Geography Reference
In-Depth Information
Table 4.1
Comparison of a small set of example hyperspectral sensors.
Acronymn
Platform
# of Channels
Spectral Range (nm)
Typical Spatial Res
ASIA Eagle
Airborne
244
400-970
Variable by height
AVIRIS
Airborne
224
400-2500
Variable by height
CASI
Airborne
288
400-1000
Variable by height
HYDICE
Airborne
210
400-2500
Variable by height
HYMAP
Airborne
126
450-2500
Variable by height
HYPERION
Spaceborne
220
400-2500
30m
IMSS
Airborne
400
+
400-800
Variable by height
PROBE-1
Airborne
128
400-2500
Variable by height
dissolved organic carbon in the nearshore water
environment. The same algorithm as Hugeunin's was
used by Karaska et al. (2004) to characterise the Neuse
River, North Carolina, using the AVIRIS hyperspectral
scanner. The authors used the technique tomap variations
in suspended chlorophyll, suspended minerals, coloured
dissolved organic carbon, and turbidity. Reference spectra
can come from laboratory measurements of water envi-
ronmental variables, or through careful field spectroscopy
studies, such as those done by Gilvear et al. (2007) to
identify river water depth and substrate type. Spectral
unmixing does not have to be a purely statistical-empirical
approach, it can be combined with physically-based
approaches to take optical physics into account.
Physically-based approaches recognise that a large
quantity of the light reaching an instrument has been
modified in several ways and by several environmen-
tal components other than the target of interest. By
understanding the nature of these components and their
interactions, it becomes possible to separate out the vari-
ous effects, quantify them, and perhaps map them. More
importantly, taking a physically-based approach allows
researchers to use physical law to predict why light does
various things in different water environments, and there-
fore we can predict ahead of time what the results of
certain river remote sensing projects might be. The key
advantage of using a physically-based approach as that
because the derived models have a physical basis they are
more robust and less site-specific than those derived using
empirical methods. This does not, however, guarantee
that they will work better for a specific mapping purpose.
While many of the optical laws and reference spectra
for various river parameters are known, the use of these to
back-calculate conditions for a given image is somewhat
complex; this problem can be solved by so-called inver-
sion methods. For example, Lee et al. (1999) provide an
early shallow coastal reflectance model/inversion model
to solve for water depth and various water properties
using hyperspectral data, with no field data other than
measured reflectance required. Another inversion algo-
rithm for in-water absorption and scattering coefficients
is suggested by Gould et al. (2001).
Mobley's Hydrolight software (Mobley, 1999; Mobley
and Sundman, 2001), a radiative transfer model based on
earlier theoretical and applied research on optics (Mobley,
1994) as well as later research (Mobley, 1999; Mobley and
Sundman, 2003; Mobley et al., 2003) provided a strong
optical physical basis for understanding the nature of light
in water, and has been used extensively in the ocean com-
munity to understand water parameters under various
lighting conditions, including depth in the shallow water
environment. Leglieter et al. (2004) and Legleiter and
Roberts (2005) used this general approach to predict the
effects of channel morphology and sensor spatial resolu-
tion on imager-derived depth estimates and other habitat
variables. Among other findings, they found that a care-
ful chosen ratio of red and green wavelengths is a stable
correlate to bottom depth. Further work by Legleiter et al.
(2009) found that depth estimates could be predictably
improved by a technique called optimal band ratio anal-
ysis (OBRA). If hyperspectral image data are available,
OBRA produces the best set of band ratios necessary to
extract water depth from imagery. Optimal band ratios
only provide relative depth maps, and other methods
must be added to convert these relative maps to absolute
depth maps (Fonstad and Marcus, 2005). Legleiter and
Roberts (2009) continued this line of research, producing
a 'forward image model' (FIM) that predicts how well a
given sensor will work for mapping a given set of stream
habitats. Such forward models have a long history in
the remote sensing of ocean habitats (for example, Mari-
torena et al., 1994), and there is definitely roomfor further
Search WWH ::
Custom Search