The Application of Neural Networks to the Floristic Classification of Remote Sensing and Gis Data in Complex Terrain

This study applies a Back-Propagation Neural Network to the task of floristic land cover classification. The input data consists of the three LANDSAT TM bands 2, 4 and 7 and the GIS based environmental variables Aspect, Elevation, Catchment, Geology and Slope. This dataset covers a 225 square-km sub scene centred near the town of Kioloa, South East, Australia. The study area is complex compnsmg a mixture of disturbed and partially cleared sclerophyll forest and rainforest on rough terrain with variable geology. Rainforest gullies and coastal heath are found along the coastal fringe thus adding to the floristic complexity confronted by the neural network. The resulting neural network classifications provide a realistic estimate of the distribution of floristic classes. The patterning is more sophisticated and less polygonal than that achieved using earlier models. The most awkward vegetation class, rainforest ecotone is handled effectively by the neural network. Misclassified pixels are allocated to either wet sclerophyll or rainforest.