COS 34-4
Tree species mapping in a diverse tropical forest with airborne imaging spectroscopy

Tuesday, August 12, 2014: 9:00 AM
314, Sacramento Convention Center
Claire A. Baldeck, Department of Global Ecology, Carnegie Institution for Science, Stanford, CA
Gregory P. Asner, Department of Global Ecology, Carnegie Institution for Science, Stanford, CA
Roberta E. Martin, Department of Global Ecology, Carnegie Institution for Science, Stanford, CA
Christopher B. Anderson, Department of Global Ecology, Carnegie Institution for Science, Stanford, CA
David E. Knapp, Department of Global Ecology, Carnegie Institution for Science, Stanford, CA
James R. Kellner, Department of Ecology and Evolutionary Biology, Brown University, Providence, RI
S. Joseph Wright, Smithsonian Tropical Research Institute, Panama
Background/Question/Methods

Mapping tree species with remote sensing is of great interest to ecologists, but the ability of remote sensing to distinguish tree species in highly diverse, closed-canopy tropical forests remains poorly understood.  Operational remote species mapping has not previously been accomplished in a diverse tropical forest, partly due to limitations imposed by traditional mapping methods.  We used imaging spectroscopy data collected by the Carnegie Airborne Observatory along with recently developed and new classification methods to identify individuals of three canopy tree species amongst a diverse background of hundreds of tree and liana species on Barro Colorado Island, Panama. First, we compared the performance of two leading methods – binary support vector machine (SVM) and biased SVM – in identifying pixels belonging to the focal species.  We used the preferred method to construct a multi-species classification to be applied to the image, and performed morphological image manipulation and segmentation to identify individual crown objects from the species predictions.  We then returned to the island to perform field validation of the predicted crowns objects.

Results/Conclusions

Both binary and biased SVM models were able to identify pixels of focal species with high pixel-level recall accuracy of 92–97% and 84–93%, respectively.  Biased SVM was chosen as the basis of the multi-species classification model because of its higher precision. Field validation of the crown predictions from the multi-species model indicated that our method had high crown-level precision, with a false positive error rate of only 0–3%. These results demonstrate the ability of high spatial and spectral resolution remote sensing to accurately map crowns of focal species within a diverse tropical forest.  This was made possible through the use of mapping techniques adapted to species detection in diverse closed-canopy forests, which will pave the way for species mapping in a wider variety of ecosystems.