The cost of time - implications of hyperspectral data volume and feature selection routines for conservation science

The recent greater availability of airborne hyperspectral imagery in the tropics has allowed for the analysis of increasingly complex analytical questions in ecology such as remote tree species identification. In comparison to species identification the spectral expression of gender in dioecious species has been generally overlooked despite its effects on plant ecophysiological functioning and the prevalence of dioecious species in the tropics. A problem often implied but not frequently addressed in these analyses is the complexity posed by the data volume collected by airborne sensors. We examine the effect of this volume specifically on feature selection routines for classification and the implication of the resultant limitations on the use of airborne hyperspectral imagery at regional operational scales. We conclude based on an examination of analytical time and the cost of high performance computing systems, that an efficient alternative for such large scale academic or NGO research is a cluster of PlayStation^trade 3s.