Increasing depth modifies the above-water spectral reflectance across all wavelengths in a characteristic way that presents a strong ‘signal’ even in the presence of
water column constituents or variable bottom reflectance. Another consequence of
the exponential attenuation with depth is that small depth changes in shallow areas
are equally resolvable as large changes in deeper water; hence the sensitivity of
bathymetric extraction methods is typically sensibly structured. If bathymetric
extraction from hyperspectral data is desired the very best methods are the multiparameter algorithms described in Sect. 4.3.6 (Fig. 4.6). Although these methods
do not require any a priori bathymetry data, without any it is difficult to assess if the
resultant map can be trusted. However, the implementation of model inversion
methods is technically challenging and at the time of writing no off-the-shelf
processing software exists. Hence the simpler bathymetry-only methods (Lyzenga
et al. 2006) may be more practical. The strength of many of these latter methods is
that if bathymetry is known for some image pixels, then parameterizing the
methods directly from image pixels ameliorates the effects of imperfect atmospheric correction or other data bias.
4.3.5 Change Detection
A distinct approach to coral reef remote sensing analysis that warrants further
development is multi-image change detection. The principle is simple, in that two
or more images at different time points are spatially aligned by geo-rectification
and regions of change are identified on a pixel-by-pixel basis. However, there are
substantial practical challenges:
• Radiometric alignment - different sensor characteristics, and illumination or
atmospheric conditions must be removed or factored out.
• Spatial alignment - benthic features on reefs occur at sub-pixel scales even for
the highest available image resolutions.
• Acquisition of two or more suitable images - costs may be prohibitive or satellite
acquisitions may not be conveniently scheduled.
A change detection approach was the basis of the demonstration of the detection of the Keppel Islands 2002 bleaching event using IKONOS data by Elvidge
et al. (2004), which remains at the time of writing probably the only peer-review
published demonstration of bleaching detection from optical satellite data. In
another example Dadhich et al. (2011) assessed post-bleaching change in coral
cover using two QuickBird images. Due to low availability of hyperspectral
satellite data, and the cost and complications of airborne data, no hyperspectral
multi-image change detection appears to have been attempted in reef environments. However, the launch of future hyperspectral satellite sensors such as EnMAP, with its 4-day revisit time, will provide a new impetus for multi-image
approaches. In addition to detecting change, using multiple images provides more
information and could help factor out variations due to atmospheric conditions and
100
J. D. Hedley
water column constituents or variable bottom reflectance. Another consequence of
the exponential attenuation with depth is that small depth changes in shallow areas
are equally resolvable as large changes in deeper water; hence the sensitivity of
bathymetric extraction methods is typically sensibly structured. If bathymetric
extraction from hyperspectral data is desired the very best methods are the multiparameter algorithms described in Sect. 4.3.6 (Fig. 4.6). Although these methods
do not require any a priori bathymetry data, without any it is difficult to assess if the
resultant map can be trusted. However, the implementation of model inversion
methods is technically challenging and at the time of writing no off-the-shelf
processing software exists. Hence the simpler bathymetry-only methods (Lyzenga
et al. 2006) may be more practical. The strength of many of these latter methods is
that if bathymetry is known for some image pixels, then parameterizing the
methods directly from image pixels ameliorates the effects of imperfect atmospheric correction or other data bias.
4.3.5 Change Detection
A distinct approach to coral reef remote sensing analysis that warrants further
development is multi-image change detection. The principle is simple, in that two
or more images at different time points are spatially aligned by geo-rectification
and regions of change are identified on a pixel-by-pixel basis. However, there are
substantial practical challenges:
• Radiometric alignment - different sensor characteristics, and illumination or
atmospheric conditions must be removed or factored out.
• Spatial alignment - benthic features on reefs occur at sub-pixel scales even for
the highest available image resolutions.
• Acquisition of two or more suitable images - costs may be prohibitive or satellite
acquisitions may not be conveniently scheduled.
A change detection approach was the basis of the demonstration of the detection of the Keppel Islands 2002 bleaching event using IKONOS data by Elvidge
et al. (2004), which remains at the time of writing probably the only peer-review
published demonstration of bleaching detection from optical satellite data. In
another example Dadhich et al. (2011) assessed post-bleaching change in coral
cover using two QuickBird images. Due to low availability of hyperspectral
satellite data, and the cost and complications of airborne data, no hyperspectral
multi-image change detection appears to have been attempted in reef environments. However, the launch of future hyperspectral satellite sensors such as EnMAP, with its 4-day revisit time, will provide a new impetus for multi-image
approaches. In addition to detecting change, using multiple images provides more
information and could help factor out variations due to atmospheric conditions and
100
J. D. Hedley