Dynamic Spatial Pattern Recognition in Count Data

Abstract

This study explores a Bayesian regression analysis for count data in the presence of spatial and temporal correlations. The contribution is to develop a regression model for count data that provides flexibility in modeling the complexity of zero-inflation, overdispersion, as well as spatial patterns in a dynamic manner. More importantly, it improves the computational efficiency via dimension reduction while handling the high-dimensional correlation structure in the data. The proposed model is applied to the survey data by the Northeast Fisheries Sciences Center (NEFSC) for estimation and prediction of the Atlantic cod in the Gulf of Maine—Georges Bank region. Both zero-inflated Poisson and negative binomial models are fitted. Model comparison shows the improvement in model fitting with consideration in the spatial-temporal correlation as well as the overdispersion in the count data.