spectral absorption coefficients, or depth and attenuation can be combined to
produce a relative benthic light product. However, the accuracy of the raw individual layers may differ widely. Depth estimation is fairly robust, and in reef
environments is always the most accurate retrieval (Dekker et al. 2011, 2009a,
2009b, 2010; Mobley et al. 2005; Lesser and Mobley 2007). In reef environments
where the water is often very clear there is little ‘signal’ to establish the optical
properties of the water column, and hence CDOM, phytoplankton and backscatter
retrievals are often invalid and simply correlate with bottom reflectance. The situation is very different in waters with higher phytoplankton or CDOM levels (Lee
et al. 2001). In waters that are not ‘optically shallow’, conversely, the bottom
reflectance cannot be determined. A further potential disadvantage of spectral
matching approaches is that the measure of spectral fit employed usually favors
retrieval of factors, such as depth, which affect the whole spectrum. Narrow
wavelength features like those discussed in Sect. 4.3.2 only weakly affect the
matching process. Benthic type mapping may benefit from a hybrid approach where
matching is weighted in wavelength regions of known pigment features, but this has
not yet been tried. In all cases basic inversion methods offer no ‘quality assurance’
and simply rely on user interpretation or validation to deduce which estimations are
reliable. The next section discusses refinements that can quantify or automate
quality assurance.
With respect to choosing a methodological approach, the methods of Lee et al.
(1998, 1999) and Mobley et al. (2005) differ substantially in the practical sense of
the forward model and inversion method used, but fundamentally the methods are
very similar. Mobley et al. (2005) populates a look-up table of reflectances using
the commercial software Hydrolight (or the associated Ecolight). Hydrolight is a
full numerical integration based model for light propagation in water, sometimes
referred to as an ‘exact’ model in that it embodies the physical theory of radiative
transfer. Lee et al. (1998, 1999) uses a more approximate forward model that gives
above-water reflectance directly as the output of a series of simple equations. This
model can be evaluated sufficiently fast to apply successive approximation algorithm such as Levenberg–Marquardt (Wolfe 1978) at each pixel. However, several
key parameters in Lee et al. (1998) were derived by multiple runs of Hydrolight,
hence with the same parameterization the models should give almost identical
results. The majority of differences are due to implementation details and discretization or local optima problems from inversion via look up tables versus
successive approximation (Dekker et al. 2011; Hedley et al. 2009a, unpublished
data). Since no off-the-shelf software currently exists for these algorithms the lookup table approach is the most straightforward solution for practitioners. While
Hydrolight (Mobley and Sundman 2000) is a commercial package, a free validated
open-source implementation of the same solution algorithm exists, PlanarRad
(Hedley 2011b). PlanarRad is functionally similar to Hydrolight but lacks chlorophyll fluorescence. Incorporating chlorophyll fluorescence may be important in
high phytoplankton environments (Tote et al. 2011) but is an insignificant factor in
most reef environments.
4 Hyperspectral Applications
103
produce a relative benthic light product. However, the accuracy of the raw individual layers may differ widely. Depth estimation is fairly robust, and in reef
environments is always the most accurate retrieval (Dekker et al. 2011, 2009a,
2009b, 2010; Mobley et al. 2005; Lesser and Mobley 2007). In reef environments
where the water is often very clear there is little ‘signal’ to establish the optical
properties of the water column, and hence CDOM, phytoplankton and backscatter
retrievals are often invalid and simply correlate with bottom reflectance. The situation is very different in waters with higher phytoplankton or CDOM levels (Lee
et al. 2001). In waters that are not ‘optically shallow’, conversely, the bottom
reflectance cannot be determined. A further potential disadvantage of spectral
matching approaches is that the measure of spectral fit employed usually favors
retrieval of factors, such as depth, which affect the whole spectrum. Narrow
wavelength features like those discussed in Sect. 4.3.2 only weakly affect the
matching process. Benthic type mapping may benefit from a hybrid approach where
matching is weighted in wavelength regions of known pigment features, but this has
not yet been tried. In all cases basic inversion methods offer no ‘quality assurance’
and simply rely on user interpretation or validation to deduce which estimations are
reliable. The next section discusses refinements that can quantify or automate
quality assurance.
With respect to choosing a methodological approach, the methods of Lee et al.
(1998, 1999) and Mobley et al. (2005) differ substantially in the practical sense of
the forward model and inversion method used, but fundamentally the methods are
very similar. Mobley et al. (2005) populates a look-up table of reflectances using
the commercial software Hydrolight (or the associated Ecolight). Hydrolight is a
full numerical integration based model for light propagation in water, sometimes
referred to as an ‘exact’ model in that it embodies the physical theory of radiative
transfer. Lee et al. (1998, 1999) uses a more approximate forward model that gives
above-water reflectance directly as the output of a series of simple equations. This
model can be evaluated sufficiently fast to apply successive approximation algorithm such as Levenberg–Marquardt (Wolfe 1978) at each pixel. However, several
key parameters in Lee et al. (1998) were derived by multiple runs of Hydrolight,
hence with the same parameterization the models should give almost identical
results. The majority of differences are due to implementation details and discretization or local optima problems from inversion via look up tables versus
successive approximation (Dekker et al. 2011; Hedley et al. 2009a, unpublished
data). Since no off-the-shelf software currently exists for these algorithms the lookup table approach is the most straightforward solution for practitioners. While
Hydrolight (Mobley and Sundman 2000) is a commercial package, a free validated
open-source implementation of the same solution algorithm exists, PlanarRad
(Hedley 2011b). PlanarRad is functionally similar to Hydrolight but lacks chlorophyll fluorescence. Incorporating chlorophyll fluorescence may be important in
high phytoplankton environments (Tote et al. 2011) but is an insignificant factor in
most reef environments.
4 Hyperspectral Applications
103