The decision to implement one or more interventions to support coral resilience under global environmental change exists within a broader decision context, in which multiple natural and human-driven forces, pressures, environmental states, and environmental and management responses interact. Therefore, such a decision should be made with consideration of other (local) stressors (e.g., poor water quality), a suite of management programs and restoration activities (e.g., management of these local stressors), and a range of human values. Within this decision context, there will likely be multiple and potentially conflicting stakeholder objectives. Additionally, there will be uncertainty about system dynamics, future conditions, and the risks and benefits of a particular decision. Different management activities, including the use of coral interventions, will vary in risk and uncertainty and in their ability to achieve defined objectives. Decision support tools and best practices exist to guide the evaluation of tradeoffs in achieving various management objectives that can be expected given the state of knowledge. An adaptive decision approach provides a structured framework for evaluating potential management actions using an iterative process that allows for continuous learning about the linked human–natural system and improvement of management decisions. The use of a decision support framework that integrates an adaptive management process requires stakeholder engagement, data collection and monitoring, and modeling efforts, but is an investment that pays off with a prioritized set of coordinated,
comprehensive, scientifically driven, and long-term strategies to achieve long- and short-term goals.
The goal of this chapter is to provide an overview of best practices for structured decision making, including high-level descriptions of specific decision analytic approaches that are useful for evaluating the consequences and tradeoffs arising from implementing particular coral interventions. An array of approaches and tools is available, with the suitability of their use being dependent on local conditions, resources, or state of knowledge. Many of these tools have been implemented in coral reef decision contexts, and example situations have been highlighted in this chapter. Though typically implementation of these tools has been limited to management of local stressors, they have utility for evaluating coral interventions as part of an overall management strategy.
Coral reef management is dynamic and complex, and the expectation is that a number of different actions will be required across temporal and spatial scales as coral reefs respond to multiple human pressures and deteriorating environmental conditions. As environmental conditions change, local reef managers and decision makers will benefit from access to new interventions that have the potential to increase the persistence and resilience of coral reefs in the face of increased thermal stress, ocean acidification, and potential disease outbreaks. Additionally, the intervention landscape itself is highly dynamic as new technologies and approaches are developed, tested, and better understood, continually providing new and improved management options. Although it may be tempting to think that these interventions can be evaluated and implemented in isolation, it is important to address the effects of traditional stressors (e.g., contaminant loadings, fishing pressures, water quality), which will influence the probability of intervention success due to the complex interactions across attributes of the coral reef ecosystem.
The DPSIR (drivers–pressures–state–impact–response) framework (Atkins et al., 2011; Kelble et al., 2013; Rehr et al., 2012) provides a useful organizing framework for structured decision making (see Figure 3.1) by linking socioeconomic factors and human activities with changes in biophysical metrics of coral reef health associated with declines or improvements in ecosystem services. The DPSIR framework provides a structure for integrating environmental and socioeconomic relationships, the understanding of which is developed based on monitoring
data, analytical studies, predictive models, expert judgment, and other studies. Driving forces are the socioeconomic and cultural factors leading to the human activities that in turn create pressures on coral reefs. These range from local drivers such as tourism and coastal land development to the global drivers contributing to climate change. The resulting pressures include local inputs of pollutants, as well as thermal stress and ocean acidification. The pressures result in degraded environmental states and coral reef condition. The impact of that degradation is a decrease in ecosystem services and the benefits provided by coral reefs to local communities. Finally, there are a set of responses, or management alternatives, that are directed at improving the state of coral reefs by affecting the pressures on, or state of, the reef. These alternatives will have differing impacts on coral reef condition. The DPSIR framework can be used as a tool for establishing the decision context with stakeholders as described in Step 1 of the process outlined in this chapter (as seen in Bradley and Yee, 2015; Yee et al., 2015).
Structured decision making is a deliberative process for evaluating how a set of alternatives can best achieve stakeholder and decision-maker objectives by maximizing benefits and minimizing potential risks (Gregory et al., 2012). It guides the management responses within the larger landscape depicted in the DPSIR framework. Effective decision making under uncertainty benefits from an adaptive and structured decision-making strategy as shown in Figure 3.2 by allowing for an iterative process of planning, acting, evaluating, and responding over dynamic spatial and temporal scales. Adaptive management consists of planning, evaluating options, establishing monitoring goals, and iteratively adjusting management plans depending on the continuing evaluation of changes in coral reef function, structure, and health (Holling, 1973; Walters, 1986; Walters and Holling, 1990). Figure 3.2 is nested in Figure 3.1 to reflect the focused decision landscape necessary to compare potential responses and interventions within a broader management context.
A strength of the structured decision-making approach is the focus on eliciting objectives, decision criteria weights, and evaluation metrics from a variety of stakeholders who may have fundamental disagreements around management of a shared resource. Decisions made on the basis of “intuition” or “gut feelings” may not reflect the complete underlying values of diverse stakeholders or objective functions of decision makers
(Addison et al., 2013; Hammond et al., 1999; Howard, 1988; Keeney and Raiffa, 1993; Runge, 2011). Structured decision making provides a way to compare different attributes and outcomes across potential alternatives in a consistent way and standardizes disparate quantitative and qualitative outcomes that are difficult to compare otherwise. The first step in any structured decision-making process is problem formulation to establish the scale of the decision context and the specific objectives, regulatory landscape, and array of stakeholders to be involved in the process (e.g., Gregory et al., 2012; Runge, 2011). The localized context of this first step is the primary reason why prescriptive, externally developed approaches are rarely useful, let alone successful.
The process of developing a decision framework can be as important an accomplishment as the application of the framework to evaluate alternatives. For example, Fletcher et al. (2015) conducted a needs assessment survey with coral reef managers to identify data needs around the relationship between climate change and coral reef health, and how those needs might best support the development of prototype decision support tools. The study was conducted over nearly 5 years with the result that the median rating by participants around a set of climate tools was only 3 (moderately useful) out of a scale of 1 to 5 (not useful to extremely useful). This was even after completing a tutorial around each of the climate tools. This exercise highlights that the application of structured decision-making approaches typically requires facilitation and engagement. While the decision tools themselves are useful, it is the engagement involving people that produces solutions. The structured decision-making process itself can be used as a two-way process whereby the decision analysts and system modelers learn from stakeholders and decision makers, and decision makers and stakeholders gain new clarity around what options they have available and what choice of strategy might deliver the greatest benefits at minimal risks given multiple objectives and value preferences.
Step 1: Identify Stakeholders, Objectives, and Risk Tolerance (Problem Formulation or Decision Structuring)
The decision context includes long- and short-term goals; ecological, economic, and social objectives (defined and measurable strategies or steps to meet these goals); and decision criteria. Problem formulation, or decision structuring, is the process of transitioning from a vague articulation of a poorly defined problem by decision makers and stakeholders to a clear definition with an associated analysis framework (von Winterfeldt and Edwards, 2007). Each stage of the adaptive management process involves interaction and collaboration among decision makers, stakeholders, scientists, and decision analysts. It is in the first problem formulation stage that these key actors and their roles are identified. Although decision makers have primary responsibility for problem formulation, stakeholder preferences are required to establish goals and objectives that will lead to the criteria against which interventions will be evaluated. Scientists and coral reef ecologists will have insight into local conditions from a biophysical and ecological perspective. Local scientific knowledge will be required to develop predictive models of how the ecosystem (and linked social system) might respond to interventions (consequences) to contribute to the identification of critical uncertainties. Local knowledge,
in collaboration with stakeholders, may also establish appropriate and targeted monitoring programs that can help assess the performance of interventions against established criteria (e.g., Nichols and Williams, 2006) and provide early warning for where an intervention might be critically needed (Dale and Beyeler, 2001). Environmental and ecological consequences then need to be translated to impacts on ecosystem services and values for people to inform how different strategy options affect multiple objectives. Problem formulation formalizes a shared understanding of the policy, legal, ecological, social, and economic drivers and how they give rise to the specific decision-making context.
The U.S. Environmental Protection Agency’s (EPA’s) Coral Reef Protection Plan has the goal of reducing anthropogenic stress on Caribbean coral reefs by promoting institutional practices that improve reef condition and directing regulatory and nonregulatory decision making toward minimizing contaminant releases to coastal systems (described in Carriger et al., 2018). Although the Coral Reef Protection Plan emphasizes conventional contaminants and stressors, it provides a good example of structuring management objectives, subobjectives, and attributes to carry through the entire adaptive management process. Table 3.1 shows the relevant objectives identified for the Coral Reef Protection Plan based on a series of stakeholder/decision-maker engagement meetings. Identifying the overall objectives and potential associated metrics (identified as subobjectives 1 and 2 in Table 3.1) is critical to establishing stakeholder priorities, selecting which interventions might be appropriate for further consideration in a decision framework, and identifying the boundaries of the decision. Failure to reach consensus (or at least reconcile different views) on what is important can lead to the failure of any intervention or restoration program.
Given that coral reef ecosystems provide a suite of services that translate to benefits and values for society (e.g., Costanza et al., 2014; Stoeckl et al., 2011), the ecosystem services provided by the reef in question will affect the risk tolerance of stakeholders and decision makers and their willingness to trade off between key values. The ecosystem services generated by coral reefs will vary from system to system, both from a biophysical perspective as well as with respect to stakeholder values (e.g., Hicks et al., 2015). EPA and others have developed assessments of ecosystem services derived from coral reefs worldwide (Brander et al., 2013; Principe et al., 2011), identifying at least 30 individual ecosystem services ranging from direct extractive uses (e.g., fishing) to direct nonextractive uses (e.g., recreational activities such as diving). These services also encompass indirect uses (e.g., reduced flooding, fish habitat) and nonuse values (e.g., existence value). The specific values that stakeholders and
TABLE 3.1 Coral Reef Protection Objectives Identified by EPA Based on Stakeholder Meetings for the Coral Reef Protection Plan
|Category||Fundamental Objective||Subobjective 1||Subobjective 2|
|Environmental||Protect, restore, and enhance ecological integrity of coral reef systems||Individual coral colonies||Endangered or threatened colonies|
|Non-endangered nor threatened colonies|
|Coral reef communities|
|Economic||Protect, restore, and enhance economic benefits from coral reef systems||Property protection from storm waves|
|Economic benefits from reef-related activities||Tourism/visitation|
|Increase employment in reef-related industries|
|Social||Protect, restore and enhance social benefits from coral reef systems||Traditional uses of reef resources||Availability of coral fish species and resources for traditional uses (e.g., festivals, local markets)|
|Traditional fishing and harvesting of reef resources|
|Human health||Protect, restore, and enhance human health benefits from coral reef systems||Protection of human lives from storm waves||Mortality from storm waves|
|Morbidity from storm waves|
|Sustenance from fisheries species|
|Governance/political commitments||Foster long-term public support and trust|
SOURCE: Recreated from Carriger et al., 2018.
decision makers place on ecosystem services generated by coral reefs will be context specific and are likely to vary both spatially and temporally.
Stakeholder and decision-maker tolerance and acceptance of intervention risks is another important attribute that will influence the evaluation of which intervention or set of interventions to try at a given time. For example, stakeholders may express a higher tolerance for risks associated with largely untested interventions if a reef is considerably more degraded (see Chapter 2 for a discussion of such context dependencies). These values are only revealed through a transparent decision-making process combined with active stakeholder engagement (Anthony et al., 2017; Kaebnick et al., 2016).
A variety of tools and approaches exist to facilitate setting up problem formulation or what has also been called decision structuring, including means–ends diagrams, objectives hierarchies, and value trees (see von Winterfeldt and Edwards, 2007, for a lengthy discussion of decision structuring; see also www.structureddecisionmaking.org). These are largely graphical methods that link actors, actions, and objectives and are used to identify the key underlying issues, select an appropriate analytical approach, and refine the analysis structure.
Step 2: Model Linkages Across Interventions, Biophysical Outcomes, and Objectives
Any proposed intervention will lead to biophysical consequences and changes in environmental states and coral reef condition. The impacts of interventions may occur at different scales from the scales at which interventions are applied. It is critical to model, as quantitatively as possible, the range of biophysical (and, as appropriate, social or economic) consequences or possible environmental states predicted to occur as a result of an intervention or set of interventions. Model design and input parameters should be tailored to specific locations at relevant spatial and temporal scales. Biophysical models should assume that coral reef systems are dynamic and characterized by stochastic variability, and that there is uncertainty in knowledge about these dynamics. Biophysical outputs can include such attributes as coral growth and cover over time, coral diversity, herbivore biomass, coral disease, macroalgae cover, and other metrics identified by researchers, stakeholders, and decision makers as critical indicators of coral reef health and resilience (e.g., Anthony et al., 2015; Maynard et al., 2017; McClanahan et al., 2012). Chapter 4 provides a simplified example of the kind of biophysical modeling required to predict environmental states as a consequence of implementation of one or more interventions. Developing this critical component of any decision framework requires an understanding of the potential effects and impacts
of the intervention. The goal is to identify which proposed set of interventions will deliver anticipated maximum benefits at minimum risks and costs, and what research and development might be required to reduce critical uncertainty and further improve strategy options.
Table 3.2 provides an overview of the most common biophysical models used to assess coral reef condition. A critical first step in the development of a quantitative model or set of quantitative models is constructing a conceptual model of coral reef interactions. This conceptual understanding of reef dynamics is critical for identifying relationships between variables driving biophysical outcomes, including feedback loops. Once the qualitative relationships and interactions have been articulated, then it is possible to determine the mathematical relationships that will quantify those interactions. This could be an equation or a set of equations, and there may already be existing modeling frameworks from the peer-reviewed literature and elsewhere that can provide a starting point, as shown in Table 3.2.
The exact form of the functional relationships between intervention implementation and environmental states or biophysical outcomes will be context- and location-specific. Those relationships are used to define one or more quantitative metric(s) against which to evaluate intervention success. Given that some of the interventions presented in Table 1.1 may be untested at field scale, it is important to note that, in many cases, the risks and impacts of intervention implementation will be poorly understood and therefore difficult to constrain in a modeling context. There may be high uncertainty around risks and benefits of untested interventions, and further, around how they might perform under different climate change scenarios. Chapter 4 provides a highly simplified proof-of-concept example that clearly demonstrates this, but also demonstrates the utility of decision analytic approaches in situations of high uncertainty through likelihood-based calculations under different contexts.
Oceanographic models translate oceanographic or climatological data across locations and through time (see Figure 3.3a) into projected outcomes for coral reefs (see Figure 3.3b), such as temperature into bleaching frequency (e.g., Donner, 2009; Logan et al., 2014) or ocean acidification into the capacity for coral calcification (e.g., Hoegh-Guldberg et al., 2007). Oceanographic models can represent the expected spatial scale of anticipated benefits of physical interventions (e.g., shading, mixing of cool water) and associated risks that relate to environmental modification. Oceanographic models that incorporate larval properties (e.g., anticipated release locations and timing, larval duration) can indicate expected connectivity among locations (e.g., Riginos, 2018; Wood et al., 2016), which
TABLE 3.2 Summary of Common Coral Biophysical Models
|Model Type||Brief Description||Examples|
|Oceanographic (stress-based) models (Figure 3.3b)||Translates environmental dynamics, such as the expected frequency of bleaching events, into predictors of coral demographics such as persistence or connectivity||Donner, 2009; Logan et al., 2014; Wood et al., 2016|
|Population genetic models (Figure 3.3f,g)||Captures evolutionary change in traits, such as thermal tolerance, as it depends on the fitness (survival and reproductive success) of different genotypes under changing environmental conditions||Baskett et al., 2009; Bay et al., 2017|
|Physiological models (Figure 3.3h)||Follows physiological dynamics of energetic exchange among a coral host, symbiotic algae, and the microbiome (e.g., dynamic energy budget models)||Cunning et al., 2017; Muller et al., 2009|
|Stage-structured population models (i.e., matrix models; Figure 3.3e)||Follows the population size distributed among discrete stages such as ages or size classes, where demographic rates (growth, reproduction, mortality) depend on stage||Edmunds and Elahi, 2007; Fong and Glynn, 1998; Hughes and Tanner, 2000|
|Integral projection population models (Figure 3.3d)||Follows the population size distributed over a continuous state such as size, where demographic rates (growth, reproduction, mortality) are a function of state and can depend on the physical and chemical environment||Edmunds et al., 2014a; Madin et al., 2012|
|Individual-based simulations||Captures demographic stochasticity by following individual-by-individual state and demographic transitions (growth, reproduction, mortality)||Langmead and Sheppard, 2004; Mumby, 2006; Mumby et al., 2006|
|Community models (Figure 3.3c)||Can follow the dynamics of multiple coral species or types as well as additional tropical reef taxa such as algae and herbivores||Blackwood et al., 2011; Fung et al., 2011; Melbourne-Thomas et al., 2011|
|Combined models||Can combine any of the elements above (stage structure, evolutionary change, multiple species in a community)||Baskett et al., 2010; Riegl and Purkis, 2009|
can inform expectations about the extent of benefits and risks from interventions as they depend on location (e.g., Hock et al., 2017).
Population genetic models (see Figure 3.3f and g) can account for the risks and benefits associated with interventions that affect evolutionary dynamics. Such effects include the benefits of increasing stress-tolerant genotypes in assisted gene flow and managed selection, the benefits of genetic diversity and the combinations of different genotypes in managed breeding, or the risks of missed evolutionary opportunities in stress-reduction
interventions such as shading and mixing of cool water. Depending on the focal stressor targeted by intervention, the evolving trait(s) might be thermal tolerance, ocean acidification tolerance, or disease resistance. Evolution of such traits might occur in different components of the coral holobiont (e.g., symbionts: Baskett et al., 2009; coral host: Bay et al., 2017). The genetic representation can take many forms, from frequencies of individual genes (which can account for the role of drift relevant to the risks and benefits of managed breeding; see Figure 3.3g) to trait distributions (Day, 2005; Turelli and Barton, 1994; see Figure 3.3f).
Physiological models (see Figure 3.3h) such as dynamic energy budget models (e.g., Cunning et al., 2017; Muller et al., 2009) can account for the risks and benefits associated with interventions that affect coral energetics (e.g., antioxidants, nutritional supplementation). For example, a model that mechanistically connects nutrient levels (carbon, nitrogen, and phosphorus) to both coral and macroalgal dynamics can assess both enhanced macroalgal growth (with the associated risk of overgrowing corals as described in Fung et al., 2011; Mumby et al., 2007) and the benefit of increased energetic competence (and hence stress resilience) from nutrient supplementation to corals (Connolly et al., 2012). Physiological models can also separate the energetic contributions of different components of the holobiont (coral host, symbiotic algae, and the rest of the microbiome), which is particularly relevant to mechanistically representing interventions that target a particular component of the holobiont (e.g., algal symbiont or microbiome manipulation).
Structured population models (see Figure 3.3d and e) are models that follow population size distributed among ages, sizes, or stages such as juveniles and adults. They can account for the risks and benefits associated with interventions that affect the coral demographic processes of growth, reproduction, and mortality. Coral age and size can significantly influence demography and stress susceptibility (Connolly and Muko, 2003), which in turn determine overall dynamics (Hughes, 1984). A common demographic benefit to several interventions is the addition of recruits or small fragments (e.g., managed breeding, assisted gene flow, gamete and larval capture and seeding, managed relocation). Models that account for population structure (e.g., Baskett et al., 2010; Riegl and Purkis, 2009), where recruits can provide significant contributions to the population size, tend to find greater sensitivity to recruitment than models that follow proportion cover (e.g., Edmunds et al., 2014b; Fabina et al., 2015), where recruits add a small amount to overall cover. In addition to more accurately assessing the potential benefits of added recruits, structured population models can help evaluate stage- or size-dependent decisions (e.g., the
relative efficacy of moving gametes versus fragments in managed relocation). Structured population models can also be particularly relevant to interventions that affect ocean acidification levels (whether abiotic or biotic) given their ability to capture ocean-acidification-dependent growth and survival on overall population dynamics (e.g., Madin et al., 2012). As with genetic models, structured population models can range from following discrete age or stage classes (matrix models, e.g., Edmunds and Elahi, 2007; see Figure 3.3e) to following size or other continuous-state distributions (integral projection models, e.g., Madin et al., 2012; see Figure 3.3d).
Community models (see Figure 3.3c) can account for the risks and benefits associated with interventions that might affect taxa beyond corals, such as macroalgal dynamics (e.g., nutrient addition, macroalgal beds to reduce ocean acidification). A common modeling framework, explored as an example in Chapter 4, is to incorporate both coral and macroalgal dynamics with implicit herbivory (i.e., herbivory rate as a function of coral cover rather than following herbivores as one of the changing populations in time; e.g., Anthony et al., 2011; Mumby et al., 2007). Extensions might incorporate explicit herbivore dynamics (e.g., Blackwood et al., 2012) or sensitivity to different functional forms of the feedback between coral cover and grazing (e.g., McManus et al., 2018). Accounting for herbivore dynamics might be particularly relevant if management considerations include fisheries management in conjunction with coral interventions. Community models can also consider multiple coral species, which might be particularly relevant to interventions focused on community-level manipulation (e.g., managed relocation—assisted migration and introduction to new areas). Models with multiple coral species can elucidate the role of coral diversity in reef resilience (e.g., Baskett et al., 2014) and therefore the risks and benefits of focusing interventions on a single target coral species as compared to multiple species in a community. For interventions that affect disease susceptibility and prevalence (e.g., antibiotics, phage therapy, managed relocation), an additional component to consider modeling explicitly is disease dynamics, such as through susceptible-infected-recovered (SIR) models.
Figure 3.3 provides an overview of these modeling types. The models included here focus on different model types across ecological scales (physiological, genetic, population, and community) given that Table 1.1 and the committee’s first report (NASEM, 2019) organized the interventions by the ecological scale and process affected. The list of model options inevitably provides a subset of possible dynamics to consider, depending on the local context, and any modeling framework might incorporate more than one of the models. For example, biophysical models of larval
dispersal among locations (or other information, such as genetic distance, that can provide anticipated connectivity; Beger et al., 2014) in combination with genetic or structured population models that include recruitment dynamics (e.g., Condie et al., 2018; Kool et al., 2010) can reveal when connectivity-enhancing interventions (e.g., assisted gene flow, assisted migration) are likely to provide benefits beyond what one might expect from natural dispersal. In addition, individual-based simulations of any of the above dynamics (e.g., Mumby, 2006, for an individual-based simulation that has both structured population and community dynamics) can account for the role of demographic stochasticity. Furthermore, beyond biophysical considerations, models might account for coupled socioecological dynamics (e.g., Thampi et al., 2018). As with any modeling exercise, it is possible to continually add realism to the model structure, which will trade off with model generality, manageability, and increased parameter uncertainty (Levins, 1966; May, 2004). A focus on capturing the dynamics that are essential to the central risks and benefits of the set of interventions under consideration can help identify the simplest possible model relevant to the decision-making process.
Each modeling framework has unique data needs for parameterization that are dependent on the ecological, spatial, and temporal scales of the dynamics modeled. Across ecological scales, data needs will include selection strength and trait-based genetic variation for genetic models, nutrient uptake and assimilation rates in physiological models, demographic rates in structured population models, and species interaction rates in community models. One resource for such parameters is the Coral Traits Database (Madin et al., 2016). The temporal scales necessary to measure these biotic parameters will depend on the timescale of the associated process, from daily for physiological responses, to annually for demographic processes, to generationally for genetic processes. Across spatial scales, data needs might range from local-scale environmental heterogeneity in stressors (e.g., temperature, pH) to basin-scale processes such as larval connectivity; the relevant spatial scale and associated processes to consider depend on the scale of the intervention and associated risks and benefits. Given inevitable parameter uncertainty, sensitivity analysis is a crucial component of model analyses as detailed below.
Any biophysical model of coral reef health and potential response to interventions is an inevitably imperfect representation of reality. Following good model development practices using established guidelines will increase stakeholder and decision-maker confidence in model projections used to evaluate potential interventions and management alternatives (for details, see Crout et al., 2008; EPA, 2009; Jakeman et al., 2006; NRC, 2007, 2012). Best practices include the following (EPA, 2009):
- Proper definition of scope and objectives of the model: The scale and complexity of models are tailored to the management problem and objectives. Hypotheses underlying the model should be tested.
- Stakeholder participation in model development: Model developers, decision makers, and model users collaborate to specify the problem and objectives to inform and develop the associated model framework.
- Conceptualizing the system: Each element of the conceptual model is clearly described (including as functional expressions, diagrams, and graphs), and the utility and science behind each element are clearly documented.
These are largely addressed in Step 1 of the adaptive management cycle. Best practices for model development also call for an iterative verification-validation-sensitivity/uncertainty analysis to provide decision makers and stakeholders with confidence that the biophysical models are fit for purpose.
Verification is the process of determining how accurately the mathematical equations are coded, including code verification (e.g., determining whether the code correctly implements the intended algorithms) and solution verification (e.g., satisfactory reproduction of coral reef behavior or comparison to measured results such as hindcasting or predicting previously observed trends and measurements) (NRC, 2012).
Validation is the process of determining the degree to which a model is an accurate representation of coral reef interactions relevant to the intended uses of the model (NRC, 2012). Cross-validation increases the confidence of model projections by evaluating the consistency of model inputs and outputs by splitting the dataset and comparing one half against the other.
Sensitivity and uncertainty analyses Sensitivity analysis is the process of identifying how model projections change as a function of model inputs. Sensitivity analysis can be local, changing one input at a time by a specific amount and recording the change in model projections; or global, using a probabilistic or other type of framework to evaluate model sensitivity to all inputs simultaneously and thereby incorporating dependencies, feedbacks, and correlations (Cariboni et al., 2007). Uncertainty analysis, which is related but different, quantifies the uncertainties associated with model inputs. The goal of uncertainty analysis is to account for all sources of uncertainty and quantify the contributions of specific sources to the overall magnitude of the model output. The source of uncertainty can be
in both the model inputs and the model structure. If model projections are highly sensitive to uncertain inputs, it may warrant obtaining further information and data to reduce the uncertainty prior to making a decision. There are formal methods available for more rigorously evaluating how much additional data should be collected in terms of cost versus benefit, for example, using value of information (Comte and Pendleton, 2018; Conroy and Peterson, 2013). Alternatively, it may not be possible to reduce the uncertainty, which might suggest using a probabilistic framework such as Bayesian networks (described in Step 3), although note that Bayesian networks are also extremely useful for quantifiable uncertainty precisely because they are probabilistic.
The proposed interventions are largely untested at scale, as discussed in Chapter 2. This may mean that we do not know and have no way of knowing all of the possible near-term effects and further downstream effects of implementing an intervention. There are likely to be interactions across interventions (some of which may be unintended), and the presence and management of local stressors will also influence the probability of intervention success. In principle, high uncertainty can be quantitatively incorporated into most models, but this presumes that we know what we do not know (i.e., that we can quantify our uncertainty). If the uncertainty cannot be quantified, it is still possible to construct different scenarios or use probabilistic modeling approaches diagnostically to answer questions such as “how high or low would the probability of a particular intermediate outcome need to be in order to push the decision one way or the other?”
Expert elicitation is a formal process for using expert opinion to reduce or quantify uncertainty in model inputs (Hemming et al., 2017). For example, Ban et al. (2014) used expert elicitation to develop estimates about the interactive effects of multiple stressors on the Great Barrier Reef, effects for which there are limited data. There was stronger consensus about certain stressor effects (such as between temperature anomalies and bleaching) and weaker consensus on others (such as the relationship between water quality and coral cover).
Example sensitivity analysis Thampi et al. (2018) developed a mathematical model of coupled socioecological interactions, in which human systems and reef systems are both represented dynamically in order to model their influence on each other, to evaluate coral reef cover in a Caribbean coral reef system. These authors developed a series of parameter planes to evaluate model sensitivity under changes in two different model parameters as shown in Figure 3.4. The plane has a model parameter on each axis, and shows the dynamics related to coral health outcomes that occur for each possible pair of parameter values while holding all other parameters
at fixed baseline values. Figure 3.4a plots the strength of injunctive social norms (φ) against the parrotfish growth rate (s), while Figure 3.4b plots human sensitivity to coral reef rarity (J) and the maximal fishing rate (σ), again holding all other model inputs at baseline values. This is only one of many ways of conducting and presenting the results of a sensitivity analysis.
Step 3: Analyze Tradeoffs in Criteria Across Alternatives
Reef managers are likely to consider a range of management alternatives, including using one or more interventions in concert with conventional restoration activities as well as taking no action. These combinations, along with uncertainty in knowledge about the reef system and future environment, will yield a range of predicted changes across alternatives with tradeoffs in their ability to meet management objectives and minimize risk. For example, some intervention strategies may support the growth of a small subset of coral species that provide fish habitat but not the solid reef structure that is needed to provide coastal protection from storm waves. If fish habitat and strong reef structure are both key objectives for different stakeholder groups, then tradeoffs need to be made to reconcile different priorities or value preferences. Additionally, as the impacts of climate change progress, the capacity for management alternatives, including both conventional management and the new interventions, to meet some objectives (e.g., biodiversity or persistence of sensitive species or reefs) may diminish, and will hence drive a need to prioritize among values and places (Anthony, 2016).
It is important to introduce tradeoffs early in the structured decision-making process because they are closely tied to the setting of objectives and to value preferences. Early reconciliation of tradeoffs can help guide strategy generation, but they also need to be revisited late in the process once the performance of different strategy options is compared (Gregory et al., 2012; Hammond et al., 1999). All decisions involve evaluating and making tradeoffs and prioritization under uncertainty and resource limitations (Bottrill et al., 2008; McDonald-Madden et al., 2008; Wilson and Law, 2016). This step integrates the predicted changes in biophysical outcomes with other decision-making criteria to evaluate tradeoffs in risks and benefits across potential management activities and interventions. The approach can be as simple as mapping results (i.e., performance metrics) against different objectives (ecological, economic, and social) for each intervention strategy in a consequence table (Groves and Game, 2016; Hammond et al., 1999). This can help managers illustrate to stakeholders which intervention strategy performs better or worse against different objectives and can inform discussion around which objectives
to prioritize. Further complexity can be added through assignment of weights for different values affected by interventions, which can help managers and stakeholders arrive at a preferred intervention strategy. There are a number of different frameworks available for evaluating alternatives to assist in determining the optimal course of action (see Table 3.3).
Multi-Criteria Decision Analysis
Multi-criteria decision analysis (MCDA) supports complex decision making by explicitly incorporating multiple and potentially conflicting criteria that are valued differently by stakeholder groups and/or decision makers. Weighted criteria are used as the basis for evaluating tradeoffs across alternatives, which will differ across interventions. In the MCDA framework, evaluation criteria will include, at a minimum, the biophysical outcomes resulting from underlying models. Figure 3.5 provides a schematic of a hypothetical MCDA in the context of coral reef management. The potential interventions are shown in blue, and represent a subset of the interventions presented in Table 1.1. The example criteria, which in this case correspond to biophysical outcomes, are listed in orange in the center and are based on the recommended monitoring outcomes for coral reef resilience from Obura and Grimsditch (2009).
Predicted outcomes and criteria (such as coral diversity, herbivore biomass, disease) shown in orange can also differ for each intervention, and the specific mathematical relationship between criteria and interventions will differ and may not be linear. It is not necessary for all criteria to apply to all alternatives. In many cases, criteria will not be weighted
TABLE 3.3 Frameworks for Evaluating Management Tradeoffs
|Multi-criteria decision analysis (MCDA)||Alternatives analysis; specify alternatives, objective function(s), criteria||Brown et al., 2001|
|Decision trees||Probabilistic representation of outcomes; can backcalculate the optimum strategy||Flower et al., 2017; van Oppen et al., 2017|
|System dynamics models||Systems-based modeling approaches; typically deterministic but time varying; capture feedback loops||Chang et al., 2008; Rocha, 2010|
|Bayesian networks (BNs) or Bayesian belief networks (BBNs)||Models based on conditional probabilities; acyclic (e.g., no feedback loops)||Ban et al., 2014; Renken and Mumby, 2009; Chapter 4 of this report|
equally depending on stakeholder preferences. This combination of changes in biophysical outcomes across interventions that are mapped to decision criteria, the differential weighting of criteria, and additional decision criteria that may not be captured by the biophysical models results in a complex decision-making landscape (involving tradeoffs) that benefits from a structured approach. Objective functions define the optimum values across criteria. For example, coral cover can always be maximized while the optimum macroalgae value could be a threshold, and cost will likely always be minimized. Given all of the inputs, criteria, and weights, MCDA will generate a ranking of alternatives.
There are many different ways to structure the analysis, and the appropriate operational format will emerge out of the adaptive management process. For example, the decision about which intervention to utilize is far more nuanced than simply considering each intervention in isolation as represented in Figure 3.5. It is far more likely that there will be a series of interventions at different times and that these will be combined with management of local stressors. That said, in an iterative, adaptive management context, first evaluating each of these interventions by biophysical, economic, and social criteria using MCDA can lead to an understanding of relative differences across interventions that could be useful for subsequent analyses. For example, it may be revealed that, based on the understanding of their risks and benefits, certain interventions are either “dominant” (will always be preferred) or “dominated” (will never be preferred). The actual alternatives analysis will more likely be structured around particular scenarios, such as conventional stressor management in conjunction with differently timed intervention activities. Because there are many different ways to structure the problem, the site-specific context requires stakeholder and participatory modeling to establish objectives, goals, and criteria in an iterative way.
Evaluating sensitivity and uncertainty, as described for the biophysical models, informs tradeoff analyses as well. Depending on the specific software program that is used, it is often possible to evaluate cutoffs where one alternative is preferred over another, and how sensitive the ranking is to specific assumptions. For example, if coral cover is given a higher weight by stakeholders and decision makers as a key biophysical outcome, and the sensitivity analysis reveals that a small change in weighting of that outcome leads to a different ranking of interventions or alternatives, then that is useful information to have and will focus the communication and discussion of results across stakeholders.
Brown et al. (2001) presented an approach for protected area management at the Buccoo Reef Marine Park (BRMP) in Tobago that incorporates multiple objectives within a decision-making framework. The authors relied on MCDA and used the process to explore the impacts of four different development scenarios. The decision context was characterized by many different users in apparent conflict and by linkages and feedbacks between different aspects of the ecosystem and the economy. Diverse stakeholders were asked to weight different criteria, which were then incorporated into a tradeoff analysis to explore different management options.
Decision trees represent another way to structure a decision. In this case, rather than evaluate tradeoffs across interventions on the basis of biophysical outcomes and criteria, a set of scenarios is developed and probabilities are assigned to specific outcomes associated with each intervention. It is then possible to identify the optimal “path.” Some decision trees are qualitative and can simply take the form of flow charts without assigned probabilities and are useful for structuring the decision context and developing conceptual models.
Example Decision Trees
Flower et al. (2017) developed a general guide and decision tree for assisting coral reef managers in understanding the ecological implications of monitoring data that could inform a management response. They developed a guide for interpreting the temporal trends of 41 coral reef monitoring attributes as recorded and published by seven of the largest reef monitoring programs. As part of this guide, they proposed several decision trees to use in evaluating monitoring results. The tree is used to potentially distinguish across causal stressors given site-specific observations.
Van Oppen et al. (2017) proposed a decision tree for determining whether to incorporate assisted evolution (a term used for a variety of interventions) into restoration initiatives as part of a management strategy to enhance climate resilience of coral reefs (see Figure 3.6). In this decision tree, a more structured decision methodology is nested under “Risk/benefit analysis,” including ecological modeling of ecosystem strategy impacts (biophysical modeling as described in this document) and socioeconomic modeling/decision support.
System Dynamics Models
System dynamics models represent causal interactions within a system or population as hypothesized mathematical relationships that represent system behaviors and interactions over time. These models capture feedback structures that generate observed behaviors and effectively integrate natural system dynamics with social constructs for policy analysis and decision making. These kinds of models often use “stocks and flows” to represent key parameters and their relationships. Stocks and flows, as a specific kind of system dynamics diagram, could represent healthy versus bleached populations, or healthy versus susceptible populations where different kinds of susceptibilities could be represented depending on the specific drivers present in the system.
Example System Dynamics Model
Bartelet and Fletcher (2017) present a systems-based model developed to explore two different hypotheses about the spread of coral viruses in the Caribbean. They developed two simulation models based on the competing hypotheses about the origins and diffusion dynamics of a coral reef virus to make inferences about possible future behaviors. Figure 3.7 presents one version of the disease diffusion model based on a hypothesis that the virus originates from outside the coral colony (e.g., the result of the disposal of untreated human sewage in surrounding waters). It depicts a model diagram in terms of stocks and flows to quantify mechanisms surrounding a coral disease outbreak in the Caribbean. In this case, stocks (boxes) represent coral populations, and the flows (arrows) represent the relationship between different kinds of populations and infection and recovery rates. The four stocks in the model in Figure 3.7 include the susceptible coral population, the infectious coral population, the recovered coral population, and the white skeleton population (e.g., where the live coral tissue has died). These models provide a structure on which to prioritize research, test assumptions and hypotheses, and evaluate implications of policy and management decisions.
Bayesian networks offer another alternative for structuring decision making and are the focus of a more complete example in Chapter 4. Key activities and hypothesized relationships are identified by nodes and connecting lines in a graphical format. These graphical models represent probabilistic and conditional dependencies among nodes and frequently, though not necessarily, imply causation. The nodes in a Bayesian network define relationships and provide the ability to conduct inference
on the relationships and dependencies across nodes. Bayesian networks are also known as Bayesian belief networks, the latter term used most prominently in cases when conditional probabilities in the network are informed by expert elicitation.
Bayesian network design may be based on a conceptual model for the specific application and can incorporate results from the underlying biophysical modeling to generate conditional probability tables to quantify the relationships across nodes. Environmental data or information on drivers and pressures can relate probabilistically to biological and ecological variables that in turn relate probabilistically to ecosystem values. Bayesian models are particularly useful for application in adaptive ecosystem management programs in which data may be limited, risks and benefits are poorly understood, and biophysical modeling inputs are poorly constrained and highly uncertain, but when there are opportunities for learning and adjustment (Chen and Pollino, 2012; Nicol and Chadès, 2017; Nyberg et al., 2006) and potentially when expert opinion can complement data (Martin et al., 2005).
Example Bayesian Networks
A number of examples of the use of Bayesian networks for coral management can be found in the literature. Carriger et al. (2019) demonstrated the process and utility of a Bayesian network for assessing delivery of ecosystem services in a hypothetical coral reef conservation situation. Franco et al. (2016) used coral reef carbonate balance (or rate of change) as the output node in construction of the Carbonate Budget BBN (CARBNET). The structure of the network was initially developed based on literature review, and then modified based on expert consultation. It includes multiple levels of parent–child node relationships from the upstream climate change and anthropogenic disturbance nodes (e.g., coastal development, atmospheric carbon dioxide), to the associated pressures on ecosystems (e.g., sediment load, sea surface temperature), to the direct effects (e.g., turbidity, coral bleaching), to the presence of bioerosive (e.g., sea urchins) or bioconstructor taxa (e.g., coral carbonate production), and finally the calcium carbonate budget outcome. The relationships among nodes, as conditional probability tables, were parameterized based on a combination of original data collection and literature review, and then the relative influence of these were evaluated with a sensitivity analysis. In another example, Ban et al. (2014) used a Bayesian network to structure their expert elicitation process to estimate the effects of the interaction of multiple stressors and related management options when data about the effects of these interactions were incomplete. Figure 3.8 provides an overview of this model and the general structure of Bayesian networks. The Capturing
Coral Reef & Related Ecosystem Services (CCRES) project has developed a publicly available Bayesian network called ReefReact1 to predict the probability of coral reef cover for several selected years across different climate scenarios. The model, originally developed for Indonesian coral reefs, includes nodes for cyclone occurrence, disease outbreak, sediment exposure, grazing rate, and algal growth rate. Coral cover is predicted for several different years.
Step 4: Select Interventions or Combination of Management Activities and Determine Evaluation Metrics
The performance of different intervention strategies against objectives in the short and/or long term will inform strategy choice. Given the new quantitative information, benefits (environmental, ecological, social, and potentially cultural), risks, costs, and tradeoffs should be reconsidered before a preferred strategy option is chosen. Once the intervention, set of interventions, or combination of management actions has been selected, it is important to establish a set of evaluation metrics for monitoring that link to the decision criteria developed based on management objectives. For example, a decision criterion might be to increase coral cover and diversity to create fish habitat (Hoey and Bellwood, 2011; Williams and Polunin, 2001). The associated evaluation metrics might then be abundance and diversity of coral and fish species measured at appropriate time intervals. Measurable evaluation metrics that relate to the biophysical models will form the basis of the monitoring program discussed in the next step. Various biological community, disturbance, ecological process, and site characteristic metrics have been proposed in the context of evaluating coral reef health and resilience (e.g., Ford et al., 2018; Lam et al., 2017; McClanahan et al., 2012; Obura and Grimsditch, 2009).
Molecular tools across the “omics”—including genomics, transcriptomics, proteomics, and metabolomics—provide important metrics of intraspecific biodiversity, short-term responses to individual and suites of stressors, gene expression under changing conditions, and metabolic homeostasis. Genetic variation is a key component of resilience of populations and determines the ability of organisms to respond to a variety of environmental parameters and stressors (Levin and Lubchenco, 2008). Proteins can mediate responses to stressors including elevated temperatures, reduced or elevated oxygen levels, and toxicant exposure. Metabolomics is the study of metabolite production, which is another indicator of an organism’s state (Lohr et al., 2019; Quinn et al., 2016; Vohsen et al., 2019). Together, these molecular techniques can provide diagnostic tools
that might be applied to evaluate the success and effectiveness of management actions and interventions. As such, they could be important tools for helping guide the allocation of limited financial, human, and institutional resources toward protecting corals and coral reefs.
In addition to the “omics,” a number of physiological and ecological metrics provide insight into coral resilience. Physiological measures include growth rates, calcification, fecundity, productivity, constituent composition (e.g., lipids), algal symbiont density, and colony morphology (Edmunds and Gates, 2002; Putnam and Edmunds, 2011). Ecological attributes include survivorship, size–age distributions and coral population demographics, percentage of coral cover, coral diversity, rugosity, and recruitment patterns (Hughes and Connell, 1999). Survivorship, growth, and recruitment in combination provide insight into fitness (Reusch, 2013), and hence the likelihood that coral species or populations can be sustained under environmental change and proposed intervention strategies. Beyond corals, associated ecosystem properties such as herbivore biomass and diversity can provide insight into ecosystem function and health (McClanahan et al., 2012). The physiological and molecular techniques are of particular value due to their ability to detect changes in real time, hours, days, and weeks versus many ecological indicators that are responsive over months, years, and decades (Aswani et al., 2015).
Steps 5 and 6: Implement Interventions, and Initiate and Sustain a Monitoring Plan
Targeted monitoring is a critical component of an adaptive management plan in order to feed back to assessment and adjustment of management strategies. Monitoring provides data on predicted biophysical metrics, both to compare to model predictions over time and thereby improve the underlying biophysical modeling, but also to demonstrate and evaluate how well objectives are being achieved. Effective monitoring programs are based on the management objectives identified in Step 1 (Legg and Nagy, 2006) and the decision criteria and evaluation metrics identified in Step 4 based on these management objectives. Some of these monitoring targets, such as the “omics” and physiological metrics, can be monitored for change within short timescales. However, long-term monitoring is required to measure changes in community- and ecosystem-level metrics such as species diversity, persistence of key reef functional groups, and resilience (including long-term recovery) to disturbance (SER, 2004). There will be a progression of metrics from near-term survival, to reproduction, to long-term persistence and resilience to monitor over time (Seddon, 1999).
Monitoring should also inform the identification of causes of strategy failure. Monitoring local stressors such as water quality or human activities could improve understanding of intervention outcomes as well as identify times when conventional management strategies should be reassessed. Monitoring the causes of failure would also involve monitoring for potential risks, such as the introduction of pathogens or nonnative species.
Steps 7, 8, and 9: Evaluate, Communicate, and Adapt
As interventions are applied and monitoring data come in, the final three steps of the adaptive management circle involve evaluating progress toward decision objectives, communicating the results of the monitoring program, and potentially adapting or revising the management strategy. The monitoring data, especially when compared to model expectations, may reveal issues that were previously unknown, or may suggest localized dynamics not captured by the original decision model (e.g., the presence or absence of particular fish species and their relevance to coral health). If unintended consequences are observed, then the underlying biophysical modeling should be updated to reflect the evolving understanding of coral reef dynamics in the context of management activities. A passive adaptive management approach consists of trying one model or approach at a time. Compared to this, an active adaptive management approach, in which multiple models or approaches are implemented in an experimental manner, enhances learning and the long-term outcome. Active adaptive management is more likely to be optimal compared to passive adaptive management under higher uncertainty, objectives that encapsulate higher risk tolerance, faster rates and broader applicability of learning, and lower costs of monitoring (McCarthy and Possingham, 2007).
A key aspect of the adaptive management approach is the ability to obtain initial data for intermediate outcomes in order to modify the management intervention accordingly. For example, a stated goal might be a 20% increase in coral cover over a period of time. There will be intermediate outcomes on the path to achieving this overall goal and a formalized process for monitoring improvement will help to inform the ongoing efficacy of management alternatives. These intermediate outcomes can be evaluated in different ways; one approach is formalized through “results chains,” a tool for helping teams clearly specify their theory of change behind the actions they are implementing (Margoluis et al., 2013).
Structured decision processes and tools exist that will help coral reef managers develop objectives and implement new and potentially risky intervention strategies. The chapter identifies a number of cases when these tools have been used to guide coral reef management. Although they have largely been utilized in the context of managing local stressors, they are easily adapted to evaluating tradeoffs across other kinds of management interventions.
Conclusion: Although many tools exist for structured decision making to evaluate interventions as part of a reef management strategy, there is no single generalizable approach and no substitute for working through a structured decision process with stakeholders in the local context. This effort provides a data- and values-informed basis for selecting and evaluating management options against a set of objectives.
Recommendation: A structured, adaptive management framework that considers all drivers and pressures affecting coral reefs should be developed to evaluate tradeoffs across alternatives and identify when and where new coral intervention(s) will be beneficial or necessary. This framework should include
- Engagement of a broad set of stakeholders to establish objectives and courses of action that reflect community values.
- Development of models tailored to the local environmental and ecological setting, management objectives, and preferred intervention options.
- Targeted monitoring of short- and long-term metrics of reef health and resilience.
- Iterative evaluation and adjustment of management strategies.