Performance of neural networks to modelling agroecological processes at different spatial scales

The ecological modeller is more than others faced with the general problem of a spatial and temporal heterogeneity of the considered real processes. B ecause of diversity of data and often vague relationships between observed and measured variables artificial neural network models (ANNMs) seem to be a promissing tool for modelling processes in agroecology. The paper discuss es advantages and restrictions of the neural network paradigm on different spatial scales, methods to evaluate the performance and to improve the vali dity of the derived ANNMs. Copyright (C) 1998 IFAC.