HOW TO FIND ORDER IN N-DIMENSIONS: TESTING THE PERFORMANCE OF ORDINATIONS AND DISTANCE MEASURES USING KNOWN ENVIRONMENTAL GRADIENTS

Ordination techniques are an essential tool in community paleoecology used to visualize relationships between community assemblages. Previous comparisons of the plethora of existing ordination methods have used modeled datasets to simulate a community gradient; here, we use three empiric datasets of different extant communities and expand on previous analyses by testing a greater variety of methods. We compare the viability of ordination techniques paired with 15 common distance measures using modern community datasets sampled along known environmental gradients. Two published datasets were selected based on robust sample sizes and an increasing acidification gradient in small lakes in the North American Great Lakes region (diatoms from Charles, 1983, and zooplankton from Shead, 2007). The effects of acidification on lacustrine communities are well established, and ecologists favor ordinations for tracking response and recovery in conservation efforts. Additionally, we collected a dataset of multiple vertical intertidal zones along a regional transect of increasing salinity (decreasing freshwater influence on the west coast of Vancouver Island, Canada). Intertidal zonation is a visually striking and classic example of a community gradient, especially in combination with the effects of a salinity gradient on marine organisms.

Visual comparison of ordinations across community types suggests strong performance of Principle Coordinates Analysis (PCO) and Nonmetric Multi-Dimensional Scaling (NMDS), fair performance of Bray-Curtis Polar ordination, and poor performance of Principle Components Analysis (PCA) and Detrended Correspondence Analysis (DCA). Pearson’s correlation of the first axis ordination scores and either pH or intertidal zone/salinity are significant for most ordination-distance metric combinations, with the strongest correlation for presence-absence distance metrics using PCO and NMDS. Canberra and Manhattan distance measures perform well, especially with PCO and NMDS. Despite longstanding use, DCA axis 1 fails to reflect reliably the known environmental gradient shaping each dataset. In conclusion, NMDS and PCO consistently perform well, while neither PCA nor DCA are appropriate for community assemblage analysis in modern or fossil systems.