has the required wavelength bands to elucidate this feature. However, this method
is currently demonstrated as anecdotally successful only in a single test case.
For airborne campaigns it is sensible to design image acquisition strategy to
minimize glint. Flight paths towards or away from the sun with solar zenith angles
of 30°–60° have been recommended (Mustard et al. 2001; Dekker et al. 2003). In
one specific example, Lesser and Mobley (2007) acquired imagery with solar
zenith 40°–55° and flight direction aligned with the sun.
4.2.6 Depth Correction
Water depth has a major effect on the spectral reflectance measured over a reef.
The range of the absorption coefficient of water over the 400–700 nm wavelengths
is high and the spectral shape of absorption is also affected by water constituents
such as CDOM (Fig. 4.2). Hence the same bottom reflectance at different depths
will give rise to very different above-water reflectances, and this is clearly a
complication in identifying the bottom type. Two simple methods can be used to
remove the effect of variable depth: (1) ‘depth correction’ requires that depth be
known across the image either from acoustic (Bejarano et al. 2010; Chaps. 8–10)
or LiDAR data (Chaps. 5–7), or for some reef topographies a simple stratification
into depth zones may be possible (Mumby et al. 2004); or (2) calculating ‘depth
invariant indices’ using an image based pre-processing method that aims to
remove the effect of variable depth by calculating new image layers from the
logarithm of pairs of bands (Lyzenga 1981; Green et al. 2000). Both depth correction and depth invariant indices require regions of homogenous benthos with
variable depth within the image. In coral reef applications sand is ideal and has the
advantage of providing a bright relatively strong signal over the visible wavelength
range. Both methods rely on the concept of approximate exponential attenuation of
light as expressed by the diffuse attenuation coefficient, k i , in each band i, such that
for depth z, the above water reflectance in band i is proportional to exp(-2k i z). For
depth correction the k i values are estimated by regression for areas of sand at a
range of depths, the exponential relationship can then adjust other pixels to a fixed
depth (Bejarano et al. 2010). For classification approaches the uniform depth does
not have to be zero. For depth invariant indices new bands are created using
r = ln(r i ) - (k i / k j ) 9 ln(r j ), where r i and r j are the reflectance in bands i and j,
and k i and k j are the diffuse attenuation coefficients at the wavelengths of those
bands. In this case it is not necessary to calculate the actual k values, the ratio (k i /
k j ) can be derived by regression between bands i and j over variable depth sand
even when actual depth is unknown (Lyzenga 1981; Green et al. 2000). Depth
invariant indices are typically used with classification (Sect. 4.3.1; Green et al.
2000) but can also be incorporated into band difference methods (Sect. 4.3.2; Isoun
et al. 2003).
94
J. D. Hedley
is currently demonstrated as anecdotally successful only in a single test case.
For airborne campaigns it is sensible to design image acquisition strategy to
minimize glint. Flight paths towards or away from the sun with solar zenith angles
of 30°–60° have been recommended (Mustard et al. 2001; Dekker et al. 2003). In
one specific example, Lesser and Mobley (2007) acquired imagery with solar
zenith 40°–55° and flight direction aligned with the sun.
4.2.6 Depth Correction
Water depth has a major effect on the spectral reflectance measured over a reef.
The range of the absorption coefficient of water over the 400–700 nm wavelengths
is high and the spectral shape of absorption is also affected by water constituents
such as CDOM (Fig. 4.2). Hence the same bottom reflectance at different depths
will give rise to very different above-water reflectances, and this is clearly a
complication in identifying the bottom type. Two simple methods can be used to
remove the effect of variable depth: (1) ‘depth correction’ requires that depth be
known across the image either from acoustic (Bejarano et al. 2010; Chaps. 8–10)
or LiDAR data (Chaps. 5–7), or for some reef topographies a simple stratification
into depth zones may be possible (Mumby et al. 2004); or (2) calculating ‘depth
invariant indices’ using an image based pre-processing method that aims to
remove the effect of variable depth by calculating new image layers from the
logarithm of pairs of bands (Lyzenga 1981; Green et al. 2000). Both depth correction and depth invariant indices require regions of homogenous benthos with
variable depth within the image. In coral reef applications sand is ideal and has the
advantage of providing a bright relatively strong signal over the visible wavelength
range. Both methods rely on the concept of approximate exponential attenuation of
light as expressed by the diffuse attenuation coefficient, k i , in each band i, such that
for depth z, the above water reflectance in band i is proportional to exp(-2k i z). For
depth correction the k i values are estimated by regression for areas of sand at a
range of depths, the exponential relationship can then adjust other pixels to a fixed
depth (Bejarano et al. 2010). For classification approaches the uniform depth does
not have to be zero. For depth invariant indices new bands are created using
r = ln(r i ) - (k i / k j ) 9 ln(r j ), where r i and r j are the reflectance in bands i and j,
and k i and k j are the diffuse attenuation coefficients at the wavelengths of those
bands. In this case it is not necessary to calculate the actual k values, the ratio (k i /
k j ) can be derived by regression between bands i and j over variable depth sand
even when actual depth is unknown (Lyzenga 1981; Green et al. 2000). Depth
invariant indices are typically used with classification (Sect. 4.3.1; Green et al.
2000) but can also be incorporated into band difference methods (Sect. 4.3.2; Isoun
et al. 2003).
94
J. D. Hedley