Models Migrating Toward Data: An Integrated Statistical Framework for Salmonid Life Cycle Modeling

Stage-structured population models are a flexible framework for synthesizing demographic information and generating predictions about life-history bottlenecks and potential responses to management interventions. Traditionally, most applications of stage-structured models to salmonids have taken a piecemeal approach to parameter estimation. Estimates of transition probabilities or recruitment functions are derived independently from disparate data sources (e.g., tagging, trapping, or dam counts) using various statistical methods and subsequently combined, precluding a formal accounting of prediction uncertainty. Here we outline an alternative approach in which all parameters are estimated jointly by fitting the model to all the available data in an integrated statistical framework. Casting the life cycle model within a Bayesian state-space framework makes it possible to distinguish process and observation error, which can substantially influence estimates of extinction risk and other predicted quantities of interest. In addition to providing a consistent method for assimilating diverse data types and propagating their joint uncertainty, the integrated approach facilitates formal comparison among alternative biological hypotheses via weights of evidence given the data. We illustrate components of the integrated approach with examples from Chinook salmon (Oncorhynchus tshawytscha) in the Salmon River, Idaho, and steelhead (O. mykiss) in the Skagit River, Washington.