Geography Reference
In-Depth Information
frames. The advantage of pushbroom arrays is that they
are simpler and cheaper, and the rigid slot means that
there are no complicated movements during each frame
taken (as there is with the whiskbroom scanner). It also
increases the exposure time and, therefore, helps coun-
teract the issues caused by the low signal to noise ratio of
hyperspectral data. The disadvantage is that each column
of pixels in the instrument array is now a separate sensor,
and through time these columns might yield slightly dif-
ferent brightness values even when flown over uniform
materials. This can produce 'bad lines' (lines that are
consistently brighter or darker than neighbouring lines)
that can complicate image processing.
The coastal zone and benthic mapping community
have had a long period of development of optical analy-
sis of water and bottom properties, and this has yielded
several approaches to mapping such environments with
specifically designed and processed sensors (Ackleson,
2003). Both empirical and theoretical research in these
environments showed that broad spectral bands provide
limited information for mapping several distinct types
of materials in the water column simultaneously, and
hyperspectral-type imagers are thus preferred in many
situations (Kutser et al., 2003). The success of hyperspec-
tral approaches in ocean environments suggested that
river applications might also be successful, if airborne
imaging spectrometers could be flown at low altitude,
yielding high spatial resolution imagery needed for the
spatial heterogeneity of most rivers.
Mertes (2002) andMarcus and Fonstad (2008) provide
an overview of the history, physical nature, and uses of
the optical remote sensing of rivers. Most spectral anal-
yses used in river remote sensing have been applied to
multispectral imagery with a low number of broad bands.
While these have been successful in projects such as map-
ping suspended sediment concentrations (Mertes, 1993),
the low number of bands limits the types of algorithm
than can be applied and, more importantly, limits the
number of properties of the river that can be extracted
from imagery. This issue is known as underspecification,
a limitation hyperspectral data excel at overcoming.
Earlier multispectral scanner imagers had been used to
characterise river water depths (Lyzenga, 1978; Lyon and
Hutchinson, 1995) and a combination of water depths
and bottom sediment type (Lyon, Lunetta, and Williams,
1992). In an attempt to broaden the types of river envi-
ronments that could be characterised through remote
sensing, Marcus (2002) was one of the first to directly
apply hyperspectral imaging to stream environments,
focusing on the mapping of in-stream aquatic habitats in
Yellowstone National Park. Other research using different
instruments and techniques soon followed, and these will
be discussed in later sections.
One set of interesting results arose early. Legleiter
et al. (2002) compared the ability of multispectral and
hyperspectral imagery and various spatial resolutions
to correctly classify in situ stream habitats, and found
that high-resolution, hyperspectral imagery significantly
out-performed other imagery. They also found that
many of the errors that were observed in the comparison
of the classified hyperspectral imagery were in fact not
errors, but situations where the field mapping (the
so called 'ground truth') was more in error than the
classified imagery.
4.3 Advantages of hyperspectral imagery
Compared with traditional multispectral imagery com-
prising a small number of bands, hyperspectral imagery
has many profound advantages. The most general advan-
tage is that there are typically far more spectral bands
available (Figure 4.2b, 4.2c) than different classes of fea-
tures being mapped (for example, water habitats, or
types of vegetation species). This situation stands in stark
contrast to most remote sensing situations with mul-
tispectral imagery, which are underspecified. While this
issue can sometimes require advanced algorithms (such as
spectral unmixing, fuzzy classification, physically-based
atmospheric normalisation, regression trees, and artificial
neural networks) to produce information-rich maps, it
allows users to precisely tune techniques to identify very
precise types of objects, such as individual species-level
identification of riparian vegetation.
As one example, having hyperspectral data allows a user
to choose exactly which pair of channels will best map
the relative water depths in clearwater streams (Legleiter
et al., 2009); multispectral band ratios lack this preci-
sion and generally have less accurate water depth map
results. Additionally, the large number of channels means
that, once the imagery has been radiometrically cali-
brated, it is possible to use optical-physical relationships
to extract continuously-varying (rather than classified)
water parameters without the need for in situ ground
data. Such a process has been used for several years in the
shallow ocean water community, but it is just starting to
be used by river researchers.
Several studies have shown that hyperspectral imagery
provides superior river habitat mapping accuracy than
habitat maps derived from multispectral imagery (for
Search WWH ::
Custom Search