Spatiotemporal smoothing and sulphur dioxide trends over Europe

摘要

Spatiotemporal models for sulphur dioxide pollution over Europe are considered within an additive model framework. A suitable description of the spatiotemporal correlation structure of the data is constructed and incorporated in the analysis of the additive model, to ensure that standard errors and other forms of analysis reflect the form of variation that is exhibited by the data. To deal with the large sample size, an updating formula based on binning is derived to provide a computationally manageable implementation of the back-fitting algorithm. Interaction terms involving space, time and seasonal effects are also considered. This requires three-dimensional smoothing which is implemented by repeated application of lower dimensional marginal smoothing operations. The properties of this form of smoothing are examined and the estimators are shown to have first-order behaviour, inherited from the marginal operations, which is equivalent to the full multivariate versions.These models and methods are applied to the sulphur dioxide data, allowing detailed and informative descriptions of the spatiotemporal patterns to be created.