Predictive models of collembolan diversity and abundance in a riparian habitat

The artificial neural network (ANN) was used in this work for modelling the abundance and diversity of hydrophilous Collembola on the microhabitat sca le. The procedure was applied to a Collembolan assemblage of the northern P yrenees. Six variables were retained to describe its structure: abundance o f the three dominant species, species richness, overall abundance of Collem bola, and Shannon index. Seven environmental variables were selected as exp lanatory variables: distance to water, soil temperature, water content, and proportion of mineral soil, moss, litter and rotten wood in the substrate. Correlations between observed values and values estimated by ANN models of the six dependent variables were all highly significant. The ANN models we re developed from 83 samples chosen at random and were validated on the 21 remaining samples. The role of each variable was evaluated by inputting fic titious configurations of independent variables and by checking the respons e of the model. The resulting habitat profiles depict the complex influence of each environmental variable on the biological parameters of the assembl age, and the non-linear relationships between dependent and independent var iables. The main results and the ANN potential to predict biodiversity and structural characteristics of species assemblages are discussed. (C) 1999 E lsevier Science B.V. All rights reserved.