Neural networks and multiple linear regression models of the abundance
of brown trout (Salmo trutta L.) on the mesohabitat scale were develo
ped from combinations of physical habitat variables in 220 channel mor
phodynamic units (pools, riffles, runs, etc.) of 11 different streams
in the central Pyrenean mountains. For all the 220 morphodynamic units
, the determination coefficients obtained between the estimated and ob
served values of density or biomass were significantly higher for the
neural network (r(2) adjusted = 0.83 and r(2) adjusted = 0.92 (p < 0.0
1) for biomass and density respectively with the neural network, again
st r(2) adjusted = 0.69 (p < 0.01) and r(2) adjusted = 0.54 (p < 0.01)
with multiple linear regression). Validation of the multivariate mode
ls and learning of the neural network developed from 165 randomly chos
en channel morphodynamic units, was tested on the 55 other channel mor
phodynamic units. This showed that the biomass and density estimated b
y both methods were significantly related to the observed biomass and
density. Determination coefficients were significantly higher for the
neural network (r(2) adjusted = 0.72 (p < 0.01) and 0.81 (p < 0.01) fo
r biomass and density respectively) than for the multiple regression m
odel (r(2) adjusted = 0.59 and r(2) adjusted= 0.37 for biomass and den
sity respectively). The present study shows the advantages of the back
propagation procedure with neural networks over multiple linear regres
sion analysis, at least in the field of stochastic salmonid ecology.