Estimating foliar biochemistry from hyperspectral data in mixed forest canopy

Estimating canopy biochemical composition in mixed forests at the level of tree species represents a critical tool for a better understanding and modeling of ecosystem functioning since many species exhibit differences in functional attributes or decomposition rates. We used airborne hyperspectral d...

Full description

Saved in:
Bibliographic Details
Published in:Forest ecology and management 2008-07, Vol.256 (3), p.491-501
Main Authors: Huber, Silvia, Kneubühler, Mathias, Psomas, Achilleas, Itten, Klaus, Zimmermann, Niklaus E.
Format: Article
Language:English
Subjects:
absorption features
Animal and plant ecology
Animal, plant and microbial ecology
aviris data
Band-depth analysis
Biological and medical sciences
Branch-and-bound algorithm
Continuum removal
ecosystem processes
Fundamental and applied biological sciences. Psychology
HyMap
imaging spectrometry data
infrared reflectance spectroscopy
national-park
nitrogen concentration
pasture quality
Synecology
Terrestrial ecosystems
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Estimating canopy biochemical composition in mixed forests at the level of tree species represents a critical tool for a better understanding and modeling of ecosystem functioning since many species exhibit differences in functional attributes or decomposition rates. We used airborne hyperspectral data to estimate the foliar concentration of nitrogen, carbon and water in three mixed forest canopies in Switzerland. With multiple linear regression models, continuum-removed and normalized HyMap spectra were related to foliar biochemistry on an individual tree level. The six spectral wavebands used in the regression models were selected using both an enumerative branch-and-bound (B&B) and a forward search algorithm. The models estimated foliar concentrations with adjusted R 2 values between 0.47 and 0.63, based on the best-sampled study site. Regression models composed of wavebands selected by the B&B algorithm always performed better than those developed with forward search. When extrapolating nitrogen concentrations from one to another study site, regression models solely based on causal wavebands (known from literature) mostly outperformed models based on all wavebands. The study demonstrates the potential of both the use of causal wavebands and of the B&B algorithm.
ISSN:0378-1127
1872-7042
DOI:10.1016/j.foreco.2008.05.011