Comparison of Neural Networks and Statistical Methods in Classification of Ecological Habitats Using FIA Data

Abstract

Two artificial neural networks (ANN) and three traditional statistical classification methods are used to classify Forest Inventory and Analysis (FIA) plots into six ecological habitats in the U.S. Northeast. Four variables (overstory and understory species composition, hardwood basal area percentage, and current FIA forest type) are identified from a list of available stand variables as the most important discriminating variables for habitat classification. The error matrix and accuracy indices are used to assess the classification accuracy of the models and to test the differences between the five classifiers. The ANN models (MLP and RBF) are superior to the traditional statistical methods such as linear discriminant analysis and minimum-distance classification. The classification accuracy of the ANN models is 90% or higher for overall classification, and exceeds 92% in five of the six habitat categories. The K-Nearest Neighbor (KNN) method classifies the six ecological habitats as accurately as the two neural network models. This study shows that the ANN models and KNN method have a great potential for the classification of ecological habitats using FIA data, due to their flexibility of modeling algorithms and robustness to the problems in FIA data such as non-Gaussian distributions, nonlinear relationships, outliers and noise in the data. For. Sci. 49(4):619–631.

Bibliographic Information