Ian Taylor^1
1NOAA
November 28, 2017 9:00 (PST): FSH 203

What next for “r4ss”, a successful but aging R package?

The r4ss package, available at https://github.com/r4ss/r4ss, is a widely used set of R tools for to support the Stock Synthesis assessment platform. Thanks to a long history of development and contributions from over 20 people, it has been effective at increasing the efficiency of model explorations and writing assessment reports. However, the accumulated technical debt of years worth of edits to aging code have made adapting to evolving uses of the package more difficult. I will provide a brief overview of the package, discuss some of the key areas needing work, and hopefully get some necessary feedback on a variety of ideas for where things could go in the future.

Haikun Xu^1
1SAFS
November 14, 2017 9:00 (PST): FSH 203

Improving stock assessments from two perspectives: time-varying selectivity and data weighting

In this talk, I will present my recent research work regarding time-varying selectivity and data weighting in stock assessments. Selectivity is a key process in stock assessments and is typically varies with both age and time. To account for autocorrelated variation in selectivity, a common phenomenon in fisheries, I developed a new semi-parametric age- and time-varying selectivity method and implemented it in Stock Synthesis. Simulation experiments showed that the new selectivity method can improve the precision of model estimates from a data-rich assessment, given that the true deviations in selectivity are 2-dimensional (age and time) autocorrelated. Using the data for North Sea herring (Claupea harengus) as a Stock Synthesis case study, I found that the deviations in fishery selectivity were highly autocorrelated across both age and time, and the new selectivity smoother improved model fit and reduced the pattern in Pearson residuals for age composition. In the second part, I compared the three existing data-weighting methods (McAllister-Ianelli, Francis, and Dirichlet-multinomial) in Stock Synthesis to offer users good practices for weighting composition data when the new age- and time-varying selectivity feature is turned on. Simulation results suggested that when selectivity varies over time, the Dirichlet-multinomial method performs best in terms of weighting composition data. Moreover, it can be combined with the new age- and time-varying selectivity feature to simultaneously weight composition data and penalize selectivity variation inside the model.

Maite Pons^1
1SAFS
November 07, 2017 9:00 (PST): FSH 203

Evaluation of data-poor stock assessments for tuna fisheries

For the “principal market tunas”, like bluefin, bigeye, yellowfin, albacore and skipjack, stock assessments are performed regularly and a variety of management measures are in place to protect these stocks from overfishing. However, there are also other Scombrid species like mackerels and bonitos (commonly referred to as “small tunas”) that account for a notable proportion of the total tuna and tuna-like species catch. These species usually are not assessed and are currently not managed. Small tunas sustain important regional commercial fisheries in many coastal communities throughout their spatial distribution and their assessment is essential to be able to provide management advice which can help to ensure long-term sustainability. For most of these species there is a shortage of data that inhibits the performance of a “full” stock assessment and this is probably the most important reason why they have not been assessed so far. In general, catch time series from official tuna Regional Fisheries Management Organizations (tRFMOs) are incomplete or pooled by group of species. But, length composition data is usually available for a high proportion of these stocks thanks to numerous sampling programs implemented by different member countries of each tRFMO. In particular, the International Commission for the Conservation of Atlantic tunas (ICCAT) have suggested that different “data-limited” approaches that use life-history characteristics and length compositions of the catch should be considered in order to provide scientific information on the status of these stocks and guide management actions. The objective of the present study is to evaluate appropriate assessments and management options for small tunas in the Atlantic Ocean. First we took a “data-rich” stock (i.e. North Atlantic Albacore) as a case study, and we applied a set of different length-based and catch-based “data-poor” assessments in order to compare model outputs. Next, we will use a Management Strategy Evaluation (MSE) to determine the best combination of data, assessment and control measures to meet management objectives. Preliminary results related to the simulation modeling and data-poor assessment performance will be presented.

Lewis Barnett^1
1NOAA
October 24th, 2017 9:00 (PST): FSH 203

Deconstructing biomass to create an index of future overexploitation risk

Fishery management reference points focus on current population size of reproducing individuals (spawning stock biomass, SSB) and the fraction of unfished egg production that would remain after fishing at a constant rate for many years (spawning potential ratio, SPR). This perspective does not consider the effect of age- and size-structure on anything except numbers of eggs (e.g., SSB) and ignores near-term variation in SSB and recruitment (e.g., SPR) that can arise from transient fluctuations in population structure, which can contribute to the short-term risk of overexploitation. We build on recent research exploring how the age structure of fished stocks changes over time, using stock assessment model output and simple simulation of age-structured models to test the utility of candidate metrics for monitoring storage effects from age structure that buffer the influence of environmental variation on recruitment and SSB. We show how two populations with identical vital rates, exploitation rates, SSB and SPR can have striking divergence in short-term stock status because of different initial age structures. Preliminary results indicate that several common descriptors of change in age structure are influenced more by recruitment variation than the age diversity within the SSB, yet we identify a set of more robust metrics for future testing in potential application as the basis for a novel limit reference point or an additional index of assessment uncertainty.

Kate Richerson^1
1NOAA
October 17th, 2017 9:00 (PST): FSH 203

Reconstructing the historical abundance of West Coast Dungeness crab (Cancer magister) using catch data

US Dungeness crab (Cancer magister) landings are valued at >$200 million annually, making it one of the most economically-important West Coast fisheries. The fishery has been managed by the states under a simple “3-S rule” (referring to size, sex, and season) since the 1940s. No stock assessment is performed, but the conventional wisdom suggests that the vast majority of legal-sized crabs are taken every season. We use catch data in the form of fish tickets and some limited logbook data in combination with depletion estimators to estimate the pre-season abundance of legal-sized crabs from 1982-2015. This work is still in progress, but results so far indicate that most legal crabs are indeed taken each season, but that populations are stable or increasing. We find evidence for density-dependent catchability, and suggest that failing to account for this can bias abundance estimates downward.

Dr. Noble Hendrix^1
1QEDA Consulting
May 09, 2017 9:00 (PST): FSH 203

A court mandated life-cycle model to evaluate California Central Valley water operation effects on Chinook salmon

In 2009, NOAA Fisheries Central Valley California Office submitted a Biological Opinion on the operations of multiple water control structures (dams, routing gates, and export facilities), that was litigated by multiple water agencies. The Biological Opinion was remanded in 2011 by the judge, who also provided a strongly worded rebuking (fueled to some degree by the testimony of SAFS own Ray Hilborn) that stated, “this is the last time NMFS will be permitted to avoid studying, analyzing, and applying a life cycle model. NMFS’s chronic failure to do so now approaches bad faith in view of the undeniable importance of the information to resolve the perennial dispute over population dynamics. ”

NOAA Fisheries Southwest Fisheries Science Center shortly thereafter initated a life cycle modeling project to predict how various salmon populations would respond to suites of management actions, including changes to flow and export regimes, modification of water extraction facilities, and large-scale habitat restoration. I was asked to help with the first model in the project, which is a life cycle model for winter-run Chinook salmon (WRLCM). The WRLCM couples water planning models and physical models (e.g., HEC-RAS, DSM2, DSM2-PTM) and fishery models to evaluate management actions on winter-run population dynamics. I will describe the general model structure, the transition equations that define the movement and survival throughout the life cycle and linkages to physical models for integrating dynamics over finer spatial and temporal scales. Because the WRLCM was developed to evaluate management actions, the model sptial and temporal structure focuses on addressing questions of hydromanagment and habitat restoration. Unfortunately, this structure is not particularly well-suited for statistical estimation and presents numerous difficulties for estimating model parameters. We implemented a state-space structure and estimated the model parameters using a variant of the Expectation-Maximization algorithm to obtain MLE estimates and are actively pursuing MCMC approaches for Bayesian estimation. The fitted model is currently being used to evaluate proposed actions as part of a 2017 NOAA Fisheries Biological Opinion, meeting the judge’s 2011 request. More importantly, the calibrated WRLCM is providing a central framework that can be used to assess restoration projects, construct adaptive management frameworks, evaluate reintroduction success, simulate new monitoring designs, and prioritize recovery actions.

Kelli F. Johnson,¹ Dr. Andre E. Punt² and Dr. James T. Thorson^3
1SAFS
²SAFS
³NWFSC

April 25, 2017 9:00 (PST): FSH 203

Moving beyond Q-Q plots for spatio-temporal models

The assessment of models, which are always wrong, largely consists of two components, assessing the relative performance of candidate models and characterizing differences between observed data and model predictions. Residuals and relative error are traditionally used to identify the presence of outliers, quantify the relevance of modelled and excluded factors, and assess overall model fit. With linear models, the Q-Q plot is a standard a metric of model fit, e.g., the fat-pen test. Unfortunately, Q-Q plots can be misleading when used to assess models that include spatially-explicit parameters, where spatial patterns in residuals can indicate model misspecification. Here, we present the results of a simulation experiment to quantify the performance of the structural similarity index as a metric of model fit. Simulations were conducted using a vector-autoregressive spatio-temporal model that estimated spatio-temporal variation in density from simulated survey data for a hypothetical species off the US West coast. The structural similarity index provided spatially-explicit information regarding lack of model fit not uncovered by traditional model assessment methods. Future work will focus on how well the structural similarity index can identify misspecification, rather than lack of fit in a properly specified model, as was done here.

John Trochta^1
1SAFS
April 11, 2017 9:00 (PST): FSH 105

An investigation of popular hypotheses on the survival of Pacific herring in Prince William Sound, Alaska using Bayesian model selection

We investigate how changing environmental conditions may impact Pacific herring (Clupea pallasii) population dynamics in Prince William Sound, Alaska. The Prince William Sound herring collapsed in the early 1990’s and have yet to recover from low abundance that has remained below the lower regulatory threshold (i.e. for closing fishing) of 20,000 metric tons. This event continues to motivate herring research, especially in improving the predictability of herring biomass dynamics. There is large uncertainty in these dynamics, where we seek to improve biomass estimates by including covariates from physical and ecological data in a Bayesian age-structured assessment model of herring biomass. Indicators of potential environmental effects (i.e. sea surface temperature, air temperature, freshwater discharge, and sea level) and species interactions (i.e. pink salmon, humpback whales, and pollock) are modeled as covariates impacting either adult or juvenile survival in the assessment model. This effort will build upon previous hypothesis testing and reviews that examined the causes of both herring collapse and recovery failure by synthesizing both old and new hypotheses with updated data, as well as new data sets. Our results help to determine which effects may be most significant, while further improving the utility of herring stock assessment models in being able to simulate biomass dynamics under different environmental conditions.

Caitlin Allen Akselrud^1
1SAFS
February 21, 2017 3:30 (PST): FSH 203

A discussion-based Think Tank covering possible modeling approaches for a difficult-to-assess species, California market squid (Doryteuthis opalescens)

This will be a fun, discussion-based Think Tank covering possible modeling approaches for a difficult-to-assess species, California market squid (Doryteuthis opalescens). Please bring your model ideas, and a phone or laptop for active participation!

Market squid are short lived (6-9 months), semelparous, aggregate spawners. A recent dissertation (Cheng, 2015) suggests multiple temporally-separated sub-population, indicating multiple spawning groups throughout the year. This potentially presents a form of “portfolio effect” for this species. [You may notice some life history similarities to salmon.] Fluctuation in market squid abundance have been correlated to ENSO cycles, and squid have specific preferences for spawning ground conditions (Zeidberg et al., 2012). Currently, there is no fishery-independent survey of market squid. Fishery catch and effort data are available [reliably] from 2000-present.

Depletion methods are often used on squid populations as fishers fish down spawning aggregations. Several problems to this approach exist for the California market squid, including: a) multiple cohorts caught simultaneously (not fishing a single cohort), b) spawning aggregations immigrate into spawning grounds over several weeks (not a closed population), and c) fishing can be driven by external economic drivers, causing problems with CPUE estimates, as effort does not necessarily reflect abundance.

Bring your thinking caps and brainstorm some possible methods with me!

Cole Monnahan^1
1SAFS
February 07, 2017 3:30pm (PST): FSH 203

Introducing the no-U-turn MCMC sampler in ADMB and TMB: faster run times for large, complex fisheries models.

Bayesian analysis is a powerful tool for statistical inference in fisheries science. However, some analyses remain impracticable due to extremely long runtimes, despite increases in processor power. This is the case for data-rich age structured assessments written in AD Model Builder (e.g., stock synthesis) and large, complex spatio-temporal models for species distributions, typically built with Template Model Builder (TMB). Recently, the software package Stan, powered by the no-U-turn (NUTS) MCMC sampler, has emerged as state of the art for Bayesian analyses and it promises faster runtimes. However, for many fisheries scientists, it is not immediately feasible (nor desirable) to rewrite ADMB/TMB models in Stan. In these cases, it makes more sense to bring the algorithms to ADMB and TMB, rather than bring the models to Stan. In this workshop I will demonstrate Stan and how to interpret the diagnostic output from NUTS. I will then present and discuss my work on integrating NUTS into both TMB and ADMB. The workshop will contain a hands-on element in TMB, so bring your laptop with it installed, and will be very informal. Interested ADMB users can test their own models but need to compile an ADMB fork (contact me about this). I will expect that attendees are familiar with NUTS (read Monnahan et al. 2016) and the brave amongst you should try Betancourt 2017.

Monnahan, C. C., Thorson, J. T. and Branch, T. A. (2016), Faster estimation of Bayesian models in ecology using Hamiltonian Monte Carlo. Methods Ecol Evol. doi:10.1111/2041-210X.12681

Betancourt, Michael. A Conceptual Introduction to Hamiltonian Monte Carlo. arXiv preprint arXiv:1701.02434 (2017).