The use of artificial neural networks to assess fish abundance and spatialoccupancy in the littoral zone of a mesotrophic lake

The present work describes a comparison of the ability of multiple linear r egression (MLR) and artificial neural networks (ANN) to predict fish spatia l occupancy and abundance in a mesotrophic reservoir. Models were run and t ested with 306 observations obtained by the sampling point abundance method using electrofishing. For each of the 306 samples, the relationships betwe en physical parameters and the abundance and spatial occupancy of various f ish species were studied. For the 15 fish species occurring in the lake, si x main fish populations were retained to perform comparisons between ANN an d MLR models. Each of the six MLR and ANN models had eight independent envi ronmental variables (i.e. depth, distance from the bank, slope of the botto m, flooded vegetation cover, percentage of boulders, percentage of pebbles, percentage of gravel and percentage of mud) and one dependent variable (fi sh density for the considered population). To determine the population asse mblage, principal component analysis (PCA) was performed on the partial coe fficients of the MLR and on the relative contribution of each independent v ariable of AWN models (determined using Garson's algorithm). The results st ress that ANN are more suitable for predicting fish abundance at the popula tion scale than MLR. In the same way, a higher level of ecological complexi ty, i.e. community scale, was reliably obtained by ANN whereas MLR presente d serious shortcomings. These results show that ANN are an appropriate tool for predicting population assemblage in ecology. (C) 1999 Elsevier Science B.V. All rights reserved.