Strategies in computer intensive analysis of vascular plant floristics with application to three US Atlantic coastal sites

摘要

We implement methods for generating null plant communities in order to have empirical probability distributions (e.g., of Jaccard floristic similarity) against which observed statistics may be compared for tests of significance. We present data for three United States Atlantic coastal sites having a conbined vascular flora of 1298 species. The flora of these sites, located from about 400 to 250 N Latitude, was geographically disjunct. Only 2% of the 1298 species was found at all three sites, 19% was found at two sites, and 79% of species was found only at one of our coastal sites, despite their physiographic similarity. This pattern was similar for the 350 alien species (27% of total species) and the 35 species of non-flowering plants. Jaccard coefficients may be used to measure floristic similarity between pairs of study sites but proper interpretation of these coefficients should be made relative to a null community. Although "randon labeling" (our method 1) is most commonly used for testing the significance of statistics observed from floristic inventories, it delivers null means and P-values widely contrasting with those generated by equally appealing methods (our methods 2-4). We found little agreement among the four methods for generating null (random) commnities. Whether an observed Jaccard similarity coefficient is smaller than, equalto, or greater than what we would expect for a randomized plant community, depends on the model used to generate the null community. Our study illustrates this expectation. There remains a need for large-scale systematic study of computer intensive methods for testing the significance of statistics comparing highly diverse plant communities.