Patterning spatial variations in fish assemblage structures and diversity in the Pilica River system

The study of fish assemblages in the Pilica River has not yet been undertaken despite data being available for various reaches. At the beginning we tried to analyse the data by detrended correspondence analysis (DCA), but obtained ordinations were not clear. On the two-dimensional scatterplot, sites from the main channel and sites from lower courses of the largest tributaries formed a long bowed "black cloud" on which single sites and their symbols were not visible. Besides, there are well-known limitations for indirect gradient analysis, such as strong distortions with non-linear species abundance relations (horseshoe effect, arch effect, missing data, noise, redundancy, outliers, disjointed data matrix, etc.) (Gauch 1982; Kenkel and Orloci 1986; ter Braak 1987; Jongman et al. 1995; Guegan et al. 1998; Palmer 2000). These limitations can be avoided by applying the artificial neural network (ANN) which has already been successfully used in ecology (Chon et al. 1996; Guegan et al. 1998, 2001; Lek and Guegan 1999, 2000; Brosse et al. 2001; Park et al. 2001a). In this study we used the self-organizing map (SOM) with an unsupervised learning algorithm, which was used in a few studies to reveal relationships within ecological communities (Chon et al. 1996, 2000a; Giraudel et al. 2000; Giraudel and Lek 2001; Brosse et al. 2001; Park et al. 2001a). The conclusions drawn from these works indicate that the SOM algorithm is fully usable in ecology as a technique for analyzing data and for community ordination (Chon et al. 1996). The fish fauna of the Pilica River was selected for the study because it is one of the best known in Poland (Penczak 1988, 1989; Penczak et al. 1995, 1996). Also some comparative research on its stability and variability has already been done (Penczak and Kruk 1999, 2000) and this fact creates a chance for determining which method used for data ordering gives a more clarified and closer-to-reality picture of the assemblage structure in the river system. In this fish fauna inventory study (1992-95) fish sampled at each site were not only counted but also weighed. It is well known that the potential energy of an ecosystem is not distributed proportionally between species (Odum 1980). Out of tens or hundreds of species forming a community, relatively few, named dominants, exert a major controlling influence, because they are ecologically successful in a given environment. Expressing the importance of populations in an ecosystem is difficult in energy units, but it is much easier in biomass. Density can be effectively used only for comparing populations of species of similar body size (Acarina, Chironomids, etc.). The aim of the study is to show how two different methods, SOM and DCA, can be advantageous to analyse complex, non-linear fish population datasets. Here we try to test if clusters of neurons distinguished by the SOM on the basis of fish biomass only, also differ in terms of species diversity, community dominance and assemblage stability.