VIRTUAL PALEOECOSYSTEM (VIRPE): A NEW COMPUTATIONAL MODEL FOR ANALYZING PALEOECOSYSTEMS AND THE ACCURACY OF MULTIVARIATE ANALYSES

Many paleocommunity studies depend on multivariate techniques to interpret ecosystem patterns and the controlling environmental and biotic variables. Communities are the result of complex mosaics of biotic interactions and environmental parameters. How reliable are various techniques for capturing the parameters that control community diversity? A new analytical program, Virtual Paleoecosystem (VirPe), was developed specifically to investigate such paleoecological issues. The program models paleocommunity membership, abundance, and attributes, permitting comparison of “actual” results with those predicted by multivariate techniques. In this first version of VirPe, we explore the accuracy of common ordination techniques (Bray-Curtis, DCA, and PCA) given a single, controlling environmental variable. In a set of simple communities, all techniques would be expected to ordinate samples on the first axis along this environmental gradient.

We created a pool of taxa, each with an environmental preference and range, body size, and recruitment rate. Multiple communities (samples) were constructed by inputting disturbance rate and available space. The algorithm randomly selects an environment for each sample, and then recruits individuals into the available space based on the recruitment probabilities, which vary with distance from the optimal environment of the given taxon. Recruitment continues until either the available space is fully occupied or the system is disturbed. Several sets of communities were generated to explore the accuracy of ordination.

When variation in taxon attributes was constrained, all techniques capture the single environmental gradient (Spearman Rank Correlation, r = 0.85-1.0, p << 0.01). With increased complexity in the community and increasing variation in taxon attributes, minor errors appear in the results of the analyses. Bray-Curtis ordination has a tendency to move groups of similar samples slightly out of rank order along the gradient, whereas DCA tends to displace individual samples more distantly out of gradient order. PCA exhibits the problems of both Bray-Curtis and DCA techniques. Later iterations of the program will allow the analyses of multiple community generations, time-averaging, and preservation potential of the living community.