Using an Atlantis model of the southern Benguela to explore the response of ecosystem indicators for fisheries management

Highlights

• The Atlantis modelling framework has been successfully applied to the southern Benguela ecosystem.

• The modelled system responds to forcing in ways that reflect observations and understanding of the real-world system.

• Ecosystem indicators in the model exhibit behaviour in line with expectations from real-world studies.

Abstract

Atlantis is a whole-of-system modelling framework developed for Management Strategy Evaluation. This paper describes an Atlantis model that was built to simulate the southern Benguela ecosystem and its major associated fisheries to assist fisheries management in the region. We divided the region into spatial zones based on hydrodynamics, current fishing management, and important ecosystem processes. We divided the biological components of the system into functional groups based on trophic interaction, life history traits and fisheries management objectives. We evaluated the model against historical data and known ecosystem interactions (such as competition and predation), and found that it simulates important ecological processes well at multiple trophic levels. We tested the model under fishing pressure scenarios and evaluated the performance of common ecosystem-level indicators. The response of the modelled system (as shown by indicators) was in line with expected behaviour of the indicators, reinforcing our confidence in the usefulness of the model.

Introduction

Ecosystem-based fisheries management (EBFM) is a holistic approach to the management of marine living resources (Larkin, 1996, Link, 2002). EBFM is centred on multi-species interactions in a variable environmental context, and includes social, economic and political elements as part of fisheries management (Larkin, 1996). Significant commitments have been made in recent years towards the implementation of Ecosystem-based Fisheries Management (EBFM) in several ecosystems around the world (Pitcher et al., 2009, Sherman et al., 2005, Keith et al., 2013). Fisheries legislation in many countries now includes ecosystem objectives, generally based on such agreements as the Law of the Seas Convention, which explicitly requires a precautionary approach in the management of marine living resources. Balanced against such ecological conservation requirements is the continuing need to engage in fishing activities for economic, nutritional and social needs (Hilborn et al., 2012, Hall et al., 2013). The development and implementation of such solutions requires a broad array of scientific support (Worm et al., 2009).

Marine fisheries management typically employs focussed mathematical models of single species which are rigorously fitted to data. In contrast, ecosystem models emphasise generality and realism over precision (sensu Levins, 1966). Ecosystem models are useful in comparing the relative performance of multiple candidate management strategies, or the performance of a single strategy under multiple alternative assumptions regarding environmental conditions or ecosystem dynamics. Although the complexity of ecosystem models may preclude rigorous validation (Oreskes et al., 1994), they nevertheless afford the only feasible means for conducting ecosystem-scale experiments (Scheffer and Carpenter, 2003), and allow for representation of multiple regimes in a single model. Fulton et al. (2011) note that the associated increases in model complexity in moving from single-species to ecosystem models will generally mean that the latter are not suitable for tactical management (such as quota setting). Rather, ecosystem models are better employed as strategic tools to support adaptive management (Fulton, 2010), and are proving particularly helpful in understanding overall ecosystem functioning, including trophic cascade mechanisms and complex non-linear system responses (Fulton et al., 2005, Plagányi and Butterworth, 2004, McDonald et al., 2008).

Management strategy evaluation (MSE) is an iterative process used to design and evaluate operational management strategies (as described in de la Mare, 1996, Butterworth and Punt, 1999, Sainsbury et al., 2000). MSE seeks to guide the selection of a management strategy by analysing the performance and trade-offs of any candidate strategy in light of management objectives. MSE has been implemented as a fisheries management tool in South Africa (Cochrane et al., 1998), and offers a useful methodology to achieve EBFM objectives (Sainsbury et al., 2000). Traditionally MSE has been done with single species, or simple multi-species models. However, basing the work around an ecosystem model has proved an effective approach (McDonald et al., 2008, Fulton et al., 2011). Models such as the one described in this paper (ABACuS, Smith, 2013) allow candidate management strategies to be evaluated in terms of quantitative performance measures such as targets and limits (Sainsbury et al., 2000, Smith et al., 2011). Any system model will incorporate assumptions about system processes (which are generally uncertain), so the use of multiple models allows for such evaluation to be made under alternative hypotheses about system structure and function and alternative model formulations (Fulton, 2010, Sainsbury et al., 2000).

The southern Benguela ecosystem is generally considered to include both the southern portion of the Benguela current itself, and also the south coast of South Africa and the Agulhas Bank (see Fig. 1). The Benguela Current extends from Cape Point all the way up the West Coast to Angola, but there is a strong biogeographical division at the Luderitz upwelling cell on the Namibian coast, which is commonly taken as the division between the northern and southern parts of the current. In our model we have instead taken the political boundary between South Africa and Namibia at the mouth of the Orange River as our northern limit, reflecting the focus of the model on fisheries management. The west coast of South Africa is characterised by strong wind-driven upwelling in the Benguela Current, which gives rise to high primary productivity, and this in turn supports large biomasses of fish species.

The Agulhas Current is a warm, fast-flowing water mass that moves southwards down the Natal coast on the east of South Africa. Along this coastline, the continental shelf is narrow, but on the south coast the Agulhas Bank extends 250 km further south than the shoreline. The main core of the Agulhas current is diverted away from the coastline along the edge of the Agulhas Bank, and waters on the Bank are relatively warm and calm, making it an ideal spawning area for many pelagic fish species.

Connecting the Agulhas Bank to the productive waters of the west coast are “transport currents”, which include a narrow jet current off the Cape of Good Hope, and also a series of ring-shaped eddies that are shed from the Agulhas Current as it retroflects southwards. These rings, along with the Good Hope Jet, transport eggs and larvae spawned on the west Agulhas Bank to the strong upwelling zone near Cape Columbine. This region is an important nursery area for juveniles of sardine, anchovy, round herring and snoek.

The fisheries management regime in the southern Benguela region has been a pioneer in both implementation of EBFM and in the ecosystem modelling to support it (Pitcher et al., 2009, Shannon et al., 2006, Shannon et al., 2010). Some of the modelling approaches already applied are detailed below:

• Substantial food-web modelling has been performed in the region using Ecopath with Ecosim (EwE, Christensen et al., 2005). Initial trophic flow models were built using Ecopath (Jarre-Teichmann et al., 1998, Shannon and Jarre-Teichmann, 1999), and a subsequent model was used to explore the effects of fishing on pelagic stock structure under various trophic control assumptions (Shannon et al., 2000). Later EwE models compared trophic flow in the southern Benguela food web between the 1980s and 1990s (Shannon et al., 2003), and investigated the drivers of regime shifts in small pelagic fish populations (Shannon et al., 2004a, Shannon et al., 2004b).

• The individual-based model OSMOSE (Object-oriented Simulator of Marine ecOSystem Exploitation, Shin and Cury, 2001) has assumptions of size-based predation and focuses on fish population dynamics. An OSMOSE model of 12 fish species in the southern Benguela was used to simulate the same fishing scenarios as the EwE model of Shannon et al. (2003), and the results were compared (Shin et al., 2004). A general consistency in the results, despite substantially different assumptions in the modelling approaches, increased confidence in the usefulness of both models. OSMOSE has also been used to explore the sensitivity of ecosystem-based indicators across a range of fishing scenarios (Travers et al., 2006).

• A frame-based model was developed to investigate regime shifts in the sardine and anchovy stocks under various scenarios of climate and fishing (Smith and Jarre, 2011). Frame-based models represent stable regimes as “frames”, and have encoded rules which govern switching between frames. They are conceptually similar to Markov state-and-transition models. Despite extreme simplification of many system dynamics, the frame-based model was useful in highlighting the combined role of climate and fishing pressure in initiating or delaying regime shifts.

• Operating models (OMs) have been developed to explore specific predator/prey dynamics in the context of MSE, such as the interaction between hake and Cape fur seal (Punt and Butterworth, 1995).

• Minimum Realistic Models (MRMs) have been developed to study networks of interaction between small groups of species, such as lobsters, sea urchins and abalone (Blamey et al., 2013).

• More broadly, Smith et al. (2011) used the existing EwE and OSMOSE models of the southern Benguela, along with models of other ecosystems, to explore the ecosystem-level impacts of fishing low-trophic-level species.

Atlantis (Fulton et al., 2004) is an ecosystem modelling framework designed for management strategy evaluation. Previous implementations of the framework have ranged in scale from small estuarine regions to several million square kilometres of ocean. The availability of an Atlantis model in the southern Benguela ecosystem has several immediate benefits. The strengths and weaknesses of the different modelling techniques can be explored in a reasonably well-understood ecosystem. In particular, this enables us to extend the previous work on EwE/OSMOSE comparisons of Shin et al. (2004). A detailed comparison of this Atlantis model with the EwE and OSMOSE models can be found in Smith et al. (in press). From an MSE perspective, the new Atlantis model covers the whole socio-ecological system more comprehensively than any existing model of the southern Benguela, and offers the potential to evaluate management options on a broader range of assessment criteria.

The implementation of Atlantis in an upwelling system also improves our understanding of the potential for usefully modelling this kind of marine system with Atlantis. Although an Atlantis model has been designed for the Californian current system (Horne et al., 2010), the model in that region was focussed on demersal rockfish and did not emphasise upwelling. In the southern Benguela several of the major target species are small pelagic fish, which occupy a vital “wasp-waist” position in the functioning of upwelling ecosystems (Cury et al., 2000). Pelagic fisheries are particularly prone to high catch variability and risk of collapse, due to high natural variability and instability of the fish populations (Fréon et al., 2005). Models such as this new Atlantis implementation are vital to evaluating both the economic implications and the ecosystem-scale impact of alternative management strategies for pelagic fisheries (Smith et al., 2011).

The model described in this paper is called “ABACuS”, or “Atlantis in the Benguela and Agulhas Current System”. The aim of this model is to explore the impact of multiple fisheries operating in the southern Benguela ecosystem. The key issue under consideration is how the exploitation of individual species affects the broader ecosystem, and how different fisheries management strategies might change that ecosystem-level impact. We are also interested in the combinatorial effect of multiple fisheries operating in the same ecosystem, which is explored more fully in Smith et al. (2015).

Atlantis is a versatile modelling framework with multiple sub-models to simulate ecology, fisheries and management. Sub-models are executed in separate code libraries and can be selectively implemented for a given simulation. The ecological component of Atlantis consists of a spatially explicit stock structure of higher trophic levels supported by a deterministic primary production model, driven by hydrodynamic forcing of nutrient and water flows.

The modelled region is represented by polygonal spatial zones with multiple depth layers. The important biological components of the ecosystem are placed into functional groups, which may consist of a single species, or may contain assemblages of multiple species as appropriate. Within the Atlantis framework, the flow of nutrients is tracked explicitly (including uptake, processing and remineralisation) through the major components of the local food web. Nutrient-, light-, space- and temperature-dependent primary production is represented using size-structured phytoplankton and macrophyte biomass pools. Planktonic movement is determined by advective transfer between the polygons, and modelled nektonic organisms can exhibit directed movement between the polygons as well as in and out of the modelled region as a whole (to represent long-distance migration for species which may be present in the region only seasonally). Lower trophic levels are modelled as biomass pools, but vertebrates are modelled with age and stock structure, with the model tracking population change and the condition of an average individual. The partitioning of the ecological species is based on the model objectives, so in the ABACuS model species which experience more directed fishing pressure tend to be in their own group. Various growth and predation functions can be implemented as data and ecological knowledge dictate, including Beverton-Holt recruitment (for fish) and fixed offspring/adult relationships (for birds, mammals and chondrichthyans). The functions used in this Atlantis model are detailed in Table 1.

Harvest, the impacts of other human activities, monitoring and regulation are each handled by separate sub-models. However, as this work is focussed on the simulation of the ecosystem, these other sub-models will not be discussed in detail here.

Assessing the state of a complex ecosystem requires simultaneous evaluation of many indicator variables (Sainsbury and Sumaila, 2002, Levin et al., 2009, Shannon et al., 2010). Model outputs from an Atlantis simulation cover a vast array of metrics, and often produce gigabytes of data. Sensitivity and uncertainty analysis based on randomised parameter sets (e.g. Gal et al. (2014)) is rendered infeasible by the hundreds or thousands of parameters in an Atlantis model, and model run times of several hours. It is therefore necessary to focus questions of model skill or model usefulness on specific objectives.

The evaluation of model skill can take many forms (Stow et al., 2009, Bennett et al., 2013). Multivariate analysis of residual structure (Allen and Somerfield, 2009) can be a useful approach for systems models, but requires a high degree of confidence in the data across the full scope of the system being modelled. A widely accepted standard for skill evaluation involves general pattern matching (Stow et al., 2009), and here this was applied across multiple categories of spatial and higher-dimension information. This study compared the biomass response of functional groups to fishing pressure with the historical records, and examined food-web restructuring under fishing pressure.

Understanding of core process that drive ecosystems is often incomplete, and data resolution may be highly variable across different levels of the system. Therefore, it can be difficult to derive metrics of system performance that are relevant at an ecosystem scale. However, the usefulness of a model is best evaluated in terms of the objectives of the modelling exercise (Starfield, 1997, Oreskes et al., 1994). The purpose of this model is to inform management strategy evaluation, and allow prospective managers to compare alternative strategies in an EBFM context. As such, the key aspect of the model that needs confirmation is how realistically the ecosystem responds to fishing pressure. To measure this, we need suitable indicators that collate such information at a system level. Such a suite of indicators should represent several functional groups, including a spectrum of fast- and slow-growing species, habitat-defining groups and target species for the fisheries (Link et al., 2002, Fulton et al., 2005, Shin et al., 2010a).

Extensive work has been done over the past decade in establishing ecosystem indicators that are robust to assumptions about system processes and useful in communicating system impact due to fishing (e.g. Fulton et al., 2005, Shin et al., 2010a, Link, 2005). If the model can produce indicator behaviour that responds to perturbation in a similar way to the expected behaviour of indicators in the real system, this is a good basis for using the model to provide practical advice to fisheries management. Shin et al. (2010b) identified six real-world indicators that fit the requirements of EBFM: mean length of fish in the community, the mean lifespan, the proportion of predatory fish in the system, the proportion of under- and moderately-exploited species, trophic level (TL) of landings and the inverse biomass coefficient of variation. Detailed descriptions of these indicators and their expected behaviour under fishing pressure can be found in Shin et al. (2010b). Under the fishing pressure tests that we applied, which involve constant fishing pressure tuned to average catches in the system, four of these indicators are appropriate for use. We also present results of total change in biomass of each group under two fishing tests.

Marine ecosystems respond to both natural forcing (such as changing climatic conditions, occurring on all timescales from daily variation in temperature and light conditions through to seasonal changes and multi-decadal shifts in climate regime) and anthropogenic forcing, which includes direct interactions such as fishing, as well as nutrient and pollutant input through rivers and dumping, etc. (Travers et al., 2007). Dynamic ecosystem simulation models are very effective in improving understanding of the key processes, components and scales at work in a system (Fulton, 2010). Such models, in conjunction with MSE, also allow us to examine the robustness of management strategies under various climate scenarios.