Succession theories and related methods for time series analysis are presented and illustrated by examples using published data. Starting from a rather general definition of succession which is understood as any directional change of vegetation in time, the phenomena can be distinguished at three different levels of perception: pattern, process and mechanism. Pattern is the phenomenon perceived and reflected in the sample data. Process is a grammar generating this pattern (Dale 1980). Both can be accessed by appropriate statistical analysis. Mechanisms are the causes of the changes. Whereas in some cases they can be inferred from patterns and processes, they cannot be observed or measured directly.
The methods and examples described include the statistical design of long-term investigations, the analysis of community dynamics using Markov chains, pattern analysis by ordination, the analysis of temporal autocorrelation and fuzzy ordination for fitting time series data to a time vector. In most methods, the results can be tested for the existence of trends using Monte Carlo simulation. Experience suggests that directional changes can only be interpreted if the data cover a reasonable number of time steps.