Neural networks to predict fish biodiversity in watercourse

Abstract

Fish communities are closely dependent on the characteristics of the aquatic environnement in which they evolve. We have investigated the use of neural networks for modelling those complex non-linear relationships. This study is based on a large database of electric fishing carried out in the whole Seine basin. The basic problems raised by the use of neural network are twofold : on the one hand the network architecture is to be chosen by a trial-and-error procedure, and on the other hand the training and test sets are to be wisely chosen. The first experiments have been driven using eight empirical input parameters on 18 different species. The results show a reasonnably good accuracy in the prediction of presence or absence of fish. Further experiments are still requiered to confirm this results, but the latter already show interesting possibilities for environmental management.