INTRODUCTION
Overfishing has had severe impacts on target fish populations (Pauly et al. Reference Pauly, Christensen, Dalsgaard, Froese and Torres1998; Christensen et al. Reference Christensen, Guénette, Heymans, Walters, Watson, Zeller and Pauly2003; Myers & Worm Reference Myers and Worm2003, Reference Myers and Worm2005; Halpern et al. Reference Halpern, Walbridge, Selkoe, Kappel, Micheli, D'Agrosa, Bruno, Casey, Ebert, Fox, Fujita, Heinemann, Lenihan, Madin, Perry, Selig, Spalding, Steneck and Watson2008). In many systems, historical reconstructions conclude that contemporary fisheries achieve only a fraction of their former yields (for example see Rosenberg et al. Reference Rosenberg, Bolster, Alexander, Leavenworth, Cooper and McKenzie2005; Cadigan & Hutchings Reference Cadigan, Hutchings, Holm, Smith and Starkey2001). Although formal fisheries management plans may mandate the restoration of depleted species to some specified level, for example a level based on maximum sustainable yield, some fraction of estimated virgin biomass or other criterion, any set of goal biomasses developed by single species methods may be incommensurate in an ecosystem context (Larkin Reference Larkin1977, Reference Larkin1996; Walters et al. Reference Walters, Christensen, Martell and Kitchell2005). A responsible management objective might be to restore entire ecosystems using a holistic approach to management, rather than through a piecemeal application of single-species biomass goals (Campbell et al. Reference Campbell, Gray, Hazen and Shackeroff2009; Pitcher Reference Pitcher2001).
Historic ecosystems have been suggested as potential whole-ecosystem restoration goals (Pitcher Reference Pitcher2001), not least because (1) as past systems existed, we know their structure was compatible with thermodynamic laws, whereas an invented ecosystem structure may not be; and (2) past abundances act as cognitive guides as to what human benefits it might be possible to recover in an almost universally depleted marine realm (Pitcher Reference Pitcher2001, Reference Pitcher2005; Pitcher et al. Reference Pitcher, Haggan, Preikshot and Pauly1999). The past may thus serve as an analogue for a restored future, which sets expectations as to the production potential, biodiversity and resilience potentially present in a healthy ecosystem. More sophisticated goals have also been evaluated; for example, ecosystems containing an optimized set of species biomasses as defined by the optimal restorable biomass concept (ORB: Ainsworth & Pitcher Reference Ainsworth, Pitcher, Nielsen, Dodson, Friedland, Hamon, Musick and Verspoor2008; Pitcher et al. Reference Pitcher, Ainsworth, Buchary, Cheung, Forrest, Haggan, Lozano, Morato, Morissette, Levner, Linkov and Proth2005). Careful consideration and debate should be given to selecting the restoration goal, which is largely a question of social priorities (Campbell et al. Reference Campbell, Gray, Hazen and Shackeroff2009). However, even if suitable goals could be established for whole-ecosystem restoration, it is doubtful whether the ecosystem can be manipulated into the desired state (Mace Reference Mace2001). The work presented here offers a first step in that direction.
We introduce a new tool for ecosystem-based fisheries management (EBFM) to help identify fishing strategies that will promote a shift in the ecosystem towards a more desired state. It is a modified version of Ecopath with Ecosim's (EwE) policy search routine, which includes updated algorithms and new user interfaces (for EwE see Polovina Reference Polovina1984; Christensen & Pauly Reference Christensen and Pauly1992, Reference Christensen, Pauly, Christensen and Pauly1993; Walters et al. Reference Walters, Christensen and Pauly1997, Reference Walters, Pauly, Christensen and Kitchell2000; for the policy search routine see Christensen & Walters Reference Christensen and Walters2004). The principal advance we have achieved is the development of the specific biomass (SB) criterion, a flexible ecological objective function that can be used to direct the policy search routine to achieve whole ecosystem restoration, while providing users with precise control over the qualities of the restoration plan. The SB criterion replaces a much simpler ecological objective function that is present in the release version of EwE software, and can work in concert with previously described economic and social objective functions to generate a wide range of restoration scenarios.
We conducted the analysis in two parts. First, we demonstrate the effectiveness of the SB criterion by attempting to restore the historic 1950 AD ecosystem of Northern British Columbia (BC) in dynamic simulations commencing with the 2000 EwE model (Ainsworth et al. Reference Ainsworth, Pitcher, Heymans and Vasconcellos2008a). Second, we evaluate candidate restoration scenarios with a more practical ecosystem goal, based on the ORB concept, which resembles a goal that society may wish to consider. The biomass configuration of this example goal ecosystem supports profitable fisheries (i.e. fisheries where revenues exceed costs) and maximizes an objective function based on long-term economic returns from fisheries, while prohibiting extinctions (see economic ORB in Ainsworth & Pitcher Reference Ainsworth, Pitcher, Nielsen, Dodson, Friedland, Hamon, Musick and Verspoor2008). Because we structured the ORB ecosystem to provide economic returns, it is designed, once restoration is achieved, to generate more sustainable benefits than either the 2000 ecosystem or the historic 1950 ecosystem; this benefit may be used to defray the costs of restoration.
METHODS
Ecopath with Ecosim
Ecopath with Ecosim is a trophodynamic marine ecosystem simulator that partitions living and non-living components of the ecosystem into sets of similar species (functional groups). It acts as a thermodynamic accounting system, tracking the flows of biomass between groups according to a diet matrix, deducting energy spent in respiration and lost as unassimilated food (Ecopath: see Polovina Reference Polovina1984; Christensen & Pauly Reference Christensen and Pauly1992; Ecosim: Walters et al. Reference Walters, Christensen and Pauly1997; Appendix 1, see supplementary material at URL http://www.ncl.ac.uk/icef/EC_Supplement.htm).
Ecosim's policy search routine (Christensen & Walters Reference Christensen and Walters2004) permits exploration of ecological, economic and social trade-offs inherent in fisheries management policies, and optimization of fleet design to achieve policy objectives. It uses a non-linear optimization algorithm (Fletcher & Powell Reference Fletcher and Powell1963) and iteratively adjusts fishing mortalities until the fleet-effort combination is discovered that maximizes fishery benefits (Christensen et al. Reference Christensen, Walters and Pauly2005). The objective function used for the task of rebuilding here includes economic and ecosystem rebuilding criteria (Eq. 1).
WECON and WREB are relative weighting factors for economic and ecosystem rebuilding. The summed terms represent net present value (NPV) of the harvest plan for each functional group (i) and gear type (j) and the ecosystem rebuilding benefits measured using the SB index for each functional group and simulation time step (t). NPV summarizes the expected stream of profits, discounting benefits far off in the future using an intergenerational discounting term (Sumaila Reference Sumaila, Pitcher and Sumaila2001, Reference Sumaila2004; Sumaila & Walters Reference Sumaila and Walters2005) at conservative discount rates for the current generation (δ = 4%) and future generations (δfg = 10%). We used the intergenerational discounting approach because it is more suitable for long-term fisheries conservation than the conventional method of discounting (Ainsworth & Sumaila Reference Ainsworth and Sumaila2005).
Northern British Columbia models
This paper employs previously published EwE models for Northern British Columbia (BC) for the years 1950 and 2000 (for full details of these see Ainsworth Reference Ainsworth2006 and Ainsworth et al. Reference Ainsworth, Pitcher, Heymans and Vasconcellos2008a). Briefly, the 1950 and 2000 models use a common structure. Fifty-three functional groups describe the ecosystem, and 11 juvenile/adult split pools represent trophic ontogeny in commercially important fish species. Basal species tend to be aggregated, while fishery targets and ecologically important species are modelled in more detail. Biomass data for 1950 and 2000 comes from Department of Fisheries and Oceans (DFO) reports, literature sources and community interviews (Ainsworth Reference Ainsworth2006). Catch data comes mainly from historical catch records and includes estimates of illegal, unreported and unregulated catch (Ainsworth & Pitcher Reference Ainsworth and Pitcher2005). Model basic parameters were evaluated using a sensitivity analysis, and the 1950 simulation was fitted to time series. Trophic vulnerabilities (species interaction rate parameters) were calibrated for the 1950 model and adapted to the 2000 model using a novel procedure that assumes stationarity in species foraging tactics (Ainsworth et al. Reference Ainsworth, Pitcher, Heymans and Vasconcellos2008a).
Fishing fleets
Larkin (Reference Larkin1996) said, ‘Existing fleets are a blunt instrument for fine tuning the relative abundance of species’. Where gear types catch multiple species, this limits the ability to structure the ecosystem for human benefit. For example, one depleted species cannot be selected for rebuilding while maintaining high levels of catch on sympatric species using unselective gear because high levels of bycatch could sabotage rebuilding efforts. Here, we refer to a hypothetical fishing fleet that pursues one species group per sector as the maximum dexterity or maxdex fishing fleet. The maxdex fleet is used to project a ‘best case’ restoration scenario, where precise manipulation of species biomasses is possible and the goal ecosystem configuration is not limited by the selectivity of fishing gear, only by the ecological interplay of species. This sets the benchmark for restoration and helps evaluation of the effectiveness of various candidate fleets to be used for the task of rebuilding.
We use two other fleets in the rebuilding scenarios here. The first is a representation of the Northern BC fleet c. 2000 (described in Ainsworth et al. Reference Ainsworth, Pitcher, Heymans and Vasconcellos2008a); the second is a ‘next-generation’ fishing fleet designed with ecologically and socially responsible criteria in mind. We refer to this as the lost valley fishing fleet (after Pitcher et al. Reference Pitcher, Heymans, Ainsworth, Buchary, Sumaila, Christensen, Knudsen, MacDonald and Muirhead2004; described for Northern BC by Ainsworth et al. Reference Ainsworth, Heymans, Pitcher and Pitcher2004). The lost valley fleet reduces bycatch and collateral habitat damage to within technologically achievable limits, and adheres, unlike most world fisheries today (Pitcher et al. Reference Pitcher, Kalikoski, Pramod and Short2009), to the FAO (Food and Agriculture Organization of the United Nations) Code of Conduct for Responsible Fisheries (FAO 1995).
A new restoration objective function
The distributed version of EwE V5.1 contains several objective functions representing economic, social and ecological benefits. It contains another standard objective function to rebuild biomass of depleted species, which is called ‘mandated rebuilding’ (Christensen et al. Reference Christensen, Walters and Pauly2005). However, that objective function is suitable only when a single species group is mandated for recovery and is not adequate when multiple species groups are to be restored simultaneously. It recognizes improvement only through biomass increases of groups, whereas selective declines may be needed to facilitate growth in species through direct or indirect trophic interactions. The ‘mandated rebuilding’ routine also assumes that restoration is equally desirable in all groups, which makes prioritization impossible.
The specific biomass (SB) criterion is an objective function designed to make any prioritization between species groups explicit, and to give the user more control over the trade-offs inherent in ecosystem restoration. The algorithm used to compute the new SB function essentially governs the trade-off between restoring biomass of a large number of easily recoverable groups, versus a smaller number of critical ‘problem’ groups. The latter are groups that are slow to increase in biomass (such as slow-growing groups) or where biomass reductions are difficult to achieve (such as groups not subject to direct harvests). The SB function uses two main variables to define the desired trade-off: the unit of ecosystem ‘improvement’ towards goal biomass and the choice of model used to calculate the value of marginal improvement towards the goal.
The SB criterion evaluates the biomass of species groups flagged by the user for rebuilding. The difference between starting biomass, when t = 0, and the goal biomass for restoration defines the potential scope for improvement. At each time step, group biomass is employed to calculate an internal term (θ) representing the proximity of the group's biomass to the rebuilding goal, where θ is a unitless multiple of the initial biomass differential between the initial and goal biomass. Initially, θ = 0 if starting biomass is less than the goal and θ = 2 if starting biomass is greater than the goal. As group biomass increases or decreases towards the goal, θ will approach 1. If starting biomass equals goal biomass, then θstart = 1. At each simulation time step, proximity to goal (θ) is calculated (Eq. 2):
![\begin{equation}
\theta = \left\{ {\begin{array}{*{20}l}
{\frac{{B_{\it current} \,{-}\, B_{\it start} }}{{B_{\it goal} \,{-}\, B_{\it start} }}} &&&& {{\rm if}\,B_{\it start} < B_{\it goal} } \\[6pt]
{2 - \left[ {\frac{{B_{\it current} \,{-}\, B_{\it start} {\rm }}}{{B_{\it goal} \,{-}\, B_{\it start} }}} \right]} &&&& {{\rm if}\,B_{\it start} > B_{goal} } \\\end{array}}\right.\end{equation}](https://static.cambridge.org/binary/version/id/urn:cambridge.org:id:binary:20160920234121178-0214:S0376892910000135:S0376892910000135_eqn2.gif?pub-status=live)
Bcurrent is the functional group's biomass at time step t, Bgoal is the goal biomass and Bstart is baseline biomass when t = 0. Each group's contribution to the SB index is calculated based on θ; groups will contribute their maximum to the objective function when θ = 1.
If we define the unit of ecosystem ‘improvement’ towards the goal strictly as a change in biomass (hereafter called the biomass criterion), then in depleted ecosystems the search will tend to advocate fishing strategies that greatly reduce fishing mortality from baseline levels. If we define improvement in terms of a per cent change towards target (hereafter called the per cent criterion), which will cause a larger number of groups to approach their targets, but less overall change in system biomass. Under the per cent criterion, proximity to target (θ) is passed directly to the marginal improvement model as θper cent for each functional group. If the unit of improvement is biomass, then θbiomass is first calculated (Eq. 3). A combined term (θcombined) can also be used for mid-range solutions (Eq. 4).
X is a weighting factor between 0 and 1.
The marginal improvement valuation model allows the user to weigh the relative contribution of a functional group to the objective function SB according to the group's current distance from the goal biomass. Under the linear valuation model (Fig. 1a; Eq. 5), all species groups are weighted equally in the calculation regardless of their distance to target.
The linear model will make whatever trade-offs are necessary to reduce biomass residuals versus the desired ecosystem configuration, minimizing ∑it |θit − 1| for each functional group i and time step t.
Figure 1 Three models describing marginal improvement in SB function. The initial ecosystem condition is taken as θ = 0 if start biomass is less than goal, or θ = 2 if start biomass is greater than goal. As the functional group approaches its goal biomass, θ approaches 1. (a) Linear model weighs a unit of improvement the same, regardless of the current biomass differential versus goal. (b) Quadratic model gives the greatest improvement to objective function when groups first begin to move towards target. (c) Asymmetric gamma model is precautionary; biomass increase towards target is more valuable to the objective function than biomass decrease.
Under the quadratic model option (Fig. 1b; Eq. 6), the greatest marginal increase in the objective function occurs when groups first begin to move towards their goal biomass. More groups will improve in the optimal fishing policies than under the linear model, but the average improvement in the proximity function θ will be lower. The quadratic model is precautionary because the objective function decreases rapidly as group biomasses drift away from their goals in either positive or negative direction.
The gamma valuation option determines marginal improvement based on a gamma function (Fig. 1c; Eq. 7). Because it is asymmetric, improvement from functional group biomass growth contributes more to the objective function than improvement from group biomass decreases; it is therefore the most precautionary option. Optimal policies will more often overshoot target biomasses than fall short.
Phi (φ) is a scaling term for the Y-axis, (γ) is the shape parameter and (β) is a scaling term for the gamma function (Γ)). The gamma model provides excellent flexibility to define the shape of the valuation curve. The model response is standardized according to arbitrary but interpretable values. Parameters are set so that the gamma objective function is worth 0.5 when proximity to goal (θ) is equal to 0.5 or 2.0. In other words, a functional group will contribute the same to the SB objective function when it is 50% short of its goal biomass as when it is 100% in excess of its goal biomass, and that value corresponds to 50% of the maximum possible contribution for that group towards the objective function. Parameters used to set this relationship were φ = 1.74, γ = 3.21 and β = 0.45.
We developed this new policy search tool as a compiled executable replacing Ecopath.exe, available from us on request. Some auxiliary parameters are accessible on the new interface, such as an ‘extinction threshold’, which allows the user to specify a minimum acceptable functional group biomass for the restoration plan, as a per cent of initial biomass. All harvest plans we evaluated employed a 5% depletion threshold, the default value in the new interface.
Restoration case study
Using the new SB objective function, we designed a restoration scenario that would convert the Northern BC ecosystem (c. 2000) to a more productive state. The goal ecosystem configuration was based on the 1950 ecosystem, optimized for long-term harvest benefits following the optimal restorable biomass (ORB) concept (Ainsworth & Pitcher Reference Ainsworth, Pitcher, Nielsen, Dodson, Friedland, Hamon, Musick and Verspoor2008; Pitcher et al. Reference Pitcher, Ainsworth, Buchary, Cheung, Forrest, Haggan, Lozano, Morato, Morissette, Levner, Linkov and Proth2005). Briefly, ORB is the equilibrium ecosystem condition that would result from fishing a historic ecosystem responsibly for a long period of time. The species composition in the ecosystem is adjusted, through application of an optimal fishery programme (Ainsworth Reference Ainsworth2006), to support the largest sustainable fishery profits at equilibrium while disallowing extinctions. We restricted the analysis to restoration plans targeting this economic ORB ecosystem since it proved to be a more competitive economic goal than the historical 1950 ecosystem and should provide a better return on the investment of restoration. Each restoration scenario is directed according to a multi-criterion objective that contains varying weights on economic and rebuilding (SB) objectives (Eq. 4).
Cost-benefit analysis
The cost-benefit analysis of the whole restoration plan can be divided into three stages (Fig. 2). In the restoration phase (α), optimal fishing mortalities estimated by the SB algorithm reduce harvest rates from the status quo level and allow the ecosystem to rebuild biomass of targeted species. The profit sacrificed from the baseline level (status quo fisheries) is considered the ‘cost’ of restoration, as these yields would otherwise have been available to resource users. The second phase is transitional (β); fishing effort is adjusted to cancel any remaining biomass accumulations. In the third phase (γ), final equilibrium harvest levels are maintained until the end of the evaluated time horizon.
Figure 2 Conceptual diagram showing cost-benefit analysis. Restoration plan consists of three phases: Optimal fishing phase for ecosystem rebuilding (α); readjustment of fishing effort to establish new equilibrium and cancel biomass accumulations (β); sustainable fishing at new profit equilibrium (γ). Costs and benefits of restoration are taken in relation to forecasted profits from the initial ecosystem, assuming status quo profit.
Exploitable biomass increases with restoration so fisheries can draw more profit sustainably from the restored ecosystem than from the original ecosystem. The difference is called the economic ‘benefit’ of restoration. We have assumed completely malleable fishing capital; there is no penalty associated with fleet restructuring, although such costs could be included where they can be estimated. Fixed and variable costs of fishing and catch value are specified by gear type (Ainsworth Reference Ainsworth2006).
Here, the restoration process begins in 2000. At the end of the restoration phase (α), which lasts 5, 7.5, 10, 15, 20 or 30 years, a new static model based on a particular time step in the simulation (in this case, the last year of the restoration phase) is created using the EII export/import procedure in Ecosim (Christensen et al. Reference Christensen, Walters and Pauly2005). We call this the transition model. In a few cases, minor changes were made to the transition models to recreate mass-balance (for example we cancelled residual biomass accumulations and adjusted predation mortality) but these had minimal effects on the species biomass values at equilibrium. During the transition phase (β), we cancelled biomass accumulations out of the transition model using another optimal fishing policy directed by the SB algorithm, which sought to hold biomasses at constant levels corresponding to the end of the restoration phase. The transition phase lasts 20 years. Harvesting then continues at the restored profit equilibrium throughout the equilibrium harvest phase (γ), which we extended to 100 years to evaluate benefits over several human generations.
RESULTS
Algorithm diagnostics
We tested the performance of the SB algorithm in achieving an ecosystem target, the historic 1950 ecosystem. The biomass criterion for unit of improvement allows a greater overall change in functional group biomass towards the target configuration than either the ‘mixed’ or ‘per cent’ unit of improvement. It allows rebuilding of heavily depleted species groups. The per cent criterion results in a more even improvement among species groups.
The linear marginal improvement valuation model produced the greatest overall change in the ecosystem, indicated by a low sum of squares versus the target configuration. The quadratic and gamma valuation models were unable to restructure the ecosystem to match the target configuration as precisely. The average per cent improvement across species groups was approximately the same for the linear and quadratic valuation models; however, the gamma model allowed more groups to exceed their target biomasses, and therefore errs on the side of caution.
The SB algorithm was able to reconfigure the ecosystem to resemble a specific goal, in this case the historic 1950 ecosystem (Fig. 3). Principal component analysis (PCA) summarized the similarity of the end-state functional group biomass vector with the target ecosystem configuration after a 50-year restoration plan. Effective ecosystem reconfiguration was possible when the biomasses of all species groups were mandated for adjustment (Fig. 3a). The PCA recognized changes in biomass so only plans optimized under the biomass unit of improvement achieved results. When fewer groups were mandated for restoration, the distinction between the biomass and per cent units of improvement became less important since the mode of action was similar for optimal fishing policies using both metrics. Fishing solutions therefore converged (Fig. 3b).
Figure 3 Principal component analysis showing ecosystem configurations after restoration. Large square represents the initial ecosystem configuration (2000); large circle represents the goal configuration (historic 1950 ecosystem); small closed circles show harvest simulations optimized for biomass recovery (unit of improvement is biomass); small open circles show additional simulations (per cent and mixed unit of improvement). (a) All groups mandated for restoration. (b) Commercial groups mandated only. Plots have been rotated to reveal restoration success along the x-axis.
The end-state functional group biomasses following restoration lay very close to their target levels of the historical 1950 ecosystem (Fig. 4). When only commercial groups were mandated for recovery, biomass closely approached the target biomass. Fewer conflicting dynamics needed to be resolved in an optimization concerning fewer groups. Of the eleven commercial functional groups mandated for recovery, the restoration process brought biomass levels to within 15% of target, on average. When all groups were mandated for recovery, groups approached within 25% of target biomass on average.
Figure 4 End-state group biomass after rebuilding relative to target 1950 goal biomass. Thin line shows 1950 goal biomass (defined as 1); thick line shows initial 2000 biomass (left) and restoration end-state biomass (right). (a) All species groups mandated for restoration. (b) Commercial species groups mandated only. Biomass values are aggregated across species groups. Radial scale is linear. Comm. Inv. = commercial invertebrates, Other Inv. = other invertebrates.
Cost-benefit analysis of restoration
Restoration scenarios that were directed entirely by the SB objective (WREB = 1, WECON = 0; Eq. 4) achieved the best restoration possible without concern for economic costs and benefits. However, by incrementing the relative weight of the economic objective, we produced a spread of restoration plans that placed increasing emphasis on the profitability of fisheries throughout the restoration process (Fig. 5). All of the fishing policies achieved a prescribed level of restoration measured as a reduction in model residuals versus the goal configuration (Fig. 5). Plans with a strong economic directive concentrated rebuilding efforts on the most valuable species groups (Fig. 5, top left), while plans with less concern for harvest benefits affected a greater number of groups resulting in higher end-state biodiversity. Biodiversity was not an explicit objective of this policy search, although that facility exists (see Kempton's Q index modified for EwE; Ainsworth & Pitcher Reference Ainsworth and Pitcher2006). Nevertheless, biodiversity increased as the ecosystem approached the goal. There was a convex relationship between profit and biodiversity, so restoration scenarios that left the ecosystem in the middle of this range achieved an advantageous compromise.
Figure 5 End-state profit and biodiversity of restoration plans (n = 39) targeting the historic 1950 ecosystem based on random initialization of fishing effort. The optimal equilibriums were calculated using various weightings in the objective function for ecosystem restoration and economic performance. All species groups are considered in the restoration objective; plans use biomass as the unit of improvement, employ a linear marginal improvement valuation model and apply the maxdex fishing fleet. Contours show reduction in sum of squares; peak represents 65% reduction. Initial ecosystem values: profit $480·km−2·yr−1; biodiversity 7.4.
NPV of restoration to the economic ORB ecosystem using the lost valley fleet and the BC fleet rebuilt the ecosystem to a minimum specified degree measured as a reduction in residuals versus target (Fig. 6). Scenarios used a spread of weightings on the economic and SB rebuilding objectives. When an economic objective was included, the rebuilding took longer to achieve the specified reduction in residuals but produced more fishery profits annually (Fig. 6). Costs of restoration were spread out over time. NPV was calculated assuming a standard discount rate of δ = 5%, similar to long-term bank interest (Government of Canada benchmark bond yielded 4.29% in April 2007, see URL http://www.bankofcanada.ca; Fig. 6). When NPV = 0, scenarios involving the lost valley fleet, which has less bycatch than the BC fleet and is able to fine-tune biomass proportions more precisely and with less waste, outperformed bank interest as an investment in natural capital (Fig. 6).
Figure 6 NPV of restoration plans achieving a minimum reduction in residuals versus goal ecosystem. Each scenario includes a restoration and harvest phase. Discount rate (δ = 5%. Target ecosystem is the 1950 economic ORB. Fleets used are: BC fleet (closed circles) and the lost valley fleet (open circles). Criterion for successful restoration is a 20% reduction in the sum of squares versus commercial species groups; this level was chosen to provide a maximum spread in points along the x-axis. Plans optimized for restoration achieve the SS reduction criterion quickly, but plans that include an economic objective achieve restoration more slowly and maintain higher profit during the rebuilding phase.
Varying restoration time
The NPV of restoration plans that targeted the economic ORB ecosystem using the BC fleet and the lost valley fleet included various weightings on the economic objective and the SB rebuilding objective, but all achieved some degree of improvement versus the target ecosystem according to a least squares criterion (Fig. 7). Longer restorations plans were able to achieve a closer match to the target ORB system, and so could deliver a greater sustained profit once the new equilibrium is reached. Using these simple equilibrium level optimal fishing mortalities to drive the simulations, most of the restoration benefit occurred within the first 15–20 years; the system then reached ‘restored’ biomass equilibrium according to the imposed optimal fleet-effort pattern, and there was little benefit in extending the restoration phase (α). Even with the unmodified BC fleet, conservative restoration plans outperformed status quo profit, but the lost valley fleet, which had less bycatch, offered the most cost-effective restoration scenarios.
Figure 7 Net present value of restoration scenarios for (a) BC fleet and (b) lost valley fleet. Goal ecosystem is economic ORB. Restoration scenarios last 100 years and include a rebuilding phase (α) of 5, 7.5, 10, 15, 20 or 30 years and an equilibrium harvest phase lasting until year 100. The profit stream is evaluated using a 5% discount rate. Broken line indicates status quo NPV (profits in year 2000 extended to year 100). Optimizations use biomass as the unit of improvement, and employ a linear marginal improvement valuation model, which provides the best conditions for profitability. Various weightings on economic and SB criteria provide a spread of possible restoration plans. The vertical spread of points along the Y-axis consists of runs optimized for economic returns at top, and runs optimized for ecosystem rebuilding at bottom.
DISCUSSION
Ecological limits to restoration
We have demonstrated application of the new restoration tool using restoration goals based on historic ecosystems. Historical states may be suitable objectives for ecological and social reasons, but many additional factors must be considered before any real-world application. Chronic impacts from overfishing may have compromised the ability of the ecosystem to recover in the short term. If trophic energy flows up the food web have become more linear and simplified because of fishing (Pauly et al. Reference Pauly, Christensen, Dalsgaard, Froese and Torres1998, Reference Pauly, Christensen, Guénette, Pitcher, Sumaila, Walters, Watson and Zeller2002), then the energy budget of coastal marine ecosystems may no longer support a broad diversity of specialized predators. In addition, if directional climate change has occurred since historic times, then restoration policies directed towards historic states will be fighting a natural shift in the assemblage. Concerns have also been raised regarding the reduced fitness of populations owing to founder effects, evolutionary change in response to fishing, irreversible species introductions and persistent changes in environmental regimes (such as nutrients or hydrography) (Pitcher 2005). Extinction problems are also recognized, as is the loss of locally adapted populations and keystone species (Pitcher Reference Pitcher2005). These issues present tangible ecological and environmental limits to restoration, some insurmountable, thus we have restricted the present study to scenarios targeting restoration goals based on relatively recent historical conditions. A trophic ecosystem modelling approach may evaluate many of these issues, but additional modelling tools are needed.
At the time of this study, the EwE policy search routine was limited to determining only equilibrium-level optimal fishing mortalities owing to an error in the way that parameters were scaled when estimating the numerical derivatives in the Fletcher-Powell optimization routine (C. Walters, UBC Fisheries Centre, personal communication 2006). All the optimizations we conducted therefore used a ‘single block’ optimization, namely a single fishing mortality assigned to each gear type and held constant throughout all simulation years. Future work could explore the use of complex multi-staged restoration plans. Non-linear and hysteretic change may prevent a complex marine system from reverting to its wilderness state once fishing pressure is removed (Scheffer et al. Reference Scheffer, Carpenter, Foley, Folke and Walker2001; Hughes et al. Reference Hughes, Bellwood, Folke, Steneck and Wilson2005; Duarte et al. Reference Duarte, Conley, Carstensen and Sánchez-Camacho2009). However, the SB algorithm may identify complex restoration schemes, and advance an alternative stable state desirable from a policy perspective.
If fisheries activity or food web dynamics maintain a current persistent state, then we might develop recovery scenarios that use direct and indirect trophic effects to promote changes in the assemblage; for example by suppressing predators whose compensatory feeding habits help maintain current ecosystem structure. However, changes in nutrification, flow patterns or other external environmental factors (as may be induced by humans or as the result of natural variation) could sabotage restoration efforts through separate and synergistic activity (Duarte et al. Reference Duarte, Conley, Carstensen and Sánchez-Camacho2009). Such environmental factors may be incorporated into EwE through use of simple time-forcing functions, but calibrating the system response against climate models (as per the Intergovernmental Panel on Climate Change scenarios) would require special care. A preliminary effort made with the current model confirmed non-linear dynamics in fisheries value and species biomass with decreasing nutrient availability. However, model-coupling facilities in EwE (version 6) will greatly increase our ability to address these non-linear dynamics: to place confidence limits on predictions of ecosystem structure following restoration, or even incorporate climate effects explicitly in the policy optimization yielding more precautionary management targets.
With these new tools, it may be possible to test regime-shift hypotheses that explain the recovery failure of Atlantic Northern cod (for example relating to environmental changes, mammal predation or poaching/bycatch: Rice & Rivard Reference Rice and Rivard2003; or depensation effects: Walters & Kitchell Reference Walters and Kitchell2001). In the case of northern BC, a pre-industrial assemblage may only be achieved through careful alteration of keystone groups and maintenance of key trophic interactions, such as the sea otter-kelp-urchin triad (Estes & Duggins Reference Estes and Duggins1995; Steneck et al. Reference Steneck, Graham, Bourque, Corbette, Erlandson, Estes and Tegner2002). We can expect idiosyncrasies specific to each ecosystem (Duarte et al. Reference Duarte, Conley, Carstensen and Sánchez-Camacho2009) and there are clearly large uncertainties involved, some of which can be evaluated in the EwE framework. In the BC example, uncertainties surrounding the density dependent feeding rates of sea otters and urchins can be evaluated (after Mackinson et al. Reference Mackinson, Blanchard, Pinnegar and Scott2003), as can dependence of juvenile fish on biogenic structure (see for example Okey et al. Reference Okey, Vargo, Mackinson, Vasconcellos, Mahmoudi and Meyer2004; Ainsworth et al. Reference Ainsworth, Varkey and Pitcher2008b). However, from a policy perspective, scientific uncertainty may not be sufficient grounds to delay restoration attempts; successful reintroductions of sea otters have already occurred in BC and elsewhere in the north-east Pacific (Riedman & Estes Reference Riedman, Estes, Van Blarian and Estes1998; Lance et al. Reference Lance, Richardson and Allen2004).
Cost-benefit analysis
There is potential to improve the bioeconomic model used by the policy search routine in this paper; for instance, the malleability of fishing capital should be carefully considered. Some fisheries scientists view overcapacity and overcapitalization as the single greatest threat to the long-term viability of fish stocks (Mace Reference Mace, Hancock, Smith, Grant and Beumer1997; Gréboval & Munro Reference Gréboval and Munro1999; Ward et al. Reference Ward, Brainerd, Freese, Mace, Milazzo, Squires, Terry, Thunberd, Travis and Walden2001). However, here we assume that there is no cost or penalty associated with fleet restructuring or decommissioning. Ongoing developments in the EwE framework, such as a recent effort to integrate a new fleet buy-back scheme into the optimization procedure (Cheung & Sumaila Reference Cheung and Sumaila2008), may permit better assessments as to the economic viability of whole ecosystem restoration.
In the economic analysis here, costs and benefits in each year were considered in relation to the status quo profit, which was the estimated fisheries profit for 2000. The profit, we assume, will remain constant over the next 50 years. The implicit assumption is that the levels of fishing mortality in northern BC are sustainable. There may be reason to doubt this where declining species are concerned, notably rockfish and other valuable demersal fish. We are therefore using a generous estimate of baseline profit. If the current level of profit actually declines over the next 50 years, the benefits of restoration would have been underestimated here, and consequently the NPV of the restoration plan.
The rebuilding plans (Fig. 6) allow some level of fishing to occur throughout the rebuilding process. This important feature could make restoration more socially acceptable to stakeholders than an approach in which fishing pressure is completely eliminated to allow quick rebuilding (see Clark & Munro Reference Clark and Munro1975) since we are not strictly eliminating fishing effort in these rebuilding plans. We are instead using selective fishing as a tool, applying fishing effort for example to remove competitors of depressed species and facilitate the growth of prey. Strategic application of fishing effort should out-perform a blanket policy of fishing cessation, and achieve faster and more economical growth of desired species (Ainsworth Reference Ainsworth2006). Moreover, economic externalities may slow down the optimal rate of rebuilding in real applications, as would the encumbering species interactions that can now be explicitly managed using this new EBFM tool.
CONCLUSIONS
Development of tools to predict marine ecosystem dynamics continues to advance the science behind restoration ecology, but the level of coordination that must be achieved among all stakeholders, and even among marine scientists, may prove to be a greater challenge to the restoration agenda than technical requirements. Ecosystem restoration is comprehensive by definition, and restoration policies will necessarily enlist many industrial and scientific partners into the process of restoring the natural system. Scientists from diverse disciplines such as ecology, sociology and economics will need to work together to breakdown barriers that have so far divided marine science into disjointed sectors (Rosenberg & McLeod Reference Rosenberg and McLeod2005). Industry especially must be committed to the task, since any tenable restoration plan is likely to impact the short-term profitability of fisheries in favour of long-term sustainability. In this respect, scenarios that allow profit during restoration might more politically palatable.
ACKNOWLEDGEMENTS
We gratefully acknowledge the following people for helpful comments on the specific biomass objective function and implementation in EwE: Villy Christensen, Carl Walters, Daniel Pauly, Rashid Sumaila (UBC Fisheries Centre), Les Lavkulich (UBC Institute for Resources, Environment and Sustainability) and Alan Sinclair (Department of Fisheries and Oceans).
