sea state. Fully understanding the confounding affect of environmental fluctuations
is a prerequisite of image based change detection that is not trivial. For example,
sediment resuspension on a reef could be mistaken for bleaching. Relatively little
is known about the optical effects of reef sediments in general (Hedley 2011a).
In the earliest days of Landsat applications on reefs, Bina and Ombac (1979)
proposed the use of multiple time point data to minimize the effect of tidal variation. Still to this date the use of multiple time series data on coral reefs has largely
been restricted to Landsat (Andréfouët et al. 2001; Dustan et al. 2002; Phinney
et al. 2002; Schuyler et al. 2006), which is clearly a function of data availability.
Despite the well-established advantages of airborne hyperspectral data for one-off
mapping (Mumby et al. 1997) operational satellite data will continue to have a role
to play. Dustan et al. (2002) and Phinney et al. (2002) made the promising
observation of reflectance changes over 20 Landsat images consistent with known
coral to algal shifts in the Caribbean. ESA’s upcoming Sentinel 2 is often posited
as a Landsat and SPOT continuity mission but in fact this instrument has five
narrow bands that are ‘hyperspectral’ in character (Table. 4.1). Models suggest it
will outperform Landsat in reef applications (Hedley et al. 2012a) and combined
with 10 m spatial resolution, 5-day revisit for coastal areas, the capability for reef
change detection using Sentinel 2 will surely be a priority for future investigation.
4.3.6 Inversion Methods
The most recent developments in shallow water remote sensing are in radiative
transfer model inversion methods, which are variously referred to as ‘physicsbased’ or ‘semi-analytical’ and largely stem from key publications by Lee et al.
(1998, 1999) and Mobley et al. (2005). The idea behind these methods is to
construct a ‘forward model’ for above-water spectral reflectance that takes a
number of input parameters including depth, the concentration of various water
column constituents such as CDOM and phytoplankton, and the choice of bottom
material, which determines the bottom spectral reflectance. For each pixel in an
image an inversion algorithm effectively runs the model backwards, to find the
best combination of input parameter values to give the closest spectral match
between the model output and the measured pixel reflectance. These methods were
developed for hyperspectral data, since numerous wavelength bands are necessary
to tease apart the spectral influence of all the different components (Fig. 4.2 and
4.3). The methods can be equally applied to multispectral data but the inherent
uncertainty will increase (see Sect. 4.4).
Figures 4.7 and 4.8 illustrate the kind of products that can be derived from these
inversion methods. The raw outputs are image layers for each of the model input
parameters: bathymetry, benthic type, and values that express the relative concentrations of CDOM or phytoplankton and the water column backscatter. From these
raw parameters it is easy to calculate secondary outputs, the bottom reflectance can
be reconstructed, water column optical properties can be recombined to give
4 Hyperspectral Applications
101
is a prerequisite of image based change detection that is not trivial. For example,
sediment resuspension on a reef could be mistaken for bleaching. Relatively little
is known about the optical effects of reef sediments in general (Hedley 2011a).
In the earliest days of Landsat applications on reefs, Bina and Ombac (1979)
proposed the use of multiple time point data to minimize the effect of tidal variation. Still to this date the use of multiple time series data on coral reefs has largely
been restricted to Landsat (Andréfouët et al. 2001; Dustan et al. 2002; Phinney
et al. 2002; Schuyler et al. 2006), which is clearly a function of data availability.
Despite the well-established advantages of airborne hyperspectral data for one-off
mapping (Mumby et al. 1997) operational satellite data will continue to have a role
to play. Dustan et al. (2002) and Phinney et al. (2002) made the promising
observation of reflectance changes over 20 Landsat images consistent with known
coral to algal shifts in the Caribbean. ESA’s upcoming Sentinel 2 is often posited
as a Landsat and SPOT continuity mission but in fact this instrument has five
narrow bands that are ‘hyperspectral’ in character (Table. 4.1). Models suggest it
will outperform Landsat in reef applications (Hedley et al. 2012a) and combined
with 10 m spatial resolution, 5-day revisit for coastal areas, the capability for reef
change detection using Sentinel 2 will surely be a priority for future investigation.
4.3.6 Inversion Methods
The most recent developments in shallow water remote sensing are in radiative
transfer model inversion methods, which are variously referred to as ‘physicsbased’ or ‘semi-analytical’ and largely stem from key publications by Lee et al.
(1998, 1999) and Mobley et al. (2005). The idea behind these methods is to
construct a ‘forward model’ for above-water spectral reflectance that takes a
number of input parameters including depth, the concentration of various water
column constituents such as CDOM and phytoplankton, and the choice of bottom
material, which determines the bottom spectral reflectance. For each pixel in an
image an inversion algorithm effectively runs the model backwards, to find the
best combination of input parameter values to give the closest spectral match
between the model output and the measured pixel reflectance. These methods were
developed for hyperspectral data, since numerous wavelength bands are necessary
to tease apart the spectral influence of all the different components (Fig. 4.2 and
4.3). The methods can be equally applied to multispectral data but the inherent
uncertainty will increase (see Sect. 4.4).
Figures 4.7 and 4.8 illustrate the kind of products that can be derived from these
inversion methods. The raw outputs are image layers for each of the model input
parameters: bathymetry, benthic type, and values that express the relative concentrations of CDOM or phytoplankton and the water column backscatter. From these
raw parameters it is easy to calculate secondary outputs, the bottom reflectance can
be reconstructed, water column optical properties can be recombined to give
4 Hyperspectral Applications
101