A workshop has been designed to evaluate several major features that are considered desirable in tuna stock assessment models that might require consideration when designing the base structure of a general stock assessment program. This will help decide what approach should be followed for the next generation tuna assessments. The topics covered include: 1) Spatial structure; 2) Tagging data; 3) Length based dynamics; 4) Random effects; and 5) Close Kin Mark-Recapture. We will also cover: The modeling framework and model components, Workflow in RTMB, and lessons learned from previous developers.
Recordings of the workshop are posted on the CAPAM YouTube Channel. Links to the recording of each presentation and copies of the slides are listed below after each abstract.
11:00 Introduction - Arni Magnusson and Mark Maunder (Chair)
Spatial structure
11:10 Assessing the Multiverse: Considerations for Explicitly Integrating Spatial Structure in Next-Generation Stock Assessments - Matt Cheng
11:40 Practical, scale-free, and generic approach to movement modelling: bridging from individual-based to population-scale dynamics - Jim Thorson
12:00 Quantitative modeling of the spatiotemporal dynamics of tuna populations - Inna Senina
12:20 Discussion (recording)
12:50 Break
Tagging data
1:10 Mixing It Up: Considerations for Integrating Tagging Data into Next Generation Tuna Stock Assessments - Dan Goethel
1:40 Go with the flow: adding movement to tuna stock assessments with admove - Tobias Mildenberger
2:00 Modelling migration: an invertebrate perspective - Simon de Lestang
2:20 Discussion (recording)
2:50 Summary
Random effects
11:00 The Essential Roles of Random Effects in Fish Stock Assessment - Anders Nielsen
11:30 Random Effects in WHAM: Use in Northeast US assessments - Tim Miller
11:50 Random effects in stock assessment models - Noel Cadigan
12:10 Discussion (recording)
12:40 Break
Close kin mark-recapture
1:00 Incorporating CKMR in tuna stock assessments - Rich Hillary
1:30 CKMR and tuna and software - Mark Bravington
1:50 Routine Collection of CKMR Data and Integration in Assessment: Considerations for Tunas - Nicholas Fisch
2:10 Discussion (recording)
2:40 Summary
Length based dynamics
11:00 Age-Length-based Stock Assessments: Solution or Just Yet Dead End - Andre Punt
11:30 Length based processes in gadget3 -Bjarki Þór Elvarsson
11:50 Model mis-specification of length-based processes among available model structures - Nick Davies
12:10 Discussion (recording)
12:40 Break
General concepts of implementing a model
1:00 Lessons learned from 40 years of Stock Synthesis; handing the baton to FIMS - Richard Methot
1:20 Discussion
1:30 Stock assessment workflow using RTMB - Darcy Webber
1:50 Discussion
2:00 Factors to consider when developing the next generation tuna stock assessment model - Arni Magnusson and Mark Maunder
2:20 Discussion
2:30 Revised Thoughts on the Next Generation Tuna Stock Assessment Model - Mark Maunder and Arni Magnusson
2:40 General discussion (recording)
CKMR and tuna and software
Mark Bravington
Drawing on my experience of over a dozen CKMR projects in diverse settings, this talk will touch on:
Inputs & outputs
Big questions for CKMR and tuna: suitability, cost, value, and obstacles.
Summarize six "tuna-like" CKMR projects, in terms of model differences.
Spatial & Sampling
Software requirements & tricks
Where to start ?
Presentation slides: MVB-CKMR-tuna-software.pdf
Random effects in stock assessment models
Noel Cadigan, Memorial University
I illustrate ways I commonly use random effects in state-space stock assessment models (SSAMs) based on a model for thorny skate, which is a data-moderate stock with no aging program and no fishery catch size compositions. The assessment model is estimated using length-based survey indices, estimates of total fishery landings, and weight- and maturity-at-length. I used random effects for F, M, survey catchability (Q), recruitment, and initial population size.
The second part of my talk reviews some theoretical and simulation research on the frequentist sampling properties of marginal maximum likelihood estimates of parameters and empirical bayes predictions of random effects in nonlinear mixed-effects regression models including SSAMs. This is the inferential framework used in ADMB and TMB. This research is particularly relevant for interpreting the results of SSAM simulation studies. I highlight that TMB standard errors are actually mean squared errors (MSEs) or prediction standard errors (PSEs).
In stock assessment the effects we model as random are usually just high dimensional parameters (HPDs) that we model as random for statistical regularization purposes. However, in this case inference should be conditional on specific values of the HDPs rather than the marginal inferences provided by TMB. There is conditional bias to worry about and the conditional variance is smaller than the marginal variance, although the conditional MSE can be approximated by the marginal MSE (i.e., PSEs). Hence, in the end we recommend confidence intervals for stock assessment quantities should be based on the TMB PSEs. It is possible to do better with a bias correction procedure we proposed, which we illustrate with some SAM case studies, but this requires further research.
Presentation slides: noel_cadigan_RE1.pdf
Assessing the Multiverse: Considerations for Explicitly Integrating Spatial Structure in Next-Generation Stock Assessments
Matt Cheng1, Daniel Goethel2, and Aaron Berger3
1University of Alaska, College of Fisheries and Ocean Sciences, Juneau, AK
2NOAA, Alaska Fisheries Science Center, Juneau, AK
3NOAA, Northwest Fisheries Science Center, Newport, OR
Spatial structure is widespread across marine populations and arises from the interplay among individual movement, recruitment dynamics, spatially-varying environmental conditions that shape demographic patterns, and heterogeneity in fishery dynamics. Although stock assessment models often rely on the assumption of a single, homogenous ‘unit stock’, there is a growing recognition of the importance of incorporating spatial dynamics to reduce biases in population estimates and ensure robust management advice. Here, we summarize the common forms of population structures encountered across marine systems, which should ideally constitute the minimum set of structural features that a next-generation tuna assessment can accommodate. Approximations for representing spatial structure within panmictic formulations are discussed (e.g., fleets-as-areas models), which remain relevant for tuna assessments when data limitations or model complexity preclude the implementation of fully spatially explicit approaches. General decision points and recommended practices for spatially-stratified stock assessments (i.e., large-area box transfer models) are outlined, with a focus on next-generation spatial considerations in recruitment processes, movement, demography, and fishery dynamics. We conclude that next-generation tuna stock assessments, should at a minimum, be capable of representing a wide range of population structures (e.g., panmictic populations, metapopulations, natal homing), flexibly model recruitment dynamics, apply parsimonious movement formulations, incorporate spatially-varying demography, and accommodate diverse fishery fleet structures. However, it remains unclear whether multi-scalar frameworks that generalize fine-scale spatial processes to spatially-stratified contexts are necessary, and how such an approach could be implemented analytically. Nonetheless, a generalized next-generation tuna assessment platform should support seamless extension from single-region models to multi-stock, multi-region assessments, thereby facilitating application across stocks spanning a spectrum of data quality and availability.
Presentation slides: Cheng_NextGen_Tuna_SptStrc_v2.pdf
Model mis-specification of length-based processes among available model structures
Nick Davies
An overview of model structures (age-, size-, and age-size-structured) is given to highlight those aspects causing mis-specification of certain length-based processes, and to identify the formulations developed in each that attempt to correctly specify or approximate them. The length-based processes considered are: individual, and inter-annual variability in growth; and, spatial heterogeneity in growth. A ranking of the mis-specification “level” is offered in respect of each model structure in terms of its caveats; with a recommendation given for that model structure having the least mis-specification. The potential for bias from these forms of mis-specification in the context of Western and Central Pacific Ocean tuna is illustrated, and include: high levels of both individual growth variability and fishing mortality; inter-annual environmental variation; and, large-scale movement.
Presentation slides: davies_kim_lenbased_misspecfctn_vsn3.pdf
Length based processes in gadget3
Bjarki Þór Elvarsson1, Jamie Lentin2, and William Butler1
1Marine and Freshwater Research Institute, Iceland
2Shuttlethread
Many key processes that shape fish population dynamics are size-based, including mortality, fleet selectivity, consumption, and maturation. For stock-assessment purposes, however, these dynamics are often assumed to depend on age rather than size, largely for computational convenience. In practice, sampling is frequently sparse and growth information limited, partly because obtaining reliable age data is difficult.
Integrated assessment models of population dynamics—such as those developed within the Gadget framework—help address these data limitations by combining information from multiple datasets and sources. Typical Gadget models track numbers at age and length at each time step, which allows key processes to be represented directly as length-based. In this presentation, we describe the Gadget approach to parameterising the length-transition matrix, show examples of how it can be adapted to different use cases and discuss computational associated costs. We also explain how data can be incorporated into the model through likelihood components. Finally, we illustrate the approach with a case study in which an age–length model is fitted without of the use of age data.
Presentation slides: gadget-lengthupdate.pdf
Routine Collection of CKMR Data and Integration in Assessment: Considerations for Tunas
Nicholas Fisch
Integrated fisheries assessments form the backbone of commercial fisheries management throughout the world, especially for Tunas. Close-Kin Mark-Recapture (CKMR) sampling offers a promising new data source to integrate within fisheries stock assessments. I wonder whether fisheries should consider routinely collecting genetic information for every fish already being collected for age composition. To this end I examine, using self-test simulations, the expected improvements in precision and accuracy of derived quantities and estimated parameters within statistical catch-at-age models when CKMR sampling of the age composition samples is conducted and the data integrated within the assessment. I examine the expected improvements across three life history types (cod-like, flatfish-like, and sardine-like) and different amounts of data available to the assessment, including the uncertainty and inclusion of an abundance index and the sample size and time series length of CKMR and age composition samples. Results suggest CKMR data can provide considerable improvements in accuracy and precision of spawning stock biomass at the end of the time series and parameters defining natural mortality and scale of the population, provided an adequate annual sample size is collected relative to the spawning abundance of the stock during the period of CKMR inference. I will discuss the paper and some of my experiences with coding CKMR integration into assessment, in addition to some tuna specific considerations.
Presentation slides: Fisch Tuna CKMR Talk short.pdf
Mixing It Up: Considerations for Integrating Tagging Data into Next Generation Tuna Stock Assessments
Daniel Goethel1, Matthew Vincent2, Matt Cheng3, and Aaron Berger4
1NOAA, Alaska Fisheries Science Center, Juneau, AK
2NOAA, Southeast Fisheries Science Center, Beaufort, NC
3University of Alaska, College of Fisheries and Ocean Sciences, Juneau, AK
4NOAA, Northwest Fisheries Science Center, Newport, OR
Tagging data provides important information on distribution, movement patterns, growth, and mortality for widely distributed and mobile tuna species. Although tag-integrated assessments can directly utilize tagging data to inform parameter estimates, tag ‘nuisance’ parameters (e.g., tag mixing, mortality, shedding, and reporting rate) must be addressed to ensure appropriate inference and avoid estimation bias. Similarly, for electronic tags, scaling from the dynamics of relatively few individuals to that of the entire population can be inappropriate. We synthesize the methods available to incorporate tagging data within spatial assessment frameworks, then highlight key features to consider for next generation tuna stock assessment models. Capabilities of existing integrated spatial assessments (e.g., conventional tagging sub-models) form the minimal requirements for future tuna assessment platforms. Next, models that integrate all tag types (e.g., electronic tags, genetic tags, biological tags, and conventional tags), including appropriate likelihood assumptions and robust scaling to population level dynamics, will be necessary. Ultimately, multiscalar frameworks that enable modeling data sources at varying spatiotemporal scales (i.e., integrating high resolution tagging modules where movement and mortality estimates scale to the broad stratified units typically assumed for other data sources) may help address tag mixing issues and potentially better inform movement estimates (e.g., through the use of preference functions linked to environmental data). Thus, next generation assessment platforms should ideally meld current high resolution tagging models with spatially explicit stratified assessments to enable joint estimation of parameters, while also allowing these modules to be run independently (given likely run-time constraints) with subsequent integration via priors or penalties on movement parameters. Developing a modular assessment platform that can easily collapse partitions and allow varying spatiotemporal resolutions among sub-models will ease exploration of adequate model complexity. Yet, consideration should be given to when more complex assessment tools are necessary to provide robust management advice, considering extant data limitations. Concomitant development of high resolution simulation and management strategy evaluation tools could help optimize data collection schemes and potentially inform simpler, more transparent management procedures.
Presentation slides: Goethel et al._tag integration considerations_final_post.pdf
Incorporating CKMR in tuna stock assessments
Rich Hillary
The robust and accurate assessment of global tuna stocks faces many challenges (contracting fishing footprints, patchily sampled commercial data, decreasing funding). CKMR has held a lot of appeal for these stocks and its limited implementation - mostly for the bluefin tuna species - has shown that is has clear potential to be transformative for the other major tuna stocks. This talk will briefly cover off on the fundamentals of the idea and how this layers onto the key stock assessment parameters and variables. Using Southern Bluefin tuna, probably the most "complete" implementation in terms of assessment integration and RFMO long-term buy-in, we outline what implementation might look like in terms of data collection, model building and long-term monitoring potential. Using a spatially-explicit CKMR design program for Indian Ocean yellowfin tuna I also address the issue of whether larger populations with more complex spatial dynamics are a barrier to implementation.
Presentation slides: Hillary pres_CKMR_tuna.pdf
Modelling migration: an invertebrate perspective
Simon de Lestang
In stock assessment modelling, simplicity is preferred, with complexity warranted only when essential for capturing key biological processes or complex reporting requirements. One approach to maintaining spatial simplicity is to use fleets as proxies for areas. However, finer spatial resolution becomes necessary when model outputs are required at smaller spatial scales or when the target species' biology varies markedly across the modelled domain. Once discrete areas are incorporated and the species exhibits movement behaviour, the additional complexity of migration must also be replicated.
Migration can be modelled based on size or age, with the choice informed by the drivers which induce movement, such as an pre-maturation migration to breeding grounds. A critical challenge emerges when animals move between areas with different growth characteristics. Most stock assessment platforms (e.g., Stock Synthesis 3) employ growth equations that describe length as a function of age. When an individual migrates to an area with different growth parameters, its length-at-age no longer conforms to the destination area's growth relationship, preventing direct application of the new growth equation. Size transition matrices, which relate current size to future size distributions, offer a practical solution by enabling differential growth rates without age dependency.
This has led to the adoption of integrated length- and age-based models that track both dimensions simultaneously using length bins and size transition matrices for growth. This framework is currently employed in Western Australia for crustacean stock assessments, including the Western Rock Lobster and Crystal Crab fisheries, successfully balancing spatial realism with operational tractability.
Presentation slides: Simonde_Modelling_Migration.pdf
Introduction
Arni Magnusson and Mark Maunder (Chair)
Arni Magnusson, Nick Davies, Graham Pilling, Paul Hamer, and Mark Maunder
We introduce the CAPAM mini virtual workshop by outlining the reasons why a next generation tuna stock assessment model is needed. The two main software programs used for assessing tunas in the EPO (MULTIFAN-CL and Stock Synthesis) are coming to the end of their development cycles due to staff retirements and the underlying programming languages becoming outdated. Evaluation of software for tuna assessments presented to the WCPFC Scientific Committee in 2025 concluded that new software specifically for tuna assessments is needed. Development of this software through a collaboration between the Pacific Community (SPC) and the Inter-American Tropical Tuna Commission (IATTC) is clearly advantageous due to their long-standing relationship, comparable datasets, comparable challenges, and overlapping stock distributions. There is also a related ongoing collaboration between SPC, IATTC, and the Danish Technical Institute (DTU) on a fine-scale spatial-temporal model of tuna tagging data. SPC will present a project proposal in August 2026 for funding the development of the next generation tuna model. This workshop will help guide the decisions about what type of software should be developed and how it will be developed.
Recording
Presentation slides: Arni 2025_12_08_capam.pdf
Factors to consider when developing the next generation tuna stock assessment model
Arni Magnusson and Mark Maunder
We outline the main factors to consider when developing the next generation tuna stock assessment model including initial decisions such as whether the project should 1) collaborate with an existing development (e.g., FIMS), 2) create a new tuna focused general model, or 3) extend an existing fine-scale spatial-temporal model (e.g., SEAPODYM, admove). We discuss several factors including 1) leadership and governance structure, 2) format of the development team, 3) the development environment, 4) code requirements, 5) programing approach and language, 6) and components to include. We also discuss the many other components of a stock assessment ecosystem that need to be included ranging from generating a conceptual model of the system and the good practices to implement it within the software package, through to diagnostics, displaying results, and providing advice using simulations in a MSE framework.
Presentation slides: Develoment concepts.pdf
Revised Thoughts on the Next Generation Tuna Stock Assessment Model
Mark Maunder and Arni Magnusson
We outline the main issues involved in developing the next generation tuna stock assessment model identified throughout the workshop. We then provide a strawman proposal to develop the next generation tuna stock assessment model. Although the strawman is based on our current thinking on the best way forward, it is only intended to promote discussion at this stage. This proposal is based on the several factors 1) funding will be limited, 2) potential uses and contributors will be limited, 3) stock assessment scientists will contribute to the development, 4) inclusion of tagging data is essential, 5) the model will be used in the short to medium term, 6) development of a simple software package is more likely to succeed. The straw man includes a) code a new model in RTMB for tropical tunas, b) Include random effects, c) do not include CKMR because there are few current applications and spatial dynamics might complicate the analysis, d) include tagging data, e) include a scalable spatial structure, f) do not include length-based dynamics since there is currently few data on growth for tropical tuna, g) use a focused development approach (i.e., limit the people involved), h) and use a single stock assessment scientist with extensive programing experience to develop the initial version of the model before letting other contribute. We also highlight that it is unclear whether including fine-scale spatial-temporal dynamics and tagging data in the stock assessment model is the best approach.
Presentation slides: Summary Maunder.pdf
Lessons learned from 40 years of Stock Synthesis; handing the baton to FIMS.
Richard Methot
Six years ago, the assessment community gathered for a CAPAM workshop in New Zealand on the topic of next generation stock assessment models. NMFS used the success of that workshop as a rationale for initiation of a project to build such a model in the U.S. using a team distributed across all six Science Centers and our Office of Science & Technology. Our Fisheries Integrated Modeling System (FIMS) project is designed to provide such a model for the assessment community. It uses a combination of C++, R, and TMB in an open-to-all model development process enabled by github. Today, the end of ADMB and SS3 is much closer than it was 6 years ago. Fortunately, the Fisheries Integrated Modeling System has made great progress to develop core capability. It is scheduled to undergo independent review in April 2026 and several stock assessments in 2026 will see side-by-side FIMS models presented. While FIMS currently is yet to develop all the specific capabilities requested for the next generation of tuna models, it does provide the framework and capability for such features to be added. Developers from the tuna modeling community can help accelerate the inclusion of needed features in FIMS.
Presentation slides: Methot SS3_Lessons_Learned_CAPAM_2025.pdf
Go with the flow: adding movement to tuna stock assessments with admove
Tobias Mildenberger
This talk introduces admove, an advection–diffusion state-space model for estimating movement and other stock-assessment-relevant quantities from tuna tagging data. Taxis, advection and diffusion can be linked to smooth habitat-preference functions of environmental covariates such as sea surface temperature and mixed-layer depth. admove can be fitted to both conventional and archival tags and allows incorporation of spatially resolved fishing effort and catch, enabling simultaneous estimation of movement rates, length-based natural and fishing mortality, and abundance indices. The model is implemented using two numerical approaches, a matrix-exponential method on a spatial grid and a continuous-space Kalman filter. We illustrate the framework using skipjack tuna in the eastern Pacific Ocean, highlighting inferred habitat preferences, seasonal movement patterns, and spatially explicit mortality and biomass estimates. Finally, we show how admove outputs can feed into tuna stock assessment models and outline ongoing work to extend the approach to other species, tag types and ocean basins.
Presentation slides: presi_mildenberger.pdf
Random Effects in WHAM: Use in Northeast US assessments
Tim Miller
In this talk, I briefly review the various ways random effects can be implemented in the WHAM software package, a state-space age-structured assessment modeling platform, we use in the Northeast US, and what random effects options are used in assessments for managed stocks. I then summarize conclusions from recent research on application of random effects in state-space age-structured assessment models including simulation studies carried out as part of a working group investigating the reliability of state-space assessment models.
Presentation slides: Tuna_slides_miller.pdf
The Essential Roles of Random Effects in Fish Stock Assessment
Anders Nielsen
A primary goal of fish stock assessment models is to provide outputs that are valuable for fishery management. Consequently, a key requirement is the ability to predict year-to-year changes in stock status and to quantify the accuracy of those predictions. At its most fundamental level, this is precisely what state-space models are designed to do. Furthermore, random effects --- the defining feature of state-space models --- offer elegant solutions to many persistent challenges in stock assessment. These include formulating flexible time-varying processes with a parsimonious number of parameters; handling missing observations; accounting for and predicting the effects of environmental variation; objectively weighting different data sources; partitioning observation error from process variation; and estimating variances of latent biological and fishery processes. While most applications to date have focused on biomass or age-structured models, the core benefits of the state-space approach --- particularly its ability to structure flexible processes and account for correlations --- are perhaps even more critical for length-based assessment models required for tuna stocks.
Presentation slides: nielsen-slides.pdf
Age-Length-based Stock Assessments: Solution or Just Yet Dead End
André E. Punt, School of Aquatic and Fishery Sciences, University of Washington
Most fishery stock assessments are based on age- or size-structured population dynamics models and are fit to a variety of data sources. Stock assessments based on age-structured models are computationally less burdensome than those based on size-structured models but often require very strong assessments to fit to size-composition data, an important data source for many hard-to-age species. Stock assessment methods based on age-structured models also fail (generally) to account for changes in length-at-age due to size-selective fishing. In contrast, stock assessments based on size-structured models account for size-selective fisheries but generally cannot represent any age-specific processes and cannot fit to age-composition and conditional age-at-length data. Age-size-structured models can however overcome these concerns at the cost of increased computational demands. This talk contrasts how age-, size- and age-size-structured models are formulated and some of the consequences of an overlying simplistic model formulation. It then summarizes past studies on the performances of each type of assessment framework and highlights challenges associated with moving to stock assessment methods based on age-size-structured population models. These challenges can be divided broadly into: (a) selecting appropriate modelling assumptions, (b) reducing the computational burden, and (c) parameter estimation and data needs.
Presentation slides: 2025-12-11 Punt - Age-length models.pdf
Quantitative modeling of the spatiotemporal dynamics of tuna populations
Inna Senina
SEAPODYM is a Eulerian modelling framework for describing basin-scale, spatiotemporal dynamics of migratory top predators under the influence of environmental variability and fishing. Today it is designed for quantitative modeling of the spatiotemporal and age-structured dynamics of tuna stocks. The underlying advection–diffusion–reaction equation with an ageing term (ADR-a), completed with appropriate initial and boundary conditions, is discretized on a regular grid and solved with an alternating-direction implicit method. The ADR-a describes three key dynamic processes being movement, reproduction, and mortality, which are environment-driven; the environment is represented by ocean forcings consisting of physical, biogeochemical variables and prey distributions.
To enable quantitative skills, the modelling framework includes parameter estimation, global sensitivity analysis, twin experiments, and likelihood profiling, and integrates multiple sub-models. Applications include: (i) simulations of population dynamics with or without exploitation; (ii) a tagged-fish movement model; (iii) spawning and feeding habitat models; and (iv) evaluation of stock connectivity across regions or EEZs. Parameter estimation (for applications i–iii) is done by minimizing the negative log-likelihood with a quasi-Newton optimizer and an exact analytical gradient provided by the model’s adjoint code. Parameters are informed by geo-referenced fisheries data (catches and length compositions), tagging data, and early-life data. Adding non-fisheries data is essential, as relying on fisheries data alone can bias estimates toward high non-directional mixing (diffusion), weak directed movement (advection), and artificially elevated variability driven by local reactions. Incorporating tagging data improves estimates of environmental preferences and movement (both advection and diffusion) and reduces the amount of predicted unobserved biomass. Early-life data, although sparse, provide a valuable information on the spatial–seasonal distribution of spawning habitat and spawning biomass. Overall, maximum-likelihood estimates produce stock sizes consistent with those from the MULTIFAN-CL stock-assessment models in the Western and Central Pacific Ocean.
Because estimated population-dynamic parameters are invariant in space and time within this mechanistic framework, SEAPODYM can be validated against independent fisheries and tagging datasets. Once validated, it becomes a powerful tool for management: testing marine spatial planning scenarios, assessing environmental effects on recruitment and stock size, and projecting climate-change impacts on abundance and distribution at basin and regional scales. It can also support stock assessments by supplying movement rates, regional abundance indices, and indicators of fishery change (e.g., effort creep). These benefits come with challenges—in particular, high dimensionality, optimization non-linearity, parameter identifiability, and biases in oceanic forcings—which must be addressed to quantify uncertainty and to enable potential use for stock assessment.
Presentation slides (updated): SEAPODYM_IS_20251209.pdf
Practical, scale-free, and generic approach to movement modelling: bridging from individual-based to population-scale dynamics
Jim Thorson
Population dynamics are driven by movement, but stock assessment has a limited toolbox for directly measuring or modelling movement relative to other ecological subfields. Here, I provide a quick introduction to advection-taxis models, which define a partial differential equation (PDE) where advection is driven by the gradient of a preference function (“taxis”). In particular, I discuss three numerical approaches for solving advection-taxis movement: using individual location as a variable (state-space Lagrangian models); discretized spatially at high resolution to solve as a Hidden Markov Model (archival tag track reconstruction); or discretized at low resolution within a “few-box” population-dynamics model. I claim that advection-taxis is a generic, parsimonious, and scale-free approach to model either fine-scale (individual-based) or coarse (population-dynamic) movement.
Presentation slides: Thorson_movement_V1-1.pdf
Stock assessment workflow using RTMB
Darcy Webber
Transitioning to RTMB has significantly streamlined stock assessment workflows for my team and me, offering seamless integration with R's ecosystem. By implementing models directly in R, we can encapsulate them into lightweight packages without relying on compiled code, thereby reducing package size, eliminating compilation delays, and enhancing portability. This approach harnesses R's robust tools, such as roxygen2 for function documentation, vignettes for comprehensive package overviews, testthat for unit testing (easily automated via GitHub Actions), and access to an extensive library of R packages. In this presentation, I will showcase excerpts from three recent RTMB-based assessments: an age-structured model for southern bluefin tuna, a length-based model for New Zealand red rock lobster, and a combined age- and length-structured model for New Zealand bluenose. These examples illustrate how RTMB promotes reproducibility, collaboration, and rapid iteration in fisheries science.
Presentation slides: Webber CAPAM_RTMB_workflow.pdf