Hyperspectral data provides substantially increased
understanding of plant biophysical and biochemical
properties relative to multispectral broadband data.
Accuracies in classifying, modeling, mapping, and monitoring
are substantially higher when specific hyperspectral
narrowbands (HNBs) and hyperspectral vegetation indices
(HVIs) are used as opposed to multispectral broadbands.
Even though this is now a well-established fact, there is
still significant knowledge gap in our understanding of the
importance hyperspectral data in study of agricultural crops
and vegetation (Thenkabail et al., 2011). Indeed, opportunities
exist for making significant knowledge advances in several
areas of hyperspectral study of vegetation and agricultural
crops such as in: 1. Establishing specific HNBs and HVIs
to quantify biophysical and biochemical properties, 2.
Overcoming Hughes’ phenomenon and data redundancy, 3.
Building hyperspectral libraries of crops and vegetation, and
4. Developing advanced automated methods of hyperspectral
data analysis. Also, increasing amounts of hyperspectral
data (e.g., entire archive of ~64,000 images of EO-1 Hyperion
available from USGS Earthexplorer (
.
usgs.gov/);see highlight article in this issue for details) are
becoming available, globally, for researchers around the world
to conduct specific studies on specific issues in different
croplands and vegetation of the world. Given this fact, the
need for focused research to better understand, model, and
map specific vegetation and agricultural crop characteristics
utilizing hyperspectral data is of great importance. In this
context, PE&RS initiated this special issue on the topic.
The special issue covers some of these advances through
seven distinct peer-reviewed articles. Highlights of these
seven peer-reviewed articles along with their key knowledge
advancement are summarized below.
Aasen et al. developed automated algorithms to process large
volumes of hyperspectral data in order to determine which
hyperspectral narrowbands (HNBs) and\or hyperspectral
vegetation indices (HVIs) hold the best information. They
developed an impressive algorithm called HyperCor to
process large volumes of hyperspectral data and discern their
information pathways. HyperCor takes hundreds or thousands
of HNBs, computes their two-band or multi-band HVIs,
correlates with biophysical and biochemical quantities of
vegetation or agricultural crops, and shows us the windows or
regions in electromagnetic spectrum providing high and low
information content. This is exactly the type of tool that we
Special Issue Foreword
Research Advances in
Hyperspectral Remote Sensing
Prasad S. Thenkabail
need to make best and most efficient use of hyperspectral data
in applications such as the vegetation, and the agricultural
crops. Their study on rice crop was conducted with 5 years
of solid data (3 years for model development and 2 years for
validation). In selecting best HNBs, it must be noted that the
HNBs which are most prominent for one biophysical quantity
may not be the most prominent for another biophysical
quantity. At times, it may not even be most prominent for the
same biophysical quantity in another date. This phenomenon
happens as a result of having narrowbands adjacent to one
another providing near similar information (e.g., 680 nm
and 690 nm are likely to have equally good correlation
with biomass). This will require us to select the most
prominent narrowband in each spectral range (e.g., a band
centered at 680 nm, with 10 nm bandwidth, could be most
frequently occurring narrowband in modeling biophysical
and biochemical properties of vegetation and agricultural
crop in 600 to 700 nm band range). Overall, they clearly
established that automated algorithms like HyperCor are
extremely valuable in analyzing biophysical and biochemical
variables of massive volumes of hyperspectral data of
agricultural crops and vegetation. They also introduced a
novel concept of multi-correlation matrices strategy (MCMS)
to select and use HNBs in HVIs. The idea here is to source
the importance of HNBs based on their significant occurrence
in different correlation matrices (CMs). They showed that
MCMS provided significantly improved accuracies in
studying rice biomass. MCMS is an interesting concept, but
requires further development. Aasen et al. developed their
models using various approaches: pooled data of various
growth stages, individual growth stages, linear models, and
non-linear models. It must be noted that robust models of
biophysical and biochemical properties of specific crops
need to be developed, ideally, taking data of across sites and
across growing stages. Such models are also often nonlinear
in nature as a result of saturation in reflectivity in full canopy
cover scenario. Also, such models need to be developed for
individual crops rather than grouping multiple crop types in
models to achieve best results that are specific and targeted.
It is well known that hyperspectral data is not panacea
for addressing complex issues of cropland and vegetation
classification, modeling, and characterization. This is because,
even though hyperspectral data provides a quantum leap in
information, discerning that information is not an easy task
given the complexities of processing massively large data
volumes, establishing redundant bands, and implementing
methods and techniques that accurately and rapidly establish
information from data. In this regard the work of Parshakov
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2014
721
