CFE 2019: Start Registration
View Submission - CMStatistics
B0726
Title: Dynamical mixture modelling of spatial processes Authors:  Thais Fonseca - Universidade Federal do Rio de Janeiro (Brazil) [presenting]
Alexandra Schmidt - McGill University (Canada)
Viviana Lobo - Universidade Federal do Rio de Janeiro (Brazil)
Abstract: Spatio-temporal processes in environmental applications are usually assumed to follow a Gaussian model, possibly after some transformation. Flexibility to the usual Gaussian assumption is added by modelling the process as a scale mixture between a Gaussian and log-Gaussian process. The scale is represented by a process and it is allowed to depend on covariates. The resultant kurtosis varies with location, allowing the time series at each location to have different distributions with different tail behaviour. Regarding the temporal dependence, a dynamical model based on state equations is assumed for the scale process and a computationally efficient estimation algorithm based on sequential Monte Carlo is proposed. An application to maximum temperature data illustrates the effects of altitude in the variability of the process and how this dependence may change over time.