Characterising spatial variability, which is of utter importance in inspection and maintenance strategies, requires comprehensive spatially distributed databases. However, in real practice, spatially distributed inspection is costly and could damage the structure if a large number of destructive tests are carried out. Therefore, the first objective of this work is to propose a methodology to extract as much informations as possible from available spatially distributed databases, in order to characterise the spatial correlation. Moreover, a preventive maintenance strategy should be supported by deterioration models able to propagate uncertainty and spatial variability. Then, the second objective of the paper is to evaluate the ability of these models to propagate uncertainties and spatial variability. The methodology is illustrated with data collected through destructive tests in a concrete wall exposed to carbonation. The database encompasses information about the concrete porosity, saturation degree , density, and carbonation depth. Recommendations are hence provided in this work for the choice of input parameters that should be modelled as random fields. These recommendations were applied and then confirmed by comparing measured and modelled spatially distributed carbonation depths. The results highlight that uncertainties in measurements and statistical uncertainties have significant impact when dealing with spatial variability.