greater if the aircraft rolls. Cross track effects are greatest when the direction of
flight is perpendicular to the solar azimuth, since then the instrument cross-track
direction is in the solar plane and direct solar reflectance is the greatest (Mobley
1994; Kay et al. 2011).
In an aquatic application several processes will cause the detected radiance to
vary with view angle: (1) at steeper angles the path through the atmosphere is
greater; (2) reflectance of the upper side of the air–water interface is highly
directionally dependent (Kay et al. 2011); (3) at steeper angles the transmitted path
through the water is greater; and (4) the benthos may exhibit non-Lambertian bidirectional reflectance function (BRDF) (Hedley and Enríquez 2010). Of these, the
effect of air–water interface reflectance is by far the most apparent (Fig. 4.2).
While atmospheric path can be corrected by atmospheric correction codes, the inwater path and benthic BRDF in aquatic applications are rarely corrected. Image
processing software packages such as ENVI (Exelis VIS 2012) may contain crosstrack correction algorithms, but typically these have been developed with terrestrial applications in mind and may be parameterized based on canopy BRDFs only
(Kennedy et al. 1997). Since the primary cross-track effect in an aquatic image is
reflection from the air water interface, a sunglint correction procedure may be used
provided there is a deep water area across the entire track from which to parameterize the visible band-NIR relationship (see Sect. 4.2.5 and Kay et al. 2009).
4.2.5 Sunglint Correction
A major source of pixel-to-pixel variation in high spatial resolution images is the
reflection of the sun from the upper side of the water surface. For pixels smaller
than surface waves, undulations in the water surface introduce bright speckle or
wave shaped ‘sunglint’ patterns (Fig. 4.4). As spatial resolution decreases below
that of waves, the effect tends to become a gradual cross-image effect (Sect. 4.2.4).
These patterns obscure benthic features and will confound classification algorithms. For model inversion algorithms (Sect. 4.3.5) the surface reflectance is a
complicating factor. The reflected component of the radiance has never penetrated
the water surface, so spectrally it carries no information about sub-surface features
and it simplifies analysis to remove it. Although note that other remote sensing
methods can usefully infer wave energy and bathymetry from the effect on wave
glint patterns (Cureton et al. 2007; Splinter and Holman 2009).
A number of sunglint removal algorithms have been proposed (Joyce 2004;
Hochberg et al. 2003b; Hedley et al. 2005; Lyzenga et al. 2006) which all rely on
the assumption that the spatial variation in a near-infrared (NIR) band is solely due
to glint. Kay et al. (2009) demonstrated that these algorithms are essentially
functionally identical and differ only in what is assumed to be the ‘base level’ NIR.
Using the depth of the oxygen absorption feature at 760 nm is an alternative
approach that has been suggested by Kutser et al. (2009), assuming the imagery
92
J. D. Hedley
flight is perpendicular to the solar azimuth, since then the instrument cross-track
direction is in the solar plane and direct solar reflectance is the greatest (Mobley
1994; Kay et al. 2011).
In an aquatic application several processes will cause the detected radiance to
vary with view angle: (1) at steeper angles the path through the atmosphere is
greater; (2) reflectance of the upper side of the air–water interface is highly
directionally dependent (Kay et al. 2011); (3) at steeper angles the transmitted path
through the water is greater; and (4) the benthos may exhibit non-Lambertian bidirectional reflectance function (BRDF) (Hedley and Enríquez 2010). Of these, the
effect of air–water interface reflectance is by far the most apparent (Fig. 4.2).
While atmospheric path can be corrected by atmospheric correction codes, the inwater path and benthic BRDF in aquatic applications are rarely corrected. Image
processing software packages such as ENVI (Exelis VIS 2012) may contain crosstrack correction algorithms, but typically these have been developed with terrestrial applications in mind and may be parameterized based on canopy BRDFs only
(Kennedy et al. 1997). Since the primary cross-track effect in an aquatic image is
reflection from the air water interface, a sunglint correction procedure may be used
provided there is a deep water area across the entire track from which to parameterize the visible band-NIR relationship (see Sect. 4.2.5 and Kay et al. 2009).
4.2.5 Sunglint Correction
A major source of pixel-to-pixel variation in high spatial resolution images is the
reflection of the sun from the upper side of the water surface. For pixels smaller
than surface waves, undulations in the water surface introduce bright speckle or
wave shaped ‘sunglint’ patterns (Fig. 4.4). As spatial resolution decreases below
that of waves, the effect tends to become a gradual cross-image effect (Sect. 4.2.4).
These patterns obscure benthic features and will confound classification algorithms. For model inversion algorithms (Sect. 4.3.5) the surface reflectance is a
complicating factor. The reflected component of the radiance has never penetrated
the water surface, so spectrally it carries no information about sub-surface features
and it simplifies analysis to remove it. Although note that other remote sensing
methods can usefully infer wave energy and bathymetry from the effect on wave
glint patterns (Cureton et al. 2007; Splinter and Holman 2009).
A number of sunglint removal algorithms have been proposed (Joyce 2004;
Hochberg et al. 2003b; Hedley et al. 2005; Lyzenga et al. 2006) which all rely on
the assumption that the spatial variation in a near-infrared (NIR) band is solely due
to glint. Kay et al. (2009) demonstrated that these algorithms are essentially
functionally identical and differ only in what is assumed to be the ‘base level’ NIR.
Using the depth of the oxygen absorption feature at 760 nm is an alternative
approach that has been suggested by Kutser et al. (2009), assuming the imagery
92
J. D. Hedley