We often model geostatistical (i.e. point-referenced data) in order to determine whether or not there are spatial patterns of autocorrelation. The object of interest is frequently an underlying spatial function giving rise to patterns of spatially correlated data. When we work with discrete observational data, a problem arises - we want to study smoothly-varying response surfaces over space, but the data themselves are not continuous and therefore we cannot specify a likelihood which is continuous in both space and response. Consequently, we often choose to reparameterize our model in terms of a latent smooth spatial surface and a link function mapping this spatial surface to the parameters of a likelihood function appropriate for discrete data.