imagery to remove the effects of variable upwelling and downwelling light streams.
This variability causes confusion between similar bottoms types (Ohde and Siegel,
2001) at various depths or in regions with dissimilar concentrations of suspended solids
(Dekker et al., 1996). Thus, the signal is not exclusively associated with a single
environmental variable; rather, it possesses attributes of a combination of parameters.
Several studies have attempted to separate these parameters to improve the ability to
discriminate between coral reef features.
In 1978, a simple image-based approach to compensate for the influence of
variable depth on water leaving radiance was developed (Lyzenga, 1978). The
technique involved removing scattering in the atmosphere and variation in the surface
of the water, connecting depth to radiance using a linear algorithm, calculating the ratio
of attenuation coefficients for different band pairs, and generating a depth-invariant
index of bottom type (Lyzenga, 1978). More recent approaches have focused on
expanding this technique (Philpot, 1989; Maritorena et al., 1994) (Table 8). Typically,
however, they are founded on several assumptions drawn from statistical relationships
between reflectance and selected attributes of the aquatic environment. Using these
techniques requires that researchers make the following assumptions: (1) that water
turbidity is consistent, and low concentrations of suspended solids are present
throughout an image scene; (2) that light attenuates exponentially with depth regardless
of depth or bottom type; (3) that downwelling and upwelling light streams can be
characterized in an identical manner and that there is no contribution by fluorescence or
backscatter; and (4) that the ratio of bottom reflectances in two bands is the same for all
bottom types within the scene (Newman, 2001). The assumption of log-linear
attenuation with depth, in particular, is problematic for bright substrates in cases in
which there may be multiple reflections between the suspended material and the
surface, with the result that the rate of change of attenuation with depth changes
(Newman and LeDrew, 2002)
One way of avoiding such assumptions is to use a radiative transfer algorithm such
as Hydrolight, which was designed for computing radiance distributions for ocean
water bodies. The algorithm is defined at the following website: http://www.sequoiasci
.com/pdf/H42Description.pdf. (The WASI program of Gege and Albert, Chapter 4 of
this book, also can be used for modeling the spectral distribution throughout the water
column.) This time-independent model computes water-leaving radiance as a function
of depth, direction, and wavelength within the water. The model requires several inputs
that are measured at the time of image capture, where possible, and include (1)
absorption and scattering coefficients; (2) water surface conditions; (3) benthic feature
characteristics such as structure and depth, and (4) sky radiance distribution.
3.3.3 Designing the Field Survey
Collecting ground confirmation data (often misleadingly referred to as ‘ground
truth’) remains a significant component of assessing the accuracy of remotely sensed
imagery. In most cases, it is unlikely that depictions of reef characteristics in imagery
alone will be sufficiently accurate for marine park planning and management,
regardless of the image analyst’s skill in the identification of marine habitats. Although
ground confirmation information is often costly and more time consuming than
airborne and satellite imaging, adequate field surveys can eliminate confusion and avert
inappropriate planning strategies based upon poor information.
A field survey of coral reef environments must be used to support image analysis
for three primary reasons: (1) to define habitats of interest; (2) to identify the spatial
264
Newman, LeDrew and Lim
This variability causes confusion between similar bottoms types (Ohde and Siegel,
2001) at various depths or in regions with dissimilar concentrations of suspended solids
(Dekker et al., 1996). Thus, the signal is not exclusively associated with a single
environmental variable; rather, it possesses attributes of a combination of parameters.
Several studies have attempted to separate these parameters to improve the ability to
discriminate between coral reef features.
In 1978, a simple image-based approach to compensate for the influence of
variable depth on water leaving radiance was developed (Lyzenga, 1978). The
technique involved removing scattering in the atmosphere and variation in the surface
of the water, connecting depth to radiance using a linear algorithm, calculating the ratio
of attenuation coefficients for different band pairs, and generating a depth-invariant
index of bottom type (Lyzenga, 1978). More recent approaches have focused on
expanding this technique (Philpot, 1989; Maritorena et al., 1994) (Table 8). Typically,
however, they are founded on several assumptions drawn from statistical relationships
between reflectance and selected attributes of the aquatic environment. Using these
techniques requires that researchers make the following assumptions: (1) that water
turbidity is consistent, and low concentrations of suspended solids are present
throughout an image scene; (2) that light attenuates exponentially with depth regardless
of depth or bottom type; (3) that downwelling and upwelling light streams can be
characterized in an identical manner and that there is no contribution by fluorescence or
backscatter; and (4) that the ratio of bottom reflectances in two bands is the same for all
bottom types within the scene (Newman, 2001). The assumption of log-linear
attenuation with depth, in particular, is problematic for bright substrates in cases in
which there may be multiple reflections between the suspended material and the
surface, with the result that the rate of change of attenuation with depth changes
(Newman and LeDrew, 2002)
One way of avoiding such assumptions is to use a radiative transfer algorithm such
as Hydrolight, which was designed for computing radiance distributions for ocean
water bodies. The algorithm is defined at the following website: http://www.sequoiasci
.com/pdf/H42Description.pdf. (The WASI program of Gege and Albert, Chapter 4 of
this book, also can be used for modeling the spectral distribution throughout the water
column.) This time-independent model computes water-leaving radiance as a function
of depth, direction, and wavelength within the water. The model requires several inputs
that are measured at the time of image capture, where possible, and include (1)
absorption and scattering coefficients; (2) water surface conditions; (3) benthic feature
characteristics such as structure and depth, and (4) sky radiance distribution.
3.3.3 Designing the Field Survey
Collecting ground confirmation data (often misleadingly referred to as ‘ground
truth’) remains a significant component of assessing the accuracy of remotely sensed
imagery. In most cases, it is unlikely that depictions of reef characteristics in imagery
alone will be sufficiently accurate for marine park planning and management,
regardless of the image analyst’s skill in the identification of marine habitats. Although
ground confirmation information is often costly and more time consuming than
airborne and satellite imaging, adequate field surveys can eliminate confusion and avert
inappropriate planning strategies based upon poor information.
A field survey of coral reef environments must be used to support image analysis
for three primary reasons: (1) to define habitats of interest; (2) to identify the spatial
264
Newman, LeDrew and Lim