What information do we gain from random sampling, and what knowledge do we benefit from systematic sampling in space and time paying attention to the spatial and temporal continuity of observations and their covariance behavior? What questions are relevant in experiments without any treatment? How can we install an experiment without any treatments? What are the analytical tools and opportunities for diagnosing “on-site” processes in the landscape, in a farmer’s field, in a plot, within a soil profile, and even within a section of the rhizosphere? Is variation bad? What is the range of representation of an individual observation in space or time? How can field observations taken at different distances, extents, and over different domains be related to each other? The objective of this contribution is to provide answers to these questions based on applied examples of experimental data collected at various spatial and temporal scales. Data on soil chemical and physical properties, biomass and plant yield are analysed in different ways, i.e., on one hand as randomly taken samples, and on the other hand as measurements that were georeferenced and therefore allow to calculate their range of spatial and temporal representation. The impact of sampling scale and data aggregation of soil water content measurements and their spatial relations to other soil properties is discussed. Temporal stability of remote sensing data reflecting the spatial pattern of biomass development in farmers’ fields is quantified, and analytical opportunities presented how to determine crop status and yield relations for prediction purposes. Rather than laying out treatments randomly in experimental fields, treatments can be arranged in a continuously changing pattern across the landscape. In this case, frequency domain based methods allow decomposing cyclic variation components. Georeferenced sampling and appropriate analytical techniques improve the understanding of on-site biogeochemical and physical processes.