(Lee et al. 1998, 1999, 2001; Mobley et al. 2005; Brando et al. 2009; Hedley et al.
2009a; Dekker et al. 2011). These methods are primarily designed for hyperspectral
data and can routinely achieve excellent results for bathymetric extractions (Hedley
et al. 2009a). Results for water column optical properties and benthic composition
are more variable but can be well estimated in some environments (Lee et al. 2001;
Mobley et al. 2005; Goodman and Ustin, 2007).
Cross-comparing image analysis methods is difficult as there is no universally
applicable classification scheme that allows direct inter-comparison of methods for
benthic mapping. Different methods may return fundamentally different types of
information. Mumby and Harbourne (1999) designed a hierarchical habitat classification scheme for use in Caribbean reefs. A hierarchical scheme has the benefit
that classes can be merged for cross comparison between methods of different
‘descriptive resolutions’ (Green et al. 1996). For coral bleaching surveys Siebeck
et al. (2006) have promoted the use of a color reference card. In a field survey or
remote sensing context time is well spent in the planning stage to devise a scheme
that will allow a meaningful merging of classes if the initial analysis proves too
ambitious.
4.1.2 Design and Operational Considerations
The term hyperspectral does not have a rigidly defined meaning, but conveys the
sense of data sources with numerous wave bands that are spectrally narrow. The
majority of published hyperspectral analyses on coral reefs have been achieved
with sensors mounted on airplanes such as the Compact Airborne Spectrographic
Imager (CASI) or the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) (Mumby et al. 2004; Goodman and Ustin 2007). Only a few satellite sensors
that could be classed as hyperspectral exist and have been used in shallow water
applications, the Hyperion sensor being the most notable to date (Lee et al. 2007).
However, the situation regarding available hyperspectral sensors is dynamic,
upcoming sensors such as the German EnMAP scheduled for launch 2013 and
NASA HyspIRI in 2015 (Table 4.1) may completely change the landscape of
hyperspectral reef applications in the near future.
The distinction between multispectral and hyperspectral is sometimes vague
and becomes increasingly blurred with the multitude of sensor designs (Table 4.1).
For example, in a fairly successful ‘hyperspectral’ demonstration Mumby et al.
(2004) used only 6 bands of a CASI dataset. Next-generation satellite sensor
families traditionally seen as ‘multispectral’ increasingly have hyperspectral-like
features, such the extra bands available in DigitalGlobe’s WorldView 2
(Table 4.1), or the narrow bands of the European Space Agency’s (ESA)
upcoming Sentinel 2 (Table 4.1; Hedley et al. 2012a). Hyperspectrally-orientated
techniques are likely to see increased use on a wider range of sensors in the future.
Most techniques are equally applicable to multispectral and hyperspectral data, the
accuracy of estimations may vary but this is not necessarily tied to the number of
4 Hyperspectral Applications
83
2009a; Dekker et al. 2011). These methods are primarily designed for hyperspectral
data and can routinely achieve excellent results for bathymetric extractions (Hedley
et al. 2009a). Results for water column optical properties and benthic composition
are more variable but can be well estimated in some environments (Lee et al. 2001;
Mobley et al. 2005; Goodman and Ustin, 2007).
Cross-comparing image analysis methods is difficult as there is no universally
applicable classification scheme that allows direct inter-comparison of methods for
benthic mapping. Different methods may return fundamentally different types of
information. Mumby and Harbourne (1999) designed a hierarchical habitat classification scheme for use in Caribbean reefs. A hierarchical scheme has the benefit
that classes can be merged for cross comparison between methods of different
‘descriptive resolutions’ (Green et al. 1996). For coral bleaching surveys Siebeck
et al. (2006) have promoted the use of a color reference card. In a field survey or
remote sensing context time is well spent in the planning stage to devise a scheme
that will allow a meaningful merging of classes if the initial analysis proves too
ambitious.
4.1.2 Design and Operational Considerations
The term hyperspectral does not have a rigidly defined meaning, but conveys the
sense of data sources with numerous wave bands that are spectrally narrow. The
majority of published hyperspectral analyses on coral reefs have been achieved
with sensors mounted on airplanes such as the Compact Airborne Spectrographic
Imager (CASI) or the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS) (Mumby et al. 2004; Goodman and Ustin 2007). Only a few satellite sensors
that could be classed as hyperspectral exist and have been used in shallow water
applications, the Hyperion sensor being the most notable to date (Lee et al. 2007).
However, the situation regarding available hyperspectral sensors is dynamic,
upcoming sensors such as the German EnMAP scheduled for launch 2013 and
NASA HyspIRI in 2015 (Table 4.1) may completely change the landscape of
hyperspectral reef applications in the near future.
The distinction between multispectral and hyperspectral is sometimes vague
and becomes increasingly blurred with the multitude of sensor designs (Table 4.1).
For example, in a fairly successful ‘hyperspectral’ demonstration Mumby et al.
(2004) used only 6 bands of a CASI dataset. Next-generation satellite sensor
families traditionally seen as ‘multispectral’ increasingly have hyperspectral-like
features, such the extra bands available in DigitalGlobe’s WorldView 2
(Table 4.1), or the narrow bands of the European Space Agency’s (ESA)
upcoming Sentinel 2 (Table 4.1; Hedley et al. 2012a). Hyperspectrally-orientated
techniques are likely to see increased use on a wider range of sensors in the future.
Most techniques are equally applicable to multispectral and hyperspectral data, the
accuracy of estimations may vary but this is not necessarily tied to the number of
4 Hyperspectral Applications
83