CANSSI at the SSC: A Statistical Science Showcase

CANSSI at the SSC: A Statistical Science Showcase
Organizer and Chair: John Braun (University of British Columbia)

ALISHA ALBERT-GREEN, University of Western Ontario
A Spatio-Temporal Cluster Process for Modelling Storm Cells  [PDF]
Storm cells are the smallest component of a storm-producing system. A cluster of such cells is referred to as a storm and a storm system consists of a cluster of storms. This research develops a model for these storm cells over space and time. Specifically, we extend the modified Thomas process, which is commonly employed for the analysis of clustered point processes, to account for the hierarchical clustering present in our data. We do this by allowing the parents to follow a doubly stochastic process, namely a log-Gaussian Cox process. This model is applied to storm cell data from the Bismarck radar station in North Dakota, USA and parameter estimation is done using minimum contrast estimation.
DEREK BINGHAM, Simon Fraser University
Design of Experiments in the CANSSI CRT ``Statistical Modeling of the World: Computer and Physical Models in Earth, Ocean, and Atmospheric Sciences''  [PDF]
Design and analysis of experiments continue to make important and far-reaching contributions to scientific investigation. Indeed, it is hard to imagine any field where it does not play a role. In the CANSSI Collaborative Research Team project ``Statistical Modeling of the World: Computer and Physical Models in Earth, Ocean, and Atmospheric Sciences'', experimental design is used for both physical and computer experiments. This talk will outline the role of experimental design for both settings, and a new design approach for exploring non-convex regions that are common in geophysical application will be presented.
RADU V. CRAIU, University of Toronto
Modern Research Topics in Copula Methods: A Report from the CANSSI Collaborative Team Project  [PDF]
We will summarize some of the research activities undertaken by the members of the Copula CANSSI Collaborative Research Team: Christian Genest, Louis-Paul Rivest, Elif Acar, Harry Joe, Johanna Neslehova, Jean-Francois Quessy, Bruno Remillard and Radu Craiu.

Focus will be placed on our own research efforts on inference for conditional copula models in which the calibration function depends on multiple covariates. We will discuss a novel approach that relies on a Gaussian process prior used in conjunction with a single index model formulation. The methodology will be illustrated using simulations and real data.