An Introduction to the Physical Basis for Deriving River Information by Optical Remote Sensing - Fluvial Remote Sensing for Science and Management - page 62

In this, more typical situation, reliable depth estimates
can be obtained using the ratio-based method described
above. Legleiter and Roberts (2005) showed that this
approach is robust to the effects of sub-pixel variations in
depth and bottom reflectance. This study also described
how spectral mixture analysis could be used to disag-
gregate aquatic and terrestrial radiance contributions to
mixed pixels along the banks, which will be more preva-
lent in coarser-resolution data. A key point made by
Legleiter and Roberts (2005) is that the ability to char-
acterise channel morphology depends in a complex and
spatially variable manner on the spatial resolution of the
sensor and the dimensions of the channel features of inter-
est. As a result, a pixel size that is adequate for one reach
might not be sufficient for other, more morphologically
complex portions of the river.
Although the radiance upwelling from a river channel
comprises a continuous spectrum, remote sensing instru-
ments typically integrate the spectral radiance over a few
discrete bands. The number, width, and location of these
bands determine how faithfully spectral variations in the
original radiance signal are reproduced by image data
and thus dictate the extent to which spectral information
can be used to infer channel attributes. High spectral
resolution is potentially quite valuable because several
important optical characteristics of rivers vary apprecia-
bly with wavelength. Most notably, the reflectance of the
streambed and the absorption and scattering processes
that determine the rate of attenuation within the water
column have well-defined spectral shapes. For bathy-
metric mapping applications, the availability of a large
number of narrow spectral bands enables selection of
wavelengths that are highly sensitive to depth, typically
due to strong absorption by pure water, but unaffected
by variations in bottom reflectance and scattering within
the water column.
Legleiter et al. (2009) introduced a simple method,
called optimal band ratio analysis or OBRA, for identify-
ing wavelength pairs that satisfy these requirements. This
technique exploits high spectral resolution data by com-
puting the image-derived quantity X defined by Equation
(3.16) for all possible combinations of bands. The result-
ing X values are then regressed against fieldmeasurements
of depth and the band combination that yields the highest
R 2 is deemed optimal for depth retrieval. OBRA of both
field spectra and simulated data from a radiative transfer
model indicated that integrating the radiance signal over
broader spectral bands did not significantly degrade the
predictive power of X vs. d regression equations, imply-
ing that high spectral resolution data are not necessarily

essential for accurate bathymetric mapping. When field
spectra were re-sampled to match the bands of specific
multispectral sensors, R 2 values were somewhat lower,
however, suggesting that the wavelength position of the
bands might be more important than their widths.
For other applications focused on substrate mapping,
high spectral resolution could be more important, if not
critical. Distinguishing among bottom types requires suf-
ficient spectral information to resolve subtle differences
between streambeds. For example, periphyton exhibits
a diagnostic chlorophyll absorption feature at 675 nm
that can be used to detect algal presence, but only if
the imaging system provides sufficient spectral detail to
resolve this feature. Moreover, higher spectral resolution
could allow the effects of depth and bottom reflectance
to be be disentangled with greater confidence where
heterogeneous substrates might act to complicate depth
retrieval. Hyperspectral data also create opportunities
to apply more sophisticated approaches such as spec-
tral mixture analysis, which could be especially helpful
along the banks. For management applications where
bathymetric mapping is the primary objective, the robust
performance of ratio-based depth retrieval implies that
more readily available, multispectral data will often be
adequate, however.
A third sensor characteristic, radiometric resolution, is
not visually apparent in the same way as spatial resolu-
tion and does not affect the structure of an image data
set in an obvious manner as does spectral resolution.
Radiometric resolution can thus be easily overlooked,
but this fundamental property of an imaging system
should be a primary consideration in any remote sens-
ing investigation. For example, the precision with which
depths can be estimated and the maximum depth that
can be detected are both largely determined by sensor
radiometric resolution, which describes an instrument's
ability to distinguish subtle variations in the amount of
upwelling radiance. The critical concept to bear inmind is
that although the original radiance signal is continuous,
image data are recorded as digital numbers. Convert-
ing the continuous signal to a discrete form necessarily
entails some loss of information. In the context of rivers,
this principle implies that a change in depth, bottom
reflectance, or volume reflectance can only be detected if
the corresponding change in radiance exceeds the fixed
amount of radiance corresponding to one digital number.
Continuing with our focus on depth retrieval, this
reality of remote sensing dictates that truly continuous
bathymetric maps cannot be derived from digital image
data. Instead, each depth estimate is associated with a