Implementation of a self-organizing map for investigation of impoundment impact on fish assemblages in a large, lowland river: Long-term study

This fish community survey conducted in the impounded Warta River at a backwater (AB) and a tailwater (CD) sites is a continuation of our former study (1985–2002). Similarly as the former, the present analysis concerns the pre-impoundment period (1985–1987) and the post-impoundment period, but the latter period is eight years longer (1988–2010). Both analyses employed a self-organizing map (SOM, an artificial neural network algorithm) to identify patterns in fish assemblages. The present analysis applied a new, presently available (free on the internet), clustering technique to obtain the higher typology levels. Neurons and fish samples were segregated into two clusters (X and Y), and four subclusters (x1, x2, y1, y2). All AB samples (from both the pre- and postimpoudment periods) and three CD samples from the pre-impoundment period were assigned to cluster X. Cluster Y included only postimpoundment CD samples. Total fish density was higher by at least one order of magnitude in Y than in X. This was accompanied by a lower density of lithophils and higher dominance (Simpson index) in Y. Also, the indicator value index (IndVal) was employed to assess the effectiveness of the present fish sample typology by distinguishing the indicator species. Most species with significant IndVal values for X were obligatory riverine ones, but for Y facultative riverine ones. The study indicated that the dam without a fish ladder has an essential impact on fish assemblage structure, which is more pronounced in the tailwater. However, the present study indicated that the rate of assemblage changes caused by the impoundment, which was demonstrated in the former study, has much abated in the first decade of the 21st century.