Tipping points—where a small perturbation triggers a large response—can occur in many complex environmental systems. They produce abrupt and sometimes irreversible change, are inherently difficult to predict, and thus pose considerable challenges to the occupants and managers of those systems. However, tipping points can also represent opportunities. Here, different mathematical types of tipping points and different environmental processes that can give rise to them are distinguished. Then, I chart the crucial role that tipping points played in creating the modern Earth system. Looking ahead, potential large-scale tipping points are briefly reviewed before highlighting systems that could harbor tipping points across mechanisms and scales. The prospects for anticipating tipping points, avoiding dangerous ones, and encouraging others are outlined. Finally, a series of virtuous tipping points are identified, which can help transform the relationships between human societies and the environmental systems we depend upon.
KeywordsThe phrase “tipping point” captures the notion that sometimes little things can make a big difference to the state and/or fate of a system (1). Tipping points can occur in all manner of complex dynamical systems, including the human body and human societies. Indeed the term tip point was first introduced in sociology in the 1950s to describe the percentage of nonwhite residents in a US city neighborhood that would trigger a “white flight,” switching the neighborhood to total occupation by nonwhites (2). This was long before the book by Gladwell (1) caused a tipping point of its own in usage of the term. Since its publication, the many adopters of the phrase tipping point have included environmental scientists trying to articulate the prospect of passing thresholds in the climate system or in ecosystems (3).
Tipping points are a particular concern in environmental systems because we humans depend on those systems—for our well-being in the case of its many ecosystems and for our very existence in the case of the Earth system. Tipping points produce abrupt system-wide change that is often difficult (and sometimes impossible) to reverse, giving them high impacts. Thus, even if their likelihood is low, they pose significant risks—in the technical sense that risk is the product of the likelihood of an event and its impacts (4). Tipping points are also difficult to predict, making them hard to manage. However, there is currently much excitement about generic early warning indicators for one important class of tipping points (3–5). These have the potential to diagnose the vulnerability of a system to being tipped, if not forecast the time of tipping (6).
For a system to exhibit a tipping point, there must be strong positive feedback in its internal dynamics. Positive feedback means a closed loop of causal connections that are self-amplifying, tending to magnify any perturbation, and for there to be a tipping point, the positive feedback must be strong enough (or nearly so) to go into “runaway,” under some conditions ( Figure 1 ). Runaway means that, in response to a given perturbation, one trip around the positive feedback loop produces an additional change that is as large as or larger than the initial perturbation, so the next trip around the loop produces an even larger response, and so on. Such a mechanism can propel an abrupt transition between different “attractor” states of a system. In contrast, the attractors themselves are characterized by negative feedback, self-damping causal loops, which counteract perturbations. Tipping points are thus often associated with systems that exhibit multiple stable states (alternative attractors) under the same boundary conditions.
Although the phrase tipping point has only been relatively recently applied to environmental systems, the underlying concept that such systems can sometimes exhibit threshold behavior has been widely appreciated for at least the past half century. For example, in 1961 a box model of two stable regimes of flow for ocean thermohaline circulation was introduced (7), with the potential for abrupt shifts between them. In the late 1960s, energy balance models showed that Earth's climate could exhibit multiple stable states of temperature and ice cover under the same incoming sunlight (8, 9) and, hence, abrupt transitions as incoming sunlight varies (as it has done over Earth's history). The concepts of multiple stable states and accompanying abrupt transitions were also introduced to ecology in the late 1960s (10) and flourished through the 1970s (11, 12). The past decade or so has seen a renaissance of interest in tipping points in ecological systems under the banner of ecosystem “regime shifts” (13). More recently, the term “critical transitions” has been used to describe tipping points in a range of complex systems (5).
Thus, the field could be reviewed by dividing it down disciplinary lines into ecological- and climate-tipping points, as these have been the main fields of recent research activity. However, both fields have been reviewed already, quite exhaustively for ecosystems (13–18), less so for climate (3, 19–21), although the connections between them have received less attention. Furthermore, there are other types of environmental systems that can exhibit tipping points: Geomorphologists have identified threshold changes between alternative landscape states (22–24), and biogeochemists have recognized the key roles of positive feedbacks in driving oceanic anoxic events (OAEs) (25, 26). These different types and scales of system (and associated disciplinary segregation) conceal some common types and features of tipping points. Hence, this review aims to draw out those commonalities and also highlight systems where tipping could occur across a range of spatial scales, involving a variety of environmental processes. Particular emphasis is placed on larger-scale tipping points because these generally pose greater risks (or opportunities).
As tipping point has been liberally used of late in an environmental context, it is important to be clear what this review is not about. It is not about all environmental positive feedbacks (i.e., self-amplifying causal loops). It focuses on cases where positive feedback is strong enough to create alternative stable states ( Figures 1 and 2a,b ), or very nearly so, whereas the majority of positive feedbacks are not strong enough to create tipping points. It is not about all regime shifts as some clearly occur within modes of variability without the passing of a tipping point (27). The review is also not about what might be termed turning points, for example, where consumption of a finite resource (such as oil or phosphorus) stops increasing and starts to decline, or when the land surface switches from being a sink to a source of carbon dioxide. Nor is the review about “points of no return” per se, such as the loss of a species or of a unique ecosystem, although some tipping points are also points of no return.
The review starts by establishing a general theoretical framework, from dynamical systems theory, which defines and categorizes types of tipping points in a broad mathematical sense. Next, some of the different types of environmental processes that can give rise to tipping points are identified. Then, the review traces the role of tipping points in creating the Earth system we know today. Turning to the future, focus is put on potential large-scale tipping points and on connections between different types and spatial scales of tipping points. Then, the general prospects for anticipating environmental tipping points, avoiding dangerous ones, and encouraging others are considered. The review concludes by arguing for the need to identify and encourage virtuous tipping points in environmental systems, in societies, and in the couplings between them.
It should already be clear that several different terms, not just tipping point, are being used in various fields (and sometimes in the same field) to describe what appears to be the same basic phenomenon: a small perturbation producing a large change in a system. But if we dig a little deeper, are there actually qualitatively different origins for (or mechanisms behind) tipping point behavior?
The conception of tipping points in climatology (3), of regime shifts in ecology (13), and of critical transitions more generally (5) comes largely from the mathematical theory of bifurcations, which dates back to the late nineteenth century (28). Of particular interest are the subset of catastrophic bifurcations that were the focus of catastrophe theory in the 1960s and 1970s (29). In fact, most of the current focus is on just one type of bifurcation: the “saddle-node” (or “fold”) bifurcation ( Figure 2b ). Steady forcing past such a bifurcation point, where an attractor disappears, causes an abrupt and discontinuous transition to an alternative attractor.
This focus on bifurcations seems rather different from the etymological origin (2) and popularization (1) of tipping points in sociology, where they are associated with the notion that spatially extended systems of discrete agents (humans in this case) can exhibit threshold behavior. Such tipping points are likened to “phase transitions” in physics, where the collective state (e.g., of neighborhood racial composition, or of matter) abruptly switches. The particular spatial structure that allows an idea or an epidemic to spread can also be described as a “percolation threshold” at which an infectious idea or agent is no longer confined in localized pockets but can spread globally across a grid (30). Different orders of phase transition are recognized in physics. In a first-order phase transition, a discontinuity occurs in an observable state variable of a system (density in the case of phase transitions of matter). In a second-order phase transition, a discontinuity occurs in the second derivative of some variable (classically, the free energy of a system); i.e., there is a kink in the gradient. In an infinite-order phase transition, changes are continuous. Phase transitions of different orders can be reversible or irreversible.
The general theory of contagion (31), i.e., the spread of social influences or biological infections, provides a theoretical bridge between these two conceptions of tipping points. In the simplest (nonspatial) models of contagion, a nonzero population of infected individuals only becomes possible past a transcritical bifurcation point ( Figure 2c ). In the language of statistical mechanics, this is a critical point at which a second-order phase transition occurs, which is reversible in this case. However, if some memory of a past (infected/uninfected) state is introduced into epidemic models, then a saddle-node bifurcation can occur beyond which—if a critical mass of the population is infected—an epidemic is possible. This can be likened to a first-order phase transition with a discontinuity in whether a finite infected population exists or not.
What these types of tipping points have in common is the notion that a steady change in some control parameter ultimately leads to a qualitative change in the system state when some threshold is passed. Catastrophic bifurcations and some phase transitions involve irreversible changes (e.g., Figure 2b ), which mean that when the control parameter is returned to the threshold value (ρcrit) it does not revert to its original state. However, other noncatastrophic bifurcations and phase transitions are reversible at the same value of the control (e.g., Figure 2c ). In general terms, all these phenomena can be grouped together as bifurcation-type tipping points, noting that these include both reversible and irreversible cases. This category also includes the special case of systems that show a large change in some state variable for a small change in control because they are close to a fold ( Figure 2a ).
Bifurcation-type tipping points carry generic early warning signals (5). If a system is subject to some source of small short-term fluctuations from which it tries to recover because it is an attractor, then the rate of recovery slows down as the system is steadily forced toward a bifurcation-type tipping point ( Figure 3a ). This critical slowing down behavior is a result of the negative feedback that maintains the attractor state weakening in strength, and positive feedback progressively taking over. It applies to nearly all bifurcations, whether they are catastrophic (e.g., Figure 2b ) or not (e.g., Figure 2c ) (32, 33), and also to the special case of systems close to a fold ( Figure 2a ). Critical slowing down is also typically, but not universally (33), accompanied by rising variance (34). Other changes in higher moments of the statistical distributions of data have also been suggested as early warning indicators but are more specific to particular types of bifurcation tipping points.
However, this does not cover all types of tipping points. Consider, for example, the situation where a system exhibits multiple attractors under its current boundary conditions, e.g., due to a fold ( Figure 2b ), and without any change in external conditions, short-term internal variability within the system causes it to switch between attractors ( Figures 2d and 3b ). The short-term internal variability is typically described as “noise,” and the resulting change is an example of a noise-induced transition (35), which here is called noise-induced tipping (36). The resulting switch can be just as rapid and profound as a bifurcation-induced tipping, and the distinction between these types of tipping is recognized in ecology (15) and climatology (34). Indeed ecologists have documented cases of compounded perturbations causing noise-induced tipping (37, 38). The only mathematical ambiguity is how to define the tipping point. It can be defined in terms of the variable of the system in question that is undergoing abrupt change (F in Figure 2d ), but its value varies with the control parameter (ρ). Thus, it is not as well defined mathematically as a bifurcation-type tipping point.
Purely noise-induced tipping points between attractors are not expected to show early warning signals ( Figure 3b ) because the stability properties of the underlying attractors are not changing (4, 34). Indeed, any kind of rapid forcing (stochastic or otherwise) between attractors will tend to eliminate the possibility of detecting early warning signals. However, the fact that noise is causing a system to sample multiple attractors can be used to determine the number of those attractors (39), and their relative stability or instability (39, 40).
In reality, noise is usually present, and boundary conditions are often changing. Hence, purely bifurcation-type ( Figure 2b ) and purely noise-induced ( Figure 2d ) tipping are abstractions. There is usually a mixture of both processes going on. Although this blurs the possibility of getting a deterministic forecast of future bifurcation, it can offer some alternative early-warning capability; for example, a noisy system that is approaching a bifurcation may show “flickering” events in which it starts to sample an alternative attractor before a more permanent transition occurs (41). More generally, changes in the number and stability of attractors over time can be monitored (39, 42), and the appearance of new attractors or disappearance of old ones can sometimes be detected before bifurcation occurs (39, 42).
Recently, a third type of tipping behavior has been distinguished, “rate-dependent tipping,” which need not be associated with either bifurcations or noise (36). Its key property is that the forcing of a system has to exceed a threshold rate for a runaway change to occur (in which a system abruptly leaves an attractor). The resulting change may ultimately be reversible (i.e., the system eventually returns to the original attractor). A worked example is the “compost bomb instability” (43), where dense concentrations of organic matter (e.g., compost heaps or hay stacks) are subject to a critical rate of external heating, which can trigger runaway breakdown involving escalating internal release of heat by organisms respiring the organic matter, sometimes resulting in a fire. Additional candidates for rate-dependent tipping have been recognized in ecology (44) and ocean circulation (45). Whether rate-dependent tipping carries early warning signals is a subject for research.
There are thus several mathematically distinct potential sources of tipping point behavior (where a small change gives rise to a large response in a system). In summary, a bifurcation tipping point may occur when steady forcing causes an attractor to bifurcate, or shift abruptly, which causes a phase transition to occur; noise-induced tipping is when an internal perturbation causes a system to leave one basin of attraction and enter a different one; and rate-dependent tipping is when a critical rate of forcing causes a long excursion from an attractor.
What these three types of tipping points have in common is the existence of strong positive feedbacks. In environmental systems, different types of processes can give rise to such feedbacks.
Purely geophysical tipping points (not involving life) involve strong positive feedbacks in the fluid dynamics and/or thermodynamics of the ocean, atmosphere, or cryosphere and their coupling to the rest of the climate. They could arise on other planets or could have arisen on Earth before the advent of life. For example, if cooling causes sea-ice cover to reach ∼30° latitude, then a small additional perturbation to ice cover can produce sufficient cooling (through increasing planetary albedo) to produce a runaway enclosure of the planet in ice ( Figure 4a ) (8, 9).
However, many processes in the Earth system are influenced by life (and have been for most of its history). Unlike most nonliving processes, organisms often have peaked responses to environmental variables, such as temperature. If they also affect the same variable sufficiently strongly, this can create the potential for alternative attractors and tipping points between them. A hypothetical example is the initiation (or collapse) of temperature regulation on “Daisyworld” (46) involving strong positive feedback between daisy cover, planetary albedo, and temperature ( Figure 4b ).
Environmental tipping points involving life can be divided into several types, including those involving biogeophysical, biogeochemical, biogeomorphological, and ecological processes.
Biogeophysical tipping points hinge upon strong positive feedback between living things and the geophysical variables that affect them, such as temperature, water supply, or fires. For example, grasslands encourage fires, which in turn tend to exclude trees, producing alternative savannah and woodland attractors ( Figure 4c ) (47, 48).
Biogeochemical tipping points hinge upon strong positive feedback between living things and the chemical variables they influence, such as nutrients or oxygenation state. For example, if the bottom waters of a deep stratified lake or ocean basin become anoxic, this triggers increased recycling of phosphorus from the sediments, which in turn fuels increased productivity above and more anoxia in a positive feedback, which can produce alternative attractors ( Figure 4d ) (25, 49).
Biogeomorphological tipping points hinge upon strong positive feedback between living things and the substrates on which they grow. For example, in tidal systems if salt-tolerant plants can establish, they slow the erosion of their substrate, enabling it to build up and produce a salt marsh above the water ( Figure 4e ) (50). Steeply incised (rapidly eroding) channels provide an alternative attractor. Mud flats can form a third attractor in which diatom-based biofilms secreting extracellular polymers suppress erosion by binding sediment together (51).
Here, biotic-interaction tipping points are defined as those involving strong positive feedback arising from the inherent propensity of life for exponential growth and the direct ecological interactions between living things within food webs. For example, a “trophic cascade” can occur when predators in a food web (sometimes including humans) suppress the abundance of their prey, thereby releasing the next trophic level from predation (52, 53). This category also includes thresholds for epidemics to occur (31). (Environmental variables are not central to biotic-interaction tipping points; hence, they are not a main focus of this review.)
Of course, real-world tipping points can involve a mixture of these types of mechanisms. For example, in shallow lakes, there is a biogeochemical feedback whereby the removal of aquatic plants increases the stirring up and recycling of phosphorus from sediments, promoting the growth of phytoplankton and turbidity, which prevents the plants from reestablishing (54). However, biotic-interaction feedback is also important with aquatic plants providing a refuge for zooplankton from fish grazing and enabling the zooplankton to keep the phytoplankton in check (55). Removing the vegetation opens the zooplankton to fish predation and can release the phytoplankton to bloom in a trophic cascade.
Salt marshes provide another example where biotic interactions between vegetation and grazers can contribute to multistability, e.g., burrowing snails mobilize sediment (enhancing erosion) as well as directly grazing vegetation, but vegetation also suppresses burrowing by snails (56). Outbreaks of grazers (carrying their own positive feedback of replication) can thus trigger runaway erosion (biogeomorphological feedback) in which salt marsh vegetation is abruptly lost (57, 58).
I now turn to Earth history to explore how tipping points have played a crucial role in creating a world in which we could evolve (59).
The origin and spread of life over 3.5 billion years ago (Bya) must have triggered environmental tipping points. The first self-replicating entity represented a small perturbation that would have undergone exponential growth until it was constrained by a lack of resources or unfavorable conditions. Like all life forms, it must have transformed its environment by taking in matter and free energy, transforming them to maintain low internal entropy, and excreting the resulting (high-entropy) waste products (60). The extent of these environmental effects would initially have been limited by abiotic supplies of material resources and free energy. Simulations suggest, however, that as life diversified, tipping points would have occurred each time a recycling loop for a given resource was closed, as this generates strong positive feedback on the growth of participants in the loop (61).
To expand further, life had to access new sources of free energy and materials. The first organisms to have truly global environmental effects would have been those that evolved anoxygenic forms of photosynthesis and hence access to the abundant source of free energy from the sun. Each time life evolved to access some underutilized resource, for example, a new electron donor for photosynthesis, this had the potential to be a tipping point. In particular, if the resulting waste products changed the environment away from what allowed existing species to live, but the organisms causing the change were able to adapt to the new environment, this could cause an “evolutionary regime shift” (62). In artificial biosphere simulations, such tipping points are accompanied by mass extinctions and separate long intervals of environmental stability (62).
The origin of oxygenic photosynthesis (using water as an electron donor) ∼2.7 Bya led ultimately to probably the most profound tipping point in the planet's history: the “Great Oxidation” of Earth's atmosphere in the interval 2.45–2.3 Bya. A simple model (63) suggests this was a tipping point between bistable states for atmospheric oxygen that coexisted under the same boundary conditions (e.g., the input of reducing material to Earth's surface from geologic processes). The tipping mechanism was biogeochemical, hinging on a strong positive feedback associated with the formation of the ozone layer: Once the atmospheric oxygen concentration is sufficient for stratospheric ozone to start to accumulate and shield the troposphere below from UV light, the consumption of oxygen by reaction with methane slows, more ozone accumulates, and there is a runaway positive feedback. Alternative models of the Great Oxidation (64) also show this strong positive feedback and an abrupt rise of oxygen. The Great Oxidation could have involved passing the bifurcation point where the low-oxygen state became unstable, but more likely it would have been noise-induced beforehand once the Earth system entered the bistable regime.
The “Snowball Earth” glaciations, which are postulated to have occurred at least once about the time of the Great Oxidation (∼2.3 Bya) and at least twice during the Neoproterozoic eon (∼0.71 Bya and ∼0.64 Bya), probably also involved passing a tipping point that separated alternative states. In this case, the tipping mechanism was geophysical, involving the increase in albedo (reflectivity) that occurs when sea ice covers the dark ocean surface generating further cooling and sea-ice spread ( Figure 4a ) (8, 9). Some combination of drivers steadily cooling the planet plus perhaps a stochastic cooling event (e.g., a volcanic eruption injecting reflective aerosols into the stratosphere) could have tipped the system. Another bifurcation-type tipping point had to be passed to exit from the Snowball Earth state, requiring the buildup of CO2 in the atmosphere (and possibly dirt darkening the sea ice) until the equator reached melting point and the runaway feedback operated in reverse. Some models show additional stable states, such as a “slush ball,” with open water around the equator (65), but transitions between multiple states all exhibit tipping points and irreversibility (66).
A second rise in oxygen (or lesser oxidation) occurred near the end of the Proterozoic, ∼0.6 Bya, in which the deep oceans finally became oxygenated. This was also the time when the first animals began to flourish, causing a profound ecological transition from very simple ecosystems dominated by cyanobacteria to essentially modern predation-based eukaryote-dominated food webs with multiple trophic levels (67). This might have involved a tipping point between turbid cyanobacteria-dominated and clear-water eukaryote-dominated attractor states on the basis of analogies with modern lakes (67). If so, oxygenation of the deep ocean (and the appearance of animals there) may have been a consequence of the origin of animal predation in the surface ocean.
The past 542 million years of the Phanerozoic eon have in some respects been remarkably stable, with atmospheric oxygen regulated by strong negative feedbacks since at least 370 Mya. Nevertheless, the ocean has shown a tendency to occasionally switch (back) to a state of widespread anoxia, with OAEs recognized in the Paleozoic era and the Jurassic and Cretaceous periods. The mid-Cretaceous (120–80 Mya) series of OAEs are the best characterized, having rapid onsets and terminations and suggesting tipping point dynamics. A potential tipping mechanism is the same strong biogeochemical positive feedback seen in modern stratified lakes (49), whereby anoxic bottom waters cause increased recycling of phosphorus from sediments, which in turn fuels increased productivity above and more anoxia ( Figure 4d ) (25). A recent model suggests that this feedback has the potential to run away globally when anoxic waters start to impinge on the bottom of the continental shelves, where ∼80% of modern phosphorus removal from the ocean occurs, at which point the ocean switches abruptly to a fully anoxic and euxinic (hydrogen sulfide producing) state (26). A systematic increase in the weathering of phosphorus from the land could have tipped the ocean into an OAE (25, 26). Furthermore, it could have forced Earth past a noncatastrophic (Hopf) bifurcation into an oscillating regime, between anoxic events and oxygenated oceans (25). This could explain why OAEs recurred roughly every 5–6 million years in the mid-Cretaceous (25).
One of the candidate drivers (25) for the postulated increase in weathering is the transition from gymnosperm-dominated to angiosperm-dominated ecosystems, which was occurring on land ∼100 Mya and has also been proposed as a tipping point between alternative attractors (68). Angiosperms can produce dense minor veins in their leaves, making them more productive than gymnosperms (69). As a result, they produce more easily decomposable litter and can supply more carbon to roots and mycorrhizae. This can trigger a biogeochemical positive feedback mechanism in which angiosperms promote soil nutrient release through both enhanced litter recycling and mineral weathering, and angiosperms preferentially benefit from this because they have higher growth rates than gymnosperms (68). An additional biogeophysical positive feedback that could have contributed to their takeover from the slower growing gymnosperms is that many angiosperms promote fires (70).
The climate was warm in the Cretaceous and remained so into the Paleogene period, which included a series of abrupt warming events. The Paleocene-Eocene thermal maximum (PETM) 55.8 Mya was the first and most dramatic. Global temperatures rose roughly 5°C within 20,000 years, accompanied by a large release of carbon, and it took ∼100,000 years for the carbon cycle and the climate to recover. There is no sign of a large trigger, suggesting tipping point dynamics. A popular hypothesis is that a modest perturbation—possibly warming caused by changes in Earth's orbit or a volcanic intrusion into carbon-rich sediments under the North Atlantic (71)—destabilized methane hydrates under ocean sediments, leading to large releases of methane, its oxidation to carbon dioxide, and further warming in a positive feedback loop. However, it is unlikely that this feedback became runaway because different locations tend to reach the point of hydrate destabilization at different times, and the warming from each individual methane release event is very small (72). Instead, orbital forcing may have triggered an ocean circulation tipping point whereby intermediate waters, under which much of the methane hydrate reservoir resides, were abruptly and synchronously warmed (73). This warming was slowly amplified by methane release (72), emptying the methane reservoir, which took hundreds of thousands of years to recharge. This model can explain why the PETM was just the first in a series of hyperthermal events of declining magnitude but increasing frequency as Earth steadily warmed until ∼50 Mya (73).
Since ∼50 Mya, the climate has cooled overall, including a particularly abrupt growth of an ice sheet on Antarctica at the Eocene-Oligocene transition ∼34 Mya. This has been modeled as a switch between attractors triggered by declining CO2: Once the climate is cold enough, mountain ice caps can abruptly spread down to lower altitudes and cover a much larger area (74) through a mixture of ice dynamics and positive feedbacks owing to increasing surface albedo and increasing surface elevation (75). This is a geophysical tipping mechanism carrying some irreversibility, as CO2 has to be increased to a higher threshold value for abrupt ice sheet retreat to occur. Early warning indicators are consistent with Antarctic glaciations having involved a bifurcation-type tipping point (76).
Meanwhile in terrestrial ecosystems, the relationships between angiosperm vegetation and fires were evolving (70) and at some point gave rise to alternative attractors (47, 48). Notably, closed broad-leaved forests exclude fire, whereas more recently evolved grasslands are great promoters of fire. Grasslands also support grazing herbivores, which can help maintain grasslands by consuming tree seedlings (59). Tipping between these attractor states could help explain the steplike expansion of grasses, especially the abrupt Miocene expansion of C4 grasses ∼8 Mya, which may have been triggered by declining atmospheric CO2, as C4 grasses outcompete C3 plants under low CO2 (59, 77).
The onset of Northern Hemisphere glaciations in the mid-Pliocene 3.2–2.7 Mya and subsequent shifts in climate dynamics have attracted several theories invoking tipping points (75), which are neatly reviewed elsewhere (78). Models suggest declining CO2 as the most plausible forcing of Greenland glaciation (79), which would likely have occurred through bifurcation tipping (80). The mid-Pleistocene transition from obliquity-driven 41-kyr glacial cycles to 100-kyr cycles (81) has also been interpreted as a bifurcation (75, 78). In this case, a (noncatastrophic) Hopf bifurcation is proposed from a stable fixed point solution (where ice volume tracks obliquity forcing) to a limit cycle in which the internal dynamics of the Earth system produce the 100-kyr periodicity in ice volume, and this is synchronized to orbital forcing (75, 78). There are many alternative models for the recent 100-kyr glacial-interglacial cycles (82–84), but they share common features. In particular, the abruptness of glacial terminations is explained in terms of a bifurcation-type tipping point carrying some irreversibility (82), consistent with early warning indicators preceding the last three terminations (76).
Within the ice age cycles, abrupt shifts have occurred in several elements of the climate system. The warmest interglacial intervals appear to have triggered West Antarctic ice sheet (WAIS) collapse (85, 86), thanks to a geophysical positive feedback that separates multiple stable states for the position of the grounding line, where the ice sheet detaches from the bedrock underwater (87). The Greenland ice sheet (GIS) shrank significantly during the penultimate (Eemian) interglacial period. In the tropics, monsoons have switched on and off, including the Indian South Asian monsoon, linked to climate changes in the North Atlantic (88), and the East Asian monsoon (89), linked to orbital forcing. The underlying strong geophysical positive feedback is the release of latent heat from moist air over the continent, which drives upwelling of that air and the overall monsoon overturning circulation (90). In simple models this produces bifurcation-type tipping points (90, 91), e.g., as net incoming radiation varies.
Most iconic, however, are the repeated Dansgaard-Oeschger abrupt warming events (or D-O events) during the last glacial cycle. They were most marked in the North Atlantic region but had widespread effects. The underlying mechanism continues to be debated, but most studies assign a key role to switches between different attractor states for the Atlantic thermohaline circulation (THC), coupled to changes in sea-ice cover (78, 92). The “null hypothesis” is that the D-O events were purely noise-induced tipping points (34, 39), consistent with an initial failure to find early warning signals (34). However, new results find evidence for some forcing toward a bifurcation-type tipping point when considering the ensemble of D-O events (93).
The final D-O event was overlain on the last deglaciation as abrupt warming into the Bølling-Allerød period ∼14.7 kya, followed by a less abrupt cooling into the Younger Dryas ∼12.8 kya (94). This was followed by a final abrupt warming at the end of the Younger Dryas ∼11.7 kya, marking the start of the Holocene (94). Early warning signals consistent with bifurcation tipping have been detected prior to the Bølling warming (76) and prior to the end of the Younger Dryas (76, 95). This, together with the remarkable stability of the Holocene, suggested that a bifurcation tipping point was passed during the last deglaciation in which the cold climate attractor lost its stability (96). However, existing analyses do not support this hypothesis (96), implying that a noise-induced tipping event may have brought about the warm climate state in which all human civilizations have developed.
Although the current Holocene interglacial period has been relatively stable in the high latitudes, it has included several abrupt changes in hydrological cycling in the tropics (97). Most famous is the termination of the “green” Sahara state in the mid-Holocene, often cited as a tipping point (13, 16). This is based on evidence for an abrupt increase in terrigenous dust influx to marine sediments west of the present desert ∼5.5 kya (98) and accompanying climate model simulations, which show an abrupt change in response to smooth orbital forcing (99). Models suggest that, under the mid-Holocene ∼6-kyr insolation, the green Sahara and the present desert state were alternative attractors (100). The positive feedback separating these attractors involves changing surface albedo, e.g., loss of vegetation in the Sahara region increasing surface albedo, leading to less incoming radiation, radiative cooling of the air, associated sinking of the air, suppression of cumulus convection and rainfall, and amplification of vegetation loss (101, 102). Internal climate variability could have produced noise-induced tipping between the two attractor states 7.5–5.5 kya, before orbital forcing destabilized the green attractor (103). However, an alternative model attributes the abrupt collapse in vegetation ∼5 kya to a threshold response of vegetation to smoothly declining precipitation rather than to a strong biogeophysical feedback (104). Other paleo-records argue for local-scale tipping points in ecosystems being passed at different times in North Africa (105–107). As yet, no early warning indicators of bifurcation have been found in data (108).
The failure of the Australian monsoon to penetrate the interior of the continent during the Holocene is possibly the first example of a human-induced climate tipping point (109) involving the removal of herbivores (40–30 kya) (triggering increases in fire) as well as deliberate burning practices (converting a tree-shrub grassland to desert scrub), and thus suppressing positive feedback between vegetation cover and rainfall (101). Human-induced extinction of megafauna is also hypothesized to have tipped Beringia from steppe grassland to moss-dominated tundra (110). No doubt there have been other smaller-scale environmental tipping points as human civilizations evolved, with the collapse of several civilizations tentatively linked to feedbacks between resource depletion and lock-in to existing practices (111). However, it is with the accessing of unutilized resources, fossil fuels, that human population and impacts on the environment have really escalated. This could represent an evolutionary regime shift (62), threatening a state shift in the biosphere (112).
This brief resume of the making of modern Earth already illustrates several mechanisms for environmental tipping points and several types of systems in which they can occur. But only the largest-scale changes tend to show up in the paleorecord. I now look ahead to the prospects for future large-scale tipping points and drill down from larger to smaller scales to explore the potential connections between different tipping mechanisms across various spatial scales.
Previous work has identified a short list of tipping elements in Earth's climate system (3). Tipping elements are at least subcontinental-scale (1,000 km or greater in length) subsystems of the Earth system that can exhibit a tipping point (3). Particular focus was on those policy-relevant elements that might be tipped this century by human activities and undergo a qualitative change to a new attractor state within this millennium. The short list includes (and was partly identified on the basis of) systems that have exhibited abrupt changes in Earth's recent past: the THC, the GIS, the WAIS, the Indian summer monsoon, and the west African monsoon, and the Sahara/Sahel region. Also included were the El Niño-Southern Oscillation (ENSO), Arctic summer sea ice, Amazon rainforest, and boreal forests. Estimates of the proximity of these tipping points have been collated (3) and elicited from experts (113). Subsequent work has suggested other candidates (20, 21) and has questioned the existence of some tipping points (e.g., ENSO, Arctic summer sea ice) or whether they are accessible during this century (e.g., Indian summer monsoon, Amazon rainforest).
To qualify as a tipping element, tipping should occur coherently across a broad spatial scale (of at least 1,000 km) because the positive feedbacks responsible operate at such a large scale, because their consequences (e.g., megafires or pathogen epidemics) could escalate to such scales, or (less likely) because the climate changes coherently across a large area and this causes a series of smaller-scale tipping points to be passed simultaneously. The first case of large-scale feedback is the most compelling and includes systems that have (or are tightly coupled to) inherently extensive fluid dynamics, covering most of the short list. However, the strength of large-scale feedbacks is debated for the Arctic summer sea ice, Sahara/Sahel, Amazon rainforest, and boreal forests (as discussed below).
The iconic example of a climate tipping point is a collapse of the THC if sufficient freshwater enters the North Atlantic to halt density-driven deepwater formation there. This probably requires >4°C global warming this century (113), although current models are biased toward stability relative to observations that suggest the ocean resides in a bistable regime (114, 115). The THC is robustly forecast to weaken, and as it does so, it may pass a nearer tipping point in which deepwater stops forming to the west of Greenland and is left only forming to the east of Greenland (45). Both tipping points are somewhat rate dependent.
The GIS may have the nearest tipping point (113) at only 0.8–3.2°C (the best estimate is 1.6°C) above preindustrial; beyond this, the ice sheet is committed to eventually disappear (80). We are at the lower end of this temperature range, and the GIS is already losing mass (3). There could be multiple attractors for ice volume, with a first transition involving retreat of the ice sheet onto land and ∼1.5 m of sea-level rise (116), whereas its total loss would add ∼7 m to the sea level. However, the transition is expected to take many centuries (and therefore does not appear abrupt), with the rate depending on the amount by which the tipping point is exceeded (80).
The WAIS is probably further from a tipping point than the GIS, but this is more uncertain (3, 113), especially in the light of new evidence that the GIS was more stable than thought during the last interglacial (117). The WAIS has the potential for more rapid change than the GIS and, hence, greater impacts. Current models and expert elicitation put the threshold for the WAIS collapse at ∼5°C warming of the surrounding ocean (85) or >4°C global warming (113). The WAIS has the potential to cause sea-level rise on the order 1 m per century and 3–4 m in total.
The ENSO changed its dynamics in 1976 or 1977 and has produced some severe El Niño events since (e.g., 1983, 1998). Its pattern has also arguably changed such that the warm water shifts from the west to the middle (rather than the east) of the equatorial Pacific (118). Models disagree over the sign of future changes in El Niño amplitude (119) but generally give no change in frequency. Some models simulate increased El Niño amplitude in the future (119), but ENSO is unlikely to either vanish or become overly strong this century (113, 120). Whether there is any underlying tipping point is highly uncertain (120).
Monsoon systems are expected to exhibit tipping points (90), and the Indian (south Asian) summer monsoon showed a pronounced weakening in the second half of the twentieth century (121). This has been linked to an atmospheric brown cloud haze created by a mixture of black carbon (soot) and sulfate aerosols, causing more sunlight to be absorbed in the atmosphere and less heating at the surface, tending to weaken monsoonal circulation (121, 122). If aerosol forcing continues, a doubling of drought frequency within a decade has been forecast (122).
The loss of Arctic summer sea ice is readily reversible in models in the dark polar winter, leading many to argue that it does not possess a bifurcation-type tipping point (123, 124). However, cloud feedbacks could create a bifurcation on the way to summer ice loss (125). Furthermore, there has already been an abrupt and persistent increase in the amplitude of seasonal ice loss since 2007 (126). This is likely linked to the flushing of thick multiyear ice out of the basin in 2007, leaving a thinner ice cap that is prone to larger retreats each summer. The ice loss is already changing atmospheric circulation patterns and contributing to extreme weather in the midlatitudes (127).
When considering biogeochemical feedback processes, the potential for tipping the ocean into an anoxic event deserves careful evaluation. Currently oxygen minimum zones are expanding globally, linked to climate warming, and increased loading of nutrients from the land is creating anoxic “dead zones” in some coastal regions. Models suggest that runaway feedback is possible should the oxygen minimum zone at intermediate depths expand such that it contacts the bottom of continental shelves, where much phosphorus currently leaves the ocean (26). However, if a transition were triggered, it would take thousands of years to complete thanks to the long residence time of phosphorus in the deep ocean.
Leading theories for the PETM warming event have led many to consider whether there could be a large-scale tipping point for release of methane from hydrates in marine sediments. An ocean circulation tipping point leading to coherent warming of intermediate depths of the ocean could perhaps trigger a large-scale methane release (73). However, current models suggest instead a sequence of small-scale tipping points as various locations reach the hydrate instability threshold at different times. Under a given forcing scenario, these can roughly double the amount of climate change on a long timescale (72).
The potential for tipping points at multiple scales emerges where the land surface and its vegetation play a key role in the feedback dynamics of a system, as is the case for the Sahara/Sahel, Amazon, and boreal forest regions. This also leads to debate about the actual scale at which tipping points can occur (107). Hence, these systems serve to illustrate the potential for different scales and sources of tipping behavior and interactions between them, starting with the Sahara/Sahel.
At the largest scale, the original suggestion (101, 102) of feedback between vegetation and precipitation in the Sahara/Sahel remains compelling, with some models suggesting that it can create bistability, at least in the southwest Sahara, under the present climate (128). In a regional climate model, increasing vegetation and soil moisture below the desert thermal low weakens the African easterly jet (which transports moisture off the content) and greatly increases precipitation (129). If this bistability and the associated tipping points are real, they offer the opportunity to tip the region into a greener Sahara state in the future either as an inadvertent consequence of climate change (130) or through a deliberate revegetation program (131). However, land-use pressure may prevent future greening (130).
At a slightly smaller scale, many people in the Sahel depend upon a tipping point in the form of a seasonal northward “jump” of the west African monsoon each July. Tipping is due to a strong positive feedback in atmospheric dynamics: When the east/west wind changes sharply in the north/south direction, this (inertial) instability causes the northward perturbation of an air parcel to generate additional northward flow (132). The result is a rapid decrease in coastal rainfall and wetting and an associated greening of the Sahel (132). However, if the THC weakens below a threshold value (∼8 sverdrups), this could cause an abrupt warming in the Gulf of Guinea, which would prevent this seasonal northward jump of the monsoon (133). The result could be a large reduction in rainfall in the Sahel (133, 134). However, in an alternative model, collapse of the monsoon actually leads to wetting of the Sahel owing to a reorganization of atmospheric circulation in which moist Atlantic air is drawn into the Sahel from the west (134). This tipping mechanism has been linked to past greening of the southwest Sahara in the mid-Holocene (129).
At intermediate spatial scales, semiarid lands such as the Sahel also exhibit alternative attractors for grassland or savannah (10–50% tree cover) vegetation states (48). Going to smaller spatial scales, biogeophysical positive feedbacks can produce alternative attractors, which can manifest as spatial patterning of vegetation (135). A key positive feedback involves stripes of vegetation on a slight slope retaining water, leaving bare ground between the stripes into which the water cannot infiltrate. Such microscale feedback between vegetation and soil moisture could interact synergistically with macroscale feedback between vegetation and precipitation (101, 102), thus contributing to large-scale bistability and therefore to tipping points (136–138). Global climate models do not yet include such local feedbacks, meaning that they may be underestimating tipping potential in, e.g., west Africa (139). More speculatively, local feedbacks might trigger a cascade of amplifying effects up to larger spatial scales (139), although no direct evidence yet exists for such a scenario.
At small scales, desertification can involve several other different types of strong positive feedbacks that often hinge on switches in the type of vegetation cover. The concentration and localization of soil nutrients by shrubs can propel the transition from semiarid grassland to patchy desert shrubs during overgrazing (140), a biogeochemical feedback. Within deserts, moving sand dunes and static vegetated dunes represent bistable alternative attractors in which the presence of vegetation reduces erosion by wind stabilizing their substrate (141), a biogeomorphological feedback. With climate change increasing wind speeds in some locations, there is the potential for abrupt remobilization of dune systems, e.g., in the Kalahari (142).
Recently the tipping point between savannah and forest has emerged as a ubiquitous bistability separated by strong positive feedback involving fire (47, 48). Furthermore, biotic-interaction feedbacks can act upon biogeophysical feedbacks to create tipping. For example, in the Serengeti ecosystem of east Africa, a trophic cascade triggered by the eradication of rinderpest in wildebeest caused a wildebeest population irruption, suppressing grasses, which in turn suppressed fires, triggering a switch from savannah to woodland (143). In future projections for Africa, local tipping points are forecast, including widespread switches from C4 grassland or savannah to C3 woodland, owing to rising CO2 (144). It is argued that larger-scale tipping points will not occur, but the model used does not include any changes in or coupling to the hydrological cycle (144), so it excludes that possibility by design.
Tipping between forest and savannah states is also a concern in tropical South America, but concerns are for a switch in the opposite direction involving dieback of parts of the Amazon rainforest. The Amazon rainforest recycles around a third to a half of its precipitation helping to maintain itself (a biogeophysical positive feedback) (145). In some models, this gives rise to alternative stable states with or without forest, especially in the drier eastern Amazon where the forest could be replaced by savannah (146). Climate change appears to be increasing variability over the Amazon basin, with extreme droughts in 2005 and 2010 but record wet years in 2009 and 2012. An overall lengthening of the dry season has been attributed to greenhouse gas forcing (147). If the drying trend continues, several model studies have shown the potential for significant dieback of up to 70% of the Amazon rainforest this century (148–151), and its replacement by savannah and caatinga (mixed shrubland and grassland) or seasonally dry forest (152).
These models do not include local positive feedback between vegetation state and fire, yet that independently leads to the prediction that large areas of the eastern Amazon basin are bistable with an alternative, stable savannah state under the present climate (47, 48). Furthermore, there is observational evidence that under extreme drought in the rainforest, there can be a phase transition to a megafire regime in which fires spread over a much larger area (153). Interactions between local-scale positive fire feedbacks, critical threshold behavior in the fires themselves, and vegetation-precipitation positive feedback could thus potentially cause Amazon tipping points.
Moving to the high latitudes, boreal regions exhibit multiple attractors defined in terms of tree cover and temperature, including boreal forest, two partial tree cover states, tundra, and steppe (154). Furthermore, positive feedbacks in boreal forests involving climate, fires, and pest invasion have been identified (155). However, the links between specific positive feedbacks and the separation of different vegetation attractor states have yet to be made. Phase transitions could potentially occur in both fire regimes and pest outbreaks. Fires in parts of the Canadian boreal plains have recently gone through a phase transition to a megafire regime, implying an abrupt and pronounced increase in fire size (156). Meanwhile, bark beetle eruptions (157) are causing widespread tree mortality in western Canada (158) and have turned the nation's forests from a carbon sink to a carbon source (159). These bark beetle eruptions are thought to involve positive feedbacks interacting across scales to cause tipping (157). Drought is also contributing to Canadian boreal forest dieback (160), and in the future, widespread dieback is predicted (161).
The survey of environmental tipping points thus far, selective as it has been, should serve to highlight their ubiquity. So, what can be done about the threat posed by dangerous environmental tipping points? And which (if any) environmental tipping points might be encouraged?
There are two distinct approaches to identifying systems that could exhibit tipping points (6). One is structural, looking for particular mechanisms and network architectures, and the other is statistical, looking for generic indicators of loss of resilience or the existence or appearance of multiple attractors. The structural approach lends itself to better-understood systems, whereas the statistical approach can be applied even when one is ignorant of the underlying positive feedback mechanisms of a system that may give rise to tipping behavior.
The structural approach to identifying potential tipping points includes, as shown here, looking for strong positive feedback in the internal dynamics of a system. When it comes to more complex spatially extended systems that can be characterized as networks, then a general recipe is to look for high connectedness and homogeneity (similarity) of nodes in a network, as opposed to modularity (of structure) and heterogeneity (of nodes) (6). An example of a highly connected, fairly homogeneous network is that of Caribbean coral ecosystems, which showed repeated recovery from local perturbations (e.g., owing to mobile organisms across the network repairing local damage) before undergoing large-scale synchronous collapse in the 1980s (6, 162). This network approach could be usefully applied to the interactions between different tipping elements of the climate system (113) to ascertain if there is any overall structural vulnerability of the Earth system. It is a fairly heterogeneous network with varying strengths (and sign) of connections (113). Hence, different tipping elements are expected to shift at different levels of environmental drivers rather than produce a synchronous global tipping point.
Of course, where one knows a system well enough, it is natural to build a mechanistic model encapsulating this process-based knowledge. Then the key to identifying potential tipping points is to capture the key feedbacks and stability regimes of the real system in the model. Once constructed, detailed models can be used to identify measurable indicators of the stability regime of a system that can be monitored in the real world. An example is the identification of a stability indicator for the THC that has been diagnosed directly from data (114, 115). This has shown that existing models are biased too stable, but as yet, this knowledge has not been used to directly recalibrate a model. However, advanced methods of data assimilation are available for climate- and weather-forecasting models and could be adapted for this purpose.
The statistical approach to identifying potential tipping points tries to derive the stability properties of a system directly from data. This includes looking for the manifestation of multiple attractors as multimodality in time series data or spatial data across environmental gradients (where space substitutes for time) and converting it into a diagnosis of the number and resilience of different attractors in a system (39, 48, 154). It also includes looking for critical slowing down as a signal of the loss of resilience of an attractor state. Recently, successful tests have been made on small-scale experimental systems where biotic-interaction feedbacks create bistability (163–165) and in ecosystem-scale lake-manipulation experiments approaching a trophic cascade (166). However, the experimenters knew their systems well enough to know that they were going to tip them, and they could perform multiple replicates. In the real world, such luxuries are not often available, nor desirable or affordable. For example, we only have one climate system and can only build up a sample size greater than one in the rare cases where multiple events have repeated in the past (93).
Hence, a crucial challenge for developing early warning methods to the point that they can be reliably used on a real system is to draw up an accurate null model of the system to test against (167, 168). This could be a long record of its past behavior prior to human interference, but these situations can be hard to find. Alternatively, it could mean using a process-based model of the system in question. Thus, a complete approach to early warning in reality will often combine the structural and statistical approaches.
Assuming we can achieve some early warning for some damaging future environmental tipping points, when should we intervene to avoid them? The answer of course depends on the system, but there are some general rules that avoidance prospects depend on: the rate that an intervention can be deployed and take effect relative to the proximity of a tipping point, the rate of approach of a tipping point, and the noise level of the system in question. The generic problem is whether an intervention action can come early enough and act fast and strongly enough to avoid a tipping point (169). This has been illustrated for slow (shoreline development) and fast (fishing) interventions in a model lake ecosystem with various early warning signals of an approaching tipping point (169).
However, the problem of inertia is particularly apposite in the climate system, where there are several sources of inertia. First, we have to want to do something to tackle emissions of climate-affecting substances, most notably CO2. After several decades of effort, there is currently no agreed international protocol, and the one that did exist (the Kyoto protocol) was ineffective. There has nevertheless been some progress on the mitigation of CO2 emissions within particular regions, but this is an inherently slow process involving socioeconomic transformation. Then, there are additional delays between changes in CO2 emissions and changes in atmospheric CO2 concentration. Crucially, CO2 concentration continues to rise until emissions fall below the amount absorbed by natural (and any geoengineered) carbon sinks. Also, there is a delay between these concentration changes altering the energy balance of Earth and a response in Earth's surface temperature. If we start deliberate CO2 removal from the atmosphere, there will be a comparable delay in the response of Earth's temperature (170).
Thus, the prospects for avoiding any nearby climate tipping points look slim. Mitigating emissions of short-lived radiative forcing agents (such as methane or black carbon aerosols) or deliberate sunlight reflection methods of geoengineering can act faster to alter the energy balance of the planet and its temperature, but one still has to factor in the social inertia to action in the first place.
Historically, it has often been the case that ecosystems pass a tipping point, taking everyone by surprise. Then, their human managers get another unpleasant surprise when they find themselves having to work far harder to reverse the resulting change than it took to tip it in the first place. Recovering shallow lakes from eutrophication is a classic example. Often attempts at reversal fail because the system is bistable ( Figure 2b ). If system recovery is achieved, it will also involve a (different) tipping point and must depend on the same strong positive feedback dynamics within the system itself. Even in the case of more reversible tipping points ( Figure 2a ), recovering a system will depend on triggering the same tipping dynamics in reverse.
This invites the concept of encouraging some environmental tipping points, where humans intervene in a social-ecological system in a way that works with its natural dynamics to tip it into a more desirable state. Indeed, there is already a mature EcoTipping Points Project that seeks to do just that (171). This positive conception of environmental tipping points (172) builds on examples such as Apo Island in the Philippines, where the creation of a small marine sanctuary ultimately reversed the near collapse of the regional fishery by triggering a range of coupled social and ecological positive feedbacks. The general idea is that good management of a social-ecological system can eliminate an undesirable attractor (173).
The idea of working with environmental tipping points to recover or create desirable system states can be broadened to landscape dynamics, for example, using deliberate vegetation efforts to prevent sand dune mobility. In its boldest form, it could be scaled up to attempt to green the southwest Sahara or Australia (131). Existing proposals involve planting and irrigating vegetation and rely on the idea that an alternative vegetated attractor state would be stable under the present climate. This could be combined with deliberate fire suppression in existing grasslands and savannahs to tip them toward woodlands. If strong positive vegetation-climate feedbacks are present, it may only need a partial forestation program to tip a full transition.
This we might call “Gaia engineering” to distinguish it as a type of geoengineering that seeks to understand and work with natural tipping point dynamics to deliberately transform large-scale subsystems of the Earth system into desired states. This of course brings with it a raft of practical and ethical reservations: Notably, do we understand the systems well enough to have confidence in the outcome of our deliberate interventions? By bringing the deliberate use of existing strong positive feedbacks into geoengineering, it would seem to reduce the predictability of outcomes and raise the already high-risk stakes. But in truth, sunlight reflection methods of geoengineering already pose these dilemmas for tipping elements of the climate system. For example, simulations of Southern Hemisphere stratospheric aerosol injection show that this could tip the Sahel into a wetter (and in due course greener) state (174).
The fact that scientists are even discussing such geoengineering is a stark reflection of our collective inability thus far to act to tackle the causes of climate change. This in turn is just one manifestation of an extreme lock in to unsustainable ways of thinking and collective action.
To conclude, it may be that we need to identify and encourage “virtuous” tipping points in human systems, and their coupling to environmental systems, to achieve sustainability. Tipping points in the realm of changing human worldviews, collective social decision making, and sociotechnical transitions may hold the key to transforming the relationships between human societies and the environmental systems we depend upon.
This starts at the fundamental level of encouraging tipping points in thinking. In the philosophy of belief revision, the theory of coherence suggests that changes in worldview happen through tipping points rather than along a continuum, invoking the analogy of a gestalt shift (175). Such tipping points in which an old view is rejected and a new one adopted have been extensively modeled (175). For example, one additional piece of evidence can be sufficient to tip a person from the view that global warming is a natural fluctuation to the view that it is caused by humans (176). In contrast, the inclusion of emotional attachments, such as avoiding restrictions on fossil-fuel use and avoiding government intervention, can explain why someone else will retain the view that global warming is a natural fluctuation in the face of the same evidence (176). If worldviews are attractors, this can help explain why the belief that humans cause global warming is stagnating or even declining in some societies while scientific evidence continues to accumulate (177). For example, many people have deeply held views that the world is essentially “just,” and dire messaging about global warming (especially if it comes without a clear solution) can then increase denial of its existence (177).
Moving up a level to collective decision making, there is currently an impasse in climate negotiations, with no global agreement on reductions in greenhouse gas emissions, despite decades of concerted effort. A game theory view on this suggests that we are stuck in a “tragedy of the commons” attractor (an inefficient Nash equilibrium), where individual (here national) short-term self-interest to not commit to the cost of emissions reduction is leading to a failure to act collectively (globally) to prevent dangerous climate change, even though that is likely to be much more costly in the long term (i.e., prevention is the efficient Nash equilibrium) (178). However, there could be a tipping point in the form of a “tipping set”—a subset of agents (nations) who by adopting climate control measures (i.e., changing from the inefficient to the efficient equilibrium) can induce all others to do the same (178). Alternatively, if a dangerous climate tipping point is clearly identified, this can change the nature of the game, tipping the negotiations into a coordination game in which collective action to avoid the tipping point is virtually assured (179). This shift in the nature of the game is unaffected by uncertainty about the impact of crossing the tipping point, but uncertainty about the location of a tipping point leads to failure to coordinate actions (179). Arguably, this is the situation at present, giving any progress on early warning efforts for climate tipping points a qualitatively important socioeconomic value.
Beyond changing worldviews and agreeing to do something lies actually transforming our collective actions, which may also involve tipping dynamics. There is often a profound lock in to current technologies in the form of existing infrastructure, subsidies, and policies. Prior investment is known to influence decisions about what course of action to take when the rational choice would be to only consider the future costs and benefits of different options. Such sunk-cost effects at a societal level can give rise to multiple attractors and tipping points between them (111). An example is the huge existing investment in fossil fuels as our dominant energy source. A world supplied by sustainable energy amounts to a different attractor state, and at least one model predicts an irreversible tipping point between them if a carbon tax reaches a critical value (180). Thus, policy can act as a trigger for social tipping, and positive feedbacks can help encourage such sociotechnical transitions, for example, the reinvestment of a fraction of renewable energy captured into the construction of more renewable energy capture devices.
Designing and encouraging such virtuous tipping points would mark a profound transition for Earth as a system, namely the emergence of teleological feedback, involving conscious foresight and planning right up to the global level within the system. This would no doubt take the Earth system on a new trajectory, and it might even alter its eventual fate. As things stand, about a billion years hence, the biosphere is forecast to collapse. As the sun steadily brightens, rock-weathering vegetation that currently cools the planet by removing CO2 will eventually start to overheat and a runaway positive feedback will be set in motion (181). However, if some intelligent life forms remain that far into the future, they may use their understanding of feedback dynamics to delay this ultimate environmental tipping point.
- 1. Environmental tipping points are ubiquitous, occurring in a variety of systems across a range of spatial and temporal scales and involving many environmental processes.
- 2. Tipping points share common underlying mathematical properties in the categories of bifurcation, noise-induced, and rate-dependent tipping points.
- 3. Environmental tipping points can arise owing to purely geophysical (abiotic), biogeophysical, biogeochemical, biogeomorphological, or species-interaction processes.
- 4. Tipping points played a key role in creating the modern Earth system in which humans could evolve, and past tipping behavior suggests systems that could exhibit future tipping points.
- 5. New methods offer the potential to anticipate unwanted environmental tipping points, but inertia in human and environmental systems makes avoiding tipping points challenging.
- 6. Some environmental tipping points present opportunities to restore degraded environmental systems or create preferred states of environmental systems.
- 7. Virtuous tipping points in human systems and their coupling to environmental systems need to be identified and encouraged to trigger a transition to sustainability.
- 1. Tipping points need to be more carefully defined, categorized, and consistently discussed in the academic literature and at the interface between science, society, and policy.
- 2. The ubiquity of rate-dependent environmental tipping points should be explored, together with the scope for identifying their early warning signals.
- 3. The scope for future landscape (biogeomorphological) tipping points to be triggered should be explored, alongside their interaction with other types of environmental tipping points.
- 4. Resolving the underlying tipping mechanism of past abrupt climate changes will help establish the predictability of their corresponding systems and the scale of risk they pose.
- 5. Efforts to anticipate future tipping points should synthesize structural and statistical approaches, including using data assimilation to recalibrate process-based models.
- 6. Intervention methods to avoid an unwanted tipping point or to reduce its impacts should be considered alongside anticipation methods to produce a socially useful early warning.
- 7. A structural assessment of the potential for future global tipping should be made that considers the network of interactions between key elements of the Earth system.
- 8. The potential for future tipping of the ocean into an anoxic event deserves careful evaluation as this has received little attention relative to the danger it could pose.
The author is not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.
I thank John Schellnhuber for first encouraging me to work on climate tipping points, Marten Scheffer for offering a broader perspective, Valerie Livina for extensive collaboration, Yadvinder Malhi for editorial input, the Natural Environment Research Council for support (NE/F005474/1), and the Royal Society for a Wolfson Research Merit Award.