PS 79-201
Forecasting population size and risk: Accounting for time-varying demography and environmental drivers

Friday, August 15, 2014
Exhibit Hall, Sacramento Convention Center
Eric R. Buhle, NOAA Northwest Fisheries Science Center, Seattle, WA
Mark D. Scheuerell, School of Aquatic and Fishery Sciences, University of Washington, Seattle, WA
Richard W. Zabel, NOAA Northwest Fisheries Science Center, Seattle, WA
Background/Question/Methods

Predictive models of population dynamics that can forecast changes in abundance over short time horizons are a critical management tool for many imperiled species. The utility of such forecasts depends in part on a characterization of their uncertainty, which allows decision-makers to weight the costs and benefits of various scenarios by their probability of occurrence. For many species in which only certain life stages are directly observable, one underappreciated source of uncertainty in population forecasts is temporal variation in age-specific vital rates, which are often assumed to be constant. Another source of uncertainty arises when environmental variables are used as leading indicators in forecasts, because the strength of their effects on vital rates may change over time. We developed a model, tailored to anadromous Pacific salmonids (Oncorhynchus spp.), that predicts adult abundance and age distribution within a cohort based on projections from observed juvenile abundance. Using a Bayesian state-space statistical framework, we allow vital rates (survival and maturation probabilities) as well as the effects of environmental covariates to fluctuate through time, and incorporate this process variability along with parameter uncertainty in the forecasts.

Results/Conclusions

We illustrate the method with applications to threatened Chinook salmon (O. tshawytscha) and steelhead (O. mykiss), and compare the forecast performance of our approach with that of a traditional forecasting model based on within-cohort age composition that ignores process error and does not include environmental covariates. Relative to the traditional model, the state-space approach shows improved forecast skill (reduced root mean squared error and posterior predictive loss). This results both from the use of environmental leading indicators to predict survival, and from avoiding the unrealistic assumption of constant age structure, which can produce substantial bias (high or low) in traditional forecasts.