|PFRP Home > Protected Species Projects List|
A General Bayesian Integrated Population Dynamics Model for Protected Species
Reports (PDF): FY
2008, FY 2007, FY
2006, FY 2005, FY
Project researchers will develop a general Bayesian integrated model for protected species based on the model developed for the New Zealand sea lion by Hilborn et al. (in prep, see also Maunder 1999; Maunder and Hilborn 2000; Breen et al. in prep). and used, with modifications, for Hector's dolphin (Davies et al. 2001). This spatial and age-structured population dynamics model allows the inclusion of priors for many model parameters and the inclusion of multiple data sets. It includes both compensatory and depensatory mortality in the first year of life. Movement between subpopulations is a function of the population size relative to carrying capacity and distance between subpopulations. The data used in this model are absolute or relative abundance estimates by subpopulation, which can be age-structured or population estimates. The model also has a forward projection component that allows for decision analysis based on human-caused mortality (interaction with fisheries) that includes parameter uncertainty and process error (e.g., stochasticity in recruitment and survival, catastrophes, and regime shifts in carrying capacity). The model will be extended to include tagging data (with multiple recaptures, e.g., banding), catch-at-age/length/stage/sex data, environmental/habitat correlations, sex, structure, and stage structure. Sex and stage-structure are added because the biological behavioral characteristics of protected species often differs between sexes and as the individual progress through different stages. The inclusion of tagging data is important because tagging data is traditionally the main source of information for estimating many biological parameters of protected species. The type of tagging data available for protected species differs from the type of data available for fish populations. Most tagging data for protected species (e.g., banding, sight-resight) includes multiple resighting of the same individual. Project researchers will use the methodology of select software packages (e.g., MARK) combined with the relevant literature (e.g., Lebreton et al., and Maunder 2001b) to develop the tagging component of the integrated model. The management strategies to be modeled may include changes in bycatch rates, fishing effort, spatial or temporal allocation of fishing effort, or habitat levels. Methods to determine the effect of management measures on the associated fishery will be investigated (e.g., Maunder et al. 2000). The model will be programmed using the AD Model Builder (ADMB) software package (Otter Research) which is ideal to implement such a complex model. ADMB is a set of C++ libraries that aid in the development of non-linear models. ADMB provides formatted reading and writing of data structures, matrix algebra, a sophisticated and efficient minimization routine based on automatic derivative calculations and provides the Hessian matrix, and a Markov Chain Monte Carlo (MCMC) routine for sampling from the Bayesian posterior distribution. ADMB has been used to develop a general spatial and age-structured models for fisheries stock assessment, marine mammals, and a general age-structured catch-at-length analysis for tuna stocks. At the core of the ADMB is the AUTODIF library that is used for the MULTIFAN-CL stock assessment model (Fournier et al., 1998). Both posterior mode (penalized likelihood) estimates and full Bayesian integration will be implemented. The model will be developed as generally as possible to allow its application to many different species and for the investigation of many different assumptions and data sets.
In the first year, the model will be applied to the Hawaiian population of black-footed albatross and the northeastern Pacific stock of spotted dolphin. The types and quality of data available for these two populations differ greatly; so they make good contrast for applying a general model. During the first year project researchers will pursue collaborations with other interested researchers who wish to apply the model to their populations. In the second year the model will be applied to these populations, possibly including other Hawaiian albatrosses, other eastern Pacific Ocean dolphins, and sea turtles.
Year 1 funding for this 2-year project to be awarded in early 2003.
This page updated August 7, 2008