A bootstrap procedure to select hyperspectral wavebands related to tannin content

Abstract

Detection of hydrocarbons in plants with hyperspectral remote sensing is hampered by overlapping absorption pits, while the ‘optimal’ wavebands for detecting some surface characteristics (e.g. chlorophyll, lignin, tannin) may shift. We combined a phased regression with a bootstrap procedure to find wavebands related to the tannin content of fresh leaves, and compared the bands thus identified with the reflectance spectra of pure quebracho tannin. For both reflectance and derivative spectra, bands were identified that can be linked to known absorption features (1471, 1640, 1710, 2175 nm) or fall within the absorption regions identified from a reflectance curve of pure quebracho tannin. We also confirmed a previously reported link between tannin content and the derivative spectra around 800 nm. The study shows that by combining bootstrapping with phased regression it is possible to determine wavebands of which the reflectance is influenced by tannin.

Acknowledgements

This study was funded as part of a PhD project by the International Institute for Geo‐information Science and Earth Observation (ITC) in the Netherlands. We thank Mike Peel for his invaluable assistance with the logistics in South Africa. We are grateful for the support of the Senalala game farm, located within the Klaserie Game Reserve, which provided housing and access to the game farms within the Klaserie Game Reserve.