(i.e., first index in each of the 6 categories in Table 3). Further,
HVIs involving multiple HNBs (Table 4) have great promise
and need further research. Typically, in biophysical and
biochemical modeling ~3 to 8 bands help attain high R-square
values before the relationship becomes asymptotic.
It is now reasonable to state, based on meta analysis, that
~12% of the HNBs in EO-1 Hyperion (i.e., for example, ~30
HNBs out of a total of 242 HNBs each of 10 nm wide from 400-
2500 nm) are non-redundant for a given application such as
in study of vegetation or agricultural crops. This would mean
that about 88% of Hyperion bands (e.g., ~212 HNBs out of a
total of 242 HNBs) are redundant. However, it must be noted
that wavebands that are redundant for one application (e.g.,
agriculture), may be very valuable in another application (e.g.,
geology).
It is obvious that there is a need for Whole Spectral Analysis
(WSA) as well as Optimal Spectral Analysis (OSA). WSA is
of great value under certain conditions such as when: (a) the
ability exists to use integrated spectra over a continuum (e.g.,
integrating spectra over 500 to 600 nm), (b) accurate spectral
libraries exist to match class spectra with target spectra from
spectral library, (c) spectral signature over an entire spectral
range such as 400 to 2500 nm wavelengths are preferred,
(d) the Hughes’ phenomenon can be overcome by using very
large training and accuracy assessment sample sizes, and (e)
massive computing power exists to overcome handling very
large data volumes. OSA (Table 2) is preferred in situations
involving factors such as when: (i) large number of HNBs are
redundant (as is often the case for a given application), (ii)
overcoming Hughes’ phenomenon (e.g., when training samples
for classification and accuracy assessment are insufficient in
dealing with very large dimensions of hyperspectral data), (iii)
specific physiologically meaningful HVIs are required, and
(iv) clear efficiency of working with non-redundant bands is
meaningful and facilitates rapid applications of data without
making significant compromise in classification or modeling
accuracies.
It is obvious that there are inadequate hyperspectral libraries
at present and there is a clear need to establish hyperspectral
libraries of vegetation and agricultural crops that take into
consideration a wide array of factors such as, for example,
crop types, genotypes, phenology, background influences, and
consistency of platform from which the data is acquired.
Spaceborne hyperspectral data acquisition is likely to be a
preferred option due to its consistency and global coverage.
Issues of cloud cover will be addressed to significant extent
through constellations acquiring data throughout the growing
period of crops, for example, along with advanced processing
schemes.
The future of remote sensing may involve regular and
routine acquisition of hyperspectral data from which broad-
bands (e.g., Landsat bands) are simulated. In such a case,
broadbands can be used for data continuity studies of existing
systems such as the Landsat or IKONOS or Resourcesat,
whereas hyperspectral data and its derivatives (e.g., specific
HNBs, HVIs, hyperspectral libraries of species types and crop
types) are used for advanced studies of agricultural crop and
vegetation classification, monitoring, modeling, and mapping.
R
eferences
Atzberger, C. 2013. Advances in Remote Sensing of Agricul-
ture: Context Description, Existing Operational Monitoring
Systems and Major Information Needs,
Remote Sensing,
5(2): 949-981.
Bajwa and Kulkarni 2011 Hyperspectral data mining, Pp. 93-
120 in Hyperspectral Remote Sensing of Vegetation, P.S.
Thenkabail, et al., Eds., ed: CRC Press- Taylor and Francis
group, Boca Raton, London, New York, 781p.
Banskota, A., Wynne, R., Thomas, V., Serbin, S., Kayastha,
N., Gastellu-Etchegorry, J. and Townsend, P. 2013. Inves-
tigating the Utility of Wavelet Transforms for Inverting a
3-D Radiative Transfer Model Using Hyperspectral Data to
Retrieve Forest LAI,
Remote Sensing,
5(6): 2639-2659.
Bian, M., Skidmore, A.K., Schlerf, M., Fei, T., Liu, Y. and
Wang, T. 2010. Reflectance spectroscopy of biochemical
components as indicators of tea (Camellia Sinensis) quality,
Photogrammetric Engineering and Remote Sensing
, 76(12):
1385-1392.
Blackburn, G.A. 2007. Hyperspectral remote sensing of plant
pigments, Journal of Experimental Botany, 58(4): 855-867.
Bolton, D.K. and Friedl, M.A. 2013. Forecasting crop yield
using remotely sensed vegetation indices and crop phenol-
ogy metrics,
Agricultural and Forest Meteorology,
173(0):
74-84.
Bruzzone, L., Conese, C., Maselli, F. and Roli, F. 1997. Multi-
source classification of complex rural areas by statistical and
neural network approaches. ,
Photogrammetric Engineering
and remote Sensing,
65(5): 523-533.
Burger, J. and Gowen, A. 2011. Data handling in hyperspec-
tral image analysis,
Chemometrics and Intelligent Laborato-
ry Systems,
108(1): 13-22.
“The future of remote sensing is
likely to involve regular and routine
acquisition of hyperspectral data
from which broadbands (e.g., Landsat
bands) are simulated. This is a
“win win” situation providing data
continuity for the past satellites and
at the same time providing advanced
spectral signatures needed for
increased understanding of global
vegetation and agriculture”.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2014
707
