See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/322650698

Modeling alternative stable states in Caribbean

coral reefs

Article in Natural Resource Modelling · January 2018

DOI: 10.1111/nrm.12157

CITATIONS READS

0 72

6 authors, including:

Elizabeth Sherman
Bennington College
14 PUBLICATIONS 129 CITATIONS

SEE PROFILE

All content following this page was uploaded by Elizabeth Sherman on 28 January 2018.

The user has requested enhancement of the downloaded file.

Received: 13 July 2017 Accepted: 27 August 2017

DOI: 10.1111/nrm.12157

Modeling alternative stable states in Caribbean

coral reefs
Julie C. Blackwood1 Colin Okasaki2 Andre Archer3 Eliza W. Matt1
Elizabeth Sherman4 Kathryn Montovan4

1 Department of Mathematics and Statistics,

Williams College, Williamstown, MA Abstract

2 Department of Mathematics, Harvey Mudd
The resilience of Caribbean coral reefs, which are an impor-
College, Claremont, CA
tant source of biodiversity and provide essential ecosystem
3 Mathematics, Statistics, and Computer Sci-

ence, Macalester College, Saint Paul, MN

services, is constantly challenged by many reef stressors
4 Department of Science and Mathematics, including ocean acidification, hurricane damage, and over-
Bennington College, Bennington, VT harvesting of herbivorous reef fish. The presence of two
Correspondence alternative stable states—a desirable state with high lev-
Julie C. Blackwood, Department
els of coral cover and its coral-depleted counterpart—has
of Mathematics and Statistics,
Williams College, Williamstown, MA. been widely documented in the literature. Increasing coral
Email: jcb5@williams.edu resilience to prevent phase shifts to the undesirable state is
a critical research priority, and mathematical models can
serve as an important tool to not only better understand
the underlying dynamics of observed coral communities,
but also to evaluate the potential impacts of stressors and
the outcome of management strategies designed to promote
coral persistence. Here, we review the existing literature of
mathematical models designed to understand the processes
that generate alternative stable states. We focus on models
that are comprised of ordinary differential equations and, at
their core, capture algal–coral dynamics.

Recommendations for Resource Managers

• Evidence for the existence of alternative stable states and
the associated presence of hysteresis implies a need for
management designed to increase the resilience of coral
reef ecosystems.
• In addition, holistic approaches to designing manage-
ment strategies are required to both increase resilience of

Natural Resource Modeling. 2018;e12157. wileyonlinelibrary.com/journal/nrm Copyright © 2018 Wiley Periodicals, Inc. 1 of 17
https://doi.org/10.1111/nrm.12157
2 of 17 Natural Resource Modeling B LACKWOOD ET AL.

coral reefs and maximize the benefits of the ecosystem

services they provide.
• Due to the intrinsic complexity and spatial variability of
coral reef ecosystems, management cannot be designed
using a “one size fits all” approach. Instead, local dynam-
ics and stressors need to be carefully considered.

KEYWORDS
community composition, hysteresis, phase shift, reef stressors

1 I N T RO D U C T I O N

Caribbean coral reefs have recently suffered significant losses in coral cover as a consequence of the
catastrophic die-off of the sea urchin Diadema antillarum in the early 1980s. The conventional wisdom
has been that overfishing of herbivorous fish led to the pre-eminence of Diadema as the last remaining
significant herbivore keeping algae at bay in the Caribbean prior to that die-off. Thus, overfishing
led to the precarious dependence of herbivory on just the one species. The effectiveness of Diadema
during the recent past in controlling macroalgae abundance has much support (Levitan, Edmunds, &
Levitan, 2014; Miller, Kramer, Williams, Johnston, & Szmant, 2009). In fact, some reports suggest that
Diadema may have been among the dominant herbivores for millennia (Lessios, Garrido, & Kessing,
2001; Levitan et al., 2014). However, the importance of these urchins as significant herbivores prior
to human disturbance has been called into question by Cramer, Dea, Carpenter, and Norris (2017),
who examined the subfossil spine assemblages of echinoids for the past 3000 years. They do confirm
that Diadema is causally related to the abundance of at least one reef-building coral species (Madracis
mirabilis), but not others (e.g., Acropora cervicornis). Moreover, recent studies suggest that bioeroding
sponges may play a significant role in coral cover (Murphy, Perry, Chin, & McCoy, 2016). Thus, the
relationship of coral abundance to herbivory, bioerosion, and macroalgae is likely to be complicated.
In addition to biotic factors, numerous abiotic factors, including temperature, ocean acidification, and
mechanical damage from hurricanes have been identified as coral reef stressors. These factors not only
affect the coral but also the animals implicated in their resilience, or their ability to recover following
a perturbation or disturbance (Holling, 1973). For example, in addition to the increased incidence
of coral bleaching due to elevated temperatures (Donner, 2011), Diadema is likely to be negatively
affected by temperatures predicted to occur due to global warming (Sherman, 2015). Nevertheless,
in order to understand the mechanisms that are driving the observed macroalgae and coral dynamics
and to uncover methods for promoting restoration and recovery, holistic approaches integrating field
research, experimental work, and mathematical modeling are needed.
Many have hypothesized that there exist two stable states for coral ecosystems: the first is a desirable
state that corresponds to high coral cover while the second is a degraded coral-depleted state that
usually has high levels of macroalgae. It is clear that anthropogenic stresses can drive phase shifts
from a state with high coral cover to one that is coral depleted, but a significant distinction is made
based on whether the phase shift is continuous (a smooth transition between states) or discontinuous (a
sudden jump from one state to another) as an exogenous factor is varied. In systems with a continuous
phase shift, only a single stable state exists across all environmental conditions. In contrast, alternative
stable states exist when there is a discontinuous phase shift such that multiple stable states can coexist
B LACKWOOD ET AL. Natural Resource Modeling 3 of 17

under the same environmental conditions. In this case, an undesirable shift between stable states due
to changing environmental conditions cannot usually be undone easily; to return to the original state
would require environmental parameters to shift beyond where the discontinuous change occurred
(i.e., hysteresis). The existence of alternative stable states for coral communities has been extensively
debated in the literature, with many arguing that the observed drastic shifts in community structure are
due to changing environmental conditions that represent a continuous phase shift (Dudgeon, Aronson,
Bruno, & Precht, 2010) and attempts to experimentally shift a community to an alternative stable state
have not been successful without changing the environmental conditions (Connell & Sousa, 1983).
In contrast, others argue that while we may not be able to see alternative stable states under identical
environmental conditions, this does not mean that they are not present in this system.
Although progress is being made, the debate between whether there are two alternative stable states
or whether the changes in community structure are the result of a continuous phase shift remains an
open question. Mathematical modeling is an important tool for investigating the existence of alternative
stable states because of the opportunity to construct a system in which the environmental conditions
are kept unchanged while the community composition is shifted. For example, Mumby, Hastings, and
Edwards (2007) developed a simple model of the interactions between coral, macroalgae, and algal
turfs. Using this model, they demonstrated that alternative stable states exist and their stability depends
on the rate at which herbivores graze macroalgae. When grazing is low, macroalage is able to spread
over turf and successfully outcompete coral leading to a stable coral-depleted state. In contrast, when
grazing of algae is sufficiently high, the growth of macroalgae is sufficiently inhibited so that coral
can recruit to and overgrow turf and stably persist at high levels. Moreover, the analysis of their model
suggests that the phase shift between these stable states is discontinuous. Mumby et al.’s model provides
a parsimonious explanation for the existence of alternative stable states, with feedbacks between coral
and algae at its core. Hereafter, our emphasis in this paper is on related modeling studies that capture
the presence of alternative stable states.
Due to the relative simplicity of the Mumby et al. model, its baseline assumptions provide a founda-
tion to build upon in an attempt to better understand the complex dynamics of coral reef ecosystems.
Consequently, a wealth of modeling studies have been performed over the past decade that build upon
similar baseline assumptions. In this paper, we introduce the Mumby et al. model in detail and its key
properties. We then review modeling efforts since then that are either directly based on the Mumby
et al. model or have similar model structures. While many other studies exist that have considered, for
example, delay differential equations (e.g., Blackwood & Hastings, 2011; Li, Wang, Zhang, & Hast-
ings, 2014) or large-scale spatial simulation models (e.g., Mumby, 2006), we focus exclusively on
models based on systems of ordinary differential equations (ODEs) that have coral–algal interactions
at their core. Further, the majority of models we review consider the dynamical impacts of stressors to
Caribbean reefs.

2 BA SELINE MODEL OF CORAL–ALGA L I NTERACTIONS

Here, we describe the influential model of coral–algal interactions first introduced by Mumby et al.
(2007). This model arguably provides the most parsimonious representation of interactions between
coral and algae that leads to alternative stable states. The model is spatially implicit and is comprised
of three ODEs that capture the dynamics of coral (𝐶), macroalgae (𝑀), and algal turfs (𝑇 ). Each of
the three state variables is described in terms of the fraction of seabed that they occupy. An important
biological assumption made is that the seabed is covered entirely by coral, macroalgae, and algal turfs
so that 𝐶 + 𝑀 + 𝑇 = 1. Conveniently, this allows the system to be completely determined only by two
4 of 17 Natural Resource Modeling B LACKWOOD ET AL.

C
T1

Coral cover equilibrium

CH

aC
d
rT

g/(M + T ) T2
T M 0
g1 g2
γT
Grazing rate, g

FIGURE 1 (A) Schematic representation of the interactions captured in the Mumby et al. (2007) framework; (B)
bifurcation diagram of the equilibrium coral cover versus the grazing rate (𝑔)
Note. Solid gray lines denote stable equilibria and dashed gray lines indicate unstable equilibria as well as saddles. This
bifurcation diagram captures the presence of alternative stable states in the model with 𝐶𝐻 denoting the desirable stable
state with high levels of coral cover. 𝑔1 and 𝑔2 , which are found explicitly in the Appendix, represent the critical grazing
rates at which the transcritical bifurcations 𝑇1 and 𝑇2 occur. These bifurcations are the points at which the system can
switch to the alternative stable state.

of the state variables. In other words, the fraction of algal turf is defined by 𝑇 = 1 − 𝐶 − 𝑀 and can
therefore be identified using the solutions for 𝐶 and 𝑀.
The model captures five basic interactions: (i) macroalgae overgrows coral at a rate 𝑎, (ii) macroalgae
spreads vegetatively over algal turfs at a rate 𝛾, (iii) coral recruits to and overgrows algal turfs at a total
rate of 𝑟, (iv) coral has a natural mortality rate of 𝑑, which then gives rise to algal turfs, and (v) grazers
feed indiscriminately on both algal types at a fixed rate 𝑔 (see schematic representation in Fig. 1A).
For the latter interaction, it is assumed that grazing crops macroalgae and gives rise to turfs whereas
grazing on algal turfs essentially maintains the turf. Using the fact that 𝑇 is entirely determined by 𝑀
and 𝐶, the model is given by:

d𝐶
= r𝑇 𝐶 − d𝐶 − a𝑀𝐶, (1)
d𝑡

d𝑀 g𝑀
= a𝑀𝐶 − + 𝛾M𝑇 . (2)
d𝑡 𝑀 +𝑇
All parameters were estimated using empirical studies, with the exception of grazing, which may vary
based on, for example, local fishing practices.
An advantage of this relatively simple model formulation is that a linear stability analysis can be
explicitly carried out. While the details are left to the Appendix, it can be shown that there are four
equilibria (a fifth equilibrium exists, but it takes on a negative value for our parameterization and is
therefore ignored here). Recalling that 𝑇 ∗ = 1 − 𝐶 ∗ − 𝑀 ∗ , the equilibria, (𝐶 ∗ , 𝑀 ∗ ), are given by:

(I) (0, 0), or local extinction of both macroalgae and coral, which gives rise to a state entirely covered
by algal turfs;
(II) (0, 𝛾−𝑔
𝛾
), or an algal-dominated state where the equilibrium level of coverage of macroalgae
increases as the grazing rate decreases;
(III) ( 𝑟−𝑑
𝑟
, 0), or local extinction of macroalgae and persistence of coral; and
̂ 𝑀),
(IV) (𝐶, ̂ where 𝐶,
̂ 𝑀̂ ≠ 0 corresponds to coexistence of both coral and macroalgae.
B LACKWOOD ET AL. Natural Resource Modeling 5 of 17

Importantly, linear stability analysis reveals that when grazing rates are sufficiently low, then equi-
librium (II) is stable and (III) is unstable so that algae is able to outcompete coral (this is the degraded
coral-depleted state). However, when the grazing rate is sufficiently high, then macroalgae is success-
fully kept at bay. In this case, the stability of these equilibria is reversed so that the state with high coral
cover (equilibrium (III)) is stable.
As pointed out by Blackwood, Hastings, and Mumby (2012), this system exhibits hysteresis. That
is, the critical grazing rate (𝑔2 in Fig. 1B) at which the system switches stability from equilibrium (II)
to (III) is different from the critical rate (𝑔1 in Fig. 1B) at which stability switches from (III) to (II)
(these critical rates are found analytically in the Appendix). In other words, both equilibria coexist and
are stable for grazing rates between these critical values. In this region of parameter space, the basins
of attraction for the stable equilibria are divided by a separatrix that connects the origin (equilibrium
(I)) and a saddle (or semistable) point in which coral and macroalgae coexist (equilibrium (IV)). These
behaviors are captured in a bifurcation diagram of the equilibrium level of coral cover versus the grazing
rate (Fig. 1B).
In the sections that follow, we describe several modeling studies that have been performed since 2007
when the Mumby et al. model was introduced. In particular, we review models that focus on coral–algal
interactions at their core and exhibit alternative stable states (i.e., discontinuous phase shifts between
stable states). We note that in many cases, we adapt and/or simplify notation to be consistent with the
model described above and therefore to ease model interpretation.

3 ALTERNAT I V E M O D E L S OF CO RAL– ALGA L

I N T E R AC T I O N S

3.1 Turf, free space, and coral recruitment

Several studies use a similar framework to Mumby et al. (2007) but make alternative assumptions on
what covers the seabed. For example, some authors have generalized the assumption that 𝑇 represents
algal turf; instead, it is often defined as 𝐹 or 𝑆 to represent a more generic term that captures any “free
space” or substrata. One such example of this comes from another influential modeling framework that
was introduced by Fung et al. in 2011. In this section, we discuss the models from Fung, Seymour, and
Johnson (2011) to demonstrate a model formulation that is similar to Mumby et al. (2007) but makes
some different key assumptions.
Fung et al. introduced three models with increasing complexity and, at their core, each of these
models was also centered on describing coral–algal interactions. Their first model considered coral,
algal turfs, and free space (denoted by 𝐹 ). In this setting, algal turfs can grow on to substrata but are
not capable of competing with coral. In the second model, the algal class was generalized to include
macroalgae and could therefore compete with existing coral. In the final model, macroalgae and algal
turfs were considered separately. Here, we describe the final and most complex model in detail.
A key difference with Mumby et al. (2007) is that Fung et al. (2011) differentiates between algal
turfs and free space and therefore requires four state variables. In this model, it is again assumed that
the seabed is entirely covered by these four variables so that the fraction of free space can be found
implicitly by 𝐹 = 1 − 𝐶 − 𝑀 − 𝑇 . Further, this model makes more explicit assumptions about coral
growth and recruitment. It is assumed that spawning corals recruit onto space at a rate 𝑙𝑠 , brooding
corals recruit onto space at a rate 𝑙𝑏 , which is also proportional to the existing coral 𝐶, and both coral
types can recruit to algal turfs at a fraction 𝜖 of the rate at which they recruit to space. Next, they
assume that coral grows laterally over both free space and algal turfs at rates 𝑟 and 𝛼𝑟 (where 𝛼 ∈ [0, 1]),
6 of 17 Natural Resource Modeling B LACKWOOD ET AL.

respectively, but this growth is limited by the presence of macroalgae. As in Mumby et al. (2007), they
assume that coral has a natural mortality rate of 𝑑, macroalgae can overgrow coral at a rate 𝑎 and turf at
a rate 𝛾, and they additionally assumed that macroalgae and algal turfs can grow over free space at rate
𝛾𝐹 and 𝜁 , respectively. They also assumed that coral growth can be directly suppressed by macroalge
at a rate 𝛽. Finally, they assumed that grazing of macroalgae and algal turfs occurs at fixed rates 𝑔𝑀
and 𝑔𝑇 , respectively. These assumptions are captured in the following model:

d𝐶
= (𝑙𝑠 + 𝑙𝑏 𝐶)(𝐹 + 𝜖𝑇 ) + 𝑟(1 − 𝛽𝑀)(𝐹 + 𝛼𝑇 )𝐶 − 𝑑𝐶 − a𝑀𝐶, (3)
d𝑡

d𝑀
= 𝛾𝑀𝑇 + 𝛾𝐹 𝑀𝐹 + a𝑀𝐶 − 𝑔𝑀 𝑀, (4)
d𝑡

d𝑇
= −𝜖(𝑙𝑠 + 𝑙𝑏 )C𝑇 − 𝑟𝛼(1 − 𝛽𝑀)T𝐶 − 𝛾𝑀𝑇 − 𝑔T𝑇 + 𝜁 𝐹 . (5)
d𝑡

In addition to considering grazing as a term that is subject to change through variation of fish abun-
dance (due to, e.g., changes in fishing effort), Fung et al. (2011) assumed that other model parameters
are also subject to change due to other stressors (e.g., nutrification and sedimentation). In all cases, they
found that the stressors can lead to decreases in coral resilience and therefore increase the likelihood of
phase shifts. However, discontinuous shifts (i.e., as exhibited in Fig. 1B at points 𝑇1 and 𝑇2 ) rather than
continuous shifts through smooth transitions to higher (or lower) coral cover only occur for parameter
values at the edge of their parameter ranges.

3.2 Extending the community composition

While the vast majority of models that capture alternative stable states focus on algal and coral interac-
tions and stressors to coral, there are other organisms that have competitive interactions with coral and
algae. Consequently, it is plausible that these species play a fundamental role in preventing (or driving)
phase shifts to undesirable states.
For example, sponges—which compete with coral—are known to play a functionally important role
in Caribbean coral reef ecosystems (Bell, 2008; Suchanek, Carpenter, Witman, & Harvell, 1983; Wulff,
2001). González-Rivero, Yakob, and Mumby (2011) therefore extended the coral–turf–macroalgae
framework to account the proportion of sponges 𝑆 occupying a given area of seabed. Similar to other
models presented in this paper, the proportion of algal turf can be found implicitly as 𝑇 = 1 − 𝑀 −
𝐶 − 𝑆. In this model, natural coral mortality is ignored (i.e., 𝑑 = 0 in Equation 1). It is also assumed
that sponges outcompete coral for space at a rate 𝑎2 , macroalgae can overgrow sponge at a rate 𝑎3 , and
sponge can overgrow turfs at a rate 𝑧. Finally, it is assumed that generalist predators may graze on both
macroalgae and sponges so that a proportion of the total grazing 𝑏 is allocated to sponges. This model
is given by

d𝐶
= r𝑇 𝐶 − a𝑀𝐶 − 𝑎2 S𝐶, (6)
d𝑡

d𝑀 g𝑀
= a𝑀𝐶 − + 𝛾M𝑇 + 𝑎3 M𝑆, (7)
d𝑡 𝑀 + 𝑇 + 𝑏𝑆

𝑑𝑆 𝑔𝑏𝑆
= 𝑎2 𝑆𝐶 + 𝑧S𝑇 − 𝑎3 M𝑆 − . (8)
𝑑𝑡 𝑀 + 𝑇 + 𝑏𝑆
B LACKWOOD ET AL. Natural Resource Modeling 7 of 17

In this model formulation, grazing pressure on macroalgae is diluted by the presence of sponges.
Consequently, under most scenarios explored by González-Rivero et al. (2011), the presence of sponges
increases the requisite level of grazing required to maintain high levels of coral cover and, conse-
quently, decreases coral resilience. However, there might also be direct positive effects of sponges on
coral cover. For example, sponges can enhance coral survival by binding the reef frame and temporar-
ily stabilizing carbonate rubble following physical damage. Therefore, additional modeling studies of
interactions between coral, algae, and sponge might be important to further our understanding of com-
plex reef interactions.
Another example of a modeling study that extends the community composition is related to recruit-
ment facilitation, which can also drive alternative states on temperate reefs. In particular, crustose
corraline algae (CCA) facilitates recruitment of herbivores (e.g., sea urchins), which in turn feed on
macroalgae that would otherwise outcompete CCA. This interaction was captured in the model of Bas-
kett and Salomon (2010), who find that the closer to obligate the relationship between CCA and sea
urchins is, the more hysteretic the system. In Baskett and Salomon (2010), CCA takes the place of
tropical corals but the ecological dynamics are similar. The major difference between the two mod-
els is dimensionality: Baskett and Salomon (2010) include herbivore populations as a phase variable
dependent upon CCA populations for recruitment, whereas Mumby et al. (2007)’s model treats herbi-
vore populations as a control parameter. Other notable differences include the lack of an intrinsic death
term for CCA and the assumption that macroalgae makes up the entirety of an urchin diet, as opposed
to Mumby et al. (2007)’s inclusion of algal turf.

4 MODELING THE E FFECTS OF REEF STRESSORS

The majority of ODE models that capture the dynamics of coral–algal interactions have been developed
to both assess the impacts of stressors on the resilience of coral reefs as well as explore the dynamical
responses to conservation efforts. For example, one major stressor to coral is the overharvesting of
herbivorous reef fish and a method for mitigating this issue is to impose restrictions or even a complete
moratorium on fishing. In this section, we describe modeling efforts that have focused on exploring
the consequences of overfishing, hurricane impacts and other physical damages, ocean warming, and
acidification. We also review the conclusions drawn about the dynamical effects of interventions. We
note that while we separate these stressors into two different sections, there is some overlap between
them. For example, some models consider combinations of multiple stressors.

4.1 Grazing rates and harvesting of herbivorous reef fish

Several modeling studies have considered the effects of changes in grazing rates on coral reef resilience.
Some of these studies have considered structural aspects of the coral–algal system that may impact
grazing rates while others directly consider the dynamics of herbivorous reef fish and their relationship
to the grazing rate.
One important aspect of coral–algal systems not captured in the original baseline model is external
coral and algal recruitment and its relationship to grazing rates. Including recruitment is one way in
which spatial aspects of the system can be considered, and an analysis of such a system was performed
by Elmhirst, Connolly, and Hughes (2009). Adapting the model of Mumby et al. (2007), they assumed
that coral and algal recruitment onto algal turfs occurs at rates of 𝜁 and 𝜖, respectively, and that recruit-
ment is proportional to the amount of settlement space (i.e., turf) available so that the model can be
8 of 17 Natural Resource Modeling B LACKWOOD ET AL.

rewritten as
d𝐶
= r𝑇 𝐶 − 𝑑𝐶 − a𝑀𝐶 + 𝜁 𝑇 , (9)
d𝑡

d𝑀 g𝑀
= a𝑀𝐶 − + 𝛾M𝑇 + 𝜖𝑇 . (10)
d𝑡 𝑀 +𝑇
These authors constructed bifurcation diagrams of the equilibrium coral cover versus grazing intensity,
similar to Figure 1(B), and then demonstrated how this bifurcation diagram is modified when 𝜁 and 𝜖
take on positive values. Including external recruitment of coral causes the transcritical bifurcation (at 𝑇1
in Fig. 1B) to become a saddle node bifurcation, creating a second stable equilibrium with small—but
positive—levels of coral cover between 𝑔1 and 𝑔2 . Additionally, the value of 𝑔2 decreases. In contrast,
external recruitment of macroalgae causes the transcritical bifurcation at 𝑇2 to become a saddle node
bifurcation and increases 𝑔1 , thereby causing a decline in coral resilience.
In other settings, the grazing term 𝑔 has been directly related to fish biomass. For example, Black-
wood et al. (2012) explicitly modeled herbivorous fish dynamics and assumed that the fish (assumed
to be parrotfish, the primary grazer of Caribbean reefs) abundance (𝑃 ) grows logistically but that there
is an associated harvesting rate 𝑓 . This can be written as
( )
d𝑃 𝑃
= 𝑠𝑃 1 − − 𝑓𝑃, (11)
d𝑡 𝐾
where 𝑠 is the per capita growth rate, 𝑓 captures mortality due to fishing, and 𝐾 is the carrying capacity.
They assumed that 𝐾 may vary with coral cover based on various assumptions. Importantly, they
defined grazing to be a function of 𝑃 (i.e., 𝑔(𝑃 )) so that sustainable levels of fishing effort could be
determined based on its impacts on the ecosystem dynamics. Consequently, this model provides a
relatively simple example of ecosystem-based management. This model formulation was extended in
Bhattacharyya and Pal (2015), who additionally accounted for external recruitment of macroalgae. In
their model, they assumed that macroalgae produces inhibitory toxins to coral growth. The authors
demonstrate that reef resilience declines with increases in macroalgae toxicity.
There are many other assumptions made in Blackwood et al. (2012) that have been relaxed in other
settings. For example, van de Leemput, Hughes, van Nes, and Scheffer (2016) argue that there is not
necessarily a single strong feedback—such as the positive feedback between coral and grazers—but
that there are several feedbacks whose strength varies in both time and space. Therefore, it is essential
to consider local dynamics of individual reefs and assess what feedbacks may be present. To test this,
van de Leemput et al. (2016) constructed a model with three weak feedbacks; that is, each feedback is
not independently capable of generating alternative stable states. The model they constructed makes
a couple of key assumptions: both coral and macroalgae can both grow through importation at rates
𝑖𝐶 and 𝑖𝑀 , and they grow proportionally to the amount of coral and macroalgae present at rates 𝑟
and 𝛾, respectively. Similar to other models described in this review, they assume that there is a natural
mortality rate of coral (𝑑). Finally, they explicitly model herbivore (assumed to include all reef grazers)
population dynamics and assume they take the same structure as Blackwood et al. (2012). Finally, it is
assumed that macrogalgae is removed at a rate proportional to the number of grazers and present algae
so that
d𝐶
= (𝑖𝐶 + 𝑟𝐶)𝑇 − 𝑑𝐶, (12)
d𝑡

d𝑀
= (𝑖𝑀 + 𝛾𝑀)𝑇 − 𝑔𝑀𝑃 , (13)
d𝑡
B LACKWOOD ET AL. Natural Resource Modeling 9 of 17

( )
d𝑃 𝑃
= 𝑠𝑃 1 − − 𝑓𝑃. (14)
d𝑡 𝐾
They then modified this model accordingly to test three different feedbacks: saturation of herbivore
consumption with sufficiently high algal coverage, competition between coral and macroalgae, and
increases in herbivores corresponding to increases in coral cover through, for example, more suitable
habitat conditions. In analyzing this model, the authors importantly concluded that it is possible for
multiple weak feedbacks to collectively generate phase shifts.
In another modeling study, Bozec, Yakob, Bejarano, and Mumby (2013) did not model grazer
dynamics explicitly (as in Equation 11) but they relaxed some assumptions about grazing and fishing
behaviors. For example, they considered two different genuses of parrotfish (Scarus and Sparisoma)
that had associated grazing rates of 𝐺scar (for Scarus) and 𝐺spar (for Sparisoma). Each of these terms
directly depended on properties related to each genus of parrotfish. For example, Bozec et al. (2013)
tested two assumptions on how parrotfish abundance—and therefore grazing rates—may change with
changes in rugosity (a measure of structural complexity of a reef ). Specifically, they tested both a linear
response and a Type II Holling response in which increases in parrotfish abundance eventually satu-
rates with a high enough rugosity. Such saturation may occur from, for example, density dependent
competition among parrotfish for food resources and habitat. Additionally, each species has different
grazing behavior; Scarus preferentially consume algal turf whereas Sparisoma graze indiscriminately
on both algal types (as in the original model). Now, the differential equation for macroalgal dynamics
can be written as
𝐺spar 𝑀 ( 𝐺spar 𝑇 )
d𝑀
= a𝑀𝐶 − + 𝛾M𝑇 − 𝐺scar + M𝑇 . (15)
d𝑡 𝑀 +𝑇 𝑀 +𝑇

Due to the way in which both species graze on turf, the rate of macroalgal overgrowth of turf is slowed
down due to grazing (as reflected in the final term in this equation). Finally, these authors also consider
losses due to hurricane impacts. These authors highlight that coral resilience increases with the strength
of the positive feedback between coral and grazing: if rugosity increases then parrotfish abundance
increases, which leads to an increase in grazing and, in turn, coral faces less competition and maintains
greater rugosity.
While the models described in this section directly look at the effect of changes in the grazing
rates, other models consider whether grazing rates near the critical grazing threshold—the grazing rate
at which the system shifts to a coral-depleted state—can be detected in time series data. For exam-
ple, Fung, Seymour, and Johnson (2013) expanded the model introduced by Fung et al. (2011) (see
Equations 3–5) to account for demographic stochasticity. They used stochastic simulations to assess
whether early warning signals (Scheffer et al., 2009) arise when the grazing rate is approaching its
critical threshold. While they conclude that some early warnings indicators appear near the critical
grazing threshold (such as an increase in variance), other early warning indicators (such as skewness)
do not.

4.2 Climate change

Climate change and related increases in ambient CO2 levels have led to increased ocean tempera-
tures, increased frequency and severity of hurricanes, and ocean acidification, each of which distinctly
impacts the coral reef ecosystem. When water temperatures are too high, coral lose beneficial symbi-
otic algae resulting in what is called coral bleaching. Bleaching events reduce coral survival, growth,
and reproduction (Hoegh-Guldberg, 1999). Warmer ocean temperatures have increased the frequency
10 of 17 Natural Resource Modeling B LACKWOOD ET AL.

and severity of bleaching events (Heron, 2016). Increasing ambient CO2 concentrations are causing
the ocean to become more acidic, which may reduce calcification rates and accelerate the disintegra-
tion of coral skeletons and calcified substrate (Hoegh-Guldberg et al., 2007). Increased temperatures
as a result of climate change can also lead to higher hurricane frequencies and intensity, which lead to
discrete physical disturbances to the organisms and calcium substrate.
There are many models of the alternative stable states between coral and algae that consider reef
stress under climate change in different ways. For example, Edmunds et al. (2014) developed a
metapopulation model to determine the relative importance of coral traits in responding to thermal
stress as a proxy for climate change. They model macroalgae, coral recruits, and coral adults on 𝑛
patches, which are connected through dispersal of coral and algal recruits. The model assumes that a
fraction of coral recruits stay in the patch they originate from and the rest are evenly distributed among
all other patches. We forgo the modeling details here, but the authors captured several processes that
could differ between coral species such as general life history traits (e.g., dispersal and mortality rates),
resistance to overgrowth by macroalgae, and tolerance to thermal stress. Through simulations and sen-
sitivity analyses, the authors identified thermal tolerance, growth rate, and lifespan of corals as the
primary predictors of coral resilience in the face of increasing thermal stress.
Following this study, Fabina, Baskett, and Gross (2015) developed a model to explore the conse-
quences of bleaching events as they increase in severity or frequency. Similarly to Edmunds et al.
(2014), these authors start with a nonspatial stage-structured model of coral and macroalgae, which
differentiates between coral adults and juvenile recruits. This stage structure of coral provides the flex-
ibility to capture stage-specific differences between responses to bleaching events. The three base dif-
ferential equations in the Fabina et al. (2015) model are omitted here, but have many components
similar to other basic differential equations models discussed in this paper. However, in their model,
grazing of macroalgae is broken into two components: one that captures density-independent grazing
at a fixed rate and another for density-dependent grazing that saturates as adult coral cover increases.
These authors added an additional layer to their model that stochastically captured bleaching events. To
develop this component of their model, they used climate models to determine how bleaching frequency
and magnitude would increase with changes in climate. Their findings indicate that severe bleaching
events lead to greater overall damage as compared to higher frequency but less severe events. Based
on their results, the authors delineate two primary findings related to reef conservation: protection
of corals that are less resistant to bleaching events should be prioritized when bleaching events are
relatively mild, and more resistant corals should be prioritized when bleaching events are severe.
In another study that investigated coral responses to thermal stress, Baskett, Fabina, and Gross (2014)
extended Mumby et al. (2007) to assume that there are up to two types of coral: one capable of with-
standing higher levels of stress (referred to as “resistant” coral, 𝐶𝑆 ), and a second that experiences
faster recovery following disturbance (referred to as “resilient” coral, 𝐶𝐿 ). In the setting constructed
by these authors, a given community can have either one type or both types present and it is therefore
assumed that 𝑇 = 1 − 𝐶𝑆 − 𝐶𝐿 − 𝑀. Additionally, the coral types compete with each other for space
which is captured by competition parameters 𝛼𝑆 and 𝛼𝐿 . The other parameters associated with each
coral type can differ according to their specific properties; for example, the rate at which macroal-
gae overgrows coral is specific to whether the coral is resistant or resilient. These authors additionally
assume that there is a density-independent external recruitment rate for both coral types in addition to
macroalgae (𝑟ex,𝑆 and 𝑟ex,𝐿 for resistant and resilient corals, respectively, and 𝑟ex,𝑀 for macroalgae).
Under these assumptions, the model is given by:

𝑑𝐶𝑆
= 𝑟ex,𝑆 𝑇 + 𝑟𝑆 (1 − 𝑀 − 𝐶𝑆 − 𝛼𝑆 𝐶𝐿 )𝐶𝑆 − 𝑑𝑆 𝐶𝑆 − 𝑎𝑆 𝑀𝐶𝑆 , (16)
𝑑𝑡
B LACKWOOD ET AL. Natural Resource Modeling 11 of 17

d𝐶 𝐿
= 𝑟ex,𝐿 𝑇 + 𝑟𝐿 (1 − 𝑀 − 𝛼𝐿 𝐶𝑆 − 𝐶𝐿 )𝐶𝐿 − 𝑑𝐿 𝐶𝐿 − 𝑎𝐿 𝑀𝐶𝐿 , (17)
d𝑡

d𝑀 𝑔
= 𝑟ex,𝑀 𝑇 + 𝑎𝑆 𝑀𝐶𝑆 + 𝑎𝐿 𝑀𝐶𝐿 − + 𝛾M𝑇 . (18)
d𝑡 𝑀 +𝑇
Finally, they stochastically implement damage to each species of coral due to the consequences of ther-
mal stress (e.g., coral bleaching). These authors conclude that coral resilience increases when both coral
types are present under two main conditions: the resilient coral does not outcompete the resistant coral,
and disturbances are sufficiently severe so that resilient corals cannot survive independently. Moreover,
the authors conclude that management needs to account for both site diversity and connectivity.
In general, modeling the dynamical effects of ocean warming and acidification is constrained by both
uncertainty and complexity in the underlying processes that drive them. Therefore, a common way in
which to explore the impacts of these stressors to reefs is through “implicit” approaches in which
the parameters within a model can vary as environmental conditions change. For example, Anthony
et al. (2011) adapted the model of Mumby et al. (2007) and used it to investigate how combinations
of different levels of CO2 and fishing pressure may impact reef communities. They included stochastic
background mortality as a proxy for accounting for other stressors that do not directly correlate with
CO2 (e.g., hurricane damage). In their parameterization, they assumed that ocean acidification impacts
coral calcification (and therefore coral growth rates), ocean warming affects coral mortality (through,
e.g., bleaching events), grazing pressure affects algal mortality rates, and nutrient enrichment affects
the growth rate of algae. Anthony et al. (2011) modified the original model by converting the coral
colonization rate of free space (𝑟) and coral mortality rate (𝑎) into functions that are parameterized
based on ocean acidification (as measured by the aragonite saturation state, which is a measure of
carbonate ion concentration) and sea surface temperature. These authors concluded that both ocean
warming and high levels of acidification are capable of lowering coral resilience.
While Anthony et al. (2011) and some of the models in Section 4.1 consider hurricane impacts as a
part of their analyses, we highlight several papers that focus on them more exclusively. For example,
Blackwood, Hastings, and Mumby (2011) considered multiple stressors to reefs: fishing and hurricane
damage. Fishing was implemented as described above (Equation 11) and they added an equation for
coral rugosity—which can be substantially impacted by hurricane impacts—so that it increases with the
presence of living coral (𝐶) and decreases from bioerosion. They then implement hurricanes stochas-
tically using stochastic jump processes such that hurricane impacts follow a Poisson distribution and
occur at a rate 𝜆. In their model, hurricanes instantaneously cause decreases in coral cover, macroalgal
cover, and rugosity. These authors concluded that management should be adapted according to local
conditions: if hurricane impacts are relatively low, stronger regulations on fishing should be enforced.
In contrast, if hurricane damage is high, efforts should be focused on restoration.
Related constructions of models with hurricane impacts have used the Dirac delta function to impose
an instantaneous reduction to coral, macroalgae, and turf (Ippolito, Naudot, & Noonburg, 2016). How-
ever, they did not consider rugosity. In another modeling study, Obolski, Hadany, and Abelson (2016)
performed a follow-up analysis to Blackwood et al. (2011) in which they performed a cost–benefit
analysis to analyze the potential for fish restocking as a control method. These authors concluded that
a moratorium on fishing coupled with restocking programs may improve reef health.
Many of the models we have discussed so far focus on just one or two environmental factors or
include them in simplified ways. Baskett, Nisbet, Kappel, Mumby, and Gaines (2010) attempt to quan-
titatively prioritize the most important actions we can take to protect corals by constructing a fairly
complex model, synthesizing and extending the models presented in Mumby et al. (2007), Baskett,
12 of 17 Natural Resource Modeling B LACKWOOD ET AL.

Gaines, and Nisbet (2009), and more. Their model accounts for many of the most important dynam-
ics on coral reefs, including symbiont thermal sensitivity, coral competition with macroalgae, size
structure of coral populations, and the potential effects of multiple symbiont strains, multiple coral
species, and multiple spatial patches. This model was able to reproduce most major historical trends
and bleaching events in the locations modeled (the Bahamas), and the authors used both model sen-
sitivity and model dynamics to inform their conclusions. While the authors of Baskett et al. (2009)
do not claim that their model should be used for quantitative prediction, they use qualitative com-
parisons of the model predictions to inform their management recommendations. They conclude
that the most important management goals should be (i) conservation of diverse coral communi-
ties rather than homogeneous but thermally tolerant communities, (ii) mitigation of anthropogenic
impacts that affect coral–macroalgal competition and juvenile coral survivorship, and (iii) protection
of both locations and connectivity to locations with oceanographic features that lead to lower thermal
stress.

5 DIS CUSSIO N

Coral reefs support an incredible amount of biodiversity and provide numerous ecosystem services
ranging from fisheries production to coastal defense from storms (Moberg & Folke, 1999). In the
Caribbean, many stressors including those related to declines in algal grazing following the great urchin
die-off of the early 1980s provide a constant challenge to the resilience of reefs. In particular, the exis-
tence of two alternative stable states—a desirable state with high levels of coral cover and a degraded
coral-depleted state—has been frequently hypothesized in the literature. Understanding the mecha-
nisms that may lead to the presence of alternative stable states as well as processes that underlie phase
shifts between them is a central research priority for managing coral reef ecosystems.
In this review paper, we highlighted several modeling studies that are centered on the most parsi-
monious explanation for the presence of alternative stable states. In particular, declines in grazing of
macroalgae by herbivorous reef fish allow macroalgae to outcompete coral and proliferate, ultimately
resulting in a phase shift to the degraded algal-dominated state. Subsequent increases in grazing gives
rise to algal turfs and thereby frees up space for coral to recruit to and then proliferate, leading to a
stable state with high levels of coral cover. Early influential models of algal–coral interactions cen-
tered almost exclusively on these processes (e.g., Fung et al., 2011; Mumby et al., 2007). Since the
initial conception of these models, many more studies have been performed that build upon the central
algal–coral interaction framework. This review provides a summary of these related modeling studies
to date, ranging from models that expand the community composition to evaluating the resilience of
coral when faced with one or more of its many stressors.
While many modeling studies exist, several considerations remain that may be important for future
research. For example, while some studies explicitly integrate algal–coral dynamics with the population
dynamics of herbivores, interactions between trophic levels of fish are typically not accounted for. Such
studies that include additional trophic interactions could reveal important considerations for designing
management strategies. As a second example, a recent paper used the Fung et al. (2011) framework to
demonstrate that the bottom-up processes of sedimentation and eutrophication (which can result from
coastal development) are capable of driving phase shifts to alternative stable states in reef ecosystems
(Arias-González et al., 2017). The importance of these stressors has also been empirically documented
(e.g., Lapointe, 1997). However, sedimentation and eutrophication have been given much less atten-
tion as potential drivers of phase shifts in modeling studies centered on algal–coral dynamics. Further
research is therefore needed to better understand potential management implications.
B LACKWOOD ET AL. Natural Resource Modeling 13 of 17

While the focus of this study is on models of ODEs, several other frameworks exist to investigate
coral–algal dynamics in more complex settings. For example, several approaches have been utilized to
explicitly model populations spatially. Mumby (2006) developed a spatial simulation model in which
coral–algal interactions occur over a large grid of cells. In another setting, Melbourne-Thomas et al.
(2011) developed a large-scale model on a spatial grid where local dynamics were determined by differ-
ential equations describing the dynamics of eight key organisms and larval transport connected patches
regionally. Spiecker, Gouhier, and Guichard (2016) implemented their integro-difference equations on
200 coupled patches, which were connected through material transport and dispersal and analyzed the
dynamics to test the effectiveness of various configurations of marine reserves.
In addition to focusing on models constructed of ODEs, our emphasis here is on models derived pre-
dominantly from studies of Caribbean coral reefs. Nevertheless, the same stressors of overfishing, bio-
erosion, elevated temperatures, increasing ocean acidity, local impacts of human habitation, etc., affect
the reefs of the Indo-Pacific as well with concomitant increases in macroalgae cover and decreases
in coral cover (Bruno & Selig, 2007; Glynn & Manzello, 2015; Nordström, Nordström, Lokrantz &
Folke, 2009). Thus, it is likely that this modeling approach is applicable to reef dynamics in the Indo-
Pacific as well. The Indo-Pacific did not experience the great Diadema die-off of the Caribbean but the
consequences of the loss of herbivory afflicts reefs in both regions. Perhaps the most important unique
stressor of the Indo-Pacific is the recurrence of Acanthaster predation outbreaks (Clements & Hay,
2017), which might need to be accommodated in models for greater predictive power in Indo-Pacific
reefs.
More generally, mathematical models serve as a complementary tool to field and experimental stud-
ies. We highlight a range of mathematical models that have contributed to the growing body of litera-
ture on coral–algal interactions and phase shifts in Caribbean reefs. Further studies that expand upon
these ideas will be important in obtaining a more holistic understanding of complex coral ecosystem
dynamics.

ACKNOW LEDGMENTS
We thank Ondrej Maxian, Anna Neufeld, and Emma Talis for helpful discussions during the early
stages of this work. Our research was supported in part by the National Science Foundation REU
Grant DMS—1347804, the Williams College Science Center, and the Clare Boothe Luce Program of
the Henry Luce Foundation.

REFERENCES
Anthony, K., Maynard, J. A., Diaz-Pulido, G., Mumby, P. J., Marshall, P. A., Cao, L., & Hoegh-Guldberg, O. (2011).
Ocean acidification and warming will lower coral reef resilience. Global Change Biology, 17, 1798–1808.
Arias-González, J. E., Fung, T., Seymour, R. M., Garza-Pérez, J. R., Acosta-González, G., Bozec, Y. M., & Johnson,
C. R. (2017). A coral-algal phase shift in Mesoamerica not driven by changes in herbivorous fish abundance. PLoS
ONE, 12, e0174855.
Baskett, M. L., Fabina, N. S., & Gross, K. (2014). Response diversity can increase ecological resilience to disturbance
in coral reefs. American Naturalist, 184, E16–E31.
Baskett, M. L., Gaines, S. D., & Nisbet, R. M. (2009). Symbiont diversity may help coral reefs survive moderate climate
change. Ecological Applications, 19, 3–17.
Baskett, M. L., Nisbet, R. M., Kappel, C. V., Mumby, P. J., & Gaines, S. D. (2010). Conservation management approaches
to protecting the capacity for corals to respond to climate change: A theoretical comparison. Global Change Biology,
16, 1229–46.
14 of 17 Natural Resource Modeling B LACKWOOD ET AL.

Baskett, M. L., & Salomon, A. K. (2010). Recruitment facilitation can drive alternative states on temperate reefs. Ecology,
91, 1763–1773.
Bell, J. J. (2008). The functional roles of marine sponges. Estuarine, Coastal and Shelf Science, 79, 341–353.
Bhattacharyya, J., & Pal, S. (2015). Hysteresis in coral reefs under macroalgal toxicity and overfishing. Journal of Bio-
logical Physics, 41, 151–172.
Blackwood, J. C., & Hastings, A. (2011). The effect of time delays on Caribbean coral-algal interactions. Journal of
Theoretical Biology, 273, 37–43.
Blackwood, J. C., Hastings, A., & Mumby, P. J. (2011). A model-based approach to determine the long-term effects of
multiple interacting stressors on coral reefs. Ecological Applications, 21, 2722–2733.
Blackwood, J. C., Hastings, A., & Mumby, P. J. (2012). The effect of fishing on hysteresis in Caribbean coral reefs.
Theoretical Ecology, 5, 105–114.
Bozec, Y. M., Yakob, L., Bejarano, S., & Mumby, P. J. (2013). Reciprocal facilitation and non-linearity maintain habitat
engineering on coral reefs. Oikos, 122, 428–440.
Bruno, J., & Selig, E. R. (2007). Regional decline of coral cover in the Indo-Pacific: Timing, extent, and subregional
comparisons. PLoS ONE, 2, e711.
Clements, C. S., & Hay, M. E. (2017). Size matters: Predator outbreaks threaten foundation species in small marine
protected areas. PLoS ONE, 12, e0171569.
Connell, J., & Sousa, W. (1983). On the evidence needed to judge ecological stability or persistence. American Naturalist,
121, 789–824.
Cramer, K. L., O’Dea, A., Carpenter, C., & Norris, R. D. (2017). A 3,000 year record of Caribbean reef urchin commu-
nities reveals causes and consequences of long-term decline in Diadema antillarum. Ecogeography, 40, 1–10.
Donner, S. (2011). An evaluation of the effect of recent temperature variability on the prediction of coral bleaching
events. Ecological Applications, 21, 1718–1730.
Dudgeon, S. R., Aronson, R. B., Bruno, J. F., & Precht, W. F. (2010). Phase shifts and stable states on coral reefs. Marine
Ecology Progress Series, 413, 201–216.
Edmunds, P. J., Adjeroud, M., Baskett, M. L., Baums, I. B., Budd, A. F., Carpenter, R. C., … Gates, R. D. (2014).
Persistence and change in community composition of reef corals through present, past, and future climates. PLoS
ONE, 9, 1–12.
Elmhirst, T., Connolly, S. R., & Hughes, T. P. (2009). Connectivity, regime shifts and the resilience of coral reefs. Coral
Reefs, 28, 949–957.
Fabina, N. S., Baskett, M. L., & Gross, K. (2015). The differential effects of increasing frequency and magnitude of
extreme events on coral populations. Ecological Applications, 25, 1534–1545.
Fung, T., Seymour, R. M., & Johnson, C. R. (2011). Alternative stable states and phase shifts in coral reefs under anthro-
pogenic stress. Ecology, 92, 967–982.
Fung, T., Seymour, R. M., & Johnson, C. R. (2013). Warning signals of regime shifts as intrinsic properties of endogenous
dynamics. American Naturalist, 182, 208–222.
Glynn, P. W., & Manzello, D. P. (2015). Bioerosion and coral reef growth: a dynamic balance. In C. Birkeland (Eds.),
Coral reefs in the Anthropocene (pp. 67–97). Neverlands: Springer, Dordrecht.
González-Rivero, M., Yakob, L., & Mumby, P. J. (2011). The role of sponge competition on coral reef alternative steady
states. Ecological Modelling, 222, 1847–1853.
Heron, S. F., Maynard, J. A., van Hooidonk, R., & Eakin, C. M. (2016). Warming trends and bleaching stress of the
world’s coral reefs 1985-2012. Scientific Reports, 6, 38402.
Hoegh-Guldberg, O. (1999). Climate change, coral bleaching and the future of the world’s coral reefs. Marine and
Freshwater Research, 50, 839–866.
Hoegh-Guldberg, O., Mumby, P. J., Steneck, R. S., Greenfield, P., Gomez, E., Harvell, C. D., … Hatziolos, M. E. (2007).
Coral reefs under rapid climate change and ocean acidification. Science, 318, 1737–1742.
B LACKWOOD ET AL. Natural Resource Modeling 15 of 17

Holling, C. S. (1973). Resilience and stability of ecological systems. Annual Review of Ecology and Systematics, 4, 1–
23.
Ippolito, S., Naudot, V., & Noonburg, E. G. (2016). Alternative stable states, coral reefs, and smooth dynamics with a
kick. Bulletin of Mathematical Biology, 78, 413–435.
Lapointe, B. E. (1997). Nutrient thresholds for bottom-up control of macroalgal blooms on coral reefs in Jamaica and
southeast Florida. Limnology and Oceanography, 42, 1119–1131.
Lessios, H., Garrido, M., & Kessing, B. (2001). Demographic history of Diadema antillarum, a keystone herbivore on
Caribbean reefs. Proceedings of the Royal Society, 268, 1–7.
Levitan, D., Edmunds, P., & Levitan, K. (2014). What makes a species common? No evidence of density-dependent
recruitment or mortality of the sea urchin Diadema antillarum after the 1983-1984 mass mortality. Oecologia, 175,
117–128.
Li, X., Wang, H., Zhang, Z., & Hastings, A. (2014). Mathematical analysis of coral reef models. Journal of Mathematical
Analysis and Applications, 416, 352–373.
Melbourne-Thomas, J., Johnson, C. R., Fung, T., Seymour, R. M., Chérubin, L. M., Arias-González, J. E., & Fulton,
E. A. (2011). Regional-scale scenario modeling for coral reefs: A decision support tool to inform management of a
complex system. Ecological Applications, 21, 1380–1398.
Miller, M., Kramer, K., Williams, S., Johnston, L., & Szmant, A. (2009). Assessment of current rates of Diadema antil-
larum larval settlement Coral Reefs. Coral Reefs, 28, 511–515.
Moberg, F., & Folke, C. (1999). Ecological goods and services of coral reef ecosystems. Ecological Economics, 29,
215–233.
Mumby, P. J. (2006). The impact of exploiting grazers (Scaridae) on the dynamics of Caribbean coral reefs. Ecological
Applications, 16, 747–769.
Mumby, P. J., Hastings, A., & Edwards, H. J. (2007). Thresholds and the resilience of Caribbean coral reefs. Nature,
450, 98–101.
Murphy, G. N., Perry, C. T., Chin, P., & McCoy, C. (2016). New approaches to quantifying bioerosion by endolithic
sponge populations: Applications to the coral reefs of Grand Cayman. Coral Reefs, 35, 1109–1121.
Nordström, A. V., Nordström, M., Lokrantz, J., & Folke, C. (2009). Alternative stable states on coral reefs: Beyond
coral-macroalgal phase shifts. Marine Ecology Progress Series, 376, 295–306.
Obolski, U., Hadany, L., & Abelson, A. (2016). Potential contribution of fish restocking to the recovery of deteriorated
coral reefs: An alternative restoration method? PeerJ., 4, e1732.
Scheffer, M., Bascompte, J., Brock, W. A., Brovkin, V., Carpenter, S. R., Dakos, V., … Sugihara, G. (2009). Early-
warning signals for critical transitions. Nature, 461, 53–59.
Sherman, E. (2015). Can sea urchins beat the heat? Sea urchins, thermal tolerance and climate change. PeerJ, 3, e1006.
Spiecker, B., Gouhier, T. C., & Guichard, F. (2016). Reciprocal feedbacks between spatial subsidies and reserve networks
in coral reef meta-ecosystems. Ecological Applications, 26, 264–278.
Suchanek, T. H., Carpenter, R. C., Witman, J. D., & Harvell, C. D. (1983). Sponges as important space competitors in
deep caribbean coral reef communities. Symposia Series for Undersea Research, 1, 55–60.
van de Leemput, I. A., Hughes, T. P., van Nes, E. H., & Scheffer, M. (2016). Multiple feedbacks and the prevalence of
alternate stable states on coral reefs. Coral Reefs, 35, 857–865.
Wulff, J. (2001). Assessing and monitoring coral reef sponges: Why and how? B Marine Science, 69, 831–846.

How to cite this article: Blackwood JC, Okasaki C, Archer A, Matt EW, Sherman E, Montovan
K. Modeling alternative stable states in Caribbean coral reefs. Nat Resour Model. 2018;e12157.
https://doi.org/10.1111/nrm.12157
16 of 17 Natural Resource Modeling B LACKWOOD ET AL.

APPENDIX A : STA BILIT Y ANALYSIS OF BA SELINE CORAL–ALGAL

MODEL
In this section, we provide a detailed stability analysis for the original model of Mumby et al. (2007).
First, we can identify the model equilibria by setting the right-hand side of Equations (1) and (2) equal
to zero or, equivalently, define:

𝑓1 = (𝑟𝑇 − 𝑑 − 𝑎𝑀)𝐶, (A1)

( 𝑔 )
𝑓2 = 𝑎𝐶 − + 𝛾𝑇 𝑀, (A2)
𝑀 +𝑇
and identify the points at which 𝑓1 = 𝑓2 = 0. Replacing 𝑇 with 1 − 𝑀 − 𝐶, from Equation (A1), we
obtain the equilibrium conditions that 𝐶 ∗ = 0 and 𝑟(1 − 𝑀 ∗ − 𝐶 ∗ ) − 𝑑 − 𝑎𝑀 ∗ = 0. Similarly, from
𝑔
Equation (A2) we have that 𝑀 ∗ = 0 and 𝑎𝐶 ∗ − 1−𝐶 ∗ ∗
∗ + 𝛾(1 − 𝑀 − 𝐶 ) = 0. From this, we can obtain
five equilibria:
{ ( ) ( ) ( }
𝛾 −𝑔 𝑟−𝑑 ) ( )
(𝐶 ∗ , 𝑀 ∗ ) = (0, 0), 0, , ̂ 1 , 𝐶̂2 , 𝑀
, 0 , 𝐶̂1 , 𝑀 ̂2 , (A3)
𝛾 𝑟

̂ 𝑖 ≠ 0 for 𝑖 = 1, 2. For the second two equilibria to have biological relevance, we require
where 𝐶̂𝑖 , 𝑀
that 𝛾 > 𝑔 (macroalgae spreads over turf faster than it is grazed) and 𝑟 > 𝑑 (coral overgrows turf
faster than it succumbs to natural mortality). The final two equilibria are found by simultaneously
𝑔
solving 𝑟(1 − 𝑀 ∗ − 𝐶 ∗ ) − 𝑑 − 𝑎𝑀 ∗ = 0 and 𝑎𝐶 ∗ − 1−𝐶 ∗ ∗
∗ + 𝛾(1 − 𝑀 − 𝐶 ) = 0. After simplifying,
these equilibria can be found explicitly as the roots of a quadratic in 𝐶. As shown in Figure 1B, the final
equilibrium only exists for a range of values of 𝑔. Additionally, only the larger of these two equilibria
is non-negative for the parameterization used in Figure 1B; therefore, hereafter we only consider the
larger of the two and redefine it as (𝐶, ̂ 𝑀).
̂
To analyze the stability of these equilibria, we first find the Jacobian matrix, 𝐽 :

⎡ 𝜕𝑓1 𝜕𝑓1 ⎤ [ ]
⎢ 𝜕𝐶 ⎥ 𝑟(1 − 2𝐶 − 𝑀) − 𝑑 − 𝑎𝑀 −(𝑟 + 𝑎)𝐶
𝜕𝑀 𝑔 𝑔
⎢ 𝜕𝑓 𝜕𝑓2 ⎥ = (𝑎 − 𝛾)𝑀 − 𝑀 𝑎𝐶 − + 𝛾(1 − 𝐶 − 2𝑀) .
⎢ 2 ⎥ (1 − 𝐶)2 1−𝐶
⎣ 𝜕𝐶 𝜕𝑀 ⎦

The stability of each equilibrium can be found by determining the sign of the eigenvalues of 𝐽 evaluated
at the equilibrium. At the origin, the eigenvalues are given by 𝜆1 = 𝑟 − 𝑑 and 𝜆2 = 𝛾 − 𝑔. By our
assumption that 𝛾 > 𝑔 and 𝑟 > 𝑑, both eigenvalues are positive. Recalling that an equilibrium is locally
unstable if both eigenvalues positive, then the origin is unstable.
For the equilibrium corresponding to healthy state with high levels of coral cover, we find that the
eigenvalues are:

𝜆1 = 𝑑 − 𝑟,
( ) 𝑔𝑟 𝛾𝑑
𝑟−𝑑
𝜆2 = 𝑎 − + .
𝑟 𝑑 𝑟
In this case, the first eigenvalue is always negative (by our assumption that 𝑟 > 𝑑). The second eigen-
value is negative only when
( ) 𝛾𝑑 2
𝑎𝑑 𝑟−𝑑
𝑔> + . (A4)
𝑟 𝑟 𝑟2
B LACKWOOD ET AL. Natural Resource Modeling 17 of 17

Recalling that an equilibrium is locally stable if all of its eigenvalues are negative valued, then this
equilibrium is only stable when the grazing rate is sufficiently high. Moreover, notice that when the
grazing rate is equal to the right-hand side of Equation (A4), it is exactly equal to 𝑔1 in Figure 1B.
For the equilibrium corresponding to a degraded coral-depleted state, we find that the eigenvalues
are:
𝛾 −𝑔
𝜆1 = (𝑟 − 𝑑) − (𝑟 + 𝑎) ,
𝛾
𝜆2 = 𝑔 − 𝛾.

While 𝜆2 < 0, 𝜆1 < 0 only when

( )
𝑎+𝑑
𝑔<𝛾 . (A5)
𝑎+𝑟
This equilibrium is locally asymptotically unstable when both eigenvalues are negative; therefore, the
coral-depleted state is stable for sufficiently low levels of grazing and unstable otherwise. Note the
grazing rate takes on the value on the right hand side of Equation (A5), this is exactly the value 𝑔2 as
displayed in Figure 1B.
The final equilibrium where both coral and macroalgal cover are nonzero, (𝐶, ̂ 𝑀),
̂ is much more
algebraically complicated, so we omit the details here. However, it can be shown that under our param-
eterization this equilibrium (when biologically relevant) is always a saddle point. Importantly, this
stability analysis demonstrates the presence of alternative stable states.

View publication stats