Future Climate Surprises - Elsevierscitechconnect.elsevier.com/.../02/Tim_Lenton1.pdf · Future Climate Surprises Tim Lenton College of Life and Environmental Sciences, University

Chapter 17

Future Climate Surprises

Tim LentonCollege of Life and Environmental Sciences, University of Exeter, Exeter, UK

Th

Co

Chapter Outline
17.1. Introduction: Probing Future Climates 489
17.2. Defining Climate Surprises
490
17.2.1. Tipping Points and Noise-Induced Transitions 490

17.2.2. Policy-Relevant Tipping Elements 492

17.3. Melting of Large Masses of Ice
493
17.3.1. Arctic Sea-Ice 493

17.3.2. Greenland Ice Sheet (GIS) 494

17.3.3. West Antarctic Ice Sheet (WAIS) 494

17.3.4. Yedoma Permafrost 494

17.3.5. Ocean Methane Hydrates 495

17.3.6. Himalayan Glaciers 495

17.4. Changes in Atmospheric and Oceanic Circulation
495
17.4.1. Indian Summer Monsoon (ISM) 495

17.4.2. El NinoeSouthern Oscillation (ENSO) 496

17.4.3. Atlantic Thermohaline Circulation (THC) 496

17.4.4. West African Monsoon (WAM) and

Sahel-Sahara 496

17.4.5. Southwest North America (SWNA) 497

17.5. Loss of Biomes
497
17.5.1. Amazon Rainforest 497

17.5.2. Boreal Forest 498

e Future of the World’s Climate. DOI: 10.1016/B978-0-12-386917-3.00017-8

pyright � 2012 Elsevier B.V. All rights reserved.

17.5.3. Coral Reefs 498

17.6. Coping with Climate Surprises
498
17.6.1. Risk Assessment 498

17.6.2. Removing the Element of Surprise 499

17.6.3. Early Warning of Bifurcations 500

17.6.4. Limitations on Early Warning 501

17.6.4.1. The Lack of Data Problem 502

17.6.4.2. The Lag Problem 502

17.6.4.3. The Noise Problem 502

17.6.5. Bifurcations in Noisy Systems 502

17.6.6. Application to Past Abrupt

Climate Changes 503

17.7. Future Climate: Surprises, Responses, and Recovery

Strategies
505
17.7.1. Mitigation 505

17.7.2. Geo-engineering 506

17.7.3. Rational Responses? 506

17.7.4. Recovery Prospects 507

17.8. Conclusion: Gaps in Knowledge
507
Acknowledgements
507
17.1. INTRODUCTION: PROBING FUTURECLIMATES

Fifteen years ago, when the first edition of this book waspublished (Henderson-Sellers, 1995), much scientificattention was directed at one source of climate surprise:changes in the Atlantic thermohaline circulation (THC;Peng, 1995). In the intervening time, scientists haveidentified many more systems that could produce climatesurprises (Lenton et al., 2008). Improvements to theobservational and palaeorecords have reinforced the viewthat climate can change abruptly at large scales.Furthermore, recent, striking developments in the climatesystem have added to the concern that human-inducedclimate change is unlikely to involve a smooth andentirely predictable transition into the future. The recordminimum area coverage of Arctic sea-ice in September2007 drew widespread attention, as has the accelerating

489

.

l

,

t

loss of mass from the Greenland and West Antarctic IceSheets (Pritchard et al., 2009; Rignot et al., 2008)Droughts have afflicted the Amazon rainforest (Phillipset al., 2009) and a massive insect outbreak has struckCanada’s boreal forest (Kurz et al., 2008b). These large-scale components of the Earth system are among thosethat have been identified as potential ‘tipping elements’ eclimate subsystems that could exhibit a ‘tipping point’where a small change in forcing (in particular, globatemperature change) causes a qualitative change in theirfuture state (Lenton et al., 2008). The resulting transitionmay be either abrupt or irreversible or, in the worst casesboth. In IPCC terms, such changes are referred to as‘large-scale discontinuities’ (Smith et al., 2009). Shouldthey occur, they would surely qualify as dangerousclimate changes (Schellnhuber et al., 2006). However, noall are equally dangerous. While, for the most part, theimpacts are clearly damaging and large, there is at least

http://dx.doi.org/10.1016/B978-0-12-386917-3.00017-8

490 SECTION j V Understanding the Unknowns

one case (greening of the Sahel) where the climatesurprise could be a pleasant one.

While the terminology of ‘climate surprises’ impliestheir predictability is limited (there is some irreducibleuncertainty), those discussed herein are not completelyunpredictable. The trigger of any tipping-point change islikely to be a combination of natural variability on top of anunderlying forcing due to human activities. Hence, one canonly talk in terms of probabilities of passing particulartipping points. However, recent expert elicitation hasobtained some useful information on these probabilities fordifferent future warming scenarios (Kriegler et al., 2009).The probabilities are imprecise but, even with the mostconservative assumptions, they indicate that in a 4�Cwarmer world it is more likely than not that at least one offive large-scale thresholds will be passed. The key messagefrom recent studies is that large climate surprises nowappearsignificantly closer, in terms of global temperature change,than they did in earlier assessments (Smith et al., 2009).

This chapter has the following aims: Firstly, itaddresses how to categorize climate surprises, reviewingexisting definitions of tipping elements and tipping points,and adding the case of noise-induced transitions. Next, thelist of potential policy-relevant tipping elements (Lentonet al., 2008) is revisited (and slightly revised) consideringthem in three categories: the melting of large masses ofice, changes in atmospheric and ocean circulation, andloss of biomes. Then, the discussion turns to how sciencecan help societies cope with climate surprises, startingwith how to assess the risk they pose, and then examiningin detail the prospects for early warning of them. Finally,the available response and recovery strategies areconsidered for societies faced with unwelcome climatesurprises.

17.2. DEFINING CLIMATE SURPRISES

The nature of surprise is that it is unexpected, and abrupt.So, ‘climate surprises’ is taken here to refer to events wherethere is a stochastic (i.e., random) component driving them,as well as a deterministic one, and where the resultingchanges are unexpectedly large, relative to the factorsdriving them. Surprises can be pleasant or unpleasant, butin the case of anthropogenic climate change it is usuallyassumed that changes are for the worst. Abrupt andunpredicted changes are seen as particularly undesirable,because they are most difficult to adapt to. In the followingsubsections, climate surprises are subdivided into thosewhere a system is forced past a ‘tipping point’ and thosewhere internal variability (noise) triggers a transition.The reason, as we will see later, is that there is someprospect of predicting ‘tipping points’ (because they havea deterministic component), whereas purely noise-induced

transitions are completely unpredictable surprises.However, the subdivision is over-idealized, and in realitya mixture of both factors is likely to be at work in futureclimate change surprises.

17.2.1. Tipping Points and Noise-InducedTransitions

In colloquial terms, the phrase ‘tipping point’ captures thenotion that “little things can make a big difference”(Gladwell, 2000, p. 1). In other words, at a particularmoment in time, a small change can have large, long-termconsequences for a system. To apply the term usefully to theclimate (or in any other scientific context), it is important tobe precise about what qualifies as a tipping point, and aboutthe class of systems that can undergo such change. To thisend, the term ‘tipping element’ has been introduced (Lentonet al., 2008) to describe large-scale subsystems (orcomponents) of the Earth system that can be switched eunder certain circumstances e into a qualitatively differentstate by small perturbations. In this context, the tippingpoint is the corresponding critical point e in forcing anda feature of the system e at which the future state of thesystem is qualitatively altered. For a system to possessa tipping point, there must be strong positive feedback in itsinternal dynamics (see Harvey, 2012, this volume).

To formalize the notion of a tipping element further(Lenton et al., 2008), it is important to define a spatial-scale; here, only components of the Earth system associatedwith a specific region or collection of regions, which are atleast subcontinental in scale (length-scale of order ~1000km), are considered. Then, for such a system to qualify asa tipping element, it must be possible to identify a singlecontrol parameter (r), for which there exists a criticalcontrol value (rcrit), from which a small perturbation(dr > 0) leads to a qualitative change in a crucial feature ofthe system (DF) after some observation time (T> 0). In thisdefinition (Lenton et al., 2008), the critical threshold (rcrit)is the tipping point, beyond which a qualitative changeoccurs e this change may occur immediately after thecause or much later.

Many scientists intuitively take ‘tipping point’ to besynonymous with a ‘bifurcation point’ in the equilibriumsolutions of a system, as schematically illustrated inFigure 17.1a. This implies that passing a tipping pointnecessarily carries some irreversibility. However, otherclasses of non-linear transition can meet the definitionabove, and one schematic example is given in Figure 17.1b.Again this shows the (time-independent) equilibriumsolutions of a system, but here they are continuous (there isno bifurcation) and, therefore, the transition is reversible.

In reality, the existence or not of a tipping point should beconsidered in a time-dependent fashion, and there could beseveral other possible types of tipping elements (for the

δρ

Control (ρ) ρcrit

ΔF

(a)

Control (ρ) ρcrit

δρ

ΔF

(b)

Syst

em fe

atur

e (F

)

Syst

em fe

atur

e (F

)

FIGURE 17.1 Two types of tipping point.

The schematics show the time-independent

equilibrium solutions of a system: (a)

a system with bi-stability passing a bifurca-

tion point, (b) a mono-stable system exhib-

iting highly non-linear change. These cases

meet the definition (Lenton et al., 2008) of

a tipping element passing a tipping point

(rcrit), where a small change in control (dr)

results in a large change in a system feature

(DF).

491Chapter | 17 Future Climate Surprises

mathematical details, see the supplementary informationof Lenton et al., 2008). Theoretically, one can constructelements that react infinitely slowly to tipping, yet do this inan entirely irreversible fashion. Also, recent work hasidentified examples of rate-dependent tipping: wherea system undergoes a large and rapid change, but onlywhen the rate at which it is forced exceeds a critical value(Levermann andBorn, 2007; SebastianWieczorek, personalcommunication, 2010).

A different class of climate surprise are noise-inducedtransitions, as illustrated in Figure 17.2. In such cases,internal variability causes a system to leave its current state(or attractor) and transition to a different state (or attractor).This does not require a bifurcation point to be passed,potentially it can occur without any change in forcing (thecontrol parameter, r). However, it does require the co-existence of multiple states under a given forcing, whichimplies that the underlying system possesses bifurcation-type tipping points. As illustrated in Figure 17.2b, onecan think of noise as pushing a system out of a valley(one stable steady state), up to the top of a hill (an unstable

Control (ρ) ρcrit

ΔF

Fcrit

(a)

F

ΔF

(b)

Syst

em fe

atur

e (F

)

Pote

ntia

l

System feature

δ

steady state), and it then rolling down the other side intoa different valley (some new stable state). One might callperching on the top of the hill (at the unstable steady state)a ‘tipping point’ and define it in terms of the correspondingvalue of the system feature (Fcrit). However, the value ofFcrit is a function of r (Figure 17.2a), so is not as well-defined as rcrit. Thus, noise-induced transitions do notstrictly fit the tipping point definition given above (Lentonet al., 2008), but they are related to it. Clearly, they area kind of surprise that could occur in the climate system andshould be considered, an example being abrupt monsoontransitions (Levermann et al., 2009).

In general, noise-induced transitions become morelikely to occur the closer one is to a bifurcation point. Thus,given that the climate system has its own internally-generated noise (familiar to us as the weather), we canexpect that, if it is approaching a bifurcation point, it willleave its present state before the bifurcation point is reached(Kleinen et al., 2003). This means that some future climatesurprises could involve a mixture of the idealized mecha-nisms shown in Figure 17.1a and Figure 17.2.

Fcrit (F)

Fδ

FIGURE 17.2 Noise-induced transition.

The schematics show a system with bist-

ability undergoing a noise-induced transi-

tion between states: (a) time-independent

equilibrium solutions (for comparison with

Figure 17.1), (b) representation of the tran-

sition within the system potential (y-axis of

(a) has become x-axis of (b)). In (b), the

wells represent stable steady states, the

hilltop represents an unstable steady state,

and the ball represents the actual state of the

system (filled black is the initial state). In

this case, a small perturbation due to noise

(dF), with no change in control (r), results

in a large change in a system feature (DF).

This suggests an alternative class of tipping

point (Fcrit).


17.2.2. Policy-Relevant Tipping Elements

The above definitions are quite general and couldconceivably be applied at any point in Earth’s climatehistory. However, the focus here is on future climatesurprises, so we need to narrow them down somewhat.Previous work (Lenton et al., 2008) has defined a subset of‘policy-relevant’ tipping elements by adding the followingconditions to our tipping element definition:

(i) Human activities are interfering with the system suchthat decisions taken within a ‘political time horizon’(TP ~ 100 years) can determine whether the tippingpoint (rcrit) is crossed.

(ii) The time to observe a qualitative change (includingthe time to trigger it) lies within an ‘ethical timehorizon’ (TE ~ 1000 years).

(iii) A significant number of people care about the fate ofthe system because either it contributes significantlyto the overall mode of operation of the Earth system,it contributes significantly to human welfare, or it hasgreat value in itself as a unique feature of thebiosphere.

FIGURE 17.3 Map of potential policy-relevant tipping elements overlain o

contents of this chapter. Question marks indicate systems whose status as po

Veronika Huber, Martin Wodinski, Timothy M. Lenton, and Hans-Joachim Sc

This definition focuses on the consequences of deci-sions enacted within this century that could lead to largechanges within this millennium. For a system to meet thedefinition of a tipping element, there needs to be sometheoretical basis for expecting it to exhibit a criticalthreshold at a subcontinental-scale and/or past evidence ofthreshold behaviour. To identify the subset of policy-rele-vant tipping elements, the conditions given above wereevaluated. For (i), the ‘accessible neighbourhood’ ofclimate out to 2100 was defined, by considering the rangeof IPCC Special Report on Emissions Scenarios (SRES)climate forcing factors and the range of resulting projectedclimate changes (IPCC, 2007a). To evaluate (ii), modelprojections and palaeo data were used, taking into accountknown shortcomings of the models. To evaluate (iii) inev-itably involves some subjective judgments. Figure 17.3shows the resulting map of the potential policy-relevanttipping elements in the climate system, updated somewhatfrom the one originally introduced (Lenton et al., 2008).

Before getting into the details of specific tippingelements, it is worth pausing to consider whether thoseidentified on the map (Figure 17.3) include all the systems

n global population density e adjusted from Lenton et al. (2008) based on

licy-relevant tipping elements is particularly uncertain. (Source: Figure by

hellnhuber.)


that might undergo noise-induced transitions in the futureand, if not, how they might be included? Conceivably therecould be systems that do not reach a bifurcation point(Figure 17.1a) due to anthropogenic climate change, butcan still be knocked out of their present state by a relativelyhigh degree of internal variability (Figure 17.2). TheHolocene climate is generally characterized as havingrelatively low internal variability (compared to, forexample, the ice age climate; cf. Berger and Yin, 2012, thisvolume). Consequently, it tends to be assumed that the‘signal’ of anthropogenic climate change will be largecompared to the ‘noise’ of internal variability. However,when one goes down to the subcontinental-scale, variabilityis much greater than in the global mean. (Furthermore,internal variability may itself vary with changes in themean climate state.) To include the possibility of noise-induced transitions, one could broaden the definition outsomewhat, by offering an alternative to condition (i):

(i-alt.) Internal variability within the ‘political timehorizon’ (TP ~ 100 years) could be sufficient to push thesystem past an unstable state (Fcrit) into a new basin ofattraction.

Strictly speaking, this criterion has not been applied incoming up with the following list of potential climatesurprises. However, both anthropogenic forcing andinternal climate variability could play a role in tippingseveral of the systems. In the following sections, potentialtipping elements are subdivided into those involvingmelting of large masses of ice, those involving changes inthe circulation of the atmosphere and ocean, and thoseinvolving the loss of unique biomes.

17.3. MELTING OF LARGE MASSES OF ICE

The concept of a threshold is intuitively obvious whenthinking about ice melting to liquid water e an example ofa first-order phase transition. However, that happens ona relatively small-scale. For major masses of ice on Earth toqualify as climate tipping elements, they must exhibita large-scale threshold due to strong positive feedbacks intheir internal dynamics, coupled to the climate.

17.3.1. Arctic Sea-Ice

The summer minimum area cover of Arctic sea-ice hasdeclined markedly in recent decades, most strikingly in2007. Observations have fallen below all IPCC modelprojections, despite the models having been in agreementwith the observations in the 1970s (Stroeve et al., 2007).Winter sea-ice is also declining in area (though lessrapidly), with a loss of 1.5 million km2 of multiyear icecoverage in the past decade (Nghiem et al., 2007). Thereis also an overall, progressive thinning of the ice cap,

with observations showing a decrease of mean wintermultiyear ice thickness from 3.6 m to 1.9 m over the pastthree decades (Kwok and Rothrock, 2009). The observeddecline in sea-ice is consistent with acceleration due tothe iceealbedo positive feedback, as exposure of thedark ocean surface causes increased absorption of solarradiation. This is warming the upper ocean andcontributing significantly to melting on the bottom of thesea-ice. Over 1979 to 2007, 85% of the Arctic region hasreceived an increase in solar heat input at the surface,with an increase of 5% per year in some regions,including the Beaufort Sea (Perovich et al., 2007). In situmeasurements in this region show that there was a threetimes greater bottom ice melt in 2007, compared toearlier years (with relatively little change in surfacemelt) (Perovich et al., 2008). Other factors contributingto record ice loss include patterns of atmosphericcirculation (Maslanik et al., 2007; Rigor and Wallace,2004) and ocean circulation (Nghiem et al., 2007), whichhave exported multiyear ice out of the Arctic basinthrough the Fram Strait, reductions in summertime cloudcover (Kay et al., 2008), and increased input of oceanheat from the Pacific (Shimada et al., 2006; Woodgateet al., 2006).

Further warming ‘in the pipeline’ raises the possi-bility that the Arctic may already be committed toa qualitative change in which the ocean becomes largelyice-free in summer (e.g., Harvey, 2012, this volume). Theyear that the North Pole becomes seasonally ice-free willlikely be seen as a ‘tipping point’ by non-experts. Whilepolitically important (cf. Taplin, 2012, this volume), it isunlikely that such a transition involves an irreversiblebifurcation (as in Figure 17.1a) (Eisenman andWettlaufer, 2009). Summer sea-ice quickly recovers oncethe climate turns cold again because of a stabilizingfeedback related to the ice growth rate (Notz, 2009). Yetit may still qualify as a tipping element because, as theice cap gets thinner, it becomes prone to larger fluctua-tions in area, which can be triggered by relatively smallchanges in forcing (Holland et al., 2006). Also, it isconceivable that loss of summer ice would involveadditional dynamical feedbacks that lead to a qualitativechange in atmosphere or ocean circulation and heattransports. If so, the impacts are likely to be felt furtherafield, for example in Europe. Loss of winter (i.e., year-round) ice is more likely to represent a bifurcation(Figure 17.1a), where the system can switch rapidly andirreversibly from one state (with seasonal ice) to another(without any) (Eisenman and Wettlaufer, 2009).However, the threshold for year-round ice loss requiresaround 13�C warming at the North Pole (Winton, 2006).Whether this could occur this century depends onanthropogenic emissions and the uncertain strength ofpolar amplification of warming.


17.3.2. Greenland Ice Sheet (GIS)

The Greenland ice sheet (GIS) is currently losing mass ata rate that has been accelerating (Rignot et al., 2008). Insummer 2007, there was an unprecedented increase insurface melt, mostly south of 70�N and also up the westside of Greenland, due to an up to 50-day longer meltseason than average with an earlier start (Mote, 2007). Thisis part of a longer-term trend of increasing melt extent sincethe 1970s. Recent observations show that seasonal surfacemelt has led to accelerated glacier flow (Joughin et al.,2008; van de Wal et al., 2008). The surface mass balance ofthe GIS is still positive (there is more incoming snowfallthan melt at the surface, on an annual average), but theoverall mass balance of the GIS is negative due to anincreased loss flux from calving of glaciers that outweighsthe positive surface mass balance. The margins of the GISare thinning at all latitudes (Pritchard et al., 2009), and therapid retreat of calving glaciers terminating in the ocean,most notably Jakobshavn Isbrae, is probably linked towarming ocean waters (Holland et al., 2008).

The GIS will be committed to irreversible meltdownif the surface mass balance goes negative, most notablybecause as the altitude of the surface declines, it getswarmer (a positive feedback). Initial assessments put thetemperature threshold for this to occur at around 3�C ofregional warming, based on a positive-degree-days modelfor the surface mass balance (Huybrechts and De Wolde,1999; Cogley, 2012, this volume). Results from an expertelicitation concur that if global warming exceeds 4�C, thereis a high probability of passing the threshold (Kriegleret al., 2009). An alternative surface energy balance modelpredicts a more distant threshold at 8�C regional warmingor ~6�C global warming (J. Bamber, personal communi-cation, 2010). However, recent work suggests the thresholdcould be much closer at 1.3�Ce2.3�C global warmingabove pre-industrial (A. Robinson and A. Ganopolski,personal communication, 2010). The actual threshold formassive GIS shrinkage must lie before the surface massbalance goes negative. A more nuanced possibility, whichis emerging from some coupled climateeice sheet modelstudies, is that there could be multiple stable states for GISvolume, and, hence, multiple tipping points (Ridley et al.,2009). Passing a first tipping point where the GIS retreatsonto land could lead to ~15% loss of the ice sheet and about1 m of global sea-level rise. As for the rate at which thiscould occur, an upper limit is that the GIS could contributearound 50 cm to global sea-level rise this century (Pfefferet al., 2008).

17.3.3. West Antarctic Ice Sheet (WAIS)

The West Antarctic Ice Sheet (WAIS) is also losing mass atpresent, and some parts, particularly those draining into the

Amundsen Sea, are thinning rapidly (Pritchard et al., 2009).While air temperatures have recently been shown to bewarming across West Antarctica (Steig et al., 2009), theshrinkage of the WAIS is more sensitive to the intrusion ofwarming ocean waters and the collapse of floating iceshelves that buttress the main ice sheet. The WAIS may bevulnerable to large-scale collapse, due to retreat of thegrounding line where the ice sheet is pinned to the bedrockbelow sea level (Mercer, 1978; Weertman, 1974). Havingbeen strongly questioned when it was first introduced, theparadigm of a potential abrupt collapse of the WAIS hasrecently gained new momentum (Vaughan, 2008). Recenttheory has confirmed the potential for multiple stable statesof the grounding line and, hence, bifurcation-type tippingpoints (Figure 17.1a) (Schoof, 2007). Also, new palaeo datahas shown that the WAIS collapsed repeatedly during the~3�C warmer world of the early Pliocene (5 Mae3 Ma)(Naish et al., 2009). Modelling supports this and suggestsfurther collapses during some (but not all) of the morerecent Pleistocene interglacials (Pollard and DeConto,2009). Furthermore, East Antarctic ice cores show anoma-lous spikes of warmth (above present) during all of the pastfour interglacial intervals, which might be explained byrepeated WAIS collapse (Holden et al., 2010). Data fromthe last (Eemian) interglacial suggest the up-to-2�C-warmerworld of the time may have had peaks of sea level up to 9 mhigher than present and rates of sea-level rise of 1.6m� 1.0mper century (Rohling et al., 2008). To achieve such rates ofsea-level rise probably required rapid grounding line retreatof the WAIS, and possibly parts of the periphery of the EastAntarctic Ice Sheet (EAIS) that are also grounded below sealevel. Current models put the threshold for WAIS collapsewhen the surrounding ocean warms by ~5�C (Pollard andDeConto, 2009), and expert elicitation concurs that if globalwarming exceeds 4�C, it is more likely than not that theWAIS will collapse (Kriegler et al., 2009).

17.3.4. Yedoma Permafrost

Continuous permafrost is the perennially frozen soil,which currently covers ~10.5 million km2 of the Arcticland surfaces but is melting rapidly in some regions. Thisarea could be reduced to as little as 1.0 million km2 by theyear 2100, which would represent a qualitative change instate (Lawrence and Slater, 2005; Harvey, 2012, thisvolume). However, permafrost did not make the originalshortlist of tipping elements (Lenton et al., 2008) becauseof a lack of evidence for a large-scale threshold forpermafrost melt. Instead, in future projections the localthreshold of freezing temperatures is exceeded at differenttimes in different localities. Yet more recent work hassuggested that at least one large area of permafrost couldexhibit coherent threshold behaviour. The frozen loess(windblown organic material) of northeastern Siberia


(150�Ee168�E and 63�Ne70�N), also known as yedoma,is deep (~25m) and has an extremely high carbon content(2%e5%); thus, it may contain ~500 PgC (Zimov et al.,2006). Recent studies have shown the potential for thisregional frozen carbon store to undergo self-sustainingcollapse, due to an internally-generated source of heatreleased by biochemical decomposition of the carbon,triggering further melting in a runaway positive feedback(Khvorostyanov et al., 2008a, 2008b). Once underway, thisprocess could release 2.0 PgCe2.8 PgC per year (mostly asCO2, but with some methane) over about a century,removing ~75% of the initial carbon stock. The collapsewould be irreversible in the sense that removing theforcing would not stop it continuing. To pass the tippingpoint requires an estimated >9�C of regional warming(Khvorostyanov et al., 2008a). However, this is a regionalready experiencing strongly amplified warming, partlylinked to shrinkage of the Arctic sea-ice (Lawrence et al.,2008). During AugusteOctober 2007, Arctic landtemperatures jumped around 3�C above the mean for thepreceding 30 years (from analysis of the HadCRUT data).Thus, the yedoma tipping point may be accessible thiscentury under high emissions scenarios as discussed inHarvey (2012, this volume).

17.3.5. Ocean Methane Hydrates

Recent model estimates suggest that up to 2000 PgC arestored as methane hydrates beneath the ocean floor (Archeret al., 2009). As the deep ocean warms, heat diffuses intothe sediment layer and may destabilize this reservoir offrozen methane. Bubbles associated with the melting ofmethane may trigger submarine landslides (Kayen and Lee,1991). This finding raises the concern that the destabili-zation of methane hydrates could result in an abruptmassive release of methane into the atmosphere. If thisscenario were plausible, methane hydrates would clearlyqualify as a policy-relevant tipping element. However,palaeoclimatic evidence makes this scenario very unlikely(Archer, 2007). Instead, the most likely impact of a meltinghydrate reservoir is a long-term chronic methane source(Archer et al., 2009; Harvey, 2012, this volume). Anadditional warming of 0.4�Ce0.5�C from the hydrateresponse to fossil fuel CO2 release is estimated, persistingover several millennia (Archer et al., 2009). This estimateis, however, subject to large uncertainties in particular withregard to the magnitude of temperature forcing required totrigger the destabilization of methane hydrates. Even ina ~1.5�C global warming scenario, ~2�C warmer ‘heatbubbles’ may persist at depth in the ocean for manycenturies (Schewe et al., 2010). In summary, a qualitativechange in this Earth subsystem is unlikely to occur ona policy-relevant timescale (as defined by Lenton et al.,2008). Yet, methane hydrates can be considered a slow and,

for societal purposes, irreversible tipping element in theglobal carbon cycle.

17.3.6. Himalayan Glaciers

It has been suggested that the HindueKusheHimalayaeTibetan (HKHT) glaciers should be added to the list oftipping elements because much of their mass could be lostthis century (Ramanathan and Feng, 2008). The loss ofmountain glaciers (Cogley, 2012, this volume) involvesa positive feedback whereby dust accumulation lowers thesurface albedo, thus accelerating melting (Oerlemans et al.,2009). Also, where snow or ice disappears altogether, thefurther lowering of albedo (iceealbedo feedback) amplifieswarming (Pepin and Lundquist, 2008). However, it needs tobe examined whether such positive feedbacks will causeHKHT mass loss to exhibit strong non-linearity in responseto warming, and therefore qualify as a climate surprise ortipping element.

17.4. CHANGES IN ATMOSPHERIC ANDOCEANIC CIRCULATION

The circulations of the ocean and atmosphere, coupledtogether and to the land surface, can exhibit differentdynamical stable states and modes of variability, withpotential thresholds between them (cf. Latif and Park,2012, this volume). They can also be particularly sensitiveto gradients of forcing as these are usually what drive thecirculations in the first place. Monsoons are a seasonalexample, initiated by more rapid heating of the land thanocean, which causes warm air to rise over the continent,creating a pressure gradient that sucks in moist air fromover the ocean, which then rises, its water condenses, andrain falls, releasing latent heat that reinforces the circula-tion (Levermann et al., 2009).

17.4.1. Indian Summer Monsoon (ISM)

The Indian Summer Monsoon (ISM) system is alreadybeing influenced by aerosol and greenhouse gas forcing.Palaeorecords indicate its volatility, with flips on and off ofmonsoonal rainfall linked to climate changes in the NorthAtlantic (Burns et al., 2003; Goswami et al., 2006; Guptaet al., 2003). Greenhouse warming, which is stronger overNorthern Hemisphere land than over the Indian Ocean,would on its own be expected to strengthen the monsoonalcirculation. However, the observational record showsdeclines in ISM rainfall, which have been linked to an‘atmospheric brown cloud’ (ABC) haze created bya mixture of black carbon (soot) and sulfate aerosols(Ramanathan et al., 2005). The ABC haze is moreconcentrated over the continent than over the ocean to thesouth, and it causes more sunlight to be absorbed in the


atmosphere and less heating at the surface. Hence, it tendsto weaken the monsoonal circulation (Meehl et al., 2008;Ramanathan and Carmichael, 2008). In simple models,there is a tipping point for the regional planetary albedo(reflectivity) over the continent which, if exceeded, causesthe ISM to collapse altogether (Levermann et al., 2009;Zickfeld et al., 2005). The real picture is likely to be morecomplex with the potential for switches in the strength andlocation of the monsoonal rains. Increasing aerosol forcingcould further weaken the monsoon but, if then removed,greenhouse warming could trigger a stronger monsoon,producing a climatic ‘roller-coaster ride’ for hundreds ofmillions of people (Zickfeld et al., 2005).

17.4.2. El NinoeSouthern Oscillation (ENSO)

The El NinoeSouthern Oscillation (ENSO) phenomenon isthe most significant natural mode of coupled oceaneatmosphere variability in the climate system. Over the pastcentury, warming has been greater in the eastern than thewestern equatorial Pacific and this has been linked to ElNino events becoming more severe (e.g., in 1983 and1998). Recently, a changing pattern of El Nino has beennoted towards ‘Modiki’ events where the warm pool shiftsfrom the West to the middle (rather than the East) of theequatorial Pacific (Ashok and Yamagata, 2009; Yeh et al.,2009). In future projections, the first coupled model studiespredicted a shift from current ENSO variability to morepersistent or frequent El Nino conditions. Now thatnumerous models have been inter-compared, there is noconsistent trend in frequency. However, in response toa stabilized 3�Ce6�C warmer climate, the most realisticmodels simulate increased El Nino amplitude (with nochange in frequency) (Guilyardi, 2006). Also, a shifttowards Modiki events has recently been forecast (Yehet al., 2009). Furthermore, palaeo data indicate differentENSO regimes under different climates of the past. Despitelarge persisting uncertainties, the probability of tippingpoint behaviour, in the sense that ENSO either vanishes orbecomes overly strong, is estimated to be rather low duringthe twenty-first century (Latif and Keenlyside, 2009, Latifand Park, 2012, this volume). The mechanisms and time-scale of any transition are unclear, but a gradual increase inEl Nino amplitude and/or a shift in location is consistentwith the recent observational record and would, neverthe-less, have severe impacts in many regions.

17.4.3. Atlantic Thermohaline Circulation(THC)

The archetypal example of a climate surprise is a reorga-nisation of the Atlantic THC, which is prone to collapsewhen sufficient freshwater enters the North Atlantic to haltdensity-driven deep water (NADW) formation there (Peng,

1995; Stommel, 1961). Modelling that minimizes artefactsarising from numerical diffusion shows that a hysteresis-type response to freshwater perturbations is a character-istic, robust feature of the THC (Hofmann and Rahmstorf,2009). However, the shutdown of the THC may actually beone of the more distant tipping points. Expert elicitationsuggests that THC collapse becomes as likely as not with>4�C warming this century (Kriegler et al., 2009). TheIPCC (2007a) views the threshold as even more remote, butrecent analysis suggests the AR4 models are systematicallybiased towards a stable THC (Drijfhout et al., 2010).Although a collapse of the THC may be one of the moredistant tipping points, a weakening of the THC this centuryis robustly predicted (IPCC, 2007a). This, in turn, will havesimilar, though smaller, effects as a total collapse. Apotential tipping point that occurs in some models isa switch of the subpolar gyre in which deep convection andNADW formation shuts off in the Labrador Sea region (tothe west of Greenland) and convection switches to onlyoccurring in the GreenlandeIcelandeNorwegian Seas(to the east of Greenland) (Born and Levermann, 2010;Levermann and Born, 2007). This would have dynamiceffects on sea level, increasing it down the eastern seaboardof the USA by around 25 cm in the regions of Boston, NewYork, and Washington DC (in addition to the global stericeffects of ocean warming) (Yin et al., 2009).

17.4.4. West African Monsoon (WAM)and Sahel-Sahara

Past intervals of severe drought in West Africa have beenlinked to weakening of the THC (Chang et al., 2008;Shanahan et al., 2009). This event seems to triggera phenomenon known as the Atlantic Nino (by analogywith El Nino events in the equatorial Pacific), involvingreduced stratification and warming of the sea surface in theGulf of Guinea. This disrupts the West African Monsoon(WAM), which is usually enhanced by the development ofa ‘cold tongue’ in the eastern equatorial Atlantic thatincreases the temperature contrast between the Gulf ofGuinea and the land to the north. In a typical year, there isalso a northward ‘jump’ of the monsoon into the Sahel inJuly, which corresponds to a rapid decrease in coastalrainfall and the establishment of the West African WesterlyJet in the atmosphere (Hagos and Cook, 2007). The jump isdue to a tipping point in atmospheric dynamics: when theeast/west wind changes sharply in the north/south direction,this instability causes the northward perturbation of an airparcel to generate additional northward flow (a strongpositive feedback).

It is not clear in which direction the WAMmight shift inthe future. The more benign alternative is a greening of theSahel-Sahara, by a mechanism that can be related toobservations of the seasonal northward shift of the WAM


(although this would restrict dust-borne nutrient supplyto the tropical Atlantic and the Amazon rainforest(Washington et al., 2009). The more dangerous option isa southward shift of the WAM. If ocean temperatureschange such that the West African Westerly Jet fails to formor is weakened below the tipping point needed to createinertial instability, then the rains may fail to move into thecontinental interior, drying the Sahel. Recent simulationssuggest a tipping point if the THC weakens below ~8 Sv,causing the subsurface North Brazil Current to reverse andan abrupt warming in the Gulf of Guinea (a persistentAtlantic Nino state). TheWAM then shifts such that there isa large reduction in rainfall in the Sahel and an increase inthe Gulf of Guinea and coastal regions (Chang et al., 2008).Such a transition is forecast in one of only three IPCC(2007a) models that produces a realistic present climate inthis region (Cook and Vizy, 2006). However, the other twomodels give conflicting responses: in one the Sahel getsmarkedly wetter despite a collapse of the WAM and in theother there is little net change.

17.4.5. Southwest North America (SWNA)

Increased humidity in a warmer world causes increasedmoisture divergence, changing global atmospheric circula-tione including poleward expansion of the Hadley cells andthe subtropical dry zones (Held and Soden, 2006; Lu et al.,2007), a development that tends to strongly reduce runoff inthese regions (Milly et al., 2005). One area that may beparticularly affected is southwest North America (SWNA),defined as all land in the region 125�We95�W and25�Ne40�N. Aridity in this domain is robustly predicted tointensify and persist in future and a transition is probablyalready underway: to something which has been describedas “.unlike any climate state we have seen in the instru-mental record” (Seager et al., 2007, p. 1183). Recently,increased SWNA aridity has been linked to the potential forincreased flooding in the Great Plains (Cook et al., 2008).The key driver is model-projected relatively higher summerwarming over land than over ocean (analogous to whatdrives seasonal monsoons). In simulations of futuredynamics, an increased contrast between the continental lowand the North Atlantic subtropical high strengthens the GreatPlains low-level jet, which transports moisture from theCaribbean to the upper Great Plains, triggering floodingthere but starving SWNA of moisture (Cook et al., 2008).However, it is unclear whether there is strong non-linearityin response to warming, and, therefore, whether drying ofSWNA qualifies as a climate surprise or tipping element.

17.5. LOSS OF BIOMES

At the ecosystem-scale there are probably many potentialthresholds related to climate change, the most obvious of

which is the ‘extinction’ (disappearance) of a unique typeof ecosystem, because it has nowhere to retreat to (forexample the Fynbos on the southern tip of Africa, orecosystems already high in mountains). Such changes areclearly a concern to policymakers, but it is not clear thatthey involve a dynamical threshold within the relevantecosystems. Also, here the focus is on the larger-scale ofwhat on land are called biomes. Biome tipping points cancome about due to local biophysical feedbacks that existbetween the land surface and climate (Claussen et al.,1999), with the Amazon rainforest and boreal forest beingleading candidates (Lenton et al., 2008). Although marineextinctions are clear in the geologic record followingclimate ‘catastrophes’ (e.g., Belcher andMander, 2012, thisvolume), relatively little attention has been directed atthe search for tipping elements in marine ‘biomes’ (or‘biogeographical provinces’).

17.5.1. Amazon Rainforest

A severe drought occurred from July to October in 2005 inthe western and southern parts of the Amazon basin, whichled the Brazilian government to declare a state of emer-gency. Despite initial ‘greening up’ of large areas of forest(Saleska et al., 2007), the 2005 drought made the Amazonregion a significant episodic carbon source, when otherwiseit has been a carbon sink (Phillips et al., 2009). The 2005drought has been linked to unusually warm sea surfacetemperatures in the North Atlantic (Cox et al., 2008).However, lengthening of the Amazon dry season is alsopart of a wider trend in seasonality, associated with weak-ening of the zonal tropical Pacific atmospheric circulationas attributed to anthropogenic greenhouse gas forcing(Vecchi et al., 2006). The trend of a lengthening dry seasonis forecast to continue, with one model predicting that the2005 drought will be the norm by 2025 (Cox et al., 2008). Ifdrying continues, several model studies have shown thepotential for significant dieback of up to ~70% of theAmazon rainforest by late this century, and its replacementby savanna and caatinga (mixed shrubland and grassland)(Cook and Vizy, 2008).

There are positive feedbacks related to the ways inwhich rainforests store and recycle water to the atmo-sphere, which could accelerate the Amazon demise:increased (decreased) forest cover leads to increased(decreased) precipitation and vice versa (Betts et al., 2004).Such a feedback opens the possibility of bistability andtipping point behaviour. Ecosystem disturbance processessuch as increased fire frequency and pest infestation couldalso amplify a transition initially driven by drought.Experts suggest Amazon dieback is more likely than not ifglobal warming exceeds 4�C (Kriegler et al., 2009).However, the Amazon rainforest may lag climate forcingsignificantly and, hence, it may be committed to some


dieback long before it is apparent. In one model, committeddieback begins at 1�C global warming above pre-industrial,even though it does not begin to be observed until globalwarming approaches 4�C (Jones et al., 2009). The existenceand extent of forecast Amazon dieback depends on thechoice of climate model (Salazar et al., 2007; Scholze et al.,2006) because not all global climate model projections givea regional, seasonal drying trend in the Amazon. A recentanalysis based on 19 GCMs has indicated that many modelstend to underestimate current rainfall and that, althoughdry-season water stress is likely to increase over thetwenty-first century, the rainfall regime of east Amazonia islikely to shift in the direction of seasonal forests, rather thansavanna (Malhi et al., 2009). Dieback is generally lesssensitive to the choice of vegetation model, but the directeffect of CO2 increasing the water use efficiency of vege-tation can have a strong effect of tending to shift thedieback threshold further away (P. M. Cox, personalcommunication, 2010).

17.5.2. Boreal Forest

The boreal forest in western Canada is currently sufferingfrom an invasion of mountain pine beetle that has causedwidespread tree mortality (Kurz et al., 2008b). This hasturned the nation’s forests from a carbon sink to a carbonsource (Kurz et al., 2008a). Fire frequencies have also beenincreasing across the boreal forest zone. In the future,widespread dieback has been predicted in at least onemodel, when regional temperatures reach around 7�C abovepresent, corresponding to around 3�C global warming.Expert elicitation concurs that above 4�C global warmingdieback becomes more likely than not (supplementaryinformation of Kriegler et al., 2009). The causal mecha-nisms include increasingly warm summers becoming toohot for the currently dominant tree species, increasedvulnerability to disease, decreased reproduction rates andmore frequent fires causing significantly higher mortality.The forest would be replaced over large areas by openwoodlands or grasslands, which would in turn amplifysummer warming and drying and increase fire frequency,producing a potentially strong positive feedback. (Awarmerfuture climate may also enable northward expansion of theboreal forest into tundra regions (Scholze et al., 2006; Sitchet al., 2008) but this is not forecast to involve thresholdbehaviour.)

17.5.3. Coral Reefs

Coral bleaching events, linked to ocean warming, havebecome much more widespread and detrimental in recentdecades, and marine biologists are talking about being ata ‘point of no return’ for tropical coral reefs (Veron et al.,2009). Ocean acidification, as a direct consequence of

rising atmospheric CO2 emissions, may also contribute tothreshold-like changes in coral reef ecosystems (Riebesellet al., 2009; Sen Gupta and McNeil, 2012, this volume).Cold-water corals that grow down to 3000 m depth will beparticularly vulnerable to acidification. They will be firstaffected as the saturation horizon of aragonite (a crystallineform of calcium carbonate) shallows because of oceanacidification. Once bathed in corrosive waters and under-saturated in aragonite, the skeletons and shells maydissolve and the reefs collapse. It has been estimated thatwith unabated CO2 emissions, 70% of the presently knowndeep-sea coral reef locations will be in corrosive waters bythe end of this century (Guinotte et al., 2006). However,whether large areas will reach a threshold together (andthus qualify as a tipping element), such as the Great BarrierReef, or the cold-water coral reef systems extending fromnorthern Norway to the west coast of Africa, warrantsfurther research.

17.6. COPING WITH CLIMATE SURPRISES

Having detailed several potential climate surprises, theoverriding question becomes ‘How should (climate) policyrespond?’ In human endeavours, the prospect of having todeal with unpleasant surprises e high-impact, relativelylow-probability events, including a strong element ofunpredictability e is not new. Think of earthquakes orhurricanes making landfall. Systems exist (albeit flawedones) for dealing with such events, and they hinge arounda risk management approach (e.g., Taplin, 2012, thisvolume). Although these are relatively short-timescale‘events’, some of the risk management principles may beusefully mapped over to climate tipping points. Risk, in theformal sense, is the product of the likelihood (or proba-bility) of something happening and its (negative) impact.So a meaningful risk assessment of tipping elements woulddemand careful assessment of the likelihood of passingvarious tipping points (under different forcing scenarios),as well as the associated impacts.

17.6.1. Risk Assessment

There already exists some information about the likelihoodof passing different tipping points as a function ofglobal temperature change. Results of a workshop andliterature review cover eight tipping elements (Lenton andSchellnhuber, 2007; Lenton et al., 2008), and a process ofexpert elicitation considered six of these under threedifferent future climate trajectories, and involved the elic-itation of imprecise probability statements from 52 experts(Kriegler et al., 2009). Useful results were obtained for fivetipping elements: the Greenland Ice Sheet, West AntarcticIce Sheet, ENSO, Amazon rainforest, and the THC. Theimprecise probability statements were then formally


combined to give lower-bound probabilities. These revealthat the likelihood of passing at least one of five tippingpoints rises from >16% under a midrange (2�Ce4�C)global warming corridor to >56% (i.e., more likely thannot) under a high-warming (>4�C) corridor. In Figure 17.4,the ‘burning embers’ diagram of Lenton and Schellnhuber(2007) is updated to summarize likelihoods as a functionof global warming, based on the expert elicitation resultsand recent literature.

There also exists some information on the impacts oftipping the different elements, but the gaps are larger andthis area needs more detailed work. A recent study hasarticulated the implications of four different tipping pointscenarios for the insurance sector (Lenton et al., 2009a),considering Amazon rainforest dieback, Indian SummerMonsoon disruption (coupled with melt of HKHT glaciers),a shift to a more arid climate in southwest North America(including loss of mountain snowpacks), and high-end sea-level rise from melting ice sheets with additional regionalsea-level rise along the northeastern seaboard of the USArelated to weakening of the THC. However, tipping pointimpacts will depend on human responses and are thusa more epistemologically contested area than assigninglikelihoods to events. The resulting ambiguity needs to bereduced if risk assessment is to be usefully pursued(Stirling, 2003).

With these caveats, let us offer an initial ‘straw man’illustration of how a tipping point risk assessment mightlook (Table 17.1). Here a scenario of partial mitigation ofgreenhouse gas emissions is assumed, leading to roughly3�C global warming by 2100. The focus is on tippingelements from the original shortlist (Lenton et al., 2008)where a threshold can be meaningfully linked to globaltemperature change (thus excluding the Indian SummerMonsoon). Likelihoods and relative impacts are assessedon a five-point scale: low, lowemedium,medium,mediumehigh, and high. Information on likelihood is taken from

Arc

tic s

umm

er s

ea-i

ce

Gre

enla

nd ic

e sh

eet

Am

azon

rai

nfor

est

Wes

t Ant

arci

tc ic

e sh

eet

Bor

eal f

ores

t

Atla

ntic

ther

moh

alin

e ci

rcul

atio

n

Glo

bal w

arm

ing

abov

e 20

00 (

°C)

0

1

2

3

4

5

6

FIGURE 17.4 Burning embers diagram summarizing current information

global warming. (Source: Updated from Lenton and Schellnhuber, 2007, base

discussed therein.)

a review of the literature (Lenton and Schellnhuber, 2007;Lenton et al., 2008) and, where available, expert elicitation(Kriegler et al., 2009). Impacts are considered in relativeterms based on an initial subjective judgment (noting thatmost tipping point impacts, if placed on an absolute scalecompared to other climate eventualities, would be high).Impacts depend on timescale and here the full ‘ethical timehorizon’ of 1000 years is considered (Lenton et al., 2008),assuming minimal discounting of impacts on futuregenerations. Likelihood and impacts are simply multipliedtogether to give a measure of risk, and a ranking emerges(Table 17.1).

What this simple tabulation readily illustrates are somefamiliar dilemmas for the would-be risk manager: rela-tively high-impact, low-probability events, such as WAMcollapse, come out with a similar risk to relatively over-impact, high-probability events, such as Arctic summersea-ice loss. However, what stand out are the high-impact,high-probability scenarios as a priority for risk manage-ment effort, in this case the melting of the Greenland icesheet, followed byWest Antarctic Ice Sheet collapse. Theserisks will be best managed by restricting the extent of futurewarming of these systems and their surrounding ocean. Ofcourse, this exercise would be better conducted witha wider team of experts and relevant stakeholders to get amore scientifically credible and socially legitimate assess-ment (cf. Taplin, 2012, this volume). It is simply hoped thatthe ranking in Table 17.1 encourages some thought andactivity in this area.

17.6.2. Removing the Element of Surprise

Faced with the risk of unpleasant climate surprises,perhaps the most useful information that science couldprovide to help societies cope is some early warning of anapproaching tipping point. There are several degrees ofearly warning, from simply identifying possible threats, to

El N

iño

Sout

hern

Osc

illat

ion

Wes

t Afr

ican

Mon

soon

Yed

oma

perm

afro

st

Yea

r 21

00 r

ange

(IP

CC

200

7)

0

0.5

1

Won’t happen

Certain

As likely as not

More likely than not

Less likely than not

on the likelihood of tipping different elements under different degrees of

d on expert elicitation results (Kriegler et al., 2009) and recent literature

TABLE 17.1 A Simple ‘Straw Man’ Example of Tipping Element Risk Assessment

Tipping Element

Likelihood of Passing

a Tipping Point (by 2100)

Relative Impact**

of Change in State (by 3000)

Risk Score

(likelihood � impact) Risk Ranking

Arctic summer sea-ice High Low 3 4

Greenland ice sheet MediumeHigh* High 7.5 1 (highest)

West Antarctic Ice Sheet Medium* High 6 2

Atlantic THC Low* MediumeHigh 2.5 6

ENSO Low* MediumeHigh 2.5 6

West African monsoon Low High 3 4

Amazon rainforest Medium* Medium 4 3

Boreal forest Low LoweMedium 1.5 8 (lowest)

*Likelihoods informed by expert elicitation.**Initial judgment of relative impacts is the subjective assessment of the author.


being able to forecast that a tipping point is imminent. Forother high-impact events, such as hurricanes or tsunamis,there are already quite sophisticated early-warningsystems in place, which climate policy could potentiallylearn from.

Where a potential tipping point threat has beenconvincingly identified, the challenge becomes ‘Can anyearly warning signs be detected before the threshold isbreached?’ Here the answer depends critically on the natureof the underlying tipping phenomenon. Bifurcation-typetipping points (Figure 17.1a) offer the best prospects forearly warning. In contrast, purely noise-induced transitions(Figure 17.2) are fundamentally unpredictable (Ditlevsenand Johnsen, 2010; Hastings and Wysham, 2010). Non-bifurcation type tipping points (Figure 17.1b) present anintermediate case; their response is expected to resemblebifurcation-type behaviour to a certain degree. The pros-pects for early warning of an approaching bifurcation arenow considered in more detail, before turning to thequestion of how to characterize systems with high levels ofnoise.

system beingforced past abifurcationpoint

FIGURE 17.5 Schematic representation of a system being forced past

a bifurcation point. The system’s response time to small perturbations, s, isrelated to the growing radius of the potential well. (Source: Figure by

Hermann Held, from Lenton et al., 2008. Copyright 2008 National

Academy of Science, U.S.A.)

17.6.3. Early Warning of Bifurcations

Physical systems that are approaching bifurcation pointsshow a nearly universal property of becoming more slug-gish in response to a perturbation (Scheffer et al., 2009;Wiesenfeld and McNamara, 1986; Wissel, 1984). Tovisualize this, picture the present state of a system as a ballin a curved potential well (attractor) that is being nudgedaround by some stochastic noise process, such as weather(Figure 17.5). The ball continually tends to roll backtowards the bottom of the welle its lowest potential stateeand the rate at which it rolls back is determined by the

curvature of the potential well. As the system is forcedtowards a bifurcation point, the potential well becomesflatter. Hence, the ball will roll back ever more sluggishly.At the bifurcation point, the potential becomes flat and theball is destined to roll off into some other state (alternativepotential well). Mathematically speaking, the leadingeigenvalue, which characterizes the rates of change aroundthe present equilibrium state, tends to zero as the bifurca-tion point is approached.

So, for those tipping elements that exhibit true bifur-cation points (e.g., Figure 17.1a), an observed slowingdown in timeseries data could provide a basis for earlywarning. This should be manifested as an increasingautocorrelation in the time-series data (in simple terms,each data point becomes more like the surrounding ones).Following this rationale, a method of examining thedecay rate to perturbations using a simple lag-1 autocor-relation function (ACF) was developed, averaging overshort-term variability in order to isolate the dynamics of the

-50

-40

-30

Tem

pera

ture

pro

xy

0.8

1

DFA

-indi

cato

r

-50000 -40000 -30000 -20000 -10000 0time [years]

0.9

0.95

1

ACF-

indi

cato

rs

with detrendingwithout detrending

500 points 500 points

FIGURE 17.6 A test of tipping point early warning methods applied to

the GISP2 Greenland ice core d18O proxy record of palaeotemperature,

following Livina and Lenton (2007) with the addition of ACF analysis. A

sliding window of 500 data points is used, and the data are non-uniform, as

indicated in the top panel. Results are plotted in the middle of the sliding

windows. The vertical dotted line in the bottom two panels indicates where

the analyses include the 500 points before the transition into the Holocene

marked in the top panel. The ACF method was applied with and without

the detrending method described in Dakos et al. (2008).


longest immanent timescale of a system (Held and Kleinen,2004). The approach was subsequently modified byusing detrended fluctuation analysis (DFA) to assess theproximity to a threshold from the power law exponentdescribing correlations in the timeseries data (Livina andLenton, 2007). At a critical threshold, the data becomehighly correlated across short and middle-range timescalesand the time series behaves as a random walk with uncor-related steps. Both methods need to span a sufficient timeinterval of data to capture what can be a very slow decayrate, and they suffer the problem that rapid forcing ofa system could alter the dynamics and override any slowingdown.

Model tests have shown that both early warningmethods work in principle, in simple (Dakos et al., 2008),intermediate complexity (Held and Kleinen, 2004; Livinaand Lenton, 2007), and fully three-dimensional (Lentonet al., 2009b) models. The challenge is to get the methods towork in practice, in the complex and noisy climate system.Initial tests found that the ending of the last ice agerecorded in the ice core data is detected as a critical tran-sition using the DFA method (Livina and Lenton, 2007).Subsequent work showed increasing autocorrelation ineight palaeoclimate timeseries approaching transitions,using the ACF method (Dakos et al., 2008).

In Figure 17.6, existing DFA analysis of a Greenlandice core record (Livina and Lenton, 2007) is comparedwith the ACF method (with or without detrending). Here,the data come from the GISP2 ice core and are the d18Ostable water isotope record, which is a proxy for past airtemperature at the ice-core site (and can also be influ-enced by changing water source temperatures andsnowfall seasonality). Increasing d18O corresponds towarming. The data in this case are sparse and unevenlyspaced (this is typical of many palaeo data records,e.g., Harrison and Bartlein, 2012, this volume) and are notinterpolated, since this can introduce a high degree ofcorrelation and overwrite any potential early warningsignal. Both approaches detect critical behaviour duringthe last deglaciation (the indicators approach or exceeda critical value of 1). However, using only the more sparsedata prior to the transition (to the left of the dotted verticalline in Figure 17.6, top panel) the upwards trends,indicative of slowing down, are rather weak (to the left ofthe dotted vertical lines in Figure 17.6, middle and bottompanels).

Recent work has emphasized that to get a reliablesignal of approaching bifurcation, one should monitorchanges in variance, as well as autocorrelation in the data(Ditlevsen and Johnsen, 2010). As a threshold isapproached and the potential becomes flatter (Figure 17.5),one intuitively expects the variance to go up (i.e., the ballto make greater departures from the local stable state). If atthe same time the system is slowing down, one must be

careful to choose a long enough time window to accuratelysample the variance. The resulting changes in variancemay provide a statistically more robust and earlier warningsignal than changes in autocorrelation (Ditlevsen andJohnsen, 2010). However, in the climate system, whichcontains many sources of inertia or ‘memory’, the twoproperties can be expected to change together. Otherpotential early warning indicators are being explored forecological tipping points and could potentially be appliedto climate. These include increasing skewness of responses(Biggs et al., 2009; Guttal and Jayaprakash, 2008), spatialvariance, and spatial skewnesses (Guttal and Jayaprakash,2009).

17.6.4. Limitations on Early Warning

It is encouraging that there is some theoretical scope forearly warning of an approaching bifurcation, but there are


considerable practical limitations on whether an effectiveearly-warning system could be deployed for specificsystems. Here three problems are highlighted.

17.6.4.1. The Lack of Data Problem

A key consideration is ‘What is the longest internal time-scale of the system in question?’ It is changes in this thatthe bifurcation early warning method is trying to detect. Inthe case of the ocean circulation or ice sheet dynamics,these timescales are long (in the thousands of years).Therefore, one needs a long and relatively high-resolution(palaeo) timeseries record for the system in question, inorder to get an accurate picture of its natural state of vari-ability from which to detect changes. Often such recordsare lacking. However, some potential tipping points havemuch faster dynamics and relatively little internal memory,for example the monsoons. For such systems, existingobservational timeseries data may be sufficient. Also, forspecific tipping elements, such as the THC, there may beother leading indicators of vulnerability that are deduciblefrom observational data (Drijfhout et al., 2010).

17.6.4.2. The Lag Problem

If a tipping element is forced slowly (keeping it in quasi-equilibrium), proximity to a threshold may be inferred ina model-independent way. However, humans are forcingthe climate system relatively rapidly, so inherently ‘slow’tipping elements, such as ice sheets and the THC, will bewell out of equilibrium with the forcing (e.g., Harvey, 2012,this volume). This means that a dynamical model simu-lating transient behaviour will also be needed to establishproximity to a threshold.

17.6.4.3. The Noise Problem

Noise (or internal variability) in a system may be such thatit does not allow the detection of any trend towards slowingdown. For example, in the case of the THC there are nowseveral years of direct observations showing that itsstrength exhibits high internal variability (Latif and Park,2012, this volume). Where internal variability is high, atipping element could exit its present state well beforea bifurcation point is reached (Figure 17.2a). Hence, fora method of anticipating a threshold to be useful, the time ittakes to find out its proximity to a threshold must be shorterthan the time in which noise would be expected to cause thesystem to change statee the ‘mean first exit time’ (Kleinenet al., 2003). A sophisticated early warning system shouldtake account of the noise level for a particular tippingelement and adjust its estimates accordingly (Thompsonand Sieber, 2010).

In systems with a high level of noise, flickering betweenstates (i.e., noise-induced transitions in both directions)may occur prior to a more permanent transition (Bakke

et al., 2009). We now turn to consider whether, in suchnoisy systems, one can detect bifurcations.

17.6.5. Bifurcations in Noisy Systems

Although individual noise-induced transitions in theclimate system are inherently unpredictable, if a system isexperiencing a relatively high level of noise and samplingseveral different states (or modes of operation), then inprinciple one can deduce how many states it has, as well astheir relative stability or instability. To do so successfullyrequires a sufficiently long timeseries that all availablestates are being sampled. If the number of states and/or thestability of states changes over time then in principle thiscan also be detected. However, to do so requires a longwindow that is moving through an even longer timeseries.These ideas are at the heart of a recently developed methodcalled ‘potential analysis’ (Livina et al., 2010, 2011). Whilethe ‘slowing down’ method described above assumes thata system subject to noise is contained within one potentialwell, and looks for signals that the corresponding state isbecoming unstable, ‘potential analysis’ assumes thata system is sampling a number of potential wells (e.g.,Figure 17.2b) and tries to deduce the number of wells andtheir stability properties.

Potential analysis assumes that the dynamics of achosen climate variable can be described by a stochasticdifferential equation (i.e., one that includes a noise term),with an underlying potential (i.e., series of wells and hill-tops) whose shape can be described by a polynomialequation. The chosen stochastic differential equation hasa corresponding FokkerePlanck equation, describing theprobability density function; crucially, this has a stationarysolution that depends only on the underlying potentialfunction and the noise level. This gives a one-to-onecorrespondence between the potential and the stationaryprobability density of the system (the potential is directlyproportional to minus the logarithm of the probabilitydensity, scaled by the square of the noise level). This allowsthe underlying potential to be reconstructed, given theprobability density of a stretch of timeseries data and anestimate of the noise level. For the mathematics, see Livinaet al. (2011).

The method is illustrated in Figure 17.7. It starts bytransforming a window of timeseries data into a histogramof the data. Next, this is converted into an empiricalprobability density of the data using a standard Gaussiankernel estimator. If the system has a single, stable state, theresulting distribution should have a single mode withsmooth sides. Any deviations from this immediatelyprovide a visual clue as to the existence of other states. Inthe example, the probability density has a distinct‘shoulder’ suggesting the existence of a second state. Next,the probability density is inverted and log-transformed.

-40000 -35000 -30000

-46

-44

-42

-40

-38

data

-46 -44 -42 -40 -380

5

10

15

20

25

30histogram

-4 -20

0.05

0.1

0.15

0.2

0.25empirical probability density

-4 -2

transformed probability density

2

3

4

-6 -4 -2 4 6

polynomial fit

2

4

6fit of 2nd orderfit of 4th orderfit of 6th order

transformed kernel

0 2 4

0 2 40 0 2-2

reconstructed potential

0

2 4

FIGURE 17.7 Illustration of the method of potential analysis for a window of timeseries data (here NGRIP d18O data 40e30 ka BP). The steps are:

obtain histogram of the data; convert into empirical probability density (using a standard Gaussian kernel estimator); transform this distribution (take

natural logarithm, invert, and scale by noise level); least-squares fit the transformed distribution with polynomial functions of increasing even order and

select the highest order before encountering a negative leading coefficient; accurately determine the coefficients using the Unscented Kalman Filter; and

reconstruct the potential.


The method then attempts to least-square fit the trans-formed distribution with polynomial functions ofincreasing, even order (starting with second order, i.e.,a quadratic equation). At some point, the least-square fitreturns a negative leading coefficient, which is not physi-cally reasonable (this happens for a sixth-order fit in theexample). So, the polynomial of highest degree before thisis encountered is taken as the most appropriate represen-tation of the probability density of the timeseries (fourth-order in the example in Figure 17.7).

The number of states in the system is then determinedfrom the number of inflection points in the fitted poly-nomial potential (and the results of this are plotted inFigure 17.8, using a colour scheme). The simplest potentialhas a single state with no inflection points. Each pair ofinflection points corresponds to an additional state. Thisapproach picks up real minima (wells) in the potential, orjust flattening of the potential. The latter, importantly, aredegenerate states corresponding to bifurcation points. Ina final step, having determined its order, the coefficients ofthe potential can be accurately estimated using theUnscented Kalman Filter (UKF; Kwasniok and Lohmann,2009). For this to work well, the noise level must beaccurately estimated, which can be done separately usinga wavelet denoising routine that separates the signal intothe potential and the noise (Livina et al., 2011). The endresult is a reconstructed potential.

17.6.6. Application to Past Abrupt ClimateChanges

In principle, the method of potential analysis just describedcan detect bifurcations in a noisy climate system. Theirtiming can never be precisely tied down, because themethod always has to work with a relatively long timewindow of data (typically of order 1000 data points).However, by moving a sliding window through a longtimeseries, one should be able to detect changes in thenumber of states over time.

To test this, potential analysis has been applied toa classic case study of abrupt climate change in Earth’srecent past: the DansgaardeOeschger events (or ‘DeOevents’ for short) (Livina et al., 2010). These were rapidclimate changes that occurred repeatedly during the lastice age, were concentrated in the North Atlantic region buthad widespread effects, and are recorded most strikingly inGreenland ice cores. Figures 17.8a and 17.8c show twosuch ice core records: the GRIP (Dansgaard et al., 1993)and NGRIP (NGRIP, 2004) d18O stable water isotoperecords, which are a proxy for past air temperature at theice-core sites, which are 325 km apart. The DeO eventsare the abrupt increases in d18O (corresponding towarming).

What do these DeO events reveal about the nature ofpast climate surprises? The precise mechanism for what

FIGURE 17.8 Ice-core d18O proxy

records of palaeotemperature over the past

60 ka at two different sites 325 km apart in

Greenland: (a) GRIP (Dansgaard et al.,

1993) and (c) NGRIP (NGRIP, 2004); both

at 20-year resolution and on the most

recent GICC05 timescale (Svensson et al.,

2008). Contour plots of the number of

detected states in the records are a function

of time and sliding window length (results

plotted at the midpoints of the sliding

windows), for (b) GRIP and (d) NGRIP,

where; red ¼ 1 state, green ¼ 2, cyan ¼ 3,

purple¼ 4. This shows the loss of a second

climate state (green-to-red transition

across a wide range of window lengths)

around 25 ka BP. (Source: Livina et al.,

2010.)


was going on in the climate system continues to be debated(Colin de Verdiere, 2006), although most studies assigna key role to changes in the Atlantic THC, coupled tochanges in sea-ice cover. Regardless of the underlyingmechanism, recent work has shown that the repeatedtransitions from cold ‘stadial’ to warm ‘interstadial’ statescan be well-described by a model of purely noise-inducedtransitions (Ditlevsen et al., 2005) (Figure 17.2). Recentanalysis suggests that the cold stadial state remains stableduring the last ice age and does not experience a bifurcation(Ditlevsen and Johnsen, 2010). However, when lookingacross the interval 70 kae20 ka BP, the warm state is

characterized as being only marginally stable (Kwasniokand Lohmann, 2009).

On examining this more closely using the method ofpotential analysis (Figure 17.8b and 17.8d), the ice corerecords are best characterized as having two states from 60ka BP to about 25 ka BP, but only one during the depths ofthe Last Glacial Maximum (LGM) (Livina et al., 2010). Itis inferred that a bifurcation occurred (Figure 17.1a)sometime prior to 25 ka BP, in which the warm interstadialstate became unstable, and later it disappeared altogether.Figure 17.9 shows this bifurcation occurring in recon-structed potentials from the NGRIP data. It is reflected in

−2 0 2 4

0

10

20

30

z

U(z

)

40−30 ka35−25 ka30−20 ka

FIGURE 17.9 Potential curves reconstructed from the NGRIP d18O

palaeotemperature proxy data: for the intervals 40 kae30 ka (blue line, as

in Figure 17.7), 35 kae25 ka (red line), and 30 kae20 ka (black line). Here

‘z’ represents anomalies in NGRIP d18O and U(z) is the derived potential,

which shows the steady states in the system and describes the rate and

direction of response of ‘z’ to deviations from steady state (according to

dz/dt ¼ edU/dz). The results show a bifurcation in which the warm

interstadial climate state becomes degenerate (unstable) and then disap-

pears, at some time prior to 25 ka. (Source: Livina et al., 2011. With kind

permission from Springer Scienceþ Business Media)


the original records as the warm events becomingprogressively shorter, until they are very short-lived indeed,and then they cease through the LGM (Figures 17.8a and17.8c). In the case of a bifurcation such as this, whichreduces the number of system states, detection (the switchfrom green to red in Figures 17.8b and 17.8d) typicallyoccurs after the potential has become degenerate anda bifurcation has occurred (Figure 17.9). That is becausea system with a second state that has lost its stability, butnot disappeared, altogether gives a ‘shoulder’ in the prob-ability density and is still best described by a fourth-orderpolynomial (e.g., the reconstructed potential for NGRIP35 kae25 ka BP in Figure 17.9). However, this means that,conversely, in a system that is gaining an extra state, themethod may pick this up before that state becomes stable(i.e., bifurcation occurs) and, hence, provide some earlywarning that a new climate state is appearing. Currentlypalaeo and observational timeseries data are being analysedto search for such cases where potential analysis indicatesa new climate state is appearing (Livina et al., 2011).

The simple model of DeO events just described does notaccount for all the features in the ice core records, inparticular, the gradual cooling that typically follows anabrupt warming (an asymmetrical response) (Figures 17.8aand 17.8c). Furthermore, it does not account for variations inthe noise level (i.e., internal variability) between the cold andthe warm states (it appears to be considerably greater in thecold state). Hence, future work can be expected to produce

more advanced models. Still, to summarize our presentunderstanding, the archetypal example of past abrupt climatechanges illustrates that the climate system has undergoneboth noise-induced transitions and bifurcations in the past.These two types of climate surprise have very differentimplications for whether individual events are predictable.

17.7. FUTURE CLIMATE: SURPRISES,RESPONSES, AND RECOVERY STRATEGIES

Assuming science could provide society with reliable earlywarning of an approaching tipping point, and that thisinformation was viewed as sufficiently credible, salient,and legitimate to warrant action, the question becomes,‘What could (or should) we do about it?’ Obvious coursesof action are to try and avoid reaching the tipping point, orto try and ‘pre-adapt’ (build resilience) to better cope withthe changes that are due to occur. Which is feasible or mostappropriate will depend somewhat on the tipping elementin question and the forcing factors responsible, but let usstart with some fairly general observations.

The natural response to warning of the prospect ofa high-impact event is to try to avoid it. However, ouroptions for effective intervention (‘steering’) in the climatesystem are actually somewhat limited. That is because thereare several sources of inertia in the Earth system e onemight liken heading towards a climate tipping point as likethe Titanic heading towards the fateful iceberg: it is a bigship and it is hard to change its trajectory quickly. Ananalogous problem of avoiding an approaching tippingpoint in an ecological system (a fishery) shows that oncethere is a reliable early warning of an approaching tippingpoint, it is probably too late for slow intervention methodsto avoid it (Biggs et al., 2009).

17.7.1. Mitigation

The conventional avoidance strategy is to mitigate theemissions of greenhouse gases, especially carbon dioxide,but this is a slow intervention. The climate is responding tothe concentration of these agents (and the resulting radia-tive forcing), not their emissions (Harvey, 2012, thisvolume). The prime mitigation target, CO2, is a very long-lived pollutant and the current policy challenge is stillframed as stopping it rising and holding a stabilizationlevel, not bringing its concentration down. Even if radiativeforcing is stabilized, the climate will continue to warm forsome considerable time because heat will still be enteringthe system (primarily the deep ocean) and slow positivefeedbacks will be operating (such as loss of methane fromhydrates, e.g., Harvey, 2012, this volume).

Potentially, radiative forcing could be reduced bymitigating emissions of short-lived greenhouse gases (e.g.,tropospheric ozone) and black carbon (BC) aerosols, whose


concentrations will decline fairly rapidly. Ozone in thelower atmosphere currently contributes 20% as much asCO2 to global warming. Rigorous global implementation ofair pollution regulations and available technologies coulddrive down emissions of ozone precursors (especiallycarbon monoxide and nitrogen oxides) quickly, producinga climate response within decades (Molina et al., 2009). BCis estimated to be the second or third largest globalwarming agent, although large uncertainties about its exactradiative forcing exist. Due to its deposition on snow andresulting positive feedbacks, BC may be responsible for asmuch as ~0.5�Ce1.4�C of the 1.9�C warming observed inthe Arctic from 1890 to 2007 (Shindell and Faluvegi, 2009)and for approximately 0.6�C of the 1�C warming in theTibetan Himalayas since the 1950 (Ramanathan andCarmichael, 2008). Halving BC emissions could be ach-ieved by 2030 with full application of existing technologies(Cofala et al., 2007). Lowering BC emissions would havethe ‘double benefit’ of reducing global warming and raisingthe temperature level at which tipping points involvingmelting of ice and snow are triggered (Hansen andNazarenko, 2004). A further ‘fast-action’ strategy toconsider is phasing down the production and consumptionof hydrofluorocarbon (HFCs) with high global warmingpotential (Molina et al., 2009).

The bulk of mitigation will always be a relatively slowway of turning the climate ship; the inertia in society when itcomes to replacing energy and built infrastructure has noteven been mentioned. The steering might be speeded up bystarting to actively remove CO2 (or other positive radiativeforcing agents) from the atmosphere (Lenton and Vaughan,2009). However, it will still take timescales of half a centurybefore CO2 concentration can be stabilized and a centurybefore temperature stops rising (Lenton, 2010). A compre-hensive analysis of the complex control problem involved isgiven elsewhere (Schellnhuber et al., 2009), based on thelimitation of cumulative CO2 emissions worldwide.

17.7.2. Geo-engineering

So, are there any faster avoidance strategies available? Thisraises the controversial topic of geo-engineering theamount of sunlight absorbed by the Earth, with the aim ofdeliberately counteracting the positive radiative forcingcoming primarily from accumulated greenhouse gases(Lenton and Vaughan, 2009). Some proponents of this typeof geo-engineering are arguing that it should be developedas a potential emergency response to the prospect of anapproaching climate tipping point. For their argument tohold up, one would need to be confident that such inter-vention could work fast enough to avoid the transgressionof a threshold. The currently much-discussed method ofinjecting aerosols (probably of sulfate) globally into thestratosphere (Crutzen, 2006) could conceivably rebalance

radiative fluxes at the tropopause, and could begin to coolthe climate within a year, if it mimicked the Mount Pina-tubo volcanic eruption. However, it would take severalyears of repeated injection to feel the full climate coolingeffect (as the temperature of the ocean mixed layeradjusted) (Wigley, 2006). To this must be added the time todevelop the technology and the infrastructure needed todeploy it, which is probably at least a decade at present. So,we are probably talking about 10e20 years for this type ofintervention to take effect. In principle, this could behelpful in avoiding an approaching temperature thresholdin a relatively sluggish system such as the Greenland icesheet. Perhaps a tailor-made regional version of radiationmanagement e ‘regio-engineering’ (H. J. Schellnhuber,personal communication, 2010) e would be the preferableoption under these circumstances.

However, geo-engineering is not going to be much usefor fast response systems or ones where the threshold is notclearly linked to global temperature, such as the westAfrican or Indian monsoons. Indeed, past volcanic aerosolinjections are known to have slowed down the hydrologicalcycle and reduced rainfall in such regions (Robock et al.,2008; Trenberth and Dai, 2007) (and the theory behind thisis robust). So, aerosol geo-engineering interventions mightpose a greater risk to some tipping elements than thereduction in risk they achieved for others. Once again,a careful risk management approach is needed to evaluatecosts, benefits, and likelihoods.

17.7.3. Rational Responses?

In a rational world, any early warning would be useful,even if it turns out to be impossible to avoid an approachingtipping point, because it would give societies some time toprepare themselves. Effective adaptation can certainlyhappen faster than mitigation can alter the climate trajec-tory, and some types of adaptation can probably happenfaster than geo-engineering could alter the climate trajec-tory. However, these may be types of adaptation, such asmass migration, that carry their own considerable risks oftriggering undesirable social ‘tipping points’, that is,conflicts rather than co-operative responses. Research on‘social tipping elements’ is urgently needed to betteranticipate this type of dynamics (Schellnhuber, 2009).

Furthermore, there is the problem that humans are notperfectly rational actors. Hence, receiving an alarm signalcarries the danger of triggering maladaptive responses,especially when our fallible, internal, human methods ofrisk assessment are at play. Such maladaptive responses toearly warning of approaching climate tipping points cannotbe ruled out (Travis, 2010). Indeed, despite the knownincidence of hurricane landfalls, there has been a demo-graphic trend of people moving to Florida. Ironically, UScitizens have also historically been leaving the central


Great Plains and moving to the southwest, which is one ofthe regions highlighted as undergoing a transition to a newclimate state less able to support agriculture and people.

17.7.4. Recovery Prospects

If societies experience climate surprises, and the corre-sponding tipping elements change state, is there any pros-pect for recovering their original state? Again the answerwill depend on the system in question, and the timescale.The definition of a tipping element includes systems thatcan exhibit reversible or irreversible changes in state(Figure 17.1). A reversible transition means that, if theforcing is returned below the tipping point, the system willrecover its original state (either abruptly or gradually). Anirreversible transition means that it will not e it takesa larger change in forcing to recover (and hence there issome hysteresis in the trajectory of the system in phasespace). An example of a transition that should be reversiblein principle is the loss of Arctic summer sea-ice; whereastransitions that could exhibit some irreversibility are theloss of large ice sheets, changes in the Atlantic THC and/orthe loss of major biomes.

Even if a transition is reversible in principle, it does notmean that the changes will be reversible in practice. A keyproblem (discussed above) is that it is rather difficult toreduce radiative forcing of the climate system, unless oneindulges in geo-engineering, and the climate further lagsthe forcing (the problem of ‘committed climate change’).In other words, global warming is hard to reverse. Even lossof Arctic summer sea-ice, which seems like a highlyreversible system because the ice regrows each winter, hassome longer-term ‘memory’. It cannot be completelyrecovered in one season because some of the ice that hasbeen lost consisted of thick, multiyear strata that takeseveral winters to accumulate.

Irreversible transitions vary in what is needed to recoverfrom them, and even with respect to whether recovery ispossible at all. The ‘strongest’ form of irreversibility isextinction e the loss of species and, hence, genetic infor-mation and diversity (e.g., Belcher and Mander, 2012, thisvolume). Extinctions would surely accompany majorbiome transitions, such as dieback of the Amazon rainforestor boreal forests, should they occur. Thus, althoughsomething resembling the current rainforest or boreal forestmight eventually grow back, they would never be the same.Ice sheets, such as those on Greenland andWest Antarctica,if lost, might eventually be recovered with appropriateforcing, but it would take timescales of tens of thousands ofyears to regrow a major ice sheet. Alternatively, humanactivities might fundamentally alter the dynamics of theclimate system, switching it out of its recent mode ofroughly 100,000-year glacial cycles and back into an earlier

(Pliocene-like) state in which Greenland lives up to itsname and remains un-glaciated, West Antarctica onlysporadically has an ice sheet, East Antarctica holds less ice,and sea levels are in the long term around 25 metres higher(Rohling et al., 2009).

17.8. CONCLUSION: GAPS INKNOWLEDGE

Existing work has probably not identified all possible futuresurprises in the climate system, so more research is defi-nitely warranted to systematically search for them. Thelatest methods for detecting threshold behaviour could beapplied to a wide range of palaeo data, as well as to theinstrumental record, and to the output of existing climatemodel runs, such as from the IPCC Assessment Reports. Auseful theoretical starting point would be to try to identifyall the potentially strong positive feedbacks in componentsof the Earth system that could be manifested at largespatial-scales e because these are a necessary condition fortipping point behaviour. Historically, climate science hasbeen good at identifying global-scale positive feedbacks ontemperature. However, the focus here is on intermediate(but still large) spatial-scales, and about feedbacks that areinternal to the dynamics of a part of the system, andsometimes have little or no effect on global temperature(although they may be triggered by it changing). In recentyears, at least one such feedback has been discovered (thepotential for runaway breakdown of yedoma permafrost),and others have been better formalized (the multiple stablestates of ice sheet grounding lines). Once identified, suchfeedbacks should then be included in Earth system models,and the phase space of the models systematically searchedfor multiple stable states at the regional-scale and othersigns of strong non-linearity. There are recent examples ofthe successful detection of multiple states in complexmodels, for example in the Amazon basin (Oyama andNobre, 2003). Further effort is encouraged in this area,although, with state-of-the-art models, it will challengecurrent computing resources.

ACKNOWLEDGEMENTS

I thank John Schellnhuber for first encouraging me to workon climate tipping points and for input to this chapter.Veronika Huber and Martin Wodinski producedFigure 17.3, with input from John, and I. Hermann Heldoriginally produced Figure 17.5. Valerie Livina performedthe analyses in, and provided, Figures 17.6, 17.8, and 17.9,and helped with Figure 17.7. Anders Levermann, VeronikaHuber, John Schellnhuber, and an anonymous refereeprovided helpful reviews.

Documents

Future Climate Surprises - Elsevierscitechconnect.elsevier.com/.../02/Tim_Lenton1.pdf · Future Climate Surprises Tim Lenton College of Life and Environmental Sciences, University