PREDICTING GRASSLAND COMMUNITY CHANGES WITH AN ARTIFICIAL NEURAL-NETWORK MODEL

Artificial neural networks are parallel processing systems with the ab ility to learn by example and generalize from inferred patterns. In th is application, a neural network model of the feedforward, backpropaga tion type is designed to predict future community composition from kno wledge of present climatic factors and species cover. Training and tes ting data are drawn from a 30-year record of the environmental and veg etative variables of a grassland community. The resulting trained netw ork is capable of forecasting accurately up to 4 years into the future . The results indicate a potential usefulness of neural network techno logy for non-mechanistic modeling in ecological research and managemen t.