3.3.1 Defining Reef Environments
Ambiguous terminology concerning features within a coral reef ecosystem has
threatened to limit the effectiveness of remote sensing. The problem is that accurate
identification of reef features is a function of the operator’s knowledge of reef
terminology as well as the spatial and spectral capabilities of the sensor (Mumby and
Harborne, 1999). The differences in terminology pose difficulties when comparisons
are drawn between different types of classified imagery, especially within images of
higher levels of detail (Mumby and Harborne, 1999). In an effort to minimize these
effects, several authors have suggested that scientists adopt a systematic, objective, and
replicable classification of reef habitats (Done, 1999; Andréfouët and Claereboudt,
2000; Edinger and Risk, 2000; Phinn et al., 2000; Mumby, 2001).
Couched within biological terminology, the term habitat is defined as “An
aggregation of different species of organisms living and interacting within the same
area” (Wasserman, 1975). Although defining a habitat explicitly would improve
classifications of remotely sensed data, the above definition implies the aggregation of
variable reef structures and biological conditions at varying spatial scales that are nondecipherable with current imaging technologies (Andréfouët and Claereboudt, 2000).
A solution to variable habitat terminology has been revealed in the development of a
framework that combines remote sensing capabilities and concepts from landscape
ecology (Phinn, 1998). Moreover, developments in the “scaling theory” have provided
direction for scaling field to image data and matching remotely sensed data to relevant
state variables (Curran et al., 1998). The procedure involves several steps that begin
with a thorough examination of the spatially distinct natural groupings of features
visible within a coral reef environment.
A study pioneering this research derived benthic classes based on an objective
approach using Agglomerative Hierarchical Classification of field data and Similarity
Percentage Analysis (Mumby and Harborne, 1999). Results indicated that the
descriptive resolution of remote sensing techniques is still dependent upon the limited
subjective decisions of the operator. In addition, the method is useful only for turbid
free images, in areas in which Jerlov 1 waters are consistent throughout the scene.
Although subjectivity plays a role in defining the habitat of interest, the limitations of
sensors abilities control the range of subjective decisions and thus influence the
selection of the more appropriate sensor (Andréfouët et al., 2002).
3.3.2 Using the Spectral Signature for Class Discrimination
Normally we assume that relatively large differences between the intrinsic spectral
signatures of corals will provide a relatively high probability that coral reef features
within an image will be successfully delineated. However, from a radiometric or sensor
perspective, often the electromagnetic energy of the target of interest can overlap
between substrate types. The problem is that the optical signal leaving the coral surface
is a function of that feature’s pigmentation, structure, branch orientation, and internal
shadowing characteristics (Holden and LeDrew, 1998; Hochberg and Atkinson, 2000;
Joyce and Phinn, 2002; Lubin et al., 2001; Hedley and Mumby, 2002), elements that
are difficult to separate out.
Studies of coral spectral signatures measured in situ typically fall into two
categories. First, signatures are often examined as a function of variations in pigment
densities that characterize the visual color of coral species (Falkowski et al., 1998).
Several studies have examined the contribution of color to measured radiance,
particularly comparing structures that lack color, which results from bleaching, to coral
structures saturated with zooxanthellae, a measure of health (Holden and LeDrew,
262
Newman, LeDrew and Lim
Ambiguous terminology concerning features within a coral reef ecosystem has
threatened to limit the effectiveness of remote sensing. The problem is that accurate
identification of reef features is a function of the operator’s knowledge of reef
terminology as well as the spatial and spectral capabilities of the sensor (Mumby and
Harborne, 1999). The differences in terminology pose difficulties when comparisons
are drawn between different types of classified imagery, especially within images of
higher levels of detail (Mumby and Harborne, 1999). In an effort to minimize these
effects, several authors have suggested that scientists adopt a systematic, objective, and
replicable classification of reef habitats (Done, 1999; Andréfouët and Claereboudt,
2000; Edinger and Risk, 2000; Phinn et al., 2000; Mumby, 2001).
Couched within biological terminology, the term habitat is defined as “An
aggregation of different species of organisms living and interacting within the same
area” (Wasserman, 1975). Although defining a habitat explicitly would improve
classifications of remotely sensed data, the above definition implies the aggregation of
variable reef structures and biological conditions at varying spatial scales that are nondecipherable with current imaging technologies (Andréfouët and Claereboudt, 2000).
A solution to variable habitat terminology has been revealed in the development of a
framework that combines remote sensing capabilities and concepts from landscape
ecology (Phinn, 1998). Moreover, developments in the “scaling theory” have provided
direction for scaling field to image data and matching remotely sensed data to relevant
state variables (Curran et al., 1998). The procedure involves several steps that begin
with a thorough examination of the spatially distinct natural groupings of features
visible within a coral reef environment.
A study pioneering this research derived benthic classes based on an objective
approach using Agglomerative Hierarchical Classification of field data and Similarity
Percentage Analysis (Mumby and Harborne, 1999). Results indicated that the
descriptive resolution of remote sensing techniques is still dependent upon the limited
subjective decisions of the operator. In addition, the method is useful only for turbid
free images, in areas in which Jerlov 1 waters are consistent throughout the scene.
Although subjectivity plays a role in defining the habitat of interest, the limitations of
sensors abilities control the range of subjective decisions and thus influence the
selection of the more appropriate sensor (Andréfouët et al., 2002).
3.3.2 Using the Spectral Signature for Class Discrimination
Normally we assume that relatively large differences between the intrinsic spectral
signatures of corals will provide a relatively high probability that coral reef features
within an image will be successfully delineated. However, from a radiometric or sensor
perspective, often the electromagnetic energy of the target of interest can overlap
between substrate types. The problem is that the optical signal leaving the coral surface
is a function of that feature’s pigmentation, structure, branch orientation, and internal
shadowing characteristics (Holden and LeDrew, 1998; Hochberg and Atkinson, 2000;
Joyce and Phinn, 2002; Lubin et al., 2001; Hedley and Mumby, 2002), elements that
are difficult to separate out.
Studies of coral spectral signatures measured in situ typically fall into two
categories. First, signatures are often examined as a function of variations in pigment
densities that characterize the visual color of coral species (Falkowski et al., 1998).
Several studies have examined the contribution of color to measured radiance,
particularly comparing structures that lack color, which results from bleaching, to coral
structures saturated with zooxanthellae, a measure of health (Holden and LeDrew,
262
Newman, LeDrew and Lim