Self-Organizing Maps in Revealing Variation in Non-obligatory Riverine Fish in Long-term Data

Abstract

Sensitiveness to the overall influence of the river channel regulation, impoundment and pollution was studied for 12 non-obligatory riverine (NOR) fish species in the Warta River, Poland, over the period of 1963–1998. Their total abundance has not considerably changed unlike the structure of their aggregation, which was revealed by a self-organizing map (SOM, Kohonen unsupervised artificial neural network) almost perfectly separating 1960s samples from 1990s ones. The greatest changes in proportion were recorded for the dominant roach Rutilus rutilus (from 28.6 to 60.0%), perch Perca fluviatilis (1.9 to 9.5%), pike Esox lucius (2.9 to 4.9%), and the former codominant bleak Alburbus alburnus (22.9 to 1.8%). They resulted from highly significant (p < 0.001), reaching even one order of magnitude, differences in the abundance of these species. Such remarkable alternations in dominant NOR ichthyofauna obviously influenced the whole fish assemblage structure, which was best illustrated by a clear vertical gradient over SOM in Simpson and Shannon indices. Thus, NOR species, both studied together and separately, can supply valuable information on the fish assemblage dynamics and should not be neglected in temporal analyses, often focused only on vulnerable obligatory riverine, migratory and/or endemic species.