You are on page 1of 8

underestimated in many models.

Concluding, there is
a strong need for MME and PPE studies into
bistability of the AMOC, accounting for freshwater
influx from rivers and melting of the Greenland ice
sheet. To date, these intercomparisons have proven
too computationally expensive, but EMICs can provide
an accurate alternative (Stouffer et al. 2006, Gregory
et al. 2005). Stommel’s   salt-advection feedback is an
important link which is hard to trace in the more
comprehensive AOGCMs and their ensemble
simulations, because of their dedication to presentclimate stability and overall underappreciation of
high-integration processes of feedback, which EMICs
do take into account (Hofmann and Rahmstorf 2009,
Claussen et al. 2002).

external basins. The conceptual setup is shown in
Figure 8.
The   elegance   of   Stommel’s model is its
commitment to symmetry. The temperatures and
salinity of box one are chosen as the exact negatives
of the other box, with zero the benchmark

4. Stommel’s  salt-advective box model
The discussed developments in climate research
seem to suggest a need for widening the spectrum of
model complexity, both in order to assess the
integration of physical processes such as feedbacks,
as well as in order to explore a wider parameter space
in ensemble simulations. Particularly the role of salt
advection in the Meridional Overturning Circulation
causes a continuous attention to simple models that
are conceptually strong in reproducing the
characteristic freshwater hysteresis response that is
found in some EMICs and AOGCMs. Simple climate
models of the ocean often involve the representation
of ocean basins in the form of interconnected boxes
through which water circulates due to density
differences   linearly   related   to   the   water’s  
temperature and salinity. Seminal work in this respect
is   Stommel’s   two-box model of thermohaline
circulation (1961).
Because in the real ocean salinity and
temperature can have entirely different flux
mechanisms at different spatial dimensions, a clear
demarcation of ocean volumes can be effective in
analysing feedback mechanisms between them.
Stommel did this by representing the MOC with
thermohaline circulation through two boxes,
connected at the top and at the bottom. Each box is in
turn connected to an external basin of constant
reference density and temperature. One box could
represent the poles or the North Atlantic, where the
other could represent the tropics, depending on the
initial conditions, thermohaline transfer coefficients,
flow parameters and reference properties of the

Figure 8. Stommel’s  two-box model setup

This gives two simple partial differential equations
that can be solved analytically:

Here, c and d represent the transfer coefficients
for temperature and salinity respectively. By making
the system dimensionless, Stommel shows that the
system depends on only three parameters in the
following equations:

Temperature and salinity are denoted by y and x, 𝜆
 represents a measure of resistance to flow between
the basins, 𝛿 the temperature-salinity relaxation time
(the ratio of their transfer coefficients) and R the
pycnal effects (the strength of density difference).
Solution of the setup yields the temperaturesalinity diagram shown in Figure 9. Clearly visible in
the diagram are the three equilibrium points, one of
which is the hysteresis bifurcation point, discussed
earlier. Perfect mixing within the basins is a crucial
elements. The imaginary mixing rods would have as
their counterpart in reality turbulent convective
mixing in the ocean. Renewed interest in the salt-

or vice-versa. The model used is a four-box temperature and salinity construction derived from Stommel’s  original   two-box model. Density anomaly lines are drawn to show the T-S ratio. initial conditions will be simulated that imitate symmetry between the basins. hysteresis is observed (Figure 10). also known as the bifurcation point beyond which the system can go from salinity driven to temperature driven. parametrised with constant flow w and v. a)is a temperature-dominated stable node. In comparison to the original model. For the purpose of examining thermohaline hysteresis effectively. we will not consider wind-stress effects in the following model. depending on the model parameters 𝛿.The main basins are connected to external basins only above the thermocline Figure 9. Adopted from Whitehead (2008). c) a salinity-dominated spiral node and b) an unstable saddle point. resulting in four variables each for salinity and temperature . where each basin has a thermocline which vertically separates two internal basins with different salinity and temperature.The thermocline allows turbulent property exchange through diffusive entrainment. effectively giving a total of four boxes . Temperature and salinity are plotted as a function of the bottom heating bath temperature. There are a few changes made in the model setup: . and subsequently suggest modelling avenues that modify the pioneer model to resemble the actual physics and geometry of the MOC. One can clearly see bistability in the system. which interact through entrainment. independent from the bottom and top flow q . 11 . Four-box salt-advection model with entrainment For the purpose of addressing the issues put forward  earlier.Each basin is divided internally by a thermocline.Each box is distinctly represented in the model.advection feedback mechanism has prompted laboratory experiments in which thermohaline hysteresis is reproduced (Whitehead 2009). It consists of 2 main basins of equal height and surface area. This is dependent on whether thermohaline hysteresis and bistability are reproduced despite the new model geometry and entrainment. in which top saltwater influx and bottom heating would cause fast convective mixing. The purpose of the model exercise is to study the effects   of   complexification   on   Stommel’s core result. which could influence the position of the bifurcation point (Guan and Huang 2008). Thermohaline hysteresis reproduced in a box experiment.  R and 𝜆.  a  more  complex  version  of  Stommel’s model is presented and evaluated upon its core results. Using an altered setup of one freshwater reservoir and one active basin. a few integral curves show the three equilibrium points the system can have. Other authors looking to conceptually re-examine   Stommel’s   box   model have pointed towards the prevalence of winddriven circulation. Figure 10. In   this   diagram   of   Stommel’s   box   of   thermohaline circulation. The setup chosen is seen in Figure 11.

We calculate the pressure balance that determines the bottom and top flow using p1 and p4 as the basin pressures at the surface. The equations governing this system are derived from standard equations of fluid property mixing (with V the volume of a basin):   = (T −T ) 𝑞 dT = (T − T ) = dt V dt T1 V The main flow q can be in two directions and in either situation the equations look very different. determined by the density difference of the two basins. and p2 and p3 at the bottom.𝒯4 𝒮4 A 𝒯1 𝒮1 A q T4 S4 T1 S1 V4 V1 = Ah1 v w T3 S3 T2 S2 V3 V2 Figure 11. The pressure difference determining the flow at the bottom of the two basins is defined by p2-p1 and p3-p4 respectively: dtqT2 (dt𝑞T + T V  – dt𝑞T ) V T dS dt dS dt dS dt dS dt cA (𝒯 − T ) V 12 . The flow q. must be equal from each box to the next so that the volume of the basins remains the same. Surface flow is constrained by parameter kt and bottom flow by kb. In addition to this regular flow. we can reduce both bottom and top flow to a single resistance parameter: cA (𝒯 − T ) V 𝑞 (T − T ) + 𝑞 (T − T ) + |𝑣|(T − T ) k=k +k : 𝑞 (T − T ) + 𝑞 (T − T ) + |𝑤|(T − T ) 𝑞 (T − T ) + 𝑞 (T − T ) + |𝑤|(T − T ) + dA (𝒮 − S ) V Solving for q. we define the main flow from the hydrostatic balance and the internal flow in the two basins through constants of entrainment v and w. this gives: The exchange of property at the surface of basins 1 and 4 is linearly dependent on the product of the surface area A and the salinity and temperature transfer coefficients d and c. The two top basins also simulate property exchange with the atmosphere through external basins with a fixed temperature and salinity. the two basins have internal entrainment flow at a specific height h which is dependent on the temperature diffusion coefficient and the surface of the thermocline. The system of equations then becomes: = 𝑞 (S − S ) + 𝑞 (S − S ) + |𝑣|(S − S ) + p p p −p p −p 𝑞 = dT dt dT dt dT dt dT dt 1 V 1 = V 1 = V 1 = V = In order to determine the flow parameters. Setup of four-box salt-advective model with separate variables and entrainment along a simulated thermocline. Entrainment can work to either counteract or contribute to the overall flow q. This problem is solved by converting them into signdependent components: 𝑞 + |𝑞| 2 𝑞 − |𝑞| 𝑞 =− 2 1 V 1 = V 1 = V 1 = V 𝑞 (T − T ) + 𝑞 (T − T ) + |𝑣|(T − T ) + 𝑞 (S − S ) + 𝑞 (S − S ) + |𝑣|(S − S ) 𝑞 (S − S ) + 𝑞 (S − S ) + |𝑤|(S − S ) 𝑞 (S − S ) + 𝑞 (S − S ) + |𝑤|(S − S ) + dA (𝒮 − S ) V −p =k 𝑞 −p =k 𝑞 = 𝜌 𝑔h + 𝜌 𝑔h = 𝜌 𝑔h + 𝜌 𝑔h (k + k )𝑞 = 𝑔 (𝜌 h + 𝜌 h ) − (𝜌 h + 𝜌 h ) (k + k )𝑞 = 𝑔 (𝜌 h + 𝜌 h ) − (𝜌 h + 𝜌 h ) Now.

δ represents the temperature-salinity relaxation time: Az (−𝑦 + R𝑥 )   (1 + z ) A (−𝑦 + R𝑥 ) + (1 + z ) A − (−𝑦 + R𝑥 )   (1 + z ) Az − (−𝑦 + R𝑥 )   (1 + z ) When we then introduce the parameter τ = ct. which corresponds to the temperature dependent timescale. 𝒮i. c. because it represents the   system’s   external   forcing. the system of equations becomes: In the choice of parameters later on. 𝑣 and 𝑤 such that they render the equations dimensionless. We can then make temperature and salinity dimensionless by taking them relative to this difference parameter: Here. The main interaction the model is concerned with is the surface property exchange. we define the previously introduced parameters k. We also determine the transfer coefficients using an abovethermocline temperature diffusion time tt: 𝜌 = 𝜌 (1 − αT + βS ) Hz (1 + z )t Hz d=δ (1 + z )t c= We then end up with a flow given by: 𝑞= 𝑔𝜌 k (−αT h + βS h − αT h + βS h ) − (−αT h + βS h − αT h + βS h ) Definition in terms of dimensionless variables In order to evaluate the model. d.   To   reduce   the   model’s   complexity. which is loosely defined as one for the pycnal effects: h = 𝑓= This turns the flow equation into the following expression: T 𝒯 S 𝑥 = 𝒮 β𝒮 α𝒯 d c We further define: 𝑦 = R= δ= d𝑥 dτ d𝑥 dτ d𝑥 dτ d𝑥 dτ ) ) ) ) 13 1 λh 1 = λh 1 = λh 1 = λh = 1 λh 1 = λh 1 = λh 1 = λh = |𝑣| (𝑦 h |𝑣| (𝑦 𝑓 (𝑦 − 𝑦 ) + 𝑓 (𝑦 − 𝑦 ) + h |𝑤| (𝑦 𝑓 (𝑦 − 𝑦 ) + 𝑓 (𝑦 − 𝑦 ) + h |𝑤| (𝑦 𝑓 (𝑦 − 𝑦 ) + 𝑓 (𝑦 − 𝑦 ) + h 𝑓 (𝑦 − 𝑦 ) + 𝑓 (𝑦 − 𝑦 ) + |𝑣| (𝑥 h |𝑣| (𝑥 𝑓 (𝑥 − 𝑥 ) + 𝑓 (𝑥 − 𝑥 ) + h |𝑤| (𝑥 𝑓 (𝑥 − 𝑥 ) + 𝑓 (𝑥 − 𝑥 ) + h |𝑤| (𝑥 𝑓 (𝑥 − x ) + 𝑓 (𝑥 − 𝑥 ) + h 𝑓 (𝑥 − 𝑥 ) + 𝑓 (𝑥 − 𝑥 ) + − 𝑦 ) + (𝒯 − 𝑦 ) −𝑦 ) −𝑦 ) − 𝑦 ) + (−𝒯 − 𝑦 ) − 𝑥 ) + δ(𝒮 − 𝑥 ) −𝑥 ) −𝑥 ) − 𝑥 ) + δ(−𝒮 − 𝑥 ) . we choose the external basin properties such that they become 𝒯1 = –𝒯4 = 𝒯. 𝒮1 = – 𝒮4 = 𝒮.We also introduce a simple form of the equation of state based on the temperature and salinity transfer coefficients α and β: Where the z parameter indicates the relative position of the thermocline in each main basin. 𝒯 will be used to derive a proportional 𝒮 from a chosen R. α and β are in the same order of magnitude and already proportional. We also introduce the following parameter for the resistance against flow: d𝑦 dτ d𝑦 dτ d𝑦 dτ d𝑦 dτ kcA λ= 𝑔𝜌 α𝒯H We assume H to represent the height of each main basin (in this model both basins are of equal height): Hz (1 + z H h = (1 + z H h = (1 + z Hz h = (1 + z λq c 𝑓= We introduce the following parameter. 𝒯i.

81 Figure 12.   so   they   are   left   in   their original dimensions. flow fluctuates at first. Figure 13. partly  based  on  Stommel’s  parameter  analysis: 1 6 R=2 1 λ= 5 e = e = e = 100 1 z =z =z= 5 δ= As remaining dimensional parameters we choose: m s kg 𝜌 = 1027 m H = 3000m A = 5 ∙ 10 m t = 100 𝒯 = 20 g = 9. freshwater influx as simulated by a specific initial salinity condition would lead to hysteresis past the bifurcation point into the spiral node. a T-S parametric plot for each basin will show how each basin approaches equilibrium values for both temperature and salinity depending   on   the   initial   conditions. For specific initial conditions.  What  is   different in the result that is presented is that the range of initial conditions that lead to the spiral node is very small (which is only reached from an initial salinity of 0.  equal   entrainment in the two boxes. In interpreting the result. For  simplicity’s  sake. the flow is expected to drop to zero. one of which is unstable (Figure 12). their location. λ parameters leads to loss of equilibria. if boundary conditions of the real ocean were perturbed such as to resemble the initial conditions of the model. Figure 13 shows the time lapse of the THC flow. as also with Stommel’s  original  model. shows the same three equilibria. Manipulation of the δ.In this model.5 +/. simulating a collapse of the THC. as Stommel showed as well.0. 14 . Control simulation with symmetrical initial conditions and high entrainment flow. Comparatively. entrainment fluxes are dependent on the temperature diffusion coefficient as well as the surface area of the thermocline: 𝑣 = cAe w = cAe2 The entrainment in this model may also be interpreted as a mixing coefficient between the bottom and top of the two main basins. As dimensionless parameters the following are chosen. R. but returns to zero eventually.2).   As   in   Stommel’s   initial model.or temperature dominated. whether certain conditions can make the basins unstable and if it is salinity. this will allow for examination of the amount of equilibrium points. These last parameters (specifically the last four) determine   the   model’s   geometry. using symmetry in the   initial   conditions   to   simulate   Stommel’s original model. Results and discussion A first solution of the model. Symmetry in the model is directly noticed because the density lines do not intersect with each other. Main flow q as a function of time. Model Initialization The system of equations is solved numerically and the time-evolution of the variables in each basin is plotted. Eventually. as well as equal heights of the thermocline and surface area are chosen.

the position and occurrence of the equilibria remains unchanged with a lower entrainment flow. Parametrisations of processes such as wind stress could be introduced in order to study the hysteresis behaviour. The effect of lower entrainment flow is shown in Figure 14. but the hysteresis result is kept. As expected. For this last part of the evaluation. Figure 14. and the presented model exercise seems to corroborate these conceptually. the other boxes have the same bistability. Entrainment flow of lower orders of magnitude (Top: e = 10. and that of climate in general. In essence.hindering of the natural hysteresis effect as long as entrainment is high enough. 5. Each box now has a distinct density profile. it appears that expansion of Stommel’s  salt-advection box model is not changed in its core result after complexification by including a linear entrainment factor and asymmetry in the initial conditions of the four boxes. although the three equilibria are found at exactly the same positions. but for the purpose of clarity only that of box one is shown in Figure 15. a wide range of different initial conditions were chosen at high entrainment flow. and many modelling efforts to date have mostly considered surface warming to indicate a linear decrease in AMOC strength in the 21st century. Density diagram of the four-box system with a selection of different initial conditions for each box. More chaotic than the symmetrical system. Bottom: e = 0. there is no 15 . The second measurable of the model was the initial condition asymmetry between boxes.1) shows more chaotic density behaviour in the system. The AMOC is potentially in a bistable regime. many of the density lines overlap and the first time steps show very fast flow. In the spectrum of models assessing the Meridional Overturning Circulation. Freshwater influx from Greenland ice sheet melt and river runoff could severely influence this seemingly Figure 15. Although the condition of non-intersecting density lines is not met. In this demonstration case. the model could be dimensionalised to resemble the conditions in actual ocean basins. Computational capacity limits a systematic exploration of the parameter space in AOGCMs over long timescales. it produces the same equilibria. Concluding. Conclusion We set out to examine two main issues of interest in the study of the Atlantic Meridional Overturning Circulation. a choice was made to have the value of entrainment be at least one order of magnitude higher than the volume of the basin. and multi-model ensembles lack a well-defined metric with which to weight individual models and interpret the acquired mean. An analysis of the literature suggests a number of possible solutions. so that its mixing effect is high. This result points towards further research into modifying the original model to more accurately represent ocean dynamics.

M. Z. 113–117. S. R. Jenkins. Nevertheless. Fichefet. Stommel’s   Box   Model   of   Thermohaline   Circulation Revisited—The Role of Mechanical Energy Supporting Mixing and the Wind-Driven Gyration. D.   and   simple   models   such   as   Stommel’s   are   clear in capturing specific mechanisms such as thermohaline hysteresis... Geophysical Research Letters 32. E. but can much more easily be tuned in their parameters and provide probabilistic uncertainty distributions. Allen. Intergovernmental Panel on Climate Change. C.. the tentative underappreciation of freshwater influx in some of the state-of-the-art multi-model ensembles calls for further research in the thermohaline hysteresis mechanism. L.   Uncertainty   Range   and Reliability of Regional Climate Changes from AOGCM Simulations via   the   ‘Reliability   Ensemble   Averaging’   (REA)   Method. R. from all sides of the modelling spectrum. The great variety in climate models comes with a need for representative future projections. IPCC Technical Paper 2 [Houghton. Geneva.   F. M. and K. J. As a final note. M. G. D. D.)]. 51 pp.J. K. 2005. 2004. present and future climate projections. L. M. Perturbed physics ensemble simulations seem to allow a systematic exploration of modelling parameter space. 1990: Processes and modelling. Meira Filho. J. M. A Model Intercomparison of Changes in the Atlantic Thermohaline Circulation in Response to Increasing Atmospheric CO2 Concentration. Stone. It shows that even when physical and geometrical expansions to the original two-box model in the form of thermocline entrainment and asymmetrical four-box initial conditions are made. E. H. et al. Weaver. Collins. A. at least in coarse resolution. 2002. EMICs 16 in the range close to AOGCMs. P. P. Lohmann. I. Annan.J. Geophysical Research Letters 38. study specific climate change mechanisms in conceptual detail. and mediation of internal variability and uncertainty. based on probabilistic intercomparison in the context of observational data. Using Ensemble Prediction Methods to Examine Regional Climate Variation under Global Warming Scenarios. Weaver. B.   2002.   L. J. Ephraums (eds. J. R. D. DOI:10. 2008. B. J. X. Pohlmann. Guan. on par with AOGCMs in their predictive capacity.. Hargreaves. Loutre. Mysak. M. Berger. J. Booth. 2010. Dixon.. Cambridge University Press. Huang. Hofmann. Rahmstorf.. Journal of Physical Oceanography 38: 909-918. 1997: An Introduction to Simple Climate Models Used in the IPCC Second Assessment Report. USA. J. H. influencing the ensemble in its totality. higher resolution. and simple climate models on the conceptual end of the spectrum. Recent freshwater forcing simulations using EMICs and incidentally even stateof-the-art AOGCMs have sparked a renewed interest in this hysteresis behaviour. A. Woollings. Smith.T. Griggs. although the model chosen to perturb could have imperfections in the core.   O. Cubasch.monostable state of the thermohaline mechanism that drives the circulation in the centuries beyond. and examine longterm climate forcing from output of AOGCMs. Bhaskaran. Ganopolski. G. T. Petoukhov. U.T. B. Lunkeit.   ‘Calculation   of   Average.. T. 2011. Webb. Crucifix. A. Quantifying Uncertainties in Climate System Properties with the Use of Recent Climate Observations. Wang. J. H. Harvey. which is why multi-model ensembles are used to provide a realistic consensus of modelling efforts.’   Journal   of   Climate 15: 1141–1158. Murphy. M. DOI: 10. J. Climate Dynamics 36: 1737-1766. et al. C. Journal of Scientific Computing 26: 467–487. Ocean Modelling 11: 174-192. Gregory J. in order to mitigate computational cost. Stouffer. Coupled Atmosphere-Ocean General Circulation Models represent the pinnacle of detail and accuracy in past. Various weighting methods to mediate the relative quality of ensemble members have been developed. R. Calov. Forest. Mokhov. We conclude by arguing for a wider scope of complexity in climate modelling. Alexeev.1029/2011GL047208. Switzerland. although an objective metric is yet to be agreed upon. R. Their development over the last decades has seen integration of more physical processes. On the Stability of the Atlantic Meridional Overturning Circulation. A. can be vitally important in assessing a variety of forcing scenarios. H. 69–91. P. 2007. IPCC. M. 2009. T. J. D. L.. Webster. J. The models of lesser complexity might have less accuracy. 2004. A. Alcamo. their computational cost is high. The Physical Science Basis. P. Maskell (eds. A. References Claussen. et al.J. The results indicate that the original model could be expanded further to introduce more realistic geometries and physical processes such as windstress forcing in order to simulate real ocean basins. Weber. I. M. Sexton. C. Harris. F. Goosse. P. United Kingdom and New York. Hawkins. B. M. and J. The modelling exercise employed in this thesis is based on continued attention to Stommel’s   saltadvective feedback mechanism. C.1029/2005GL023209.)]. M.   Mearns. Contribution of Working Group I to the Fourth Assessment . Giorgi. L. S. B. which is why the modelling community has turned to models of lesser complexity in order to assess feedbacks and long timescale forcing scenarios. J. B. the classical bistability of the simulated thermohaline circulation is kept. Goldstein. Earth System Models of Intermediate Complexity: Closing the Gap in the Spectrum of Climate System Models.. Allison. Rougier.... M. M. Gregory. Feedbacks and Forcings: a Comparison of Perturbed Physics and Multi-Model Ensembles. A. Sokolov. 2002.. NY.. Proceedings of the National Academy of Sciences of the United States of America 106: 2058420589. De Cuevas. Probabilistic Formulations for Transferring Inferences from Mathematical Models to Physical Systems. pp. S. Bistability of the Atlantic overturning circulation in a global climate model and links to ocean freshwater transport. In: Climate Change: The IPCC Scientific Assessment [Houghton.. EMICs have been shown to be. W. Climate Model Errors. Science: 295.. and J. Stone. R. F.. G.G. Climate Dynamics 18: 579-586.

A. M.. Schneider. DOI:10. O. An Integrated Assessment of Changes in the Thermohaline Circulation.J. Levermann. 2011. M. Crucifix. E. S. Hofmann. Aina. R. 1988. 471–523.. Taylor. A. Challenges in Combining Projections from Multiple Climate Models. USA. K. Selten. M. et al. L. Journal of the American Statistical Association 104: 97-117. B. M.. Investigating the Causes of the Response of the Thermohaline Circulation to Past and Future Climate Changes. Kettleborough. Geophysical Research Letters 31. Simulation of Abrupt Climate Change Induced by Fresh Water Input to the North Atlantic Ocean. Reviews of Geophysics 45: 8755-8785. Appenzeller. 2011. B. Thermohaline Convection with Two Stable Regimes of Flow.B. and A. Influence of Vertical Mixing on the Thermohaline Hysteresis: Analyses of an OGCM. Bulletin of the American Meteorological Society 88: 1383-1394. 2000. H. 1982. Furrer.. 2002. G. Whitehead.000 Years. R. R. 2005.. A. Influence of CO2 Emission Rates on the Stability of the Thermohaline Circulation. Kamenkovich. Geophysical Research Letters 38. C. S.. Coupled OceanAtmosphere Models with Flux Correction. What is the thermohaline circulation? Science 298: 1180-1181. 2007. Nature 378: 165–167.T. Evidence for Slow Mixing Across the Pycnocline from an Open Ocean Tracer Release Experiment.J. Estimates of Meridional Atmosphere and Ocean Heat Transports.. Mysak. Mearns.. The Thermohaline Ocean Circulation—A System with Dangerous Thresholds? Climate Change 46: 247–256. DOI:10. Rahmstorf. Nychka. McAvaney. D.J. Collins. Knutti. Uncertainty in Predictions of the Climate Response to Rising Levels of Greenhouse Gases. L.. Nature 419: 207–214. 2001: Model evaluation.. I. M. F.. NY. 2005. and R. Evaluation of Different Methods to Assess Model Projections of the Future Evolution of the Atlantic Meridional Overturning Circulation. Journal of Physical Oceanography 33: 1707-1721. 1997.. K. L. Lohmann. A.1007/s00382-011-1110-5. Nychka. Lohmann. M. Abe-Ouchi. C. M. Manabe. Mapping Model Agreement on Future Climate Projections. Stocker. R. C. 1993. A.. Hagedorn. P. 2007. 2000. Schmittner. W. M. C. Cambridge University Press. B. 2007. Journal of Physical Oceanography 39: 1231-1244.. L.. K. M. J. J. A. J.. Gregory. 862–865. Schneider. Ganopolski. Hydrology and Ocean Circulation.. A. et al. S. Nature: 388. C. S. E. Rooth. A. Progress in Oceanography. Latif. 1961. Thermohaline circulation hysteresis: A Model Intercomparison. Delworth. J. Proceedings of the National Academy of Sciences of the United States of America 97: 1347-1350. M. W. On the driving processes of the Atlantic meridional overturning circulation. pp. Tebaldi. Arblaster. G. T. Journal of Climate 23: 2739-2755. Geophysical Research Letters 37. J. Shutts. Hasselmann. B. Climate Model Genealogy. Weber SL. G. 2011. Model Projections of the North Atlantic Thermohaline Circulation for the 21st Century Assessed by Observations. Climate Calculations with a Combined OceanAtmosphere Model.. 2006. Stouffer. Dijkstra. T. K. Wunsch.. 2001. and J. The WCRP CMIP3 Multi-Model Dataset: A New Era in Climate Change Research. O. Drijfhout. R. Marsh. M. Vellinga.. A.. M. Meehl... R. Journal of Climate 18: 1524-1539. Stott. C. R.. R.. et al.. Prange. and M. Representing Model Uncertainty in Weather and Climate Prediction. Meehl. Law C. DOI:10. Journal of Climate 20: 843–855. J.. J. P. A. Z.. S. Climatic Change 96: 489-537. Allen. G. F. A. Sausen. Kuhlbrodt. R. Weigel. Dixon. Abrupt Climate Change and Thermohaline Circulation: Mechanisms and Predictability. Abrupt Transitions and Hysteresis in Thermohaline Laboratory Models.. R. H. M. J. F. Schmittner. Knutti. Rahmstorf. Hofmann. 2003. Schmittner. 2004. S. 2001. Branstator.1029/2005GL023655. M.C. Smith. 2002. B.1029/2011GL049863. S. 1969. Journal of Climate 23: 4175-4189. and K. C. T.. 2002. Latif. Cambridge University Press. (eds. R. J. R. P. J. A. S.J. Goosse. S. Journal of Climate 14: 3433– 3443. 2010. Ganopolski.. DOI: 10. Stouffer. Covey.Report of the Intergovernmental Panel on Climate Change. A. and T. 2004. M. 2005. 2010. Rahmstorf S. Journal of Climate 20: 2121-2133. A. Doblas-Reyes. Trenberth. United Kingdom and New York. T.1029/2005GL024368. Yin. M. Montaya.)]. 17 . S. G. McAvaney. F. C.. M. M. Tebaldi.. G. D. Weaver. J. United Kingdom: Chapters 8 and 10. Quantifying Uncertainty in Projections of Regional Climate Change: A Bayesian Approach to the Analysis of Multimodel Ensembles. Otto-Bliesner. Eby. R. and K. Sundby. What Might We Learn from Climate Forecasts? Proceedings of the National Academy of Sciences of the United States of America 99: 2487–2492. Knutti. Knutti. Crucifix. Stainforth. S. Paul. 2008. 11. In: Climate Change 2001: The Scientific Basis. DOI:10. J. 2009. Geophysical Research Letters 38. Rahmstorf. A. B. Griesel. 2009. Latif. The Modern and Glacial Overturning Circulation in the Atlantic Ocean in PMIP Coupled Model Simulations. G. R.. W..1029/2004GL020739.. Ocean Circulation and Climate During the Past 120. K. Kliphuis. Bondeau. Bryan. Caron. Watson A.1029/2011GL046864. Wang. Peltier. W. Jung. Raper. Smith. Climate Dynamics 17 : 601–613. Tebaldi. R. Tropical Origins for Recent and Future Northern Hemisphere Climate Change. S. B. Use of an UpwellingDiffusion Energy Balance Model to Simulate and Diagnose AOGCM Results. A. Liniger. Smith. 2007. B. Harris. et al. Manabe. Climate of the Past 3: 51–64. Annual Review of the Earth and Planetary Sciences 33: 163-193. Mitchell. Jaeger. J. Journal of Climate 19: 1365–1387. T. J. 1995. C. Murakami. L. Gregory. Cramer. Booth. Mearns. A. Tellus: 31. Nature 364:701–703. Palmer.. Masson. R.. Rahmstorf. Nature 433: 403–406. M. DOI:10. A. H.. Tebaldi.M. D. T. Climate Dynamics. Estimates of Uncertainty in Predictions of Global Mean Surface Temperature. Jackson.. L. C. B. N. A. Journal of the Athmospheric Sciences 26: 786789. 2005. Leutbecher. Cermak. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change [Houghton. A. F. Bayesian Modeling of Uncertainty in Ensembles of Climate Models. J. Stommel. Osborn. 224-230. Link. Vikebø. Zickfeld. Christensen. Knutti. Kuhlbrodt. B. Geophysical Research Letters 32. Climate Dynamics 2 : 145– 163. M. Stouffer. D. L. 131–149. Ledwell J. Risks of Weighting in Multimodel Climate Projections. R. Marotzke.. The Sensitivity of the Meridional Overturning Circulation to Modelling Uncertainty in a Perturbed Physics Ensemble Without Flux Adjustment. T. Barthel. C. 2007.