REEVALUATING THE UTILITY OF DETRENDED CORRESPONDENCE ANALYSIS AND NON-METRIC MULTIDIMENSIONAL SCALING FOR ECOLOGICAL ORDINATION

Ordination is widely used in ecology and paleoecology for characterizing gradients in community composition. Recent ecological literature documents the controversy over the relative merits of two of the most common methods, detrended correspondence analysis (DCA) and non-metric multidimensional scaling (MDS), with most preferring MDS over DCA. One set of criticisms of DCA stems from theoretical concerns, such as the inelegance of detrending and rescaling and their possible distorting effects, the implied chi-square distance metric and its exaggeration of the distinctiveness of samples with several rare taxa, and the assumption that the rate of turnover along a gradient is constant. A second group stems from two studies that simulated coenoclines and coenoplanes and compared the effectiveness of ordination methods at recovering those gradients. Although in most cases, DCA performed comparably to MDS, those studies emphasized the few cases in which the performance of MDS was superior. Subsequent studies have exaggerated the claims of these two studies and, as a result, harsh criticisms of DCA are now common in the ecological literature. Here, we present new coenoplane simulations that test - and contradict - several of the claims made about the relative merits of DCA and MDS. In particular, MDS is not able to recover simple, one-dimensional gradients as effectively as DCA. Also, DCA is effective in many cases at recovering complex multidimensional gradients and is not restricted to simple, one-dimensional gradients. Finally, MDS does not necessarily perform better than DCA in recovering complex gradient structure, a result borne out in field studies as well as simulations. Given the potential for both techniques to perform poorly under some circumstances, we recommend performing both types of ordinations and evaluating their effectiveness by their interpretability and through comparison with an external data set of environmental information.