Geography Reference
In-Depth Information
Also shown in Figure 3.8a are reflectance spectra for
various bank materials encountered along the same river
(Legleiter et al., 2009). Even for studies focused specifically
on in-channel attributes such as depth and substrate type,
the banks cannot be ignored. Along the margins of the
channel, the water-leaving radiance signal of interest will
inevitably be mixed with radiance from adjacent terres-
trial features. The severity of this problem depends on the
dimensions of the channel relative to the image pixel size
and also on the spectral contrast between submerged areas
and various bank cover types. This issue was examined in
detail by Legleiter and Roberts (2005), who attempted to
'unmix' near-bank pixels using reflectance spectra of pure
aquatic and terrestrial end-members. The field spectra in
Figure 3.8a indicate that dry sediments are brighter than
their submerged counterparts but still relatively dark,
particularly for coarser cobble, which was essentially
spectrally flat. Finer sand-sized sediment was brighter
and had a more pronounced increase in reflectance with
increasing wavelength; woody debris had a similar spec-
tral shape but somewhat higher reflectance than sand.
The most spectrally distinct bank cover type was riparian
vegetation, with low reflectance in the blue and red due
to chlorophyll absorption and much higher reflectance in
the near-infrared. The much greater reflectance of vege-
tation than water at these wavelengths makes this bank
cover type spectrally distinct from the channel. Legleiter
and Roberts (2005) thus found that pixels along vegetated
banks could be unmixed successfully. Conversely, gravel
bar surfaces were much more difficult to distinguish due
to their spectral similarity to the wetted channel. These
results imply that separating the aquatic signal of interest
from the composite radiance from a mixed pixel will
be more challenging for certain portions of the riverine
landscape than for others. Under some circumstances,
such as shallow, braided channels with numerous, un-
vegetated gravel bars, even identifying the water's edge
could be difficult. For bathymetric mapping applications,
an important consequence of terrestrial contamination
is the common occurrence of negative depth estimates
for shallow water near the banks. Typical, inverse rela-
tionships between depth and radiance result in negative
depth estimates when applied tomixed pixels that include
gravel, sand, or especially riparian vegetation. The devel-
opment of improved methods of characterising channel
margins remains an important challenge in the remote
sensing of rivers.
3.4 Inferring river channel attributes
from remotely sensed data
With this overview of the pertinent radiative transfer
processes and optical properties, we are now prepared
to examine how river information can be derived from
remotely sensed data. We have chosen to focus our dis-
cussion on one of the more mature applications of remote
sensing to fluvial systems, the estimation of water depth
from passive optical image data. This topic has received
considerable research attention due to the fundamental
role of spatially distributeddata onflowdepths ingeomor-
phic investigations, habitat assessments, and monitoring
activities. Moreover, depth is perhaps the most tractable
river attribute to infer from image data, and several other
applications, such as substrate mapping, are facilitated
by, if not dependent upon, bathymetric information.
3.4.1 Spectrally-basedbathymetric
mappingviabandratios
Although numerous depth retrieval methods have been
applied to rivers with varying degrees of success, a semi-
empirical approach based on a simple spectral band ratio
has been shown to have a number of important advan-
tages. The physical basis of this technique has been exam-
ined in detail elsewhere (Legleiter et al., 2004; Legleiter
and Roberts, 2005; Legleiter et al., 2009), and only a
summary of these findings is presented here. Recall our
discussion of Equation (3.14), which separated the total,
at-sensor radiance signal
L
T
(
) into its component parts.
The component of primary interest for depth retrieval is
the bottom-reflected radiance
L
B
(
λ
), which depends not
only on depth
d
but also on the irradiance reflectance of
the streambed
R
B
(
λ
λ
). Some means of distinguishing vari-
ations in depth from variations in the brightness and/or
composition of the substrate is thus necessary. The basic
principle behind band ratio-based bathymetric mapping
is that spectral differences in
R
B
(
) are much smaller
than the order-of-magnitude spectral differences in the
rate at which light is attenuated by the water column
(Figure 3.8b). As a result, the bottom reflectance in two
bands,
λ
λ
2
, is similar for different bottom types,
and the ratio
R
B
(
λ
1
and
λ
2
) is relatively unaffected by
substrate heterogeneity (e.g., Dierssen et al., 2003). In
contrast, for clear-flowing streams the rate of attenua-
tion primarily depends on the absorption coefficient of
λ
1
)
/
R
B
(
Search WWH ::
Custom Search