Among the responses of marine species and their ecosystems to climate change, abrupt community shifts (ACSs), also called regime shifts, have often been observed. However, despite their effects for ecosystem functioning and both provisioning and regulating services, our understanding of the underlying mechanisms involved remains elusive. This paper proposes a theory showing that some ACSs originate from the interaction between climate-induced environmental changes and the species ecological niche. The theory predicts that a substantial stepwise shift in the thermal regime of a marine ecosystem leads indubitably to an ACS and explains why some species do not change during the phenomenon. It also explicates why the timing of ACSs may differ or why some studies may detect or not detect a shift in the same ecosystem, independently of the statistical method of detection and simply because they focus on different species or taxonomic groups. The present theory offers a way to predict future climate-induced community shifts and their potential associated trophic cascades and amplifications.
An ecosystem regime shift is often defined as a substantial and relatively rapid shift between two contrasting stable states . A common explanation to the origin of some shifts refers to the theory of alternative stable states [2–4]. The theory stipulates that, for a given system, some alternative stable states or attractors are possible and that the shift from one system to another depends on the size of the attraction basin and the strength of both positive and negative feedbacks. The transition from one state to another is difficult to predict, and the return to initial environmental conditions is not sufficient for the system to switch back in the case of hysteresis . Although some stepwise ecosystem changes may well be caused by the existence of several attractors, their existence remains difficult to prove . Some processes and alternative stable states leading to such a phenomenon have been identified in lakes and in some marine ecosystems [4,5,7]. When the system is controlled by engineer species, such as in coral-dominated ecosystems , or a keystone species, such as in seaweed-structured marine ecosystems , alternative stable states (coral/macroalgae and seaweed/deforested states) and their causes can be well explained . Algal overgrowth in Caribbean coral reefs is mainly attributed to (i) nutrient loading, which stimulates algal growth and (ii) overfishing that reduces the number of herbivorous fish that controls algal proliferation ; similar mechanisms have also been invoked to explain regime shifts in kelp forest ecosystems. In contrast to these local benthic ecosystems, the processes and mechanisms leading to abrupt community shifts (ACSs) in pelagic ecosystems remain difficult to both identify and understand [1,9] and scientists have mainly progressed on the detection of patterns [10–13].
While climate and temperature have repeatedly been assumed to be at the origin of regime shifts in pelagic ecosystems such as those of the Pacific Ocean  and the North Sea [11,14], or more recently in the northeast Atlantic and its adjacent seas [12,13], the mechanisms by which a climate signal may bring about these ecosystem responses [15,16] are unclear. Even the existence of relatively stable states has also been questioned for some systems (e.g. North Sea), which led some authors to prefer the term abrupt community/ecosystem shift (ACS or AES) . Investigating long-term changes in the state of calanoid assemblages in the North Sea, Beaugrand & Ibanez  revealed an abrupt shift in the 1980s between two apparent stable states. As the length of the time series increased, however, the existence of stable states became less apparent  and long-term changes in the community state were more comparable to climatic vacillations . This led Beaugrand et al.  to propose to base the detection of these shifts by the change in the multi-scale multivariate (multispecies) temporal variance of the community, the approach removing the necessity of having a stable state; the higher the variance signature, the greater the magnitude of the shift .
This paper shows that ACSs do not always originate from the transition from one alternative stable state to another but may instead result from the interaction between climate-induced environmental changes and the species ecological niche at the community level. A theory is proposed to explain in detail how an ACS may be instigated and some generalizations are propounded. The theory is then tested in the North Sea using data on copepods sampled by the Continuous Plankton Recorder (CPR) survey.
The ecological theory teaches us that each species can be characterized by its ecological niche. Although different definitions of the niche concept exist , one common and operational definition is the range of tolerance of a species when several factors are taken together . According to this definition, each species can be distinguished by an n-dimensional envelope, which determines its response to spatial and temporal environmental changes. One advantage of the concept of the niche is that it can be seen as a macroscopic elementary property, not only integrating the molecular, physiological and behavioural characteristics of a species but also allowing the consideration of the emergence of new ecological properties at the species level. In practice, the niche is assessed from the physiological or biogeographic knowledge of a species using only a limited number of environmental dimensions .
At a large scale (regional to basin scale), temperature is a key environmental driver that should affect all species, especially ectotherms (i.e. 99.9% of all species on the Earth) [24,25]. In this paper, we will primarily focus on thermal niche, although it is well known that other environmental parameters (e.g. photosynthetically active radiation, bathymetry, oxygen, mixed layer depth) may also play a role . A one-dimensional (thermal) niche with a Gaussian shape has a few remarkable points (figure 1). When the species is at the centre of its thermal niche (between TS1 and TS2; S for stable), the species is resistant to climate-induced temperature changes, the resistance being dependent on the shape of the niche, the breadth of the thermal window, and more specifically the breadth of the optimal zone [27,28]. When the species is present in an environment with a thermal regime that corresponds to the edge of its thermal niche (THV; HV for high variability), it becomes highly sensitive to climate-induced changes in temperatures. Towards the limits of the niche, the variability levels off (TD; D for detection) before reaching the lethal temperature TL. At this level of abundance, the influence of climate becomes difficult to detect and the location of TD depends on the noise-to-signal ratio . Such a theory, which may be coined the Macroecological Theory for the Arrangement of Life (METAL theory) enables us to understand why the species' responses to climate-induced temperature change vary in space and time. It has been applied to plankton (Calanus finmarchicus and Acartia tonsa), fish (the Atlantic cod Gadus morhua) and seabirds (the little auk Alle alle) [28–31]. It has also been used to understand the global ecogeographical pattern in biodiversity (foraminifers and copepods) and the Rapoport's effect (i.e. the increase in the mean latitudinal range of a species with latitude) , as well as to connect phenological and biogeographical shifts at the species level and long-term shifts at the community level . The shape of the niche was not always Gaussian in all these studies. Some were determined by non-parametric ecological niche models  and some niches were even rectangular to relax the constraints related to the shape of the niche .
3. A theoretical framework
A simple model can show why ACSs may also result from the individual response of each species to climate-induced changes in temperature. The procedure developed in Beaugrand et al.  was transformed to produce a number of hypothetical (Gaussian) niches for both eurytherms and stenotherms. The response curve of the expected abundance E of a pseudospecies s to change in sea temperature in a given site i and time j was modelled as follows : 3.1with Ei,j,s the expected abundance of a pseudospecies s at location i and time j; cs the maximum value of abundance for species s fixed to one; xi,j the value of temperature at location i and time j; us the thermal optimum and ts the thermal amplitude for species s (figure 1).
Figure 2 shows some examples of niches that can be generated from equation (3.1) along a thermal gradient. Eight niches were created in this example with thermal optima of us = 12°C (black) and us = 14°C (blue) for more psychrophiles, us = 18°C (green), us = 22°C (orange) and us = 24°C (red) for more thermophiles. The thermal amplitude was ts = 2°C for more stenotherms (dashed line) and ts = 4°C for more eurytherms (filled line). Substantial shifts in the abundance of these pseudospecies, expected from the knowledge of their thermal niche, took place when temperature was artificially increased from 17.5°C to 18.5°C (middle of time series). A single degree of temperature rise was sufficient to trigger abrupt changes in the abundance of some eurytherms and stenotherms, especially those with an initial thermal regime close to THV (figure 2a,b; an increase for more thermophiles in orange and a decrease for more psychrophiles in blue). No shift was observed for pseudospecies with an initial thermal regime close to Topt (green). The magnitude of the change was greater for stenotherms because they are more sensitive to temperature (figure 2a,b). However, stenotherms with an initial thermal regime close to the edge of their thermal niche (outside TD) exhibited only minor changes in abundance (figure 2a,b; black and red).
The same expectations emerged when the abundance of pseudospecies were assessed from a time series of temperature having the same variability as long-term changes in temperatures observed in the North Sea for the period 1958–2010 but a different average (mean average = 17.5°C versus 10.2°C in the North Sea; figure 2c). When the eight pseudospecies were combined and their long-term changes analysed by standardized principal component analysis (PCA on the table years × pseudospecies), the first principal component (PC, 91.4% of the total variance) revealed major changes in this pseudocommunity, including a stepwise and rapid shift at the end of the 1980s (figure 2d). A change-point analysis  detected a shift in 1988 (figure 2d; in red), corresponding to the timing of the regime shift originally proposed in the North Sea . In such ecosystems, the framework predicts that the response of the community is directly a function of the magnitude of the thermal shift and the degree of stenothermy in the community.
To demonstrate that a community composed of stenotherms should be theoretically more sensitive to long-term changes in temperature than eurytherms, we created strict stenothermic (thermal amplitude ts varying from 1°C to 2°C, every 0.2°C) and eurythermic (thermal amplitude ts varying from 9°C to 10°C, every 0.2°C) pseudocommunities with different thermal optima (figure 3). For a thermal optimum varying between 8 and 12°C (figure 3a(i),b(i)), long-term changes in expected abundance varied more for stenotherms (figure 3a(ii)) than eurytherms (figure 3b(ii)). Despite this, a PCA performed on each pseudocommunity gave similar results (first PC in figure 3a(iii),b(iii)). Similar results were found for pseudocommunities with a thermal optimum varying between 5 and 17°C (figure 3c,d) and 0 and 25°C (figure 3e,f). In particular, long-term changes in the expected abundance of stenothermic species crossed each other (figure 3e(ii)) at the time of two ACSs detected by change-point analysis ca 1988 and 1998 (figure 3e(iii)). The two first PCs were also similar, although the second shift (1998) was more apparent for the stenothermic than the eurythermic pseudocommunity (figure 3e versus 3f(iii)). Note also that the second shift was not detected in the simplified pseudocommunity of the first example (figure 2d).
The amplitude of year-to-year changes or first differences (d(2i+1)/2 = xi+1 − xi with i varying between 1 and t − 1) were then calculated for all long-term changes in the expected abundance of stenotherms (figure 3e(ii)) and eurytherms (figure 3f(ii)) to estimate the theoretical sensitivity of these two pseudocommunities to changes in temperature (figure 4). Only species with an average expected abundance above 0.1 was considered in this analysis. The year-to-year amplitude of the stenothermic pseudocommunity (ranging between −0.6 and 0.6) was higher than the year-to-year amplitude of the eurythermic pseudocommunity (ranging between −0.1 and 0.1). This analysis demonstrates that the sensitivity of a community should depend upon the degree of stenothermy/eurythermy of species.
The application of the METAL theory in the context of regime shift leads to several predictions: (i) a stepwise shift in temperature leads to an ACS; (ii) a long-term shift in temperature will be amplified at the community scale by the interaction between the thermal niche of a species and the thermal regime, this being more prominent when the number of stenotherms is higher in the community or when the thermal regime of the region is close to the point THV of eurytherms (figures 1 and 2); (iii) some species will not exhibit a stepwise response to change in temperature during an ACS (e.g. species close to their optimum and species at the edge of their thermal niche outside TD).
4. Test of the theoretical framework for the North Sea ecosystem
(a) Generation of the pseudocommunities
As in Beaugrand et al. , we produced pseudospecies that were able to colonize the North Sea (4° W, 10° E, 51° N, 60° N), so long as they could withstand monthly changes in sea surface temperatures (dataset ERSST_V3). All species had a unique thermal niche after the principle of competitive exclusion of Gause . The generation of pseudospecies was made using equation (3.1) for the period 1958–2009. Many pseudospecies were produced, each having a unique thermal niche with an optimum temperature ranging from 4°C to 25°C by 4°C increment and a thermal amplitude ranging from 0.1°C to 10°C by 2°C increment. However, only pseudospecies with an annual relative (i.e. expressed as percentage) abundance greater than 0.01 and a presence greater than 10% for all years of the period 1958–2009 were kept .
(b) Comparison with long-term changes in the abundance of copepods
A first comparison was performed for the time period 1958–2009. The annual mean of all North Sea copepods sampled from the CPR survey was calculated. A total of 52 species, which had an annual relative (i.e. expressed as percentage) abundance greater than 0.01 and a presence greater than 10% for all years of the period 1958–2009 , were retained. As described above, changes in the community state were inferred from the calculation of the first PC of a standardized PCA performed on the abundance (log-transformed) matrix 52 years × 52 species or taxa. The first PCs from the PCA performed on pseudospecies and copepods were then compared.
As more pseudospecies were generated than observed copepods (86 pseudospecies versus 52), we chose 52 pseudospecies at random with which to examine long-term changes in the pseudocommunity state in the North Sea. The pseudocommunity state was assessed by the first PC of a standardized PCA on the table 52 years (1958–2009) × 52 pseudospecies. We repeated the selection of the pseudospecies 10 000 times and recalculated each time the first PC on pseudospecies.
(a) Predicted long-term changes from the model
A total of 10 000 expected long-term community states were generated (figure 5; curves in black). Positive values reflected a warmer dynamic regime and negative values a colder dynamic regime. Most long-term expected patterns exhibited a warm dynamic regime between 1958 and 1962 and then a colder dynamic regime until the end of the 1980s, periods well identified in previous studies . The cold-ocean climate anomaly of 1978–1982 was also correctly reproduced . After an abrupt shift at the end of the 1980s, the system remained in a warm dynamic regime and a second abrupt shift was detected at the end of the 1990s [12,41].
(b) Comparison with long-term changes in the abundance of copepods
When long-term changes in copepod abundance were examined by a PCA (figure 5a; in red), the first (approx. 1988) and second (approx. 1998) shifts were well synchronous to expected patterns. The first PC explained 25.96% of the total variance. Patterns of changes predicted by the theoretical framework were closely related to observed long-term changes (figure 5a). The observed warm dynamic regime of the period 1958–1962 was, however, less apparent. The correlation coefficients between the 10 000 long-term expected patterns and observed long-term community shifts in copepod abundance were rmean = 0.72 (rmin = 0.68 and rmax = 0.76). All correlations were significant after accounting for temporal autocorrelation at the threshold p < 0.05 .
(c) Comparison with long-term changes in calanoid assemblages
It is remarkable that the pattern predicted by the theoretical framework was also closely related to observed long-term changes (percentage of explained variance of the first PC: 32.46%) in calanoid assemblages (shift in 1962, similar oscillations during the colder dynamic regime; figure 5b). The predicted 1988 shift was observed earlier when calanoid assemblages were examined, however (figure 5a versus 5b). The correlation coefficients between the 10 000 long-term expected patterns and observed long-term community shifts in calanoid assemblages were rmean = 0.47 (rmin = 0.43 and rmax = 0.55). All correlations were significant after accounting for temporal autocorrelation at the threshold p < 0.05.
Current theories suggest that shifts from one ecosystem state to another may be related to the presence of attractors [2,6]. Although they are difficult to identify in the field, there is evidence that attractors play a key role in the catastrophic shifts of some natural systems . For example, attractors have been suggested for lakes where the system alternates between clear and algae bloom phases and in some marine systems, e.g. shifts from a cod- to clupeid-dominated system state in the Baltic Sea . In the benthic realm, regime shifts are also frequently reported at a local scale and some are well explained by referring to the theory of alternative stable states . However, using a simple framework, this paper shows that another mechanism may be at the origin of some climate-induced ACSs reported in the scientific literature (e.g. the North Sea), without the need to invoke the theory of alternative stable states.
Here I propose that climate-induced temperature changes may also cause ACSs through the interaction between temperature changes and the thermal niche of each species composing a community. From the simple framework used in this study, two theorems can be proposed. The first theorem states that ‘providing that the niche has a Gaussian shape and that there is no species interaction, a substantial increase (e.g. a 1°C of mean annual temperature) in the thermal regime of an ecosystem triggers an ACS. The magnitude of the community shift depends on the extent of the temperature change and the degree of eurythermy (stenothermy) of the species composing a community. A higher degree of stenothermy increases the sensitivity of the community to temperature change’. In practice, the first theorem should be robust from departure to the condition on the niche shape. It is well known that the shape of the niche of some species is not Gaussian and may exhibit large differences in shape . However, the niche must be unimodal. The magnitude of the observed shift is likely to strongly depend on the noise-to-signal ratio, which is influenced by the quality of the data (e.g. sampling and species identification). The noise-to-signal ratio influences the location of the point TD along the thermal niche (figure 1). This phenomenon may also be at the origin of substantial changes in the timing of a shift. Such a problem is more likely to happen when the change in the thermal regime is small.
The second theorem states that ‘during a climate-driven ACS, the response of species is individualistic, depending upon the characteristics of their thermal niche, the initial thermal regime and the magnitude of the thermal shift. It follows that not all species are expected to show a shift (e.g. species located around Topt or outside TD) and that some may react earlier (stenotherms for an initial thermal regime close to THV) than others (eurytherms or species with an initial thermal regime close to Topt or outside TD)’. This second theorem has strong practical implications. According to species or taxonomic group, the magnitude and the timing of the shift may vary, independently of the type of statistical procedures that also influence the timing . In the North Sea, it has been suggested that the timing of the shift varies according to the selection of species, taxonomic group and species assemblage [11,14,17,18]. It is also possible that some studies focusing on the same ecosystem find a shift, whereas others do not, independently of other issues related to statistical technics or sampling programmes. This prediction was observed in this study. Using a small number of pseudospecies, we found a unique shift (ca 1988; figure 2d). By contrast, when a higher number of pseudospecies was examined, we detected two shifts (ca 1988 and 1998, figure 3e,f).
Two further assertions are proposed here. The first assertion stipulates that climate-caused ACSs are highly correlated to temperature changes, the intensity of the correlation depending on species composition and the noise-to-signal ratio. Although this first assertion needs to be better evaluated, it is expected that if patterns of temperature change exhibit a long-term trend, a cyclical variability and a strong year-to-year variability (very common pattern), long-term changes in the community state may show vacillations characterized by a succession of periods of low and higher temporal variance . An example of such vacillations is shown in figure 5a (black and red curves). When only a relatively small number of years are considered, these vacillations may look like a (community) regime shift; a community regime shift can be defined as a special case of ACS between two stable community states (i.e. state with no temporal autocorrelation).
The second assertion stipulates that the sharpness of the ACS depends not only on the abruptness of the temperature change but also on the degree of eurythermy (stenothermy) of species. A high degree of stenothermy may lead to a more stepwise shift whereas a large degree of eurythermy may lead to a longer shift. This was shown in figure 3e,f and suggested in figure 4. Here also, the noise-to-signal ratio can interact and move the exact theoretical timing of the shift .
Other mechanisms have been proposed to explain how climate-induced temperature change may affect aquatic ecosystems [44,45]. In shallow lakes, North Atlantic Oscillation-induced changes in temperature can affect the occurrence of clear water phases . Using a periodically forced algae–zooplankton model, Scheffer et al.  proposed that both plankton and food web dynamics can be modulated by temperature by altering the seasonal patterns and the biomass of phytoplankton and their interactions with zooplankton and fish. It is, however, unknown whether such a mechanism can be translated to the marine environment. Using time series investigated to characterize the 1976/1977 Pacific regime shift, Hsieh et al. showed that whereas time series of physical variables (e.g. Pacific Decadal Oscillation index, Southern Oscillation Index) were linear stochastic, biological time series (e.g. California Cooperative Oceanic Fisheries Investigations copepods, sockeye salmon, chum salmon) exhibited a nonlinear signature. These results supported the idea of a nonlinear amplification of stochastic hydro-climatic forcing. However, the authors did not provide any mechanisms to explain how the nonlinear amplification takes place. The theory proposed here shows that ACSs may arise from the nonlinear interaction between the species niche and the year-to-year fluctuations in temperature.
The theory outlined in this study is simple. It is well known that ecosystems and their biodiversity are influenced by a multitude of environmental parameters, not only temperature. The proposed theory is here tested using a one-dimensional (thermal) niche. Although this thermal dimension is probably important for all species , it is by no means the only dimension that conditions the species’ response to environmental (even climate) changes . Climate also has a strong influence through photosynthetically active radiation, the degree of stability of the water column, oxygen concentration and the amount of both macro- and micro-nutrients [46,47]. Therefore, changes in atmospheric forcing, which also affect oceanic circulation at many spatial scales, may have a strong effect on the ecosystem state . It is likely that the consideration of all these dimensions will reinforce the theory rather than weakening it, although this point requires more investigation.
The proposed theory does not consider species interaction yet. Species interactions are often involved in the positive feedbacks that shift a system from one state to another . Because of these species interactions, climate-induced species shifts may propagate through the food web [41,49]. Both bottom-up or top-down controls have been suggested in the pelagic environment [50,51]. Trophic cascades may be observed if a key or engineer species is affected by temperature changes. These environmentally induced trophic cascades should not be confounded with anthropogenic trophic cascades related to pollution or overfishing  for which the proposed theory is not relevant. The likelihood of natural trophic cascades may increase at times of substantial temperature change. The proposed framework suggests that the null hypothesis of a common response of all species to climate-induced changes in temperature should also be tested when climate-induced trophic cascades are hypothesized . Kirby & Beaugrand  proposed the term trophic amplification, when climate-induced temperature changes have a significant influence on each species, and that this influence is also exacerbated through the trophic food web.
Other mechanisms have been proposed to explain observed regime shifts in some exploited marine ecosystems . These hypotheses generally invoke fishing, which may unbalance species interactions and lead to persistent ecosystem changes. The prey-to-predator loop is based on the fact that many prey feed on the eggs or larval stages of their predators . When the predator declines because of an external force such as fishing, this can lead to the outbreak of its prey, which may in turn prevent the recovery of the predator because of predation pressure at the egg or larval stage. A key assumption here is the reduction of the predator by fishing. Note that this hypothesis may also hold if the predator has been affected by adverse environmental conditions. The predator pit loop is a hypothesis that is based on the empirical observation that two opposite states seem to occur for forage (prey) fish population, one characterized by a low abundance (refuge) and another characterized by outbreak abundance . This hypothesis invokes prey–predator interactions. When the prey becomes low, the predator searches for alternative preys and the specific mortality rate of the prey diminishes. This virtual refuge allows the species to survive and to subsequently recover. When the prey becomes more abundant, the predator will only catch a constant fraction corresponding to the satiation point and the prey is likely to continue to increase. In some circumstance, this can lead to an explosive increase. These two hypotheses explain the persistence of an ecosystem shift but are less effective in explaining the origin of the change, especially in the absence of fishing. Trophic interactions could be implemented in the present theory and would probably lead to an increased sensitivity of the system to temperature change.
This paper provides evidence that some climate-induced ACSs may result from the nonlinear interaction of the thermal niche of species with temperature change at the community level. Because temperatures are expected to rise rapidly with global climate change in the next decades, these ACSs are likely to increase in both frequency and intensity. The term regime shift has often been used in the past to describe stepwise changes in the ecosystem/community state. Because time series were relatively short (i.e. a few decades), two full dynamic regimes (apparently stable states) and a shift were frequently observed [10,11,18]. Since then, time series have increased in length and have started to reveal more complex temporal patterns . The deterministic theory outlined in this paper explains why such complex patterns forms. Soon, it may be become apparent that we are observing ecosystem/community state vacillations, where ACSs alternate with periods of more relative stability. These periods of relative stability may not be confounded with stable states. The present theory offers a way to predict future climate-induced ACSs that may be at the origin of trophic cascades and amplifications.
One contribution of 16 to a Theme Issue ‘Marine regime shifts around the globe: theory, drivers and impacts’.
- © 2014 The Author(s) Published by the Royal Society. All rights reserved.