| Summary: | Globally, the number of threatened species is increasing and conserving them is a high priority to the scientific community. Assessing the status of these protected species is challenging due to missing, contested, and or contradictory data streams. Integrated models (IMs) provide a statistical framework for combining disparate data sources to inform derived variables that managers use to measure recovery efforts. This dissertation offers several novel application of IMs for assessing vital rates and population status of loggerhead sea turtles, Chinook salmon, and marine mammals in the United States. While Chapters 2 and 3 adhere to the statistical application of IMs, Chapters 4 and 5 use heuristic bioenergetic IMs to estimate predation rates of threatened Chinook salmon populations by protected marine mammal populations. The first data chapter examines how integrating mark-recapture, and skeletochronology data can improve estimates of growth for a population where ages are not directly observable. Using simulation experiments, I demonstrate that the integrated model estimates for the ages and growth parameters are unbiased for sample sizes higher than 50, and there is little misspecification of the growth process using AIC. For Western Atlantic loggerheads (Caretta caretta), the integrated model with the lowest AIC is the von Bertalanffy growth process with the mean growth coefficient (μ_k) equal to 0.076 year-1 and a mean asymptotic straight carapace length (μ_(L_∞ )) equal to 92.1 cm which is similar to other recent studies. The parameters for the growth model with integrated data are similar to those using only skeletochronology data; however, the uncertainty of the age estimates are smaller. The estimated mean age at stranding and final recapture of tagged turtles is 13.5 years and 14.6 years, respectively. Assuming the size at sexual maturity is 95% of the asymptotic size, the mean and 95% predictive interval for age at sexual maturity is 38 (29, 49) years. For populations where individual growth is variable, and the age of the animals is unmeasurable, this modeling approach of integrating multiple sources of information provides valuable insight for managers. In Chapter 3, I examine how integrated models incorporating spatiotemporal covariance between the time-series of observations in a network of streams can reduce bias in the parameters and derived variables of the model. Using simulation experiments, I test the ability of the estimation model (EM) to recreate the parameters of the operating model (OM) with temporal, spatial and spatiotemporal processes (e.g., redd density, proportion female, pre-spawn survival) that are similar to the processes used to estimate the abundance of Chinook salmon (Oncorhynchus tshawytscha) in tributaries of the upper Snake River Basin, US. I then applied the EM to a network of 19 tributaries in the upper Snake River Basin that have some combination of redd count surveys, mark-recapture, and sex and pre-spawn survival information from carcass surveys. From the simulation experiments, failure to match the spatiotemporal covariance for the processes in the EM and OM resulted in biased parameter estimates and increased uncertainty in spawner abundance. Applying the model to the upper Snake River Basin, the EM with the lowest AIC included spatiotemporal covariance for the total abundance, the redd density, the proportion female, and the pre-spawn survival. Relative to models with no spatiotemporal covariance, the model with the lowest AIC resulted in reduced uncertainty for the processes in years with missing data. This chapter supports including spatiotemporal processes in models where time and distance are unlikely to lead independent observations. Chapter 4 is an integrated ecosystem model to evaluate the marine mammal predation of Chinook salmon in the inland waters of Puget Sound, Washington. Using a bioenergetics model, diet composition, and predator abundance, I estimate the daily consumption rates of Chinook salmon by resident killer whales, California sea lion, Steller sea lions, and harbor seals. The 2015 Chinook salmon biomass consumed by killer whales is nearly equivalent to pinnipeds (570 metric tons versus 625 metric tons, respectively), but the number of Chinook salmon consumed by pinnipeds is higher than killer whales (8.75 million versus 83 thousand). The significant difference in the number of Chinook salmon consumed is due to size selectivity of harbor seals, which eat primarily juvenile smolts. Compared to 1970, I estimate the smolt mortality for Puget Sound Chinook salmon stocks from harbor seals has increased from 1.8% to 22.4%. This chapter demonstrates the importance of non-stationary interspecific interactions when evaluating species recovery. As more protected species respond positively to recovery efforts, managers should attempt to assess trade-offs between these recovery efforts and the unintended ecosystem consequences of predation and competition on other protected species. Chapter 5 extends the integrated bioenergetics model in Chapter 4 to pinniped and killer whale populations throughout the northeast Pacific. Using the same bioenergetics approach, this integrated model includes temporal and spatial distributions for the predators and Chinook salmon prey. I estimate that the biomass of Chinook salmon consumed by the marine mammals has increased ~150% since 1970, while the biomass of Chinook salmon harvest from commercial fisheries has declined ~50% during the same period. Changes in predator consumption are not uniform across the domain. Large killer whale population increases in British Columbia, southeast Alaska, and western Alaska have led to commensurate increases in consumption. Conversely, pinniped populations have increased along the entire west coast, but we only predict significant increases in consumption in the Salish Sea and Columbia River. Long-term management strategies for Chinook salmon will need to consider potential conflicts between recovery for the aggregate predator populations and locally endangered predator and prey populations. We can improve the management of protected species by using IMs to resolve information from multiple data streams. While researchers should consider extensive simulation testing to ensure that the parameters of their IM are estimable, there is no guarantee that IMs are the correct representation of the natural processes. Furthermore, the raw data needed to build and IM can be challenging to access. Past research treating summarized quantities as data or parameters for a population model limits our ability to determine if the lack of model fit is due to the inputs or the model itself. In the first two chapters, I offer IMs with explicit representations of the uncertainty in the models and the data. In the next two chapters, I offer IMs with a heuristic approach based on summarized data and statistics. Moving forward, future IM research efforts, especially in the protected species arena, will require continued collaboration between researchers and agencies, increased transparency of the modeling process, access to the raw data, and efforts by quantitative scientists to find novel solutions to critical ecological questions.
|
|---|