Advancements and Best Practices in Quantitative Population Modeling

Science informs management of living biological resources through population modeling. In both fisheries and wildlife management, a broad array of techniques are used to assess the condition of living resources and advise on sustainable management decisions. These quantitative techniques range in complexity, and the approach used in a given scenario is typically dictated by the available data. As new modeling techniques, data sources, and advanced computing resources become available, the range of available analytical techniques continues to expand. However, with increasing capability comes increasing responsibility to understand what is available, what is appropriate for a given situation, and how the approach should be applied given the available information and scientific understanding. This joint conference presents an opportunity for participants from marine, terrestrial, and freshwater backgrounds to share successes in resource management as well as broaden the conversation about emerging techniques and best practices in population modeling. We encourage participation in this symposium from government scientists and management agencies, industries, non-governmental organizations, technology-partners, research institutions, citizen-scientists, academics (including students), and other stakeholders with a diversity of experiences. Through presentation and discussion, this symposium will bring participants together to discuss the advances and challenges associated with developing population modeling techniques, determining when to apply them, and translating them into effective resource management policies, as well as hear about current and past successes in those endeavors.

8:00AM A Bad Practices Guide to Modelling Growth in Fisheries Stock Assessment
  Mark Maunder, Kevin Piner, Hui-Hua Lee
Growth is one of the key biological processes included in population dynamics models used in contemporary fishery stock assessment and is generally considered the most well understood and best estimated. However, our knowledge about growth may not be as good as we believe nor as good as required for estimating the population dynamics. Within dynamic models, the use of growth can be portioned into two components: 1) converting quantities from weight into numbers and vice versa (the stock assessment model is typically in numbers, but the catch and other quantities in weight) and is important for yield calculations; 2) and fitting to length composition data. The former might be well informed in many cases because these analyses are dependent on the most numerous age classes (for which growth is often well know), while the latter is highly dependent on the asymptotic length (which is often poorly known). Several issues have been identified that can bias growth estimates in fisheries stock assessment models including: length-based sampling, length-based selectivity, age-based availability, aging error, appropriately combining data, correlated residuals, and incorrect functional forms. These issues are discussed in the context of stock assessment modeling and management advice.
8:20AM Cohort Resonance: How Age Structured Populations Respond to Environmental Variability
  Louis W. Botsford, J. Wilson White, Alan Hastings, D. Patrick Kilduff, Annie E. Schmidt, Lauren Yamane, Mikaela Provost
In the early 2,000s population biologists discovered that the response of age structured models with density-dependent recruitment could be understood as a filter that was more sensitive to some environmental frequencies than others: frequencies near the inverse of a generation time, and 2) very low frequencies; hence the name Cohort Resonance. Others showed that this could explain why population variability increased with fishing. It is also explains the phenomenon of cyclic dominance in sockeye salmon. Because both paleo records and GCMs show changes in the frequency of physical indicators, the question of their effect on wildlife populations often arises. A cohort resonance approach showed surprisingly that an increase in the frequency of ENSO would reduce variability and the probability of extinction in the population of Brandt’s cormorants on the Farallon Islands. The prediction of considerable variance at the cohort frequency in the temperature of a California salmon spawning stream raised the question of whether that would increase probability of extincton. Surprisingly it increased variability, but not extinction. The different roles in extinction, of the two cohort resonance frequency bands raised the question of how different life history characteristics cause the two, a question answered for Pacific salmon and Atlantic cod.
8:40AM Dynamics of Fish Production in a Rapidly Changing Ecosystem: Coupling Age Structured Stock Assessment and Fish Bioenergetics Models of Lake Trout in the Main Basin of Lake Huron
  Ji X. He, James R. Bence, Charles P. Madenjian
We coupled age structured stock assessment and bioenergetics models to describe dynamics of lake trout production in the main basin of Lake Huron. Production to biomass (P:B) ratios were calculated from annual production and beginning-of-the-year biomass. P:B ratios varied among juveniles (ages 1-2), subadults (ages 3-5), and adults (> age-5). These ratios did not change after the 2003 population collapse of the alewife, which had been a major prey fish. Lake trout adult biomass and production built up when fishery harvest was less than 50% of subadult production. Juvenile and subadult biomass and production decreased substantially after 2003 because increased recruitment of wild lake trout did not fully offset a decline in hatchery recruitment. Adult biomass peaked during 2006-2007, adult production peaked in 2010, and the increases occurred when annual fishery harvest was less than 50% of adult production. Production of adult lake trout was sustained after the alewife population collapse as evidenced by these constant P:B ratios. Such stability reflects adaptability of lake trout as the native apex predator and adequate harvest control during major changes in structure and function of the ecosystem, while the effects of decreased growth and reduced mortality on production offset each other.
9:00AM Application of State Space Stock Assessment Modeling to Lake Whitefish (Coregonus clupeaformis)
  Emily M. Liljestrand, James R. Bence, Jonathan J. Deroba
State space modeling (SSM) can improve upon widely applied approaches to fitting population models, such as penalized likelihood, because process error (resulting from natural variation in the system) and observation error (resulting from measurement technique) are each estimated, rather than assuming one is known or calculated from the other. We used an age-based SSM, where “states” are abundances and quantities (such as catchability and natural mortality) that change over time as a function of the previous state and process error. The time series of observed data depend on the states and observation error. We apply this approach to Lake Whitefish (Coregonus clupeaformis) catch and effort data in the 1836 Treaty-ceded waters of the Great Lakes. We compare model fits when process errors follow a random walk (RW) or white noise (WN) structure. We present model estimates and how they compare to results from methods currently used in treaty-water management that do not estimate process and observation errors. We will also present preliminary and planned simulation testing, and discuss the implications for future application. This research is part of a program intending to determine to what extent theoretical advantages of SSMs can be realized in the face of multiple time-varying processes.
9:20AM Bayesian Hierarchical State-Space Models for Life Cycle Modeling and Abundance Predictions with an Application to Delta Smelt
  Leo Polansky, Ken B. Newman, Lara Mitchell, William E. Smith
Life cycle population modeling seeks to quantify vital rates such as recruitment and life-stage specific survival, and to describe them as functions of covariates. Stage structured models are a natural mathematical framework for life cycle modeling, and hierarchical state-space models in turn embed these mathematical models into ones useful for statistical inference. Life cycle modeling of Delta Smelt (Hypomesus transpacificus), an annual Osmerid species endemic to the San Francisco Estuary, revealed several outstanding general methodological challenges for application of hierarchical state-space models. These included biological process parameter estimability given biased and noisy abundance indices, predictor variable measurement uncertainty, and the need to compare millions of models representing the different combinations of top down and bottom up processes acting on life stage specific vital rates. Simulation experiments were carried out to test the performance of Bayesian inference and stochastic search variable selection given these challenges. We applied the methodology to 21 years of Delta Smelt abundance indices to identify whether and how different habitat features explained their vital rates. Finally, we developed a population viability analysis framework that included vital rate covariate joint distributions, parameter uncertainty, and process noise to quantify covariate effect sizes and forecast abundances under different management regimes.
09:40AM Break
1:10PM Exploring Alternatives to the Multinomial for Fitting Composition Data within a Stock Assessment Simulation
  Nicholas Fisch, Robert Ahrens, Ed Camp, Kyle Shertzer
Fisheries stock assessments have traditionally represented age and size composition data using the multinomial likelihood, however the multinomial cannot account for the correlations and overdispersion that exist in the data. Not accounting for these phenomena can affect assessment performance. Methods to remedy this have included down-weighting composition data within assessment and using alternative likelihoods to the multinomial. Down-weighting composition data in stock assessment is laborious and does not ultimately account for the correlation in the residuals, and alternative likelihoods for composition data have not all been evaluated using stock assessment simulation. To evaluate the performance of alternative likelihoods in fitting composition data, we developed a spatially-explicit age-structured simulation model that mimics correlation structure observed in composition data. We fit assessment models to simulated data and assessed the performance of various composition likelihoods in estimating stock dynamics and quantities of management interest. We expect to present preliminary results regarding model performance with regards to likelihood choice and the degree of spatial clustering of data. This study may have important implications for the determination of which likelihood is appropriate conditional on how data were sampled, and could provide information on how one might sample data to more appropriately match a likelihood.
1:30PM Projections for Adaptive Management of Biomass and Yield in Marine Protected Areas
  Caren Barceló, J. Wilson White, Louis W. Botsford, Alan Hastings
Implementation of marine protected areas (MPAs) to benefit biodiversity and fishery yield is becoming more widespread globally. In most cases, the implementation and management of MPAs is not coordinated with existing conventional fishery management efforts. Consequently, the dependence of fishery population dynamics on: 1) the spatial pattern of the MPAs, and 2) connectivity through larval dispersal between MPAs and fished areas, is not incorporated into the population dynamic considerations of conventional fishery management. Currently the greatest management need for an integrated understanding of spatio-temporal fishery dynamics is to describe the short-term ‘transient’ effects immediately following implementation of MPAs. Previously we have projected transient responses for adaptive management of California’s MPA network, using an approach that accounts for the tempo of ‘filling in’ of the age structure previously truncated by fishing. We extend those results to the adaptive management of a fisheries outside MPAs, projecting the response of fishery yield forward through the “filling-in” period and beyond. We show how population persistence and yield depend on existing and adjusted levels of harvest, as well as the level of connectivity afforded by larval dispersal, in a way that reflects the value of Spawning Potential Ratio (SPR) maintained in conventional fishery management.
1:50PM Reconstructing Salmon Runs to Sustainable Fisheries Management
  Lingbo Li, Pieter Van Will
Sustainable management is challenging when and where multiple populations of varying productivity are harvested together. To avoid management errors that may cause serious conservation harm to those depleted populations and economic harm to the abundant population fisheries, salmon managers need to understand the expected stock-specific harvest rates at each management area. Here we employed 10 co-migrating chum salmon stocks with a large range of abundance returning to spawning grounds in the southern British Columbia and Puget Sound as an example to show how our Quantitative Population Model (QPM) can help make management decisions. The model was developed based on catch, escapement, and genetic stock identification (GSI) data using R and AD Model Builder. We applied a Bayesian approach incorporating experts’ knowledge on arrival timing and abundance of each stock. We included all possible pathways, which were determined based on GSI and escapement data, for each stock from their initial arrival area to their spawning grounds. We presented preliminary stock-specific estimates and sensitivity analyses. We finally discussed how model results can be used to advise on sustainable management decisions and how the model can be adapted to a variety of fisheries.
2:10PM A Temporally-Stratified Extension of Space-for-Time Cormack-Jolly-Seber for Migratory Fish
  Dalton Hance, Russell Perry, Adam C. Pope, John M. Plumb
Understanding drivers of temporal variation in demographic parameters is a central goal of mark-recapture analysis. To estimate survival of migrating animal populations, “space-for-time” mark-recapture employs fixed sampling locations in space to monitor marked populations as they migrate rather than the standard practice of using fixed sampling points in time. Under the space-for-time model structure, modeling the effect of time-varying covariates on model parameters is complicated due to unknown passage times for individuals that are not detected at monitoring sites. To overcome this limitation, we extended the Cormack-Jolly-Seber (CJS) framework to estimate temporally-stratified survival and capture probabilities by including a discretized arrival time process in a computationally-efficient Bayesian framework. Our model allows flexibility in the functional form including incorporating temporally stratified covariates and random effects. We demonstrate our framework by estimating daily survival, capture, and arrival probabilities at each dam of the Federal Columbia River Power System based on hundreds of thousands of individually tagged juvenile salmon released into the Snake River, USA. We illustrate the relevance of our model to fisheries managers by deriving quantities of interest that cannot be otherwise obtained with standard CJS approaches, such as smolt-to-adult return ratios based on day of passage at a given dam.
2:30PM Abundance and Outmigration Timing for Naturally-Produced Snake River Fall Chinook Salmon over a Period of Population Recovery
  John M. Plumb, Russell Perry, Dalton Hance, Ken Tiffan
ESA-listed fall Chinook salmon in the lower Snake River, Idaho have undergone marked increases in abundance from near-extirpation levels. This recovery may be attributed to reduced harvest, stable minimum spawning flows, summer flow augmentation, predator control, increased hatchery supplementation, improved dam passage structures, and the implementation of summer spill operations. Estimating natural fish abundance was complicated by the protracted migration of juveniles, varying marking rates of hatchery juveniles at release, variation in daily fish collection at Lower Granite Dam, and no fish sampling in winter. We used a hierarchical Bayesian time-varying Cormack-Jolly-Seber model to estimate fish arrival and collection at the dam that could leverage information from different data sources to estimate abundance We then used a log-normal kernel to impute fish abundances passing the dam during the winter without fish sampling. From 1992 to 2017, adult and juvenile fish abundances increased by an order of magnitude, concomitant with reduced subyearling outmigration timing and lower fractions of yearling outmigrants. Adult-to-smolt recruitment indicates that current adult returns are either at or above capacity for juvenile production measured at the dam.
2:50PM Refreshment Break
3:20PM Integrating the Effect of Time-Varying Covariates within a Proportional Hazards Survival Model for Juvenile Chinook Salmon in the Yakima River, WA
  Adam C. Pope, Tobias J. Kock, Russell W. Perry, Amy C. Hansen
Mark-recapture methods are commonly used to estimate demographic rates such as population survival. Fitting parameters of these models as a function of covariates has long been a staple of these analyses. However, in space-for-time mark-recapture models applied to migrating fishes, time of arrival and exposure time for non-detected individuals is unknown, and so incorporating the effect of time-varying individual covariates over exposure histories has remained challenging. This limitation motivated us to consider novel techniques to integrate over missing travel time data, including the use of proportional hazards to model travel time as a function of time-varying covariates. We analyzed data from a mark-recapture study on juvenile Chinook salmon migrating in the Yakima River in Washington State using a Cormack-Jolly-Seber model framework and complete data likelihood structure. Imputation of unobserved travel times via MCMC allowed us to estimate survival via a proportional hazards model. The effect of time-varying covariates such as river discharge were integrated over the exposure history of each individual. We found a link between exposure history to these covariates and survival. This application provides a valuable example for integrating temporal covariates to evaluate the relationship between environment, exposure history, and demographic vital rates in other populations of interest.
3:40PM Survival Estimates from an N-Mixture Model for Young-of-the-Year Largemouth Bass within the Sacramento-San Joaquin Delta
  Brock Huntsman, Frederick Feyrer, Matthew Young
The Sacramento-San Joaquin Delta in California has experienced dramatic ecological changes that have coincided with the proliferation of non-native Largemouth Bass (Micropterus salmoides, LMB). The change from dynamic tidally-influenced habitat to stable lacustrine habitat, along with non-native aquatic vegetation expansion, has been hypothesized responsible for favorable conditions for LMB survival and recruitment. We developed an open N-mixture model to test this LMB hypothesis across four lakes with varying degrees of tidal influence within the delta. We focused our investigation on young-of-the-year (YOY) LMB because this is a critical stage during LMB development. The N-mixture model was constructed from monthly boat electrofishing surveys conducted in 2010, 2011 and 2014 within each tidal lake. We found detection efficiency increased and decreased with effort and turbidity, respectively; while YOY LMB survival was highest in habitats with high submerged aquatic vegetation density and intermediate water temperatures (~22.5 ℃). Surprisingly, we did not detect a strong conductivity effect on LMB survival, but this result was likely an artifact of our sites not exceeding physiological conductivity limits. Our results are consistent with previous delta studies suggesting that LMB benefit from the stable conditions potentially created by rapid colonization by non-native submerged aquatic vegetation.
4:00PM Harvest Slots As a Management Tool to Improve Marine Recreational Fishing Opportunities and Sustainability in Gulf of Mexico Red Snapper
  Erin Bohaboy, Shannon Cass-Calay, William Patterson
The Gulf of Mexico Red Snapper (Lutjanus campechanus) stock was depleted to historically low levels by the 1990s and remains under a rebuilding plan despite years of restrictive fishing regulations, including daily bag limits, minimum size requirements, and seasonal closures in the recreational fishery and individual fishing quotas in the commercial fishery. Harvest slots (a minimum and a maximum size regulation for harvested fish) combined with sufficiently high post-release survival could potentially shift fishing mortality to younger, smaller fish while maintaining older mature female fish in the stock that contribute disproportionally to spawning biomass. We performed stock projection simulations in Stock Synthesis version 3.30 to quantify the effects of slot limits (minimum length at retention = 16–20″, maximum length at retention = 22–34″ or no maximum length limit) and discard mortality (0–30%) on catches and stock status. Changes in projected catch, harvest, and season length were highly dependent on assumptions regarding the distribution of fishing mortality and effort among recreational and commercial fleets. Results suggest that a wide slot limit (e.g., 16–32″) combined with reductions in discard mortality (25–50%) may result in sufficient increases to recreational harvest season length to warrant consideration of this previously unused management tool.
4:20PM Measuring Ecological Changes in the Context of Causal Analysis
  Andrew Deines, Ann Michelle Morrison, Roxolana Kashuba
Fisheries and wildlife scientists are often called upon to assess ecosystem “health” or “integrity.” However, in practice these vague terms do not automatically correspond to specific and measurable community metrics. There is a large assortment of such metrics available to attempt to quantify the components and properties of an ecosystem, which emerge at various levels of biological organization. But it is not always obvious which metrics most appropriately evaluate a given assessment goal. In the context of causal analysis, or identifying the cause of an observed ecological state, it is useful to separate metrics that are better suited to integrating across multiple drivers of ecosystem change (e.g. benthic indexes may respond to multiple chemical and physical disturbances) from metrics better suited to diagnosing particular drivers (e.g. fish size distribution responds strongly to fishing). To this end, we report on our review of common metrics used in ecological assessments of fish and fisheries at different levels of biological organization (i.e., individuals, populations, assemblages/communities, and ecosystems). We also discuss quantitative approaches to evaluate differences in metric values between treatment and control groups, changes in time series, and in assessing casual inference
4:40PM How Well Do You Really Know Your Fish Population? Exploring Methods to Assess Freshwater Fish Populations
  Alan Ward, Weston Pearce
Being able to correctly assess fish populations in a given water is the first, and possibly most important, step in modeling these populations. Fisheries managers have a variety of methods available to assess freshwater fish populations, however, many are stuck utilizing past routines for various reasons. Attempts have been made in recent years to standardize sampling methods across jurisdictional boundaries so that comparisons can be made on larger scales. In the State of Utah, we made a sweeping statewide switch in fishery monitoring methods to accommodate the efforts of the American Fisheries Society to standardize sampling methods. However, managers at Strawberry Reservoir found the methods employed to be severely lacking. During 2017 and 2018 we employed alternative monitoring methods in order to test the ability of our standardized netting protocols, as well as historic protocols, to assess fish populations. We also used a variety of fish marking techniques and other sampling methods to better describe fish population structure and dynamics. We found that many of our fish species were not being adequately sampled with the standardized and traditional methods. Our new methods allow for much more robust samples, better population structure and size estimates, with reduced sampling effort.

Organizers: Jeffrey Vieser, Melissa A. Karp, Patrick Lynch, Daniel Goethel, Aaron Berger
Supported by: Jeffrey Vieser, Melissa Karp

Location: Reno-Sparks CC Date: September 30, 2019 Time: 8:00 am - 5:00 pm