Category Archives: Fisheries Think Tank

Think Tank: Murdoch McAllister

Developing management options to deal with spasmodic recruitment in Atlantic redfish
Murdoch McAllister, UBC Institute for the Oceans and Fisheries
Tuesday, February 3rd, 2025, 9:30AM

Virtual

Webinar Link: https://attendee.gotowebinar.com/register/8831598901906187862

Both short-lived and long-lived fishes are known to exhibit so-called spasmodic recruitment, a population dynamics pattern in fisheries characterized by long, consecutive periods of low recruitment interrupted by infrequent, irregular, and intense pulses of high recruitment. For up to several decades such fish stocks may remain at low abundance. Very large recruitment events typically arrive unanticipated and come as a surprise to both industry and managers. Spasmodic behaviors have been observed recently on both coasts of North America, for example, in Pacific Ocean Bocaccio Rockfish, California Sardine and Anchovy, and Canadian Atlantic redfish and capelin. Given their highly unpredictable behaviors, spasmodic fish stocks present substantial challenges and also new opportunities for the design of fishery management plans for fisheries for such stocks when large recruitment events are confirmed. I summarize findings from my research group’s study of Canadian Atlantic redfish on characterizing stock trends and evaluating candidate management procedures.

Think Tank: James Thorson

Tutorial for adding nonstationarity, nonlinearity, and statistical interactions to dynamic structural equation models
James Thorson, NOAA Northwest Fisheries Science Center
Tuesday, January 13th, 2025, 9:30AM

FISH 203

Webinar Link: https://attendee.gotowebinar.com/register/8831598901906187862

Dynamic structural equation models (DSEM) provide a fast and general approach to incorporate ecological knowledge in multivariate time-series analysis, e.g., when linking covariates to stock-assessment models. However, DSEM has previously been restricted to stationary, linear, and additive relationships, whereas many ecological systems are nonstationary, nonlinear, and include statistical interactions.
In this workshop, we introduce “moderated SEM” and quickly summarize how moderated DSEM can incorporate nonstationarity, nonlinearity, and statistical interactions. We will then walk through three R-package vignettes in package dsem, showing (1) random slopes linking the Pacific Decadal Oscillation to local temperature, (2) nonlinear (Lotka-Volterra) predator-prey interactions for Wolf-Moose or Paramecium-Didinium systems, and (3) quadratic temperature-
dependence in zooplankton dynamics for Lake Washington. We encourage participants to have a laptop with R and Rtools installed, and/or pre-installing the dsem@dev branch using `remotes::install_github(“james-thorson-NOAA/dsem@dev”)`.

Think Tank: Maia Kapur

Deep Fish: Neural Network Models for the Coming Decades of Fisheries Science and Management
Maia Kapur, UW School of Aquatic and Fisheries Sciences Alum
Tuesday, December 2nd, 2025, 9:30AM

FISH 203

Webinar Link: https://attendee.gotowebinar.com/register/6627634441098973530

ABSTRACT TBD

Think Tank: Michael Kinneen

MASE-Based Model Selection and Misspecification Detection in Stock Assessment Models
Michael Kinneen, University of Washington
Tuesday, October 28th, 2025, 9:30AM

FISH 203

Webinar Link: https://attendee.gotowebinar.com/register/4179512428227788384

ABSTRACT TBD

Think Tank: Andre Punt

Best practice for estimating natural mortality with reference to fish stocks off southeast Australia
Andre Punt, University of Washington
Tuesday, October 14th, 2025, 9:30AM

FISH 203

Webinar Link: https://attendee.gotowebinar.com/register/334591734150911065

Natural mortality (M) is a key parameter in age- and size-structured methods of fish stock assessment because estimates of biomass in absolute terms and relative to reference points are sensitive to its value. M can be pre-specified based on “indirect” methods, estimated with a prior, and estimated without a prior. However, there is an absence of best practice guidelines for how to treat M within stock assessments. Five alternative broad categories of methods for treating M in stock assessments (unconstrained estimation, estimation with a prior, the “lowest plausible” and “highest plausible” values based on indirect methods, and the results of the Hamel-Amax indirect method) are compared for ten stocks in Australia’s Southern and Eastern Scalefish and Shark Fishery using likelihood profiles, retrospective analyses and hindcast skill. There is no method that performs best in all cases. The results support a best practice where estimation with a prior should be the default unless diagnostics suggest that the population dynamics or the observation model is clearly mis-specified (e.g., an estimate of M that differs markedly from the mean of a prior based on longevity information). It is also best practice to conduct sensitivity analyses and use decision tables to highlight the effects of incorrectly assumed values of M when mis-specification is suspected and M is pre-specified using a longevity-based method.

Think Tank: Caitlin Stern

Impacts of Spatiotemporal Index Standardization on the Stock Assessment of Norton Sound Red King Crab
Caitlin Stern, Alaska Department of Fish and Game
Tuesday, September 30th, 2025, 9:30AM

FISH 203

Webinar Link: https://attendee.gotowebinar.com/register/8919857799779167325

Indices of abundance are key inputs to stock assessment models, influencing model outputs that shape fisheries management decisions. Design-based methods for producing these indices from fishery-independent surveys come with potential drawbacks, including highly variable estimates when spatial dependence in survey catch is present yet excluded from calculations, and biased results when survey sampling deviates from survey design. Implications for stock assessment models include the potential for better predictive ability when using model-based rather than design-based indices. The stock assessment for red king crab (Paralithodes camtschaticus) in Norton Sound, Alaska, uses design-based indices of abundance based on three fishery-independent surveys. These surveys cover different years and areas, and within a survey the locations sampled sometimes varies among years, suggesting a need for spatiotemporal standardization to improve abundance estimation. Here, I evaluate the effects on key outputs of the stock assessment model of using design-based survey indices versus a spatiotemporally standardized model-based index of abundance. Fits to size composition data were similar for models using design-based versus model-based indices, while fits to standardized fishery catch per unit effort indices were poorer for models using model-based indices. Retrospective patterns were less extreme for models using the spatiotemporal model-based index, indicating an improved ability to predict stock biomass in the terminal year. The estimated scale of the population was lower across the time series when using the model-based index. While estimated stock status was comparable or higher when using the spatiotemporal model-based index, model-estimated abundance was reduced, as was recommended harvest.

Think Tank: Cole Monnahan

Title TBD
Cole Monnahan, NOAA Alaska Fishery Science Center
Tuesday, January 7th, 09:30 AM PST

FISH 113

Diagnosing common sources of lack of fit to composition data in fisheries stock assessment models using One-Step-Ahead (OSA) residuals

Modern fisheries stock assessments rely on multiple sources of data to estimate population parameters and management quantities. Age and length data provides demographic information on recruitment strengths and mortality rates and is commonly fit as numbers- or proportions-at-age or -length. These data are inherently correlated among categories (bins) and residuals are therefore not independent. One-Step-Ahead (OSA) residuals have been proposed as a statistically valid replacement for the commonly used (but incorrect) Pearson residuals. However, there is no clear best practice for how to efficiently evaluate and diagnose model fit when using OSA residuals. We begin by providing an intuitive introduction to OSA residuals for multinomial data, including why Pearson residuals are incorrect and can mislead and why OSA residuals are correct statistically. We then demonstrate how OSA residuals are calculated and why this leads to some counterintuitive properties. Next, we illustrate a series of cases with and without sources of model misspecification common in stock assessments based on age compositions from a simplified fished population. For each case, we provide a comparison of visual and quantitative diagnostics based on the properties of OSA residuals. Model misspecification was identifiable in all cases using a combination of statistical analysis of OSA residuals, aggregate fits, and/or visual inspection of (scaled) Pearson residuals. However, the statistical power to detect minor model misspecification depends on the sample size, the number of bins, and the number of years of data. Analysts accustomed to visual inspection of Pearson residuals may have to adjust their interpretation for OSA residuals, and adopt new diagnostic tools to be able to identify common types of model misspecification properly. By illustrating common problems and accurate interpretation when the correct answer is known, we provide guidelines for model diagnostics using OSA residuals that can serve as a starting point for evaluating more complex operational stock assessments.

Webinar Link: https://attendee.gotowebinar.com/register/8099446196859219552

ABSTRACT TBD