THE USE OF ARTIFICIAL NEURAL NETWORKS TO PREDICT THE PRESENCE OF SMALL-BODIED FISH IN A RIVER

1. Discriminant factorial analysis (DFA) and artificial neural network s (ANN) were used to develop models of presence/absence for three spec ies of small-bodied fish (minnow, Phoxinus phoxinus, gudgeon, Gobio go bio, and stone leach, Barbatula barbatula). 2. Fish and ten environmen tal variables were sampled using point abundance sampling by electrofi shing in the Ariege River (France) at 464 sampling points. 3. Using DF A, the percentage of correct assignments, expressed as the percentage of individuals correctly classified over the total number of examined individuals, was 62.5% for stone leach, 66.6% for gudgeon and 78% for minnow. With back-propagation of ANN, the recognition performance obta ined after 500 iterations was: 82.1% for stone leach, 87.7% for gudgeo n and 90.1% for minnow. 4. The better predictive performance of the ar tificial neural networks holds promise for other situations with non-l inearly related variables.