Schooler, S.L. and LeMoine, M., 2023. Integrated Population Models with application to Skagit River Chinook Recovery Evaluation. Skagit River System Cooperative, Burlington, WA. pp. 26.

To understand fish and wildlife population dynamics we must determine which internal and
external factors most influence demographic rates, and therefore drive changes in populations.
Integrated population models (IPMs) incorporate multiple data types throughout animals’ life
cycles, and incorporate separate process variability and sampling uncertainty, allowing for
increased accuracy of parameter estimates. The goal of this project was to develop, apply, and
evaluate a Bayesian state-space life cycle IPM for Skagit River Chinook Salmon (Oncorhynchus
tshawytscha) to enable inference about population dynamics from multiple sources of life stage
specific data including age composition, smolt abundance, escapement, and harvest. We used
non-least squares models to test effects of density dependence on freshwater and marine
productivity and determined that freshwater productivity was best described by a non-density
dependent function, while marine productivity was best described by a Ricker model. To
determine the ability of the IPM to detect influences of environmental covariates on productivity,
we compiled data on temperature, river flow, weather, and ocean condition indices. We
developed a state-space life cycle IPM that incorporated freshwater and marine life stages with
environmental covariates, and included functions for random process and observation error. We
evaluated if our model successfully described and linked freshwater and marine life stages of
Chinook Salmon using model convergence and fit metrics, comparison to parallel non-least
squares model results, and examined the ability of the model to correctly identify parameters
from simulated datasets. We found that the IPM successfully converged, fit the data well, and
appeared to estimate similar covariate coefficients and non-density dependent productivity to
single-stage NLS models. However, productivity and density dependence for simulated datasets
were poorly predicted, indicating poor parameter identifiability. Future work on this model will
involve improving parameter identifiability, further exploring the effects of covariates on life
cycle stages, and adding additional data.