PS 81-186 - Models migrating toward data: Viability analysis for Pacific salmon using integrated population models

Friday, August 11, 2017
Exhibit Hall, Oregon Convention Center
Eric R. Buhle, Conservation Biology, NOAA Northwest Fisheries Science Center, Seattle, WA; Quantitative Consultants, Inc., Boise, ID, Mark D. Scheuerell, School of Aquatic and Fishery Sciences, University of Washington, Seattle, WA and James T. Thorson, Northwest Fisheries Science Center, National Oceanic and Atmospheric Administration, Seattle, WA

Population viability analysis (PVA) remains a critical component of conservation management for many imperiled species, providing a quantitative framework for comparing relative risks under alternative hypothesized states of nature or management scenarios. Recent developments in population-dynamics modeling emphasize the need to separate observation and process error and show that conflating these sources of uncertainty can lead to biased parameter estimates and substantively affect projected dynamics. The six species of Pacific salmon and steelhead trout (Oncorhynchus spp.), many populations of which are listed under the US Endangered Species Act, are a representative example of the use of PVA to assess current status and prioritize recovery planning efforts. However, most PVA models for Pacific salmonids currently use a two-stage approach to parameter estimation that does not distinguish observation from process error. We developed an integrated population model (IPM) of single life-stage (adult-to-adult) dynamics in which the process model can accommodate time-varying age structure, the presence of hatchery-origin fish, and multiple populations sharing common environmental influences. The observation model involves only a few basic data components, which are widely available for many Pacific salmonid populations. Parameters are estimated in a hierarchical Bayesian framework using R and Stan software.


We compare the IPM approach to a traditional two-stage spawner-recruit analysis by run reconstruction (RR) using data from 24 populations of spring/summer Chinook salmon in the Snake River Evolutionarily Significant Unit (ESU), modeled either separately or as a multi-population ensemble. We show that the IPM produces weaker (and biologically more plausible) estimates of the strength of density-dependence. Projections of 50-year quasi-extinction risk are more conservative (i.e., pessimistic) under the IPM than under the RR, and the ranking of relative risk among populations differs under the two approaches. At the ESU level, the RR model suggests that under average environmental conditions, the average population could withstand an additional 85% mortality before the deterministic long-range growth rate falls below 1, compared to 60% under the IPM. We conclude that integrated modeling offers a coherent, flexible statistical framework that can be extended to more complex salmonid life cycles. Its advantages over the traditional two-stage approach include the ability to handle missing data, a seamless connection between parameter estimation and simulation, and the ability to distinguish true dynamical variability from observation noise. These differences have significant implications in the PVA context. Our code is freely and publicly available as an R package.