Cole Monnahan^1
1SAFS
February 26th, 2019 9:00 (PST): FSH 203

Overcoming long Bayesian run times in integrated fisheries stock assessments

Abstract: Bayesian inference is an appealing alternative to maximum likelihood estimation, but estimation can be prohibitively long for integrated fisheries stock assessments. Here, we investigated potential causes of long run times for four U.S. assessments written in AD Model Builder (ADMB), both custom built and Stock Synthesis models. Between regularization, parallel chains, and the more efficient no-U-turn sampler we found models ran thousands of times faster. In this talk, I will go into detail about the causes of slow run times, the largest of which is the effect of over-parameterization on posterior geometry and how that interacts with MCMC sampler efficiency. I will also discuss the technical details of the default Metropolis and no-U-turn samplers in ADMB and how tuning parameters can dramatically increase efficiency, all in the context of real stock assessments. Finally, I will demonstrate the new R package adnuts which presents a convenient framework for running parallel chains and exploring convergence with the interactive tool ShinyStan. Throughout I will contrast ADMB’s (and TMB’s) Bayesian capabilities with the popular program Stan. The goal of this informal, hands-on talk will be for assessment scientists to leave with the knowledge and tools necessary to try Bayesian inference on their assessments. Participants wishing to follow along should have a model build with ADMB v12.0 and the github.com development branch of adnuts installed.

Grant Adams^1
1SAFS
January 29th, 2019 9:00 (PST): FSH 203

Sensitivity of multispecies statistical catch-at-age models to inclusion of annual recruitment deviations as random effects and diet estimation

The development of analytical tools that address ecosystem based fisheries management (EBFM) needs is essential for rationally managing fisheries resources in the face of changing climate and ecological processes. Multispecies statistical catch-at-age (MSCAA) models are one tool of particular value for providing both tactical and strategic EBFM advice by accounting for trade-offs among fisheries for multiple species. MSCAA links multiple single-species age-structured models via predator-prey interactions conditioned on bioenergetics and diet information. Within MSCAA models, annual recruitment are commonly estimated as deviations away from a mean process and predator-prey interactions conditioned on predator-prey suitability. For most MSCAA models, penalized likelihood approaches have been used to constrain recruitment deviations away from zero and suitability assumed to be known. Such an approach requires the a priori selection of recruitment variance and limit uncertainty in predation mortality. However, this can limit our understanding of the variation in recruitment and extent of predation and may also have an impact on estimates of biological reference points. Alternatively, recruitment deviations can be treated as random effects and suitability estimated., allowing for the estimation of variance terms and annual predation mortality. Using a MSCAA developed for groundfish in the Bering Sea (CEATTLE for climate-enhanced, age-based model with temperature-specific trophic linkages and energetics), I evaluate the sensitivity of MSCAA models to the incorporation of annual recruitment deviations as random effects and suitability estimation. This research addresses the need for a greater understanding of parameter uncertainty in MSCAA models prior to further implementation in fisheries management.

Noble Hendrix¹ and Erica Fleishman²

¹QEDA Consulting
²Dept. of Fish, Wildlife and Conservation at Colorado State University
October 30, 2018 9:00 (PST): FSH 203

Modeling the spatio-temporal dynamics of one little bugger – delta smelt in the San Francisco Estuary

Delta smelt are an annual estuarine fish that complete their life-cycle in the San Francisco Estuary. They were listed as threatened under the Endangered Species Act in 1993 and have continued to decline since then. As a result, there is interest in determining how management of water flow through the estuary may affect the population dynamics of the species. Focus has turned to factors that affect survival and movement in the fall and in the spring, and here we are interested in evaluating the fall. We have constructed a spatially explicit model that estimates survival and movement as functions of covariates in four regions of the estuary. This framework allows an explicit examination of hypotheses regarding the role of environmental drivers on these processes. The model also includes the probability of capture for spatial replicates within each region. We have developed the model in Stan and have fit to simulated data with well-behaved statistical properties; however, the catch data are zero-inflated and over dispersed. We investigate several alternatives including zero inflated Poisson and zero inflated negative binomial for the likelihood and present the results of fitting several conceptual models developed from recovery plans. We are interested in evaluating approaches for model averaging and model goodness of fit. With respect to the latter, we are developing metrics that highlight spatio-temporal patterns in model weaknesses to help generate the next round of conceptual models.

Margaret Siple^1
1SAFS
October 16, 2018 9:00 (PST): FSH 203

Simulating marine mammal bycatch impacts in data-poor situations

Simulation can be used to explore the expected outcomes of management actions when data on marine mammal abundance and bycatch are sparse or unavailable. For governments interested in evaluating their marine mammal bycatch with respect to regulations or instituting new changes (e.g., gear modifications or new bycatch limits), simulation can be a useful tool for exploring how marine mammal populations might respond to these actions. One of the work products of the Ocean Modeling Forum’s Marine Mammal Bycatch Working Group will be a tool that allows users to visually explore the impacts of different bycatch management actions on marine mammal abundance, given limited information about the fishery or the marine mammal of interest (i.e., abundance, current bycatch rate, and productivity).

The tool allows users to compare outcomes among various bycatch and starting abundance scenarios. It operates at different levels of complexity, using inputs chosen by the user in an age-structured model that projects abundance into the future, given estimates for abundance and bycatch rates. Users can apply specific life history parameters or choose from a list of life history types (humpback whale, bottlenose dolphin, bowhead whale, pinniped, or the ‘generic’ life history type used in the original simulations for PBR). Outputs include abundance trajectories under different management scenarios, and performance with respect to management objectives.

In the future, the tool will also be able to be used to find the strategy most likely to meet certain management objectives and explore the impacts of additional mortality besides bycatch (e.g., ship strikes). This talk will demonstrate the tool’s current capabilities and discuss some of the expansions to the current model, including ways to incorporate non-bycatch mortality and how to “back-calculate” strategies based on management objectives. I would also like to solicit feedback on some of the options for visualizing outputs.

Chantel Wetzel^1
1NWFSC
October 2, 2018 9:00 (PST): FSH 203

Summiting the steepness mountain: Current and alternative approaches in determining steepness for West Coast rockfish stocks

Productivity, termed steepness, in West Coast rockfish stocks has historically been difficult to estimate within stock assessment models. Since 2007, West Coast Estimating the rockfish assessments have used a regional meta-analysis prior to estimate or fix the steepness parameter. However, the information about the steepness parameter has varied among assessments for the same species over time and the number of species included in the meta-analysis has varied since 2007 resulting in variability in the prior mean for steepness being applied in assessments for rockfish species for each West Coast groundfish assessment cycle. The assumption regarding steepness is highly influential in the assessment model and the majority of assessments for rockfish species have been unable to estimate steepness resulting in the parameter being fixed at the prior mean. This was a major issue in the 2017 Pacific ocean perch assessment where the model results were deemed implausible when steepness was fixed at the prior mean. The steepness value applied in the final model was determined using a new approach for West Coast rockfish stocks where a model-ensemble approach with alternative values of steepness were equally weighted. This presentation will examine the approach that was used in the Pacific ocean perch assessment, explore the results of this approach being applied to other rockfish stocks, and the potential challenges to this approach.