APPLICATION OF NEURAL NETWORKS TO MODELING NONLINEAR RELATIONSHIPS INECOLOGY

Two predictive modelling principles are discussed: multiple regression (MR) and neural networks (NN). The MR principle of linear modelling o ften gives low performance when relationships between Variables are no nlinear; this is often the case in ecology; some variables must theref ore be transformed. Despite these manipulations, the results often rem ain disappointing: poor prediction, dependence of residuals on the var iable to predict. On the other hand NN are nonlinear type models. They do not necessitate transformation of variables and can give better re sults. The application of these two techniques to a set of ecological data (study of the relationship between density of brown trout spawnin g sites (redds) and habitat characteristics), shows that NN are clearl y more performant than MR (R(2) = 0.96 vs R(2) = 0.47 Or R(2) = 0.72 i n raw variables or nonlinear transformed variables). With the calculat ion power now currently available, NN are easy to implement and can th us be recommended for modelling of a number ecological processes.