Most long-term investigations of vegetation are designed to monitor change in species abundance and composition. This often leads to interpretational problems due to oscillations caused by environmental fluctuations within and between seasons. To improve the interpretability of such investigations, it is proposed to consider in addition to the original data also processes that can be observed with help of derived parameters only, e.g. changes in the resemblance pattern of a set of relevés. Such processes are, for instance, compositional convergence and divergence of data sets, subsets, centroids of subsets or the crispness of a group structure.
Since several of the processes may occur simultaneously, there is a need to carry out the analysis at different scales or levels of aggregation, using several methods to measure species performance as well as different plot sizes and plot arrangements. The latter allows to recognize and measure change in spatial correlation of relevés over time.
The importance of the scale of observation is illustrated with help of two sets of field data. In the first, where no interpretable trends can be detected, convergence of vegetation types emerges in aggregated data only and plot size has no influence. In the second, spatial relationship of relevés indicates the existence of a gradient. This is then destroyed by superimposed management treatments. It is therefore concluded that time data should be analysed at different levels of aggregation and the location of plots considered as an important analytical help in future monitoring projects.