Options for integrating ecological, economic, and social objectives in evaluation and management of fisheries

Overview

Descriptive multivariate models are methods that measure the status of a community, food web, or ecosystem using commonly available data (Link et al., 2002). Example data and metrics include total yield in species groups (groundfish, pelagics), size-spectra, diversity indices, and species richness. In ecosystem modeling, multivariate methods are used to identify patterns in data that can provide insight into key processes and the potential impacts of fishing on marine communities and ecosystems (Link et al., 2002; Rice, 2000). RAPFISH, or “rapid assessment of fisheries,” is one approach developed to assess fisheries sustainability in terms of social, ethical, economic, and ecological aspects (Pitcher & Preikshot, 2001).

Evaluation

RAPFISH and other multivariate methods score high in their ability to integrate diverse types of information, but they rank poorly against all other criteria for applied decision support (Table 2, Figure 3). This is largely due to the fact that these methods ignore uncertainty, have unknown reliability, and are not related to ongoing management decisions (Rice, 2000). They are not used in tactical management; their primary value is to provide a holistic understanding of the nature of fisheries problems and to bound more complex multispecies methods (Kerr & Ryder, 1989).

Overview

Dynamic multispecies models address problems of interacting species or fisheries that are not typically included in single-species models. They address two distinct problems: biological interactions in the form of predation and competition (trophic dynamics), and technical interactions, which address issues of joint capture of species by multiple fishing units in time and space (Murawski, 1984; Pope, 1979). While it is recognized that both types of interaction can, and likely do, occur simultaneously in many fisheries, multispecies research has focused primarily on understanding the implications of biological interactions on the productivity of target species (Magnijsson, 1995; Murawski, 1984).

Models that address biological interactions are dynamic extensions to single-species methods that typically allow for predation mortality and prey-related growth (Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000). Multispecies virtual population analysis (MSVPA) is perhaps the best-known multispecies approach. It has been used to study predation in marine fish assemblages and to refine estimates of predation-induced natural mortality (for detailed examples, the reader is referred to the ICES multispecies studies reviewed by Pope (1991) and Magnijsson (1995)). MSVPA has led to at least two important insights about both the biology and management of fish communities (Magnijsson, 1995). First, that natural mortality is higher than assumed in single-species assessments, and is highly variable from year to year. Second, when evaluated in a multispecies context, changes in management tactics can have counter-intuitive effects on fish populations. For example, single-species methods (e.g., yield-per-recruit analyses) indicate that increases in mesh size decrease fishing impacts on young fish, but multispecies analysis shows increased predation on younger fish as a result of an increased abundance of predators (Magnijsson, 1995).

Technical interactions encapsulate institutional, operational, and economic aspects of multistock fisheries (Hilborn, 1976), stocks that are exploited by more than one fishing unit (Pope, 1979), by-catch or incidental mortality (Murawski, 1991), and competitive interactions within and between fisheries that affect catch and economic performance (Ulrich et al., 2001). Technical interactions models have traditionally provided results based on equilibrium analyses, which has greatly restricted their application to real-world problems (Magnijsson, 1995; Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000). A further limitation is that much of the research has been directed toward identifying “optimal” management solutions in a multispecies/multifleet context. These models are best considered to be tools that can be used to evaluate the trade-offs between components of the fishery system (e.g., stocks, species, and fleets) given clear management objectives (Hilborn, 1976; Murawski, 1984).

Recent applications of multispecies methods are directed at incorporating a range of biological interactions in stock assessment and management. For example, Hollowed, Bax et al. (2000) and Hollowed, Ianelli et al. (2000) evaluated mortality by top predators such as arrowtooth flounder (Atheresthes stomias) and Steller sea lion (Eumetopias jubatus) in the stock assessment model for Gulf of Alaska pollock (Theragra chalcogramma) and found that estimates of natural mortality were both higher and more uncertain than is commonly assumed in standard single-species methods. Failing to account for this can lead to biased estimates of stock status and overstated confidence in management advice. In addition to these impacts on the quality of management advice, predator–prey interactions also affect the robustness of management targets and reference points. In particular, biological reference points for forage species appear to be moving targets that depend on the level of predation mortality (Collie & Gislason, 2001).

In spite of such insights, these methods are not widely used in management because they are subject to high statistical and structural uncertainty (i.e., they capture only a small part of the true dynamics of fish communities), and because they provide advice for which there is relatively little understanding or demand in the current management systems (Collie & Gislason, 2001; Magnijsson, 1995). European Union fisheries management may be a possible exception. For example, the MYFISH Project explored social, economic, ecological, and institutional trade-offs with the aim of informing multispecies management (myfishproject.eu; Rindorf et al., 2017). In general, however, the management value of these biological models appears to be limited to providing general predictions of average patterns and relationships (i.e., reference points for prey species should be conditioned on the level of predation mortality), improving biological realism in depictions of natural mortality and growth (Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000; Magnijsson, 1995), and “ground-truthing” inputs for simpler, single-species models (Rice et al., 1991).

Evaluation

Both classes of multispecies model described above rank poorly in terms of the criteria for holistic decision support (Table 2, Figure 3). Their limited management application likely reflects a lack of stakeholder participation at the problem structuring stage, leading to both a low level of understanding by non-ecologists and low level of integration into the current management context. Additionally, while multispecies methods aim to broaden the ecological perspective of fish populations, they do not generally integrate social and economic aspects of fisheries. However, it is possible to develop models that consider these holistic aspects of fishing. Punt et al. (2010) developed a dynamic bio-economic assessment model that accounts for the biological and economic aspects of capturing three prawn species in Australia's northern prawn fishery. The authors noted that the model was unlikely to be used to set real-world management targets, in part due to the fact that the theories underlying social and economic concepts are difficult to apply to actual fisheries, and introduce significant additional uncertainty into the management problem (Dichmont et al., 2010). Managing for a broader set of criteria (such as economic yield) requires agreement of all stakeholders on how the management targets will be defined, estimated, and applied (Dichmont et al., 2010). Thus, in the absence of high levels of stakeholder participation, multispecies methods are unlikely to be broadly used in management.

Overview

Aggregate species (“production”) models represent the relationship between fishing effort and yield for aggregate species groups and fish communities (Gaichas et al., 2012; Pope et al., 2006). They differ from multispecies models in that they do not represent the dynamics of individual species, but represent the aggregate production of a community or ecosystem (Fogarty, 2014). These models are used to estimate key management parameters such as the system-level equilibrium yield (MSY) and therefore provide a practical approach for implementing ecosystem-based management principles (Fogarty, Overholtz, & Link, 2012). This approach operates from the perspective that it is not necessary (or possible) to fully understand ecosystem structure and function before implementing ecosystem-based management (Fogarty, 2014).

At present, aggregate production methods are not commonly used in management, but they might be useful in data-poor fisheries. At a minimum, they are considered a viable alternative to methods that attempt to accurately represent potentially dozens of interacting species (Hilborn & Walters, 1992). The key advantage of aggregation is that it draws on the fact that total fish yield, size-structure, and biomass tend to be highly conservative properties of ecosystems and fish communities (Kerr & Ryder, 1989). It is therefore possible to develop ecosystem-level management protocols to ensure that system-level stability, variability, and resilience are maintained. Approaches to assembling species complexes in these models can depend on management objectives, and include grouping according to taxonomic affinity, habitat preference (e.g., demersal species), functional group, size-class, and simply combining across all species. However, the most robust aggregation rules (in terms of ability to balance conservation and economic outcomes) combine species with similar productivity, species interactions, and sensitivity to environmental forcing (Gaichas et al., 2012).

Evaluation

Similar to the multivariate and multispecies models described above, aggregate species models were developed primarily by biologists to address the ecological needs of ecosystem-based fisheries management (EBFM). Thus, while they can integrate a variety of biological data, they are not designed to integrate social and economic data, or to be used as tools for stakeholder engagement (Table 2, Figure 3). They are commonly used as strategic, direction-setting methods (e.g., World Bank 2009). However, relative to the other multispecies and ecosystem methods, the chief advantage of aggregate species models is that they can be used to provide tactical management advice that is robust, cost-effective, and which could, with some modification, be used in the current management system.

Overview

Ecosystem models are mathematical representations of ecological systems that can be used to screen policy options under a broader set of ecological criteria than is typically considered in single-species management (Walters & Martell, 2004). They can predict the dynamics of exploited marine ecosystems, identify unintended consequences of management decisions, and highlight trade-offs between and among ecological objectives (Essington & Plaganyi, 2013). The ecological modeling literature is expansive and includes debate on the “optimal” level of model complexity, integration of ecological theories with approaches to model development (e.g., top-down versus bottom-up), trade-offs between biological accuracy and utility of the models, and data availability (and lack thereof), to name only a small subset of topics and issues (Allen & Fulton, 2010; Christensen & Walters, 2004; Essington, 2004). Ecosystem modelers recognize the need for cross-disciplinary engagement in model development (e.g., Allen & Fulton, 2010); however, there is relatively little discussion of engaging stakeholders and managers in the process. Some have suggested that multispecies and ecological models preclude collaboration because they are costly to develop (in terms of time and data) and because the models and results are overly complex and counter-intuitive (Table 2; Magnijsson, 1995).

Several classes of ecosystem model are identified in the literature (e.g., Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000; Plagányi, 2007), but three have captured the broad interest of fisheries scientists and to a lesser extent, managers. Aggregate or whole ecosystem models account for all trophic levels in an ecosystem, and describe feeding interactions and energetic pathways within ecological communities. This category includes the Ecopath with Ecosim (EwE) suite of models, which is the most widely used approach to date (Walters, Christensen, & Pauly, 1997). Dynamic system models involve the most detailed representations of ecosystem structure and can represent both top-down (i.e., predation and fishing) and bottom-up (prey and primary production) effects on key ecological processes and interactions. Models in this class typically depict biogeochemical processes, coupled oceanographic-ecological models, and detailed tropho-dynamic relationships (Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000). Atlantis is perhaps the most detailed and best-recognized dynamic system (Fulton et al., 2011). It is unique among ecosystem models because it was developed to address practical problems related to the evaluation of ecological indicators and tools for EBFM, it includes models of how decisions are actually made in a given fishery (Plagányi, 2007), and it has the capacity to predict stakeholder behavior in evaluations of the impact of changing regulations and policies (Fulton et al., 2010). It is intended for whole-of-system management strategy evaluations (see below). However, in spite of its broad scope and integrative approach, Atlantis is not recommended for tactical management (Fulton, 2010). Hollowed, Bax et al. (2000) and Hollowed, Ianelli et al. (2000) suggest that complex ecological models such as Ecosim and Atlantis are not an improvement over single-species models for providing advice on the impacts of fishing, in large part due to a lack of the data required to parameterize these complex models, as well as a poor understanding of the key processes and interactions that govern ecosystem dynamics. As a result, ecosystem model predictions are highly uncertain and of limited use for tactical fisheries management (Essington, 2004).

Minimum realistic models (MRM) and models of intermediate complexity for ecosystem assessments (MICE) reduce the structure of ecosystem models to the minimum required to inform management (Butterworth & Punt, 1999; FAO 2008; Plagányi, 2007; Plagányi et al., 2014). By accounting for a restricted set of biological interactions in a specific ecosystem, MRM and MICE address a key limitation of ecosystem models—namely that scientific uncertainty around model predictions increases with model complexity (Plagányi & Butterworth, 2004). These “minimal” models explicitly account for uncertainty and their performance can be evaluated using standard statistical diagnostic tools. However, only MICE include ecological, environmental, and human components of the system and are considered to be comprehensive tools for ecosystem assessment (Plagányi et al., 2014). In contrast, MRM are essentially dynamic multispecies models that have the potential to be used for tactical decision making, but are seldom used for that purpose (Plagányi, 2007; Plagányi et al., 2014). The “human” component of MICE includes behavior and decision making in the form of fleet dynamics, location choice, and investment at the scale of a fishery; however, social and economic assessments have not been included in any tactical application to date (Plagányi et al., 2014). Thus, with respect to holistic evaluation, MRM and MICE have not yet overcome the problems inherent in full ecosystem models (Table 2). However, given their focus on tractability and robustness they are likely to be valuable tools for this purpose in the near future.

Evaluation

All classes of ecosystem models explored in this study rank poorly in terms of their utility for holistic decision support (Table 2, Figure 3). However, it is important to note that full ecosystem models were never intended to compete with or replace single-species assessment and tactical advice (Christensen & Walters, 2004; Hollowed, Bax et al., 2000; Hollowed, Ianelli et al., 2000). They are superior to single-species methods for providing strategic analysis and insight related to ecosystem function, impacts of harvesting on food webs, long-term cumulative impacts of policies and natural variability, and for highlighting gross inconsistencies in fisheries management objectives and expectations (Table 2; Christensen & Walters, 2004; Plagányi & Butterworth, 2004; Fulton, 2010). A significant challenge to their use in management is that the current system does not typically demand the type of strategic advice that could be provided by ecosystem models.

Overview

Bayesian belief networks (BBNs) graphically depict causal relationships among key components of a management system (Cain, 2001). They represent uncertainty about both the natural resource and its response to management intervention by probabilistically representing relationships between system variables (McCann, Marcot, & Ellis, 2006). They are structurally similar to influence diagrams, which are tools for understanding links between objectives, alternatives, and consequences (Gregory et al., 2012). The key difference between the two is that influence diagrams are used in the context of optimization and providing the “best” solution, whereas BBNs are intended to promote shared understanding of a management problem (Cain, 2001). Influence diagrams are tools for decision making, while BBNs are tools for decision support.

The applications of BBNs in resource management are relatively few, but include land-use planning, water quality management, fisheries management, and water use management (Duespohl, Frank, & Doell, 2012; McCann et al., 2006). They have widest application in integrated water management and planning, which is characterized by many of the same issues as EBFM: the existence of multiple objectives that span a variety of environmental, economic, social, and political activities and scales, the need to adopt a system-level approach to management, and a requirement for stakeholder participation to ensure implementation and compliance (Cain, 2001).

The utility of BBNs as a tool for facilitation of stakeholder participation and integration of diverse knowledge is emphasized in the literature (Table 2, Figure 3; Cain, Batchelor, & Waughray, 1999; Duespohl et al., 2012). In addition to their graphical construction, which clearly shows the assumed relationships among system variables, their use of Bayesian statistics means that BBNs can synthesize both empirical data (including missing observations) and expert knowledge in representations of modeled system and alternative beliefs about how impacts and interventions propagate through that system (Kuikka et al., 1999; Rahikainen et al., 2014). They are additionally highly interactive and can be developed quickly using a number of computationally efficient commercial modeling software packages (e.g., WinBUGS; Lunn, Thomas, Best, & Spiegelhalter, 2000). The capacity for rapid, cost-effective development of alternative models is perhaps their chief advantage over the other modeling approaches we evaluated.

Outcomes of BBN models are measured in terms of utility for system variables that represent diverse preferences, risk attitudes, and/or monetary values. Utilities are quantitative values that can be used to illustrate trade-offs between competing objectives, allow alternative management options to be compared, and increase transparency in the management process. For example, Levontin, Kulmala, Haapasaari, and Kuikka (2011) developed a BBN of the Baltic salmon management system to evaluate performance of alternative management options in terms of their impacts on a diverse set of variables that included social capital, salmon stock structure and conservation, commercial catch, and recreational in-river fishing opportunities. Utilities for each variable were normalized such that values close to 1 represented the most desirable state. Management options were subsequently ranked according to each component utility and for the total utility score, given different weighting for each variable (e.g., showing preference for conservation over economic outcomes). BBNs used in this manner can increase transparency about management preferences and improve collective understanding of the management process.

BBNs have some important caveats that limit their use in management (Table 2). First is that construction requires specification of probabilities for all system variables and interactions (Olson, Wagner, & Willers, 1990). This can be difficult to obtain, and potentially misleading, if based on expert judgment that is not obtained using structured approaches (Marcot et al., 2001). The problem of model parameterization may be particularly challenging for large numbers of system variables (McCann et al., 2006). The reliability of BBNs is an additional challenge, because quantitative and probabilistic understanding of causal relationships within social–ecological systems is generally poor, regardless of whether it is based on empirical studies or expert knowledge (Duespohl et al., 2012). Furthermore, use of discrete variables to represent system states and impacts (e.g., low, medium, high) leads to low precision of results and outcomes that are vague and difficult to interpret. Continuous variables can be used to improve precision, but only for well-studied relationships (Borsuk, Schweizer, & Reichert, 2012). Another significant limitation is that BBN do not represent temporal dynamics, feedbacks, or spatial variability (McCann et al., 2006). As a result, BBNs cannot be used to address highly dynamic management problems that are often encountered in resource management (Castelletti & Soncini-Sessa, 2007). Their static nature means that BBNs are not well suited for tactical decision making and forecasting the implications of alternative management actions. They are best used for strategic decision support and planning.

Evaluation

BBN have a demonstrated ability to integrate social, economic, and ecological objectives and to provide cost-effective, strategic information to decision makers. In addition, this method could be used in the current management system, particularly at the strategic planning stage, which requires a comprehensive picture of the objectives, trade-offs, and uncertainties to be addressed in the planning process. They therefore score highly against criteria for generation and use of information (Table 2). While relatively easily understood, this method is limited in terms of providing scientifically robust advice on the likely impacts of management interventions. We represent this with a low score against criteria for transmission of information (Table 2).