Abstract

Data from egg sampling surveys often contain a mixture of zero observations and large count or density values, often with a high proportion of zeros. We will consider a particular dataset giving walleye Pollock egg counts from the western Gulf of Alaska, from Kodiak Island to Unimah Pass, in the years 1978-2000. The main interest will be to predict the intensity of eggs as a process varying in space and time (both within and between years). In this talk I will discuss a Bayesian approach for the analysis of such data. The excessive number of zeros in the data is taken into account by the use of a two stage modelling approach, resulting in a zero-inflated hierarchical space-time model. An underlying intensity process is assumed to both influence the probabilities of zeros and the amount of eggs in non-zero observations. Dependence of covariates and spatio-temporal correlations are taken into account through the modelling of the underlying intensity process. Fitting is performed through Markov Chain Monte Carlo simulations. Results for the walley Pollock egg counts will be presented.

This talk will be based on joint work with Ingunn Tvete and Bent Natvig at the University of Oslo and Lorenzo Ciannelli at Oregon State University.