Large point referenced datasets are common in the environmental and natural sciences. The computational burden in fitting large spatial datasets undermines estimation of Bayesian models. We explore several improvements low-rank and other scalable spatial process models including reduction of biases and process-based modeling of “centers” or “knots” that determine optimal subspaces for data projection. These include point-process models as well as nonparametric Dirichlet process models. We also consider alternate strategies for handling massive spatial datasets. One approach concerns developing process-based super-population models and merging them with Bayesian finite-population sampling techniques for spatial and spatial-temporal data. We also explore model-based simultaneous dimension-reduction in space, time and the number of outcome variables leading, essentially, to flexible classes of spatial-temporal factor models. Flexible and rich hierarchical modeling applications in forestry are also demonstrated.

Keywords: Hierarchical models; Markov chain Monte Carlo; Predictive processes; Super-population models

Biography: Sudipto Banerjee is Professor of Biostatistics in the School of PublicHealth,University of Minnesota. His research focuses upon statistical modeling and analysis of geographically referenced datasets, Bayesian statistics (theory and methods), interface between statistics and Geographical Information Systems, and statistical computing. He has published over sixty peer-reviewed journal articles, several book chapters and has co-authored a book titled “Hierarchical Modeling and Analysis for Spatial Data”. In 2009 he was honored with the Abdel El Sharaawi Young Investigator Award from the The International Environmetrics Society. This is awarded to a young investigator below the age of 40 who has made outstanding contributions to the field of environmetrics.