on glacial-interglacial timescales, a sea-level proxy is required for the calibration. To estimate the uncertainty due to the differences in the age-scales of the sea-level proxy and the ice-core record, the sea level/ocean volume forcing (Peltier & Fairbanks, 26) (thick dark and light blue lines and circles) of the transient deglaciation simulation is shifted relative to the CO 2 forcing (Monnin et al., 21) (black line, virtually the same age-scale as the δkr atm measurements) by ±1 kyr and ±5 kyr (thin blue solid and dashed lines). We distinguish two sea level scenarios due to the uncertainty of the measurements at around 13 kyr B.P. b) Relationship between δkr atm and global mean ocean temperature for both sea level scenarios without a shift (thick dark and light blue lines, as in Fig. 4.1) and with a shift of ±1 kyr (thin lines). The dark gray area denotes the calibration uncertainty due to the uncertainty of the sea level record (up to ±.15 C). The light gray area indicates the uncertainty due to age scale uncertainty during the last deglaciation (up to ±.3 C). Simulations without geothermal heat flux relative to modern geothermal heat flux (black square) and with doubled geothermal heat flux (black circle) shift the calibration curve by less than.15 C. c) A sea level forcing shift of ±5 kyr increases the uncertainty to ±.45 C.

112 11 4. NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE δkr atm ( ) atm CO 2 sensitivity LGM 2 C ECS LGM 3 C ECS LGM 4 C ECS T oc ( C) Figure 4.4: Analysis of the calibration uncertainties due to equilibrium climate sensitivity (ECS) uncertainty. Hegerl et al. (27) give a 66 % probability for the ECS to lie between 2 C and 4.5 C. Triangles: LGM simulations including the sea-level contribution to the δkr atm change for 2 C, 3 C, and 4 C ECS, respectively (simulations with 4.5 C ECS are not possible with the current model setup because certain model parametrizations are no longer valid under the cold glacial conditions). The lines denote a hypothetical linear transition from LGM to modern climate with respect to temperature and sea-level change. The calibration uncertainty is derived from the possible range of mean ocean temperature for a particular δkr atm value. It is estimated to be ±.35 C (dashed lines). the glacial-interglacial δkr atm change is smaller, whereas the sea-level contribution remains the same. Thus the sea-level contribution becomes more important. For the δkr atm to mean ocean temperature relationship a linear transition from LGM to modern climate with respect to temperature and sea-level is hypothesized. The relationships between between δkr atm and mean ocean temperature of the 4 C ECS and the 3 C ECS cases differ less. As the temperature contribution increases, the importance of the sea-level contribution is reduced. A calibration uncertainty due to the ECS uncertainty of ±.35 C is derived from the possible range of mean ocean temperature for a particular δkr atm value (Fig. 4.4). Because for high ECS simulations the sea-level contribution to the δkr atm change becomes small relative to the temperature contribution, we argue that a 4.5 C ECS simulation would not further increase the calibration uncertainty Age-Scale Uncertainties As the eustatic sea levels of previous glacial and interglacial periods relative to modern are well known (approximately 12 m for the LGM, +4-6 m for the last interglacial period (Jansen et al., 27, and references therein) but with recent studies pointing to larger values (Kopp et al., 29)), global mean ocean temperature can be reconstructed accurately. Agescale uncertainties become important during glaciations and deglaciations where sea-level and ocean temperatures were subject to major changes. Here, age-scales uncertainties are quantified by shifting the sea-level proxy relative to the CO 2 forcing time-series by ±1 kyr and by ±5 kyr in the transient simulations. Because of methane-based age-scale synchronizations between ice-cores (Blunier et al., 1998), the CO 2 record is virtually on the same age-scale as ice core δkr atm measurements (the uncertainty is approximately 1 years for the last deglaciation (Blunier et al., 27)).

113 4.5. SENSITIVITY OF δkr ATM TO OCEAN MIXING 111 During the last deglaciation, where radiocarbon dating and annual layer counting is still possible, age-scale uncertainties do not exceed 1 yrs (Reimer et al., 29; Blunier et al., 27; Rasmussen et al., 26; Andersen et al., 26). The uncertainties of the ice core age scales arise from uncertainties in layer counting, the ice age to gas age difference, and methane synchronization. Shifting the ocean-volume forcing by ±1 kyr results in calibration curves which differ by up to.6 C (thin blue lines and light gray area in Fig. 4.3b). The calibration curve uncertainty is therefore estimated to be ±.3 C. Age-scale uncertainties are substantially larger during earlier times. Shifting the ocean-volume forcing by ±5 kyr increases the uncertainty to ±.45 C (light gray area in Fig. 4.3c). The uncertainty of the Barbados sea level record alone leads to a calibration curve uncertainty of ±.15 C (dark gray area in Fig. 4.3b) Geothermal Heat Flux Geothermal heat flux contributes to the global ocean temperature. Because δkr atm only records heat input at the surface ocean but the geothermal heat source is at the seafloor, the noble gas signal decouples from the temperature signal at the heat source and hence poses an additional uncertainty factor. Note that within the ocean the gases are strongly undersaturated because of the high ambient pressure. Thus, a deep-ocean temperature increase, e.g. due to geothermal heat would not lead to outgassing. In order to estimate the uncertainty from changes in geothermal heat flux, a simulation is performed where modern geothermal heat flux is taken into account by applying the forcing field of Pollack et al. (1993) re-gridded to the Bern3D model grid. In this case the ocean mean geothermal heat flux is 96 mwm 2. In a second simulation, a doubled geothermal heat flux forcing is used. In all other simulations described in this paper, geothermal heat flux is neglected. The global mean ocean temperature difference between zero and modern geothermal heat flux is.17 C and δkr atm =.2 (Fig. 4.3b). The difference between doubled and modern heat flux is.21 C for mean ocean temperature and δkr atm =.6. The decoupling of the noble gas signal from the temperature signal at the heat source causes a shift of the calibration curve of less than.15 C in both cases (Fig. 4.3b). Since a local temperature change at the heat source propagates to the ocean surface, the noble gases pick up a fraction of the signal. Because we consider the doubling and the shutdown of geothermal heat flux as extreme scenarios, we conclude that variations in geothermal heat flux add an uncertainty of less than ±.15 C. 4.5 Sensitivity of δkr atm to Ocean Mixing in Equilibrium and Transient Simulations Equilibrium Simulations To determine the dependence of δkr atm on the magnitude of ocean mixing, a large number of model runs are performed where both the wind-stress parameter is varied from 4 % to 14 % of the present-day values and atmospheric CO 2 from 4 % to 28 % of the pre-industrial value (Figs. 4.5a-c). The standard diapycnal diffusivity of 1 5 m 2 s 1 is used. While wind stress changes affect mixing within the ocean as well as global ocean temperature, changes in CO 2 mainly affect ocean temperature. The result is a suite of climate states with different ocean temperatures and ocean mixing timescales, in the following expressed as AMOC strength (Figs. 4.5d-f). All simulations are run for 3, years into equilibrium.

114 NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE windstress relative to modern (%) a) AMOC (Sv) δkr atm ( ) d) atm CO 2 relative to pre-industrial (%) AMOC (Sv) global mean ocean T ( C) windstress relative to modern (%) b) atm CO 2 relative to pre-industrial (%) 4.5 global ocean T ( C) AMOC (Sv) δkr atm ( ) global mean SST ( C) e) windstress relative to modern (%) c) atm CO 2 relative to pre-industrial (%) atm T ( C) AMOC (Sv) δkr atm ( ) global mean atm T ( C) f) Figure 4.5: Sensitivity of δkr atm to ocean mixing. Every point represents a simulation of an ensemble which is run into equilibrium with a wind-stress parameter between 4 % and 14 % of the modern value and an atmospheric CO 2 concentration between 4 % and 28 % of the pre-industrial value. The reference pre-industrial steady state is represented by a larger bullet. a) AMOC strength, b) global ocean temperature, and c) atmospheric temperature of every ensemble member. d-f) δkr atm as a function of the mixing timescale (expressed as AMOC strength) and mean ocean temperature, mean SST, and mean atmospheric temperature, respectively.

115 4.5. SENSITIVITY OF δkr ATM TO OCEAN MIXING 113 The equilibrium timescale of both krypton and nitrogen is approximately 75 years for the modern ocean as determined by a simulation where the model is run into equilibrium after doubling CO 2 in the atmosphere. The equilibrium timescale depends on the strength of the circulation. Figure 4.5d demonstrates that δkr atm depends on mean ocean temperature but hardly on the AMOC strength as long as the simulations are run into equilibrium. δkr atm also correlates with global mean SST (Fig. 4.5e) and with global mean atmospheric temperature (Fig. 4.5f). However, using δkr atm as a proxy for SST and atmospheric temperature would additionally require information on the strength of ocean mixing. The simulations redone with a higher diapycnal diffusivity of m 2 s 1 result in the same conclusions Dansgaard-Oeschger Simulations A suite of idealized Dansgaard-Oeschger event simulations are performed in order to provide an estimate of the δkr atm signal which is produced by abrupt climate change and associated ocean temperature changes. In our simulations we consider periodic shutdowns of the AMOC of a pre-selected duration in order to investigate whether these events could be detected in the ice cores. The question is of interest because the mixing timescale of δkr atm is an important factor on these short timescales. Mixing timescales of δkr atm increase during weak AMOC and Southern Ocean overturning cell conditions. The simulations are separated into simulations where the North Atlantic (referred to as NA runs), the Ross Sea (RO), the Weddell Sea (WE) and both Ross and Weddell Seas (RW) are perturbed. In the NA case, several 24-kyr simulations are performed where the North Atlantic from 5 to 7 N is perturbed with a freshwater flux F FW during τ years. The freshwater injection shuts down the AMOC. After τ years, the perturbation is reversed and freshwater is removed from the same ocean region for another τ years, whereby τ ranges from 2 to 4 years and F FW =.1 Sv. The AMOC recovers and overshoots. This periodic freshwater forcing is applied for the entire duration of the simulation. Note that the SSS change due to the freshwater perturbation has an effect on noble gas solubility. However, because the freshwater discharges associated with the Dansgaard-Oeschger events are not well known, we chose the smallest amount of freshwater required for the AMOC shutdown in order to minimize this effect. We achieve this by bringing the model close to the AMOC threshold by reducing the Atlantic-to-Pacific freshwater flux correction from.17 Sv (Ritz et al., 211) to.9 Sv. This reduces the AMOC from 14 Sv to 1.5 Sv at steady state. However, note that freshwater is removed from the North Atlantic during AMOC on phases of the simulations. This leads to an AMOC strength of approximately 19 Sv. Note that the ocean volume change caused by the freshwater discharge is not taken into account when calculating δkr atm. AMOC strength, Southern Ocean meridional overturning cell strength, mean atmospheric temperature, mean ocean temperature and δkr atm time series are displayed for a section of the run in Fig Peak-to-peak values of the oscillating δkr atm as a function of τ are shown in Fig In the Southern Ocean perturbation simulations, the Atlantic-to-Pacific freshwater flux correction is not modified. Simulations are performed for the same range of τ as in the NA runs and two cases for F FW =.5 and.1 Sv are considered. Unlike in the North Atlantic case, the freshwater perturbation is too small to shut down the Southern Ocean overturning cell completely. The overturning cell weakens from 18 Sv to approximately 14 Sv for F FW =.5 Sv and to approximately 11 Sv for F FW =.1 Sv in all cases RO, WE and RW (Fig. 4.6g). The AMOC weakens gradually during the perturbation. Therefore, its strength depends on the duration of the perturbation. Again, during the recovery phases of the simulations, fresh-

116 NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE North Atlantic Perturbation τ=.5 kyr τ=2 kyr Southern Ocean Perturbation SO MOC (Sv) AMOC (Sv) 22 b) c) f) g) h) d) ocean T ( C) δkratm ( ) time (kyr) i) e) δkratm ( ) τ=4 kyr atm T ( C) atm T ( C) SO MOC (Sv) AMOC (Sv) 1 τ=2 kyr 16 a) 2 ocean T ( C) τ=.5 kyr τ=4 kyr j) Weddell Sea Ross Sea Ross & Weddell Sea 18 2 time (kyr) 22 Figure 4.6: The response of δkratm to idealized Dansgaard-Oeschger simulations. a-e) A section of three of the seven 24 kyr long idealized Dansgaard-Oeschger simulations NA where.1 Sv of freshwater is injected into the North Atlantic. f-j) Dansgaard-Oeschger simulations WE where the Weddell Sea is perturbed by.1 Sv of freshwater. a, f) AMOC strength. b, g) Southern Ocean meridional overturning circulation strength. For τ = 2 kyr and τ = 4 kyr a 1-yr running average is calculated for better visibility. c, h) Mean atmospheric temperature. d, i) Global mean ocean temperature. e, j) δkratm. In panel j, for τ = 4 kyr, δkratm of the Ross Sea perturbation (RO) and of the Ross and Weddell Sea perturbation (RW) simulation are also displayed to emphasize the similarity of the different Southern Ocean perturbation simulations.

117 4.5. SENSITIVITY OF δkr ATM TO OCEAN MIXING 115 δkr atm peak to peak ( ) North Atlantic (.1 Sv) Ross Sea Weddell Sea Ross & Weddell Sea.5 Sv.1 Sv τ (kyr) Figure 4.7: Peak-to-peak change of δkr atm during idealized Dansgaard-Oeschger simulations as a function of freshwater perturbation duration τ (see also Figs. 4.6e,j) Thick line:.1 Sv North Atlantic perturbations (NA); squares: Ross Sea perturbations (RO); triangles: Weddell Sea perturbations (WE); diamonds: Ross and Weddell Sea perturbations (RW). For the Southern Ocean,.5 Sv (dashed lines) and.1 Sv (solid lines) perturbations are performed. water is removed from the ocean leading to stronger overturning cell strengths as compared to the standard values. Fig. 4.7 clearly reflects an increase of the peak-to-peak δkr atm change ( δkr atm ) with increasing perturbation duration τ. In the North Atlantic perturbation case, δkr atm =.65 for τ = 4 years. Note that both a stronger perturbation and taking into account the ocean volume change would increase δkr atm. A closer look at the transient evolution of NA mean ocean temperature (Fig. 4.6d) and δkr atm (Fig. 4.6e) reveals slight differences. In the model, the freshwater perturbation causes a brief rise in mean ocean temperature before it drops. This rise is associated with a drop in Southern Ocean overturning cell strength (Fig. 4.6c), because the entire Indopacific Ocean warms when it is less fed with the cold waters of the Southern Ocean. With a short delay, the warming is overcompensated by the cooling of the entire Atlantic due to the absence of relatively warm NADW. δkr atm also increases prior to the decrease. However, the drop begins earlier compared to mean ocean temperature. The origin for this decoupling lies partly with the freshening of the North Atlantic due to the freshwater discharge which increases solubility, and partly with the fact that in the model the AMOC shutdown leads to a SST reduction in the North-East Atlantic by up to 7 C together with a SSS reduction by around 3 psu. Both freshenings increase noble gas surface ocean solubility independent of temperature. The AMOC shutdown also leads to more sea ice in the North Atlantic and slightly less in the Southern Ocean, but this does not influence δkr atm. Still, the temperature is the major driver of δkr atm. Comparing the North Atlantic perturbation scenarios to the Southern Ocean perturbation simulations reveals substantial differences in the behavior of global mean ocean temperature (Figs. 4.6d and 4.6i) and δkr atm (Figs. 4.6e and 4.6j). In experiments RO, WE and RW, the Indopacific Ocean warms substantially due to the weakening of the Southern Ocean overturning caused by the Southern Ocean freshwater perturbation and as described above. The deep Atlantic also warms during the perturbation due to the lack of cold AABW. Because

118 NOBLE GASES AS PROXIES OF MEAN OCEAN TEMPERATURE the AMOC weakens only slightly, the upper Atlantic cools only moderately. The result is a warming of the global ocean and therefore an increase of δkr atm. The δkr atm response to the different Southern Ocean perturbation simulations is similar but the amplitudes vary depending on the perturbation region and strength (Fig. 4.6j). δkr atm values from.18 to.51 are reached for τ = 4 years (Fig. 4.7). 4.6 Conclusions We conclude that atmospheric noble gas concentrations are suitable proxies of global mean ocean temperature. Simulated global and annual mean atmospheric noble gas to nitrogen ratios for the last glacial maximum are δkr atm =.77, δxe atm = 2.36, and δar atm =.2. These simulated glacial-interglacial variations are ten to one hundred times larger than the seasonal variations of the present-day atmosphere. For δar atm, these seasonal variations have been measured at several globally distributed sites showing amplitudes of.12 to.37 (Battle et al., 23) and.5 to.15 (Keeling et al., 24). Seasonal global mean variations of δar atm are simulated to be.22. On glacial-interglacial timescales, changes in ocean volume and the associated change in sea-level pressure affect δkr atm substantially, even though glacial ocean volume is only 3 % smaller than during interglacial periods. Therefore, the sea-level history must be taken into account in the reconstruction. During glaciation and deglaciation periods, the timing of ocean temperature change relative to ocean volume change becomes important. Simulations of the last deglaciation using alternatively two scenarios of the Barbados uplift-corrected eustatic sea-level record of Peltier & Fairbanks (26), and the benthic δ 18 O stack ocean volume proxy of Lisiecki & Raymo (25) lead to an uncertainty of the calibration curve of ±.15 C due to the choice of the sea-level record. Because of uncertainties in the age-scales between the icecore and the sea-level record, we estimate an uncertainty of the δkr atm calibration curve of ±.3 C for the last deglaciation and up to ±.45 C for earlier transitions. During stadials and interstadials when sea-level was relatively constant and the values are well-established, the uncertainties are smaller. The uncertainty of the equilibrium climate sensitivity substantially adds to the calibration uncertainty, because for different climate sensitivities the relative contributions of ocean temperature and sea level to the δkr atm change are shifted. The uncertainty is estimated to be ±.35 C. The uncertainty which arises from geothermal heat flux changes does not exceed ±.15 C. Better constraints on equilibrium climate sensitivity and geothermal heat flux reduce these uncertainties. Sea ice has the potential to decouple δkr atm from ocean temperature by preventing airsea gas exchange. However, in the model this occurs only when wind-stress is reduced to below 6 % of the modern forcing, thus under extreme climate conditions of very weak ocean mixing combined with extensive sea-ice cover in the Southern Ocean to approximately 65 S throughout the year. But because wind-driven gyre transport is too weak in the model, Southern Westerlies and consequently the Antarctic Circumpolar Current must face an even larger change than the model suggests to decouple δkr atm from ocean temperature. δkr atm shows little dependence on the strength of ocean mixing as long as the climate is in equilibrium. Various idealized Dansgaard-Oeschger simulations provide an estimate of the δkr atm signal which is produced by abrupt climate change. The resulting δkr atm signal depends on the interval between subsequent abrupt changes and reaches.65 for a North Atlantic perturbation of 4 years when ignoring the effect of sea-level change.

119 4.7. ACKNOWLEDGMENTS 117 The climate model simulations presented here have implications for future measurements and their precision in order to constitute a useful paleoclimatic proxy. Together with the calibration uncertainties discussed in this paper, the precision of future ice core measurements will determine the usefulness of the noble gas proxies. The mean values and uncertainties of first results by Headly & Severinghaus (27) are.7 ±.3 and.14 ±.93 for the late Holocene and 1.34±.37 for the LGM. These uncertainties are high when compared with the expected signals from Dansgaard-Oeschger events simulated with this model. This suggests that measurements with significantly smaller uncertainty would be required in order to make noble gases a powerful paleoclimatic proxy for global ocean temperature changes. The low LGM value of the ice-core measurement compared to the simulated LGM signal hints towards cooler LGM temperatures than simulated with the model. This is supported by the simulated glacial-interglacial atmospheric temperature difference of 4.7 C, which is at the lower end of the observation based estimates of 4 C to 7 C (Jansen et al., 27). This possible underestimation could be due to a higher ECS than the 3 C assumed in the standard model setup, but also due to other uncertainties within the model such as the LGM ice-sheet extent and processes not included in the model, for example changes of the atmospheric dust load. The ability to reconstruct past global mean ocean temperature has major implications for climate research. First, the knowledge of past global mean ocean temperature gives a very valuable constraint to better tune paleoclimate models. Then, the evolution of ocean heat uptake can be deduced and with the ocean as the main storage of incoming energy (land surface and glaciers can be neglected), the evolution of the radiative imbalance of the atmosphere can be estimated. 4.7 Acknowledgments This study was funded by GRACCIE (CONSOLIDER-INGENIO 21). We thank J. Severinghaus, H. Fischer, T. Kellerhals and M. Leuenberger for fruitful discussions and comments. Appendix: Saturation Concentration and Schmidt Number Coefficients of Xenon Wanninkhof (1992) provides the Schmidt number of all gases discussed in this paper except xenon as a third order polynomial fit to observations Sc = A B(T T ) + C(T T ) 2 D(T T ) 3, (4.5) where T is the sea-surface temperature in Kelvin and T = K. We calculate the coefficients for xenon as described by Wanninkhof (1992) using data of Jähne et al. (1987) and obtain the coefficients A = ,B = K 1, C = K 2, and D = K 3. The saturation concentration of xenon in µmolkg 1 seawater as a function of temperature and salinity is fitted to the data of Wood & Caputi (1966) using the polynomial C s =.98µmol kg 1 exp(a + A 1 T s + A 2 T 2 s + S(B + B 1 T s )), (4.6) where S is the salinity in psu, ( ) K + T T T s = log, (4.7) T

120 118 BIBLIOGRAPHY and the coefficients A = , A 1 = , A 2 = , B = psu 1, and B 1 = psu 1 (Severinghaus, personal communication 21; Hamme & Severinghaus, 27). Bibliography Andersen, K. K., Svensson, A., Johnsen, S. J., Rasmussen, S. O., Bigler, M., Röthlisberger, R., Ruth, U., Siggaard-Andersen, M.-L., Steffensen, J. P., Dahl-Jensen, D., Vinther, B. M., & Clausen, H. B., 26. The Greenland Ice Core Chronology 25, ka. Part 1: constructing the time scale, Quaternary Science Reviews, 25, Barker, S., Cacho, I., Benway, H., & Tachikawa, K., 25. Planktonic foraminiferal Mg/Ca as a proxy for past oceanic temperatures: a methodological overview and data compilation for the Last Glacial Maximum, Quaternary Science Reviews, 24, Battarbee, R. W., 2. Palaeolimnological approaches to climate change, with special regard to the biological record, Quaternary Science Reviews, 19(1-5), Battle, M., Bender, M., Hendricks, M. B., Ho, D. T., Mika, R., McKinley, G., Fan, S. M., Blaine, T., & Keeling, R. F., 23. Measurements and models of the atmospheric Ar/N 2 ratio, Geophysical Research Letters, 3, Blunier, T., Chappellaz, J., Schwander, J., Dällenbach, A., Stauffer, B., Stocker, T. F., Raynaud, D., Jouzel, J., Clausen, H., Hammer, C., & Johnsen, S., Asynchrony of Antarctic and Greenland climate change during the last glacial period, Nature, 394, Blunier, T., Spahni, R., Barnola, J.-M., Chappellaz, J., Loulergue, L., & Schwander, J., 27. Synchronization of ice core records via atmospheric gases, Climate of the Past, 3, Brassell, S. C., Applications of biomarkers for delineating marine paleoclimatic fluctuations during the Pleistocene, in Organic Geochemistry: Principles and Applications, pp , eds Engel, M. H. & Macko, S. A., Plenum Press. Brassell, S. C., Eglinton, G., Marlowe, I. T., Pflaumann, U., & Sarthein, M., Molecular stratigraphy a new tool for climatic assessment, Nature, 32, Bryan, S. P. & Marchitto, T. M., 28. Mg/Ca-temperature proxy in benthic foraminifera: New calibrations from the Florida Straits and a hypothesis regarding Mg/Li, Paleoceanography, 23, PA222. Dansgaard, W., Johnsen, S. J., Clausen, H. B., Dahl-Jensen, D., Gundstrup, N. S., Hammer, C. U., Hvidberg, C. S., Steffensen, J. P., Sveinbjörnsdottir, A. E., Jouzel, J., & Bond, G., Evidence for general instability of past climate from a 25-kyr ice-core record, Nature, 364, Dymond, J. H. & Smith, E. B., 198. The virial coefficients of pure gases and mixtures: a critical compilation, Clarendon Press, Oxford. Elderfield, H., Yu, J., Anand, P., Kiefer, T., & Nyland, B., 26. Calibrations for benthic foraminiferal Mg/Ca paleothermometry and the carbonate ion hypothesis, Earth and Planetary Science Letters, 25, Enting, I. G., On the use of smoothing splines to filter CO 2 data, Journal of Geophysical Research, 92, 1,977 1,984. Ganachaud, A. & Wunsch, C., 2. Improved estimates of global ocean circulation, heat transport and mixing from hydrographic data, Nature, 48, Gruber, N., 24. The dynamics of the marine nitrogen cycle and its influence on atmospheric CO 2, in The Ocean Carbon Cycle and Climate, pp , eds Follows, M. & Oguz, T., NATO ASI Series, Kluwer Academic, Dordrecht, The Netherlands. Gruber, N., 28. The marine nitrogen cycle: overview and challenges, in Nitrogen in the marine environment, pp. 1 5, eds Capone, D. G., Bronk, D. A., Mulholland, M. R., & Carpenter, E. J., Academic Press, Amsterdam, The Netherlands, 2nd edn.

121 BIBLIOGRAPHY 119 Hamme, R. C. & Emerson, S. R., 24. The solubility of neon, nitrogen and argon in distilled water and seawater, Deep-Sea Research Part 1, 51, Hamme, R. C. & Severinghaus, J. P., 27. Trace gas disequilibria during deep-water formation, Deep-Sea Research Part 1, 54, Headly, M. A. & Severinghaus, J. P., 27. A method to measure Kr/N 2 ratios in air bubbles trapped in ice cores and its application in reconstructing past mean ocean temperature, Journal of Geophysical Research, 112, D1915. Hegerl, G. C., Zwiers, F. W., Braconnot, P., Gillett, N. P., Luo, Y., Marengo Orsini, J. A., Nicholls, N., Penner, J. E., & Stott, P. A., 27. Understanding and attributing climate change, in Climate Change 27: The Physical Science Basis. Working Group I Contribution to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pp , eds Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tignor, M., & Miller, H. L., Cambridge University Press. Jähne, B., Heinz, G., & Dietrich, W., Measurement of the diffusion coefficents of sparingly soluble gases in water, Journal of Geophysical Research, 92, Jansen, E., Overpeck, J., Briffa, K. R., Duplessy, J.-C., Joos, F., Masson-Delmotte, V., Olago, D., Otto- Bliesner, B., Peltier, W. R., Rahmstorf, S., Ramesh, R., Raynaud, D., Rind, D., Solomina, O., Villalba, R., & Zhang, D., 27. Palaeoclimate, in Climate Change 27: The Physical Science Basis. Working Group I Contribution to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pp , eds Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tignor, M., & Miller, H. L., Cambridge University Press. Johnsen, S. J., Dahl-Jensen, D., Gundestrup, N., Steffensen, J. P., Clausen, H. B., Miller, H., Masson- Delmotte, V., Sveinbjornsdottir, A. E., & White, J., 21. Oxygen isotope and palaeotemperature records from six Greenland ice-core stations: Camp Century, Dye-3, GRIP, GISP2, Renland and NorthGRIP, Journal of Quaternary Science, 16(4), Jouzel, J., Masson-Delmotte, V., Cattani, O., Dreyfus, G., Falourd, S., Hoffmann, G., Minster, B., Nouet, J., Barnola, J. M., Chappellaz, J., Fischer, H., Gallet, J. C., Johnsen, S., Leuenberger, M., Loulergue, L., Luethi, D., Oerter, H., Parrenin, F., Raisbeck, G., Raynaud, D., Schilt, A., Schwander, J., Selmo, E., Souchez, R., Spahni, R., Stauffer, B., Steffensen, J. P., Stenni, B., Stocker, T., Tison, J.-L., Werner, M., & Wolff, E., 27. Orbital and millennial Antarctic climate variability over the past 8, years, Science, 317, Keeling, R. F., Blaine, T., Paplawsky, B., Katz, L., Atwood, C., & Brockwell, T., 24. Measurement of changes in atmospheric Ar/N 2 ratio using a rapid-switching, single-capillary mass spectrometer system, Tellus B, 56, Kopp, R. E., Simons, F. J., Mitrovica, J. X., Maloof, A. C., & Oppenheimer, M., 29. Probabilistic assessment of sea level during the last interglacial stage, Nature, 462, Lisiecki, L. E. & Raymo, M. E., 25. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ 18 O records, Paleoceanography, 2, PA13. McDermott, F., 24. Palaeo-climate reconstruction from stable isotope variations in speleothems: a review, Quaternary Science Reviews, 23, Mélières, M.-A., Martinerie, P., Raynaud, D., & Lliboutry, L., Glacial-interglacial mean sea level pressure change due to sea level, ice sheet and atmospheric mass changes, Global and Planetary Change, 3, Monnin, E., Indermühle, A., Dällenbach, E., Flückiger, J., Stauffer, B., Stocker, T. F., Raynaud, D., & Barnola, J.-M., 21. Atmospheric CO 2 concentrations over the Last Glacial Termination, Science, 291, Müller, P. J., Kirst, G., Ruhland, G., von Storch, I., & Rosell-Melé, A., Calibration of the alkenone paleotemperature index U K 37 based on core-tops from the eastern South Atlantic and the global ocean (6 N-6 S), Geochimica et Cosmochimica Acta, 62(1), Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 26. Water mass distribution and ventilation time scales in a cost-efficient, three-dimensional ocean model, Journal of Climate, 19,

122 12 BIBLIOGRAPHY Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 28. Modeled natural and excess radiocarbon: Sensitivities to the gas exchange formulation and ocean transport strength, Global Biogeochemical Cycles, 22, GB311. Orr, J. C., On ocean carbon-cycle model comparison, Tellus, 51B, Overpeck, J. T., Webb, T., & Prentice, I. C., Quantitative interpretation of fossil pollen spectra: Dissimilarity coefficients and the method of modern analogs, Quaternary Research, 23, Peltier, W. R. & Fairbanks, R. G., 26. Global glacial ice volume and Last Glacial Maximum duration from an extended Barbados sea level record, Quaternary Science Reviews, 25, Pollack, H. N., Hurter, S. J., & Johnson, J. R., Heat flow from the Earth s interior: analysis of the global data set, Reviews of Geophysics, 31, Quinby-Hunt, M. S. & Turekian, K. K., Distribution of elements in sea water, EOS, 64, Rasmussen, S. O., Andersen, K. K., Svensson, A. M., Steffensen, J. P., Vinther, B. M., Clausen, H. B., Siggaard-Andersen, M. L., Johnsen, S. J., Larsen, L. B., Dahl-Jensen, D., Bigler, M., Röthlisberger, R., Fischer, H., Goto-Azuma, K., Hansson, M. E., & Ruth, U., 26. A new Greenland ice core chronology for the last glacial termination, Journal of Geophysical Research, 111, D6. Reimer, P. J., Baillie, M. G. L., Bard, E., Bayliss, A., Beck, J. W., Blackwell, P. G., Ramsey, C. B., Buck, C. E., Burr, G. S., Edwards, R. L., Friedrich, M., Grootes, P. M., Guilderson, T. P., Hajdas, I., Heaton, T. J., Hogg, A. G., Hughen, K. A., Kaiser, K. F., Kromer, B., McCormac, F. G., Manning, S. W., Reimer, R. W., Richards, D. A., Southon, J. R., Talamo, S., Turney, C. S. M., van der Plicht, J., & Weyhenmeyer, C. E., 29. INTCAL9 AND MARINE9 radiocarbon age calibration curves, -5, years Cal BP, Radiocarbon, 51, Ritz, S. P., Stocker, T. F., & Joos, F., 211. A coupled dynamical ocean - energy balance atmosphere model for paleoclimate studies, Journal of Climate, 24, Sarmiento, J. L. & Gruber, N., 26. Ocean biogeochemical dynamics, Princeton University Press, Princeton and Oxford. Schlesinger, W. H., Biogeochemistry, an analysis of global change, Academic Press, 2nd edn. Schouten, S., Hopmans, E. C., Schefuss, E., & Damste, J. S. S., 22. Distributional variations in marine crenarchaeotal membrane lipids: a new tool for reconstructing ancient sea water temperatures?, Earth and Planetary Science Letters, 24, Schouten, S., Hopmans, E. C., Forster, A., van Breugel, Y., Kuypers, M. M. M., & Damste, J. S. S., 23. Extremely high sea-surface temperatures at low latitudes during the middle Cretaceous as revealed by archaeal membrane lipids, Geology, 31(12), Siddall, M., Rohling, E. J., Almogi-Labin, A., Hemleben, C., Meischner, D., Schmelzer, I., & Smeed, D. A., 23. Sea-level fluctuations during the last glacial cycle, Nature, 423, Wanninkhof, R., Relationship between wind speed and gas exchange over the ocean, Journal of Geophysical Research, 97, Weiss, R. F., 197. The solubility of nitrogen, oxygen and argon in water and seawater, Deep-Sea Research, 17, Weiss, R. F. & Kyser, T. K., Solubility of krypton in water and seawater, Journal of Chemical and Engineering Data, 23, Wood, D. & Caputi, R., Solubilities of Kr and Xe in fresh and sea water, Tech. rep., U.S. Naval Radiological Defense Laboratory, San Francisco, CA.

123 Chapter 5 Atmospheric Radiocarbon during the Last Deglaciation A Modeling Perspective Abstract During the last 3, years, the atmospheric 14 C ( 14 C atm ) has decreased from approximately 7 to according to reconstructions from tree-rings, fossil corals, and varved marine sediments. However, the reason for this large decline has not yet been successfully explained. Major contributions to the decline that have been discussed in the literature include a reduction in the radiocarbon production rate, or a decreased residence time of radiocarbon in the ocean after the last glacial maximum. The latter may be caused by a weaker meridional overturning circulation, and poorly ventilated, radiocarbon-depleted abyssal water masses during glacial times that mix with the rest of the ocean during the deglaciation. Many research groups have been searching for these radiocarbon-depleted water masses in recent years, some with success. Here, a reduced complexity coupled ocean-atmosphere carbon-cycle climate model is used to simulate changes in 14 C atm inferred from the two production rate reconstructions available today, due to rapid reductions of the Atlantic meridional overturning circulation, and due to the difference between the glacial and the modern state of the ocean. In three scenarios, last glacial maximum (LGM) radiative forcing and carbon cycle conditions are established and the modeled benthic-planktonic radiocarbon age differences are compared to reconstructions. Depending on the scenario, the LGM-to-modern 14 C atm difference ranges between 4 and 219. However, more simulations must be performed where more processes involved in the glacial-interglacial carbon-cycle change are included in order to consolidate this range. If the production rate reconstruction of Laj et al. (22) can be believed, then the resulting changes in 14 C atm follow the 14 C atm reconstructions back to 2 kyr BP. However, the observed 14 C atm decline from 3 to 2 kyr BP cannot be explained. The radiocarbon production rate reconstruction of Muscheler et al. (24) results in a 14 C atm history that is several hundred per mil lower than the 14 C atm reconstructions and that does not reproduce the 14 C atm decline during the deglaciation. 5.1 Introduction A question yet unsolved by the paleoclimate community is the reason for the large decline of atmospheric radiocarbon ( 14 C) during the last deglaciation as found in several reconstructions (Fig. 5.1e). Broecker & Barker (27) give a summary of the possible mechanisms involved in the atmospheric radiocarbon change while focusing on the Mystery Interval from 17.5

124 ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION to 14.5 kyr BP (before present, before year 195 AD). The first discussed mechanism is variations in the radiocarbon production rate at the top of the atmosphere. Radiocarbon is produced by cosmogenic radiation, i.e. by protons, that interact with the particles of the atmosphere to produce pions, neutrons, electrons etc. Atmospheric nitrogen may capture such a neutron whereby it forms radiocarbon and a proton. The radiocarbon atom then oxidizes rapidly to 14 CO 2. The radiocarbon production rate is proportional to the flux of cosmogenic particles which depends on the strength of the geomagnetic and solar magnetic fields (Masarik & Beer, 1999). The geomagnetic field shields the atmosphere from cosmic rays, the solar magnetic field on the other hand modulates the galactic cosmic ray flux. The produced atmospheric radiocarbon is exchanged into the ocean and the biosphere, where it decays with a half-life of 573 years. The second mechanism discussed by Broecker & Barker (27) is the release of an isolated radiocarbon-depleted carbon reservoir in the abyssal ocean. The hypothesis is that large water masses have been isolated in the abyssal ocean during glacial times thereby hardly mixing with overlying water masses, thus allowing radiocarbon to decay to very low concentrations. After a strengthening of the deep ocean circulation during the deglaciation, these water masses start to mix with the rest of the ocean. The radiocarbon depleted carbon finally exchanges with the atmosphere leading to a decrease of the atmospheric radiocarbon concentration. Other mechanisms discussed by Broecker & Barker (27) are changes in the marine reservoir age (a measure proportional to the radiocarbon concentration difference between the air and the surface ocean, see Chapter 6), the release of radiocarbon stored in sediments during glacial times, and a massive injection of radiocarbon-free methane stored in organic-rich ocean-margin sediments into the atmosphere. They also discuss the validity of the atmospheric radiocarbon reconstructions as well as the possibility for the radiocarbon half-life to be significantly greater than the currently accepted value of 573 years. They conclude that the observed atmospheric radiocarbon drop is well documented in several independent records (Fig. 5.1), that a scenario of a large radiocarbon-free methane injection into the atmosphere can be ruled out, that an increase of the radiocarbon half-life is unlikely, but that reconstructions of the radiocarbon production rate are in disagreement and that a radiocarbon-depleted carbon reservoir in the abyssal ocean is plausible. The hypothesis of a radiocarbon depleted ocean reservoir is popular, because it is scientifically accepted that the residual between glacial and modern atmospheric CO 2 (CO 2,atm, Fig. 5.1c) was stored in the ocean during glacial periods. If the outgassed CO 2 was radiocarbon depleted, then there is a chance of the existence of an abyssal carbon reservoir. In the recent years many research groups have made efforts to find radiocarbon depleted waters within the ocean. Marchitto et al. (27) and Stott et al. (29) claim to have found evidence for the upwelling of radiocarbon-depleted waters in the eastern Pacific during Heinrich event 1 (H1, from about 16.8 to 14.6 kyr BP; Hemming, 24; Rasmussen et al., 26) and during the Younger Dryas (YD, from about 12.8 to 11.6 kyr BP; Rasmussen et al., 26). Rose et al. (21) and Bryan et al. (21) find similar evidence from the southwestern Pacific and the Arabian Sea, respectively. Broecker & Clark (21), Magana et al. (21), and De Pol-Holz et al. (21), on the other hand, find no evidence of radiocarbon depleted waters in the central equatorial Pacific, the eastern equatorial Pacific, and off the coast of Chile, respectively. From a sediment core of the Atlantic sector of the Southern Ocean, Skinner et al. (21) find evidence that the deep water circulating around Antarctica was more than two times older than today relative to the atmosphere, and Thornalley et al. (211) report extremely radiocarbon-depleted water masses from the North Atlantic.

125 5.1. INTRODUCTION atm C ( ) normalized C production rate a) 1.6 Laj et al., 22 d) 1.5 Muscheler et al., normalized 14 C production rate 5 Hughen et al., 26 Intcal4 (Reimer et al., 24) Fairbanks et al., 25 Intcal9 (Reimer et al., 29) 4 Hoffmann et al., 21 Hughen et al., 26 Fairbanks et al., 25 b) e) 14 C ( ) atm Laj et al., 22 Muscheler et al., 24 Hoffmann et al., Intcal4 (Reimer et al., 24) Intcal9 (Reimer et al., 29) H1 B-A YD 3 c) 3 f) atm CO 2 (ppm) 25 2 atm CO 2 (ppm) Time (kyr BP) H1 B-A YD 15 1 Time (kyr BP) 5 Figure 5.1: Reconstructions of a) radiocarbon production rate, where the original data was splined with a cut-off period of 3 kyr, b) atmospheric 14 C, and c) atmospheric CO 2 (Monnin et al., 21; Siegenthaler et al., 25; Lüthi et al., 28) for the last 5 kyr. d-f) as a-c, but for the last 2 kyr. Note that the Intcal4 and Intcal9 atmospheric 14 C records are compilations of several data sets. Intcal9 includes the data set of Hughen et al. (26). The purpose of this study is to discuss the different mechanisms that might have played a role in the observed deglacial decline of atmospheric radiocarbon and to quantify possible contributions of each mechanism using a reduced-complexity coupled ocean-atmosphere carbon-cycle climate model with a three-dimensional representation of the ocean. The study focuses on changes in the radiocarbon production rate, on changes in the ocean circulation and on the release of radiocarbon-depleted carbon during the deglaciation. Unfortunately, the model fails to simulate realistic representations of the glacial climate with respect to the current knowledge of atmospheric CO 2 concentrations and the state of the ocean. At this point, suggestions are made for future studies when the knowledge on the processes involved in the glacial-interglacial atmosphere-ocean carbon exchange will be more advanced.

126 ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION 5.2 Model Setup For this study the Bern3D coupled ocean-atmosphere model is used. The ocean component is a frictional geostrophic balance ocean model (Edwards et al., 1998; Müller et al., 26) with a horizontal resolution of 36 by 36 grid cells and 32 depth layers. One model year is simulated in 48 time steps. The atmosphere is described by an energy and moisture balance model of equal spatial and temporal resolution as the ocean (Ritz et al., 211). Sea ice is dynamically calculated. The model is run with a prognostic carbon cycle following protocols of OCMIP-2 (Orr et al., 1999), Tschumi et al. (28), and Parekh et al. (28). Radiocarbon is simulated as separate dissolved inorganic carbon (DIC) and organic carbon (DOC) tracers DIC-14 and DOC-14, respectively (Müller et al., 28). T. Tschumi and P. Parekh (personal communication 28) report too high modeled tracer concentrations in the Southern Ocean when using the air-sea gas transfer velocity proposed by the OCMIP-2 formulation which is proportional to the squared surface wind speed. As a consequence, a formulation was used by Tschumi et al. (28) and Parekh et al. (28) where the gas transfer velocity is no longer wind-speed dependent. Here, a more realistic formulation is used where the air-sea gas transfer velocity is proportional to the surface wind speed: k = C 1 C 2 u (Sc/66) 1/2 (1 A i ), (5.1) where C 1 = is a scaling factor to ensure that the global annual mean airsea gas exchange remains equal to the formulation of OCMIP-2, C 2 =.81 is a scaling factor proposed by Müller et al. (28) for better agreement between modeled ocean tracer inventories and observations, u is the cross-calibrated, multiplatform (CCMP) ocean 1 m surface wind speed climatology from the Physical Oceanography Distributed Active Archive Center (PO.DAAC, data available online at Sc is the tracer dependent Schmidt number, and A i the fractional sea-ice cover. The terrestrial biosphere is simulated by the box model of Siegenthaler & Oeschger (1987). It is composed of 4 wellmixed compartments of ground vegetation and leaves (1 GtC), wood (5 GtC), detritus (12 GtC), and soils (15 GtC). The carbon stocks and exchange fluxes remain constant. The model spinup involves spinup of the physics of the coupled atmosphere-ocean model (Ritz et al., 211) followed by a 1 kyr spinup of the carbon-cycle. During the spinup, atmospheric radiocarbon is fixed to 14 C =, where ( 14 ) 14 R C = , (5.2) R std with 14 R std = the pre-industrial atmospheric 14 C-to- 12 C ratio (Karlén et al., 1964) and 14 R the 14 C-to- 12 C ratio at any given time. After the spinup, the model specific radiocarbon production rate P is determined for the pre-industrial state by solving the steady state radiocarbon balance equation for the production rate P and averaging P over a 5-year control simulation: P = λi atm +F ao A o +F at A t, where λ = 1/8267 yr 1 is the decay constant of radiocarbon (Godwin, 1962), I atm the atmospheric radiocarbon inventory, F ao and F at the net atmosphere-ocean and atmosphere-terrestrial biosphere radiocarbon fluxes, respectively, A o the global sea-surface area, and A t the global land-surface area. The prescribed CO 2,atm of the greenhouse gas radiative forcing is decoupled from the CO 2,atm calculated by the carbon cycle. Last glacial maximum (LGM) conditions are simulated by prescribing the radiative CO 2,atm to 18 parts per million (ppm), CH 4 to 35 parts per billion (ppb), and ice-sheet albedo according to Peltier (24). Freshwater is removed from the ocean due to the presence of the ice sheets in order to represent the more saline glacial ocean. Since the Bern3D model is a rigid-lid model, salt is added to the ocean instead of removing freshwater. Additionally,

127 5.3. RADIOCARBON PRODUCTION RATE 125 LGM aeolian iron fluxes are applied (Mahowald et al., 26). In one case, the terrestrial biosphere is kept constant at modern conditions. In the other case, the soil carbon stock is reduced by 64 PgC. The terrestrial glacial carbon pool is estimated to contain 3-7 Pg less carbon than today (Köhler & Fischer, 24, and references therein). The value of 64 PgC is chosen for comparison reasons with the study of Bouttes et al. (211). The carbon fluxes of the terrestrial biosphere are kept constant at modern conditions. Note that the model used in this study does not include a marine sediment diagenesis model. 5.3 Radiocarbon Production Rate There are currently two data sets available of temporal changes in radiocarbon production rate of the past 6 kyr. The first is provided by Muscheler et al. (24) and is based on 1 Be measurements in Greenland ice cores (Fig. 5.1a). As radiocarbon, 1 Be is produced by cosmogenic radiation at the top of the atmosphere. However, as opposed to radiocarbon, 1 Be is attached to aerosols and removed from the atmosphere after a mean residence time of 1-2 years (Raisbeck et al., 1981). The second record is provided by Laj et al. (22) and is based on marine sediment paleointensity records that are proportional to the strength of the Earths magnetic field (Fig. 5.1a). The two production rate reconstructions show large differences prior to 15 kyr BP, but both records represent important features such as the Laschamp-event at around 4 kyr BP, a geomagnetic excursion where the geomagnetic field intensity was substantially reduced and the orientation of the Earth s magnetic field changed temporarily. In order to quantify the atmospheric 14 C ( 14 C atm ) changes due to the changes in radiocarbon production rate, four simulations are performed that span the last 6 kyr where the modeled radiocarbon production rate determined during the model spinup is scaled by the normalized production rate by either Laj et al. (22) or Muscheler et al. (24). Two simulations are performed using the production rate of Laj et al. (22). In the first, the original data are used, and in the second, the model is forced with a splined version with a cut-off period of 3 kyr to remove short-term variability. The third and fourth simulations are analogous but forced by the production rate of Muscheler et al. (24) (Fig. 5.2). Because of the radiocarbon half-life of 573 years, a radiocarbon signal remains in the climate system for several thousand years. Thus, in order to correctly simulate the radiocarbon distribution at a given time, the radiocarbon evolution of at least the prior 3 kyr must be simulated. Therefore, only the model results of the last 3 kyr are robust. As can be presumed from the discrepancies between the two production rate reconstructions prior to 15 kyr BP, the results of the model simulations differ substantially with 14 C atm differences of about 3 between 3 and 2 kyr BP. The simulation forced by the Muscheler et al. (24) relative production rate also remains several hundred per mil below the 14 C atm reconstructions and does not suggest a decline of 14 C atm during the deglaciation. On the other hand, the simulation forced with the Laj et al. (22) relative production rate closely follows the 14 C atm reconstructions from the LGM at around 2 kyr BP to the present and remains at a stable level between 3 and 4 prior to 2 kyr BP in accordance to the reconstruction by Hoffmann et al. (21). However, the 14 C decline from 3 to 2 kyr BP suggested by the Intcal9 reconstruction (Reimer et al., 29) is not reproduced. To this date it is not clear which of the two production rate reconstructions is more realistic, nor can one be rejected. The same is true for the different 14 C atm reconstructions. In the next section, the possibility of a radiocarbon-depleted ocean reservoir during glacial times is discussed.

128 ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION C Production: Laj et al., 22 original 14 C Production: Laj et al., 22 spline 14 C Production: Muscheler et al., 24 original 14 C Production: Muscheler et al., 24 spline Intcal9 (Reimer et al., 29) Hoffmann et al., C ( ) atm Spinup 3 Time (kyr BP) 2 1 Figure 5.2: Model simulations of the last 6 kyr where the model is forced with the relative radiocarbon production rate of Laj et al. (22) (blue lines) and Muscheler et al. (24) (red lines). In one case the model is forced with the original data sets (light colored lines), in the other case the model is forced with a splined production rate (3 kyr cut-off period, dark colored lines). The model results are compared to the Intcal9 data set (black line, Reimer et al., 29) and to a new data set by Hoffmann et al. (21) (gray line). The first 3 kyr of the simulation are considered as a spinup for the second 3 kyr of the simulation. Every data set is plotted on its own age scale. For the radiocarbon production rate forcing of the model simulations, the GRIP ss9 age scale is used. 5.4 Radiocarbon Distribution of the LGM Ocean Depending on the glacial state of the ocean, the residence time of carbon within the ocean might have been increased, leading to a more radiocarbon-depleted ocean carbon inventory as compared to the pre-industrial state. If the atmospheric production rate is assumed constant, then consequently 14 C atm must have been higher. If this was the case, then during the deglaciation, when the ocean circulation changes to its modern state, 14 C atm is reduced because more radiocarbon enriched carbon is taken up by the ocean. During glacial times, atmospheric CO 2 was at 18 to 19 ppm, i.e., 9 to 1 ppm lower than the preindustrial value of about 28 ppm. The corresponding amount of carbon was stored most likely in the ocean. If the ventilation of the glacial ocean was truly weaker, then the extra carbon released to the atmosphere during the deglaciation would additionally have lowered 14 C atm. Several mechanisms have been proposed to explain the increased ocean carbon inventory during glacial times (for reviews, see Sigman & Boyle, 2; Archer et al., 2; Fischer et al., 21; Sigman et al., 21). Although all mechanisms transfer CO 2 from the atmosphere to the ocean, they have different effects with respect to radiocarbon. Here it is proposed to use a reduced-complexity climate model to perform a sensitivity study where a glacial CO 2,atm concentration of 19 ppm is invoked by several combinations of the proposed mechanisms. The glacial-ocean radiocarbon distribution of every solution can be compared to reconstructed benthic-planktonic radiocarbon age differences (τ BP ) to validate the modeled

129 5.4. RADIOCARBON DISTRIBUTION OF THE LGM OCEAN 127 LGM-to-modern 14 C atm change. τ BP = 1 ( 14 ) λ ln R oc (z = ) 14 R oc (z = H), (5.3) where 14 R(z) is the 14 C-to- 12 C ratio at the surface (z = ) and bottom (z = H) of the ocean. H represents the depth of the water column and varies spatially. Unfortunately, with the mechanisms tested below, it is not possible to reach CO 2,atm conditions of 19 ppm, unless the model state is pushed to unrealistic solutions. Still, in order to illustrate the procedure, these unrealistic solutions are discussed. More work needs to be done where more possible mechanisms involved in the glacial-interglacial CO 2,atm change are taken into account Possible Mechanisms to Reduce Atmospheric CO 2 The model is brought to glacial conditions by applying glacial greenhouse gas concentrations, ice sheet extent and aeolian iron fluxes. The terrestrial biosphere is kept at modern conditions. This model state is henceforth referred to as STDLGM. Simulated CO 2,atm is 254 ppm in STDLGM (when the terrestrial biosphere is reduced by 64 PgC, CO 2,atm is increased to 34 ppm). Note that the prescribed CO 2,atm of the greenhouse gas radiative forcing is decoupled from the CO 2,atm calculated by the carbon cycle. The lower CO 2,atm compared to modern CO 2,atm arises from the higher solubility of the cooler ocean, the extended presence of sea ice that reduces air-sea gas exchange, the response of the biogeochemistry to the cooler conditions and different ocean circulation state, and from the increased aeolian iron fluxes. Three additional adjustments are made to the model to further reduce CO 2,atm. In the first, a brine rejection parametrization recently proposed by Bouttes et al. (21) is implemented into the model. Bouttes et al. (21) argue that very dense brines that are formed as a consequence of sea-ice formation sink from the surface ocean to the abyss along the Antarctic Shelf thereby hardly mixing with surrounding waters. This mechanism leads to a stratification of the deep ocean. As this is a sub-gridscale process, it is not automatically represented by the model and must therefore be parametrized. A parameter η determines the fraction of the brines that sink to the abyss. The rest is mixed in the surface ocean. The second process that is taken into account is the northward shift of the Southern Hemisphere westerly winds during the LGM (Toggweiler et al., 26). According to their hypothesis, this reduces the wind stress over the Southern Ocean and consequently the Antarctic Circumpolar Current. This leads to a decrease of the Southern Ocean overturning cell strength which in turn increases the residence time of carbon within the ocean. This process is applied by reducing the Southern Ocean wind stress relative to modern conditions. Third, the Martin-curve exponent, a biogeochemical parameter, is changed to deepen the LGM remineralization depth of exported particulate organic carbon (POC) (Kwon et al., 29). Using the Climber-2 EMIC, Bouttes et al. (211) manage to tune their model into a glacial state with a CO 2,atm of 19 ppm and a surface-to-deep δ 13 C gradient that matches reconstructions of the LGM. In order to achieve this result, the brine rejection parametrization of Bouttes et al. (21) is applied as well as an iron-fertilization parametrization, because an iron cycle is not explicitly included in their carbon-cycle module. As a third process, they have applied a stratification-dependent vertical diffusivity. An LGM simulation without these three processes leads to a CO 2,atm of 257 ppm. Note that this result includes a 64 Pg glacial loss of carbon of the terrestrial biosphere. Carbonate compensation by a sediment diagenesis model compensates the additional carbon input of the terrestrial biosphere. Without carbonate compensation, CO 2,atm is increased to 298 ppm in their model. The brine rejection

130 ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION parametrization, where η =.6, additionally reduces CO 2,atm by 42 ppm. Iron fertilization by increased aeolian iron fluxes reduces CO 2 by 15 ppm, and the effect of the stratification dependent vertical diffusivity reduces CO 2 by 8 ppm. Here, the same approach is tested with the Bern3D model. However, the process of carbonate compensation is not simulated because of the lack of a sediment diagenesis model. Modern terrestrial biosphere conditions are applied in order to begin with a comparable CO 2,atm concentration to the initial condition of Bouttes et al. (211) (254 ppm in STDLGM vs. 257 ppm). Since iron is explicitly simulated, the iron fertilization parametrization cannot be applied. The application of the brine rejection parametrization at the ocean cells adjacent to Antarctica with η =.6 reduces CO 2,atm by only 11 ppm to 243 ppm. If η = 1, CO 2,atm is reduced by 17 ppm. Thus, the effect of the brine rejection parametrization is by far weaker than in the results of Bouttes et al. (211). Applying the brine rejection formulation not only to the model cells adjacent to Antarctica, but to all model cells with η = 1, CO 2,atm is reduced by 25 ppm. Setting the diapycnal diffusivity from 1 5 m 2 s 1 to very low 1 8 m 2 s 1 throughout the ocean (a simplification of the stratification dependent diffusivity with a more extreme effect) reduces CO 2,atm by another 3 to 9 ppm. To conclude, the mechanisms proposed by Bouttes et al. (211) that lead to CO 2,atm concentrations of 19 ppm are not nearly as efficient in the Bern3D model unless carbonate compensation is responsible for the large discrepancy. Reducing the wind stress south of 34 S in the Bern3D model reduces CO 2,atm by a maximum of 14 ppm to 24 ppm. In this case, the Southern Ocean wind stress is reduced to 3 % of the modern values. Unfortunately, also the combination of brine rejection parametrization together with the reduction of Southern Hemisphere westerlies does not permit a reduction to glacial CO 2,atm conditions in the Bern3D model Extreme Scenarios where Atmospheric CO 2 is Brought to 19 ppm With even more extreme scenarios, three glacial model states are realized where CO 2,atm is lowered to approximately 19 ppm (Table 5.1). In the first, the wind stress is reduced globally to 35 % of modern conditions (simulation is referred to as WINDONLY). This lowers CO 2,atm to 192 ppm and increases 14 C atm to 219. In the second, global wind stress is scaled down to 62 % of modern conditions, and the brine rejection parametrization is applied to the entire ocean with η = 1 (WINDBRINE). This reduces CO 2,atm to 19 ppm and increases 14 C atm to 121. In the third realization, the POC remineralization depth scale is increased from an e-folding depth of 246 m to 2266 m (POCREM). This change affects the carbon cycle, but not the ocean physics. CO 2,atm is reduced to 194 ppm, 14 C atm is increased by 4. Note that with the inclusion of a marine sediment diagenesis module, a much smaller change of the e-folding depth is required to reach a CO 2,atm of 19 ppm (R. Roth, personal communication 211). Although these simulations represent very unrealistic cases, they illustrate the method proposed here to assess the possible range of the LGM-to-modern 14 C atm change. Benthic-planktonic age differences as well as the Atlantic meridional overturning circulation (AMOC) of the modern ocean and of the three glacial states are displayed in Fig Simulated modern benthic-planktonic age differences vary between and 2 years and compare well with the observations. The three LGM simulations differ substantially from each other with respect to benthic-planktonic age differences. Largest values of up to 4 years in the mid-latitude Pacific and up to 35 years in the North Atlantic are simulated in the WINDONLY scenario. These large values arise because of the reduced overturning circulation in the Atlantic and Pacific due to the low wind stress. Reconstructions from the

131 RADIOCARBON DISTRIBUTION OF THE LGM OCEAN 129 Benthic-planktonic age differences (yr) AMOC (Sv) Modern Depth (km) Latitude WINDONLY Depth (km) Latitude WINDBRINE Depth (km) Latitude POCREM Depth (km) Latitude Figure 5.3: Benthic-Planktonic radiocarbon age differences (τ BP = λ 1 ln( 14 R oc(z = )/ 14 R oc(z = H)), where 14 R is the 14 C-to- 12 C ratio, and z = and z = H stand for the surface and bottom of the ocean, respectively, with H the depth of the water column) for the modern model state and three LGM scenarios (Table 5.1) as well as the corresponding Atlantic meridional overturning circulation (AMOC). The model results are compared to modern measurements and LGM reconstructions of Broecker et al. (199), Broecker et al. (24), Robinson et al. (25), Broecker & Clark (21), Barker et al. (21), Skinner et al. (21), and Thornalley et al. (211). Often, LGM τ BP reconstructions lie within a range of values. Then the range is indicated by two colors representing the minimum and maximum value, respectively.

132 13 5. ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION Table 5.1: Summary of the LGM model scenarios. Note that the CO 2,atm of the radiative forcing is decoupled from the CO 2,atm calculated by the carbon cycle (third column). The resulting 14 C atm is given in the last column. Name Settings CO 2,atm 14 C atm STDLGM radiative CO 2,atm = 18 ppm, CH 4,atm = 35 ppb; 254 ppm 1 LGM ice sheet extent; LGM aeolian iron fluxes WINDONLY STDLGM settings; 35 % of modern wind stress 192 ppm 219 WINDBRINE POCREM STDLGM settings; 62 % of modern wind stress; global ocean η = 1 brine rejection STDLGM settings; POC remineralization e-folding depth from 246 m to 2266 m 19 ppm ppm 4 North Atlantic partially agree with the model results. However, the range given by the reconstructions is large. The reconstructions from the Pacific suggest much lower values than the simulation. Unfortunately, no reliable reconstructions are available from the open Pacific Ocean. The AMOC is almost shut down in this simulation. However, paleoceanographic reconstructions and paleoclimatic model intercomparison projects do not suggest such a weak and shallow LGM AMOC (Lynch-Stieglitz et al., 27; McManus et al., 24; Otto-Bliesner et al., 27). Benthic-planktonic age differences are generally lower in the WINDBRINE scenario reaching values of up to 3 years in the midlatitude Pacific. The data suggest lower values than simulated in the Pacific and higher values in the North Atlantic. Due to the extreme parameter choice for the brine rejection parametrization, the simulated ocean above 2 to 3 km depth is extremely stratified. Consequently, the AMOC is shut down. Note that due to the weak overturning circulation and therefore the weak poleward heat transport in the WINDONLY and WINDBRINE scenarios, sea-ice extends towards lower latitudes compared to the STDLGM case. Consequently, mean atmospheric and ocean temperatures are also cooler. Because in scenario POCREM only the biogeochemistry is modified but not the physics, the ocean circulation remains equal to the STDLGM ocean state. Benthic-planktonic age differences only exceed 2 years in a few Pacific cells. Modeled Pacific ages compare well to the reconstructions. However, reconstructions of the North Atlantic suggest older ages than the simulation. To summarize, when the radiocarbon production rate is assumed constant, the LGM-tomodern 14 C atm difference ranges between 4 and 219 in the three extreme scenarios. However, the results have not yet been weighted by matching the model state to the available reconstructions. Also, more processes involved in the LGM-to-modern CO 2 change must be included and many combinations of these processes must be tested in order to obtain a robust range for the LGM-to-modern 14 C atm difference. For the weighting procedure, the model states can additionally be compared to LGM reconstructions of δ 13 C (Bickert & Mackensen, 24; Curry & Oppo, 25), of cadmium-to-calcium ratios (Cd/Ca; Marchitto & Broecker, 26), and of sea-surface temperatures (SST; MARGO Project Members, 29). Note that simulated phosphate can be converted to Cd/Ca ratios according to Elderfield & Rickaby (2).

133 5.5. ABRUPT DEGLACIAL AMOC CHANGES Abrupt Deglacial AMOC Changes During the last deglaciation, it is believed that the AMOC strength was substantially reduced during H1 and during the YD (Bond et al., 1992; McManus et al., 24; Maier-Reimer & Mikolajewicz, 1989; Stocker et al., 1992; Rahmstorf, 1994). Previous modeling studies have attributed an increase of 14 C atm of up to 4 to an AMOC shutdown (Stocker & Wright, 1996; Mikolajewicz, 1996; Marchal et al., 21; Delaygue et al., 23; Butzin et al., 25; Singarayer et al., 28). Apart from the study by Singarayer et al. (28), these studies have not explicitly simulated DIC-14 and DIC-12 separately, but only the isotopic ratio 14 R assuming a constant DIC-12 distribution. Here, a sensitivity study is performed where freshwater of varying quantities from.1 to.28 Sv is discharged into the North Atlantic from 5 N to 7 N for 37 years to simulate the AMOC slowdown during H1 followed by an AMOC recovery period and a second freshwater forcing period of 1 years to simulate the YD (Fig. 5.4a). The response of 14 C atm and CO 2,atm to the freshwater perturbations is analyzed. Note that the terrestrial carbon stock remains constant in the simulations. However, recent model simulations by Bozbiyik et al. (211) with a comprehensive carboncycle climate model show that the carbon pool of the terrestrial biosphere is reduced when the AMOC slows down due to freshwater perturbations into the North Atlantic. The simulations are started at 2 kyr BP from LGM radiative forcing conditions as in STDLGM, but modern carbon cycle conditions, i.e. 14 C atm = and CO 2,atm = 278 ppm at the beginning of the simulation. The radiative forcing is kept constant at LGM conditions throughout the 12 kyr of each simulation. Depending on the amount of freshwater discharge, the AMOC weakens by 4 to 1 Sv in both H1 and YD cases, where the 1 Sv decrease corresponds to a full AMOC shutdown (Fig. 5.4c). However, not only the AMOC weakens, but also the Southern Ocean overturning circulation cell strength (SOMOC) which is a measure of how much Antarctic Bottom Water flows into the deep Indopacific and deep Atlantic Oceans (Fig. 5.4d). 14 C atm rises by up to 64 in the H1 case and by up to 48 in the YD case (Fig. 5.4e). Note that the 14 C atm results are by up to 6 lower when calculated from the isotopic ratio tracer 14 R and by up to 7 higher when the terrestrial biosphere is not taken into account. The 14 C atm increase results from the stratification of the ocean, where the upwelling of low 14 C waters is reduced leading to higher surface ocean 14 C, and therefore to a lower ocean uptake of high 14 C atm. The stratification also leads to a reduction of CO 2,atm (Fig. 5.4f). This is in contrast to the observations that show a CO 2,atm rise during H1 and the YD (Fig. 5.1f). It is thought that the observed CO 2,atm was released from the Southern Ocean due to increased upwelling of deep waters (Anderson et al., 29; Denton et al., 21; Lee et al., 211). This conflict between the observations and the freshwater perturbation simulations raises the question whether freshwater perturbations are suitable measures to reduce AMOC in a model, especially when the perturbation must be maintained to keep a low AMOC strength. Moreover, sea-level reconstructions rather show low sea-level rise during H1 and the YD compared to the Bølling/Allerød period and the early Holocene (Siddall et al., 21, and references therein). It is argued that if only the Atlantic Ocean was more stratified during the H1 and the YD but SOMOC remained constant or was even more vigorous, then the 14 C atm excursion was lower than indicated by the model results presented here, or even negative. This is demonstrated in a model simulation where the freshwater input into the North Atlantic is compensated in the Southern Ocean to prevent stratification in the Southern Ocean. In the simulation, the AMOC shuts down, but the SOMOC increases. The 14 C atm increase is substantially smaller than in the corresponding simulation without freshwater compensation, and CO 2,atm varies only by a few ppm (blue lines in Fig. 5.4).

134 5. ATMOSPHERIC RADIOCARBON DURING THE LAST DEGLACIATION.4 a) 1 b) 5 AMOC (Sv) 2 c) d) e) 15 f) C ( ) relative sea level (m).2 SOMOC (Sv) Freshwater (Sv) atm pco2 (ppm) Time (kyr BP) Figure 5.4: Sensitivity study of changes in 14 Catm and CO2,atm following abrupt changes of the AMOC. a) In order to slow down the AMOC, the North Atlantic from 5 N to 7 N is perturbed by a freshwater discharge. Two perturbations follow each other. The duration of the first corresponds to the duration of H1, the second corresponds to the duration of the YD. Gray lines: Simulations where the YD freshwater perturbation is.26 Sv in all simulations. Black lines: Simulations where the H1 freshwater perturbation is.26 Sv in all simulations. Blue line: Simulation where the freshwater input into the North Atlantic is compensated in the Southern Ocean. Thus, there is no sea-level rise in this simulation. b) The relative sealevel rise. c) The maximum strength of the AMOC. d) The maximum strength of the SOMOC. e) The change in 14 Catm. f) The change in CO2,atm.

135 5.6. CONCLUSIONS 133 Either way, the 14 C atm excursions decay away after a few thousand years, so the short term circulation changes do not contribute to the LGM-to-modern 14 C atm difference. 5.6 Conclusions LGM-to-modern changes in 14 C atm have been discussed with respect to changes in the radiocarbon production rate, to the difference between the glacial and the modern state of the ocean and to short term changes of the meridional overturning circulation. The two available reconstructions of the radiocarbon production rate differ substantially. If the production rate reconstruction of Laj et al. (22) can be believed, then the resulting changes in 14 C atm follow the 14 C atm reconstructions back to 2 kyr BP. Short term fluctuations of 14 C atm of several tens of per mil in the 14 C atm reconstructions during this period are likely to be explained by rapid changes of the AMOC and SOMOC and by the low resolution of the production rate reconstruction. Prior to 2 kyr BP, 14 C atm forced by the production rate of Laj et al. (22) follows the 14 C atm reconstruction of Hoffmann et al. (21) but not the Intcal9 reconstruction that is primarily based on Cariaco Basin data of Hughen et al. (26) during this period. It remains to be seen how the Intcal community will include the Hoffmann et al. (21) data set into the next version of the Intcal 14 C atm reconstruction. The radiocarbon production rate reconstruction of Muscheler et al. (24) results in a 14 C atm history that is several hundred per mil lower than the 14 C atm reconstructions and that does not reproduce the 14 C atm decline during the deglaciation from 2 to 1 kyr BP. In the three simulations of potential LGM ocean states where CO 2,atm is reduced to about 19 ppm, 14 C atm is increased to values from 4 to 219. This 14 C atm difference is an additional contribution to the LGM-to-modern 14 C atm difference next to the radiocarbon production rate changes. Although these values result from unrealistic scenarios, they represent a possible range of the 14 C atm change that can be achieved with this model. Still, a full sensitivity study with many more simulations must be performed where more processes involved in the glacial-interglacial carbon-cycle change are included in order to consolidate this range. Also, the results of each simulation should be weighted by comparing the SST and ocean tracer distributions to LGM reconstructions of SST, benthic-planktonic age differences, δ 13 C and Cd/Ca. In these simulations, a sediment diagenesis model should be included to account for carbonate compensation, and the lower glacial terrestrial carbon pool should be taken into account. Including a sediment diagenesis model generally leads to smaller 14 C atm changes (R. Roth, personal communication 211). A potentially important contributor to the LGM-to-modern 14 C atm decline is deglacial changes of permafrost peatlands. When permafrost thaws during the deglaciation, radiocarbon-depleted carbon is released from the peatlands to the atmosphere. Future simulations with a comprehensive land vegetation model are necessary to quantify this contribution to the LGM-to-modern 14 C atm change. Finally, it has been argued that increased volcanism during the second half of the last deglaciation has contributed to the CO 2,atm rise (Huybers & Langmuir, 29). Because the carbon released in this way is radiocarbon-free, it further contributes to the 14 C atm decrease (R. Roth et al., in preparation). Besides the quantification of the contribution of radiocarbon-depleted carbon reservoirs in the ocean and on land to the LGM-to-modern 14 C atm change, more work needs to be done to substantially reduce the uncertainties of the radiocarbon production rate and 14 C atm reconstructions in order to solve the deglacial radiocarbon mystery.

136 134 BIBLIOGRAPHY Bibliography Anderson, R. F., Ali, S., Bradtmiller, L. I., Nielsen, S. H. H., Fleisher, M. Q., Anderson, B. E., & Burckle, L. H., 29. Wind-driven upwelling in the Southern Ocean and the deglacial rise in atmospheric CO 2, Science, 323, Archer, D., Winguth, A., D., L., & Mahowald, N., 2. What caused th glacial/interglacial pco 2 cycles?, Reviews of Geophysics, 38, Barker, S., Knorr, G., Vautravers, M. J., Diz, P., & Skinner, L. C., 21. Extreme deepening of the Atlantic overturning circulation during deglaciation, Nature Geoscience, 3, Bickert, T. & Mackensen, A., 24. Last Glacial to Holocene changes in South Atlantic deep water circulation, in The South Atlantic in the Late Quaternary: reconstruction of material budget and current systems, pp , eds Wefer, G., Mulitza, S., & Ratmeyer, V., Springer-Verlag, Berlin. Bond, G., Heinrich, H., Broecker, W., Labeyrie, L., McManus, J., Andrews, J., Huon, S., Jantschik, R., Clasen, S., Simet, C., Tedesco, K., Klas, M., Bonani, G., & Ivy, S., Evidence for massive discharges of icebergs into the North-Atlantic Ocean during the last glacial period, Nature, 36, Bouttes, N., Paillard, D., & Roche, D. M., 21. Impact of brine-induced stratification on the glacial carbon cycle, Climate of the Past, 6, Bouttes, N., Paillard, D., Roche, D. M., Brovkin, V., & Bopp, L., 211. Last Glacial Maximum CO 2 and δ 13 C successfully reconciled, Geophysical Research Letters, 38, L275. Bozbiyik, A., Steinacher, M., Joos, F., Stocker, T. F., & Menviel, L., 211. Fingerprints of changes in the terrestrial carbon cycle in response to large reorganizations in ocean circulation, Climate of the Past, 7, Broecker, W. & Barker, S., 27. A 19 drop in atmosphere s 14 C during the Mystery Interval (17.5 to 14.5 kyr), Earth and Planetary Science Letters, 256, Broecker, W. & Clark, E., 21. Search for a glacial-age 14 C-depleted ocean reservoir, Geophysical Research Letters, 37, L1366. Broecker, W. S., Peng, T.-H., Trumbore, S., Bonani, G., & Wolfli, W., 199. The distribution of radiocarbon in the glacial ocean, Global Biogeochemical Cycles, 4, Broecker, W. S., Barker, S., Clark, E., Hajdas, I., Bonani, G., & Stott, L., 24. Ventilation of the glacial deep Pacific Ocean, Science, 36, Bryan, S. P., Marchitto, T. M., & Lehman, S. J., 21. The release of 14 C-depleted carbon from the deep ocean during the last deglaciation: Evidence from the Arabian Sea, Earth and Planetary Science Letters, 298, Butzin, M., Prange, M., & Lohmann, G., 25. Radiocarbon simulations for the glacial ocean: The effects of wind stress, Southern Ocean sea ice and Heinrich events, Earth and Planetary Science Letters, 235, Curry, W. B. & Oppo, D. W., 25. Glacial water mass geometry and the distribution of δ 13 C of ΣCO 2 in the western Atlantic Ocean, Paleoceanography, 2, PA117. De Pol-Holz, R., Keigwin, L., Southon, J., Hebbeln, D., & Mohtadi, M., 21. No signature of abyssal carbon in intermediate waters off Chile during deglaciation, Nature Geoscience, 3, Delaygue, G., Stocker, T. F., Joos, F., & Plattner, G.-K., 23. Simulation of atmospheric radiocarbon during abrupt oceanic changes: sensitivities to model parameters, Quaternary Science Reviews, 22, Denton, G. H., Anderson, R. F., Toggweiler, J. R., Edwards, R. L., Schaefer, J. M., & Putnam, A. E., 21. The Last Glacial Termination, Science, 328, Edwards, N. R., Willmott, A. J., & Killworth, P. D., On the role of topography and wind stress on the stability of the thermohaline circulation, Journal of Physical Oceanography, 28, Elderfield, H. & Rickaby, R. E. M., 2. Oceanic Cd/P ratio and nutrient utilization in the glacial Southern Ocean, Nature, 45,

137 BIBLIOGRAPHY 135 Fairbanks, R. G., Mortlock, R. A., Chiu, T.-C., Cao, L., Kaplan, A., Guilderson, T. P., Fairbanks, T. W., Bloom, A. L., Grootes, P. M., & M.-J., N., 25. Radiocarbon calibration curve spanning to 5, years BP based on paired 23 Th/ 234 U/ 238 U and 14 C dates on pristine corals, Quaternary Science Reviews, 24, Fischer, H., Schmitt, J., Lüthi, D., Stocker, T. F., Tschumi, T., Parekh, P., Joos, F., Köhler, P., Völker, C., Gersonde, R., Barbante, C., Le Floch, M., Raynaud, D., & Wolff, E. W., 21. The role of Southern Ocean processes on orbital and millenial CO 2 variations - A synthesis, Quaternary Science Reviews, 29, Godwin, H., Half-life of radiocarbon, Nature, 195, 984. Hemming, S. R., 24. Heinrich events: Massive late Pleistocene detritus layers of the North Atlantic and their global climate imprint, Reviews of Geophysics, 42, RG15. Hoffmann, D. L., Beck, J. W., Richards, D. A., Smart, P. L., Singarayer, J. S., Ketchmark, T., & Hawkesworth, C. J., 21. Towards radiocarbon calibration beyond 28 ka using speleothems from the Bahamas, Earth and Planetary Science Letters, 289, 1 1. Hughen, K., Southon, J., Lehman, S., Bertrand, C., & Turnbull, J., 26. Marine-derived 14 C calibration and activity record for the past 5, years updated from the Cariaco Basin, Quaternary Science Reviews, 25, Huybers, P. & Langmuir, C., 29. Feedback between deglaciation, volcanism, and atmospheric CO 2, Earth and Planetary Science Letters, 286, Karlén, W., Olsson, I. U., Kallberg, P., & Kilicci, S., Absolute determination of the activity of two 14 C dating standards, Arkiv for Geofysik, 4, Köhler, P. & Fischer, H., 24. Simulating changes in the terrestrial biosphere during the last glacial/interglacial transition, Global and Planetary Change, 43, Kwon, E. Y., Primeau, F., & Sarmiento, J. L., 29. The impact of remineralization depth on the air-sea carbon balance, Nature Geoscience, 2, Laj, C., Kissel, C., Mazaud, A., Michel, E., Muscheler, R., & Beer, J., 22. Geomagnetic field intensity, North Atlantic Deep Water circulation and atmospheric 14 C during the last 5 kyr, Earth and Planetary Science Letters, 2, Lee, S.-Y., Chiang, J. C. H., Matsumoto, K., & Tokos, K. S., 211. Southern Ocean wind response to North Atlantic cooling and the rise in atmospheric CO 2: Modeling perspective and paleoceanographic implications, Paleoceanography, 26, PA1214. Lüthi, D., Le Floch, M., Bereiter, B., Blunier, T., Barnola, J. M., Siegenthaler, U., Raynaud, D., Jouzel, J., Fischer, H., Kawamura, K., & Stocker, T. F., 28. High resolution carbon dioxide concentration record 65, 8, years before present, Nature, 453, Lynch-Stieglitz, J., Adkins, J. F., Curry, W. B., Dokken, T., Hall, I. R., Herguera, J. C., Hirschi, J. J.-M., Ivanova, E. V., Kissel, C., Marchal, O., Marchitto, T. M., McCave, I. N., McManus, J. F., Mulitza, S., Ninnemann, U., Peeters, F., Yu, E.-F., & Zahn, R., 27. Atlantic meridional overturning circulation during the Last Glacial Maximum, Science, 316, Magana, A. L., Southon, J. R., Kennett, J. P., Brendan Roark, E., Sarnthein, M., & Stott, L. D., 21. Resolving the cause of large differences between deglacial benthic foraminifera radiocarbon measurements in Santa Barbara Basin, Paleoceanography, 25, PA412. Mahowald, N. M., Muhs, D. R., Levis, S., Rasch, P. J., Yoshioka, M., Zender, C. S., & Luo, C., 26. Change in atmospheric mineral aerosols in response to climate: Last glacial period, preindustrial, modern, and doubled carbon dioxide climates, Journal Of Geophysical Research-Atmospheres, 111, D122. Maier-Reimer, E. & Mikolajewicz, U., Experiments with an OGCM on the cause of the Younger Dryas, Tech. Rep. 39, Max-Planck-Institut für Meteorologie, Hamburg, Germany. Marchal, O., Stocker, T. F., & Muscheler, R., 21. Atmospheric radiocarbon during the Younger Dryas: production, ventilation, or both?, Earth and Planetary Science Letters, 185,

138 136 BIBLIOGRAPHY Marchitto, T. M. & Broecker, W. S., 26. Deep water mass geometry in the glacial Atlantic Ocean: A review of constraints from the paleonutrient proxy Cd/Ca, Geochemistry, Geophysics, Geosystems, 7, Q123. Marchitto, T. M., Lehman, S. J., Ortiz, J. D., Flückiger, J., & van Geen, A., 27. Marine radiocarbon evidence for the mechanism of deglacial atmospheric CO 2 rise, Science, 316, MARGO Project Members, 29. Constraints on the magnitude and patterns of ocean cooling at the Last Glacial Maximum, Nature Geoscience, 2, Masarik, J. & Beer, J., Simulation of particle fluxes and cosmogenic nuclide production in the Earth s atmosphere, Journal of Geophysical Research, 14, 12,99 12,111. McManus, J. F., Francois, R., Gherardi, J.-M., & Keigwin, L. D., 24. Collapse and rapid resumption of Atlantic meridional circulation linked to deglacial climate changes, Nature, 428, Mikolajewicz, U., A meltwater-induced collapse of the conveyor belt thermohaline circulation and its influence on the distribution of 14 C and δ 18 O in the oceans, Tech. Rep. 189, Max-Planck-Inst. für Meteorologie, Hamburg, Germany, Monnin, E., Indermühle, A., Dällenbach, E., Flückiger, J., Stauffer, B., Stocker, T. F., Raynaud, D., & Barnola, J.-M., 21. Atmospheric CO 2 concentrations over the Last Glacial Termination, Science, 291, Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 26. Water mass distribution and ventilation time scales in a cost-efficient, three-dimensional ocean model, Journal of Climate, 19, Müller, S. A., Joos, F., Edwards, N. R., & Stocker, T. F., 28. Modeled natural and excess radiocarbon - sensitivities to the gas exchange formulation and ocean transport strength, Global Biogeochemical Cycles, 22, GB311. Muscheler, R., Beer, J., Wagner, G., Laj, C., Kissel, C., Raisbeck, G. M., Yiou, F., & Kubik, P. W., 24. Changes in the carbon cycle during the last deglaciation as indicated by the comparison of 1 Be and 14 C records, Earth and Planetary Science Letters, 219, Orr, J. C., Najjar, R., Sabine, C. L., & Joos, F., Abiotic-HOWTO, Gif-sur-Yvette, France, Internal OCMIP Report, 25pp., Otto-Bliesner, B. L., Hewitt, C. D., Marchitto, T. M., Brady, E., Abe-Ouchi, A., Crucifix, M., Murakami, S., & Weber, S. L., 27. Last Glacial Maximum ocean thermohaline circulation: PMIP2 model intercomparisons and data constraints, Geophysical Research Letters, 34, L1276. Parekh, P., Joos, F., & Müller, S. A., 28. A modeling assessment of the interplay between aeolian iron fluxes and iron-binding ligands in controlling carbon dioxide fluctuations during Antarctic warm events, Paleoceanography, 23, PA422. Peltier, W. R., 24. Global glacial isostasy and the surface of the ice-age earth: The ICE-5G (VM2) model and GRACE, Annual Review of Earth and Planetary Sciences, 32, Rahmstorf, S., Rapid climate transitions in a coupled ocean-atmosphere model, Nature, 372, Raisbeck, G. M., Yiou, F., Fruneau, M., Loiseaux, J. M., Lieuvin, M., & Ravel, J. C., Cosmogenic 1 Be/ 7 Be as a probe of atmospheric transport processes, Geophysical Research Letters, 8, Rasmussen, S. O., Andersen, K. K., Svensson, A. M., Steffensen, J. P., Vinther, B. M., Clausen, H. B., Siggaard-Andersen, M. L., Johnsen, S. J., Larsen, L. B., Dahl-Jensen, D., Bigler, M., Röthlisberger, R., Fischer, H., Goto-Azuma, K., Hansson, M. E., & Ruth, U., 26. A new Greenland ice core chronology for the last glacial termination, Journal of Geophysical Research, 111, D612. Reimer, P. J., Baillie, M. G. L., Bard, E., Bayliss, A., Beck, J. W., Bertrand, C., Blackwell, P. G., Buck, C. E., Burr, G., Cutler, K. B., Damon, P. E., Edwards, R. L., Fairbanks, R. G., Friedrich, M., Guilderson, T. P., Hughen, K. A., Kromer, B., McCormac, F. G., Manning, S., Bronk Ramsey, C., Reimer, R. W., Remmele, S., Southon, J. R., Stuiver, M., Talamo, S., Taylor, F. W., van der Plicht, J., & Weyhenmeyer, C. E., 24. IntCal4 terrestrial radiocarbon age calibration, -26 cal kyr BP, Radiocarbon, 46,

139 BIBLIOGRAPHY 137 Reimer, P. J., Baillie, M. G. L., Bard, E., Bayliss, A., Beck, J. W., Blackwell, P. G., Ramsey, C. B., Buck, C. E., Burr, G. S., Edwards, R. L., Friedrich, M., Grootes, P. M., Guilderson, T. P., Hajdas, I., Heaton, T. J., Hogg, A. G., Hughen, K. A., Kaiser, K. F., Kromer, B., McCormac, F. G., Manning, S. W., Reimer, R. W., Richards, D. A., Southon, J. R., Talamo, S., Turney, C. S. M., van der Plicht, J., & Weyhenmeyer, C. E., 29. INTCAL9 AND MARINE9 radiocarbon age calibration curves, -5, years Cal BP, Radiocarbon, 51, Ritz, S. P., Stocker, T. F., & Joos, F., 211. A coupled dynamical ocean - energy balance atmosphere model for paleoclimate studies, Journal of Climate, 24, Robinson, L. F., Adkins, J. F., Keigwin, L. D., Southon, J., Fernandez, D. P., Wang, S. L., & Scheirer, D. S., 25. Radiocarbon variability in the western North Atlantic during the last deglaciation, Science, 31(5753), Rose, K. A., Sikes, E. L., Guilderson, T. P., Shane, P., Hill, T. M., Zahn, R., & Spero, H. J., 21. Upper-oceanto-atmosphere radiocarbon offsets imply fast deglacial carbon dioxide release, Nature, 466, Siddall, M., Abe-Ouchi, A., Andersen, M., Antonioli, F., Bamber, J., Bard, E., Clark, J., Clark, P., Deschamps, P., Dutton, A., Elliot, M., Gallup, C., Gomez, N., Gregory, J., Huybers, P., Kawarnura, K., Kelly, M., Lambeck, K., Lowell, T., Mitrovica, J., Otto-Bliesner, B., Richards, D., Stanford, J., Stirling, C., Stocker, T., Thomas, A., Thompson, B., Tornqvist, T., Riveiros, N. V., Waelbroeck, C., Yokoyama, Y., & Yu, S. Y., 21. The sea-level conundrum: case studies from palaeo-archives, Journal of Quaternary Science, 25, Siegenthaler, U. & Oeschger, H., Biospheric CO 2 emissions during the past 2 years reconstructed by deconvolution of ice core data, Tellus, 39, Siegenthaler, U., Stocker, T. F., Monnin, E., Lüthi, D., Schwander, J., Stauffer, B., Raynaud, D., Barnola, J. M., Fischer, H., Masson-Delmotte, V., & Jouzel, J., 25. Stable Carbon Cycle Climate Relationship During the Late Pleistocene, Science, 31, Sigman, D. M. & Boyle, E. A., 2. Glacial/interglacial variations in atmospheric carbon dioxide, Nature, 47, Sigman, D. M., Hain, M. P., & Haug, G. H., 21. The polar ocean and glacial cycles in atmospheric CO 2 concentration, Nature, 466, Singarayer, J. S., Richards, D. A., Ridgwell, A., Valdes, P. J., Austin, W. E. N., & Beck, J. W., 28. An oceanic origin for the increase of atmospheric radiocarbon during the Younger Dryas, Geophysical Research Letters, 35, L1477. Skinner, L. C., Fallon, S., Waelbroeck, C., Michel, E., & Barker, S., 21. Ventilation of the Deep Southern Ocean and Deglacial CO 2 Rise, Science, 328, Stocker, T. F. & Wright, D. G., Rapid changes in ocean circulation and atmospheric radiocarbon, Paleoceanography, 11, Stocker, T. F., Wright, D. G., & Mysak, L. A., A zonally averaged, coupled ocean-atmosphere model for paleoclimate studies, Journal of Climate, 5, Stott, L., Southon, J., Timmermann, A., & Koutavas, A., 29. Radiocarbon age anomaly at intermediate water depth in the Pacific Ocean during the last deglaciation, Paleoceanography, 24, PA2223. Thornalley, D. J. R., Barker, S., Broecker, W. S., Elderfield, H., & McCave, I. N., 211. The deglacial evolution of North Atlantic deep convection, Science, 331, Toggweiler, J. R., Russell, J. L., & Carson, S. R., 26. Midlatitude westerlies, atmospheric CO 2, and climate change during the ice ages, Paleoceanography, 21, PA25. Tschumi, T., Joos, F., & Parekh, P., 28. How important are Southern Hemisphere wind changes for low glacial carbon dioxide? A model study, Paleoceanography, 23, PA428.

140

141 Chapter 6 Modeling the Effect of Abrupt Ocean Circulation Change on Marine Reservoir Age Stefan P. Ritz, Thomas F. Stocker, and Simon A. Müller Published in Earth and Planetary Science Letters, Volume 268, pp , 28.

142 14 6. OCEAN CIRCULATION CHANGES AND MARINE RESERVOIR AGE Available online at Earth and Planetary Science Letters 268 (28) Modeling the effect of abrupt ocean circulation change on marine reservoir age Stefan P. Ritz a,b,, Thomas F. Stocker a,b, Simon A. Müller a,1 a Climate and Environmental Physics, Physics Institute, University of Bern, Bern, Switzerland b Oeschger Centre for Climate Change Research, University of Bern, Bern, Switzerland Received 5 July 27; received in revised form 15 January 28; accepted 15 January 28 Available online 4 February 28 Editor: M.L. Delaney Abstract Radiocarbon surface reservoir age is required for precise dating of marine organisms. Although often assumed constant, changes in atmospheric radiocarbon content, ocean circulation, and ocean mixing imprint changes on this quantity. The spatial and temporal variations of marine surface and bottom reservoir ages in response to a shutdown and subsequent recovery of Atlantic meridional overturning circulation (MOC) are analyzed using a cost-efficient, three-dimensional ocean circulation model. Generally, surface reservoir age changes are limited to the Atlantic Ocean with a reduction of about 1 yr after the MOC shutdown, followed by a slow increase and a peak at the time of MOC resumption. Parameter sensitivity studies with respect to the roles of gas exchange, diffusivity and North Atlantic ice cover show that ice cover has the largest effect on the transient evolution of surface reservoir age during the shutdown. Our model results agree well with a recent reconstruction of surface reservoir age changes during the Younger Dryas when we reduce the rate of gas exchange in the model and include a parametrization of seasonal ice cover in the North Atlantic. 28 Elsevier B.V. All rights reserved. Keywords: reservoir age; Younger Dryas; radiocarbon; MOC; overturning; circulation 1. Introduction Knowing the marine surface reservoir age, defined as the difference in the 14 C age between the atmosphere and the surface ocean, is essential for dating purposes in marine archives such as sediment cores and corals. Hereafter, it will be referred to as reservoir age. For instance, terrestrial organic matter is dated using 14 C. Due to changes in the carbon cycle and in the production rate of 14 C over time, a 14 C calibration curve is used to convert 14 C age to calendar age (Reimer et al., 24; Fairbanks et al., 25). In order to date surface ocean biogenic Corresponding author. Climate and Environmental Physics, Physics Institute, University of Bern, Bern, Switzerland. address: ritz@climate.unibe.ch (S.P. Ritz). 1 Now at Earth and Environmental Sciences, Open University, Milton Keynes, UK. matter, such as planktonic foraminifera found in sediment cores, an equivalent calibration curve for the surface ocean is needed. Since this is not available, the calibration curve for the atmosphere is used and an estimated reservoir age is added (Stuiver and Braziunas, 1993), often assumed to be constant and similar to modern, bomb-corrected reservoir ages. The purpose of this paper is to provide information on spatial and temporal differences of reservoir age during abrupt climate change by using a cost-efficient, zonally resolved ocean model (Müller et al., 26). We investigate the fingerprint of a shutdown of the Atlantic meridional overturning circulation (MOC) on surface and bottom reservoir ages and examine the reservoir age sensitivity to various model parameters. In a recent study, Bondevik et al. (26) presented a continuous record of reservoir age changes before, during and after the Younger Dryas (YD) stadial (approximately 12.8 to 11.6 ka before present (BP, before 195); Gulliksen et al., 1998; X/$ - see front matter 28 Elsevier B.V. All rights reserved. doi:1.116/j.epsl

143 141 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) Hughen et al., 2; Friedrich et al., 24; Southon, 24), comparing 14 C ages of terrestrial plant fragments and marine shells found in the same sediment cores on Norway's west coast. They found an increase in reservoir age from 4 to 6 yr in the early YD and a drop by 3 yr immediately following the YD Holocene transition. Earlier modeling studies have analyzed the response of reservoir age to sea ice cover, wind speed and Atlantic MOC changes using simple box models and zonally averaged Earth system models (Bard et al., 1994; Stocker and Wright, 1996, 1998; Delaygue et al., 23). They found changes of more than 4 yr in the North Atlantic attributed to changes in convection and sea ice coverage. When the North Atlantic is ice covered for ten months out of the year, gas exchange flux is significantly reduced, thus contributing to an increase in reservoir age of 2 to 3 yr (Bard et al., 1994). In a modeling study using a three-dimensional ocean model, Butzin et al. (25) found bottom reservoir age variations of up to 1 yr in the North Atlantic when the MOC shuts down. For the surface reservoir age, they did not find the increase reported by Bard et al. (1994). Here, we address changes in the surface and bottom reservoir age distribution to YD type events and compare the results to the reservoir age reconstructions of Bondevik et al. (26). 2. Model setup and simulations For this study we use the Bern3D ocean model, a costefficient, seasonally forced, three-dimensional frictional geostrophic ocean model (Müller et al., 26). It consists of grid boxes in the horizontal direction and 32 layers in depth. The model is run under mixed boundary conditions for temperature and salinity. The temperature fields are taken from Levitus and Boyer (1994), and the salinity fields from Levitus et al. (1994). The atmosphere is described by one wellmixed box. Air sea gas transfer velocity and a climatology of fractional sea ice cover for the modern climate, based on the OCMIP-2 protocol (Walsh, 1978; Zwally et al., 1983; Orr, 1999), modulate the atmosphere-to-ocean 14 C flux. In the ocean model, 14 C is transported as 14 R= 14 C/ 12 C. The 12 C concentration for the atmosphere (pco 2 ) and the ocean (total inorganic carbon) are held constant at the pre-industrial values of 278 ppm and 2 mol m 3, respectively. The 14 C production rate in the atmosphere is held constant as well. Thus, we do not account for changes in solar activity and geomagnetic field intensity. A 26,-yr spinup, which includes diagnostics of the salinity fields for the mixed boundary conditions and 14 C production rate, preceded all simulations of this study. The spinup is required in order to obtain a steady state solution for the 14 C distribution. In order to diagnose the 14 C production rate, atmospheric Δ 14 C is held constant. At steady state, the production rate in the model is then equal to the decay of the 14 C content in the atmosphere and the ocean. First, a steady state control run (CTRL) has been performed. It will be addressed throughout the paper when comparisons to the steady state are made. Next we present a simulation, STDR, in which freshwater was injected into the Greenland Iceland Norwegian (GIN) sea (Tarasov and Peltier, 25) from 3 to W and 55 to 7 N for 2 yr with a triangular evolution in time (Fig. 2c). The freshwater pulse shuts down the Atlantic MOC resulting in a reversed overturning circulation in the model (Fig. 1g). Further simulations with larger pulses have shown that besides the effect of the MOC shutdown, the perturbation has an effect on the Antarctic Bottom Water cell and hence on the 14 C distribution, which leads to larger longterm changes in reservoir age. Here we are primarily interested in the effect of the MOC shutdown. In contrast to most complex ocean models, our model has multiple equilibrium states for the current parameter settings. This requires a negative freshwater perturbation in order for the MOC to recover (Fig. 2c). The MOC recovery was triggered to occur 12 yr after the shutdown; this is comparable to the duration of the YD. Similar to the freshwater pulse which causes the MOC shutdown, this perturbation may have direct model-specific responses on 14 C. But in contrast to many simpler ocean models, the MOC does not overshoot when recovered from the shutdown state by a negative freshwater perturbation (e.g., Stocker and Wright, 1996). For parameter sensitivity studies, five additional simulations have been performed with an MOC scenario similar to STDR. First, in a run referred to as KGAS, the air sea gas transfer velocity used in OCMIP-2 (Orr, 1999) was reduced by 19%, as suggested by Müller et al. (in press). In a second run, which will be referred to as PCO2, atmospheric pco 2 was reduced to 238 ppm. This corresponds to the value at the onset of the YD (Monnin et al., 21). In a manner similar to the gas transfer velocity, pco 2 affects the 14 C exchange flux between the atmosphere and the ocean. In the third run the diapycnal diffusivity was increased from 1 5 m 2 s 1 to m 2 s 1 (DIFF). This increases the maximum annual-mean MOC value from 14. Sv (Sverdrup; 1 Sv =1 6 m 3 s 1 ) in STDR to 16.5 Sv, and thus allows us to investigate whether an MOC shutdown from this higher state results in a larger reservoir age change. Since a higher diapycnal diffusivity strengthens top-to-bottom mixing, we expect to see higher surface reservoir ages. Because diapycnal diffusivity is an important model tuning parameter, it is important to understand its effects on reservoir age. Note that the parameters discussed so far are held constant throughout the run and do not vary when the MOC changes. In order to start the simulations from steady state, a separate spinup preceded each of these runs. In two runs the effect of sea ice on reservoir age is considered. Since the presence of sea ice hinders equilibration of atmospheric and surface ocean 14 C concentrations, higher reservoir ages are expected in regions covered by sea ice. In these runs we additionally prescribed a fractional sea ice cover in the Northern Hemisphere which we will refer to as equivalent sea ice cover. Because the model does not include a dynamic sea ice component, we simulated its effect on 14 Cby artificially setting the atmosphere ocean 14 C transfer to zero in the Arctic Ocean and in the Atlantic north of 63 N in run ICE1 (which matches estimates of the sea ice extent for the YD by Koç et al., 1993), and north of 56 N in run ICE2, respectively. This equivalent sea ice cover does not affect the exchange of heat and freshwater. Equivalent sea ice is activated, i.e. 14 C

144 OCEAN CIRCULATION CHANGES AND MARINE RESERVOIR AGE 24 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) Fig. 1. a, Marine surface reservoir age distribution of CTRL. Large reservoir ages occur in regions of upwelling of old, 14 C-depleted deep waters, whereas regions of deep water formation and subduction show young ages. b d, STDR marine surface reservoir age anomalies. Marked in black outlines are the regions of largest changes as well as the region of the Cariaco basin and Barbados; b, Short-term variations due to the Atlantic meridional overturning circulation (MOC) shutdown (1 yr after the start of the perturbation (hereafter referred to as ASP) compared to CTRL); c, Long-term variations due to the MOC shutdown (year 11 ASP compared to CTRL); d, Short-term variations due to the recovery of the MOC (change in reservoir age between year 11 ASP and year 12 ASP). e h, Southern Ocean (SO) and Atlantic MOC (in Sv, 1 Sv=1 6 m 3 s 1 ) for CTRL, year 1 ASP, year 11 ASP and year 12 ASP; g, the MOC is reversed during the off state.

145 143 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) exchange is suppressed, when the maximum Atlantic MOC sinks below 7 Sv. The suppression is removed when MOC strength rises above this threshold. The equivalent sea ice cover remains seasonally constant. Note that this equivalent sea ice is in addition to the modern OCMIP-2 sea ice coverage and does not replace it. The timing of the freshwater perturbations for all runs mentioned above is the same as in STDR and summarized in Table 1. Table 1 Summary of the model simulations Simulations Notes max F pos max F neg CTRL 5 yr steady state (control) run STDR Standard run.3 Sv.42 Sv KGAS Air sea gas transfer velocity reduced by 19%.3 Sv.42 Sv PCO2 Atmospheric pco 2 reduced from 278 ppm.3 Sv.42 Sv to 238 ppm DIFF Diapycnal diffusivity increased from.3 Sv.56 Sv 1 5 m 2 s 1 to m 2 s 1 ICE1 Equivalent sea ice cover (north of 63 N).3 Sv.42 Sv added during the shutdown state ICE2 Equivalent sea ice cover (north of 56 N).3 Sv.42 Sv added during the shutdown state BEST Air sea gas exchange as in KGAS, pco 2 as in PCO2, equivalent sea ice cover as in ICE1.3 Sv.42 Sv Except for CTRL, all simulations have a duration of 24 yr with a triangular freshwater perturbation starting 1 yr after the start of the run and lasting for 2 yr (Fig. 2c). The negative freshwater perturbation starts at year 11 after the beginning of the positive freshwater perturbation and lasts for 2 yr. F max pos and max stand for the maximum positive and negative freshwater input fluxes of the F pos perturbations. Marine surface reservoir age (τ s ) is defined as s s ¼ 1 14 k ln R atm 14 ; ð1þ R oc ðz ¼ Þ marine bottom reservoir age as s b ¼ 1 14 k ln R atm 14 ; ð2þ R oc ðz ¼ HÞ where λ=1/8267 yr 1 is the decay constant for 14 C(Godwin, 1962), and 14 R atm and 14 R oc (z) are the 14 Cto 12 C ratios for the atmosphere, surface ocean (z=) and bottom ocean (z= H), respectively. H represents the depth of the water column and varies spatially. As the surface ocean 14 C concentration is determined by measuring 14 C of planktonic foraminifera, 14 R oc is actually the mean 14 C concentration of the euphotic zone, whose depth varies spatially from a few meters to around 2 m in the open ocean. In the model, 14 R oc (z=) is derived by taking the mean concentration of the top two layers which correspond to a depth of approximately 8 m. 14 C concentrations are expressed in as D 14 C ¼ 14 R x; R st ð3þ Fig. 2. Marine surface reservoir age time series of STDR and its behavior to a reduced air sea gas transfer velocity by 19% (KGAS), reduced atmospheric pco 2 to the pre-yd value of 238 ppm (PCO2), increased diapycnal diffusivity from 1 5 m 2 s 1 to m 2 s 1 (DIFF), and to increased sea ice cover when in circulation off state (ICE1 and ICE2, seasonally constant equivalent ice cover north of 63 N and north of 56 N, respectively) (Table 1) for the North Atlantic site E (a), the North Atlantic site W (b) and the African west coast site N (c). The timing and amount of the freshwater perturbations are indicated in panel c. Note that DIFF needs a larger maximum negative freshwater perturbation of.56 Sv for the MOC to recover (see Table 1). where 14 R st = is the pre-industrial atmospheric 14 C to 12 C ratio (Karlén et al., 1964). In CTRL we set the atmospheric Δ 14 Cto. 3. Reservoir age response to freshwater forcing 3.1. Steady state and general temporal evolution of reservoir age Simulation CTRL displays a wide range of surface reservoir ages from about 25 yr in the Atlantic from 1 to 5 N to over

146 OCEAN CIRCULATION CHANGES AND MARINE RESERVOIR AGE 26 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) yr in the Indian and Southern Oceans and in the eastern equatorial Pacific (Fig. 1a). As the atmospheric 14 C concentration is spatially constant, the variations reflect spatial differences in ocean concentrations. High reservoir ages correlate with upwelling regions where old, 14 C-depleted waters reach the surface. In the Arctic Ocean, the high reservoir ages are due to a combination of sea ice cover and a low horizontal mixing rate. In order to detect regions with high reservoir age changes during the YD, reservoir ages of CTRL and of STDR were compared to one another shortly after the MOC shutdown (1 yr after the start of the perturbation, hereafter referred to as ASP), as well as 11 yr ASP, shortly before MOC resumption (short-term variations, Fig. 1b and long-term variations, Fig. 1c, respectively). In a second step the short-term variations due to the MOC recovery in STDR were analyzed (year 12 ASP compared to year 11 ASP, Fig. 1d). For the short-term variations due to the shutdown, reservoir ages decrease in the entire Atlantic by about 1 yr (Fig. 1b). The Southern, Indian and Pacific Oceans remain mostly unaffected. The Northern Hemisphere Atlantic reservoir age increases to values greater than those in CTRL for the long-term variation (Fig. 1c). The largest changes are found north of 5 N in the western Atlantic and at the North African west coast from 1 to 3 N. In the Southern Hemisphere the African west coast from 5 to 2 S shows the highest changes of around 1 yr compared to CTRL. The Southern and Indian Ocean reservoir ages decrease during this step by approximately 5 yr. The short-term reservoir age changes due to the recovery of the Atlantic MOC mainly consist of a large increase in the North Atlantic of up to 4 yr (Fig. 1d) Regions with particularly large reservoir age variations The evolution of the reservoir ages is investigated further in regions with particularly large age variations. These are the northwest and northeast Atlantic, as well as the northern and southern west coast of Africa (regions marked in Fig. 1). The three Northern Hemisphere Atlantic sites show similar behavior (Fig. 2): A drop in reservoir age directly after the shutdown of the MOC is followed by an increase to higher values than in CTRL. The ocean has not reached steady state yet at the time of the negative freshwater perturbation. This perturbation results in a very fast and large reservoir age increase followed by a decrease to CTRL reservoir ages. The reservoir age of the southern site of the African west coast (Fig. 3) drops rapidly by 15 yr and increases afterwards slowly by 3 to 4 yr. The negative freshwater perturbation then brings the reservoir ages back to the values of CTRL. In the following, the reasons for the modeled reservoir age changes are discussed in more detail for these regions on the basis of STDR. For North Atlantic site W, after the Atlantic MOC shutdown, surface waters remain at the surface of the ocean and have more time to take up 14 C from the atmosphere. This increases the concentration in the surface ocean and consequently atmospheric Δ 14 C. The rise of both atmospheric and surface ocean 14 C concentrations, and the fact that the ocean increase leads the atmospheric increase (Fig. 4), results in a brief decrease of the reservoir age in the entire Atlantic, especially along the path of the southward flowing waters. This can be derived from Eq. (1). In the model, the Atlantic MOC is reversed when in the off state (Fig. 1g), creating a region of upwelling in the northwestern Atlantic. As a result, reservoir age in this region increases. The longer the MOC remains in the off state, waters that are more depleted in 14 C reach the surface. Hence, reservoir age increases until the negative freshwater perturbation is added and the MOC recovers. At this time, convection recommences, resulting in the transport of a large amount of deep water to the surface very quickly. This results in an abrupt reservoir age increase of about 3 yr. After the MOC shutdown, reservoir age decreases at North Atlantic site E for the same reasons as at site W. Here too, the reservoir age increases during the second stage. In contrast to Fig. 3. Evolution of the marine surface reservoir age for the African west coast site S and the Cariaco basin/barbados in STDR and KGAS. The reduced air sea gas transfer velocity in KGAS leads to a better match between the control reservoir ages and modern reservoir age measurements for the Cariaco basin (indicated by the black vertical bar; Hughen et al., 24), and averaged Holocene measurements for Barbados (gray vertical bar; Fairbanks et al., 25), respectively.

147 145 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) Fig. 4. Atmospheric and surface ocean Δ 14 C (upper panel) and marine surface reservoir age (lower panel) for North Atlantic site W of STDR. Large changes in reservoir age occur when atmospheric and surface ocean Δ 14 C behave differently. For instance, the reservoir age drop at the onset of the YD can be assigned to the fact that the ocean Δ 14 C increase leads the atmospheric increase. the northwestern Atlantic, this is due to the advective surface flux and mixing processes, which transport old waters from west to east. The reservoir age increase is not as steep as at site W, probably explained by the mixing processes of the eastward flowing waters. At site N off the African west coast, upwelling becomes dominant during the off state. After a small drop in reservoir age, old waters start reaching the surface and continuously increase reservoir age. The slow reservoir age increase continues until the MOC recovers. The subsequent abrupt reservoir age change is influenced by two effects: The southward flow of low Δ 14 C surface waters, which increases reservoir age and the weakening of the upwelling in this region, which leads to a reservoir age decrease. During the MOC on state (i.e., in CTRL) strong upwelling in the Atlantic prevails only at the African west coast site S. Therefore, the CTRL reservoir age is remarkably higher than elsewhere in the Atlantic Ocean (Figs. 1a and 3). Upwelling is interrupted during the off state. The reservoir age adapts to the adjacent values, which are much lower. After the abrupt reservoir age drop, it increases slightly as the ocean approaches a new steady state. The upwelling recovers when the MOC switches on again, and the reservoir age rises to its CTRL value Reservoir age response in the Caribbean 14 C calibrations such as those of Reimer et al. (24) and Fairbanks et al. (25) are based on 14 C-measurements in tree rings extending back to approximately 12 ka BP. For the YD period and beyond, the Reimer et al. (24) calibration is based on 14 C data from marine varves from the Venezuelan Cariaco basin (Hughen et al., 2) (two sediment cores at 1.7 N 65. W) and on corals from various locations. The reconstruction of Fairbanks et al. (25) is based on corals from an offshore reef core collection from Barbados (at 13.1 N W). In order to link these marine data to the atmospheric values, a constant reservoir age of 42 yr is used for the Cariaco basin, while a constant reservoir age of 365 yr is used for the Barbados corals. Other work by Kromer et al. (24), based on a floating tree-ring chronology, report higher reservoir ages for the Cariaco basin during the Allerød and a reservoir age drop of approximately 2 yr at the onset of the YD. A larger drop is reported by Muscheler et al. (in press). In our model the reservoir age for the Cariaco and the Barbados regions (Fig. 3, region marked in Fig. 1) drops by approximately 1 yr at the onset of the YD followed by a slower increase to values larger than in CTRL. The abrupt recovery shows only a short rise of about 5 yr before decreasing again to its CTRL value Marine bottom reservoir age Bottom reservoir age is of importance for dating benthic foraminifera from sediment cores. Analogous to the surface reservoir age, the short- and long-term bottom reservoir age variations due to the MOC shutdown and the short-term variations to the MOC recovery are discussed here (Fig. 5). The short-term bottom reservoir age changes to the MOC shutdown are less than 1 yr in most parts of the global ocean (Fig. 5a). A bottom reservoir age increase of more than 2 yr occurs only south of Greenland, where convection ceases abruptly and the ventilation in the water column is cut off. The largest increase of 78 yr is also found in this region. The long-term anomalies (comparison of 11 yr ASP to CTRL, Fig. 5b) show a distribution similar to the short-term anomalies, but changes are larger. Again, in the entire Pacific, Indian and Southern Oceans the marine bottom reservoir age remains constant. A small reservoir age decrease of 1 to 2 yr is found in the low latitude Atlantic Ocean. The largest increases of 1 to 15 yr occur in the Caribbean Sea and the

148 OCEAN CIRCULATION CHANGES AND MARINE RESERVOIR AGE 28 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) C ages as a function of the Atlantic MOC strength. In agreement with our results, they found largest changes in the Caribbean Sea of approximately 1 yr after a shutdown of the MOC. Similar results were reported by Butzin et al. (25). 4. Parameter sensitivities Fig. 5. STDR marine bottom reservoir age anomalies. a, Short-term variations due to the Atlantic MOC shutdown (year 1 after the start of the perturbation (ASP) compared to CTRL). b, Long-term variations due to the MOC shutdown (year 11 ASP compared to CTRL). c, Short-term variations due to the recovery of the MOC (change in reservoir age between year 11 ASP and year 12 ASP). Note that the color bars of the panels are not identical. Gulf of Mexico as well as South of Greenland, where the model shows bottom reservoir ages up to 17 yr older than in CTRL. This reservoir age increase throughout the YD can be explained by the 14 C decay of these waters, which are not advected away in this state of the overturning circulation (Fig. 1g). The MOC recovery causes an abrupt bottom reservoir age decrease of 7 to 1 yr to the southeast of Greenland where convection reappears, which restarts the ventilation process in the water column (Fig. 5c). The bottom reservoir age increase southwest of Greenland is primarily a signal from the surface ocean, since the average depth of this region is only 32 m (see also Fig. 1d). Because the model simulations start and end in the modern state, the long-term reservoir age anomalies due to the MOC recovery are the reverse of those produced when the MOC shuts down. In a previous modeling study using the UVic Earth System Climate Model, Meissner et al. (23) analyzed top-to-bottom In addition to STDR, we have run the model with various alternative parameter settings as described in the model section (Fig. 2). In KGAS, the simulation with the reduced air sea gas transfer velocity, we find an 8 to 1 yr shift to higher reservoir ages in all three regions. This change is expected, because a slower air sea gas exchange increases the 14 C concentration difference between the atmosphere and the ocean, which by definition corresponds to a higher reservoir age. When running the model with the pre-yd atmospheric pco 2 value of 238 ppm instead of 278 ppm (PCO2), reservoir age results in values 5 to 8 yr higher than in STDR in all three regions. Since the air sea gas transfer velocity and pco 2 both affect the air sea gas exchange, their effect on reservoir age is very similar. According to Monnin et al. (21) pco 2 increases approximately linearly to 265 ppm during the YD. Accounting for this pco 2 change over time would result in a linear reduction of the reservoir age offset between PCO2 and STDR (not shown here). For DIFF, control state reservoir ages are approximately 5 yr higher in the Indian Ocean and 5 to 1 yr higher in the Pacific compared to STDR. The largest anomalies of approximately 15 yr are found north of Australia. Reservoir ages are 1 to 2 yr lower off the coast of Antarctica. For the longterm variations due to the shutdown of the MOC, the reservoir age increase is about 5 yr larger for the Atlantic compared to STDR (Fig. 2a and c). Anomalies in the other regions are smaller. The reservoir age shift to higher values in most regions shows that higher diapycnal diffusivity enhances the vertical exchange of deep 14 C-depleted waters with the surface ocean. The large drop in reservoir age at the onset of the YD found in both North Atlantic sites has not changed significantly compared to STDR. Thus, the larger MOC drop in DIFF is not reflected in higher reservoir age anomalies. Finally, the two simulations with additional YD sea ice cover in the Northern Hemisphere are examined. Fig. 2 shows that equivalent ice cover has a substantial effect on reservoir age. As soon as the ice cover appears, the reservoir age rises rapidly in ICE2 (seasonally constant equivalent ice cover north of 56 N), especially in the two North Atlantic regions. The reason for this increase is that during the circulation off state, the Arctic waters are not advected away and 14 C is prevented from exchanging with the atmosphere. When the circulation recovers, the reservoir age at North Atlantic site E is so high that it is not increased further by the convection induced ventilation of old waters from the deep ocean to the surface. In ICE1, where the equivalent ice cover only appears north of 63 N, the reservoir age rise is less rapid. Interestingly, adding equivalent sea ice cover to CTRL increases reservoir age in the North Atlantic sites only by approximately 5 yr. These findings are in agreement with the previous modeling work of Schmittner (23), who finds increased bottom 14 C

149 147 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) ages in the global ocean for glacial conditions due to a decrease in air sea exchange of 14 C, which is caused by increased Southern Ocean sea ice cover. 5. Comparison of the model results with reconstructions Bard et al. (1994), Austin et al. (1995), Haflidason et al. (1995) and Bondevik et al. (21) have dated the reservoir age at the time of a volcanic ash layer in various sediment cores of the North Atlantic. This ash layer is named the Vedde Ash Bed and has a calendar age of 12.5 to ka BP (Bondevik et al., 26) which places it in the middle of the YD event. In four sediment cores, Bard et al. (1994) measured a reservoir age of 7 to 8 yr (Fig. 6). This is strongly supported by the results of Austin et al. (1995) and Haflidason et al. (1995), who measured a reservoir age of around 7 yr northwest of Scotland and 8 yr southwest of Norway, respectively. Bondevik et al. (21) obtained a reservoir age of around 6 yr for the west coast of Norway. Here we compare modeled reservoir age changes for the entire YD period to the reservoir age record of Bondevik et al. (26) from sediment cores from two basins, which are now bogs, from the Norwegian west coast (at Kulturmyra, 62 2 N 5 39 E and Kvaltjern, 6 25 N 4 59 E) as well as to the results for the Vedde Ash Bed. For the model-data comparison we start with STDR (Fig. 6). It should be noted that the model run starts and ends in a modern state, while the reconstructions have start and end states that are both different from the modern state. Again, we do not take into account the long-term decrease of atmospheric Δ 14 C due to possible reductions in 14 C production Fig. 6. STDR and BEST reservoir age time series (preceded by a control run) for North Atlantic site E compared to reconstructions by Bondevik et al. (26) for the YD at the Norwegian west coast (black dots) and by Bard et al. (1994) for the North Atlantic from 4 to 7 N (gray bars, both for present-day and 12 ka BP). In BEST, the reduction of the standard air sea gas transfer velocity, as used in OCMIP-2 (Orr, 1999), by 19% (Müller et al., in press) and the usage of pre-yd pco 2 increases reservoir age by approximately 15 yr. The additional sea ice cover during the circulation off state results in a faster increase of reservoir age after the initial drop. Model year was pinned to the data such that a good match between model and data was obtained for the YD period. (Marchal et al., 21). The modeled results may therefore deviate qualitatively and quantitatively from the reconstructions (as for example at the end of the YD). From 13.5 to 13.2 ka BP in the Allerød, the reconstructions show reservoir ages of 5 to 6 yr. At the end of the Allerød, the reservoir age decreases by approximately 1 yr. It must be pointed out, though, that only one data point supports this reservoir age decrease. Because the onset of the YD (dotted line in Fig. 6) was dated based on a lithology of various sediment cores (Bondevik et al., 26), it is not clear whether this reservoir age change is actually due to the MOC slowdown. Model year was pinned to the data by Bondevik et al. (26) such that a good match between model and data was obtained for the YD period. STDR starts in a lower, modern state reservoir age and decreases at the onset of the YD due to the MOC shutdown. The rate of reservoir age increase from 12.9 to 12. ka BP is well represented by the model. The abrupt MOC recovery in the model simulation leads to a rapid increase in reservoir age to about 6 yr, which is lower than the reconstructed values. The rapidity and magnitude of reconstructed and modeled reservoir age decrease after the MOC recovery are also very similar. For present-day (pre-bomb) reservoir ages, Bard et al. (1994) find values of 3 to 4 yr for low latitudes and 4 to 5 yr for the North Atlantic from 4 to 7 N. STDR underestimates these values by 1 to 15 yr (Figs. 1a and 6). However, we found that in KGAS, reservoir age is shifted to values about 8 to 1 yr higher (see Section 4) and thus fits much better to the reconstructions for the present-day ocean by Bard et al. (1994). The modeled reservoir age distribution of KGAS also matches fairly well to reservoir age data calculated from the natural Δ 14 C of the GLODAP data set (Key et al., 24). The modeled values are typically 5 to 1 yr lower than the results from GLODAP, though. The match deteriorates in the western Atlantic and in the North Pacific, where the modeled reservoir ages are 2 and 4 yr lower, respectively. For model-data comparison for the YD period, atmospheric pco 2 of the model should be set as in PCO2. This is not the case in STDR. And although the YD sea ice extent has not been well quantified as yet by paleoclimatic reconstructions, we find it reasonable to add an additional ice cover such as in ICE1. The comparison of the reservoir age reconstructions by Bondevik et al. (26) with BEST, a simulation which combines KGAS, PCO2 and ICE1 (Table 1, Fig. 6), shows much better agreement than STDR. Note that these results are preliminary because of the highly simplified annual-mean representation of sea ice in the model. Because the reduced air sea gas transfer velocity in KGAS and reduced pco 2 in PCO2 merely result in a reservoir age shift, the reservoir age anomalies in Fig. 1 remain valid for KGAS and PCO2. For the Southern Ocean south of 6 S, Bard et al. (1994) found present-day reservoir ages from 5 to 12 yr. Simulated reservoir ages in KGAS for this region range from 6 to 12 yr. The present-day reservoir ages predicted by our model agree well with the model results by Butzin et al. (25), who use the Hamburg LSG ocean circulation model. However, for a change in MOC, the surface reservoir age results reported here show considerably better agreement to the observed reservoir age changes in the North Atlantic than the results of Butzin et al. (25).

150 OCEAN CIRCULATION CHANGES AND MARINE RESERVOIR AGE 21 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) Conclusions We find that abrupt changes in ocean circulation, such as a complete shutdown and recovery of the Atlantic MOC, can alter marine surface and bottom reservoir ages by several hundred years. Regions of largest changes (North Atlantic N5 N and the African west coast) are all affected by modulations of either upwelling strength or vertical mixing. Because the reservoir age changes in these regions are so large, we also find substantial reservoir age variations in the surrounding area of the mentioned locations. Signals of reservoir age anomalies are transported by advection and diffusion. We find a marine surface reservoir age drop of 1 yr at the onset of the YD for the Cariaco Basin and Barbados, which are key paleoceanographic sites where varved sediments and corals are available and used for 14 C calibration curves (Hughen et al., 2, 26; Fairbanks et al., 25). Our model results agree fairly well with present-day and YD reconstructions by Bard et al. (1994) and Bondevik et al. (26), especially when reducing the air sea gas transfer velocity by 19%, as suggested by Müller et al. (in press). However, the simulated abrupt marine surface reservoir age increase at the end of the YD is not seen in the reconstructions by Bondevik et al. (26). There is the possibility that such a transient change of relatively short duration might not be resolved in the records of Bondevik et al. (26), given the large uncertainties and the limited time resolution of the reservoir age reconstruction. On the other hand, this would support the idea of a rather slow MOC recovery throughout the YD (Stocker et al., 27) rather than an abrupt recovery at the end of the YD as simulated by this model. Marine bottom reservoir age anomalies are larger than those of surface reservoir age. The model shows YD bottom reservoir ages up to 15 yr larger than modern ages in the Caribbean Sea and the Gulf of Mexico, and up to 17 yr larger south of Greenland. The abrupt MOC recovery leads to a rapid bottom reservoir age drop of 7 to 1 yr southeast of Greenland. Parameter sensitivity studies show that increased diapycnal diffusivity tends to result in higher control state (modern) surface reservoir ages, especially in the Pacific with largest anomalies of 2 yr to the North of Australia because of increased vertical mixing. Surface reservoir age changes due to the MOC shutdown are about 5 yr larger in the Atlantic in the simulation with increased diapycnal diffusivity. The differences to STDR are marginal in all other regions. Increased North Atlantic sea ice cover during the YD has a major effect on reservoir age when the Atlantic MOC is in the off mode. In the model a seasonally constant Atlantic equivalent ice cover north of 56 N increases North Atlantic reservoir age by up to 7 yr over 1 yr, although it is not certain if the sea ice extent during the YD was so far south (Koç et al., 1993). In the model simulation with sea ice cover only north of 63 N, the increase in reservoir age is substantially slower. Due to the very coarse latitudinal resolution of our model at high latitudes and the poorly constrained YD sea ice extent, our simulations with increased sea ice extent only give a firstorder estimate of the potential role of sea ice on reservoir age. In our simulations we were able to show large-scale changes in marine reservoir age in response to abrupt MOC changes and that the effect of sea ice needs to be considered. In this study, the influence of changes in the 14 C production rate have not been taken into account. More work needs to be done to better represent sea ice and include 14 C production variability in order to estimate their effect on marine surface and bottom reservoir age changes in greater detail. Acknowledgments This study was funded by the Swiss National Science Foundation. We thank S. Bondevik for providing his reservoir age reconstructions and comments. We also thank P. Parekh and three anonymous reviewers for many useful comments that helped to improve the paper. References Austin, W.E.N., Bard, E., Hunt, J.B., Kroon, D., Peacock, J.D., The 14 C age of the Icelandic Vedde Ash: implications for Younger Dryas marine reservoir age corrections. Radiocarbon 37 (1), Bard, E., Arnold, M., Mangerud, J., Paterne, M., Labeyrle, L., Duprat, J., Méllères, M.A., Sønstegaard, E., Duplessy, J.C., The North Atlantic atmosphere sea surface 14 C gradient during the Younger Dryas climatic event. Earth Planet. Sci. Lett. 126 (4), Bondevik, S., Mangerud, J., Gulliksen, S., 21. The marine 14 C age of the Vedde Ash Bed along the west coast of Norway. J. Quat. Sci. 16 (1), 3 7. Bondevik, S., Mangerud, J., Birks, H., Gulliksen, S., Reimer, P.J., 26. Changes in North Atlantic radiocarbon reservoir ages during the Allerød and Younger Dryas. Science 312, Butzin, M., Prange, M., Lohmann, G., 25. Radiocarbon simulations for the glacial ocean: the effects of wind stress, Southern Ocean sea ice and Heinrich events. Earth Planet. Sci. Lett. 235, Delaygue, G., Stocker, T.F., Joos, F., Plattner, G.-K., 23. Simulation of atmospheric radiocarbon during abrupt oceanic changes: sensitivities to model parameters. Quat. Sci. Rev. 22, Fairbanks, R.G., Mortlock, R.A., Chiu, T.-C., Cao, L., Kaplan, A., Guilderson, T.P., Fairbanks, T.W., Bloom, A.L., Grootes, P.M., M.-J., N., 25. Radiocarbon calibration curve spanning to 5, years BP based on paired 23 Th/ 234 U/ 238 U and 14 C dates on pristine corals. Quat. Sci. Rev. 24, Friedrich, M., Remmele, S., Kromer, B., Hofmann, J., Spurk, M., Kaiser, K.F., Orcel, C., Küppers, M., 24. The 12,46-year Hohenheim oak and pine treering chronology from central Europe a unique annual record for radiocarbon calibration and paleoenvironment reconstructions. Radiocarbon 46 (3), Godwin, H., Half-life of radiocarbon. Nature 195, 984. Gulliksen, S., Birks, H.H., Possnert, G., Mangerud, J., A calendar age estimate of the Younger Dryas Holocene boundary at Krakenes, western Norway. Holocene 8 (3), Haflidason,H.,Sejrup,H.P.,Kristensen,D.K.,Johnsen,S.,1995.Coupledresponse of the late-glacial climatic shifts of northwest Europe reflected in Greenland ice cores: evidence from the northern North Sea. Geology 23 (12), Hughen, K.A., Southon, J.R., Lehman, S.J., Overpeck, J.T., 2. Synchronous radiocarbon and climate shifts during the last deglaciation. Science 29, Hughen, K.A., Southon, J.R., Bertrand, C.J.H., Franz, B., Zermeño, P., 24. Cariaco basin calibration update: revisions to calendar and 14 C chronologies for core PL7-58PC. Radiocarbon 46 (3), Hughen, K., Southon, J., Lehman, S., Bertrand, C., Turnbull, J., 26. Marinederived 14 C calibration and activity record for the past 5, years updated from the Cariaco Basin. Quat. Sci. Rev. 25, Karlén, W., Olsson, I.U., Kallberg, P., Kilicci, S., Absolute determination of the activity of two 14 C dating standards. Ark. Geofys. 4, Key, R.M., Kozyr, A., Sabine, C.L., Lee, K., Wanninkhof, R., Bullister, J.L., Feely, R.A., Millero, F.J., Mordy, C., Peng, T.-H., 24. A global ocean carbon climatology: results from GLODAP. Glob. Biogeochem. Cycles 18.

151 149 S.P. Ritz et al. / Earth and Planetary Science Letters 268 (28) Koç, N., Jansen, E., Haflidason, H., Paleoceanographic reconstructions of surface ocean conditions in the Greenland, Iceland and Norwegian seas during the last 14 ka based on diatoms. Quat. Sci. Rev. 12, Kromer, B., Friedrich, M., Hughen, K.A., Kaiser, F., Remmele, S., Schaub, M., Talamo, S., 24. Late glacial 14 C ages from a floating, 1382-ring pine chronology. Radiocarbon 46 (3), Levitus, S., Boyer, T., NOAA Atlas NESDIS 3: World ocean atlas Tech. Rep. Volume 4: Temperature, U.S. Department of Commerce: National Oceanic and Atmospheric Administration. Levitus, S., Burgett, R., Boyer, T., NOAA Atlas NESDIS 3: World ocean atlas Tech. Rep. Volume 3: Salinity, U.S. Department of Commerce: National Oceanic and Atmospheric Administration. Marchal, O., Stocker, T.F., Muscheler, R., 21. Atmospheric radiocarbon during the Younger Dryas: production, ventilation, or both? Earth Planet. Sci. Lett. 185, Meissner, K.J., Schmittner, A., Weaver, A.J., Adkins, J.F., 23. Ventilation of the North Atlantic Ocean during the Last Glacial Maximum: a comparison between simulated and observed radiocarbon ages. Paleoceanography 18 (2), 123. Monnin, E., Indermühle, A., Dällenbach, E., Flücklger, J., Stauffer, B., Stocker, T.F., Raynaud, D., Barnola, J.-M., 21. Atmospheric CO 2 concentrations over the Last Glacial Termination. Science 291, Müller, S.A., Joos, F., Edwards, N.R., Stocker, T.F., 26. Water mass distribution and ventilation time scales in a cost-efficient, three-dimensional ocean model. J. Climate 19, Müller, S.A., Joos, F., Plattner, G.-K., Edwards, N.R., Stocker, T.F., in press. Modelled natural and excess radiocarbon-sensitivities to the gas exchange formulation and ocean transport strength. Global Biogeochemical Cycles. Muscheler, R., Kromer, B., Björk, S., Svensson, A., Friedrich, M., Kaiser, K.F., Southon, J., in press. Tree rings and ice cores reveal 14 C calibration uncertainties during the Younger Dryas. Nature Geoscience. doi:1.138/ ngeo.xxxx. Orr, J.C., On ocean carbon-cycle model comparison. Tellus 51B, Reimer, P.J., Baillie, M.G.L., Bard, E., Bayliss, A., Beck, J.W., Bertrand, C., Blackwell, P.G., Buck, C.E., Burr, G., Cutler, K.B., Damon, P.E., Edwards, R.L., Fairbanks, R.G., Friedrich, M., Guilderson, T.P., Hughen, K.A., Kromer, B., McCormac, F.G., Manning, S., Bronk Ramsey, C., Reimer, R.W., Remmele, S., Southon, J.R., Stuiver, M., Talamo, S., Taylor, F.W., van der Plicht, J., Weyhenmeyer, C.E., 24. IntCal4 terrestrial radiocarbon age calibration, 26 cal kyr BP. Radiocarbon 46, Schmittner, A., 23. Southern Ocean sea ice and radiocarbon ages of glacial bottom waters. Earth Planet. Sci. Lett. 213, Southon, J., 24. A radiocarbon perspective on Greenland ice-core chronologies: can we use ice cores for C-14 calibration? Radiocarbon 46 (3), Stocker, T.F., Wright, D.G., Rapid changes in ocean circulation and atmospheric radiocarbon. Paleoceanography 11, Stocker, T.F., Wright, D.G., The effect of a succession of ocean ventilation changes on 14 C. Radiocarbon 4 (1), Stocker, T.F., Timmermann, A., Renold, M., Timm, O., 27. Effects of salt compensation on the climate model response in simulations of large changes of the Atlantic meridional overturning circulation. J. Climate 2, Stuiver, M., Braziunas, T.F., Modeling atmospheric 14 C influences and 14 C ages of marine samples to 1, BC. Radiocarbon 35 (1), Tarasov, L., Peltier, W.R., 25. Arctic freshwater forcing of the Younger Dryas cold reversal. Nature 435 (2), Walsh, J., A data set on Northern Hemisphere sea ice extent, Glaciological Data, World Data Center for Glaciology (Snow and Ice), Report GD-2, pp Zwally, H.J., Comiso, J., Parkinson, W., Campbell, W., Carsey, F., Gloerson, P., Antarctic Sea Ice, : Satellite Passive Microwave Observations. NASA. 26 pp.

152

153 Chapter 7 Outlook In this thesis, the Bern3D model is substantially improved by adding an energy and moisture balance atmosphere model to the existing ocean model. This addition enables the model to perform transient simulations in the past and into the future. The computational efficiency of the model permits model studies with multiple simulations on glacial-interglacial timescales, but also probabilistic assessments of future climate change can be performed with this model. In the applications presented, the understanding is improved of the behavior of noble gases in the atmosphere-ocean system, of the changes in marine reservoir age to abrupt changes in ocean circulation, and of the response of ocean temperatures to changes in atmospheric temperatures and ocean circulation. A new method is presented to reconstruct past changes in meridional overturning circulation, and in multiple simulations the glacial-to-modern change in 14 C atm is discussed. However, in every topic, further improvements can be made to increase the robustness and reduce uncertainties of the model results, and new, emerging questions can be answered. 7.1 The Bern3D Model Future steps in the development of the physical core of the coupled Bern3D model could include the redistribution of the model cells. Currently, the model cells are arranged such that the surface area of every model cell is of equal size. This has the benefit that the area term need not be included in the equations, but it has the disadvantage that the grid design is not flexible and not adjustable to specific needs. In particular, the high latitudes, especially the Arctic Ocean, the GIN Sea and the Labrador Sea, are poorly resolved in latitudinal direction. This in turn leads to inaccurate locations of deep water formation in the ocean, etc. The redistribution of the cells would enable a better resolution of the high latitudes without substantial losses in the computational efficiency of the model. However, this would require a complete re-formulation of the numerical part of the model. With the higher resolution of the Arctic Ocean, the bucket model of land hydrology (Section 2.2.4) might become an option to account for evaporation over land surfaces and snow cover. An alternative way to include the snow-albedo feedback is to tune the simple snow-albedo parametrization described in Section However, a sensitivity study should test the relevance of snow-albedo under modern and LGM climatic conditions. When running the Bern3D model with the land vegetation model component, the land albedo map should be replaced by the land albedos of the different plant functional types of the vegetation model. Finally, the incorporation of a dynamical ice-sheet model should be considered. For future model studies that investigate the mechanisms of glacial inceptions and glacial terminations,

154 OUTLOOK fully-coupled Earth System Models of Intermediate Complexity must be used where ideally the orbital parameters are the only external forcing factors. These studies will require models that can perform simulations of several thousand years. However, a realistic representation of the behavior of the ice sheets is doubtful due to the coarse resolution of the Bern3D model that provides the boundary conditions such as precipitation, atmospheric temperatures, ocean temperatures, and sea level. Especially the precipitation fields are poorly represented by the two-dimensional energy and moisture balance model. 7.2 Model Applications As in many model studies, in the AMOC reconstruction method presented in Chapter 3, a reduced-complexity climate model is used to develop and explore the method, because there is essentially no limitation on the number of simulations. After the methodological development, more complex climate models can be used to validate the method and to reduce the uncertainty of the outcome. In the particular case of the AMOC reconstruction, a 18 kyr model simulation is performed. Possibly, the duration of the simulation must be reduced when using a more complex model. A sensitivity study with the Bern3D model can examine the sensitivity of the regression coefficients (the output of the simulation) to the duration of the simulation. However, robustness of conclusions is an issue if only a model of reduced complexity is used for method development. In the noble gas study (Chapter 4), several factors of uncertainty have been detected, one of them being the equilibrium climate sensitivity (ECS). Here we refer to ECS as the equilibrium climate sensitivity of a modern state. The fact that δkr atm, δxe atm, and δar atm are dependent on ECS can potentially be exploited to narrow the estimated range of ECS. ECS is currently estimated to be between 2 C and 4.5 C (with 66 % probability) with a best estimate of 3 C (Knutti & Hegerl, 28). For this study, ice-core measurements can be compared to model simulations where different ECSs are applied. It is advisable to compare the results at times of minor sea-level changes, e.g., during Marine Isotope Stages (MISs). Preliminary work on this topic is displayed in Figs. 7.1 and 7.2 where modeled δkr atm values of the MISs of the last 8 kyr are calculated for ECSs from 2. C to 4. C. The model simulations are 8- kyr simulations with prescribed orbital parameters, greenhouse gases, and ice sheet extent as described in Chapter 2. However, more work needs to be done to take into account the uncertainties within the model such as, e.g., the prescribed ice-sheet extent. The precision of the future ice-core measurements will determine whether this method can narrow the currently estimated ECS range. The work on the LGM-to-modern 14 C atm changes (Chapter 5) reveals that the simulation of the LGM with respect to the current knowledge of the carbon cycle and ocean circulation is not as simple as suggested by Bouttes et al. (211). More work needs to be done to detect and quantify processes that lead to the glacial carbon-cycle state. A major challenge is that the resolution of global climate models is often too coarse to capture the proposed mechanisms. Hence, parametrizations of these processes must be implemented into the models. The robustness of such parametrizations should be tested by implementing them into several climate models. Once realistic representations of the LGM are feasible, the work of this chapter can be resumed to quantify the possible range of 14 C atm change that follows from the LGM-to-modern difference of the ocean circulation. The study on the changes of marine reservoir age to abrupt changes in the AMOC (Chapter 6) is done with the ocean-only Bern3D model, and the marine reservoir age is derived from

155 7.2. MODEL APPLICATIONS 153 Rel. sea level (m) δkr atm ( ) Marine Isotope Stages a) b) c) 4 2 Mean oc. temp. ( C) Time (kyr BP) Figure 7.1: 8-kyr noble gas simulation where ECS is set to 3 C. a) Modeled δkr atm. b) Modeled global mean ocean temperature. c) Prescribed relative sea-level. Here, the Lisiecki & Raymo (25) benthic δ 18 O stack is assumed to be proportional to relative sea-level. The red (interglacials) and blue (glacial maxima) lines indicate where the modeled δkr atm should be compared to observations, i.e., where sea-level changes are small. The corresponding MIS numbers are also displayed (Sarnthein & Tiedemann, 199)..4 δkratm ( ) Marine Isotope Stages glacial maxima interglacials Equilibrium Climate Sensitivity ( C) Figure 7.2: δkr atm results from five 8-kyr simulations. In every simulation, the equilibrium climate sensitivity (ECS) is set between 2 C and 4 C. The symbols represent the δkr atm at the Marine Isotope Stages (MISs) of the last 8 kyr. At the MISs, changes in sea level were relatively small (Fig. 7.1). The numbering of the MISs is done according to Sarnthein & Tiedemann (199). The model results can be compared to ice-core measurements to gain information on the ECS.

156 154 BIBLIOGRAPHY the radiocarbon-to-carbon ratio tracer 14 R. With the EBM, the study could be extended to analyze changes in marine reservoir age throughout the last 5 kyr. This is of particular interest, because the current 14 C atm reconstruction Intcal9 (Reimer et al., 29) relies to a large part on marine-derived data from the Cariaco Basin (Hughen et al., 26) where a constant marine reservoir age is assumed. However, in this study marine reservoir ages should be calculated from the DIC-14, DOC-14, DIC-12, and DOC-12 tracers instead of from 14 R, because in the 14 R implementation CO 2,atm is assumed constant. Hence, the model should incorporate the full carbon cycle. In this case the same problems arise as in the simulations of the LGM-to-modern 14 C atm change. Bibliography Bouttes, N., Paillard, D., Roche, D. M., Brovkin, V., & Bopp, L., 211. Last Glacial Maximum CO 2 and δ 13 C successfully reconciled, Geophysical Research Letters, 38, L275. Hughen, K., Southon, J., Lehman, S., Bertrand, C., & Turnbull, J., 26. Marine-derived 14 C calibration and activity record for the past 5, years updated from the Cariaco Basin, Quaternary Science Reviews, 25, Knutti, R. & Hegerl, G. C., 28. The equilibrium sensitivity of the Earth s temperature to radiation changes, Nature Geoscience, 1, Lisiecki, L. E. & Raymo, M. E., 25. A Pliocene-Pleistocene stack of 57 globally distributed benthic δ 18 O records, Paleoceanography, 2, PA13. Reimer, P. J., Baillie, M. G. L., Bard, E., Bayliss, A., Beck, J. W., Blackwell, P. G., Ramsey, C. B., Buck, C. E., Burr, G. S., Edwards, R. L., Friedrich, M., Grootes, P. M., Guilderson, T. P., Hajdas, I., Heaton, T. J., Hogg, A. G., Hughen, K. A., Kaiser, K. F., Kromer, B., McCormac, F. G., Manning, S. W., Reimer, R. W., Richards, D. A., Southon, J. R., Talamo, S., Turney, C. S. M., van der Plicht, J., & Weyhenmeyer, C. E., 29. INTCAL9 AND MARINE9 radiocarbon age calibration curves, -5, years Cal BP, Radiocarbon, 51, Sarnthein, M. & Tiedemann, R., 199. Younger Dryas-style cooling events at glacial terminations I-VI at ODP site 658: associated benthic δ 13 C anomalies constrain meltwater hypothesis, in Paleoceanography, vol. 5, pp

157 Appendix A Abbreviations Table A.1: Some abbreviations used in this thesis. AABW Antarctic Bottom Water AGCM Atmosphere general circulation model AOGCM Atmosphere-ocean general circulation model AMOC Atlantic meridional overturning circulation Ar Argon BP Before present, before year 195 AD 14 C Radiocarbon CH 4 Methane CO 2 Carbon dioxide δd, δ 18 O Hydrogen and oxygen isotopic composition δ 13 C, 14 C Carbon isotopic compositions δar atm, δkr atm, δxe atm Ar/N 2, Kr/N 2, and Xe/N 2 ratios relative to modern ratios EBM Energy and moisture balance model EMIC Earth System Model of Intermediate Complexity GCM General circulation model H1 Heinrich event 1 kyr Thousand years Kr Krypton LGM Last glacial maximum MOC Meridional overturning circulation Myr Million years N 2 Nitrogen N 2 O Nitrous oxide NADW North Atlantic Deep Water OGCM Ocean general circulation model POC Particulate organic carbon SO Southern Ocean SOMOC Southern Ocean meridional overturning circulation SSS Sea-surface salinity SST Sea-surface temperature TOA Top of the atmosphere Xe Xenon YD Younger Dryas yr Year

158

159 Acknowledgments I would like to thank......thomas Stocker for giving me the opportunity to write this thesis. I very much enjoyed the possibility to work on so many aspects of climate modeling, beginning with the development and tuning of a climate model, which is the best way to learn the possibilities and weaknesses of a model. I also learned a lot from the broad choices of modeling applications, from the implementation and thorough testing of a new paleoclimatic proxy to the testing of hypotheses inferred by proxy data and the synthesis between model and proxy data to gain better information about the climate system. Thomas always encouraged me to attend international conferences and workshops to exchange newest research achievements with respected scientists from all over the world....fortunat Joos for giving me insight into the details of the Bern3D model, climate modeling in general, and the carbon cycle during lunch, coffee breaks, and whenever else I needed his help.... Marco Steinacher, Simon Müller, Tobias Tschumi, Markus Gerber, and Raphael Roth for the enlightenment and many discussions on the contents, implementations and development of the Bern3D model code, and Johannes Rempfer, Reidun Gangstø, and Payal Parekh for their contributions to the Bern3D model. I am proud of the current status of the Bern3D model. It has definitively found its niche in the hierarchy of climate models available worldwide.... Raphael Roth for his collaboration in the work on deglacial atmospheric radiocarbon changes (Chapter 5).... Kay Bieri for his impeccable work as a system administrator. Because of his efficiency, he even finds time to help the modelers with writing scripts that increase their quality of life.... my room mates and former room mates Johannes Rempfer, Beni Stocker, Renato Spahni, Anil Bozbiyik, Sybille Zürcher, Payal Parekh and Kasper Plattner for the good time and laughs we had when not staring into the computer monitor....my fellow student Adrian Schilt who has accompanied me in my studies since the Gymnasium. Since he was always a couple of months ahead of me, I was able to profit from administrative tips and up-to-date LaTeX templates. I always enjoyed taking a day off together for paragliding in the Bernese Alps or the Jura....Doris Rätz for her administrative work....all the members of the Climate and Environmental Physics department for a wonderful time at lunch, during coffee breaks and after-work beers....my parents for their generous support throughout my 27 years of education....rahel Schneider for her love and continuous high spirits.

160

161 Publications Ritz, S. P. & Stocker, T. F., 211. Noble gases as proxies of mean ocean temperature: sensitivity studies using a climate model of reduced complexity, Quaternary Science Reviews, submitted. Ritz, S. P., Stocker, T. F., & Müller, S. A., 28. Modeling the effect of abrupt ocean circulation change on marine reservoir age, Earth and Planetary Science Letters, 268, Ritz, S. P., Stocker, T. F., & Joos, F., 211. A coupled dynamical ocean - energy balance atmosphere model for paleoclimate studies, Journal of Climate, 24, Siddall, M., Stocker, T. F., Henderson, G. M., Joos, F., Frank, M., Edwards, N. R., Ritz, S. P., & Müller, S. A., 27. Modeling the relationship between 231 Pa/ 23 Th distribution in North Atlantic sediment and Atlantic meridional overturning circulation, Paleoceanography, 22, PA2214. Wanner, H., Solomina, O., Grosjean, M., Ritz, S. P., & Jetel, M., 211. Structure and origin of Holocene cold events, Quaternary Science Reviews, submitted.

162

163 Erklärung gemäss Art. 28 Abs. 2 RSL 5 Name/Vorname: Ritz Stefan Matrikelnummer: Studiengang: Physik Bachelor Master Dissertation Titel der Arbeit: Development of a Reduced-Complexity Climate Model and Applications on Glacial-Interglacial Timescales Leiter der Arbeit: Prof. Dr. Thomas F. Stocker Ich erkläre hiermit, dass ich diese Arbeit selbständig verfasst und keine anderen als die angegebenen Quellen benutzt habe. Alle Stellen, die wörtlich oder sinngemäss aus Quellen entnommen wurden, habe ich als solche gekennzeichnet. Mir ist bekannt, dass andernfalls der Senat gemäss Artikel 36 Absatz 1 Buchstabe o des Gesetzes vom 5. September 1996 über die Universität zum Entzug des auf Grund dieser Arbeit verliehenen Titels berechtigt ist. Bern, 1. Mai 211

164

165 Curriculum Vitæ Personal Data Name Stefan Ritz Date of birth December 26, 1983 Place of origin Ferenbalm BE Education Primary School, Austin TX, USA Primary School, Bolligen, Switzerland Secondary School, Bolligen, Switzerland Gymnasium Kirchenfeld, Bern, Switzerland 6/22 Matura with a specialization on Physics and Applied Mathematics Studies in Physics at the University of Bern, Switzerland (subsidiary subjects: Mathematics and Computer Sciences) 12/25 Start of master thesis in Physics at the division of Climate and Environmental Physics, Physics Institute, University of Bern, Switzerland (Thesis advisor: Prof. Dr. Thomas F. Stocker) Assistant at the division of Climate and Environmental Physics, Physics Institute, University of Bern, Switzerland 3/27 Master of Science in Physics from the University of Bern, Switzerland 4/27 Start of Ph.D. thesis at the division of Climate and Environmental Physics, Physics Institute, University of Bern, Switzerland (Thesis advisor: Prof. Dr. Thomas F. Stocker)