The development and the release of sensors able to provide data with a high spatial resolution (> 4000 points.ha-1 for some of them) in agriculture raises new questions as to how to represent the information. This study proposes adaptation and application of a theoretical framework to high resolution data in agriculture to explain how changes in sampling unit (SU) affects: i) the total variance of the data, ii) the nugget effect (C0) as well as iii) the proportion of variance corresponding to spatially correlated component (C1). The theoretical approach has been validated on simulated data with known characteristics and on real data sets (total soluble solids of grapes just before harvest). Application of this methodology demonstrated that for the type of data under consideration, increasing the SU resulted in an expected decrease in C0 and C1 simultaneously and in an independent decrease in C0. Our study demonstrates that it is not possible to find a SU that maximizes C1 while minimizing C0. It also demonstrates that for a given increase in the SU, the decrease in C0 depends only on the sample support (the size of the SU). The choice of an optimal SU can therefore be based on the resolution of the available information and the C0 value of the raw data.