Application of multivariate and other statistical methods to the analysis of marine communities from the Maltese Islands

Abstract

Marine communities from the Maltese Islands were modelled, and it was found that species abundances are distributed as normal curves along marked environmental gradients. Parameters defining range of occurrence, abundance at mode, skewness and kurtosis were extracted and modelled. The results were then inputted into a simulation programme to create simulated communities similar in structure to those occurring locally. From these simulated communities, samples were taken using a known sampling pattern, and after truncation and transformation the data set was analysed by Cluster Analysis and Non-Metric Multidimensional Scaling, two multivariate statistical techniques, that have in recent years been found very robust in the analysis of marine communities. It was found that removal of species which have abundances of less than 3% (of the total abundance of all species within a sample) in all samples, and so that a data set with not less than 30 species is obtained helps analysis as it removes rare species that are a considerable source of noise. Transformation of abundances to 4t1t root was found to be necessary to decrease the differences in the effects of rare and dominant species on similarity measures. The Kulezynski similarity measure was found to be very robust, followed closely by the Bray-Curtis similarity measure. The Relative Manhattan and Chord metrics were less robust. These results were used to build an 'optimum protocol' for the analysis of local marine ecology data. Molluscs from a local benthic study were identified to species level, and the data from the whole study were analysed. A polluted site at Ta' Xbiex was found to be significantly different in community structure from others in less polluted areas. The analysis revealed differences between communities at different bottom distances. Data sets from three other previous local studies were re-analysed with these methods, and the differences between the results obtained in the re-analysis and those obtained in the original analysis are discussed. Species in the above-mentioned data sets were aggregated at different taxonomic levels, and the effect of this procedure on the amount of information that could be extracted by Cluster Analysis, Non-metric Multidimensional Scaling and ANOSIM tests was assessed. Aggregation at the genus level is safe, and at times beneficial, however caution must be exerted in aggregation to family and order level. Aggregation to order level was not found to be acceptable. An original computer programme, OECOSTAT, was developed for manipulation and analysis of community ecology data sets. This programme, used extensively in this research, is described.