Ordination of simulated complex forest succession: A new test of ordination methods

Abstract

A model of a 1/12th ha forest stand, FORET, generated 10 000 years of simulated species succession. Approximately the first third of these results were analyzed by principal component analysis as if they were collected field data to give the trajectory of the community particle in a collapsed species space. The ordination axis orientation was performed on a dispersion matrix and correlation matrix between species. In both cases, however, the eigen vectors were applied to the data matrix which had not been transformed to unit species variance. This facilitated comparison of species dispersion and correlation structure; it emerged they were very different. Correlation structure gave large weights to understory species while dispersion emphasized the dominant overstory species. This implies a decomposition of simulated stand behavior into overstory and understory, even though such decomposition was not formally built into the model. This decomposition would seem to pertain to real vegetation.

Principal component analysis was able to express insightful differences between data structure with and without the unit variance transformation implicit in the correlation matrix. This flexibility of the ordination method proved valuable in uncovering unsuspected ordering principles in the model. Complex simulated data allow the ordination technique to demonstrate its capacity to generate new hypotheses, which hypotheses can then be simply validated by a return to the structure of the model but with the hindsight of the analysis. The generation of new hypotheses is not possible if the simulation is of a simple coenocline; on the other hand, ordination of test field data does not allow the simple validation of new hypotheses, for in the field there is not a defined algorithm to which the researcher can return.