Elsevier

Fisheries Research

Volume 217, September 2019, Pages 46-57
Fisheries Research

Full length article
Extending integrated stock assessment models to use non-depensatory three-parameter stock-recruitment relationships

https://doi.org/10.1016/j.fishres.2017.07.007Get rights and content

Abstract

Stock assessments based on the integrated paradigm often include an underlying stock-recruitment relationship. This, along with estimates of fishery selectivity and biological parameters, allows the biomass and fishing mortality associated with Maximum Sustainable Yield (BMSY and FMSY respectively) to be calculated. However, the estimates of these quantities may differ from the proxies assumed in the harvest control rules that are used to provide management advice. Moreover, the estimated values for BMSY and FMSY are related functionally in population dynamics models based on 2-parameter stock-recruitment relationships such as the commonly used Beverton-Holt or Ricker relationships. Use of 2-parameter stock-recruitment relationships (SRRs) consequently restricts the ability to fully quantify the uncertainty associated with estimating BMSY and FMSY because 2-parameter SRRs restrict the potential range of values for BMSY/B0. In principle, BMSY/B0 and FMSY can be more independent if the stock-recruitment relationship is more general than these 2-parameter SRRs. This paper outlines eleven potential 3-parameter stock-recruitment relationships and evaluates them in terms of whether they are able to match a wide range of specifications for BMSY (expressed relative to unfished spawning stock biomass, B0) and FMSY (expressed relative to natural mortality, M). Of the eleven 3-parameter stock-recruitment relationships considered, the Ricker-Power stock-recruitment relationship is found to best satisfy the characteristics of (a) being able to mimic a wide range of BMSY/B0 and FMSY/M values, (b) not to lead to negative recruitment for biomasses between 0 and B0, and (c) not to lead to increasing recruitment while approaching the limit of zero population size. Bayesian assessments of three example groundfish species off the US west coast (aurora rockfish, petrale sole, and cabezon) are conducted using Simple Stock Synthesis based on the Beverton-Holt and Ricker-Power stock-recruitment relationships to illustrate some of the impacts of allowing for a 3-parameter stock-recruitment relationship.

Introduction

Many fishery jurisdictions conduct stock assessments in which population dynamics models are fit to monitoring data. These assessments provide various types of information for use in management. For example, they inter alia provide estimates of current spawning stock biomass, unfished spawning stock biomass, current spawning stock biomass relative to reference points, and the inputs for harvest control rules selected to provide scientific recommendations regarding limits on catch and effort. Several jurisdictions rely on estimates of the biomass and fishing mortality corresponding to Maximum Sustainable Yield (BMSY and FMSY respectively). BMSY defines the target biomass (e.g., in the USA; Punt et al., 2006), or forms the basis for calculating the target biomass (e.g., in Australia; Rayns, 2007), while the target (or limit) fishing mortality is often set equal to FMSY or some fraction thereof.

It is well known that the performances of management strategies are not very sensitive to the choice of the proxies for BMSY in harvest control rules (e.g., Clark, 1991, Clark, 2002; Hilborn, 2010, Punt et al., 2014a). Many of the “integrated” methods of stock assessment (Fournier and Archibald, 1982, Maunder, 1998, Maunder and Punt, 2013) are based on population dynamics models include an underlying stock-recruitment relationship (SRR; Dichmont et al., 2016), and can provide estimates of BMSY and FMSY. However, these estimates are rarely used when applying harvest control rules. This can be concerning to stakeholders, especially when the estimates of BMSY and FMSY from the assessment differ substantially from the proxies used to provide management advice. In principle, it is possible to parameterize population dynamics models in terms of the value of BMSY (usually expressed relative to the average unfished B0, i.e., BMSY/B0) and FMSY (expressed relative to natural mortality M) (i.e., by adopting a “leading parameters” approach (Schnute and Kronlund, 1996, Schnute and Richards, 1998, Martell et al., 2008)).

Martell et al. (2008) show how it is possible to parameterize population dynamics models based on FMSY and MSY when the SRR is of the Beverton-Holt or Ricker form. Unfortunately, simply parameterizing conventional stock assessments in terms of FMSY and MSY (see Dichmont et al., 2016 for a review of the packages on which model-based approaches to stock assessment used in the US are based) will not address several other issues related to how productivity is modelled in the population dynamics models on which stock assessments are based.

  • Most integrated stock assessments are based on the Beverton-Holt or Ricker form of the SRR (with constant recruitment a special case of a Beverton-Holt SRR).1 Unfortunately for these functional forms, specifying BMSY/B0 is the same as determining the productivity of the population as quantified by FMSY (see Fig. 1 for examples of the relationship between FMSY and BMSY/B0 for the three example species of this paper; Section 2). Thus, the use of 2-parameter SRRs restricts the potential range of values for BMSY/B0.

  • Methods of assessment wherein the population size at which MSY is achieved relative to the unfished size is pre-specified are often conducted for cetaceans (e.g., de la Mare, 1989, Punt, 1999), and the results of these assessments often suggest that BMSY/B0 (usually expressed in numbers rather than biomass) is poorly determined when attempts as made to estimate this quantity from monitoring data (e.g., Givens et al., 1995, Punt and Wade, 2012). In contrast, the use of 2-parameter SRRs, as is typically the case for assessments of fish stocks, means that it is inferred that BMSY/B0 is well determined by the data if FMSY is well-determined by the data (or that BMSY/B0 is actually pre-specified if FMSY is pre-specified).

  • Many stock assessments, particularly for data-poor scenarios, are now based on methods of stock assessment that quantify uncertainty using Bayesian methods (e.g., Simple Stock Synthesis – SSS: Cope, 2013; Depletion-Based Stock Reduction Analysis – DB-SRA: Dick and MacCall, 2011). Many assessment methods based on age- or size-structured population dynamics models that include integrated SRRs are parameterized in terms of the steepness of the SRR (h: the proportion of unfished recruitment, R0, when spawning stock biomass is reduced to 20% of B0: Francis, 1992) and some measure of unfished biomass. However, the relationship between steepness and BMSY/B0 is non-linear (Fig. 1) so placing (for example) a uniform prior on steepness is not equivalent to the placing such a prior on BMSY/B0. Moreover, the range for BMSY/B0 possible for either of the Beverton-Holt or Ricker SRRs is only a subset of the possible U[0,1] range (Punt et al., 2014a, Mangel et al., 2013; Fig. 1).

The issues outlined above suggest the need for stock assessments to include 3- as well as 2-parameter SRRs so a) there are models with no implicit relationship between BMSY/B0 and FMSY/M, b) it is possible for the population dynamics model to be parameterized in terms of any values of BMSY/B0 and FMSY/M, and c) priors can be placed on these ‘leading parameters’. This paper therefore describes thirteen potential SRRs (two 2-parameter and eleven 3-parameter SRRs), some of which have been proposed in the literature in the past. It then evaluates these relationships in terms of their capacity to achieve a broad range of values for FMSY (expressed in this paper as relative to natural mortality, and thus dimensionless) and BMSY/B0, and whether they exhibit desirable properties of SRRs, such as being defined for the entire range of possible values for B/B0. The paper finally explores the implications of conducting data-poor (catch only) assessments based on the Beverton-Holt SRR and the “best” of the eleven 3-parameter SRRs for three stocks representing contrasting life history types.

Section snippets

Example species

The analyses in this paper are based on information for three example species: aurora rockfish (Sebastes aurora), petrale sole (Eopsetta jordani), and cabezon (Scorpaenichthys marmoratus) off the US west coast. These three species cover the range from relatively short-lived and fast growing (cabezon) to very long-lived and slow growing (aurora rockfish). Table 1 lists the values for the biological parameters (growth, natural mortality and fecundity) for these species. Fishery selectivity is

Candidate stock-recruitment relationships

Table 2 lists the thirteen SRRs considered in this paper. It contains two 2-parameter SRRs (Beverton-Holt and Ricker), along with eleven 3-parameter SRRs. The SRRs are parameterized in terms of the average unfished spawning stock biomass, B0, and the steepness of the SRR. Thus, all of the SRRs in Table 2 satisfy R0 = f(B0) and hR0 = f(0.2B0), where f denotes the stock-recruitment relationship. The SRRs in Table 2 include ten forms that generalize the Beverton-Holt and Ricker SRRs by raising

Computing MSY

In equilibrium, the fully-selected fishing mortality corresponding to MSY is defined by:dC(F)dF|F=FMSY=0where C(F)=C˜(F)R(F), C˜(F) is yield-per-recruit as a function of fully-selected fishing mortality, and R(F) is recruitment when fully-selected fishing mortality2 is F (Sissenwine and Shepherd, 1987). The

Use in stock assessments

The results in the previous section suggest that the Ricker-Power SRR (and to a lesser extent the Pella-Tomlinson SRR) are able to capture a wide range of possible shapes for the yield function. The Pella-Tomlinson SRR has the undesirable feature that recruitment can increase as B → 0 (i.e., when BMSY/B0 is low). It also predicts negative recruitment for some values of B > B0 (Table 2). The Ricker-Power appears to exhibit no undesirable properties for 0 < B ≤ B0, but is undefined for all values

Selection of a SRR

This paper introduced eleven potential 3-parameter SRRs, five of which are newly formulated relationships and not found elsewhere. Six SRRs, including four of the five newly formulated relationships, were excluded because they include depensation over part of the range of biomasses between 0 and B0, and consequently there is not a one–one relationship between spawning biomass-per-recruit and recruitment, which is needed to calculate γ and h given FMSY/M and BMSY/B0. Although this problem could

Acknowledgements

AEP was partially funded by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative agreement No. NA15OAR4320063, Contribution No. 2017-097. Malcolm Haddon (CSIRO), Owen Hamel and Lewis Barnett (both NOAA Fisheries), and three anonymous reviewers are thanked for their comments on an earlier version of this paper.

References (44)

  • W.G. Clark

    F35% revisited ten years later

    N. Am. J. Fish Manage.

    (2002)
  • J.M. Cope et al.

    Status and Future Prospects for the Cabezon (Scorpaenichthys Marmoratus) as Assessed in 2003

    (2003)
  • J. Cope et al.

    Data-Moderate Stock Assessments for Brown, China Copper, Sharpchin, Stripetail, and Yellowtail Rockfishes and English and Rex Soles in 2013

    (2015)
  • D.H. Cushing

    Dependence of recruitment on parent stock in different groups of fishes

    J. Cons. Int. Explor. Mer.

    (1971)
  • D.H. Cushing

    Dependence of recruitment on parent stock

    J. Fish. Res. Board Can.

    (1973)
  • R.B. Deriso

    Harvesting strategies and parameter estimation for an age-structured model

    Can. J. Fish. Aquat. Sci.

    (1980)
  • W.K. de la Mare

    Report of the Scientific Committee. Annex L. The model used in the HITTER and FITTER programs (Program FITTER.SC40)

    Rep. Int. Whal. Comm.

    (1989)
  • D. Fournier et al.

    A general theory for analyzing catch at age data

    Can. J. Fish. Aquat. Sci.

    (1982)
  • D.A. Fournier et al.

    AD Model Builder: using automatic differentiation for statistical inference of highly parameterized complex nonlinear models

    Optim. Methods Softw.

    (2012)
  • R.I.C.C. Francis

    Use of risk analysis to assess fishery management strategies: a case study using orange roughy (Hoplostethus atlanticus) on the Chatham Rise, New Zealand

    Can. J. Fish. Aquat. Sci.

    (1992)
  • G.H. Givens et al.

    Assessment of the Bering Chukchi-Beaufort Seas stock of bowhead whales using the BALEEN II model in a Bayesian Synthesis Framework

    Rep. Int. Whal. Commn.

    (1995)
  • K.F. Johnson et al.

    Summary statistics for the comparison of the currently implemented Catch Limit Algorithm (CLA) and the variant proposed by Norway

    J. Cet. Res. Manage.

    (2016)
  • Cited by (14)

    • Stock-recruitment models from the viewpoint of density-dependent survival and the onset of strong density-dependence when a carrying capacity limit is reached

      2022, Fisheries Research
      Citation Excerpt :

      The stock-recruitment model is an essential component of fishery stock assessment models and different assumptions can have substantial impact on management advice (e.g., comparison of current status to management reference points). Many stock-recruitment models have been developed (e.g., Ricker, 1954; Beverton and Holt, 1957; Deriso, 1980; Shepherd, 1982; Schnute, 1985; Barrowman and Myers, 2000; Taylor et al., 2013; Maunder and Deriso, 2013; Punt and Cope, 2019; Mikhailov et al., 2021), but in practice only a few are commonly used. Unfortunately, stock-recruitment data are often highly variable and provide little information on the values of the parameters of the stock-recruitment model, let alone its form (e.g., Zhou, 2007).

    • Model-based estimates of reference points in an age-based state-space stock assessment model

      2020, Fisheries Research
      Citation Excerpt :

      To illustrate the general applicability of the methods, reference points were calculated for data from the 2019 Northeast Arctic cod assessment (ICES, 2019) using 12 recruitment functions. The recruitment models included two time-series models, four two-parameter models, four three-parameter models, and two spline models; thereby, representing a wide range of model classes (see, e.g., Punt and Cope, 2019 for a list of other models not considered here). Estimates for reference points that do not directly depend on recruitment (i.e., FStatus Quo, F0.1, FMax, and F35%) were practically identical for all 12 models.

    • On the dangers of including demographic analysis in Bayesian surplus production models: A case study for Indian Ocean blue shark

      2020, Fisheries Research
      Citation Excerpt :

      The OM (and the actual assessment for Indian Ocean blue shark) was based on the Beverton–Holt stock-recruitment relationship for which BMSY/K is almost always below 0.5 (Punt et al., 2014). Further work could explore alternative stock-recruitment relationships (e.g., Punt and Cope, 2019) for which the production function is more representative of long-lived species such as sharks for which BMSY/K is believed to be larger than 0.5, and hence for which the prior for Pella–Tomlinson p parameter would need to include values larger than 1. Finally, the purpose of this paper is not to denigrate the use of SPMs, particularly in data-poor situations.

    View all citing articles on Scopus
    View full text