1 Introduction
1.1 Climate tipping elements
Climate change can cause abrupt and irreversible environmental and societal change . Several climate subsystems have been identified at risk of undergoing qualitative and often irreversible change when critical thresholds of global warming are transgressed . Such subsystems are termed tipping elements (TEs), and examples include the Atlantic Meridional Overturning Circulation (AMOC), polar ice sheets, tropical rainforests, permafrost regions, and the marine biosphere. Nonlinear changes can occur at tipping points (TPs), where a slight change in a parameter or a small perturbation of a system's state can cause a large change in the system, driving it to transition into a completely different (often undesirable) state. From a dynamical system point of view, a tipping point can be reached when passing a critical value of a control parameter, for example the atmospheric CO concentration, which affects the equilibrium states of the system.
In the context of this paper, we refer to a tipping element as any climate subsystem that shows threshold behavior (at its so-called tipping point) beyond which self-amplifying feedbacks to forcing reorganize the system qualitatively (e.g., from an ice-covered to an ice-free state in Greenland). This means that, at the tipping point, a small shift in the background climate can trigger a large-scale qualitative system change. Once the tipping element is triggered, its self-amplifying feedbacks dominate the dynamics of the tipping element during the tipping process. This definition is taken from and includes large-scale climate tipping elements such as the AMOC and polar ice sheets , where the associated feedbacks (e.g., salt–advection, melt–elevation, or ice–albedo) are well known. Our definition also includes more regional bistabilities between savanna and forest vegetation in the Amazon region. In addition, we also consider elements that can show nonlinear behavior but it is speculated whether they should be considered tipping elements (El Niño–Southern Oscillation: ENSO, Arctic sea ice, and Indian summer monsoon in this paper). These entities are important due to their connections to tipping elements and for Earth system stability but also due to their potentially dramatic consequences on regional and local scales including impacts on the biosphere and human societies. We therefore do not restrict ourselves to the most plausible tipping elements in this review, but also include nonlinear components like Arctic sea ice, ENSO, and monsoon systems that can act as mediators of tipping events in the Earth system.
Tipping processes involving several tipping elements can also be found in Earth's history: during the last ice age, repeated abrupt shifts, so-called Dansgaard–Oeschger events, occurred between cold and warm phases lasting 1000–4000 years . While mostly polar and Northern Hemisphere elements (sea ice, ocean circulation, atmospheric dynamics) appear to have been involved , the climate impact of these shifts was global .
1.2 Interactions in the Earth's climate system
Most climate subsystems are linked via circulation systems in the ocean and atmosphere, which leads to statistical associations between them in their natural variability, often called teleconnections. For example, ENSO, the monsoon systems, and Atlantic multi-decadal variability form global modes of climate variability . In addition, sea surface temperature variability in the North Pacific coupled to tropical variability (ENSO) is transferred to other regions via atmospheric teleconnections and amplified on longer timescales by the large ocean heat capacity . Similarly, multi-decadal variability originating in the North Atlantic Ocean , which is believed to be partly connected to the AMOC , has a global expression in sea surface temperature patterns due to the interaction of slow oceanic and fast (but large-scale) atmospheric processes . Hence most tipping elements and other nonlinear components (such as ENSO or Arctic sea ice) are not isolated form each other but are either connected directly or mediated via changes to the background state . Via such connections between tipping elements (see Fig. 1), tipping in one subsystem – the leading subsystem – can therefore cause tipping in another one – the following subsystem . Here we call the linkages between tipping elements and/or other nonlinear components tipping interactions, whether they have a stabilizing or a destabilizing effect. The most extreme case is the situation in which the tipping of element A causes a subsequent tipping of element B. In this paper, we define a sequence of events involving several nonlinear components of the Earth system as tipping cascades . These tipping cascades can come in various forms dependent on the ordering of tipping elements
For example, a disintegration of the Greenland Ice Sheet can lead to an abrupt AMOC shift, while an abrupt change in AMOC strength can lead to an intensification of ENSO. We do not restrict our definition to specific spatial scales, timescales, or severity of impact of the tipping elements. Therefore, the slow local invasion fronts in spatial (eco)systems would also be considered, leading to a rapid local change in response to the slow system-wide and large-scale changes
Due to the many nonlinearities in the climate system, it is also conceivable that components of the Earth system, though not necessarily tipping elements on their own, could mediate or amplify nonlinear transitions in one component, also creating larger-scale impacts in other components. A prominent example is Arctic summer sea ice cover, which shows an almost linear response to the CO forcing and is not expected to show tipping behavior under anthropogenic forcing . Nevertheless it can still sharpen and amplify transitions in the ocean–atmosphere–cryosphere system . Note, however, that Arctic winter sea ice extent does show very different states at the same atmospheric CO concentration and may be considered a tipping element . Because Arctic summer and winter sea ice interacts with the AMOC, we take both of them into account as one node in our network.
On the other hand, an abrupt transition in one tipping element may also stabilize other climate subsystems
1.3 Motivation and structure of this work
While most TEs that have been proposed so far are clearly regional (with some being large-scale), there are significant knowledge gaps with respect to their tipping probability, impact estimates, timescales, and interactions. The main reasons for these knowledge gaps and therefore large uncertainties, are among others, (i) that experiments from comprehensive and process-based models are sparse (however, future projects plan to systematically increase the number of such experiments such as TIPMIP
See
The main part of this paper reviews the current knowledge of interactions between specific pairs of tipping components (Sect. 2). In Sect. 3, we discuss three paleoclimate candidates of tipping sequences that involve more than one tipping element: one from the more distant past (Eocene–Oligocene transition; approximately 34 million years ago), one from the more recent past (Dansgaard–Oeschger events, Bølling–Allerød warm period, and Heinrich events; during and since the last glacial period), and a paleoclimatic perspective on interactions between the AMOC and Amazon rainforest. Next, we map out the present state of modeling tipping sequences with respect to the role of complex Earth system models and more conceptualized approaches (Sect. 4). Lastly, we discuss current research gaps and ways forward from a knowledge, modeling, and data perspective. We also discuss the value of newly arising methods from machine learning and Earth observation and how they could complement the present research on interacting climate tipping elements. Finally, we conclude with recent progress on tipping cascades and interactions between tipping elements (both Sect. 5). Each of the sections (Sects. 1–5 and Sects. 2.1–2.7, 3.1–3.3) in this paper have been written by a different group of experts in the respective field (the co-authors of this study).
2 Interactions between climate tipping elements and nonlinear climate components
2.1 Interactions across scales in space and time
In this section, we lay out the current state of the literature on the interaction processes between components that are known to show nonlinear behavior or are even suspected tipping elements. The summary of Sect. 2.2–2.8 is shown in Table 1 and Figs. 1, 2, and 3. We show that these elements are not isolated entities but interact across the entire globe (Fig. 1). Not only do the interactions span global distances, but the elements themselves are also systems of the sub-continental up to (nearly) global spatial scale that may tip on temporal scales of months up to millennia; i.e., tipping elements interact across scales in space and time (Fig. 2) . The respective processes of the interactions can be found in Table 1 (summarized in Fig. 3), alongside an estimation of the interaction direction and, if available, an estimation of their strength (based on the detailed literature review of Sect. 2.2–2.8).
Some tipping elements are of sub-continental spatial scale (e.g., coral reefs or the Greenland Ice Sheet), while others cover significant portions of the globe (e.g., AMOC). Also, the temporal scales differ vastly among the different climate tipping elements: some of them are considered fast tipping elements once a tipping process has been initiated (tipping on the order of months, to years or decades, to centuries, e.g., Amazon rainforest and AMOC), while others are considered slow tipping elements (tipping on the order of centuries to millennia, e.g., Greenland Ice Sheet). These individual dynamics in space and time of the individual tipping elements are therefore also important for their interactions (mapped out in Fig. 2).
Table 1
List of links between elements (tipping elements and other nonlinear components) discussed in this paper. We list directed links with a short summary of the main physical mechanism(s) underlying the specific connection. We estimate whether the effect is stabilizing or destabilizing the impacted element. Because uncertainties in tipping linkages are large, we assess the uncertainty and the level of confidence following IPCC guidelines using the agreement among different studies as well as the overall evidence . If there is incipient evidence in the literature, we also add an indication of the qualitative strength of the response (weak, moderate, strong). Otherwise the link and/or its strength is set as unclear. Lastly, key references for the specific link are noted. Abbreviations: AABW – Antarctic Bottom Water, AMAZ – Amazon rainforest, AMOC – Atlantic Meridional Overturning Circulation, ENSO – El Niño–Southern Oscillation, GIS – Greenland Ice Sheet, ISM – Indian summer monsoon, PERM – permafrost, SST – sea surface temperature, WAIS – West Antarctic Ice Sheet, WAM – West African monsoon.
| Link between elements | Physical processes | Response(destabilizing,stabilizing,unclear) (strength) | Level of agreement(level of evidence) | Key references |
|---|---|---|---|---|
| GIS AMOC(Sect. 2.2.1) | Freshwater influx from GIS into the North Atlantic weakens the AMOC (consistent across models), uncertain how much freshwater flux is necessary to shut down AMOC | Destabilizing (strong) | High agreement(robust evidence) | , , |
| WAIS AMOC(Sect. 2.2.1) | Competing effects: (i) freshwater from the Southern Ocean and associated AABW reduction imply a strengthening of the AMOC (deep-ocean adjustment, likely on short timescales (years to decades) through wave dynamics; (ii) freshwater from the Southern Ocean reaches North Atlantic and weakens the AMOC (likely on timescales from decades to centuries); (iii) increasing wind intensity over Southern Hemisphere and sea ice cover over Southern Ocean can strengthen the AMOC | Unclear, likelytimescale-dependent(weak/moderate) | Low agreement(medium evidence) | , , , |
| AMOC GIS(Sect. 2.2.2) | Reduced northward heat transport in the Atlantic implies substantial cooling of the Northern Hemisphere | Stabilizing (strong) | High agreement(robust evidence) | , |
| AMOC WAIS(Sect. 2.2.2) | Reduced northward heat transport in the Atlantic may lead to warming of the Southern Ocean, destabilizing WAIS | Destabilizing (unclear) | Medium agreement(limited/mediumevidence) | , |
| GIS WAIS(Sect. 2.2.3) | Sea level rise through GIS melt destabilizes WAIS, potentially significantly since most of the WAIS bedrock lies below sea level (marine ice sheet instability) | Destabilizing(moderate) | High agreement(limited/mediumevidence) | , |
| WAIS GIS(Sect. 2.2.3) | Sea level rise through WAIS melt destabilizes GIS, potentially weakly since most of the GIS bedrock lies above sea level | Destabilizing (weak) | Medium agreement(limited evidence) | , |
Continued.
| Link between elements | Physical processes | Response(destabilizing,stabilizing,unclear) (strength) | Level of agreement(level of evidence) | Key references |
|---|---|---|---|---|
| Arctic sea ice AMOC(Sect. 2.3.1) | Studies modeling contemporary climate find that AMOC weakens as a result of Arctic sea ice decline and due to southward advection of lighter water to the North Atlantic convection sites (reasons: 1. strong anomalous heat flux into the ocean due to lowering albedo and 2. a small and transient contribution of freshwater influx from melting Arctic sea ice) | Destabilizing(weak/moderate) | Medium agreement(medium evidence) | , , , |
| AMOC Arctic sea ice(Sect. 2.3.1) | A weaker AMOC results in lower Atlantic Ocean heat transport and slows the pace of the current Arctic sea ice loss | Stabilizing(weak/moderate) | Medium agreement(medium evidence) | , , |
| Arctic sea ice GIS/PERM(Sect. 2.3.2) | Disappearance of Arctic summer sea ice leads to 0.3–0.5 C of additional regional warming over Greenland and the permafrost region; regional warming seems mostly relevant for coastal parts of Greenland – also, the coastal permafrost region is most affected by Arctic sea ice retreat due to increased erosion | Destabilizing (weak) | Medium agreement(very limited evidence) | , , , |
| AMOC AMAZ(Sect. 2.4) | A collapsing AMOC would reduce (increase) SSTs in the North (South) Atlantic and cause a southward shift of the tropical rain belt; therefore, precipitation would decrease in the northern part of the Amazon and increase in the southern part, but exact locations of precipitation changes are model-dependent | Unclear (moderate) | Medium agreement(medium evidence) | , , , |
| AMOC ENSO(Sect. 2.5.1) | A weakening AMOC increases the Equator-to-pole temperature gradient and may therefore strengthen the trade winds in general; feedbacks generating ENSO depend on the Pacific background climate, i.e., on trade wind strength – therefore, ENSO might be found more often in a positive condition | Destabilizing (weak) | Medium agreement(limited evidence) | , , |
| ENSO AMOC(Sect. 2.5.1) | Enhanced El Niño can decrease Atlantic hurricane intensity and so limit wind-induced AMOC forcing | Unclear (unclear) | Low agreement(limited evidence) | , |
Continued.
| Link between elements | Physical processes | Response(destabilizing,stabilizing,unclear) (strength) | Level of agreement(level of evidence) | Key references |
|---|---|---|---|---|
| ENSO AMAZ(Sect. 2.5.2) | Changes in ENSO amplitude and frequency will affect rainfall and temperature variability (including extreme conditions such as droughts) in various regions of the South American continent; this may regionally lead to rainforest dieback and transition to degraded savanna-like ecosystems – regions in the Andes and western Amazonia seem less vulnerable | Destabilizing (strong,regionally different) | High agreement(robust evidence) | , |
| ENSO WAIS(Sect. 2.5.3) | More frequent (or stronger) El Niño events cause more frequent atmospheric blocking situations, leading to West Antarctic ice melt events; this is followed by warm marine air anomalies over West Antarctica, especially over Ross and Amundsen Sea embayment regions | Destabilizing (unclear,probably weak/moderate) | Medium agreement(medium evidence) | , , |
| ENSO coral reefs(Sect. 2.5.4) | ENSO drives abnormally high ocean temperatures, which are superimposed on already warming oceans, leading to severe regional to global bleaching events; with increased warming, bleaching is decoupling from warm ENSO phases, becoming increasingly frequent during cold ENSO phases (La Niña) and at quasi-annual frequencies at many locations; at thresholds of 1.5 and 2 C warming, the expected die-off percentage is 70 %–90 % and 90 %–99 %, respectively | Destabilizing (strong) | High agreement(robust evidence) | , , |
| AMOC/ENSO monsoonsystems (Sect. 2.6) | An AMOC weakening would cause a southward shift of the Intertropical Convergence Zone and the tropical rain belt (e.g., WAM and ISM); ENSO shifts in a warmer climate could lead to an opposing effect on its relationships with the monsoon systems – for example, the ISM–ENSO relationship becomes weaker, while the relationship between WAM and ENSO becomes stronger with the warming climate | Unclear (monsoonsystem dependent)(unclear) | Low agreement(very limited evidence) | , , |
| PERM hydrologicalcycle AMOC(Sect. 2.7) | Under global warming, the PERM region could become wetter and lakes could emerge; if these lakes drain out, the global hydrological cycle becomes weaker, and then the AMOC weakens | Destabilizing forAMOC (unclear) | Low agreement(very limited evidence) | , |
Figure 1
Interactions between tipping elements on a world map. All tipping elements discussed in this review article are shown together with their potential connections. The causal interactions links can have stabilizing (blue), destabilizing (red), or unclear (gray) effects. For some elements, it is speculative whether they are tipping elements on their own (such as ENSO or Arctic sea ice) and they are denoted as such (blue outer ring) but they are included as nonlinear Earth system components if they play an important role in mediating transitions towards (or from) core tipping elements (compare Sect. 1.2). Tipping elements that exert a notable feedback on global mean temperature when they tip are denoted by a red inner ring. This temperature feedback can be positive (i.e., amplifying warming, as likely for the permafrost, the Arctic sea ice, the Greenland and West Antarctic ice sheets, the Amazon rainforest, and ENSO) or negative (i.e., dampening warming, as likely for the AMOC).
[Figure omitted. See PDF]
Figure 2
Interactions between tipping elements across scales in space and time. Temporal scales are transitioning times of a disintegrating tipping element from months up to millennia. Spatial scales denote the system size from sub-continental to (nearly) global scales. Transitioning times are taken from , and spatial scales from . The causal links can be stabilizing (blue), destabilizing (red), or unclear (gray). Some tipping elements are particularly speculative (such as ENSO or the Arctic sea ice) and denoted as nonlinear Earth System components (blue border). Tipping elements that exert a feedback on the global mean temperature when they tip are depicted with an additional red inner circle.
[Figure omitted. See PDF]
Figure 3
Interactions between tipping elements and other nonlinear components. Summary matrix of interactions in Figs. 1 and 2. More details on the specific interactions can be found in Table 1 and Sect. 2.2–2.7. Columns denote the element from which the interaction originates, and rows denote the tipping element which is affected by the interaction. We separate three different types of effects: a stabilizing link (blue box), a destabilizing link (red box), and an unclear or competing link (gray box). White boxes denote no (or an unknown) link. The strengths of the links are separated into four groups: strong (S), moderate (M), weak (W), and unclear (U) if a strength estimate is lacking.
[Figure omitted. See PDF]
2.2 Interactions between ice sheets and AMOCThe AMOC, Greenland Ice Sheet (GIS), and West Antarctic Ice Sheet (WAIS) are core tipping elements and are threatened by increasing CO emissions . Moreover, GIS, AMOC, and WAIS interact on very different timescales ranging from decades to multiple centuries. While some of those links might be stabilizing, others are destabilizing and would allow for the possibility of large-scale cascading events.
2.2.1 Effects of disintegrating ice sheets on AMOC
Greenland Ice Sheet to AMOC. The AMOC depends on the formation of dense water in the high latitudes of the North Atlantic. In its present state, this process is widely sustained by the positive salt–advection feedback – as the AMOC transports salt northward, a higher surface water density is maintained in this region. As GIS melting increases, the associated discharge of freshwater in the ocean would result in a decrease in the surface water density, inhibiting the formation of dense waters through deep convection and thereby weakening the circulation. As less salt is transported to the North Atlantic, the salt–advection feedback implies a self-sustained freshening of the high latitudes of the North Atlantic, which, in the worst case, can result in the collapse of the AMOC. On top of this classical positive feedback, there is a wide range of other feedbacks related to the AMOC, either negative (heat advection feedback, e.g., ) or positive (evaporation feedback). An overall destabilizing impact of GIS melting on the AMOC is mostly consistent across models, where adding freshwater in the North Atlantic
West Antarctic Ice Sheet to AMOC. In the case of a freshwater release in the Southern Hemisphere originating from West Antarctica, different opposing processes are at play that could affect the AMOC. These effects have been identified to act on different timescales and depend on the state of the circulation . First, the weakening of AABW formation might lead to enhancement of the AMOC through the so-called ocean bipolar seesaw related to deep-ocean adjustment through oceanic large-scale waves . Second, the increase in wind intensity over the Southern Hemisphere, related to an increase in sea ice cover , might also help to enhance the AMOC. Third, if large enough, the release of freshwater in the Southern Ocean might eventually reach the North Atlantic on a longer timescale (centuries), possibly weakening the AMOC. As a result, the impact of a WAIS collapse on the AMOC is still unclear, as most models show either a slight weakening
2.2.2 Effects of a collapsing AMOC on ice sheets
An AMOC collapse would imply decreased northward heat transport, leading to a substantial cooling of the Northern Hemisphere, along with warming in the Southern Hemisphere . Cooling the high latitudes of the North Atlantic would stabilize the GIS, possibly allowing for a safe overshoot of the GIS tipping point . Conversely, the related warming of the Southern Ocean represents a destabilizing impact on the WAIS, being susceptible to these warmer ocean waters via the ice shelves and their buttressing effect on upstream ice flow .
2.2.3 Direct interactions between Greenland and West Antarctic ice sheets via sea level feedbacks
It is known that an increase in sea level has an overall destabilizing influence on marine-based sectors of ice sheets, possibly triggering or enhancing the retreat of their grounding line . In the case of an ice sheet collapse, the induced sea level rise from ice sheets would vary locally depending on gravitational effects, rotational effects, and mantle deformation . Overall, sea level rise is expected to negatively impact both the GIS and WAIS, but more strongly the latter where most of the bedrock lies well below sea level .
2.3 Arctic sea ice interactions
2.3.1 Interactions between AMOC and Arctic sea ice
The strength of the AMOC is controlled by the deep convective activity at different sites in the North Atlantic Ocean (Labrador, Irminger, and Nordic seas), which is largely driven by its high surface density .
Changing Arctic sea ice cover can modulate the latter, and thus the AMOC, mainly in two ways . First, it alters radiative heating and ocean–atmosphere heat loss via changing albedo. More precisely, as the Arctic sea ice area has substantially decreased in the past 40 years, especially during summer months , the open-water fraction of the Arctic Ocean has increased and will continue to do so in the future . This has led to an increase in the absorption of solar radiation and to subsequent ocean warming, which can propagate to convection areas. Second, changes in Arctic sea ice alter the ocean density by brine rejection during sea ice formation or conversely by freshening from sea ice melt. In particular, the recent decrease in Arctic sea ice area, together with the ice loss from the Greenland Ice Sheet (Sect. 2.2.1), has added freshwater to the Arctic Ocean, although the trend in freshwater content has slowed down during the past decade . According to model simulations performed by , these warm and fresh anomalies coming from sea ice melting could propagate southward to the subpolar North Atlantic Ocean. These would affect the deep ocean at the main convection sites by reducing the surface density and would thus weaken the AMOC. The estimated timescale of this propagation is multi-decadal .
The AMOC can also affect Arctic sea ice via the transport of warm water to the North Atlantic Ocean and subsequently to the Arctic Ocean via the Barents Sea opening and Fram Strait. A weaker AMOC could result in lower ocean heat transport and increased Arctic sea ice area . The estimated timescale of this effect is approximately 1 year . However, recent observations show that the ocean heat transport to the Arctic has increased, especially on the Atlantic side . Thus, the effect of a decreasing AMOC may merely slow down the pace of ongoing increases in ocean heat transport and the associated decrease in Arctic sea ice . Despite this tight link between the AMOC and Arctic sea ice, Arctic summer sea ice cannot be considered a tipping element. A tipping point in the Arctic Ocean would mean that it loses so much sea ice that the reduced albedo results in enough warming to prevent sea ice from forming again once melted. However, according to model simulations in which a summer ice-free Arctic Ocean is simulated, Arctic sea ice recovers within 2 years, suggesting that the ice–albedo feedback is alleviated by large-scale recovery mechanisms . Winter sea ice extent does show very different states at the same atmospheric CO concentration, attributed to AMOC strength, and may therefore be considered a tipping element .
2.3.2 Effect of Arctic sea ice on the Greenland Ice Sheet and permafrost
Besides an interaction with AMOC, melting Arctic sea ice cover has a direct effect via regional warming on further high-latitude tipping elements such as the Greenland Ice Sheet and the permafrost. In the case of Arctic summer sea ice loss, which may occur during the second half of this century , additional warming levels caused by this loss are on the order of 0.3–0.5 C regionally over Greenland and the permafrost . Regional warming levels may be higher if Arctic winter sea ice additionally disappears. Further, it has been found that partial Arctic sea ice loss has a limited effect for Greenland warming patterns and is mainly relevant for coastal parts of Greenland .
At the same time, Arctic sea ice loss leads to increased coastal permafrost erosion , which operates as follows: (1) abrupt changes in summer–autumn sea ice retreat from the permafrost coast lead to (2) an increase in waves, resulting in (3) abrupt increases in erosion rates (2–4 times higher). Thus, (4) there is a potential cascading risk of carbon releases locally to the ocean and atmosphere due to the coastal permafrost collapse
2.4 Effects of AMOC changes on the Amazon rainforest
The strength of the AMOC exerts a substantial influence on the climate of tropical South America, most importantly on rainfall and its seasonal distribution. This in turn affects the state and stability of another potential tipping element in the Earth system: the Amazon rainforest.
The most important large-scale effect of the AMOC on rainfall in the Amazon works via the pattern of SSTs in the Atlantic and the associated shifts in the ITCZ (Intertropical Convergence Zone) and the tropical rain belt. There is widespread agreement that a reduction or even collapse of the AMOC would lead to reduced SSTs in the North Atlantic and increased SSTs in the South Atlantic . This change is caused by the reduction in the AMOC-related northward ocean heat transport and is amplified by wind–evaporation feedbacks . The changed SST pattern in turn affects atmospheric circulation by strengthening the Northern Hemisphere Hadley cell, particularly during boreal winter . As the location of the tropical rain belt depends on the cross-equatorial energy flux and the atmosphere energy input close to the Equator , a weakened AMOC together with a persistent Southern Ocean warming lead to a southward migration of the tropical rain belt, depending on the CO forcing trajectory . Hence, AMOC weakening may cause a tropical rain belt shift. This southward shift would cause a substantial reduction in rainfall over northern South America and an increase in rainfall over the portion of the Amazon located in the Southern Hemisphere, as well as over northeastern Brazil, which is directly affected by the tropical rain belt . Nevertheless, over the Amazon basin, the extent of this migration is model-dependent
To conclude, although different Earth system models have different biases in the location, shape, and strength of the tropical rain belt, they generally agree on the AMOC-collapse-induced increase in precipitation over the southern portion of the Amazon and northeastern Brazil . Given that the forests in the southern half of the basin contribute mostly to the rainfall generation over the basin , one could speculate that this would lead to a stabilization of the Amazon, given that a substantial fraction (24 %–70 %, , and references therein) of the rainfall of the basin is nonetheless produced by local moisture recycling. Furthermore, it has been shown that the altered tropical rain belt dynamics throughout the year would mean a reduction of rainfall mostly during the current wet season (peaking around March) and an increase in the dry season, peaking around September . Importantly, the consequences for the rainforest of a more equalized annual cycle are unclear. More generally, the full spectrum of rainforest stressors including societal-driven pressures, such as land use changes driving deforestation, has to be taken into account when assessing AMOC effects over the Amazon rainforest .
2.5 Influences of ENSO on proposed tipping elements
The El Niño–Southern Oscillation (ENSO) is the most important mode of climate variability on interannual timescales, fundamentally affecting regional and global atmospheric and oceanic circulation . The response to climate change of ENSO itself still remains debated, mainly because there are multiple (positive and negative) feedback processes in the tropical Pacific ocean–atmosphere system, whose relative strength determines the response of ENSO variability . Further, recent studies disagree about the future frequency of El Niño phases under global warming . In particular, a decreasing frequency of El Niño phases under global warming was suggested by a global climate model resolving mesoscale oceanic eddies and consequently reduced biases in the tropical oceanic mean state . Although it is debated or even unlikely whether ENSO should be considered a tipping element itself , it exerts important feedbacks on other global tipping elements. Through its global teleconnections, ENSO has the potential to influence multiple Earth system components including the AMOC, the Amazon rainforest, and the West Antarctic Ice Sheet. Changes in ENSO amplitude or frequency could alter the strength of (stabilizing or destabilizing) feedbacks within other (remote) tipping elements. Therefore, in this section, we discuss possible interactions between ENSO and other tipping elements.
2.5.1 Interactions between ENSO and AMOC
Various physical mechanisms have been discussed to explain how a decline in strength or complete shutdown of the AMOC could affect ENSO variability, mostly in terms of the amplitude of ENSO. An AMOC decline typically leads to cooling in North Atlantic surface temperatures, which affects the global atmospheric circulation, including the trade winds in the tropical Pacific. In many GCMs, upon decline of the AMOC, the northeasterly trade winds are intensified and the Intertropical Convergence Zone (ITCZ) is displaced southwards, eventually leading to an intensification of ENSO amplitude through nonlinear interactions . While the response of the trade winds and ITCZ to AMOC decline seems to be relatively robust within different (generations of) GCMs, the response in ENSO magnitude or frequency is much more model-dependent: trade winds can also affect the thermocline depth in the eastern tropical Pacific, thereby weakening the ENSO . By that, the zonal structure of winds and stratification is affected, leading to zonal shifts of variability patterns but no significant change in amplitude . Alternatively, weaker air–sea coupling due to altered trade winds affects the relevant tropical Pacific feedback balance such that the growth rate of ENSO is significantly reduced . In another model study, where physically based, conceptual models of the AMOC (Stommel box model, ) and ENSO (Zebiak–Cane model, ) are coupled via the trade wind strength, it was found that an AMOC collapse intensifies the tropical Pacific trade winds and shifts the ENSO system further into its oscillatory mode (i.e., amplitude increase) . It should be noted that most GCMs still exhibit severe biases in tropical temperature patterns, partly caused by oceanic mesoscale processes that are not properly resolved , which complicates the understanding of the fate of ENSO under greenhouse gas increase, but also under AMOC changes.
The reversed pathway, i.e., ENSO impacting the AMOC, likely also exists, but also depends on several atmosphere–ocean processes which may not be adequately resolved in models. A relatively robust teleconnection exists between an El Niño event and the negative phase of the North Atlantic Oscillation (NAO) in (late) winter . The statistical relationship between the AMOC and the NAO in GCMs depends on the subpolar North Atlantic background state; the AMOC is less sensitive in models that have extensive sea ice cover in deep-water formation areas, while in models with less sea ice cover, the background upper-ocean stratification largely determines how sensitively the AMOC reacts to surface buoyancy forcing . As for ENSO, the unbiased representation of the North Atlantic mean state represents a significant challenge for GCMs, in part due to insufficient resolution of mesoscale ocean eddies.
2.5.2 Influences of ENSO on the Amazon rainforest
The frequency and amplitude of ENSO variability have changed on decadal to centennial timescales in the past . In recent years, extreme El Niño events combined with global warming have become increasingly associated with unprecedented extreme drought and heat stress across the Amazon basin , leading to increases in tree mortality, fire, and dieback . Imposing the surface temperature pattern of a typical El Niño event in a global atmosphere–vegetation model suggests increased drought and warming in the Amazon rainforest region , which could enhance rainforest dieback and transition to degraded and fire-prone, savanna-like ecosystems in some regions. This could turn further parts of the Amazon rainforest from a carbon sink to a carbon source.
These destabilizing effects from ENSO on the Amazon rainforest are overlaid with direct climate change effects, land use change, deforestation, and human-made fires in the Amazon rainforest because the Amazon rainforest produces much of its own rainfall
2.5.3 Influences of ENSO on the WAIS
Recent significant surface melt events on West Antarctica were associated with strong El Niño phases . It has been proposed that these melt events were caused by atmospheric blocking, eventually leading to warm air temperature anomalies over West Antarctica that pass the melt point of parts of the ice sheet . Using reanalysis data, satellite observations, and hindcasting methods, strong indications have been found that the Ross and Amundsen Sea embayment regions are most affected by El Niño phases . In addition, it has been observed that, while ice shelves experience an increase in height (because accumulation height gains exceed basal melt height losses), they suffer from a decrease in mass (because basal ice loss exceeds ice gain from accumulation) due to increased ocean melting during significant El Niño occurrences in the Amundsen and Ross Sea area . Further, it is important to note that El Niño phases are not immediately transferred to surface melting in Antarctica but only after some time lag on the order of months .
Taken together, this adds to the growing body of literature indicating that a disintegration of the West Antarctic Ice Sheet, especially along the Ross–Amundsen sector, would be favored by strong El Niño phases and tipping risks may increase if El Niño phases become more frequent or intense under ongoing climate change . This may be concerning in particular because the Amundsen region is where the most vulnerable glaciers of the West Antarctic Ice Sheet are located such as the Pine Island or Thwaites glacier .
2.5.4 Influences of ENSO on warm-water coral reefs
ENSO drives abnormally high sea surface temperatures (seasonal heat waves above summer maxima baselines), which are superimposed on already warming oceans. Anomalous heat destabilizes the relationship between host corals and their symbiotic dinoflagellate algae (zooxanthellae), resulting in severe bleaching and mortality across multiple species of corals on spatial scales exceeding thousands of kilometers. While ENSO is geographically modulated by other ocean dipoles (e.g., Atlantic oscillation, Indian Ocean) , the Pacific signal is dominant and El Niño warm phases have been related to global episodes of extreme heat stress since the 1970s, e.g., 1979–1980, 1997–1998, and 2014–2017 . As global warming progresses and oceans become significantly warmer, the incidence of mass bleaching is decoupling from the El Niño warm phase , with warmer conditions compared to 3 decades ago . The global recurrence of bleaching has been reduced to an average of 6 years , sooner than expected from climate models and satellite-based sea temperatures. With warming temperatures and shortened intervals between major bleaching, multiple human stressors, and ocean acidification, the recovery time for mature assemblages of corals is now insufficient across most regions . At the scale of the Great Barrier Reef the emission of volatile sulfur compounds by corals adds to the local atmospheric aerosol load, increasing low-level cloud albedo and reducing warming . This breaks down during physiological stress and bleaching, potentially reinforcing thermal stress in a positive feedback loop. The potential contribution of this biologically derived feedback loop to local clouds, sea surface temperature, and coral bleaching is uncertain; this, however, needs validation in other locations and determination of any contribution to climatic conditions at larger spatial and temporal scales. While recovery from repeated bleaching events has been observed , the thresholds of global mean warming of 1.5 C (70 %–90 % loss of coral reefs globally) and 2 C (90 %–99 % loss) appear to still hold .
2.6 Effects of AMOC and ENSO changes on tropical monsoon systems
Future climate projections show a weakening of the AMOC, which can be substantial in its impact on the regional and global climate via ocean–atmosphere connection . Evidence from modeling and paleo-reconstructions has shown interhemispheric, low- to high-latitude climate connections via ocean–atmosphere linkage for heat and moisture transport
As described above, AMOC weakening leads to a southward shift of the ITCZ , which impacts the various monsoon systems worldwide , as also visible in paleorecords
An AMOC weakening has also been shown to strengthen the Indo-Pacific Walker circulation via cooling of the equatorial Pacific and warming of the Southern Hemisphere–Antarctic climate on a multi-decadal timescale
Both relationships, between ENSO and the Indian summer monsoon and also between ENSO and the West African monsoon, need to be further tested in paleoclimate reconstructions from both warm and cold climate states to gain a better understanding of how an abrupt change in AMOC may have an effect on ENSO and/or on tropical monsoon systems. This would allow for a more robust predictability of tropical monsoon rainfall patterns in the future.
2.7 Effects of permafrost regions on the global hydrological cycle
The permafrost regions have accumulated substantial amounts of ice in the soils. With ground ice melting away in a warmer climate, permafrost landscapes are experiencing drastic hydrological changes. The presence of ice modulates the thermophysical soil properties as well as infiltration rates and the vertical and lateral movement of water through the ground, which is often poorly represented in current Earth system models and therefore exhibits large inter-model differences. Hence, uncertainty exists about high-latitude regions becoming wetter or drier in the future due to soil hydrology representation in state-of-the-art Earth system models . They could either (i) turn into a wetter and cooler state with many freshwater systems and lakes, which support increasing land–atmosphere moisture recycling and cloud cover, reducing ground temperatures, or (ii) turn into a drier state as newly formed lakes could drain, with less moisture recycling supporting less cloud cover and a warmer surface . Which parts of the Arctic will be wetter or drier in the future is uncertain, but the differences between the potential Arctic hydroclimatic futures could be very pronounced. As recently shown by , the drier and warmer permafrost state would lead to less sea ice, a reduced pole-to-Equator temperature gradient, and a weaker AMOC. The drier state has more boreal forest extended to the north, with a higher frequency and extent of forest wildfires. In comparison with the wetter state, the drier Arctic state also shifts the position of the Intertropical Convergence Zone, which results in higher precipitation in the Sahel region and potentially also in the Amazon rainforest region. Increased forest and vegetation cover in these regions would be the consequence . Therefore, shifts in permafrost hydrology could affect climate tipping elements far beyond Arctic boundaries. Permafrost changes may impact the hydrological cycle with far-reaching impacts.
2.8 Interactions between multiple tipping elements and planetary-scale cascades
Assembling the individual links mentioned before in Sect. 2.2–2.7 gives rise to the possibility of tipping cascades involving more than two elements. These could lead to large changes at the regional and even planetary scale. Plausible examples are Dangaard–Oeschger (D/O) events (Sect. 3.2.2). Another example comes from the study of the last interglacial period, for which proxies for sea ice, polar ice sheets, AMOC, boreal forest, and permafrost indicate abrupt changes . Although the dating uncertainties make it difficult to determine the causal structure of a potential cascade, positive feedbacks between these TEs could explain the amplified polar temperatures and sea level rise obtained from reconstructions (8 C in Greenland and 6–9 m sea level rise compared to today) .
On a larger scale, tipping cascades could be responsible for driving the Earth system into completely different climatic states that have been identified in paleo-data , climate models of intermediate complexity , and general circulation models at coarse spatial resolution . For example, a tipping cascade involving ocean circulation and ice sheets might have been responsible for a transition from a greenhouse to an icehouse state at the Eocene–Oligocene boundary (Sect. 3.1).
While unlikely, a major concern regarding the future may be that a cascade involving several tipping elements and feedbacks could lock the Earth system in a pathway towards a hothouse state with conditions resembling those of the mid-Miocene (hothouse hypothesis: 4–5 C, 10–60 m sea level compared to the pre-industrial baseline) . Feedbacks that affect global temperature could involve albedo changes (e.g., through ice sheet or sea ice loss), and additional CO and CH emissions (through permafrost thawing, methane hydrates release) may lead to additional warming on medium to long timescales . In a worst-case (and unlikely) scenario, it has been speculated that a regional breakup of stratocumulus decks at atmospheric CO levels above 1200 ppm could translate into a large-scale temperature feedback leading to a warming of roughly 8 C .
Timescales are crucial when discussing hothouse scenarios. A potential hothouse state in the next centuries seems implausible in light of the current state of research. For example, in climate projections up to 2100, CMIP6 (Coupled Model Intercomparison Project) GCMs show no evidence of nonlinear responses on the global scale. Instead, they show a nearly linear dependence of global mean temperature on cumulative CO emissions . Similarly, in a recent assessment, concluded that a tipping cascade with large temperature feedbacks over the next couple of centuries remains unlikely and that while the combined effect of tipping elements on temperature is significant for those timescales, it is secondary to the choice of anthropogenic emissions trajectory. However, this does not completely rule out the possibility of a hothouse scenario in the longer term. Indeed, tipping events are not necessarily abrupt on human timescales. Positive feedbacks could have negligible impacts by 2100, for example on global mean temperature and sea level rise, but still influence Earth system trajectories on (multi-)millennial timescales . To the authors' knowledge, no multi-millennial simulation with a model incorporating the elements and feedbacks described above has yet been conducted to test this scenario. This calls for experiments across the model hierarchy. EMICs (Earth system models of intermediate complexity) in particular, and AOGCMs at coarse spatial resolution, offer an interesting trade-off as they include representations of most tipping elements while still allowing for multi-millennial simulations.
3 Possible examples of interactions between tipping elements from a paleoclimatic perspective
In this section, we outline tipping sequences that have been observed in the distant and less distant past. We discuss how the Eocene–Oligocene transition and interactions since the last (inter)glacial periods (Bølling–Allerød, Dansgaard–Oeschger, Heinrich events) may fit into the framework of a tipping cascade.
3.1 Interactions in the distant past: Eocene–Oligocene transition
The formation of a continent-scale ice sheet on Antarctica during the Eocene–Oligocene transition about 34 million years ago is known as Earth's greenhouse–icehouse transition (see Fig. 3). Following a cooling over tens of millions of years, this shift to a new climate state would have been visible from space, as Antarctic forests were replaced by a blanket of ice, and seawater receded from the continents, changing the shapes of coastlines worldwide. The climate transition is recorded as a shift in the oxygen isotopic composition of microscopic fossil shells in marine sediment cores, which reflects a combination of deep-sea cooling and continental ice growth . It had global consequences for Earth’s flora and fauna, both in the oceans and on land .
Figure 4
Conceptual linkages between changes in the Earth system associated with the Eocene–Oligocene transition 34 million years ago. External drivers were the slow changes in ocean gateways caused by tectonic plate movement and slow changes in Earth's orbital configuration. The interactions and feedbacks within the Earth system act on different timescales, which makes the complete sequence of events complicated, but overall these processes resulted in Earth's greenhouse–icehouse transition. Since all links are uncertain, they are denoted with dashed lines (compare Fig. 5).
[Figure omitted. See PDF]
This climate transition has been identified as a possible paleoclimate example of cascading tipping points in the Earth system . Examples of climatic tipping elements in this case consist of global deep-water formation, the Antarctic Ice Sheet, polar sea ice, monsoon systems, and tropical forests. In a conceptual model, the first part of the oxygen isotope shift is attributed to a major transition in global ocean circulation, while the second phase reflects the subsequent blanketing of Antarctica with a thick ice sheet .
The global ocean circulatory system was showing tentative signs of change a few million years before the climate transition, likely caused by changing ocean gateways in the North Atlantic . Neodymium isotopes do suggest that a precursor to North Atlantic Deep Water reached the Southern Hemisphere close to the Eocene–Oligocene transition, perhaps signaling the onset of Atlantic Meridional Overturning Circulation (AMOC) . However, the exact timing remains uncertain and may not correlate with the onset of the oxygen isotope shift. Indeed, the first part of the isotope shift is associated with a cooling of both deep-sea temperatures and low-latitude sea surface temperatures, which therefore more likely reflects a change in radiative forcing . However, this does not preclude AMOC onset preconditioning the system for glaciation through heat piracy in the Southern Ocean, with the exact timing of the transition set later by a favorable orbital configuration .
In general, biomes in Earth's greenhouse state reflect warmer and wetter conditions than the icehouse state of the early Oligocene, but many of these seemed to have changed gradually as climate cooled in the Eocene, making it difficult to identify vegetation tipping elements following the glaciation of Antarctica . The mammalian fossil record, which is coupled to vegetation through diet, suggests more acute changes in the early Oligocene. The Grande Coupure (i.e., “the Big Break”), is a long-known mammalian extinction–origination event during Eocene–Oligocene times involving large-scale migrations of Asian mammals into Europe . Thought to signal a combination of changing climate and floral changes, this abrupt faunal turnover might reflect crossing of an ecosystem tipping point caused by the crossing of a climate tipping point: a climate–ecology tipping cascade. Mammal extinctions seem to be particularly widespread in Afro-Arabia and linked to loss of dietary diversity . This finding is consistent with the idea that biomes in this subtropical region are tippable elements . Other evidence of vegetation biomes having tipped includes a transition from warm–temperate to cool–temperate rainforests in southeastern Australia . Monsoon systems, sensitive to forcing and to large-scale reorganizations of the climate system, might have been important for explaining the shifts in these respective vegetation biomes. Moreover, simulations of the late Eocene climate suggest the existence of a strong monsoon-like climate over the Antarctic continent; without a major reorganization of such an atmospheric circulation regime, ice growth on Antarctica seems very unlikely .
The glaciation of Antarctica also produced a sea level fall of several tens of meters , causing shallow seaways to recede, turning many marine regions into continental habitats, which experienced particularly strong seasonality . The associated reduction of the marine carbonate factory in previously submerged tropical shelf seas caused the calcite compensation depth to deepen by more than 1 km, turning enormous swathes of seafloor white as the sinking calcite shells of plankton no longer dissolved in shallow depths .
In summary, Earth's greenhouse–icehouse transition was likely associated with a range of interactions between components of the Earth system that are debated as potential tipping elements. Determining the extent to which these reflect a cascading series will require a major data-modeling effort, with improved correlations between marine and terrestrial records, as well as better constraints on the rate and magnitude of change within a range of tipping elements.
3.2 Interactions during and since the last glacial period3.2.1
Bølling–Allerød
Towards the end of the last ice age, a very prominent event is recorded in numerous geological archives. The Bølling–Allerød (B/A) started at 14.7 ka with abrupt warming in the Northern Hemisphere (with polar atmospheric circulation shifts over a few years followed by temperature increase in Greenland of around 10 C in the subsequent decades – ) in response to a reinvigoration of the AMOC and lasted until 12.9 ka. The B/A is an example of pronounced interactions between Earth system components and cascading impacts in the Earth system . At the onset of the B/A, atmospheric CO and CH concentrations rapidly increased over a few decades in response to AMOC strengthening, abrupt warming and permafrost thaw , and moisture changes
The trigger for the rapid amplification of ocean circulation and the associated abrupt impacts at the B/A transition has been a focus of debate, with opinions divided between an essentially linear response to the (possibly abrupt) cessation of freshwater forcing
Gradual changes observed in key climatic variables (e.g., CO and global temperature) during sustained periods of cold across the surface North Atlantic (as occurred prior to the B/A onset) were a persistent feature of glacial terminations throughout the last 800 kyr
Equally rapid as B/A and of comparable magnitude, transitions known as Dansgaard–Oeschger (D/O) events (Fig. 4) occurred repeatedly during glacial periods throughout much of the late Pleistocene
During the abrupt warming phases of D/O cycles, an abrupt decrease in Arctic and North Atlantic sea ice cover likely contributed to the onset of convection and a rapid resurgence of a much weaker, and potentially even collapsed, AMOC . D/O-type changes in coupled climate models also feature a rapid disappearance of sea ice that precedes the abrupt AMOC strengthening . Lastly, D/O could also be self-sustained oscillations as recent literature pointed out . Thus, the D/O warmings may potentially comprise a tipping cascade . However, such a cascading interaction may depend on the climate background state, and it is unclear whether North Atlantic sea ice cover during the last glacial period can be considered a tipping element.
Figure 5
Interactions at the end of Heinrich Stadial 4 (HS4). (a) Climate proxy indices spanning the transition from HS4 into Dansgaard–Oeschger (D/O) event 8 (time goes from left to right). From top to bottom: AMOC strength , Norwegian Sea ice cover , Greenland temperature , North Atlantic SST , dust accumulation in Greenland , Asian monsoon intensity , and South American monsoon intensity . The horizontal red bar indicates the period when the ITCZ assumed a more southerly position . The hatched region spans the transition from HS4 to D/O8 and represents an estimate of the relative age uncertainty among the records shown (i.e., it is generally not possible to tell which changes occurred earlier or later within the overall sequence). Vertical arrows indicate the direction of increase for each parameter. (b) Interactions between the ocean, atmosphere, and land during the end of HS4. Linkages with comparably high uncertainty are denoted with dashed arrows.
[Figure omitted. See PDF]
3.2.3 Heinrich eventsWhile the exact causes and mechanisms of the B/A transition and D/O events are still under debate, Heinrich events are better understood. They occurred during some of the cold stadial phases mentioned above and were associated with major reorganization of ocean circulation in the North Atlantic (for a review see ). During Heinrich events, large masses of ice were released from the Laurentide Ice Sheet, leading to a dramatic freshening of the North Atlantic Ocean and enhanced suppression of deep-water formation and the AMOC.
They can be understood as a phenomenon involving two tipping elements, the Laurentide Ice Sheet and the AMOC. The ice fluxes from the Laurentide Ice Sheet have been described as a binge–purge oscillator , where a period of strong ice accumulation (the binge phase) is followed by a period of rapid ice loss (the purge phase). During the binge phase, ice is generally thought to be frozen to the bottom and thus immovable. As the ice sheet gets thicker, basal temperatures increase until the pressure melting point of the basal ice is reached. The resulting meltwater production lubricates the bed and enables sliding of the ice. This may already be sufficient to initiate the purge phase, though further triggers like ocean subsurface warming probably also played an important role in destabilizing marine-terminating portions of the Laurentide Ice Sheet . The purge phase lasts until the ice sheet has become too thin to maintain basal temperatures above the pressure melting point, thus refreezing and stopping the ice flow. The resulting ice stream flows into the Atlantic Ocean, and as the resulting icebergs melt, Atlantic surface waters are freshened to the point where the AMOC cannot be sustained and collapses.
The mechanisms sketched above have been demonstrated in a number of transient model experiments using Earth system models of intermediate complexity and complex ice sheet–atmosphere–ocean general circulation models . However, not all details have yet been resolved; the exact trigger mechanism (and threshold) initiating the purge phase, for example, has not yet been identified .
3.3 A paleoclimate perspective on the resilience of the Amazon rainforest
Two historical analogs have provided some (albeit not fully consistent) insights into the response of the Amazon to reductions in rainfall: Heinrich events during the last glacial period and the mid-Holocene. As mentioned in the previous section, Heinrich events are remarkable intervals during the last glacial period in which the AMOC seems to have substantially weakened in response to iceberg release in the North Atlantic . Paleoclimate data from these events are of great help to evaluate the processes suggested by climate model simulations of AMOC slowdown. , using an isotope proxy from a sediment core collected offshore of the Amazon River mouth, showed savanna intrusions into the Amazon rainforest during repeated Heinrich events. The intrusions of savanna occurred in northern Amazonia and validate the suggested decrease in precipitation over that region in response to AMOC weakening . Further precipitation and, even more importantly, vegetation reconstructions with appropriate age models and sufficient temporal resolution will help clarify the southward extent of the drying of northern Amazonia due to an AMOC collapse, as well as its consequences for the rainforest.
used pollen, charcoal, and speleothem oxygen isotope proxy data to reconstruct the response of the Amazon forest during the mid-Holocene, when precipitation was relatively low . Their analysis suggests that the Amazon was resilient to rainfall reductions as high as projected by climate models for the rest of the century. However, it also has to be considered that in the study of temperature and land use were similar to pre-industrial conditions, whereas future warming and deforestation will act as additional stressors that affect the surface water balance by increasing potential evapotranspiration and decreasing precipitation recycling , while also increasing the chances of fire and thus the possibility of the Amazon being converted into a degraded, open ecosystem.
4 Modeling tipping element interactions and cascading transitions
Modeling interactions between tipping elements and potential tipping cascades in the climate system is a difficult task. A key challenge is to accurately capture feedback mechanisms between different climatic components. In addition, each climate subsystem evolves over spatial scales and timescales that can span orders of magnitude from decades to centuries for biosphere components and from centuries to millennia for the large ice sheets on Greenland and Antarctica (see Fig. 2). The ideal tool to study the interaction between tipping elements would therefore be a high-resolution, comprehensive Earth system model based on general circulation models for the atmosphere and ocean, with a sea ice component, dynamic vegetation, and interactive ice sheets and carbon cycle. Moreover, the model should be computationally fast to allow the representation of the slow processes and to run comprehensive ensembles for taking into account the uncertainties in key parameters . However, such a universal tool does not exist. Instead, a hierarchy of models of different complexity is needed to explore the interactions between tipping elements on different temporal and spatial scales . The state-of-the-art Earth system models that are usually employed in climate change projections, the models in the CMIP6 , are the first choice to investigate processes developing over centennial timescales. Since these models usually include dynamic vegetation, they are suitable to explore the interactions between abrupt changes in ocean circulation, Arctic sea ice, and vegetation cover. However, CMIP6 GCMs are computationally expensive and most of them do not include interactive ice sheets, which limits the applicability of these models to study interactions between tipping elements and potential tipping cascades that involve slow deep-ocean dynamics or ice sheets. Also, they show some limitations to how vegetation is represented, especially in tropical areas
For studying feedbacks on millennial timescales or longer, one possible solution is to use Earth system models with coarser spatial resolution, allowing for faster simulations , or Earth system models of intermediate complexity (EMICs, ). A downside of these models is that the interactions between tipping elements are necessarily less realistic and some nonlinear processes need to be parameterized at sub-grid level. In particular, EMICs are faster than GCMs of comparable spatial resolution, since they make use of some approximations in the representation of the atmosphere and/or ocean dynamics and can, for instance, be applied to investigate climate–ice sheet interactions on multi-millennial timescales .
An alternative technique to speed up complex models, and therefore enable them to explore feedbacks on longer timescales, is offline (asynchronous) coupling, which has been applied to represent vegetation–climate and ice sheet–climate interactions . One example of asynchronous coupling is when the atmosphere evolves with fixed vegetation cover and eventually the latter is updated to the equilibrium conditions of the former . However, despite being less computationally expensive, feedback mechanisms and thus tipping phenomena and cascades are better represented when dynamical (synchronous) coupling is implemented .
It is important to note that such complex process-based models with sufficient integration of tipping elements (e.g., in EMICs or GCMs) are now starting to become available, for instance when ice sheet models are dynamically integrated in Earth system models or when new versions of EMICs are developed . This will open up new possibilities for the simulation of cascading transitions.
Tipping phenomena and cascades at the regional scale, however, may be investigated with different approaches . For instance, they can be investigated using observation-based simulations coupled to energy or hydrological balance models at high spatial resolution as has been done for tipping cascades in the Amazon rainforest . Regional implementations of ice shelf–ocean interaction exist to obtain improved estimates of basal melt and to include small-scale processes, like the presence of ocean eddies that can affect the overall stability of the system and potentially intensify transitions or cascades. It is indeed possible to run regional climate models (RCMs) at horizontal grid resolutions of a few kilometers, thus providing more accurate spatiotemporal distributions of climatic variables like precipitation and temperature than GCMs. An alternative modeling framework is to apply grid refinement over a specified region of interest in a GCM, which avoids inconsistencies between the different dynamical cores and physical parameterizations used in RCMs and GCMs, and (more importantly for tipping phenomena or cascades) it allows for two-way interactions between the refinement region and the global domain . With such regionalized modeling approaches it could be possible to empirically detect local to super-regional (cascading) regime shifts .
Also, more conceptual approaches based on differential equations or box models are frequently used for studying tipping events and cascades for present-day climate and paleoclimates . While such models offer a unique way of unraveling the complex dynamics of interacting tipping elements, it is not guaranteed that results of conceptual models can be confirmed by complex models, since the former consider only a limited subset of dynamical variables and nonlinear processes of the climate system . For example, simple models that do not include space are suggested to overpredict the existence of tipping points, while spatial pattern formation phenomena might prevent such tipping when space is explicitly included .
Lastly, modeling of cascading effects from the physical system to society and the economy as well as vice versa is still missing from most state-of-the-art Earth system models, requiring urgent development
5 Discussion and conclusion
As anthropogenic global warming continues, tipping elements are at risk of crossing critical thresholds . Several assessments have investigated the risk of crossing critical thresholds of individual tipping elements, whereas interactions between tipping elements have only more recently been taken into account, mostly by conceptual models
-
Out of the discussed interactions in this review, 13 are assessed as destabilizing, while 2 are stabilizing and 4 are of unclear sign (see Figs. 1 and 3). This means that the majority of interactions are destabilizing. While confirmation or rejection through future research is necessary, the possibility cannot be ruled out that interactions between climate tipping elements destabilize the Earth system in addition to climate change effects on individual tipping elements. Therefore, tipping cascades cannot be ruled out when tipping thresholds of the first tipping elements are transgressed through ongoing global warming.
-
The core tipping elements with the lowest thresholds are the large ice sheets on Greenland and West Antarctica. They have been suggested as the initiators of tipping cascades . However, those are also the tipping elements with the largest tipping time (several centuries up to millennia). Thus, if their tipping points are only transgressed for a limited amount of time (overshoot), cascading tipping risks are also reduced
see the Supplement . -
However, if global temperatures reach global warming levels beyond 2.0 C (or stay between 1.5 and 2.0 C for a centennial timescale), more and more fast tipping elements like the AMOC or the Amazon rainforest would also be at risk of tipping and could then initiate a cascading transition on a faster timescale
see the Supplement .
As such, our results are in line with earlier assessments, in particular on timescales of tipping element interactions and possible cascading transitions, but also the current state of our knowledge gaps for tipping cascade risks . Summarized, much uncertainty remains and most studies as of now rely on conceptual models rather than detailed process-based Earth system models. Further, it is important to point out that most of the agreement and evidence in Table 1 refer to interactions in the context of natural climate variability or gradual forced climate change. To what extent the climate system can be considered nonlinear enough to produce tipping points, but on the other hand linear enough to extrapolate these interactions from small amplitudes to extreme changes related to tipping, needs to be tested. Therefore, assessing the overall stability of the Earth system, and the possibility of a chain of nonlinear transitions, will require more detailed assessments of interactions, their effect strengths, timescales, and state dependence.
While there is more and more research on individual thresholds of climate tipping elements, substantial uncertainties prevail in the existence and strength of many links between tipping elements. In order to decrease such uncertainties, we propose four possible ways forward. (i) Observation-based approaches: satellite observations, reanalysis, and paleoclimate data sets may be evaluated using correlation measures or more advanced causal inference approaches
Clearly all of these strategies have their strengths and limitations. Both (i) and (ii) could benefit from extending the established notion of correlation measures in climate networks to causal measures such as causal inference methods, for instance informed by Granger causality or conditional (in)dependences . A prominent approach for causal inference has been applied in climate science using the so-called PCMCI algorithm , which is a constraint-based method that considers (lagged) partial correlation to establish links between the considered variables. Such methods can be used to check whether identified correlations indeed correspond to a causal relation, where it is important to take into account all possible confounding factors. To do that one needs to start from the physical processes involved, for instance informed by conceptual models, and test this as a hypothesis for the causal relations . Another example of such a causal method is the Liang–Kleeman information flow method, which is based on the rate of information transfer in dynamical systems and has been rigorously derived from the propagation of information entropy between variables . The method has been recently applied to Earth system processes, including Antarctic surface mass balance and Arctic sea ice . However, to apply such methods to the tipping point context it is important to know the limitations. One of the assumptions made is that of stationarity of the links between the variables considered, which may not be true once a tipping point is crossed. Another difficulty may be the different timescales of the tipping elements where, e.g., ice sheets are very slow compared to the Amazon rainforest or AMOC.
Limitations of the approaches (i)–(iv) further include the following. First, data from recent Earth observation efforts (e.g., for AMOC by RAPID and OSNAP – – or for ice stream velocities by satellite laser altimetry from ICESat/ICESat-2 – ) may need to be extended to cover more variables relevant to Earth system tipping elements as well as better covering the relevant temporal and spatial scales. Paleoclimate data can partially compensate for such disadvantages at the cost of the data being hard to retrieve with availability and abundance far from perfect. Second, complex Earth system models may not include all relevant interaction processes between tipping elements or are often computationally too expensive to run large-scale ensembles that could take into account and propagate all relevant uncertainties. However, kilometer-scale models are under development and tipping point simulations on this scale will likely become feasible within the next decade due to current developments and further growing computational resources . And third, risk analysis approaches include accounting for theoretical knowledge properties of these types of physical dynamical systems. Therefore, different approaches should complement each other, requiring experts to combine observations, reconstructions, and novel computational strategies individually, but also potentially through a formalized elicitation. Taken together, all approaches mentioned above are required to obtain more reliable estimates of existential risks such as those posed by potential tipping events or even cascades . They could be used to inform an emulator for tipping risks taking into account properties of individual tipping elements as well as their interactions. In addition, there are also large uncertainties among the known interactions as discussed above, and also not all interactions are known or quantified (known unknowns versus unknown unknowns).
Further, in certain systems, there are forcings of non-climatic origin that could interact with climate change and lead to tipping and thus to interactions and possibly cascades with other elements. For instance, land use change and, specifically, deforestation are threatening the Amazon and decreasing its resilience to climate change
Taken together, assessing and quantifying tipping element interactions better have great potential to advance suitable risk analysis methodologies for climate tipping events and cascades, especially because it is clear that tipping elements are not isolated systems and appropriate risk assessment tools are yet to be developed.
Data availability
The data sets used in Fig. 5 can be found in the following references for (top to bottom) AMOC strength (Henry et al., 2016), Norwegian Sea ice cover (Sadatzki et al., 2020), Greenland temperature (North Greenland Ice Core Project members (NGRIP), 2004), North Atlantic SST (Martrat et al., 2007), dust accumulation in Greenland (Ruth et al., 2007), Asian monsoon intensity (Cheng et al., 2016), and South American monsoon intensity (Kanner et al., 2012).
The supplement related to this article is available online at:
Author contributions
NW and AvdH designed the study. NW and ASvdH led the writing of the paper with input from all authors. NW, ASvdH, NJS, YA, SB, JTB, VB, TK, RW, VC, AK, HC, and CL designed the figures of this paper. All authors have reviewed and edited the final version of the paper.
Competing interests
At least one of the (co-)authors is a member of the editorial board of Earth System Dynamics. The peer-review process was guided by an independent editor, and the authors also have no other competing interests to declare.
Disclaimer
For the EU projects the work reflects only the authors' view; the European Commission and their executive agency are not responsible for any use that may be made of the information the work contains. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.
Special issue statement
This article is part of the special issue “Tipping points in the Anthropocene”. It is a result of the “Tipping Points: From Climate Crisis to Positive Transformation” international conference hosted by the Global Systems Institute (GSI) and University of Exeter (12–14 September 2022), as well as the associated creation of a Tipping Points Research Alliance by GSI and the Potsdam Institute for Climate Research, Exeter, Great Britain, 12–14 September 2022.
Acknowledgements
This review article has been carried out within the framework of the Global Tipping Points Report 2023. We thank Timothy Lenton, David Armstrong McKay, and Sina Loriani for many fruitful discussions within this framework. Nico Wunderling and Jonathan F. Donges acknowledge support from the European Research Council advanced grant project ERA (Earth Resilience in the Anthropocene, ERC-2016-ADG-743080). Jonathan F. Donges is grateful for financial support by the project CHANGES funded by the German Federal Ministry for Education and Research (BMBF) within the framework “PIK_Change” under grant 01LS2001A. Ricarda Winkelmann acknowledges financial support via the Earth Commission, hosted by FutureEarth. The Earth Commission is the science component of the Global Commons Alliance, a sponsored project of Rockefeller Philanthropy Advisors, with support from Oak Foundation, MAVA, Porticus, Gordon and Betty Moore Foundation, Herlin Foundation, and the Global Environment Facility. The Earth Commission is also supported by the Global Challenges Foundation. Victor Couplet is funded as a research fellow by the Belgian National Fund of Scientific Research (F.S.R. – FNRS). Ann Kristin Klose and Ricarda Winkelmann acknowledge support by the European Union's Horizon 2020 research and innovation program under grant agreement no. 820575 (TiPACCs) and no. 869304 (PROTECT). Thomas Kleinen acknowledges support through the project PalMod, funded by the German Federal Ministry of Education and Research (BMBF), under grant no. 01LP1921A. Maura Brunetti acknowledges financial support from the Swiss National Science Foundation (Sinergia project no. CRSII5_180253). Coastal and marine sections would like to acknowledge the generous support of the BNP PARIBAS Foundation for the CORESCAM project, part of the 2019 call on Biodiversity and Climate Change, and the USAID for support for the SWAMP project. John T. Bruun gratefully acknowledges the UK Research Councils funded Models2Decisions grant (M2DPP035: EP/P0167741/1), ReCICLE (NE/M004120/1), and STFC Spark Award (ST/V005898/1), which helped fund his involvement with this work. Anna von der Heydt and Swinda K. J. Falkena acknowledge funding by the Dutch Research Council (NWO) under a Vici project to Anna von der Heydt (with project number VI.C.202.081 of the NWO Talent program). Anna von der Heydt has worked under the program of the Netherlands Earth System Science Centre (NESSC), financially supported by the Ministry of Education, Culture and Science (OCW). Anna von der Heydt, Robbin Bastiaansen, and Sacha Sinet acknowledge funding from the European Union's Horizon 2020 research and innovation program under grant agreement no. 820970 (this paper is TiPES paper no. 232) and under the Marie Skłodowska-Curie grant agreement no. 956170 (CriticalEarth). Caroline H. Lear acknowledges NERC funding for SWEET grant NE/P019102/1. Jonathan Barichivich acknowledges funding by the European Research Council (ERC) under the Horizon Europe research and innovation program (ERC-starting grant CATES, grant agreement no. 101043214). Cristiano M. Chiessi acknowledges financial support from FAPESP (grants 2018/15123-4 and 2019/24349-9) and CNPq (grant 312458/2020-7). Yevgeny Aksenov and Stefanie Rynders acknowledge support from the following projects: COMFORT (grant agreement no. 820989) under the European Union's Horizon 2020 research and innovation program; the EC Horizon Europe project OptimESM “Optimal High Resolution Earth System Models for Exploring Future Climate Changes” under grant 101081193 and UKRI grant 10039429; EPOC, EU grant 101059547; UKRI grant 10038003; and the UK NERC projects LTS-M BIOPOLE (NE/W004933/1), CANARI (NE/W004984/1), and Consequences of Arctic Warming for European Climate and Extreme Weather (ArctiCONNECT, NE/V004875/1). Yevgeny Aksenov and Stefanie Rynders acknowledge the use of the ARCHER UK National Supercomputing and JASMIN. David Docquier is funded by the Belgian Science Policy Office (BELSPO) under the RESIST project (contract no. RT/23/RESIST). Mara Baudena acknowledges the Italian National Biodiversity Future Center (NBFC) – National Recovery and Resilience Plan (NRRP), Mission 4 Component 2, Investment 1.4 of the Italian Ministry of University and Research, funded by the European Union – NextGenerationEU (project code CN_00000033). Matteo Willeit acknowledges financial support by the German climate modeling project PalMod supported by the German Federal Ministry of Education and Research (BMBF) as a Research for Sustainability initiative (FONA) (grant nos. 01LP1920B and 01LP1917D). Didier Swingedouw acknowledges financial support by the RRI “Tackling Global Change” from the University of Bordeaux and by the INSU/LEFE DECORATING and the UKRI DECADAL projects.
Financial support
This research has been supported by the European Research Council, H2020 European Research Council (grant nos. 743080, 820575, 869304, 820970, 101043214, 820989, and 101059547), the Bundesministerium für Bildung und Forschung (grant nos. 01LS2001A, 01LP1921A, 01LP1920B, and 01LP1917D), the Natural Environment Research Council (grant nos. NE/P019102/1, NE/W004933/1, NE/W004984/1, and NE/V004875/1), the Swiss National Science Foundation (grant no. CRSII5_180253), Research Councils UK (grant nos. EP/P0167741/1, NE/M004120/1, and ST/V005898/1), the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (grant no. VI.C.202.081), the Fondation BNP Paribas (grant no. CORESCAM), the United States Agency for International Development (grant no. SWAMP), the Netherlands Earth System Science Centre (grant no. NESSC), the HORIZON EUROPE Marie Skłodowska-Curie Actions (grant no. 956170), the Fundação de Amparo à Pesquisa do Estado de São Paulo (grant nos. 2018/15123-4 and 2019/24349-9), the Conselho Nacional de Desenvolvimento Científico e Tecnológico (grant no. 312458/2020-7), UK Research and Innovation (grant no. 10039429), and the Belgian Federal Science Policy Office (grant no. RT/23/RESIST). This work was supported by the TipESM project funded by the European Union's Horizon Europe research and innovation program under grant agreement no. 101137673 as well as the Swedish Research Council (DNR 2008-2859).
Review statement
This paper was edited by Jesse Abrams and reviewed by B. van der Bolt, Steven Lade, and two anonymous referees.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2024. This work is published under https://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
Climate tipping elements are large-scale subsystems of the Earth that may transgress critical thresholds (tipping points) under ongoing global warming, with substantial impacts on the biosphere and human societies. Frequently studied examples of such tipping elements include the Greenland Ice Sheet, the Atlantic Meridional Overturning Circulation (AMOC), permafrost, monsoon systems, and the Amazon rainforest. While recent scientific efforts have improved our knowledge about individual tipping elements, the interactions between them are less well understood. Also, the potential of individual tipping events to induce additional tipping elsewhere or stabilize other tipping elements is largely unknown. Here, we map out the current state of the literature on the interactions between climate tipping elements and review the influences between them. To do so, we gathered evidence from model simulations, observations, and conceptual understanding, as well as examples of paleoclimate reconstructions where multi-component or spatially propagating transitions were potentially at play. While uncertainties are large, we find indications that many of the interactions between tipping elements are destabilizing. Therefore, we conclude that tipping elements should not only be studied in isolation, but also more emphasis has to be put on potential interactions. This means that tipping cascades cannot be ruled out on centennial to millennial timescales at global warming levels between 1.5 and 2.0
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details
; Anna S von der Heydt 2
; Aksenov, Yevgeny 3
; Barker, Stephen 4
; Bastiaansen, Robbin 5
; Brovkin, Victor 6
; Brunetti, Maura 7
; Couplet, Victor 8 ; Kleinen, Thomas 6
; Lear, Caroline H 4
; Lohmann, Johannes 9
; Roman-Cuesta, Rosa Maria 10 ; Sinet, Sacha 2
; Swingedouw, Didier 11
; Winkelmann, Ricarda 12
; Anand, Pallavi 13
; Barichivich, Jonathan 14
; Bathiany, Sebastian 15
; Baudena, Mara 16
; Bruun, John T 17
; Chiessi, Cristiano M 18
; Coxall, Helen K 19
; Docquier, David 20
; Donges, Jonathan F 1
; Swinda K J Falkena 21
; Klose, Ann Kristin 12
; Obura, David 22 ; Rocha, Juan 23
; Rynders, Stefanie 3
; Norman Julius Steinert 24
; Willeit, Matteo 25
1 Earth System Analysis and Complexity Science, Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association, Potsdam, Germany; Stockholm Resilience Centre, Stockholm University, Stockholm, Sweden; High Meadows Environmental Institute, Princeton University, Princeton, NJ, USA
2 Department of Physics, Institute for Marine and Atmospheric research Utrecht (IMAU), Utrecht University, Utrecht, the Netherlands; Centre for Complex Systems Studies, Utrecht University, Utrecht, the Netherlands
3 National Oceanography Centre, Southampton, United Kingdom
4 School of Earth and Environmental Sciences, Cardiff University, United Kingdom
5 Department of Physics, Institute for Marine and Atmospheric research Utrecht (IMAU), Utrecht University, Utrecht, the Netherlands; Department of Mathematics, Utrecht University, Utrecht, the Netherlands
6 Department of Climate Dynamics, Max Planck Institute for Meteorology, Hamburg, Germany
7 Group of Applied Physics and Institute for Environmental Sciences, University of Geneva, Geneva, Switzerland
8 Earth and Life Institute, UCLouvain, Louvain-la-Neuve, Belgium
9 Physics of Ice, Climate and Earth, Niels Bohr Institute, University of Copenhagen, Denmark
10 European Commission, Joint Research Center, Sustainable Resources, Forests and Bioeconomy Unit., Ispra, Italy
11 Environnements et Paléoenvironnements Océaniques et Continentaux (EPOC), Univ. Bordeaux, CNRS, Bordeaux INP, France
12 Earth System Analysis and Complexity Science, Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association, Potsdam, Germany; Institute of Physics and Astronomy, University of Potsdam, Potsdam, Germany
13 School of Environment, Earth and Ecosystem Sciences, The Open University, Milton Keynes, United Kingdom
14 Laboratoire des Sciences du Climat et de l'Environnement (LSCE), LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay, Gif-sur-Yvette, France; Instituto de Geografía, Pontificia Universidad Católica de Valparaíso, Valparaíso, Chile
15 Earth System Analysis and Complexity Science, Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association, Potsdam, Germany; Earth System Modelling, School of Engineering and Design, Technical University Munich, Munich, Germany
16 Centre for Complex Systems Studies, Utrecht University, Utrecht, the Netherlands; National Research Council of Italy, Institute of Atmospheric Sciences and Climate (CNR-ISAC), CNR, Torino, Italy; National Biodiversity Future Center, Palermo, Italy
17 Faculty of Environment, Science and Economy, University of Exeter, Exeter, United Kingdom
18 School of Arts, Sciences and Humanities, University of São Paulo, São Paulo, Brazil
19 Department of Geological Science, Stockholm University, Stockholm, Sweden; The Bolin Centre for Climate Research, Stockholm, Sweden
20 Royal Meteorological Institute of Belgium, Brussels, Belgium
21 Department of Physics, Institute for Marine and Atmospheric research Utrecht (IMAU), Utrecht University, Utrecht, the Netherlands
22 CORDIO East Africa, Mombasa, Kenya
23 Stockholm Resilience Centre, Stockholm University, Stockholm, Sweden; Anthropocene Laboratory, The Swedish Royal Academy of Science, Stockholm, Sweden
24 NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, Bergen, Norway; CICERO Center for International Climate Research, Oslo, Norway
25 Earth System Analysis and Complexity Science, Potsdam Institute for Climate Impact Research (PIK), Member of the Leibniz Association, Potsdam, Germany