Model averaging to streamline the stock assessment process

Millar, Colin P.; Jardim, Ernesto; Finlay, Scott; Osio, Giacomo Chato; Mosqueira, Iago; Alzorriz, Nekane

doi:10.1093/icesjms/fsu043

Cited by 18 publications

(8 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most common of these are the so-called "information criteria" metrics, such as the Akaike information criterion (AIC; Burnham and Anderson, 2002), which aim at finding a balance between the goodness-of-fit and the complexity of the model. Such approaches have recently made their way into marine science, and are being used regularly in, for example, stock-assessment (Millar et al, 2015;Ianelli et al, 2015). Cross-validation techniques have also been used extensively in the marine literature and are closely related.…”

mentioning

confidence: 99%

Uncertainties in projecting climate-change impacts in marine ecosystems

Payne

Barangé

Cheung

et al. 2015

ICES Journal of Marine Science

142

112

View full text Add to dashboard Cite

Projections of the impacts of climate change on marine ecosystems are a key prerequisite for the planning of adaptation strategies, yet they are inevitably associated with uncertainty. Identifying, quantifying, and communicating this uncertainty is key to both evaluating the risk associated with a projection and building confidence in its robustness. We review how uncertainties in such projections are handled in marine science. We employ an approach developed in climate modelling by breaking uncertainty down into (i) structural (model) uncertainty, (ii) initialization and internal variability uncertainty, (iii) parametric uncertainty, and (iv) scenario uncertainty. For each uncertainty type, we then examine the current state-of-the-art in assessing and quantifying its relative importance. We consider whether the marine scientific community has addressed these types of uncertainty sufficiently and highlight the opportunities and challenges associated with doing a better job. We find that even within a relatively small field such as marine science, there are substantial differences between subdisciplines in the degree of attention given to each type of uncertainty. We find that initialization uncertainty is rarely treated explicitly and reducing this type of uncertainty may deliver gains on the seasonal-to-decadal time-scale. We conclude that all parts of marine science could benefit from a greater exchange of ideas, particularly concerning such a universal problem such as the treatment of uncertainty. Finally, marine science should strive to reach the point where scenario uncertainty is the dominant uncertainty in our projections.

show abstract

mentioning

confidence: 99%

Uncertainties in projecting climate-change impacts in marine ecosystems

Payne

Barangé

Cheung

et al. 2015

ICES Journal of Marine Science

142

112

View full text Add to dashboard Cite

show abstract

“…We focused on selecting a single model through this process, rather than a model averaging approach (Millar et al, 2015), because we wanted a tool that would efficiently explore managers' questions about ecosystem scenarios and trade-offs, as well as one that could be updated with new data regularly. Maintaining multiple models with different input streams through a model-averaging approach would make this tool less efficient in practice.…”

Section: Discussionmentioning

confidence: 99%

Balancing Model Complexity, Data Requirements, and Management Objectives in Developing Ecological Reference Points for Atlantic Menhaden

et al. 2021

View full text Add to dashboard Cite

Atlantic menhaden is an important forage fish and the target of the largest fishery along the US East Coast by volume. Since 1999, managers at the Atlantic States Marine Fisheries Commission, stakeholders, and scientists have been interested in developing ecological reference points (ERPs) that account for menhaden’s role as a forage species. To accomplish this, we developed a suite of modeling approaches that incorporated predation on menhaden and changes in productivity over time and allowed for evaluation of trade-offs between menhaden harvest and ecosystem management objectives. These approaches ranged in complexity, from models with minimal data requirements and few assumptions to approaches with extensive data needs and detailed assumptions. This included a surplus production model with a time-varying intrinsic growth rate, a Steele-Henderson surplus production model, a multispecies statistical catch-at-age model, an Ecopath with Ecosim (EwE) model with a limited predator and prey field, and a full EwE model. We evaluated how each model could address managers’ objectives and compared outputs across the approaches, highlighting their strengths, weaknesses, and management utility. All models produced estimates of age-1 + biomass and exploitation rate that were similar in trend and magnitude to the single-species statistical catch-at-age model, especially in recent years. While the less complex models were relativity easy to implement and update, they lacked key elements needed to manage multiple species simultaneously. More complex models required a wider array of data and were more difficult to update within the current management time-frames, but produced a more useful framework for managers. Ultimately, an EwE model of intermediate complexity coupled with the existing single-species assessment model was recommended for use in management.

show abstract

“…Integrating across multi-model results has several advantages for the stock assessment process [ 7 ]. For example, the choice of natural mortality model was shown to have a strong impact on the estimates of recruitment and fishing mortality.…”

Section: Discussionmentioning

confidence: 99%

“…It can be thought of as a model-weighting algorithm where the weights are based on the support for the model in the data and where each model represents a different, plausible hypotheses. A variety of model averaging approaches have been proposed: frequentist and Bayesian, simple and complex [ 7 ]. One of the key questions is how to weight the models when averaging over them [ 8 ].…”

Section: Introductionmentioning

confidence: 99%

An Applied Framework for Incorporating Multiple Sources of Uncertainty in Fisheries Stock Assessments

et al. 2016

Self Cite

View full text Add to dashboard Cite

Estimating fish stock status is very challenging given the many sources and high levels of uncertainty surrounding the biological processes (e.g. natural variability in the demographic rates), model selection (e.g. choosing growth or stock assessment models) and parameter estimation. Incorporating multiple sources of uncertainty in a stock assessment allows advice to better account for the risks associated with proposed management options, promoting decisions that are more robust to such uncertainty. However, a typical assessment only reports the model fit and variance of estimated parameters, thereby underreporting the overall uncertainty. Additionally, although multiple candidate models may be considered, only one is selected as the ‘best’ result, effectively rejecting the plausible assumptions behind the other models. We present an applied framework to integrate multiple sources of uncertainty in the stock assessment process. The first step is the generation and conditioning of a suite of stock assessment models that contain different assumptions about the stock and the fishery. The second step is the estimation of parameters, including fitting of the stock assessment models. The final step integrates across all of the results to reconcile the multi-model outcome. The framework is flexible enough to be tailored to particular stocks and fisheries and can draw on information from multiple sources to implement a broad variety of assumptions, making it applicable to stocks with varying levels of data availability The Iberian hake stock in International Council for the Exploration of the Sea (ICES) Divisions VIIIc and IXa is used to demonstrate the framework, starting from length-based stock and indices data. Process and model uncertainty are considered through the growth, natural mortality, fishing mortality, survey catchability and stock-recruitment relationship. Estimation uncertainty is included as part of the fitting process. Simple model averaging is used to integrate across the results and produce a single assessment that considers the multiple sources of uncertainty.

show abstract

Model averaging to streamline the stock assessment process

Cited by 18 publications

References 18 publications

Uncertainties in projecting climate-change impacts in marine ecosystems

Uncertainties in projecting climate-change impacts in marine ecosystems

Balancing Model Complexity, Data Requirements, and Management Objectives in Developing Ecological Reference Points for Atlantic Menhaden

An Applied Framework for Incorporating Multiple Sources of Uncertainty in Fisheries Stock Assessments

Contact Info

Product

Resources

About