Climate-induced coral bleaching events are a leading threat to coral reef ecosystems and can result in coral–macroalgal regime shifts that are difficult to reverse. It is unclear how different factors causally influence regime shift or recovery trajectories after a bleaching event. Here, we use structural causal modeling (SCM) and its application of directed acyclic graphs (DAGs) to determine how key factors affect regime shift versus recovery potential across coral reefs in Seychelles, which were severely impacted by bleaching events in 1998 and 2016. Our causal models reveal additional causal drivers of regime shifts, including initial macroalgal cover, wave exposure, and branching coral cover. We also find that reduced depth and structural complexity and increased nutrients increase the likelihood of regime shifting. Further, we use a DAG-informed predictive model to show how recovering reefs are expected to change after a recent 2016 bleaching event, suggesting that three out of 12 recovering reefs are expected to regime shift given their predisturbance conditions. Collectively, our results provide the first causally grounded analysis of how different factors influence postbleaching regime shift versus recovery potential on coral reefs. More broadly, SCM stands apart from previous observational analysis and provides a strong framework for causal inference across other observational ecological studies.

INTRODUCTION

Climate-induced coral bleaching is currently one of the leading threats to coral reef ecosystems and is expected to be an increasingly frequent stressor for coral reefs in the future (Hughes et al., 2018). A potential long-term consequence of climate-induced bleaching events is that they can lead to a coral–macroalgal regime shift, whereby the benthic composition abruptly transitions from a coral-dominated reef to one dominated by macroalgae (Graham et al., 2015). Regime shifts have become a key concern for coral reef conservation as they represent substantial change and degradation of coral reefs worldwide, which are often difficult to reverse (Bellwood et al., 2004). For example, climate-driven regime shifts have led to altered trophic structure, diversity, and species composition of reef fish communities (Hempson et al., 2017; Robinson, Wilson, Jennings, & Graham, 2019), as well as increased catch instability and fishery dependence on herbivorous fish (Robinson, Wilson, Robinson, et al., 2019). It is important to note that not all coral reefs shift toward algal domination after a bleaching event (e.g., Gilmour et al., 2013), and past research has found correlations between key predictor variables and regime shift versus recovery trajectory (Graham et al., 2015). However, these findings were not grounded in causal inference, the deliberate use of specific methods to infer causation (Pearl, 2009).

A literature review of causal inference in coral reef ecology (Appendix S1: Section S3) shows that no observational studies to date have employed causal inference methods to determine relationships for reef regime shifts; however, most studies used causal language to communicate their results (e.g., “the effect of X on Y”). With the development of structural causal modeling (SCM; Pearl, 2009), there is an opportunity to revisit these analyses to understand causal effects of factors influencing regime versus recovery trajectories on coral reef ecosystems. SCM is a causal inference method that can be used to determine causal relationships from observational data. It uses directed acyclic graphs (DAGs) to visualize the causal structure of a system under study, which is then used to guide covariate selection required for observational causal inference (see Methods section for details). Already, DAGs have been applied across several ecological studies, leading to more informed insights across study systems (Cronin & Schoolmaster, 2018; Schoolmaster et al., 2020).

Here, we employ SCM to determine how key factors have influenced recovery versus regime shift trajectories after a widespread bleaching event in Seychelles. The mass coral bleaching event of 1998 reduced coral cover by over 90% across 21 coral reef sites in the inner Seychelles (Graham et al., 2015). Postdisturbance trajectories in cover of coral and macroalgae resulted in approximately half of these sites recovering live coral, while the other half shifted toward macroalgal domination. In addition, the 2016 bleaching event further impacted reefs across Seychelles, reducing coral cover by 70% on those reefs that had recovered from the 1998 disturbance (Wilson et al., 2019). It is currently unclear whether reefs that recovered from the 1998 bleaching event will recover a second time, or undergo regime shifts to macroalgal dominance. Applying SCM to this unique and well-studied system, our study addresses two research questions: (1) Does SCM lead to additional insights on the causal drivers of regime shifts following the 1998 bleaching event in Seychelles? and (2) Which of the reef sites that recovered following 1998 are expected to regime shift as a result of the 2016 bleaching event? By employing SCM, our study aims to better understand the causal factors influencing coral–algal regime shift dynamics.

METHODS

Ecological surveys

Seychelles were impacted by a widespread climate-induced coral bleaching event in 1998 (Goreau et al., 2000). Pre- (1994) and postbleaching (2005, 2008, 2011, and 2014) surveys of 21 coral reefs throughout the inner Seychelles Islands were conducted using identical methods (see Graham et al., 2015 for details). Coral reefs were categorized as either regime shifting or recovering based on data collected before and 16 years after the 1998 bleaching event. Regime shifting reefs had postdisturbance macroalgal cover greater than coral cover and patterns through time showing high and/or increasing cover of macroalgae over time. In contrast, recovering reefs had postdisturbance coral cover greater than macroalgal cover and patterns through time showing high and/or increasing levels of coral cover over time. In total, 12 reefs were classified as recovering and nine were classified as regime shifting (see Graham et al., 2015 for details).

Causal framework

Pearl's SCM (Pearl, 2009) framework uses DAGs to visually represent the causal structure of a system under study. Specifically, nodes within a DAG represent variables, with directed arrows between nodes representing possible causal effects (e.g., X → Y shows that X affects Y). A lack of arrow assumes no causal relationship between variables, and these represent our priori assumptions about where causality cannot occur (Elwert, 2013). DAGs must be acyclic, meaning that they cannot contain bidirectional relationships or a feedback loop where a variable either directly or indirectly causes itself (Elwert, 2013). However, DAGs may still represent ecological systems with bidirectional relationships by more finely articulating the temporal sequence of events (Greenland et al., 1999). DAGs are also nonparametric, making them compatible with a wide range of statistical analyses (Glymour & Greenland, 2008).

The first step of a SCM is to create a DAG (Step 1, Figure 1). DAGs should be created and justified based on the accumulation of domain knowledge, which can include expert opinion and past and ongoing research. DAGs should include all measured and unmeasured variables required to depict the system or process under study, as well as all common causes of any pair of variables included in the DAG (Glymour & Greenland, 2008; Spirtes et al., 2001). Here, we have created a DAG representing how variables may influence regime shift versus recovery trajectories in Seychelles (Figure 2) based on ecological knowledge, and our own past and ongoing research.

Details are in the caption following the image — **FIGURE 1**
Open in figure viewer PowerPoint

Workflow for going from directed acyclic graphs (DAGs) to causal inference under the structural causal model (SCM) framework

Several key factors are assumed to influence regime shift trajectory after a climate-induced bleaching event: marine protected area (MPA), depth, nutrients, branching coral, structural complexity, herbivorous fish biomass, wave exposure, and predisturbance macroalgal cover (Figure 2). We note that we used 1994 (predisturbance) data for branching coral, macroalgal cover (not included in the Graham et al., 2015 analysis), herbivorous fish biomass, and structural complexity because we wanted to know how their condition prior to the 1998 bleaching event would influence regime shift versus recovery trajectory and to resolve any bidirectional relationships that may exist between our predictor variables and response (Glymour & Greenland, 2008). A detailed rationale of each directed arrow in our DAG is presented in Appendix S1: Section S1.1.

Once a DAG is created, it can be checked for DAG-data consistency (Step 2, Figure 1). Simply put, a specified DAG will have (often many) independencies between variables (e.g., A is independent—or d-separated—from B, if C is adjusted for) that must be compatible with the dataset it represents (Pearl, 2009). If all implied independencies are compatible with the data, it provides support for a DAG. We tested our DAG for DAG-data consistency using the R package “dagitty” (Textor et al., 2016), which verified that all 32 independencies implied by our DAG were consistent with our observational data (Appendix S1: Section S1.2).

A finalized DAG is then used to guide covariate selection required to determine causal effects. This is the critical step that separates SCM from correlative observational studies. Specifically, a graphical procedure known as the backdoor criterion guides covariate selection required to determine the causal effect of X on Y (Pearl, 2009; Appendix S1: Section S1.3). In short, the backdoor criterion instructs us to block all noncausal pathways (i.e., backdoor paths) between our predictor and response variable of interest, while leaving all causal paths open. As such, the application of backdoor criterion eliminates common statistical biases that can otherwise plague observational studies, including confounding, overcontrol, and collider bias. Here, we employ the backdoor criterion to guide covariate selection for each predictor variable expected to influence recovery versus regime shift trajectory (Step 3, Figure 1, Appendix S1: Section S1.3). We note that only the effect size of the predictor variable of interest is interpreted for its associated model, with additional variables acting as required controls. This approach differs from the “causal salad” model (McElreath, 2020) commonly used throughout ecology—including in our own work—where all assumed predictor variables are placed into one model and subsequently interpreted.

Once the backdoor criterion is applied for covariate selection, researchers must choose an appropriate statistical model (Step 4, Figure 1). As DAGs are nonparametric, they make no assumptions about the distribution of variables (e.g., normal) or the functional form of effects (e.g., linear, nonlinear, stepwise), making them compatible with a wide range of statistical methods. Here, we applied a Bayesian logistic regression analysis to each of our causal models, where the response variable was 0 for recovering sites and 1 for regime shifting sites. We standardized our data by subtracting the mean of each variable and dividing by 2 SD in order to assess relative effect sizes of our predictor variables (Gelman & Hill, 2007). We ran our models using the “rethinking” package on R, using weakly informative priors. Our final Bayesian logistic regression models (one for each predictor variable) were as follows:

Y_{i} \sim Bernoulli (p_{i})

MPA model : logit (p_{i}) = β_{0} + β_{1} MPA

Nutrient model : logit (p_{i}) = β_{0} + β_{1} Nutrient + β_{2} Depth

Herbivore biomass model : logit (p_{i}) = β_{0} + β_{1} Herbivore biomass + β_{2} MPA + β_{3} Structural complexity

Branching coral model : logit (p_{i}) = β_{0} + β_{1} Branching coral + β_{2} MPA + β_{3} Depth + β_{3} Wave

Depth model : logit (p_{i}) = β_{0} + β_{1} Depth

Structural complexity model : logit (p_{i}) = β_{0} + β_{1} Structural complexity + β_{2} Branching coral

Wave model : logit (p_{i}) = β_{0} + β_{1} Wave

Initial macroalgae model : logit (p_{i}) = β_{0} + β_{1} Macroalgae + β_{2} Herbivore biomass + β_{3} Depth + β_{3} Nutrients + β_{4} Wave

Priors (for standardized data):

β_{0} \sim Cauchy (0, 10)

β_{1, \dots, N} \sim Student_t (4,0,2.5)

DAG-informed predictive model

We further created a causally guided predictive model to determine how recovering reef sites would be expected to respond to subsequent bleaching events in 2016 (n = 12). Predictor variables included all factors that were found to directly influence this response based on our previous DAG-based analysis, which were depth, nutrient, branching coral cover, structural complexity, wave exposure, and initial macroalgal cover (see Results section). This approach captures all relevant variables assumed to influence a response variable of interest, without including or excluding variables that can influence predictive estimates noncausally. We excluded herbivorous fish biomass because our results suggested that our coarse biomass metric may not be representative of herbivore grazing effects (see Discussion section). We employed a Bayesian logistic regression model, with 0 for recovering sites and 1 for regime shifting sites. Our final predictive Bayesian logistic model is specified as:

Y_{i} \sim Bernoulli (p_{i})

Predictive model : logit (p_{i}) = β_{0} + β_{1} Depth + β_{2} Nutrient + β_{3} Branching coral + β_{4} Structural complexity + β_{5} Macroalgae + β_{6} Wave

Priors (for unstandardized data):

β_{0, \dots, N} \sim N (0, 10)

We used data from the 1998 bleaching event to train our model. A posterior predictive check (McElreath, 2020) showed that our predictive model was able to correctly identify the trajectory of 90% (19/21) of sites after the 1998 bleaching event. Our trained model was then used to predict recovery versus regime shift trajectory following the 2016 bleaching event using 2014 data.

RESULTS

Causes of regime shifts

Our causal models show that depth and structural complexity decreased the likelihood of a climate-induced regime shift following the 1998 bleaching event in Seychelles (Figure 3). Similar to Graham et al., 2015, we find that deeper and structurally complex reefs are more resilient against climate-induced bleaching events. Our nutrient causal model also shows that high nutrients (low carbon : nitrogen ratios) increase the likelihood of regime shifts (Figure 3). In addition to these insights, our causal models revealed several factors that influenced regime shift trajectory, which were not evident in our past correlative study (Graham et al., 2015). Importantly, higher initial macroalgal cover increased the likelihood of regime shifting and had the strongest effect size (and largest variation) of all predictor variables (Figure 3). High wave exposure was also shown to increase the likelihood of regime shifting (Figure 3). To a lesser extent, both higher herbivorous fish biomass and branching coral cover increased the likelihood of regime shifting, while MPA status was found to have no effect (Figure 3).

Predictions for future regime shifts

Our predictive model suggests that out of the 12 sites that recovered from the 1998 bleaching event in Seychelles, five reefs have a greater than 50% probability of regime shifting following the second bleaching event in 2016 (Figure 4a). These reefs include St Anne P, St. Anne G, Praslin NE P, Mahe NW C, and Mahe E P (Figure 4a). Of these, three reefs are particularly vulnerable, with over 60% probability of regime shifting: St Anne P (67%), St. Anne G (84%), and Mahe E P (81%; Figure 4a).

DISCUSSION

Causal versus correlative analysis

Our causal models indicate multiple additional factors that influenced regime shift versus recovery trajectory, including initial macroalgal cover, wave exposure, and branching coral cover, which were not evident from our past correlative analysis on the same study system (Graham et al., 2015), with most of the same variables. The main difference between causal versus correlative analyses lies in covariate selection. Our previous analyses placed all assumed predictor variables into one model (i.e., a “causal salad” model), and subsequently removed all variables that did not show an effect to arrive at a final model that included: herbivore biomass, depth, structural complexity, and nutrients, as well as juvenile coral density, which was not considered in the current analysis (see next paragraph; Graham et al., 2015). In comparison, here we used our ecological knowledge to create our DAG and subsequently applied the backdoor criterion to build a causal model for each of our predictor variables. The backdoor criterion guides covariate selection required to determine causal relationships from observational data, theoretically equivalent to a randomized controlled experiment (Pearl, 2009; Appendix S1: Section S1.3). Given a DAG, the application of the backdoor criterion removes spurious correlations that may otherwise plague observational studies.

DAGs also allow us to carefully think about which observational data may best suit our causal questions, reflecting the causal structure of our DAG (Figure 2). In our previous study, we used postdisturbance 2005 data for herbivorous fish biomass, to account for the short-term disturbance effects on herbivore biomass (Graham et al., 2015). However, this effect could be bidirectional, since reef sites that are in the process of regime shifting can lead to increased postdisturbance herbivorous fish biomass. Such bidirectionality is prohibited in a DAG-based analysis, and the presence of bidirectional relationships in a regression analysis can lead to erroneous results due to the presence of simultaneity bias (Merton, 1968). To remedy this, we used predisturbance 1994 data for herbivore biomass, allowing for a directed arrow pointing from herbivorous fish biomass to regime shift trajectory. We previously used postdisturbance juvenile coral density (2011 data) as a predictor variable (Graham et al., 2015). While postdisturbance juvenile coral density data may influence regime shift trajectory, it may also be influenced by the regime shift versus recovery process itself. Indeed, higher postdisturbance juvenile coral density may be more a process of recovery (Gilmour et al., 2013; Hughes et al., 2010) than a factor that influences regime shift versus recovery trajectories. Given this, we excluded postjuvenile coral density as a predictor variable in our DAG and analysis. In addition, we included predisturbance macroalgal cover as a predictor variable in our DAG, which was not previously considered, but ended up having the strongest effect size (Figure 3).

In our review of all coral reef regime shifts, we found that observational studies did not apply causal inference (Appendix S1: Section S3). Whereas some studies did not include any covariate adjustments, and others employed a causal salad approach, across all studies, causal analysis and the consideration of the overall causal structure between variables of interest were missing, including in our own work (Appendix S1: Section S3). Here, the application of DAGs and the backdoor criterion provides a formal causal framework that can guide future model selection on coral reef regime shift studies. To further demonstrate how the backdoor criterion can lead to improved causal estimates, we compare our results guided by the backdoor criterion with other statistical models, including those that do not include any covariate adjustments and a causal salad model (Appendix S1: Section S2). Our estimates vary significantly across different models, highlighting the potential for causal models across observational studies (Appendix S1: Section S2).

Additional insights

Our causal models reveal several key additional factors that affected regime shift versus recovery trajectory that were not evident in our previous study (Figure 3). In particular, we find that higher initial (predisturbance) macroalgal cover, a variable not included in Graham et al. (2015) increases the likelihood of regime shifting (Figure 3). Following a disturbance event, additional space made available through coral mortality can lead to macroalgal expansion and subsequent inhibition of coral recruitment (McCook et al., 2001). In Seychelles, the rate of coral recovery has been strongly negatively associated with the rate of macroalgal cover increase (Wilson et al., 2012). Initial macroalgal cover may lead to postdisturbance coral–algal shifts in two ways. First, reef macroalgae may have higher thermal tolerance than coral species, with some macroalgal species experiencing no mortality under elevated temperatures (Anderson, 2006). Therefore, established macroalgae may remain intact following a bleaching event, creating a competitive starting point for macroalgal expansion. Second, various macroalgal species exhibit limited dispersal ability, with propagule settlement and recruitment remaining close to the source population (Capdevila et al., 2018). As such, having higher levels of macroalgae already established within a site may create a strong basis for macroalgal recruitment following a disturbance event. To our knowledge, the impact of initial macroalgal cover prior to disturbance on reef recovery dynamic has yet to be investigated elsewhere. Future research should examine whether this pattern is generalizable across other reef ecosystems, as well as the mechanisms that underlie this process.

Our results also indicate that wave exposure increased the likelihood of regime shifting (Figure 2). Wave-exposed reefs with higher water flow can favor macroalgal growth through increased exposure to and uptake of inorganic carbon and nutrients, which can lead to higher photosynthesis and growth rates of macroalgae (Hurd, 2000). Wave action can also limit coral growth and larval settlement (Gove et al., 2015), as well as remove coral through colony dislodgement and abrasive damage (Madin & Connolly, 2006). On the contrary, higher flow rates associated with greater wave exposure have also been attributed to reduced bleaching susceptibility and faster recovery of corals through the passive diffusion of harmful oxygen radicals that can accumulate in corals under high sea surface temperature (SST) and irradiance (Nakamura et al., 2003; Nakamura & van Woesik, 2001). However, McClanahan et al. (2005) show that increased flow rate is correlated with increased bleaching intensity in Mauritius, reasoning that higher flow rates occurred in regions with lower variation in water temperature, which in turn can increase bleaching susceptibility (Safaie et al., 2018). Collectively, these factors may promote postdisturbance coral–algal regime shifts at wave-exposed reef sites. For example, wave exposure was found to be the main determinant of a coral–algal regime shift following a catastrophic typhoon disturbance in Micronesia (Ross et al., 2015). In Seychelles, the recovery rate for recovering reef sites has also been negatively associated with increased wave exposure (Robinson, Wilson, & Graham, 2019).

We found a positive association between herbivorous reef fish biomass and regime shift trajectory (Figure 3). Following a climatic disturbance, herbivorous reef fish is expected to limit coral–algal shifts through grazing pressure, which limits the growth of macroalgae and enhances coral recruitment through creating space for larval settlement (Hughes et al., 2007; McCook et al., 2001; Mumby & Harborne, 2010). Yet in Seychelles, the predisturbance biomass of herbivorous fish seems to be positively correlated with regime shift occurrence. Previous studies in Seychelles have shown that higher herbivore biomass postdisturbance was correlated with a reduced likelihood of regime shifting (Graham et al., 2015), but a slower recovery rate on recovering reefs (Robinson, Wilson, & Graham, 2019). Coarse biomass metrics of herbivorous fish biomass, which combine distinct functional groups, may not be representative of true grazing effects on coral reefs, which are tightly linked to size structure and functional composition of herbivore assemblage (Nash et al., 2015; Robinson, Wilson, & Graham, 2019; Steneck et al., 2018). Future studies in Seychelles can address this gap by looking at more accurate grazing metrics, which have been shown to clarify the effect of herbivorous reef fish across other coral reef systems (Steneck et al., 2018). In general, and particularly along reefs where fishing has drastically reduced herbivorous fish biomass, it is expected that lower levels of herbivory will limit recovery and lead to more coral–algal shifts following climatic disturbance (Hughes et al., 2007; Mumby & Harborne, 2010).

Higher initial branching coral cover was also shown to increase the likelihood of regime shifting (Figure 3). Branching corals are often vulnerable to bleaching given their lower heat tolerance (Loya et al., 2001). Following a climate-induced bleaching event, this can lead to higher loss of branching coral; for example, surveys following the 2016 bleaching event in Seychelles found that 95% of Acropora and Pocillopora colonies were either bleached or recently dead (Wilson et al., 2019). Ultimately, reefs with high predisturbance branching coral cover may result in a large coral mortality and the subsequent availability of open space, creating favorable conditions for macroalgal growth and dominance, which may overwhelm the effect of grazing pressure from herbivorous reef fish (Williams et al., 2001). Moreover, once dead, the structures provided by branching corals erode rapidly, impacting reef fish and other organisms (Sheppard et al., 2002). On the contrary, branching corals are often fast-growing (Darling et al., 2012) and can play a critical role in coral recovery following disturbance, as demonstrated by the recovering reefs in Seychelles (Robinson, Wilson, & Graham, 2019; Wilson et al., 2019) and the Great Barrier Reef (Linares et al., 2011). How this trade-off between heat tolerance and growth rates relates to longer term patterns and predictions of coral recovery may depend on the frequency and intensity of disturbances. Given that severe bleaching events are now expected every 6 years (Hughes et al., 2018), branching coral cover may ultimately be at a disadvantage over heat-tolerant corals (Kubicek et al., 2012; Kubicek et al., 2019), with reefs with higher branching coral cover potentially being more vulnerable to coral–algal regime shifts.

Predicting recovery versus regime shift trajectory post-2016 bleaching event

Our predictive model suggests that out of the 12 sites that recovered from the 1998 bleaching event in Seychelles, five reefs show a greater than 50% probability of regime shifting following the second bleaching event in 2016, with three reefs having a probability above 60% (Figure 4a). These reef sites are vulnerable to coral–algal regime shifts due to a combination of factors. For example, St Anne P, which shows a 67% chance of regime shifting, had (according to 2014 data) very high branching coral, low structural complexity, and high wave exposure (Figure 4c,e,f). In comparison, St Anne G, showing a 84% change of regime shifting, had high wave exposure and very high macroalgal cover (Figure 4f,g). Last, Mahe E P, which had a 81% chance of regime shifting, had high nutrients (low carbon : nitrogen), low structural complexity, and high wave exposure (Figure 4b,e,f). We note that our predictions may underestimate reef vulnerability to the 2016 bleaching event as our predictive model was “trained” using results from the first bleaching event in 1998. Indeed, Seychelles reefs may now be under unstable equilibria (May, 1977; Scheffer et al., 2001), essentially requiring less cumulative stress to drive coral–algal shifts (Mumby & Hastings, 2008).

Our predictions suggest that several factors can come together to influence regime shift versus recovery trajectories on coral reefs impacted by subsequent bleaching events. Given that severe bleaching events are now expected every 6 years (Hughes et al., 2018), conservation and management efforts may benefit from prioritizing locations where the rate of warming and threat of frequent bleaching is lowest (van Hooidonk et al., 2016) and where recovery from climatic disturbances is most likely (Côté & Darling, 2010; Graham et al., 2020). Specifically, reefs with increased depth, resilient coral species, and structural complexity, and low macroalgae, nutrients, and wave exposure may be more resilient against future climatic disturbances in Seychelles. Deeper, structurally complex granitic reefs with higher cover of heat-tolerant massive corals and low macroalgae may be important areas for future conservation efforts in this region (Dajka et al., 2019; Graham et al., 2006). Collectively, incorporating these findings into management efforts may aid in prioritizing potentially resilient coral reefs amidst our current environmental and climate change crisis.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Open Research

DATA AVAILABILITY STATEMENT

Data and underlying code (Arif, 2021) are available from Figshare: https://doi.org/10.6084/m9.figshare.14981235

Supporting Information

REFERENCES

Anderson, B. 2006. “Response of Tropical Marine Macroalgae to Thermal Stress.” MSc thesis, Florida Atlantic University.
Google Scholar
Arif, S. 2021. Causal Drivers of Climate-Mediated Coral Reef Regime Shifts.” Figshare. Data and Code. https://doi.org/10.6084/m9.figshare.14981235.v1.
Google Scholar
Bellwood, D., T. Hughes, C. Folke, and M. Nystrom. 2004. “Confronting the Coral Reef Crisis.” Nature 429: 827–33.
10.1038/nature02691
CASPubMedWeb of Science®Google Scholar
Capdevila, P., C. Linares, E. Aspillaga, J. Riera, and B. Hereu. 2018. “Effective Dispersal and Density-Dependence in Mesophotic Macroalgal Forests: Insights from the Mediterranean Species Cystoseira zosteroides.” PLoS One 13(1): e0191346.
10.1371/journal.pone.0191346
PubMedWeb of Science®Google Scholar
Côté, I., and E. Darling. 2010. “Rethinking Ecosystem Resilience in the Face of Climate Change.” PLoS Biology 8(7): e1000438.
10.1371/journal.pbio.1000438
PubMedWeb of Science®Google Scholar
Cronin, J., and D. Schoolmaster, Jr. 2018. “A Causal Partition of Trait Correlations: Using Graphical Models to Derive Statistical Models from Theoretical Language.” Ecosphere 9(9): e02422.
10.1002/ecs2.2422
Web of Science®Google Scholar
Dajka, J., S. Wilson, J. Robinson, K. Chong-Seng, A. Harris, and N. Graham. 2019. “Uncovering Drivers of Juvenile Coral Density Following Mass Bleaching.” Coral Reefs 38: 637–49.
10.1007/s00338-019-01785-w
Web of Science®Google Scholar
Darling, E., L. Alvarez-Filip, T. Oliver, T. McClanahan, and I. Côté. 2012. “Evaluating Life-History Strategies of Reef Corals from Species Traits.” Ecology Letters 15: 1378–86.
10.1111/j.1461-0248.2012.01861.x
PubMedWeb of Science®Google Scholar
Elwert, F. 2013. “ Graphical Causal Models.” In Handbook of Causal Analysis for Social Research, edited by S.L. Morgan, 245–73. Dordrecht: Springer.
10.1007/978-94-007-6094-3_13
Google Scholar
Gelman, A., and J. Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge: Cambridge University Press.
Google Scholar
Gilmour, J., L. D. Smith, A. Heyward, A. Baird, and M. Pratchett. 2013. “Recovery of an Isolated Coral Reef System Following Severe Disturbance.” Science 340: 69–71.
10.1126/science.1232310
CASPubMedWeb of Science®Google Scholar
Glymour, M., and S. Greenland. 2008. “ Causal Diagrams.” In Modern Epidemiology. Philadelphia: Lippincott Williams & Wilkins.
Google Scholar
Goreau, T., T. McClanahan, R. Hayes, and A. Strong. 2000. “Conservation of Coral Reefs after the 1998 Bleaching Event.” Conservation Biology 14: 5–15.
10.1046/j.1523-1739.2000.00011.x
Web of Science®Google Scholar
Gove, J., G. Williams, M. McManus, S. Clark, J. Ehses, and L. Wedding. 2015. “Coral Reef Benthic Regimes Exhibit Non-linear Threshold Responses to Natural Physical Drivers.” Marine Ecology Progress Series 522: 33–48.
10.3354/meps11118
Web of Science®Google Scholar
Graham, N., S. Jennings, A. MacNeil, D. Mouillot, and S. Wilson. 2015. “Predicting Climate-Driven Regime Shifts Versus Rebound Potential in Coral Reefs.” Nature 518: 94–7.
10.1038/nature14140
CASPubMedWeb of Science®Google Scholar
Graham, N., J. Robinson, S. Smith, R. Govinden, G. Gendron, and S. Wilson. 2020. “Changing Role of Coral Reef Marine Reserves in a Warming Climate.” Nature Communications 11: 2000.
10.1038/s41467-020-15863-z
CASPubMedWeb of Science®Google Scholar
Graham, N., S. Wilson, S. Jennings, N. Polunin, J. Bijoux, and J. Robinson. 2006. “Dynamic Fragility of Oceanic Coral Reef Ecosystems.” Proceedings of the National Academy of Sciences 103: 8425–9.
10.1073/pnas.0600693103
CASPubMedWeb of Science®Google Scholar
Greenland, S., J. Pearl, and J. Robins. 1999. “Causal Diagrams for Epidemiologic Research.” Epidemiology 10: 37–48.
10.1097/00001648-199901000-00008
CASPubMedWeb of Science®Google Scholar
Hempson, T., N. Graham, A. MacNeil, A. Hoey, and S. Wilson. 2017. “Ecosystem Regime Shifts Disrupt Trophic Structure.” Ecological Applications 28(1): 191–200.
10.1002/eap.1639
PubMedWeb of Science®Google Scholar
Hughes, T., Kristen D. Anderson, Sean R. Connolly, Scott F. Heron, James T. Kerry, Janice M. Lough, Andrew H. Baird, et al. 2018. “Spatial and Temporal Patterns of Mass Bleaching of Corals in the Anthropocene.” Science 359(6371): 80–3.
10.1126/science.aan8048
CASPubMedWeb of Science®Google Scholar
Hughes, T., N. Graham, J. Jackson, P. Mumby, and R. Steneck. 2010. “Rising to the Challenge of Sustaining Coral Reef Resilience.” Trends in Ecology & Evolution 25: 633–42.
10.1016/j.tree.2010.07.011
PubMedWeb of Science®Google Scholar
Hughes, T. P., Maria J. Rodrigues, David R. Bellwood, Daniela Ceccarelli, Ove Hoegh-Guldberg, Laurence McCook, Natalie Moltschaniwskyj, Morgan S. Pratchett, Robert S. Steneck, and Bette Willis. 2007. “Phase Shifts, Herbivory, and Resilience of Coral Reefs to Climate Change.” Current Biology 17(4): 360–5.
10.1016/j.cub.2006.12.049
CASPubMedWeb of Science®Google Scholar
Hurd, C. 2000. “Water Motion, Marine Macroalgal Physiology, and Production.” Journal of Phycology 36: 453–72.
10.1046/j.1529-8817.2000.99139.x
CASPubMedWeb of Science®Google Scholar
Kubicek, A., B. Breckling, O. Hoegh-Guldberg, and H. Reuter. 2019. “Climate Change Drives Trait-Shifts in Coral Reef Communities.” Scientific Reports 9: 3721.
10.1038/s41598-019-38962-4
PubMedWeb of Science®Google Scholar
Kubicek, A., C. Muhando, and H. Reuter. 2012. “Simulations of Long-Term Community Dynamics in Coral Reefs—How Perturbations Shape Trajectories.” PLoS Computational Biology 8(11): e1002791.
10.1371/journal.pcbi.1002791
CASPubMedWeb of Science®Google Scholar
Linares, C., M. Pratchett, and D. Coker. 2011. “Recolonisation of Acropora hyacinthus Following Climate-Induced Coral Bleaching on the Great Barrier Reef.” Marine Ecology Progress Series 438: 97–104.
10.3354/meps09272
Web of Science®Google Scholar
Loya, Y., K. Sakai, K. Yamazato, Y. Nakano, H. Sambali, and R. van Woesik. 2001. “Coral Bleaching: The Winners and Losers.” Ecology Letters 4: 122–31.
10.1046/j.1461-0248.2001.00203.x
Web of Science®Google Scholar
Madin, J., and S. Connolly. 2006. “Ecological Consequences of Major Hydrodynamic Disturbances on Coral Reefs.” Nature 444: 477–80.
10.1038/nature05328
CASPubMedWeb of Science®Google Scholar
May, R. 1977. “Thresholds and Breakpoints in Ecosystems with a Multiplicity of Stable States.” Nature 269: 471–7.
10.1038/269471a0
Web of Science®Google Scholar
McClanahan, T., J. Maina, R. Moothien-Pillay, and A. Baker. 2005. “Effects of Geography, Taxa, Water Flow, and Temperature Variation on Coral Bleaching Intensity in Mauritius.” Marine Ecology Progress Series 298: 131–42.
10.3354/meps298131
Web of Science®Google Scholar
McCook, L., J. Jompa, and G. Diaz-Pulido. 2001. “Competition between Corals and Algae on Coral Reefs: A Review of Evidence and Mechanisms.” Coral Reefs 19: 400–17.
10.1007/s003380000129
Web of Science®Google Scholar
McElreath, R. 2020. Statistical Rethinking: A Bayesian Course with Examples in R and Stan. Boca Raton, FL: CRC Press.
10.1201/9780429029608
Google Scholar
Merton, R. 1968. “The Matthew Effect in Science.” Science 159(3810): 56–63.
10.1126/science.159.3810.56
CASPubMedWeb of Science®Google Scholar
Mumby, P., and A. Harborne. 2010. “Marine Reserves Enhance the Recovery of Corals on Caribbean Reefs.” PLoS One 5: e8657.
10.1371/journal.pone.0008657
CASPubMedWeb of Science®Google Scholar
Mumby, P., and A. Hastings. 2008. “The Impact of Ecosystem Connectivity of Coral Reef Resilience.” Journal of Applied Ecology 45: 854–62.
10.1111/j.1365-2664.2008.01459.x
Web of Science®Google Scholar
Nakamura, T., and R. van Woesik. 2001. “Water-Flow Rates and Passive Diffusion Partially Explain Differential Survival of Corals during the 1998 Bleaching Event.” Marine Ecology Progress Series 212: 301–4.
10.3354/meps212301
Web of Science®Google Scholar
Nakamura, T., H. Yamasaki, and R. van Woesik. 2003. “Water Flow Facilitates Recovery from Bleaching in the Coral Stylophora pistillata.” Marine Ecology Progress Series 256: 287–91.
10.3354/meps256287
Web of Science®Google Scholar
Nash, K., N. Graham, S. Jennings, S. Wilson, and D. Bellwood. 2015. “Herbivore Cross-Scale Redundancy Supports Response Diversity and Promotes Coral Reef Resilience.” Journal of Applied Ecology 53: 646–55.
10.1111/1365-2664.12430
Web of Science®Google Scholar
Pearl, J. 2009. Causality: Models, Reasoning and Inference, 2nd ed. Cambridge: Cambridge University Press.
10.1017/CBO9780511803161
Google Scholar
Robinson, J., S. Wilson, and N. Graham. 2019. “Abiotic and Biotic Controls on Coral Recovery 16 Years after Mass Bleaching.” Coral Reefs 38(6): 1255–65.
10.1007/s00338-019-01831-7
Web of Science®Google Scholar
Robinson, J., S. Wilson, S. Jennings, and N. Graham. 2019. “Thermal Stress Induces Persistently Altered Coral Reef Fish Assemblages.” Global Change Biology 25(8): 2739–50.
10.1111/gcb.14704
PubMedWeb of Science®Google Scholar
Robinson, J., S. Wilson, J. Robinson, C. Gerry, J. Lucas, and C. Assan. 2019. “Productive Instability of Coral Reef Fisheries after Climate-Driven Regime Shifts.” Nature Ecology and Evolution 3(2): 183–90.
10.1038/s41559-018-0715-z
PubMedWeb of Science®Google Scholar
Ross, G., C. Doropoulos, and M. Zupan. 2015. “Phase Shift Facilitation Following Cyclone Disturbance on Coral Reefs.” Oecologia 178: 1193–203. https://doi.org/10.1007/s00442-015-3282-x
10.1007/s00442-015-3282-x
PubMedWeb of Science®Google Scholar
Safaie, A., N. Silbiger, T. McClanahan, G. Pawlak, D. Barshis, and J. Hench. 2018. “High Frequency Temperature Variability Reduces the Risk of Coral Bleaching.” Nature Communications 9: 1671.
10.1038/s41467-018-04074-2
PubMedWeb of Science®Google Scholar
Scheffer, M., S. Carpenter, J. Foley, J. Folke, and B. Walker. 2001. “Catastrophic Shifts in Ecosystems.” Nature 413: 591–6.
10.1002/(SICI)1099-1646(199901/06)15:1/3<43::AID-RRR535>3.0.CO;2-Q
CASPubMedWeb of Science®Google Scholar
Schoolmaster, D., Jr., C. Zirbel, and J. Cronin. 2020. “A Graphical Causal Model for Resolving Species Identity Effects and Biodiversity–Ecosystem Function Correlations.” Ecology 101(8): e03070.
10.1002/ecy.3070
PubMedWeb of Science®Google Scholar
Sheppard, C., M. Spalding, C. Bradshaw, and S. Wilson. 2002. “Erosion vs Recovery of Coral Reefs after 1998 El Nino: Chagos Reefs, Indian Ocean.” Ambio 31: 40–8.
10.1579/0044-7447-31.1.40
PubMedWeb of Science®Google Scholar
Spirtes, P., C. Glymour, and R. Schein. 2001. Causation, Prediction, and Search, 2nd ed. Cambridge, MA: MIT Press.
10.7551/mitpress/1754.001.0001
Google Scholar
Steneck, R. S., P. J. Mumby, C. MacDonald, D. B. Rasher, and G. Stoyle. 2018. “Attenuating Effects of Ecosystem Management on Coral Reefs.” Science Advances 4: eaao5493.
10.1126/sciadv.aao5493
PubMedWeb of Science®Google Scholar
Textor, J., B. Zander, M. Gilthorpe, M. Liskiewicz, and G. Ellison. 2016. “Robust Causal Inference Using Directed Acyclic Graphs: The R Package ‘dagitty’.” International Journal of Epidemiology 45(6): 1887–94.
PubMedWeb of Science®Google Scholar
van Hooidonk, R., Jeffrey Maynard, Jerker Tamelander, Jamison Gove, Gabby Ahmadia, Laurie Raymundo, Gareth Williams, Scott F. Heron, and Serge Planes. 2016. “Local-Scale Projections of Coral Reef Futures and Implications of the Paris Agreement.” Scientific Reports 6: 39666.
10.1038/srep39666
PubMedWeb of Science®Google Scholar
Williams, I., N. Polunin, and V. Hendrick. 2001. “Limits to Grazing by Herbivorous Fishes and the Impact of Low Coral Cover on Macroalgal Abundance on a Coral Reef in Belize.” Marine Ecology Progress Series 222: 187–96.
10.3354/meps222187
Web of Science®Google Scholar
Wilson, S., N. Graham, R. Fisher, J. Robinson, K. Nash, K. Chong-Seng, N. Polunin, R. Aumeeruddy, and R. Quatre. 2012. “Effect of Macroalgal Expansion and Marine Protected Areas on Coral Recovery Following a Climatic Disturbance.” Conservation Biology 26(6): 995–1004.
10.1111/j.1523-1739.2012.01926.x
PubMedWeb of Science®Google Scholar
Wilson, S., J. Robinson, K. Chong-Seng, J. Robinson, and N. Graham. 2019. “Boom and Bust of Keystone Structure on Coral Reefs.” Coral Reefs 38: 625–35.
10.1007/s00338-019-01818-4
Web of Science®Google Scholar

Citing Literature

Volume13, Issue3

March 2022

e3956

Journal list menu

Causal drivers of climate-mediated coral reef regime shifts

Abstract

INTRODUCTION