Application of Spectral and Environmental Variables to Map the Kissimmee Prairie Ecosystem Using Classification Trees

Abstract

This paper compares a variety of classification tree-based approaches to map 10 vegetation cover classes and a single built-up class in the Kissimmee Prairie Ecosystem, an endangered grass-shrubland landscape in south-central Florida (USA). This comparison is provided to identify an effective and replicable mapping methodology and facilitate the ongoing regional-scale management and monitoring of grass-shrubland ecosystems. Results showed that the best-performing models included environmental variables, due to the ability of these variables to help distinguish spectrally similar classes. The highest overall proportional accuracy of 81% was the result of incorporating linear spectral mixture analysis and geo-environmental variables into the classification tree.