SpatioTemporal: An R Package for Spatio-Temporal Modelling of Air-Pollution

Modelling of Gaussian spatio-temporal processes provide ample opportunity for different model formulations, however two principal directions have emerged. The data can be modelled either as a set of spatially varying temporal basis functions or as spatial fields evolving in time. This package provides maximum-likelihood estimation and crossvalidation tools for the first case. Development of the package was motivated by the need to provide accurate spatio-temporal predictions of ambient air pollution at small spatial scales for a health effects study. The package provides tools for extracting temporal basis functions from the data. It handles incomplete and highly unbalanced spatio-temporal sampling designs and allows for a flexible set of covariates and covariance structures to capture the spatial variability in the temporal basis functions and in the spatio-temporal residuals. Further, the package provides bias corrected predictions for log-transformed data, cross-validation tools and rudimentary MCMC-routines to asses the modelfit. Here we describe the package, providing a brief summary of the theory, but focusing our attention on an example illustrating how the package can be used for model fitting and cross-validation analysis; the example is based on data included in the package.