A mixture model-based approach to the classification of ecological habitats using Forest Inventory and Analysis data

A Gaussian mixture model (GMM) is used to classify Forest Inventory and Analysis (FIA) plots into six ecological habitats in the northeastern USA. The GMM approach captures intra-class variation by modeling each habitat class as a mixture of subclasses of Gaussian distributions. The classification is achieved based on the appropriate posterior probability. The GMM classifier outperforms a traditional statistical method (i.e., linear discriminant analysis or LDA), and produces similar overall accuracy rates to a commonly used neural network model (i.e., multi-layer perceptrons or MLP). For the classifications of individual ecological habitats, however, MLP produces better (or same) producersʹ classification accuracies for five of the six ecological habitats than does GMM. But the GMMʹs accuracy rates are more consistent (92%–97%) across the six ecological habitats than those of the MLP model (82%–99%). This study shows that GMM offers an attractive alternative for modeling the complex stand structure and relationships between variables in mixed-species forest stands.