A Comparative Study on Ordination Methods in Ecological Community Analysis.

Multivariate analysis of species-compositional data has been recently used to clarify the compositional change pattern of ecological communities, to find out major environmental factors influencing ecological communities, and to support environmental monitoring and planning. Among multivariate approaches, ordination was studied in the present study. General ordination methods were compared one another and effectiveness of them was evaluated. Two sets of species-compositional data, which have already been studied to detect compositional change patterns and environmental factors associating them, were analyzed by six popular ordination methods, namely principal component analysis (PCA), reciprocal averaging (RA), detrended correspon dence analysis (DCA), multidimensional scaling (MDS), polar ordination (PO) and canonical correspondence analysis (CCA). As for CCA, which needs both compositional and environmental data, the influence of replacing environmental data by random values was also studied. The conclusion is as follows. When compositional variation is small (the number of species in the analyzed data set is about three times of the average species richness of a sample or less), PCA with appropriate data transformation performs best. DCA and MDS also show good performance in this case. When the environmental gradients are longer and no association among them exists, DCA will produce usable results. If the gradients have some association with each other, the result of DCA may be distorted while that of MDS will not. In this case it is recommended to confirm the DCA result by CA. It should be noticed that sample scores and species scores are separately obtained by MDS. In many cases this charac teristics is quite inconvenient for community analysis. It should also be considered that the result of MDS may vary with the similarity index used. It was indicated that CCA was more effective method than the other methods studied if enough environmental information is included in the calculation. The effectiveness, however, may decrease when environmental data are not prepared sufficiently.