Dealing with missing covariate data in fishery stock assessment models

Covariates are now commonly used in fisheries stock assessment models to provide additional information about model parameters, but their use can be complicated by missing values. A wide range of covariates have been used (e.g. environment, disease, predation, food, pollutants) to model different processes (e.g. recruitment, natural mortality, growth, catchability). Several approaches are available to deal with missing covariate values. We illustrate a likelihood based approach to deal with missing covariate data when including covariates into fisheries stock assessment models. The method treats the missing covariate values as parameters from a random effects distribution. The parameters of the random effects distribution are estimated based on the observed values of the covariate. The true likelihood is implemented by integrating across the missing value random effect and, in our stock assessment example, a random effect for unexplained variation in recruitment using Laplace approximation. Simulation analysis is used to test the performance of the method and compare it to alternative approaches: (1) ignoring the covariate altogether, (2) ignoring the years with missing covariate values, (3) substituting the missing values with the mean of the observed values, and (4) estimating the missing values as free parameters. We apply the simulation analysis to a linear regression and a statistical catch-at-age stock assessment model. The simulation analysis results indicate that the random effects method for dealing with missing covariate data works moderately well, but it does not provide a substantial benefit over other less complex methods.     

Assessing the green sea urchin (Strongylocentrotus drobachiensis) stock in Maine, USA

The green sea urchin (Strongylocentrotus drobachiensis) fishery is of great importance to Maine’s economy. The fishery took off in the late 1980s as a result of expanding export markets, but has experienced substantial decline in landings since 1992 because of large decreases in urchin stock abundance. Fishery-independent surveys have not been conducted, and no formal stock assessment has been done prior to this study. Using the data collected from the fishery and urchin life history parameters derived from scientific studies, we conducted a formal stock assessment for the urchin stock. A stochastic observation-error length-structured model is used to describe the dynamics of the sea urchin population. A robust Bayesian approach is used for estimating fishery parameters because of concerns of possible outliers in fisheries data and mis-specification of priors. This study shows that the current stock is only 10% of the virgin stock biomass and that the exploitation rate is close to 40% suggesting that a large reduction in exploitation rate is necessary.     

Using the best of two worlds: A bio-economic stock assessment (BESA) method using catch and price data

Reliable stock assessments are essential for successful and sustainable fisheries management. Advanced stock assessment methods are expensive, as they require age- or length-structured catch and detailed fishery-independent data, which prevents their widespread use, especially in developing regions. Furthermore, modern fisheries management increasingly includes socio-economic considerations. Integrated ecological-economic advice can be provided by bio-economic models, but this requires the estimation of economic parameters. To improve accuracy of data-limited stock assessment while jointly estimating biological and economic parameters, we propose to use price data, in addition to catches, in a new bio-economic stock assessment (‘BESA’) approach for de-facto open access stocks. Price data are widely available, also in the Global South. BESA is based on a state-space approach and uncovers biomass dynamics by use of the extended Kalman filter in combination with Bayesian estimation. We show that estimates for biological and economic parameters can be obtained jointly, with reliability gains for the stock assessment from the additional information inherent in price data, compared to alternative assessment methods for data-poor stocks. In a real-world application to Barents Sea shrimp (Pandalus borealis, Pandalidae), we show that BESA benchmarks well also against advanced stock assessment results. BESA can thus be both a stand-alone approach for currently unassessed stocks as well as a complement to other available methods by providing bio-economic information for advanced fisheries management.     

Estimating uncertainty in fish stock assessment and forecasting

A variety of tools are available to quantify uncertainty in age‐structured fish stock assessments and in management forecasts. These tools are based on particular choices for the underlying population dynamics model, the aspects of the assessment considered uncertain, and the approach for assessing uncertainty (Bayes, frequentist or likelihood). The current state of the art is advancing rapidly as a consequence of the availability of increased computational power, but there remains little consistency in the choices made for assessments and forecasts. This can be explained by several factors including the specifics of the species under consideration, the purpose for which the analysis is conducted and the institutional framework within which the methods are developed and used, including the availability and customary usage of software tools. Little testing of either the methods or their assumptions has yet been done. Thus, it is not possible to argue either that the methods perform well or perform poorly or that any particular conditioning choices are more appropriate in general terms than others. Despite much recent progress, fisheries science has yet to identify a means for identifying appropriate conditioning choices such that the probability distributions which are calculated for management purposes do adequately represent the probabilities of eventual real outcomes. Therefore, we conclude that increased focus should be placed on testing and carefully examining the choices made when conducting these analyses, and that more attention must be given to examining the sensitivity to alternative assumptions and model structures. Provision of advice concerning uncertainty in stock assessments should include consideration of such sensitivities, and should use model‐averaging methods, decision tables or management procedure simulations in cases where advice is strongly sensitive to model assumptions.     

Integrating recreational fisheries data into stock assessment: implications for model performance and subsequent harvest strategies

Understanding the impacts of recreational fishing on commercially fished stocks is becoming increasingly relevant for fisheries managers. However, data from recreational fisheries are not commonly included in stock assessments of commercially fished stocks. Simulation models of two assessment methods employed in Australia's Commonwealth fisheries were used to explore how recreational fishery data can be included, and the likely consequences for management. In a data‐poor management strategy for blue eye trevalla, Hyperoglyphe antarctica (Carmichael), temporal trends in recreational catch most affected management outcomes. In a data‐rich age‐structured stock assessment for striped marlin, Kajikia audax (Philippi), estimates of stock status were biased when recreational catches were large or when the recreational fishery targeted different size classes than the commercial fishery and these data were not integrated into the assessment. Including data from recreational fishing can change perceptions of stock status and impact recommendations for harvest strategies and management action. An understanding of recreational fishery dynamics should be prioritised for some species.     

Generic solutions for data‐limited fishery assessments are not so simple

The majority of the world's fisheries, by number, are data‐poor/limited, and there is a growing body of literature pertaining to approaches to estimate data‐limited stock status. There are at least two drivers for assessing the status of data‐limited fisheries. The first is to try to understand and report on the global or regional status of fisheries across many stocks. The second is to attempt to assess individual data‐limited stocks, for status reporting and/or guiding management decisions. These drivers have led to attempts to find simple, generic, low‐cost solutions, including the broad application of generically parameterised models, and the blanket application of a single, or limited number of possible, analytical approach(es). It is unclear that generic methods function as intended, especially when taken out of their original design context or used without care. If the intention is to resolve individual stock status for the purposes of management, there is concern with the indiscriminate application of a single method to a suite of stocks irrespective of the particular circumstances of each. We examine why caution needs to be exercised, and provide guidance on the appropriate application of data‐limited assessment methods (DLMs). We recommend: (a) obtaining better data, (b) using care in acknowledging and interpreting uncertainties in the results of DLMs, (c) embedding DLMs in harvest strategies that are robust to the higher levels of uncertainty in the output of DLMs by including precautionary management measures or buffers and (d) selecting and applying DLMs appropriate to specific species’ and fisheries’ data and context.     

Quantifying uncertainty in the wild‐caught fisheries goal of the Ocean Health Index

Sustainability indices are proliferating, both to help synthesize scientific understanding and inform policy. However, it remains poorly understood how such indices are affected by underlying assumptions of the data and modelling approaches used to compute indicator values. Here, we focus on one such indicator, the fisheries goal within the Ocean Health Index (OHI), which evaluates the sustainable provision of food from wild fisheries. We quantify uncertainty in the fisheries goal status arising from the (a) approach for estimating missing data (i.e., fish stocks with no status) and (b) reliance on a data‐limited method (catch‐MSY) to estimate stock status (i.e., B/B_MSY). We also compare several other models to estimate B/B_MSY, including an ensemble approach, to determine whether alternative models might reduce uncertainty and bias. We find that the current OHI fisheries goal model results in overly optimistic fisheries goal statuses. Uncertainty and bias can be reduced by (a) using a mean (vs. median) gap‐filling approach to estimate missing stock scores and (b) estimating fisheries status using the central tendency from a simulated distribution of status scores generated by a bootstrap approach that incorporates error in B/B_MSY. This multitiered approach to measure and describe uncertainty improves the transparency and interpretation of the indicator and allows us to better understand uncertainty around our OHI fisheries model and outputs for country‐level interpretation and use.     

基于Bootstrap的ELEFAN方法在评估方氏云鳚群体生长参数中的应用

渔业数据有限性是小型渔业资源评估所面临的常见问题。电子体长频率分析(electroniclengthfrequency analysis,ELEFAN)常用于年龄数据难以获取或缺失的渔业,但该方法的可靠性尚待检验。本研究根据2013―2018年春、秋季共11个航次的海州湾底拖网调查数据,分别使用传统的ELEFAN与结合Bootstrap的ELEFAN方法,比较了2013―2015年与2016―2018年两个时间段内海州湾方氏云鳚(Pholis fangi)群体von-Bertalanffy生长方程中参数之间的变化。结果显示,在海州湾海域,方氏云鳚的生长参数具有显著变化, 2013―2018年,群体的极限体长变小,生长速率加快,说明海州湾方氏云鳚群体近年来呈现小型化的趋势。相比传统的ELEFAN方法,结合Bootstrap的ELEFAN方法能够给出较为稳健的参数估计,受采样随机性的影响较小,可以较好地应用于数据缺乏的小型渔业中。本研究加深了对方氏云鳚种群动态的认识,并推动了基于体长频率的生长参数估算方法在数据有限资源评估中的应用。     

Maximum likelihood and Bayesian approaches to stock assessment when data are questionable

This study examines the use of age-structured maximum likelihood and Bayesian approaches for stock assessment of the Namibian monkfish, Lophius vomerinus, resource with questionable data, in which time series are short, abundance indices are variable, and research data conflict with commercial data. Bayesian approaches with both noninformative and informative priors are investigated to determine if they enhance estimation stability. Three data scenarios are assessed: commercial and research survey data, research survey data only, and commercial data only. Both statistical approaches show that resource abundance has decreased with exploitable biomass estimated at approximately 44% of pristine levels. The maximum likelihood and the Bayesian approach with noninformative priors result in similar estimates. As the abundance data contained little information pertaining to possible density dependence within the stock–recruit relationship, only a Bayesian approach with informative priors reduces uncertainty in the steepness parameter h. Estimated management quantities are sensitive both to the set of data sources and whether prior information was informative or not. The strengths of the Bayesian approach include the integration of prior information with uncertain data, the exploration of data conflicts, and the ability to show the uncertainty in estimates of management parameters. Its weakness is that estimation stability is dependent on the choice of priors, which alters some posterior distributions of management quantities.     

Guidelines for incorporating fish distribution shifts into a fisheries management context

The impacts of climate change have been demonstrated to influence fisheries resources. One way climate has affected fish stocks is via persistent shifts in spatio‐temporal distribution. Although examples of climate‐forced distribution shifts abound, it is unclear how these shifts are practically accounted for in the management of fish stocks. In particular, how can we take into account shifting stock distribution in the context of stock assessments and their management outputs? Here, we discuss examples of the types of fish stock distribution shifts that can occur. We then propose a decision tree framework of how shifting stock distributions can be addressed. Generally, the approaches for addressing such shifts fall into one of three main alternatives: re‐evaluate stock identification, re‐evaluate a stock unit area, or implement spatially explicit modelling. We conclude by asserting that the approach recommended here is feasible with existing information and as such fisheries managers should be able to begin addressing the role of changes in stock distribution in these fish stocks. The implications of not doing so could be notably undesirable.     

Integrating the standardization of catch-per-unit-of-effort into stock assessment models: testing a population dynamics model and using multiple data types

Standardization of catch-per-unit-of-effort (CPUE) data can be integrated into stock assessment methods. We apply this method to the stock of trevally (Pseudocaranx dentex) off the west coast of New Zealand to address: (1) whether the stock assessment model explains all of the annual variation in the CPUE data, and (2) the impact on the assessment results of how the catch-at-age data are weighted. If not all of annual variation in CPUE is explained by the stock assessment model, the assessment may be statistically inadequate. The inadequacy may be in the representation of the population dynamics, in the relationship between CPUE and abundance, or due to additional variation in CPUE left unexplained by the independent variables. Catch-at-age data often have too much influence on the estimated abundance trajectory, so the sample size used in the catch-at-age likelihood function is often reduced when applying age-structured stock assessment methods. The integrated approach automatically places more weight on the CPUE data compared to the catch-at-age data, and may therefore provide an alternative to arbitrarily downweighting catch-at-age data.     

Inland fish stock assessment: Applying data‐poor methods from marine systems

Inland fisheries can be diverse, local and highly seasonal. This complexity creates challenges for monitoring, and consequently, many inland fish stocks have few data and cannot be assessed using methods typically applied to industrial marine fisheries. In such situations, there may be a role for methods recently developed for assessment of data‐poor fish stocks. Herein, three established data‐poor assessment tools from marine systems are demonstrated to highlight their value to inland fisheries management. A case study application uses archived length, catch and catch‐per‐unit‐effort data to characterise the ecological status of an important recreational brown trout stock in an Irish lake. This case study is of specific use to management of freshwater sport fisheries, but the broader purpose of the paper was to provide a crossover between marine and inland fisheries science, and to highlight accessible data‐poor assessment approaches that may be applicable in diverse inland systems.     

基于CMSY和BSM的东海区重要渔业种类资源评估

有限数据方法(data-limited method)可结合少量易获得数据和相关生物学信息对渔业资源状况、生物学参考点以及生物量等进行评估,已经成为全球区域性渔业管理组织和资源评估学者的关注热点。本研究采用基于渔获量的最大可持续渔获量(catch-based maximum sustainable yield,CMSY)和基于贝叶斯状态空间的Schaefer产量模型(Bayesian Schaefer production model,BSM)评估了东海区19个重要经济种类的资源状况,并提出了基于最大可持续渔获量(maximum sustainable yield,MSY)的渔业管理建议。结果显示,19个种类中有1个种类衰竭,3个种类严重衰退,5个种类过度捕捞,5个种类轻度过度捕捞,5个种类健康。种群状态长期评估结果表明,处于生物可持续水平的鱼类种群占比已由1980年的95%下降至2019年的26%。同时对CMSY和BSM方法的结果进行了比较,整合单位捕捞努力量渔获量(catch per unit effort,CPUE)数据的BSM方法导致了置信区间较宽,并调节了生物量轨迹的变化形态...     

鱼类自然死亡系数评估研究进展

自然死亡系数是渔业资源评估中不可或缺的重要参数,其准确度直接决定了资源评估结果的可靠性,进而影响渔业管理策略的制定。本文从生活史参数、标志回捕和年龄结构3个方面列举了国内外自然死亡系数的常用评估方法,讨论了相关方法的优缺点及影响因素,并以犬齿牙鲆(Paralichthys dentatus)和中国近海鱼类为例对比分析不同模型的计算结果。在此基础上,着重介绍了Pauly经验公式在中国近海主要经济鱼类自然死亡系数评估中的应用进展及存在问题。根据渔业资源调查和研究数据现状,认为现阶段使用Pauly经验公式评估中国近海经济鱼类自然死亡系数具有积极作用。     

CPUE standardisation and the construction of indices of stock abundance in a spatially varying fishery using general linear models

Construction of annual indices of stock abundance based on catch and effort data remains central to many fisheries’ assessments. While the use of more advanced statistical methods has helped catch rates to be standardised against many explanatory variables, the changing spatial characteristics of most fisheries data sets provide additional challenges for constructing reliable indices of stock abundance. After reviewing the use of general linear models to construct indices of annual stock abundance, potential biases which can arise due to the unequal and changing nature of the spatial distribution of fishing effort are examined and illustrated through the analysis of simulated data. Finally, some options are suggested for modelling catch rates in unfished strata and for accounting for the uncertainties in the stock and fishery dynamics which arise in the interpretation of spatially varying catch rate data.     

Determining long‐term changes in a skate assemblage with aggregated landings and limited species data

Sustainable fisheries management requires assessment of exploited populations and communities. Traditional fisheries stock assessment methods need species‐specific input data, which for skates have only recently become available in Europe. To overcome this limitation, a Bayesian multispecies biomass production model was developed. In addition to aggregated landings, input data are short time series with species‐specific information (landings and biomass indices). Applying the approach to four main skate species and a group of two skate species, all managed together in the Bay of Biscay (Northeast Atlantic), long‐term changes in the skate assemblage composition were identified. Since the 1990s, Leucoraja naevus became increasingly dominant, while the contributions of the other three species (Raja brachyura, Raja clavata and Raja montagui) declined. The abundance of the grouped Leucoraja fullonica and L. circularis has also strongly decreased, suggesting long‐term overexploitation. All species except this species group are expected to increase over the next decade under current harvest rates. Currently, the species considered here are managed under a single fishing quota making it unlikely that the group of the two most depleted species will recover soon. The multispecies modelling approach bears promise for other harvested assemblages for which only grouped harvest information is available for certain periods.     

Ecosystems say good management pays off

Understanding the strengths and weaknesses of alternative assessment methods, harvest strategies and management approaches are an important part of operationalizing single‐species and ecosystem‐based fisheries management. Simulations run using two variants of a whole‐of‐ecosystem model for the Southern and Eastern Scalefish and Shark Fishery (SESSF) area shows that (a) data‐rich assessments outperform data‐poor assessments for target species and that this performance is reflected in the values of many system‐level ecosystem indicators; (b) ecosystem and multispecies management outperforms single‐species management applied over the same domain; (c) investment in robust science‐based fisheries management pays dividends even when there are multiple jurisdictions, some of which are not implementing effective management; and (d) that multispecies yield‐oriented strategies can deliver higher total catches without a notable decline in overall system performance, although the resulting system structure is different to that obtained with other forms of ecosystem‐based management.     

适用于数据缺乏渔业的资源评估方法研究进展

开展渔业资源评估研究是制定渔业可持续发展策略的重要前提, 而数据有限是全球渔业资源评估面临的普遍挑战。传统资源评估方法具有数据需求量大、要求高等特点, 无法应用于数据缺乏渔业的资源评估中。数据缺乏方法(data-limited method)可结合少量易获得数据和相关历史生物学信息对渔业资源状况、生物学参考点以及资源量等进行评估, 已经成为全球区域性渔业管理组织和资源评估学者的关注热点。本文在分析数据缺乏资源评估方法文献的基础上, 对近几年开发的数据缺乏评估方法模型结构、模型输入输出、假设以及不确定性来源和分析方法等进行了回顾和归纳, 同时对数据缺乏评估方法未来的研究重点和方向进行了展望。分析认为, 数据缺乏渔业资源评估研究仍处于发展阶段, 若渔获量和体长数据均可用, 则应考虑整合了两种类型数据的模型, 如 LIME (length-based integrated mixed effects, 基于体长的综合混合效应)模型和 SSS (simple stock synthesis, 简化资源整合) 模型。建议今后研究中应加强以下几个方面的工作: (1)积极开展长时间、多海域、全覆盖的渔业资源独立科学调查, 以获得具有代表性的样本数据; (2)对现有数据缺乏模型进一步优化, 综合考虑各种因素对评估结果的影响; (3) 进行完整、准确的基础生物学研究, 获得较为准确的历史生物学信息, 从而降低评估结果的不确定性; (4)开展基于渔获量模型和基于体长模型的模拟测试研究, 提高模型对统计偏差和数据质量问题的包容性。     

种群模拟在渔业资源评估中的研究现状及展望

随着渔业资源评估理论、数理统计方法和计算机技术的进步, 资源评估模型朝着多样化和复杂化不断发展, 其中种群模拟技术是检测模型适用性和局限性的重要手段。该技术由种群仿真理念发展而来, 通过模拟“真实”种群的方式, 对资源评估结果和管理策略进行有效的评价和预测, 并凭借可结合海洋环境因子、鱼类洄游空间分布以及多鱼种渔业进行资源评估的特性, 已成为开发新资源的重要评估方法之一。为此, 本文对种群模拟的结构和发展过程进行了回顾, 对该技术的核心组成部分操作模型和常见的四类误差(过程误差、观测误差、模型结构误差和管理误差)展开分类讨论。此外, 本文还结合近年来迅速发展的数据缺乏和数据适中模型的特点, 根据实际应用案例对种群模拟的作用和使用前景进行梳理, 并就种群模拟技术发展中存在的主要问题和潜在解决办法提出分析和建议。     

The assessment of fishery status depends on fish habitats

At the crux of the debate over the global sustainability of fisheries is what society must do to prevent over‐exploitation and aid recovery of fisheries that have historically been over‐exploited. The focus of debates has been on controlling fishing pressure, and assessments have not considered that stock production may be affected by changes in fish habitat. Fish habitats are being modified by climate change, built infrastructure, destructive fishing practices and pollution. We conceptualize how the classification of stock status can be biased by habitat change. Habitat loss and degradation can result in either overly optimistic or overly conservative assessment of stock status. The classification of stock status depends on how habitat affects fish demography and what reference points management uses to assess status. Nearly half of the 418 stocks in a global stock assessment database use seagrass, mangroves, coral reefs and macroalgae habitats that have well‐documented trends. There is also considerable circumstantial evidence that habitat change has contributed to over‐exploitation or enhanced production of data‐poor fisheries, like inland and subsistence fisheries. Globally many habitats are in decline, so the role of habitat should be considered when assessing the global status of fisheries. New methods and global databases of habitat trends and use of habitats by fishery species are required to properly attribute causes of decline in fisheries and are likely to raise the profile of habitat protection as an important complementary aim for fisheries management.