Anisotropic spatial sampling designs for urban pollution

Abstract

Isotropic processes form an inadequate basis in modelling many spatially distributed data. In particular environmental phenomena often have strong anisotropic spatial variation, especially when the regions monitored are very large. We extend a recently proposed optimal sampling strategy by assuming a spatial anisotropic random field as the basis for the data generator mechanism. The procedure is based on a geographical space transformation indicated by Sampson and Guttorp. We discuss the optimal design and we develop a sequential procedure for selecting a network of monitoring stations in environmental surveys. Some data on sulphur dioxide pollution in Padua (Italy) are analysed to illustrate the method.