By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Due to essential maintenance the subscribe/renew pages will be unavailable on Wednesday 26 October between 02:00- 08:00 BST/ 09:00 – 15:00 SGT/ 21:00- 03:00 EDT. Apologies for the inconvenience.
 The influence of the large-scale ocean circulation on Sahel rainfall is elusive because of the shortness of the observational record. We reconstructed the history of eolian and fluvial sedimentation on the continental slope off Senegal during the past 57,000 years. Our data show that abrupt onsets of arid conditions in the West African Sahel were linked to cold North Atlantic sea surface temperatures during times of reduced meridional overturning circulation associated with Heinrich Stadials. Climate modeling suggests that this drying is induced by a southward shift of the West African monsoon trough in conjunction with an intensification and southward expansion of the midtropospheric African Easterly Jet.
 Life in the semiarid Sahel belt of tropical North Africa strongly depends on the availability of water and has been frequently affected by shifts to more arid climate, at least since the Pliocene [deMenocal, 1995]. The most recent drought occurred in the early 70s and 80s of the last century with partial recovery during the late 90s [e.g., Nicholson, 2000]. Historical records suggest that Sahel droughts result from changes in the large-scale distribution of sea surface temperature [e.g., Lamb, 1978; Folland et al., 1986] which (among other factors) is influenced by the heat transport due to the Atlantic meridional overturning circulation (AMOC) [Newell and Hsiung, 1987]. The contribution of the AMOC to the long-term variability of Sahel precipitation has not yet been demonstrated. Since the AMOC underwent substantial variations during the late Quaternary [McManus et al., 2004], high-resolution sediment records from ocean margin settings offer the opportunity to study the response of continental climate to changes in ocean circulation.
 The continental slope off Northern Senegal is an ideal site to study the history of Sahel drought, because it records the varying input of eolian dust and fluvial sediments from the adjacent African continent [Koopmann, 1981; Sarnthein et al., 1981] (Figure 1). Dust with particle sizes up to 200 μm [Stuut et al., 2005] is mobilized in the Sahel and the western Sahara [Grousset et al., 1998; Jullien et al., 2007] and transported offshore mainly by continental trade winds and the Saharan Air Layer [Prospero and Carlson, 1981]. Recent ship-based dust samples collected off Senegal and Mauritania between 13 and 20°N indicate that 44–83% of the dust is deposited at grain sizes larger than 10 μm [Stuut et al., 2005]. By contrast, 95% of the terrigenous sediments delivered by the Senegal River have grain sizes below 10 μm [Gac and Kane, 1986]. The Fe/K ratio of atmospheric dust samples [Stuut et al., 2005] shows a close relation to precipitation, with values around 2.01 (∼28°N–14°N, n = 7, SD = 0.41) in the Sahel-Saharan area and values around 3.88 (∼6°N–7°S, n = 11, SD = 0.41) in the tropics (Figure 2). This increase in Fe/K ratios toward the tropics reflects the increasing amount of dust derived from deeply chemically weathered terrains [Moreno et al., 2006] with relatively high concentrations of iron and aluminum in comparison to the more mobile potassium.
 The material transported with the Senegal River also has a very distinct geochemical signature (Table 1). Compared to the chemical composition of atmospheric dust in the Sahel [Orange et al., 1993; Stuut et al., 2005], suspended sediments in the Senegal River show significantly higher Fe/K and Al/Si ratios [Gac and Kane, 1986].
Table 1. Mean Fe/K and Al/Si Ratios of Suspended Matter From the Senegal River Moutha
Samples were taken in the period from August 1981 to November 1982 [Gac and Kane, 1986] of atmospheric dust in Dakar (mean of yearly averages from 1984 to 1987 from Orange et al. ) and atmospheric dust offshore Senegal [Stuut et al., 2005] close to the position of GeoB9508–5. n, number of observations; SD, standard deviation.
 The Senegal River mainly drains the western part of Guinea. Sediment input from the Senegal River is highly dependent on the total water discharge and mainly occurs during the rainy summer season [Kattan et al., 1987]. Both dust mobilization and fluvial input are controlled by the background climate; the most recent multidecadal Sahel drought has been associated with an increase in dust mobilization and export over the Atlantic [Prospero and Lamb, 2003] and a decrease in Senegal River discharge by more than 50% with respect to the long-term mean [Kattan et al., 1987].
 Here we present a 57-ka-long record of terrigenous sedimentation from the continental slope off Senegal, westward of the Senegal River mouth. Variations in the composition of the terrigenous material indicate that Sahel megadroughts occurred during Heinrich Stadials and were associated with cold North Atlantic sea surface temperatures during times of reduced meridional overturning circulation. We study the physics behind the changes in West African hydrology by means of a freshwater-hosing experiment using a fully coupled climate model. Our model results suggest that North Atlantic sea surface temperature and West African rainfall are linked through shifts in the positions of the monsoon trough and the midtropospheric African Easterly Jet.
2. Material and Methods
2.1. Measurements on Core GeoB9508–5
 Our 965-cm-long gravity core GeoB9508–5 was retrieved from the continental slope off Senegal at about 15°29.90N/17°56.88W from 2384 m water depth (Figure 1). Bulk sediment samples were taken every 2.5 cm downcore, washed over 150 and 63 μm sieves and dried in an oven at 60°C. From the size fraction >250 μm of each sample, 1–10 specimens of Cibicidoides wuellerstorfi were picked for isotope analyses. The isotopic composition of the foraminiferal shells was measured using a Finnigan MAT 252 mass spectrometer equipped with an automatic carbonate preparation device. The working standard gas (Burgbrohl CO2) was calibrated against Vienna PDB (VPDB) by using the National Bureau of Standards (NBS) 18, 19 and 20 standards. Internal precision, based on replicates of an internal limestone standard, was better than 0.07‰. From the total of 391 measurements, 10 outliers were rejected.
 The age model of core GeoB9508–5 is based on 12 radiocarbon ages on mixed samples of planktonic foraminifera picked from the >150 μm fraction. All dates were measured at the Leibniz-Laboratory for Radiometric Dating and Stable Isotope Research in Kiel. Raw ages were corrected for a reservoir age of 400 years; they were then converted to calendar ages using the “Fairbanks0107” calibration curve [Fairbanks et al., 2005] for ages smaller than 40,000 (Table 2). Seven additional age points were introduced between the fixed radiocarbon ages by alignment to the benthic δ18O record of core MD95–2042 [Shackleton et al., 2004] (Figure 3a).
Table 2. Radiocarbon Dates and Age Model of Core GeoB9508–5a
Core Depth (cm)
14C AMS age (−400) (ka B.P.)
Calendar Age (ka B.P.)
AMS ages were corrected by a reservoir age of 400 years and then converted to calendar ages using the Fairbanks0107 calibration curve [Fairbanks et al., 2005]. AMS, Accelerator Mass Spectrometer.
 Samples for grain size analyses were taken every 5 cm downcore. In order to isolate the terrigenous fraction from the deep marine sediments, several pretreatment steps were undertaken to remove different biogenic constituents. Organic carbon was removed by adding 10 ml H2O2 (35%) to approximately 750 mg of bulk sediment. Reaction was sped up by boiling the mixture. Boiling was continued until reaction stopped and excess H2O2 was decomposed into H2O and O2. Subsequently, CaCO3 was removed by boiling the sediment sample in 100 ml demineralized water for 1 min with 10 ml HCl (10%). The sample was diluted with demineralized water until pH = 7. Subsequently, biogenic opal was removed by adding 6 g NaOH to the sample in 100 ml water, and boiled for 10 min. The sample was diluted again with demineralized water until pH = 7. As a last step before the analysis, the sediment sample in 100 ml was boiled shortly with 300 mg of the dispersing agent Na4P2O7*10H2O. All samples were measured with a Coulter laser particle sizer LS200, resulting in 92 size classes from 0.4–2000 μm at a 5 cm (∼250 years) downcore sampling interval.
 Samples for geochemical measurements were taken at 4 cm intervals. The sediment material was dried at 200°C, powdered and homogenized. Single element concentrations were determined on 4 g of dry subsamples by energy dispersive polarization X-ray fluorescence (EDP-XRF) spectroscopy using a SPECTRO XEPOS instrument (see Wien et al.  for a detailed description of the setup). The device was operated with the software Spectro X-Lab Pro (Version 2.4), using the Turboquant method [Schramm and Heckel, 1998]. Analytical quality was assessed by repeated analyses of the certified standard reference material MAG-1 [Govindaraju, 1994]. The measured values were within 2% of the accepted value for Si, Al, K, Ca, and Fe and within 5% for Ti. The standard deviation of replicates was less than 2%.
2.2. Setup of Model Experiments
 Numerical experiments were performed with an adjusted version of the “paleo release” of the NCAR (National Center for Atmospheric Research) Community Climate System Model CCSM2.0.1. The global climate model is composed of four components representing atmosphere, ocean, land, and sea ice. The resolution of the atmospheric component is given by T31 (3.75° by 3.75° transform grid) spectral truncation for 26 layers, while the ocean has a mean resolution of 3.6° by 1.6° with 25 levels. The latitudinal resolution of the oceanic model grid is variable, with finer resolution near the equator (∼0.9°). This version of CCSM2.0.1 is referred to as CCSM2/T31x3a [Prange, 2008].
 A control run was performed in which we adopted the atmospheric composition of 1990 AD and initialized the model with modern observational data sets. An asynchronous integration technique was used to achieve a statistical equilibrium of the climate system within 300 years of model integration [Prange, 2008]. This spin-up phase was followed by a 200-year-long synchronous integration, the second half of which serves for model-data analysis of the control climate in the present study.
 The present-day control run was perturbed by a freshwater flux of 0.1 sverdrup (Sv, 106 m3 s−1) into high northern seas. The surface freshwater forcing was applied to the Labrador Sea, the Nordic Seas, the Arctic Ocean as well as the Hudson and Baffin bays. The continuously perturbed model was integrated (synchronously) for almost 450 years. Averages over the last 100 years of the water-hosing experiment were used for further analysis.
3.1. Downcore Variability of Benthic δ18O, Grain Size, and Elemental Ratios
 The age model of the core indicates an age at the base of the core of about 57 ka and mean sedimentation rates of about 17 cm ka−1. Between about 57 and 15 ka B.P., the benthic δ18O record is characterized by a series of events with relatively low benthic δ18O values (Figure 3). It has been previously shown [Shackleton et al., 2000] that these events coincide with warm temperatures over Antarctica, the so-called Antarctic Isotope Maxima (AIM) [European Programme for Ice Coring in Antarctica (EPICA), 2006] (Figures 4d and 4e), and are probably due to a combination of ice volume and deep-water temperature changes that occur synchronously with temperature changes over Antarctica.
 The Holocene section of this core contains only small amounts of coarse-grained dust (Figure 3b). Al/Si and Fe/K ratios are highest in the mid-Holocene, when the values are very close to the modern composition of Senegal River suspension (Figure 4 and Table 1). Both ratios decrease gradually toward the present.
 The glacial section of the downcore record of GeoB9508–5 is characterized by a series of abrupt increases in grain size associated with decreases of Al/Si and Fe/K ratios starting at about 49, 41, 31, 26, 19 and 13 ka B.P (Figures 3 and 4). These events coincide with the most prominent Antarctic Isotope Maxima (AIMs 1, 2, 8, 12) and their Northern Hemisphere counterparts, the Heinrich Stadials 1–5 and the Younger Dryas.
 During the Heinrich Stadials both low Al/Si and Fe/K ratios and the low amount of fine material (<20%) are consistent with the deposition of atmospheric dust and indicate a reduced contribution from Senegal River suspension. Generally, the amount of fine material (<10 μm) is highly correlated (R2 = 0.9 when interpolated to 500-year intervals) with Al/Si ratios.
3.2. Modern Atmospheric Circulation in CCSM2/T31×3a
 CCSM2/T31×3a simulates a robust Atlantic overturning circulation. Approximately 10 Sv of deepwater formed in the North Atlantic are exported to the Southern Ocean. A detailed description of the overall model performance in simulating the global climatology and ocean circulation can be found elsewhere [Prange, 2008].
 For the present study, the model's skill in simulating the West African monsoon circulation is of paramount importance. Recently, a comprehensive analysis of coupled general circulation models (CGCMs) revealed that many global state-of-the-art models failed to capture the major features of the West African monsoon circulation under modern boundary conditions [Cook and Vizy, 2006]. Eight of the 18 examined CGCMs did not even reproduce the summer migration of the tropical rain belt onto the West African continent. Figure 5a shows climatological near-surface winds and precipitation over West Africa for the summer season July–September as derived from reanalysis [Kalnay et al., 1996] and observational [Legates and Willmott, 1990] data, respectively. Transporting moisture onto the continent across the Guinean coast, the northward low-level monsoon flow penetrates as far north as 20°N, where it converges with dry northerly winds at the monsoon trough. Summer precipitation over West Africa has two distinct regional maxima. One is centered on the west coast between ∼5°N and ∼12°N, another is near the Cameroon highlands in the eastern Guinea coastal region. Both the wind and precipitation patterns are rather well captured by CCSM2/T31×3a (Figure 5b). Even though the winds over the Sahara are somewhat stronger than in the reanalysis, their flow direction is satisfactorily simulated. Convergence with the southerly monsoon winds takes place at ∼20°N. Summer precipitation maxima reside on the African continent. The location of the west coast maximum is fairly well reproduced, albeit the amount of rainfall is underestimated by the model. The Cameroon maximum is also too weak and too far inland.
 In winter (January–March) the northerly dry Harmattan winds penetrate as far south as ∼10°N in the reanalysis data (Figure 5c). In the Sahel, observed rainfall approaches zero, while the zonal band of maximum rainfall is located over the Gulf of Guinea. Even though the Harmattan winds are stronger and penetrate farther south in the CCSM2/T31×3a control run, the overall patterns of low-level winds and rainfall are well simulated (Figure 5d).
 The atmospheric circulation at higher tropospheric levels is also well captured by CCSM2/T31×3a. Figure 6 shows a cross section of the mean zonal wind velocity on the Greenwich meridian from reanalysis data and model output for the summer season. The West African monsoon is depicted as a low-level westerly flow. Above the low-level westerlies is the African Easterly Jet (AEJ) with maximum wind speeds between 700 and 500 hPa. The African Easterly Jet is the equatorward portion of the Saharan High, the divergence center that overlays the near-surface continental thermal low in both the model and the reanalysis data. In the upper troposphere (higher than 300 hPa) the lowermost part of the Tropical Easterly Jet (TEJ) is visible in both the model and reanalysis data.
 Compared to climatological data, summer rainfall is undersimulated south of ∼15°N and somewhat too high to the north of it. In CCSM2/T31×3a, 65% of the annual rainfall in the West African Sahel (i.e., west of 10°E in the latitude belt 10°N–20°N) occurs during the summer months July–September, while only 1.5% takes place during January–March. The precipitation climatology [Legates and Willmott, 1990] suggests that about 70% (1%) of the annual West African Sahel rainfall occurs during July–September (January–March). In summary, CCSM2/T31×3a simulates a realistic West African monsoon circulation and captures the major features of the West African precipitation climatology.
3.3. Response of the Atmospheric Circulation to AMOC Slowdown
 The continuous but small (0.1 Sv = 0.1·106 m3 s−1) freshwater input to the northern North Atlantic and Arctic Oceans induces a gradual decline of the AMOC. Deepwater formation in the North Atlantic slowly decreases and eventually falls below 2 Sv after ∼300 years of integration. The maximum northward heat transport in the North Atlantic decreases from 0.6 PW [Prange, 2008] to 0.32 PW in the hosing experiment. In the South Atlantic, the heat transport even changes its direction (i.e., it becomes southward in response to the freshwater perturbation). Summer precipitation and runoff over the West African Sahel parallels the decrease of the AMOC and the associated heat transport, suggesting an almost linear relationship between overturning strength and Sahel rainfall (Figure 7). The simulated drying is not restricted to the Sahel. Rather, it affects the entire West African region including the Guinea coast (Figure 8a). The slowing of the AMOC induces surface cooling in the North Atlantic realm which extends far over Europe, the Mediterranean, and northern North Africa (Figure 9). During summer (July–September), the associated positive sea level pressure anomaly results in a southward shift of the West African monsoon trough (also referred to as the Intertropical Convergence Zone) by about 3–4° latitude (Figure 8b). As a result, monsoonal rainfall retreats from the northern portion of the West African Sahel. Reduced evapotranspiration from the drier land surface (a reduction of ∼10 W m−2 in averaged summer surface latent heat flux over the West African Sahel) and less cloudiness (plus ∼15 W m−2 of averaged summer net solar flux at the surface) amplify the Sahel surface warming which is initiated by the anomalous southward low-level transport of warm Saharan air (note that reduced latent heat and enhanced shortwave fluxes at the surface are essentially balanced by increased sensible heat and net longwave fluxes). This warming results in a steepening of the meridional surface temperature gradient between the Sahel and the relatively cool Guinean coast. According to the thermal wind balance, the zonal midlevel circulation intensifies at the altitude of the African Easterly Jet (Figures 8c and 10). Near 15°N, where the surface wind anomaly converges (Figure 8b), an anomalous ascent of air occurs below the level of condensation. This upward flow supplies the anomalous easterly midlevel jet with relatively dry air from the north, while over southern West Africa, the accelerated midlevel easterly flow enhances the export of moisture from the continent to the Atlantic Ocean (Figure 8d). The enhanced moisture divergence is associated with a further reduction in rainfall over West Africa.
 The annual rainfall over the West African Sahel decreases by ∼25% in our water hosing experiment. This value is comparable to the precipitation decrease that took place during the recent Sahel drought of the 1970s/80s. Basically, drying of the Sahel takes place in all seasons in the hosing experiment. However, since the major portion of Sahel rainfall occurs during July–September (see above), the summer response of the African hydrological cycle to AMOC slowdown is by far most important.
 Today terrigenous sediments deposited on the continental slope of Senegal mainly stem from atmospheric dust and river input. These transport processes signify two different precipitation regimes: dust input mainly occurs during dry conditions whereas river input occurs when precipitation is high in the drainage basin of the Senegal River. Since our core is located at the boundary between both regimes it reflects the relative proportions of fluviatile and eolian sediments and should be a sensitive recorder of Sahel precipitation. This interpretation is consistent with the general decrease of Al/Si and Fe/K ratios during the late Holocene along with a gradual drying trend over North Africa in response to weakening insolation forcing [e.g., Kröpelin et al., 2008].
 The salient feature in gravity core GeoB9508–5 is series of abrupt increases in grain size associated with decreased Al/Si and Fe/K ratios. These events must be interpreted as periods with much lower sediment discharge from the Senegal River together with an increase in atmospheric dust input as a consequence of drought. Geomorphological evidence for a much drier climate in much of northern Senegal is given for the most recent multimillennial period of drought, occurring between about 19 and 15 ka B.P at the onset of the last deglaciation. During this time much of Senegal was covered by the so-called Ogolian Dunes [Michel, 1973]. The presence of these dunes as far south as 14°N suggests a southward shift of the corresponding climate zone by 4–5° and provides additional evidence for aridity in the West African Sahel. Further evidence for aridity in the Sahel, at least during the Younger Dryas, comes from several lake records [Gasse and van Campo, 1994; Gasse, 2000].
 Within the accuracy of the age model the repeated and abrupt initiation of dry Sahel climates occurs synchronously with the coldest sea surface temperatures in the northeastern Atlantic during the past 60 ka [de Abreu et al., 2003]. These extremely cold sea surface temperatures only occur during Dansgaard-Oeschger Stadials associated with ice rafting (Heinrich) Events in the Northern Hemisphere. The longest drought with an approximate duration of about 4 ka occurs at the onset of the last deglaciation together with Heinrich Stadial 1 in the North Atlantic [Sarnthein et al., 2000] and a pronounced warming (AIM 1) of the same duration over Antarctica [EPICA, 2006] (Figure 4e). It is generally accepted [EPICA, 2006] that the antiphasing of temperatures over Greenland and Antarctica is due to reductions in the AMOC with a decrease of the heat export from the South Atlantic to the North Atlantic and a subsequent cooling of much of the North Atlantic surface. Hence, our data strongly suggest a relation between the strength of the AMOC and sea surface temperature in the North Atlantic on the one hand and Sahel precipitation on the other hand.
 Previous studies [e.g., Street-Perrott and Perrott, 1990; Mulitza and Rühlemann, 2000; Chiang and Koutavas, 2004; Dahl et al., 2005; A. C. Itambi et al., Millennial-scale northwest African droughts relates to H events and D O Cycles: Evidence in marine sedients from off-shore Senegal, submitted to Paleoceanography, 2008] interpreted drought in the tropics by a southward shift of the Intertropical Convergence Zone (ITCZ). Our experiments indeed suggest a southward shift of the ITCZ (i.e., monsoon trough) during AMOC slowdown by a few degrees latitude over West Africa. It is however problematic to explain the widespread nature of drought exclusively with an ITCZ shift. Today, the ITCZ seems to be effectively independent of the system that produces most of the rainfall. The tropical rainbelt (which is often confused with the ITCZ) is in fact produced by a deep core of ascent lying between the axes of the AEJ and the TEJ, some ten degrees of latitude south of the ITCZ [Nicholson and Grist, 2003]. Interestingly, a southward shift of the ITCZ has been rejected as a cause of the multidecadal Sahel drought during the second part of the last century [Citeau et al., 1989].
 Also, a southward shift of the ITCZ as the sole reason for drought would require increased precipitation to the south of the present-day ITCZ. Such a pattern would be in disagreement with late Quaternary paleolimnological records from Lake Bosumtwi (∼6°N), Ghana, which indicate dry conditions at the Guinean coast during millennial-scale periods of reduced AMOC [Peck et al., 2004; Shanahan et al., 2006]. Likewise, paleoceanographic records off Nigeria (∼3°N) [Weldeab et al., 2007], Congo (∼6°S) [Schefuß et al., 2005] and Angola (∼12°S) [Dupont et al., 2008] do not show indications of increased precipitation during Heinrich Stadials. The simulated drying in our water hosing experiment is not restricted to the Sahel. Rather, it affects the entire West African region including the Guinea coast (Figure 5a) in agreement with the paleoclimatic evidence, but in contrast to the simulation of Dahl et al. . For this reason, the proposed mechanism which involves an intensified moisture export by the midlevel African Easterly Jet is a reasonable explanation for the observed precipitation pattern.
 It must be noted that our model setup includes a comprehensive land surface component with sophisticated soil-vegetation biogeophysics and hydrology [Bonan et al., 2002; Oleson et al., 2004], but fixed vegetation. An interactive vegetation cover would likely worsen the simulation of the present monsoon climatology. Prescribing the vegetation provides for a more reliable simulation of the climatology, although potentially important feedbacks are excluded in the water-hosing experiment. Even though vegetation-climate feedbacks could conceivably act to amplify the response of Sahel precipitation to a remotely forced perturbation [Charney, 1975; Zeng et al., 1999], there is no obvious reason why the basic dynamical mechanisms of Sahel drying deduced from our CGCM water-hosing experiment should be affected fundamentally by vegetation dynamics [cf. Hales et al., 2006]. We also note that the assumption of a strong positive vegetation-precipitation feedback over northern Africa has recently been challenged [Levis et al., 2004; Liu et al., 2007, 2006; Wang et al., 2008]. Essentially, the same holds for dust radiative feedbacks [Yoshioka et al., 2007]. Excluding vegetation and dust feedbacks, our approach can be considered as a “maximum simplicity model” to simulate a physically consistent mechanism of Sahel drying in response to a weakening of the AMOC.
 In the present study, no attempts have been made to simulate and perturb a glacial climate state. Simulations of glacial West African climate can hardly be validated. Moreover, it has recently been shown that the simulation of glacial African rainfall is mostly model dependent [Braconnot et al., 2007]. Within the framework of the Paleoclimate Modeling Intercomparison Project PMIP-2, simulations of the Last Glacial Maximum with five different CGCMs yielded summer Sahelian rainfall anomalies ranging between –42% and +16% depending on the model. We therefore suspect that trustworthy simulations of the glacial West African monsoon dynamics are currently not available. As a note of caution, however, we emphasize again that varying boundary conditions (i.e., ice sheet distribution, orbital parameters, atmospheric composition) through Marine Isotope Stages 2 and 3 have not been taken into account in our modeling study.
 Our study suggests a close relation between AMOC, North Atlantic sea surface temperature and Sahel precipitation. Predictions of the future rainfall in the Sahel are highly uncertain and range from wetter conditions to much drier conditions [Douville et al., 2006; Intergovernmental Panel on Climate Change (IPCC), 2007]. However, climate projections show that the thermohaline overturning will probably slow down in response to anthropogenic-induced warming [IPCC, 2007]. From the results of this study, it seems likely that the future of precipitation in the Sahel will strongly depend on the behavior of the AMOC and its influence on the spatial structure of global warming.
 Constructive comments by two anonymous referees greatly improved the paper. Thanks to Monika Segl and Birgit Meyer-Schack, Wolfgang Bevern, and Helena Filipsson for help with isotope analyses. AMS 14C datings were done by the staff of the Leibniz-Laboratory for Radiometric Dating and Stable Isotope Research in Kiel. The climate model experiments were performed on the IBM pSeries 690 Supercomputer of the Norddeutscher Verbund für Hoch- und Höchstleistungsrechnen (HLRN). This work was funded through the DFG Research Center/Excellence Cluster “The Ocean in the Earth System.”