Project Design and Scoping is the very beginning stage of project formulation before hypotheses, objectives, and the full scope of research are defined. Developing a coupled model at this stage implies integration of both natural and social scientists in designing the project scope. This multi-disciplinary collaboration is necessary to create a fully Two-way coupled model of the SES that can meaningfully answer complex ecological and social questions that do not impact exclusively either the ecological or social system. The purpose of coupling at this stage is to better identify questions, objectives, and performance metrics of interest to the management and stakeholder communities. Only when social and natural scientists along with managers and resource users work together from the project design and scoping phase can truly integrated two-way coupled models be developed. An example of such a Two-way coupled model is an end-to-end model (Box 1). The development of integrated and coupled social and natural systems models can often span multiple years. It is very infrequent that two existing models can be fully integrated to produce Two-way coupled end-to-end models which address the relevant management priorities. Hence, this long timeframe to co-develop research questions and models should be reflected explicitly in funding priorities and approaches to enable this type of integrated research². The utility of end-to-end models in a management context is the ability to quantify the impact(s) of dynamic system processes on the distribution and abundance of living marine resources and the overarching impact on long- term ecosystem resilience [e.g., the role of interspecies interactions on developing sustainable, harvest policies (Masi et al., 2018), the frequency of stock assessment updates (Hutniczak et al., 2019), and how ecosystem resilience impacts (conflicting) human use of the ecosystem (e.g., extractive vs. non-extractive use) (Weijerman et al., 2016)].

Other examples of a two-way coupling modeling approach that can be particularly effective early in the project life cycle are conceptual and causal models (see Box 2). Conceptual and causal models are usually used to make sense of relationships and linkages within a system; these linkages are often developed in consultation with stakeholders to facilitate stakeholder participation and integrate diverse sources of knowledge of the system (Düspohl et al., 2012). This collaboration allows for a common understanding of all system aspects and gives individuals who are impacted by resource management decisions an opportunity to include relationships that are important to them. Furthermore, tapping into the collective knowledge of a large and diverse group of resource users can lead to a robust understanding of the SES and overcome scientific data gaps with regard to linkage (Aminpour et al., 2020). Thus, when projects are designed from the outset with important ecological and societal issues determining the outcomes of interest, these models can be fully coupled and address the entire SES, in addition to impacts on the ecology and outcomes on the fishery and society and vice versa.

The Model Development phase typically occurs once a team has been established focused on a general research topic and a model is developed for a funding proposal. More often than not, coupling at this stage occurs in a limited two-way fashion between a very complicated social and/or economic model and a relatively simplistic representation of the ecology (see Box 3 on Bioeconomic models) or a similarly complicated ecological and/or biological model with a simple representation of the human system [see Box 4 on Management Strategy Evaluations (MSEs)].

The intent of coupling the ecological component of the model with a social component during the model development phase of a project is to assist the team in identifying the best modeling approaches to answer the set of already defined questions. In this phase, two-way coupling of the models is possible, but the questions that the coupled model will be able to answer will necessarily be a subset of those the entire interdisciplinary research team would have developed had they been involved from the start. However, this may be the best approach to take in some circumstances with well-defined questions that are somewhat limited in scope and a complete picture of the impact on the full SES is unnecessary or not feasible given regulatory or time constraints. The Bio-economic Length Age Structured Tool or “BLAST” Model (Lee et al., 2017) is one such example. It is a bioeconomic model which combines a utility-theory consistent model of recreational fishing demand and an age-structured stock dynamics model to help provide harvest advice to fisheries managers, initially developed in the Northeast United States.

Often, the way social and economic coupling occurs in this stage in an effort at implementing EBFM policies are through models of fisher behavior to better understand the economic and social costs and benefits of specific management rules and regulations (Abbott and Haynie, 2012; Abbott et al., 2015; Reimer et al., 2017; see Box 5). These models can either be created prospectively (as a fishery management council, regional planning body, or other management body is considering the impact of multiple alternatives) or through a retrospective analysis of the economic and social impacts of a management or environmental change.

The class of models that are coupled at the model development stage, whether One-way coupled or limited Two-way coupled, tend to have moderate or moderate-high management traction, which is intuitive because these are often the type of models that are created to address specific management problems. As a result, these models tend to focus on modeling fishing fleet dynamics (Branch et al., 2006; Watson and Haynie, 2018) and the impact of regulations, such as spatial or temporal closures (Abbott and Haynie, 2012; Reimer and Haynie, 2018), on the fishing industry and on society at large (Sanchirico et al., 2013).

This is the phase of a project where the modeling team has a model developed to explain some real world phenomenon and is trying to assess the degree to which their model reflects reality or to assess the potential implications of the model to society. This phase can happen during or after the initial publication of the basic natural science or social science model manuscript(s) which often form the basis for creating an integrated SES model where the coupling occurs during the Model Assessment phase. The Integrated Social Vulnerability class of models (see Box 6) often represent One-way coupling of natural and social science models via the integration of a model of the risk to a natural and/or man-made hazard [such as climate change (Hare et al., 2016)] and a model of social vulnerability and/or adaptive capacity (Jepson and Colburn, 2013) specifically in regards to the risk from climate change and sea-level rise to coastal communities, as in Colburn et al. (2016). That study is an example of both a qualitative one-way coupling in integrating the species and community diversity metrics as well as a quantitative one-way coupling between sea-level rise risk and the number of marine businesses affected.

In addition to coupling to describe the social impacts of changes in the marine environment, coupling at the Model Assessment phase can also help describe the economic impacts of these changes to the society as a whole, often through the use of One-way coupled Regional Economic Impact Models (see Box 7). These models have been integrated similarly as quantitative one-way coupled models where a climate-informed stock assessment model is used to generate a series of projections of future stock biomass and catch of the projection period (Ianelli et al., 2011) and the changes in fisheries yield is then used as an input in a dynamic computable general equilibrium (CGE) model of the Alaska fisheries and non-fisheries economy (Seung and Ianelli, 2016, 2019).

The human activity most commonly included in existing United States EBFM coupled modeling efforts is fishery catch. However, coupling at the Management Strategy Assessment stage usually implies that catches are not driven by any kind of behavioral model of fishers but rather based upon simple assumptions of fisheries mortality rates. Coupling at this stage is often a simple quantitative one-way relationship in which a series of alternative catch projections are multiplied by some fixed price to assess potential “economic impacts” of these different catch projections. These models can have some utility in fisheries and stocks in which the total TAC is caught nearly every year (full utilization) and catch projections are unlikely to change the relative prices across alternative target species. However, these models may perform poorly when the catch projections have an impact on the overall size of the catch over time through size-based targeting and production strategies among the fleet, as shown by Chen (2018) in the Bering Sea pollock fishery. These models may also perform poorly in situations where bycatch or quota constraints of other species jointly caught with the target species of interest may result in lower than full TAC utilization, as can happen in the New England and Bering Sea groundfish fisheries, more often prior to the implementation of catch shares (Brinson and Thunberg, 2013, 2016). As this coupling occurs so late in the process, it generally still only provides meaningful information about the ecological impacts of proposed management strategies or environmental changes but the impacts to society and the fishery are largely through narrative description. Thus these types of models may be useful to assess the management strategies across ecological objectives, but are unlikely to provide substantial useful information about the social or economic impacts of these ecological outcomes. Integrated models that account for management and fishing responses to changing physical and economic responses will provide more realistic projections and understanding of uncertainty (e.g., Hollowed et al., 2020; Reum et al., 2020).

The therMizer model was developed in order to better understand the effects of rising ocean temperature and changing plankton communities on fish size and abundance (Woodworth-Jefcoats et al., 2019). It is a size-based food web model with individual species represented, gear-specific fishery, and effects of temperature on metabolism and aerobic scope. It can incorporate dynamic fishing scenarios and the output can be used to estimate changes in catch value as a result of modeled climate and/or fishing scenarios, but is not intended to explain changes in fisheries behavior from any non-ecological basis.

Initial modeling results from the Alaska Climate Integrated Modeling (ACLIM) project are focused on key fisheries management areas of concern about climate change and species distribution shifts in the Eastern Bering Sea (Hermann et al., 2019; Holsman et al., 2020; Reum et al., 2020). The ACLIM project relies on repeated communication with stakeholders and managers to assess potential climate change effects as well as potential management and fleet responses to the changes they are currently experiencing (Hollowed et al., 2020). As the ACLIM project develops further, connections will be made to integrate impacts beyond commercial fisheries to create a series of integrated end-to-end models that explore a suite of climate scenarios with a variety of fisheries fleet dynamics models and potential management instruments.

Although less prevalent than their commercial counterparts, recreational fishery bioeconomic models are also employed in assessing the impact of alternate regulations on fisheries and stocks. In the Northeast United States, a multispecies SES for cod and haddock is used to assess the impact of differing possession and size limits and seasonal closures on both stocks, as well as changes in the recreational welfare derived from this mixed recreational fishery due to the regulations (Lee et al., 2017). The model provides two-way coupling between an economic recreational demand module based on choice experiment survey data and an age-structured stock dynamics module. Recreational landings and discards are estimated based on fishing regulations, stock structure, and recreational effort derived from expected utility maximization. The resultant fishing mortality is passed to the stock dynamics module, which allows estimation of alternate stock trajectories based on variability around initial conditions, uncertainty in recruitment, and changes in regulations. These changes in stock conditions will, in turn, affect future recreational fishing behavior.

Researchers at the Northeast Fisheries Science Center developed the model and a joint Northeast Fishery Management Council (NEFMC)/Mid-Atlantic Fishery Management Council (MAFMC) panel reviewed the model in 2012. The authors note challenges to employing the model in support of resource management decision-making includes time lags in and uncertainty around the scientific information used, as well as inflexibilities in the management system, which results in a condensed period for model updating and an undermining of stakeholder trust in the management process due to the use of outdated information to assess current conditions. Results from the first implementation of the model in support of fisheries management indicate that although changes in regulations had substantial impacts on the recreational welfare generated, minimal long-run conservation value was derived from even the most draconian alternatives assessed. This result actually facilitated management uptake, as it was viewed favorably by stakeholders, minimizing concerns around the adoption of a novel approach for specification setting which might have otherwise hampered adoption. Since 2013, this model has been employed to support the selection of recreational groundfish measures for the Gulf of Maine in each round of specification setting. Revised versions of the choice experiment were conducted in 2014 and 2019. The simulation model has also since been further refined at the request of the NEFMC to allow for analysis of slot limits and regulations that vary within the year or by fishery mode.

To better understand ecosystem function and the deep connections local stakeholders have with the marine environment, conceptual models of the SES can be developed in partnership between scientists and resource users at a local scale, such as for the case of the community of Sitka, Alaska, and Sitka Sound (Rosellon-Druker et al., 2019), or at a regional-or ecosystem-scale (Harvey et al., 2016). These “place-based” SES models have been developed as part of an Integrated Ecosystem Assessment (IEA) in which the two-way coupled conceptual models are used to understand the multifaceted nature of well-being in local communities and to generate a set of feasible indicators of community well-being related to their interactions with the marine environment (Szymkowiak and Kasperski, 2021). The repeated interaction with the community of Sitka has helped generate trust between the researchers and community members, and they appreciate the availability of performance metrics that reflect how they interact with the marine environment, but these models and metrics have not been designed to support any specific management decision and are generally limited to providing local context for decision makers.

Kaplan and Leonard (2012) offer one example of a simple One-way coupling from an end-to-end model to a regional input-output model. In this work, catch projections from Atlantis (Fulton et al., 2011) ecosystem model scenarios (Kaplan et al., 2012) were passed to the IOPAC input-output model (Leonard and Watson, 2011). This allowed the authors to evaluate the economic impact (in terms of jobs and income in the broader economy) stemming from changes in port-level landed revenue; landed revenue was assumed to be the product of catch and constant price per port. This coupling to IOPAC allowed outputs from the end-to-end model to be translated to direct effects (on the seafood sector), indirect effects (on suppliers to the seafood sector), and induced effects (related to broader household spending), rather than only reporting landed revenue as a modeling endpoint. The Atlantis ecosystem model scenarios tested effects of fishing gear shifts and spatial closures. Although the Atlantis model projection period was 20 years, the coupling to the input-output model was made only for years 1 and 15, largely due to the caveats described above related to the static nature of input-output models and their lack of behavioral responses.

The coupled approach of Kaplan and Leonard (2012) has been replicated with other end-to-end models, and gained traction with fishery management audiences, but the approach is not “operational,” i.e., it is not routinely delivered as a management product. Fay et al. (2019) recently applied a similar coupled approach in the Northeast United States, coupling an Atlantis model to the NERIOCOM input-output model (Steinback and Thunberg, 2006). The Atlantis-IOPAC coupling of models has been presented to the Pacific Fishery Management Council, council subcommittees, and review panels (Kaplan and Marshall, 2016). A recent application by Hodgson et al. (2018) considered port-level effects of ocean acidification on revenue, income, and employment. IOPAC is routinely updated for use by the Pacific Fishery Management Council, so further coupling is possible and is likely a constructive way to frame ecosystem modeling results, in particular because input-output models are commonly used throughout the United States by policy makers (including outside fisheries).

A review of key factors which facilitated management uptake, as detailed in Supplementary Appendix Table 1, indicates some commonalities across models successfully used in management support. Most obvious is when stakeholders and managers request answers to specific questions that necessitate the development of a coupled model, regardless of the types of models or coupling employed. Somewhat less apparent and equally necessary is the need to have models which function within management timelines and are able to test management-relevant policy instruments. For tactical advice, this reality usually translates into a need to develop relatively lightweight models which can iterate combinations of policies quickly to inform the development of management actions. For strategic advice, models need to realistically capture the dynamics of resources most closely associated with human actions under management, such that stakeholders and managers glimpse their perception of the system in model outputs, which can build trust in its function. Providing either tactical or strategic advice in this manner necessitates coupled SES models, in that it is the interplay of biology and human behavior which determines the success or failure of policy instruments. Timing plays a key role in management uptake, and good models are often left unused because they may not be completely developed in time for management actions.

In light of this, building models based on recurring demands helps to ensure the often long lead time necessary to develop models does not unduly interfere with their adoption. Many times uptake is as much a fortuitous confluence of events as careful planning. As such, having a developed model which can answer a scientifically interesting and seemingly policy-relevant question ready when it becomes important from a manager’s perspective can also bear fruit. Retrospective analyses of prior management or environmental shocks also provide valuable insights into probable human responses to future management alternatives and environmental shocks such as climate change.

One of the advantages of recognizing the need for coupled modeling is that generic modeling support activities can be applied to all disciplines. For example, the extensive use of a variety of models for stock assessments has led to NOAA support for the development of a modeling toolbox infrastructure, which includes the NOAA Fisheries Integrated Toolbox and current supports a number of the SES models described in this manuscript. This integrated toolbox infrastructure is being designed to support stock assessment, ecosystem, economic and human dimension models. Sharing a basic requirement for provision of model metadata, version control, model sharing, and other aspects will facilitate model coupling. Similarly, the need and investment for access to high-speed computing for coupled earth-system models will apply to coupled SES models.