At some stage, investigations of regional variability in climatological information require the analysis of spatial structure and interpolation between data points. In this paper we study both spatial analysis and interpolation using the example of temperature observations from a fixed surface meteorological network. Initial analyses indicate that the mean field can be modeled with a deterministic term plus a nonstationary stochastic component. The deterministic component models the influence of site location in terms of geographic coordinates, elevation, and local environment. The stochastic component presents anisotropy and nonstationarity through second-order moments and accounts for high variability between subareas in the studied region. The heterogeneous spatial structure of the region is modeled with a technique that transforms the site geographic map into a new site representation with an isotropic and stationary covariance structure. The transformed representation provides valuable information on the influence of microclimatic conditions in the spatial covariance structure. The model captures the main spatial structure characteristics over the region and allows optimal prediction of the spatial mean field at unobserved locations.