Bayesian Stock Assessment Using a Nonlinear State-Space Model

This paper presents a novel Bayesian approach to sheries stock assessment based on biomass dynamics models. These use a time series of annual catch and eeort data to model and forecast the current and future biomass of the stock. We integrate biomass dynamics models into the framework of nonlinear state-space methodology and thus allow for both observation and process error. In contrast to the ML approach via Kalman ltering which relies on linearity of state and observation equations and normal errors, the Bayesian paradigm can eeciently handle nonnormal errors as well as nonlinear transitions. To sample from the posterior distribution , we employ the Gibbs sampler in conjunction with the adaptive rejection Metropolis sampling algorithm. We illustrate this new methodology using the delay diierence model and demonstrate its superiority over existing techniques for analysing biomass dynamics models.