This thesis investigates the factors determining coral reef benthic and fish communities in a subsistence only fishing environment in the Lau island group, Fiji, using a number of different modelling techniques and evaluates their performance in analysis of large multi-species communities. Applied modelling techniques included Kohonen self-organizing maps (SOM) (software developed by the author), principal component analysis (PCA), redundancy analysis (RDA), partial causal modelling and general linear models (GLM). A central question was in how far Kohonen maps (SOM) could be applied as an alternative of a PCA in analysis of large community data.

Fish and benthic data were analysed separately and in relation to each other, environmental factors included the fishing pressure (FPI), longitude and latitude, structural environment and the abundance of crown-of-thorns starfish. Variations in the benthic environment were largely determined by a contemporaneous crown-of-thorns starfish outbreak; however the results of the SOM did not support the hypothesis that a higher risk of a crown-of-thorns starfish outbreak is associated with a higher FPI. Fish community structure was neither determined by FPI nor spatial variation, but by the benthic and structural environment. However, the degree to which fish community structure was explained varied with fish-family. While variation of damselfish community structure could be explained to 26.9% by the available environmental variables, variations in wrasse and butterflyfish community structure were only explained to a degree of 5.9% and 4.1% respectively, indicating the need of different sampling strategies or differing biological factors.

SOM proved to be a useful tool in analysing large community data because of its facilities to handle non-linear data.