Predicting the abundance of minnow Phoxinus phoxinus (Cyprinidae) in the River Ariège (France) using artificial neural networks

Abstract

The study of abundance of small-bodied species of fish such as minnow is important because these species play an important role in the food-web dynamics of small streams. In this work, we propose the use of an Artificial Neural Network (ANN) to the modelling and prediction of abundance in minnow Phoxinus phoxinus using 10 environmental microhabitat variables: distance from the bank, percentage of boulders, pebbles, gravel, sand, mud, marl, cover respectively, depth and velocity. A total of 372 points were randomly chosen from a total of 465 electrofished point samples to establish a ANN model. A validation holdout of the training of the ANN was undertaken with testing on 93 other sampling points. On the test set, the prediction performance was 92%. Our study showed the advantages of the back-propagation procedure of the neural network in the field of stochastic approaches to ecology of coarse fishes. The limitations of the neural network approaches as well as statistical and ecological perspectives are discussed.