Thordis Thorarinsdottir | Norwegian Computing Center | Norway
Spatial statistics is generally viewed as being comprised of three major branches: (1) continuous spatial variation, with the continuous process observed at individual locations or on a grid; (2) discrete spatial variation, including lattice and areal unit data; (3) spatial point patterns. Due to the complexity of the modelling, spatial processes (in particular continuous ones) are often modeled in two stages with the marginal modelling performed separately from the modelling of the spatial error correlation between locations. As the aim of the modelling is commonly to assist decision making, in particular in spatial prediction settings, quantification of the uncertainty in the various modelling components is a critical factor. We will give an overview of various approaches to model spatial variation with illustrative examples, focusing on the interplay between model choice, data dimensionality and data availability.