Artificial neural networks as a tool for predicting fish community composition in rivers (2000).

Scardi M.°, Lek S.*, Lim P.^, Di Dato P.° & Oberdorff T.°°

° Department of Zoology, University of Bari, Via Orabona 4, 70125 Bari, Italy
* CNRS - UMR 5576, CESAC, Université Paul Sabatier, Bat. IVR3, 118 Route de Narbonne, F-31062 Toulouse Cedex, France
^ ENSAT, Equipe Environnement Aquatique \& Aquaculture, Avenue de l'Agrobiopole, BP 107, 31326 Castanet Tolosan, France
°° Muséum National d'Histoire Naturelle, Laboratoire d'Ichtyologie Generale et Appliquée, 43 Rue Cuvier, F-75231 Paris Cedex 05, France

 Fish community composition has been successfully modeled by means of artificial neural networks (ANNs), using environmental variables as predictors. Two applications of this technique are presented, based on different levels of integration of the information about fish community composition.

Local species richness was modeled in the Garonne river (France), using only 3 environmental predictive variables, whereas presence or absence of 30 fish species was modeled in north-eastern Italian rivers using 26 environmental predictive variables. The results were very good in both cases: the ANN models were able to explain 82\% of the variance in local species richness and to correctly predict presence or absence of fish species in more than 85\% of the cases.

These results pointed out that ANNs can play a very important role in the study of ecological communities.