A seminar by Professor Louis-Paul Rivest from Laval University
Title: Spatio-temporal modeling of fish stocks in the presence of outliers
With Hughes Benoit, Department of Fishery and Oceans of Canada, Ariane Boivin and Julien Kiswanda Compaore Université Laval
Abstract: The presentation focusses on the evolution of a cod stock (Gadus morhua L.) in the Gulf of St-Lawrence, Canada that has been monitored by trawl surveys for about 20 years. During that period, the survey area has not been covered uniformly by the sampled sites and one objective of this presentation is to investigate whether modern spatio-temporal statistical models can be used to correct the deficiencies of the sampling design thereby permitting a better understanding of the stock evolution over the period. The base model for abundance is a generalized Poisson regression with a latent spatio-temporal Gaussian process. Methods for fitting such a model to large samples will be discussed. The impact of large abundances, that cannot be accounted for by a standard Poisson-lognormal model, will also be investigated. Generalizations of the standard Poisson regression that account for such large values are proposed. The results of a simulation study comparing the prediction of the latent process at unobserved points for various specifications of the latent process are presented.