location of the habitats, where spectral signatures may be taken; and (3) to test the
accuracy of a classified image (Green et al., 2000). A variety of field survey methods
may be employed (Table 9).
3.3.4 Capitalizing on Multi-Temporal Coverage
Following the launch of Landsat 7 ETM, a series of images of coral reefs from
various regions has been collected systematically as a part of the Long Term
Acquisition Plan (Gash et al., 2000). The Plan is the first major attempt to target
isolated coral reef ecosystems repetitively. Users are being afforded a unique
opportunity to focus their energies on change detection methods without the
complications associated with mixing different image types of varying spatial or
spectral specifications (Palandro et al., 2001).
Andréfouët et al. (2001) made an early assessment of the potential of the Landsat 7
ETM+ to detect change within a reef environment. Images were acquired within a brief
time frame during which there was no major disturbance to reefs. The lack of
disturbance permitted researchers to assess the stability of the images through time and
to estimate biases that may be associated with change detection strategies. Having
assumed that areas are large enough to provide an unmixed signal, the authors
suggested that there is difficulty in detecting changes from one object to another even
when the acquired images have identical specifications. The reason for this is that any
occurring change must be visible and cover a large enough areal extent to saturate
pixels. The effects of a hurricane, for example, would be easily identified since such a
storm would destroy living structures and transform a heterogeneous environment into
a homogeneous platform. Therefore, depending on the physical or biological shift
within a coral reef ecosystem–disturbance, phase, or strategy–spectral differences may
or may not be dissimilar enough to be detected. Un-mixing techniques are cited as a
potential solution because they would allow observers to detect intra-pixel changes, but
may be realistically applicable to no more than three classes at a time–sand,
background, and foreground, for instance.
Several change-detection techniques applied to images of coral reef environments
are derivatives of land or ice-based techniques (Lunetta and Elvidge, 1999). The Getis
statistic, for example, can be used to examine the change in reef homogeneity. This
measure considers the value of the reflectance within a single pixel and the relationship
between that pixel and the surrounding reference pixels. It has been hypothesized
that a healthy coral reef ecosystem will be heterogeneous and display negative
autocorrelation, while a disturbed reef similar in bottom type over a large area will be
spatially homogeneous and display a positive autocorrelation (LeDrew et al., 2004).
Two other measures available to researchers are principal components analysis
(Mas, 1999) and the Mahalanobis distance classification (Palandro et al., 2001), which
are regularly used prior to comparing images to detect change. It is generally accepted
that there are four aspects of change detection that are important when monitoring
natural resources: (1) detecting that a change has occurred; (2) identifying the nature of
the change; (3) measuring the areal extent of the change; and (4) assessing the spatial
pattern of the change (Klemas, 2001).
266
Newman, LeDrew and Lim
accuracy of a classified image (Green et al., 2000). A variety of field survey methods
may be employed (Table 9).
3.3.4 Capitalizing on Multi-Temporal Coverage
Following the launch of Landsat 7 ETM, a series of images of coral reefs from
various regions has been collected systematically as a part of the Long Term
Acquisition Plan (Gash et al., 2000). The Plan is the first major attempt to target
isolated coral reef ecosystems repetitively. Users are being afforded a unique
opportunity to focus their energies on change detection methods without the
complications associated with mixing different image types of varying spatial or
spectral specifications (Palandro et al., 2001).
Andréfouët et al. (2001) made an early assessment of the potential of the Landsat 7
ETM+ to detect change within a reef environment. Images were acquired within a brief
time frame during which there was no major disturbance to reefs. The lack of
disturbance permitted researchers to assess the stability of the images through time and
to estimate biases that may be associated with change detection strategies. Having
assumed that areas are large enough to provide an unmixed signal, the authors
suggested that there is difficulty in detecting changes from one object to another even
when the acquired images have identical specifications. The reason for this is that any
occurring change must be visible and cover a large enough areal extent to saturate
pixels. The effects of a hurricane, for example, would be easily identified since such a
storm would destroy living structures and transform a heterogeneous environment into
a homogeneous platform. Therefore, depending on the physical or biological shift
within a coral reef ecosystem–disturbance, phase, or strategy–spectral differences may
or may not be dissimilar enough to be detected. Un-mixing techniques are cited as a
potential solution because they would allow observers to detect intra-pixel changes, but
may be realistically applicable to no more than three classes at a time–sand,
background, and foreground, for instance.
Several change-detection techniques applied to images of coral reef environments
are derivatives of land or ice-based techniques (Lunetta and Elvidge, 1999). The Getis
statistic, for example, can be used to examine the change in reef homogeneity. This
measure considers the value of the reflectance within a single pixel and the relationship
between that pixel and the surrounding reference pixels. It has been hypothesized
that a healthy coral reef ecosystem will be heterogeneous and display negative
autocorrelation, while a disturbed reef similar in bottom type over a large area will be
spatially homogeneous and display a positive autocorrelation (LeDrew et al., 2004).
Two other measures available to researchers are principal components analysis
(Mas, 1999) and the Mahalanobis distance classification (Palandro et al., 2001), which
are regularly used prior to comparing images to detect change. It is generally accepted
that there are four aspects of change detection that are important when monitoring
natural resources: (1) detecting that a change has occurred; (2) identifying the nature of
the change; (3) measuring the areal extent of the change; and (4) assessing the spatial
pattern of the change (Klemas, 2001).
266
Newman, LeDrew and Lim