Building a state-space life cycle model for naturally produced Snake River fall Chinook salmon

Abstract

In 1992, Snake River basin fall Chinook salmon (Oncorhynchus tshawytscha) were listed for protection under the U.S. Endangered Species Act (NMFS 1992) and the population remained below 1000 individuals until 2000. Since then, returns from natural production has rebounded to over 20,000 spawners owing to a host of factors including reduced harvest (Peters et al. 2001), stable minimum spawning flows (Groves and Chandler 1999), summer flow augmentation (Connor et al. 2003), predator control (Beamesderfer et al. 1996), hatchery supplementation (Rosenberger et al. 2017), improved juvenile passage structures (Adams et al. 2014), summer spill operations (Perry et al. 2006; Adams et al. 2008), and periods of favorable ocean conditions and food availability (Logerwell et al. 2003; Peterson et al. 2014). Given this change in abundance coincident with numerous management actions and fluctuation in environmental drivers, quantifying which factors contributed to the observed rebound in natural production can provide critical insights into future management actions for this at-risk population. Multistage life cycle models provide a powerful analytical framework for understating how each life stage of a population contributes to population growth rate (Moussalli and Hilborn 1986; Greene and Beechie 2004). Multistage models may also be used as an analytical framework to explicitly estimate demographic parameters of a population model. This approach has an advantage over single-stage stock-recruitment models by allowing population growth rates to be partitioned among life stages rather than aggregated over an entire life cycle. Such partitioning allows for estimating 1) stage-specific density dependence, and 2) stage-specific effects of environmental factors or management actions. For example, Zabel et al. (2006) estimated parameters of a multistage model used in the context of a population viability analysis for spring/summer Chinook salmon in the Snake River, but such an approach has yet to be applied to fall Chinook salmon in the Snake River basin. Typically, data informing estimates of abundance at particular “check points” in the life cycle determines the complexity of the multistage model that can be fit to the data. For fall Chinook salmon, we are developing a two-stage model that encompasses: 1) upstream passage of spawners at Lower Granite Dam (LGR) to the subsequent downstream passage of their progeny at the dam, and 2) downstream passage of juveniles at LGR to their subsequent return from the ocean and passage at the Dam 2‒6 years later. This approach partitions the life cycle of fall Chinook salmon both spatially and temporally, which allows us to fit and compare alternative models with covariates specific to each stage. Our previous report to the ISAB (Zabel et al. 2013) detailed methods for estimating abundance of naturally produced adults and juveniles passing Lower Granite Dam, which provides the requisite data for fitting a two-stage model. The intent of this report is to describe the structure of the two-stage life cycle model, present preliminary results from fitting the model to data, and outline future directions and developments. As is clear from the diversity of models presented in this report, “life cycle models” range from very simple theoretically based population models (e.g., the Beverton-Holt stock- recruitment model) to very complex spatially explicit simulation models linked to hydrosystem hydrodynamic models (e.g., the COMPASS model for a single transition in a life cycle model, Zabel et al. 2008). We chose to develop a model of intermediate complexity that casts the two- stage life cycle model in a state-space framework (Newman et al. 2014). We chose to use a state-space framework implemented in a Bayesian framework because: • It provides both a statistical estimation framework for retrospective statistical analysis and a stochastic simulation framework for prospective analysis to evaluate alternative management actions. • Abundance estimates are uncertain. A state-space framework accounts for observation uncertainty in the abundance estimates and other data (e.g., age structure) while simultaneously estimating process uncertainty. • It allows for missing data. By drawing missing data from an appropriate probability model, uncertainty owing to missing data can be propagated without having to omit data or assume fixed values for missing data. Thus, a two-stage state-space life cycle model for fall Chinook salmon strikes an appropriate balance between model complexity, tractability, and applicability given the goals of performing both retrospective and prospective analysis to guide future management of this population.