Prediction of habitat quality using ordination and neural networks

The development of an automatic classification model for prediction of conservation value is described. The classifier combines ordination and neural network (NN). The classifier was trained to predict the probability of a sample being of potential conservation interest. The neural network was trained on a priori classified data and used sample scores derived from ordination for prediction. The complexity of the NN classifier and the selection of the optimal ordination method were guided by cross-validation of a series of candidate models. The conservation value of a test data set was predicted by the NN classifier, and this classification was evaluated in terms of species richness, nativeness. rarity, and beta diversity. Finally, we evaluated the capability of the approach to handle new samples, not included in the ordination. These samples were derived from habitats of threatened vascular plants. and they were all successfully predicted to be valuable. It is shown that the combination of ordination and neural networks successfully reproduces the a priori classification. It is further demonstrated on a test data set which the classifier discriminates with respect to traditional Measures of conservation interest such as rarity, nativeness. and diversity. The developed method may he seen its it promising approach to assessment of biological integrity at the scale of plant communities. and further opportunities for its application are suggested.