Fingerprints of changes in the terrestrial carbon cycle in response to large reorganizations in ocean circulation

. CO 2 and carbon cycle changes in the land, ocean and atmosphere are investigated using the comprehensive carbon cycle-climate model NCAR CSM1.4-carbon. Ensemble simulations are forced with freshwater perturbations applied at the North Atlantic and Southern Ocean deep water formation sites under pre-industrial climate conditions. As a result, the Atlantic Meridional Overturning Circulation reduces in each experiment to varying degrees. The physical climate ﬁelds show changes qualitatively in agreement with results documented in the literature, but there is a clear distinction between northern and southern perturbations. Changes in the physical variables, in turn, affect the land and ocean biogeochemical cycles and cause a reduction, or an increase, in the atmospheric CO 2 concentration by up to 20 ppmv, depending on the location of the perturbation. In the case of a North Atlantic perturbation, the land biosphere reacts with a strong reduction in carbon stocks in some tropical locations and in high northern latitudes. In contrast, land carbon stocks tend to increase in response to a southern perturbation. The ocean is generally a sink of carbon although large reorganizations occur throughout various basins. The response of the land biosphere is strongest in the tropical


Introduction
Records from various climate proxies, especially Greenland ice cores and sediments from the North Atlantic, suggest that there have been large and abrupt changes in the climate during the last glacial period (Stocker, 2000;Rahmstorf, 2002;Clement and Peterson, 2008). Those transitions occurred on the time scale of a few decades to as little as a few years. The Greenland ice core from North GRIP contains 26 such abrupt warming events with amplitudes locally of up to 16 • C (NorthGRIP Members, 2004;Huber et al., 2006;Steffensen et al., 2008). These are known as Dansgaard-Oeschger (D-O) events (Dansgaard et al., 1984;Oeschger et al., 1984). There are also accompanying intense cold periods preceding these events which have a corresponding signal in the sediment cores from the northern Atlantic that are marked by distinct layers of ice rafted debris (Heinrich, 1988;Bond et al., 1993;Hemming, 2004). These surges of large amounts of ice into the sea, known as Heinrich events, are associated with some of the coldest temperatures in Greenland. Moreover, high-resolution measurements on Antarctic ice cores indicate that each of the D-O events has a less abrupt counterpart in the south (EPICA Community Members, 2006), likely due to the operation of the bipolar seesaw (Stocker and Johnsen, 2003). The last event in this sequence was the Younger Dryas cold event (12.7-11.6 cal kyr BP) and its termination with an abrupt warming (D-O 0). Signatures of those climate events have also been found in other climate archives, such as the isotopic and pollen records in the lake and marine sediments (Eicher et al., 1981;Ruddiman and McIntyre, 1981;Bond et al., 1993;Yu and Eicher, 1998;Ammann et al., 2000;Baker et al., 2001;Prokopenko et al., 2001;Voelker, 2002;Hughen et al., 2004;Barker et al., 2009) and European and Asian loess records (Ding et al., 1999;Porter, 2001;Rousseau et al., 2002) with a geographic distribution from Europe to Asia and beyond.
Published by Copernicus Publications on behalf of the European Geosciences Union. Many of the characteristics of an abrupt climate change event can be simulated by coupled climate models that are perturbed by anomalous freshwater fluxes applied in the North Atlantic (Bryan, 1986;Mikolajewicz, 1996;Schiller et al., 1997;Manabe and Stouffer, 1999;Marchal et al., 1998;Timmermann et al., 2003;Knutti et al., 2004;Zhang and Delworth, 2005;Stocker and Marchal, 2000;Stouffer et al., 2006), which can cause the collapse of the Atlantic Meridional Overturning Circulation (AMOC).
The purpose of this paper is to determine the dependence of the climate and carbon cycle to the freshwater perturbations of different origins and identify fingerprints of these responses, by using a comprehensive atmosphere-ocean global circulation model coupled with a land surface model. Moreover, we quantify the changes in the carbon cycle during a collapse of the AMOC under pre-industrial conditions, considering both northern and southern possible origins. In this respect, both ocean response and the response of the land biosphere have been investigated with a special focus on a region of South America, where the model simulates particularly strong responses.
The reasons behind these abrupt changes found in the paleoclimate records have been surmised already in 1984 as a result of the nonlinear nature of the ocean-climate system . One such nonlinearity in the system is the existence of different modes of the thermo-haline circulation. At least some of the past abrupt climate changes, including the Younger Dryas event which caused intense cooling around the northern Atlantic and had climatic impacts over the globe, are considered to be a result of such rapid reorganizations (Boyle and Keigwin, 1987;Duplessy et al., 1988;Broecker, 1997;Clark et al., 2002). It is also documented in the records of 14 C and 10 Be that the ocean's ventilation in the North Atlantic was slowed during these cold periods (Hughen et al., 2000;Muscheler et al., 2000).
Any interruption in this ocean-wide circulation would have climatic effects both on regional and global scales. Those effects include intense cooling in the Northern Hemisphere, centred around Northern Europe and Greenland spreading to the northern Pacific (Okumura et al., 2009); changes in the marine ecosystem in the Atlantic (Schmittner, 2005) and sea level in the North Atlantic (Levermann et al., 2005); changes in precipitation patterns over the tropics due to the shift of the Inter-tropical Convergence Zone (ITCZ) (Vellinga and Wood, 2002;Dahl et al., 2005) and changes in the El Niño-Southern Oscillation phenomenon (Timmermann et al., 2005. The tropics have an important role in the abrupt climate change events, that is globalizing the Northern Hemispheric phenomenon of the AMOC shutdown through reorganizations in the ocean and the atmosphere (Chiang, 2009).
Understanding the response of the global carbon cycle to a large freshwater input into the ocean is necessary in order to explain the changes in the CO 2 concentration in the atmosphere during those abrupt events. During the intense cooling events in Greenland and the more gradual Antarctic warm events, atmospheric CO 2 records show small but significant variations. Antarctic warm events A1 to A4 had seen atmospheric concentrations of CO 2 rise by about 20 ppmv (Stauffer et al., 1998;Indermühle et al., 2000), and by about the same amount during the much shorter Younger Dryas cold event (Monnin et al., 2001).
There are two different ideas regarding the source of this increase, that is either the ocean or the change of vegetation cover on land. While some modelling experiments have suggested that this atmospheric CO 2 increase was due to an oceanic release of carbon Schmittner and Galbraith, 2008), others suggest that it was due to a land carbon release (Köhler et al., 2005;Obata, 2007;Menviel et al., 2008). Ocean outgassing, for instance, can explain the increasing atmospheric CO 2 levels if the cooling of the sea surface is constrained to the high northern latitudes, the warming in the Southern Ocean is more pronounced and the contribution from land is not taken into account . Schmittner and Galbraith (2008) also identified the ocean as the source of the atmospheric CO 2 increase during abrupt climate change events. Due to the absence of a complex atmospheric component, however, their model is probably limited in representing tropical precipitation changes that have a potentially large impact on the land biosphere.
The other possible contributor to the carbon cycle changes during abrupt climate change events is the land biosphere. Köhler et al. (2005) have found that under both pre-industrial and pre-Younger Dryas conditions, atmospheric CO 2 concentration rises due to the release of carbon from land, with the rise in the latter being slightly less pronounced.
More recently, Obata (2007) employed a general circulation model coupled with a simple land surface model to simulate an AMOC shutdown which caused a release of carbon from land resulting in an atmospheric CO 2 increase. In another study, using an earth system model of intermediate complexity (LOVECLIM), Menviel et al. (2008) suggested that in the event of an AMOC shutdown, the ocean acted as a carbon sink and the land as a carbon source under both pre-industrial and LGM (Last Glacial Maximum) conditions. Our results support their conclusions in many ways and extend it further by offering a clearer picture of the reaction of the land biosphere. An alternative location for applying freshwater perturbations is also a feature of our study which gives insight as to which hemisphere might have triggered such events in the past. The fingerprints of each trigger (northern or southern) are evident in the South American continent, which is the centre of action for carbon cycle changes on land. A comparison of our model results with paleo-records in that region is also provided.
The paper is organised as follows; in Sect. 2 a brief description of the model and the experiments is given, in Sect. 3 the results of our experiments are explained, which is followed by the discussion and conclusions in Sect. 4.
A. Bozbiyik et al.: Fingerprints of changes in the terrestrial carbon cycle 321 2 Methods

Model description
The model used in this study is a modified version of the CSM1.4-carbon climate model developed by the National Centre for Atmospheric Research (NCAR) in Boulder, USA. It is a fully coupled 3-D climate model that consists of land, ocean, atmosphere and sea ice components integrated via a flux coupler without flux adjustments (Boville and Gent, 1999).
The CSM 1.4-carbon source code is available electronically on the CCSM website (http://www.ccsm.ucar.edu/ working groups/Biogeo/csm1 bgc/). The detailed description of the model is given by Doney et al. (2006) and Fung et al. (2005). Further information on the sensitivity of the model to external forcing can be found in the literature Frölicher et al., 2009;Frölicher and Joos, 2010;Steinacher et al., 2009Steinacher et al., , 2010.
The atmospheric component CCM3 of the model has a spectral truncation resolution of approximately 3.75 • (T31 Grid) and 18 vertical levels with 10 in the troposphere and 8 in the stratosphere (Kiehl et al., 1998).
The ocean component is called the NCAR CSM Ocean Model (NCOM) and has 25 vertical levels with longitudinal resolution of 3.6 • and latitudinal resolution between 0.8 • to 1.8 • (T31x3 Grid) (Gent et al., 1998). Since the original version of CSM1.0, modifications have been made on horizontal and vertical diffusivity and viscosity to improve the equatorial ocean circulation and inter-annual variability.
The sea-ice component has the same resolution as the ocean component and the land component has the resolution of the atmosphere component. The overall water cycle is closed through a river runoff scheme.
In the fully coupled carbon-climate model, atmospheric CO 2 is a prognostic variable whose balance is determined by exchange fluxes with the land and ocean . The carbon-cycle in the ocean is based on the OCMIP-2 biotic carbon model (Najjar et al., 1992). The main differences between the original OCMIP-2 model and this model are that the biological source-sink term has been changed from a restoring formulation to a prognostic formulation and iron has been added as a limiting nutrient together with a parametrization for the iron-cycle (Doney et al., 2006).
The land biogeochemistry in the CSM1.4-carbon model is a combination of the NCAR Land Surface Model (LSM) (Bonan, 1996) and the Carnegie-Ames-Stanford Approach (CASA) biogeochemical model (Randerson et al., 1997), both of which are very well documented in the literature. Carbon is recycled in the CASA model following the life cycles of plant functional types (PFTs) through carbon assimilation via photosynthesis and carbon release via mortality, decomposition and microbial respiration . There are 3 soil texture types and 14 PFTs with fractional coverage of up to four PFTs within each model grid-box. Carbon assimilation is calculated by the LSM by estimating stomatal conductance of CO 2 and water vapour in the leaves that are in shaded or directly lit conditions (Sellers et al., 1996). The net primary productivity (NPP) is fixed as 50% of the gross primary productivity (GPP) and is calculated by LSM to be allocated to three alive biomass pools of leaf, wood and roots. The allocation of NPP to these biomass pools is climate dependent, i.e., more of the NPP is allocated to the roots under water-limited conditions, while under light-limited conditions leaves are the preferred choice of biomass pool (Friedlingstein et al., 1999). In addition to three alive biomass pools, there are 9 dead biomass pools with leaf mortality contributing to the surface litter pool, root mortality contributing to the soil litter pool and wood mortality contributing to the coarse woody debris pool. The rest of the 9 dead biomass classes includes dead surface and soil microbial pools and slow and passive pools. The rate of transfer between different carbon pools is climate sensitive, determined by soil temperature and soil moisture saturation.
The carbon cycle is fully coupled to the water and energy cycles such that changes in the temperature and soil moisture calculated by LSM affect the NPP, allocation and decomposition rates and changes in the leaf area fraction calculated by CASA affect GPP transpiration and albedo. A terrestrial CO 2 fertilization effect is inherent to the model because carbon assimilation via the Rubisco enzyme is limited by the internal leaf CO 2 concentration that is dependent on the atmospheric CO 2 concentration. Thus, the productivity increases with the atmospheric CO 2 concentration, eventually saturating at high CO 2 levels (Doney et al., 2006).
Other land surface processes that can affect atmospherebiosphere interactions, but are not implemented in this study include: explicit nitrogen cycle, fires, volcanic eruptions, dynamic vegetation change and anthropogenic land cover change.

Experimental setup
The model used in this study was brought to steady state with the 1000-year spin-up procedure undertaken by Doney et al. (2006). The 1000-yr integration is nearly stable with a minimal drift in the deep ocean. There are known biases in this spin-up, which include a cold bias in the surface air temperature (SAT) over the continental interior in the Northern Hemisphere, precipitation anomalies in the tropics such as the formation of the ITCZ over the Pacific as two bands of excess precipitation, and too much or too little precipitation over land in some tropical regions of South America, Central Africa and Southeastern Asia. These biases in the physical climate also lead to corresponding anomalies of NPP and carbon storage on land that include an underestimation of NPP in higher latitudes and an overestimation in lower latitudes. Nevertheless, overall global NPP compares well with the reconstructed pre-industrial levels of NPP. The simulated climatologies in carbon inventory and fluxes resemble those determined from available observations. Atmospheric CO 2 excursions are small, 4 ppm over several centuries, and no abrupt changes are found during the integration. The spin-up stops at the 1820 AD atmospheric CO 2 levels and our experiments start at this point in time. Additional runs with two of the freshwater settings have been performed starting from a slightly different point in time at the steady state in order to account for the effects of short term variability in the climate system. The starting points for those additional runs are chosen with a different ENSO state than the original runs.
The freshwater hosing experiments (Table 1) are designed to investigate the response of the ocean and climate system to a freshening of the surface water around the key deep water formation sites in the Northern and Southern Hemispheres under pre-industrial conditions of atmospheric CO 2 (278 ppm). Perturbations applied simulate a change in the freshwater budget of the high-latitude ocean as a result of a surge in glacial melt-water from Greenland, Antarctica, or other pre-industrial continental glacial formations such as the Laurentide icesheet, which is responsible for the abrupt climate events of Younger Dryas 12.7-11.6 cal kyr BP and another smaller one during Holocene around 8200 yr BP ( Barber et al., 1999).
It is important to note here that even though we compare our results with paleoclimate reconstructions, the boundary conditions used in our experiments are not glacial boundary conditions which had significantly different levels of atmospheric CO 2 , as well as differences in temperature, precipitation and biome distribution, especially in mid-and highlatitudes. In the Discussion section, we address this issue by comparing it to a previous study by Menviel et al. (2008).
For each perturbation the freshwater input was assumed to be a rectangular pulse uniform over 100 years. The duration of each run is 300 years including the duration of the perturbations. The three different sites where the freshwater is applied are the northern Atlantic Ocean between the latitudes of 50 • N and 70 • N (including the Labrador Sea) and the Weddell Sea and Ross Sea in the Southern Ocean.
The perturbation, which is actually a negative salinity flux as there is no actual water volume in the parametrization of the model, is set to correspond to a freshwater flux of 1.0 Sv (in two of the experiments smaller perturbations of 0.5 Sv and 0.3 Sv are applied) distributed uniformly across the area of the perturbation. There has been no salt compensation performed . The amounts of the freshwater fluxes are highly idealized and do not directly correspond to the recorded events in the past.

Global average
The most direct response of the climate system to a freshwater perturbation is a reduction in the maximum strength Table 1. List of experiments given together with the region where freshwater is applied and the size of the perturbation. Different starting years for perturbations in the ensemble experiments refer to the date in the Control and are chosen according to the ENSO index in order to account for the effect of natural variability in the system. In addition to the ensemble experiments, three sensitivity simulations have been performed to investigate the effect of a different perturbation size and/or freshwater input region. In response to a freshwater flux from near Antarctica (1.0 Wed and 1.0 Ros), the strength of the North Atlantic MOC slightly increases in the beginning, which is followed by a reduction of smaller magnitude than in the 1.0 NA experiment. The mechanism behind this is explained by Stouffer et al. (2007) as the dilution effect, i.e., freshwater input from the Southern Ocean, does not stay there but is rather transported to the other parts of the ocean. This is evidenced by the time evolution of salinity. The result of the freshwater perturbation is a decrease in the global mean surface air temperature (SAT) in all experiments (Fig. 1b). Together with the changes in the precipitation fields, these two climatic variables cause the atmospheric CO 2 concentration to increase by up to 20 ppmv in the 1.0 NA experiments and to decrease by up to 10 ppmv in the Southern Ocean experiments (Fig. 1c).

Physical response
Because temperature and precipitation changes impact the carbon cycle, we first present changes in these two variables.
In the 1.0 NA experiment annual mean surface air temperature in the Northern Atlantic region decreases drastically (Fig. 2, left column) with reduced heat input from lower latitudes, as a result of the shutdown of the MOC in the Atlantic. Anomalies of up to −15 • C are observed over the North Atlantic between Iceland and Scandinavia. The cooling is not confined to the North Atlantic region but further extends to the high latitudes over North America and Asia. In line with the idea of inter-hemispheric redistribution of heat, the Southern Hemisphere exhibits warm anomalies of up to 3 • C over the Southern Pacific, Atlantic and Indian oceans and up to 6 • C over South America. In 1.0 NA, this general pattern stays roughly the same even 200 years after the end of the freshwater input, except for a temporary spike in the SAT coinciding with the end of the perturbation. The reason for that warming lies in the subsurface warming in the North Atlantic as a result of the increased stratification. A resumption of the vertical mixing, once the freshwater perturbation is switched off, brings up water to the surface from the subsurface warm pool which was created during the AMOC shut-down. By that time, this subsurface warm pool is mixed with the surface water and, subsequently, the atmosphere. However, this mixing is insufficient to restart the AMOC because of the diluted state of the surface water. Even though there is a slight increase in the salinity during this exchange, it is not enough to bring it to the original values. To a lesser extent, the warming is also caused by the delayed warming in the Southern Hemisphere, as it takes time for the warming to penetrate towards Antarctica.
Generally, the duration and severity of the atmospheric cooling (and warming) is dependent on the size of the freshwater flux. However, South America remains the region where the strongest warming occurs in all the North Atlantic experiments.
In the 1.  Hemisphere near Antarctica and the strongest warming is in the Northern Hemisphere near Greenland. But in the 1.0 Ros experiment, partial cold anomalies can also be seen in the Northern Hemisphere high latitudes. Because of this more widespread cooling in the 1.0 Ros experiment, the range of global average SAT anomaly is comparable to the 1.0 NA experiment, even though local anomalies are not as severe.
Overall, the regional amplitudes of the anomalies are smaller for the Southern Ocean perturbations. This is in line with the fact that there is a smaller reduction in the AMOC strength, as mentioned in the previous section. This can be due to the fact that freshwater is diluted in the larger volume of the Southern Ocean. Thus, a larger freshwater input is needed in order to achieve a response similar in magnitude to the 1.0 NA experiment.
As air temperature and sea surface temperature play a great role in the determination of the precipitation patterns, a change in those fields also cause a change in the amount and distribution of precipitation over the globe (Fig. 2, right column). The largest precipitation anomalies occur mainly at low latitudes and over the oceans, where water vapour availability is the greatest. But for the following analysis of the land biosphere, the most important changes are in South America and Africa. In the 1.0 NA experiment, total annual precipitation anomalies of up to 1.6 m are recorded in those locations.
The distribution of the most severe precipitation anomalies near the equator is consistent with what is expected from a shift of the ITCZ. Northern Atlantic perturbations cause a southward shift of the ITCZ as a result of the cooling in the Northern Hemisphere, whereas perturbations from the Southern Ocean cause a northward shift. This northward migration of the ITCZ was also simulated in a study by Menviel et al. (2010). In that study, freshwater was applied uniformly over the Southern Ocean leading to similar changes in the ocean circulation and deep water formation. A decrease in the SST and SAT values in the Southern Hemisphere accompanied by subsurface warming in the Southern Ocean is also one of the commonalities.

Changes in the global carbon inventories
Climatic change, as a result of a freshwater perturbation as in our experiments, affects the distribution of carbon in the three main reservoirs of land, ocean and atmosphere. The freshwater perturbations from the northern deep water formation site cause an increase in the atmospheric carbon inventory, while the southern perturbations cause a decrease and associated changes in each carbon inventory (Fig. 3). The response of the land carbon stocks to the three perturbations from the North Atlantic amounts to a decrease by several tens of GtC. The magnitude of this decrease is proportional to the strength of the freshwater perturbation. It is also important to note here that in the control run a drift of about −6 GtC over 300 years is recorded in the ocean. This does not invalidate our experiments as the magnitude is relatively small. The negative carbon drift in the ocean is slightly masking the carbon uptake by the ocean in all experiments.
Changes in the land carbon stocks immediately affect the atmosphere, while the time required for this perturbation to reach the much larger inventory of the ocean is longer. The land carbon stocks decline as a response to the 1.0 NA perturbation causing the atmospheric CO 2 to increase, which, in time, is partially taken up by the ocean. Yet, in the time frame of our simulations a larger than expected portion of the carbon emitted from land remains in the atmosphere. This indicates that the ocean does not behave like a passive carbon sink, in which case it would be expected to take up a much larger proportion of the carbon emitted to the atmosphere. Instead, after the initial increase, total carbon in the ocean stays quite stable until the end of the experiment. This points to a reorganization of the ocean carbon cycle, which leads to a new equilibrium with the atmosphere. In the experiments with a freshwater input from Antarctica, the change in the land carbon stocks is an increase. The peak values of the increase in the land carbon pools are about 20 GtC for experiment 1.0 Ros and about 12 GtC for 1.0 Wed, resulting in a reduced atmospheric CO 2 concentration. However, these changes are not permanent and return to their original values soon after the perturbations stop. Changes in the ocean carbon inventory are smaller, reaching about 10 GtC by the end of the experiments and cannot be easily distinguished from the variations in the control. It is safe to assume that a southern perturbation on a scale comparable to our experiments (that is to say less intense than a northern perturbation) creates changes in the land biosphere that are relatively short-lived. The response of the ocean to decreasing atmospheric CO 2 concentrations would be releasing carbon to balance it. Whereas, the slight increase in the oceanic carbon instead of an expected decrease leads us to believe that the changes in the carbon cycle in the ocean play a role in this experiment, too. Hence, the stable behaviour of the ocean carbon inventory during the first 140 years of the 1.0 Ros experiment is probably due to the fact that the competing effects of the reorganization of the carbon cycle and the decrease of the atmospheric CO 2 concentrations due to land uptake cancel each other out.
A similar experiment performed with the LOVECLIM climate model did not lead to any significant changes in atmospheric CO 2 .

A more regional look
The response of the land biosphere shows clear latitudinal dependencies accompanied by the strong latitudinal coupling of the climate parameters, temperature and precipitation. Figure 4 shows the zonal averages of the changes in the temperature and precipitation fields over land, and total land carbon stocks per latitude. Changes in the carbon stocks closely follow the changes in the precipitation in lower latitudes, while in higher latitudes temperature anomalies become more dominant. This is not surprising given the fact that, as a result of an MOC shutdown, in the higher latitudes temperature anomalies are larger than precipitation anomalies and the lower latitudes experience substantial changes in precipitation due to the shift of the ITCZ and the exponential dependence of saturated water vapour pressure to temperature. Hence, small temperature changes translate into large differences in precipitation in the tropics where the ambient temperature is higher (Bard, 2002). Nevertheless, globally, precipitation anomalies are responsible for most of the change in the carbon stocks on land as most of the carbon emissions stem from the large vegetation pool of the low latitudes. However, Fig. 4 cannot capture the inhomogeneous distribution of changes within those climatic zones. Therefore, we consider a snapshot of the distribution of the changes in the carbon stocks both on land and in the ocean (Fig. 5). In the ocean, most of the carbon ends up in the Atlantic Ocean. However, this does not necessarily mean an increased air-sea gas exchange in this region. The main reason for the carbon build-up in the Atlantic is the increased transport of DIC-rich Antarctic Bottom Water and a smaller amount comes from the air-sea gas exchange. The decreased primary production makes a small negative contribution. This reorganization also causes the reduction of total carbon in the other oceans, especially in the northern part of the Pacific and Indian Oceans. The contribution from the air-sea gas exchange is positive in the Pacific and Indian oceans, whereas in the Southern Ocean outgassing prevails. The carbon stocks in the North Pacific are influenced negatively by the increased circulation at depth due to the onset of deep water formation in this region. The initiation of North Pacific meridional overturning as a result of the AMOC shutdown has also been reported in previous modelling and reconstruction studies Okazaki et al., 2010;Menviel et al., 2011). This is an interesting feature which requires further investigation with regards to the underlying mechanisms. However, this is beyond the scope of the present study.
On land, the biggest changes occur in the tropical regions of South America, Africa and southeastern Asia, with the high latitudes contributing to anomalies smaller in magnitude but more widespread. As has been shown in Fig. 4, the regions near the equator exhibit large precipitation anomalies due to the shift of the ITCZ, either positive or negative.
In order to account for the life-cycle of the land vegetation and different cycling time-scales of carbon, the model includes various carbon pools, as mentioned in the methods section. The biggest change occurs in the vegetation pool (Fig. 6). Carbon inventories in the vegetation decrease in northern latitudes and in northern South America. The response of the soil carbon is determined by the competing influences of the input from the vegetation pool as well as the microbial overturning in the soil, which is reduced due to the lower temperatures. The sum of these two influences at high latitudes, where bigger proportions of carbon are stored in the soil, is a small increase in soil carbon. Where the temperature change is positive, as is the case in parts of South America, the proportion of the soil carbon compared to the carbon stored in the vegetation cover is small. In northern South America soil carbon content amounts to about 8 kg m −2 on average, whereas average carbon stored in the vegetation cover is more than 20 kg m −2 . Because of these factors, the contribution of the soil carbon pool to the changes in the atmospheric CO 2 concentration remains small.
In order to quantify the correlation between the anomalies in various climatic variables and the changes in the carbon stocks on land, a linear regression analysis has been performed. As the vegetation carbon pool is the biggest contributor to those changes and it is directly affected by the changes in the net primary production (NPP), which in turn is determined by the climate, NPP is chosen as the variable of interest for this comparison. Figure 7 shows the the sensitivity of the NPP to those variables, given here with the colour shading. This reveals the latitudinal dependency of the sensitivity of NPP, especially to temperature anomalies; there is a positive dependence in the colder climates of high latitudes, whereas it is negative in the warmer tropical regions. Also shown in the figure is the correlation between NPP and three www.clim-past.net/7/319/2011/ climatic variables, temperature, precipitation and soil moisture. Soil moisture shows the highest correlation with NPP, as it is a composite variable that is determined by both precipitation and temperature. Correlation is generally higher in low latitudes for precipitation and soil moisture, while temperature shows a better correlation in the high latitudes. Nevertheless, South America stands out as a region of high correlation for each of the variables mentioned, including temperature.

South America -a more detailed analysis
In our model simulations, the contribution of the South American continent to the global atmospheric CO 2 rise (or fall, in the case of southern perturbations) is disproportionately high compared to the rest of the world. About half of all the losses in the land carbon pool is from the northern part of South America in the 1.0 NA experiment, as well as a comparable fraction of the gains in the 1.0 Ros experiment ( Table 2). The net change in the total carbon stocks in the northern part of South America (the region marked in Fig. 8 top left) is more than 40 GtC by the end of the 100-year perturbation. The driving factors of this substantial change are the increasing temperature and reduced precipitation, the combined effect of which is the transformation of one of the wettest climates on land into an arid desert-like climate unable to sustain the carbon-rich rain-forest type vegetation.
Precipitation decreases by about 400 mm yr −1 and the surface air temperature rises by 1.4 • C on average (Fig. 9). Soil carbon also decreases significantly, but this amounts to only half of the change in the vegetation carbon pool.
The response in the northern South America is a combination of changes in the precipitation and the air temperature, which is influenced by the prevailing winds. Over the ocean, where water vapour availability is primarily governed by the sea surface temperature, a positive correlation between the SAT and precipitation is apparent in most places. Over land, on the other hand, wind-driven transport of moisture from the sea is crucial. The north-easterlies that carry moist air to northern South America are weakened (Fig. 10) and, due to the colder SST in the northern equatorial Atlantic, their moisture content is reduced, which makes the region drier. This, in turn, results in a substantial drop in the latent heat flux ( Fig. 9 last panel), most of which is due to reduced canopy transpiration as a result of the decrease in rain-forest vegetation. Consequently, a larger portion of the heat is transferred as sensible heat, raising the SAT. Removal of the rain-forest type vegetation, therefore, creates a very important feedback which further reduces the evapotranspiration and the latent heat flux.
The dramatic change in the northern part of South America also persists in the 1.0 Ros experiment, though with an opposite sign. As in the 1.0 NA experiment, a greater portion of the total anomaly in this experiment is in the vegetation carbon pool and it follows the changes in the climate field; higher precipitation and lower SAT values that lead to higher soil moisture.
The rest of the continent is also affected significantly by the changes in the climate system. Among them are the western part of the continent and northeastern Brazil, both of which respond oppositely relative to the northern part of the continent such that, NPP increases in the 1.0 NA experiment and decreases in the 1.0 Ros experiment. Those changes, basically, follow the shift of the ITCZ. The opposite impact (relative to the northern South America) of the ITCZ displacement on the climate and vegetation of the northeastern Brazil during the Younger Dryas (Wang et al., 2004) and Heinrich events (Dupont et al., 2010) is also evident in the records of the past climate. The high-latitude regions of South America do not play an important role, in general. Although the southern parts of South America show a decrease in the carbon stocks in 1.0 Ros experiment due to the lower temperatures, compared to the other regions, it does not add up to significant values. This is simply because of the initially small size of the carbon pool in that region. Figures 8 and 9 reveal the characteristic property of a South American response to a reduction in the thermohaline circulation, which is a distinct and opposite reaction at the two different locations of the freshwater perturbation. 1.0 NA and 1.0 Ros experiments produce opposite responses not only in the north of the continent, but also in the rest of the continent (central/northeastern Brazil and possibly the western part of the continent) that form a dipole relationship with the north. Currently, the ITCZ over South America is located northward of the continent, and northeastern Brazil has a semi-arid climate. During reduced AMOC, the north end of this continental dipole responds in a positive way to a North Atlantic perturbation while the other end responds negatively, and vice-versa in the case of a southern perturbation. This kind of a decoupling of the precipitation response within the continent is also documented in other studies (Cruz et al., 2009).
ΔCarbon (kg/m2 ) Fig. 8. Snapshots of the anomalies of various climatic variables and terrestrial carbon by the end of the perturbation (decadal average) for the experiments 1.0 NA (left column) and 1.0 Ros (right column). Anomalies are calculated as the 97-106 years of experiments minus the 100-year averaged Control. The variables shown from top to bottom are as follows, SAT ( • C), precipitation (mm year −1 ), soil moisture (mm 3 mm −3 ), NPP (g m −2 day −1 ) and Total Carbon (kg m −2 ). Stippled areas indicate where the ensemble mean anomaly is significantly different from zero at the 67% level (Student's t-Test).

Comparison of South American paleoclimate reconstructions with the model results
A comparison of several paleoclimate reconstructions of precipitation anomalies during the Younger Dryas period (Table 3) with our model results shows a good agreement between proxy records and the 1.0 NA experiment responses in most locations (Fig. 11), which supports the robustness of the ITCZ-shift hypothesis and the existence of a dipole relation between the north of the continent and eastern and southern Brazil, as suggested by Wang et al. (2007). Such a dipole seems to exist between the north and the west of the continent too (Martin et al., 1997). This increase in precipitation in the west of the continent is also apparent in the proxy records. Differences in the boundaries of the precipitation increase and decrease exist, yet it should be kept in mind that the ability of the model to make regional predictions is limited by its resolution. Overall, the difference in behaviour of the global carbon cycle, together with the site-specific responses in South America give us a way to interpret the origin of the freshwater input. Thus, for future reconstruction studies considering the anti-phase relationship between different parts of the continent and also the opposite responses to the origin of the freshwater input should help us obtain a clearer picture of what has actually happened during such events.

Discussion and conclusions
In our model simulations the response of the climate and the land biosphere to a collapse or a reduction in the AMOC can be divided into two categories according to where the freshwater perturbation is applied. All the perturbations from the North Atlantic Deep Water formation region cause similar responses, which are opposite to those caused by the perturbations from the Weddell Sea and the Ross Sea. Weddell Sea and Ross Sea responses are also different from each other, Ross Sea creating a much more widespread cooling and a clearer precipitation signal. That seems to be because the Ross Sea is a more important player in the creation of the AABW in our model than the Weddell Sea and, therefore, the perturbation in the Ross Sea has a stronger effect on the global climate. We note, however, that the specific locations of deep water formation are model dependent.
The most significant changes in precipitation occur around the tropics near the ITCZ, whose position is sensitive to shifts in SST. This, in turn, creates large changes in the carbon stocks in these locations. Additionally, the fact that large amounts of carbon are stored in low latitudes causes these precipitation anomalies to amplify changes in carbon stocks. Köhler et al. (2005) forced the Lund-Potsdam-Jena (LPJ) model with an output from freshwater experiments with the ECBILT-CLIO model. They found large changes in carbon stocks in the boreal zone and relatively small carbon stock changes in the tropics in contrast to our results. There are a variety of differences between the two studies. Vegetation dynamics is explicitly simulated in the LPJ, whereas vegetation distribution is prescribed in the NCAR CSM1.4carbon model. On the other hand, interactions and feedbacks between vegetation and climate, such as those related to albedo and the water cycle, are represented in the coupled NCAR model, but not in the forced runs with LPJ. The 332 A. Bozbiyik et al.: Fingerprints of changes in the terrestrial carbon cycle atmospheric dynamics in the cost-efficient ECBILT-CLIO is represented in a simplified manner which limits its ability to simulate dynamics in the tropics.
While the southward shift of the ITCZ after the northern perturbations is well studied with climate models and supported by paleoclimate records (Leduc et al., 2009), we show that the opposite is also true, that is to say, the northward shift of the ITCZ in response to a southern perturbation. Moreover, this has direct consequences for the tropical rainforest type of vegetation in that region. The very direction of this shift determines the direction of the change in the atmospheric CO 2 concentration through its effects on the land carbon pool in the low latitudes.
The magnitude of CO 2 increase during the Heinrich events and also during the Younger Dryas event was around 20 ppmv, as recorded in ice-cores (Indermühle et al., 2000;Monnin et al., 2001). According to our results, all of this amplitude can be explained by the carbon release from the land biosphere while the ocean acts as a carbon sink. The isotopic signature of the CO 2 from the ice cores also points to a land origin for the increase during Younger Dryas (Smith et al., 1999). Moreover, in a recent comprehensive simulation of the last deglaciation, covering the Heinrich Event 1 (H1), with a coupled atmosphere-ocean general circulation model, Liu et al. (2009) successfully reproduced the major features of this cooling event. They found strong cooling  Table 3. List of the paleoclimate reconstruction studies of precipitation change in and around South America during the Younger Dryas event, as used in the compilation given in Fig. 11. Reconstructed anomalies during Younger Dryas are given in comparison with our model results from the experiment 1.0 NA. A wide range of climate proxies are used in these reconstructions including, marine sediments, lake sediments, ice cores and speleothems. Pollen records as written in parenthesis are indicative of a change in the vegetation. These are shown in the figure as squares to distinguish from the rest of the proxies (circles).   Servant et al. (1999), 18. van't Veer et al. (2000, 19. Wang et al. (2004), 20. Wang et al. (2007). in the Northern Hemisphere with a milder warming in the south, and reduced precipitation in the Cariaco basin (northern South America). The good agreement of the general patterns of change indicates that a comparison of a paleoclimate event under glacial conditions (H1) and our simulations under pre-industrial conditions is reasonable.
The net atmospheric CO 2 increase in our experiments is comparable to, but more than, that of Obata (2007) with similar initial conditions. Even though the changes in the climatic variables and the NPP show a strong resemblance, their magnitudes differ in some areas. These include a more wide-spread cooling in the Northern Hemisphere in our experiments, and a smaller land biosphere response to the precipitation anomalies in eastern Asia. Additionally, a larger negative anomaly is recorded in northern part of South America in our study. Yet, the general effect of the ITCZ-shift is robust in both studies. The differences may be attributed to the more limited representation of the land biosphere in the model used by Obata (2007).
The exact amount of the contribution of the land biosphere to this atmospheric CO 2 increase, however, should be taken with caution since the glacial vegetation cover on land was different than that implemented here. Our experiments are done under pre-industrial conditions with a larger vegetation carbon pool than during the glacial times. In a previous study (Menviel et al., 2008), in which the experiments were done under both pre-industrial and glacial boundary conditions, it has been shown that the differences in the amplitudes of the individual contributions from the land and ocean carbon pools may lead to an opposite net effect on the atmospheric CO 2 , even though the nature of each contribution is qualitatively the same. In their study, irrespective of the initial state, the roles of the ocean as a carbon sink and of the land as a carbon source remain unchanged. Also, the changes on land are very similar in both cases, that is, a reduction in the carbon stocks in the high and mid-latitudes of the Northern Hemisphere and in the tropics north of the equator and an increase to the south. Yet, the emissions from land under glacial conditions are weaker than under the pre-industrial conditions. That is probably due to the lower moisture content of the glacial atmosphere, which leads to the dampening effects of the ITCZ shift (Menviel et al., 2008), and relatively large gains in primary production in some regions such as eastern Asia and southern North America. Compared to our experiments, the main differences are the larger increases in carbon stocks in the Southern Hemisphere and the abovementioned regions in the north, which might be due to some model specific differences as well as the initial conditions. Nevertheless, as the patterns of the anomalies are very similar in both studies, it is safe to assume that our results are relevant for paleo-reconstructions as a possible indirect way to distinguish between the sources of freshwater discharge in abrupt cooling events, because the northern Atlantic and Antarctic perturbations have distinct implications for the land biosphere. Globally, the atmospheric CO 2 signal is different; an increase for the North Atlantic case and a decrease for the Antarctic case. Regionally, the South American continent proves to be in a particularly suitable position to record such past events, as the movement of the ITCZ -either to the south or to the north -would create distinguishable and, at some places, opposite responses.
In addition to what has been presented in the results section, there are some other notable features that are observed in response to the freshwater perturbations, such as the North Pacific Deep Water formation, or the strengthening of Southern Hemispheric westerlies (Fig. 10). A more detailed investigation of those responses may be undertaken in a future study.
The results of this study have also implications for future anthropogenic climate change which, as many modelling studies show, is to cause a reduction in the AMOC (Meehl et al., 2007). The effects of such a reduction can be substantial for climate and for low latitude ecosystems including, but not limited to, the rain-forests. It is also important to note that, the changes in the carbon cycle during such an event would possibly contribute to the increase in the atmospheric carbon and hence, operate as a weak positive feedback to the global warming, in addition to that associated with outgassing from a warmer ocean .