Samuel Johnson^1
1 Simon Fraser University

Novermbet 22, 2016 9:00 (PST): FSH 203

Simulation Testing Hierarchical Multi-Stock Assessment Models for a BC Flatfish Complex

I am interested in finding the conditions that favour joint stock assessments for multiple flatfish species using hierarchical models. Hierarchical multi-level modeling approaches are being increasingly applied in stock asssessment to share information across stocks and species. These so-called “Robin Hood” approaches borrow information from data-rich to potentially improve assessments for data poor species; however, little research exists to demonstrate the boundary conditions where hierarchical approaches outperform single species models. In this talk, I present a hierarchical surplus production modeling approach for a multi-species flatfish complex in British Columbia, Canada. If applicable, the hierarchical modeling approach could help provide more up-to-date assessments for species last assessed in the 1990s!

Dr. Kerry Naish¹, Dr. Marine Brieuc¹, Charles Waters¹, Dr. Maureen Purcell^2
1SAFS
²Western Fisheries Research Center, USGS
May 17, 2016 9:00 (PST): FSH 105

The genetic basis of pathogen resistance across outbred salmon populations: a comparative analysis using Random Forest

Key questions in both aquaculture and conservation of wild populations are whether genomic regions associated with specific traits are conserved across lines or populations, and to what extent trait-linked markers might be used to predict population responses to selection. In aquaculture, the ability to improve unrelated lines using common trait-linked markers provides the industry with tools necessary to maintain several diverse breeding populations. In conservation, predicting differential population responses to anthropogenic change provides a way to evaluate different remedial actions. Infectious hematopoietic necrosis virus (IHNV) is a rhabdovirus endemic to the northeast Pacific that can cause morbidity and mortality in the commercial rainbow trout and Atlantic salmon aquaculture industries, as well as in supplementation hatcheries. We have investigated the genetic basis of IHNV resistance across different populations of steelhead salmon Oncorhynchus mykiss, in a broader effort to characterize the role of differential host response in the epidemiology of the pathogen. Using the rainbow trout 50K SNP chip, we performed genome wide association analyses across five outbred populations challenged with the pathogen. Traits within these populations may be encoded by many segregating loci of small effect, whose interactions may be influenced by dominance and epistasis. Therefore, we used Random Forest because this non-parametric approach has the advantage of identifying suites of interacting loci that explain phenotypic variation collectively, but may not have large effect sizes on their own. Here, we report several genomic regions associated with resistance and evaluate their conservation across populations. We also review the key steps, analytical approaches, and lessons we have learned in the application of Random Forest to genome wide association studies in outbred salmon populations, with a view to improving genomic resources available to the community as a whole.

Dr. Jon Reum^1
1AFSC
May 03, 2016 9:00 (PST): FSH 105

Incorporating climate in to a multispecies size spectrum model of the Eastern Bering Sea

I’ve been working with Kirstin Holsman, Kerim Aydin, and Anne Hollowed at the AFSC on developing a multispecies size-based food web model of the Eastern Bering Sea. I’ll give an overview of the model and how I’m trying to parameterize it. The model is part of a larger effort to evaluate food web and fisheries responses (and associated uncertainty) to climate change by using a variety of ecological models that vary in complexity and structural assumptions.

Dr. Carey R. McGilliard^1
1AKFSC
April 19, 2016 9:00 (PST): FSH 105

Evaluating alternative harvest strategies for Alaska groundfish fisheries based on estimates of the probability of overfishing

Increasingly, management systems require that catch limits account for scientific uncertainty by specifying a buffer whereby the catch limit is lower than that determined based on an assumption of perfect information. A “P* approach” is a method whereby catch limits are specified such that the probability that a future fishing mortality (F) exceeds the F corresponding to maximum sustainable yield (F_MSY) remains at or below a pre-specified value (P*). Some sources of scientific uncertainty are typically ignored when calculating catch limits, leading to underestimates of uncertainty about population biomass. A management strategy evaluation was developed to examine the performance of a P* approach that included accounting for typically ignored uncertainties. The P* approach was considered because it seems mandated by U.S. law; however, our results demonstrate that the approach has concerning properties. First, the amount of uncertainty that is taken into account is arbitrary and influences resulting catch limits. Secondly, small changes in the acceptable probability of overfishing can result in large changes to resulting catch limits. Third, the estimated and true probability of overfishing diverge as additional uncertainty is taken into account. Simulated cases will be shown to demonstrate each of these properties. Alternative harvest policies will be discussed.

Dr. Ingrid Spies^1
1AKFSC
February 16, 2016 9:00 (PST): FSH 203

The Alaska arrowtooth flounder stock assessment: how can we make it better?

Coauthors: Steve Barbeaux and Miriam Doyle

A single stock assessment model written in AD Model Builder is used for the Gulf of Alaska and Bering Sea/Aleutian Island stocks of arrowtooth flounder. The age-structured model begins in 1961 in the GOA and 1976 in the BSAI. The modeled population includes individuals from age 1 to age 21, with age 21 defined as a “plus” group (all individuals age 21 and older). The age composition of the species shows fewer males relative to females as fish increase in age, which suggests higher natural mortality (M) for males. To account for this process, natural mortality is fixed at 0.2 for females and 0.35 for males in the model. The model is fit to time series of catch, survey indices of abundance (3 surveys in the BSAI: Bering Sea slope, Bering Sea shelf, and Aleutian Islands), and estimates of age and length composition from fishery and survey. Model parameters are estimated by maximizing the log likelihood of the data, and there are likelihood components for: survey biomass, lengths from the fishery and the survey, survey ages, catch, recruitment, and female and male fishery selectivity.

There are several aspects of arrowtooth distribution and life history that have not been considered in the stock assessment, and this presentation is designed to stimulate a discussion on how to improve the assessment. Models suggest that the population sizes have increased 8-fold in the BSAI and four-fold in the GOA over the past few decades. The model does not incorporate carrying capacity, but there are some signs that population growth is decreasing. In addition, the maximum age and sex ratio of arrowtooth flounder appears different by region, which may contradict the assumption of different natural mortality rates for males and females.

Dr. Owen Hamel^1
1NWFSC
February 02, 2016 9:00 (PST): FSH 213

Investigating measures of fishing impact for use in informing management

NOTE THE ROOM CHANGE == FSH 213

Coauthors: Christopher M. Legault, Grant G. Thompson and Richard D. Methot, Jr.

Fishery management is informed by measures of the impacts of fishing on fish stocks. However, while a number of these measures have been proposed and used, none is able to capture all the short- and long-term impacts of fishing on fish stocks. One set of such measures are fishing intensity metrics, which under one definition are invariant to changing population structure, but because of that very attribute may poorly reflect the impact of fishing on a stock in any one year. Other common fishing intensity metrics are either inconsistent measures given changes in fishery selectivity (e.g. apical fishing rate F or exploitation rate U) or reflect only long term equilibrium effects (e.g. Spawning Potential Ratio (SPR) or Equilibrium Stock Depletion (ESD)). Other measures, which we call “fishing impact metrics”, may better reflect the short-term impacts of fishing or the impacts of short-term fishing. This may be especially important in certain situations, such as when fisheries target strong cohorts, and therefore fishing intensity metrics that are independent of the beginning of year numbers-at-age may underestimate both short- and long-term impacts. We will discuss a number of alternative fishing intensity and impact metrics with the goal of finding metrics which reflect short-term impacts under a variety of fishing strategies, as well as to allow for comparison across stocks and across alternative stock assessment methodologies.

Dr. Darcy Webber^1,2
1QUANTIFISH
²SAFS, UW
January 19, 2016 9:00 (PST): FSH 213

Exploring Bayesian state-space age-structured fisheries population dynamics models

Fisheries stock assessment models use incomplete observational data and knowledge of the system we are modeling to provide estimates of key management parameters, and require that we assume that the resulting uncertainty is properly reflected in the outputs. Current models typically follow a ‘deterministic’ approach. They are deterministic in the sense that the population follows sets of equations that deterministically define the population state from a previous state, without stochastic error.
State-space models provide an alternative approach to implementing fisheries assessment models. They can incorporate both observation uncertainty and the uncertainty that arises from modelling a simplification of reality — usually referred to as process error. Specifically, state-space models relate observations to unobserved states (e.g. numbers of fish in this context) through stochastic observation equations. Stochastic transition equations define how the unobserved states are assumed to evolve over time. Because state-space models incorporate both observation and process error explicitly, they may be able to help us better quantify the uncertainty of parameters of interest for management.
Until recently, state-space models have not been more frequently applied because they have been technically difficult to implement. However as computational methods in general and Bayesian methods in particular, become more sophisticated, the approach is becoming increasingly tractable.

Michelle Jones^1
1Simon Fraser University
January 05, 2016 9:00 (PST): FSH 203

Maturing estimates of maturity – probabilistic maturation reaction norms for British Columbia sablefish (Anoplopoma fimbria)

Predicting the size and age at which fish will mature is a key factor in estimating stock productivity. In fisheries stock assessment, this relationship is usually defined by estimating maturity ogives, which describe the proportion of the stock mature at a given age or length. However, length is a function of growth. Thus, differences in growth rates among years may cause apparent changes in maturation rates. While consistent spatial variation is expected in maturity ogives, in the case of sablefish (Anoplopoma fimbria), significant temporal variation in maturity ogives is unexpected and is large enough to cause substantial concern to stock assessment modeling. Here, we examine Probabilistic Maturation Reaction Norms (PMRNs) as an alternative for estimating sablefish maturity schedules. PMRNs condition the probability an individual will mature over a set time interval given its age, size, and survival, which may reduce confounding among each of these parameters, providing a temporally-stable model. If PMRNs can provide a stable relationship to estimate sablefish maturity schedules, then uncertainty in stock assessment models may be significantly reduced.

Dr. James Ianelli^1
1AKFSC
November 24, 2015 9:00 (PST): FSH 203

Data [un]limited assessments? Minding the gaps…

A critical and seemingly less emphasized component of fisheries research involves sampling design and efficiency. Also, understanding how sampling designs and effort has changed over time is an important element for specifying scale parameters (e.g., variances or sample sizes) for stock assessment modeling. In this discussion session we’ll illustrate some big-data tools (i.e., some rudimentary Hadleyisms) and approaches used to estimate variability from sampling to achieve process error estimates so they can be evaluated relative to risk-averse control rules. Examples include novel approaches to modeling fishery weight-at-age for Bering Sea pollock and growth increments related to bottom temperature for yellowfin sole from the same region.

Caitlin I Allen Akselrud^1
1SAFS
November 10, 2015 9:00 (PST): FSH 203

Application of an age-length structured population dynamics model to data for Eastern Bering Sea Tanner crab (Chionoecetes bairdi)

Fishery stock assessments are frequently based on either age-structured population dynamics or length-structured dynamics. We have developed and implemented a model to account for both age and length dynamics for an individual fishery. An age-length assessment model is unique in its ability to capture the dynamics of fishing mortality and natural mortality on fish populations, which are functions of both length and age. By accounting for both factors, the age-length model should reduce estimation bias present in models that only account for age (or length) dynamics, and assist managers in achieving statutory goals for fishery management.
Simulation studies have showed that the age-length models can provide essentially unbiased estimates of biomass trajectories and have low sensitivity to the precision of the data. Here we evaluate these models using the actual data for the Eastern Bering Sea Tanner crab (Chionoecetes bairdi) fishery. The data include catches, fishing effort, and survey index, as well as length-composition and conditional age-at-length. Survey and fishery data (including several fleets) are available for each stock.