3.2 Multispectral Analysis and Classification
3.2.1 Types of Analysis
Remote sensing of coral reefs can be divided into two categories: direct and
indirect (Andréfouët and Riegl 2004; Andréfouët et al. 2005; Table 3.2). Direct
remote sensing is where the coral reef itself is the target of remote sensing, while
indirect remote sensing refers to studies that focus on the oceanic and atmospheric
environment surrounding the reef. Targets that can be measured by multispectral
remote sensing are described in Mumby et al. (2004).
Direct remote sensing. The objectives of direct remote sensing are the biotic and
morphologic features (Table 3.2), where benthic features such as habitats may be
of primary interest to management. Multispectral satellite sensors with blue bands
can differentiate three to six habitat classes with reasonable overall accuracy
(60–75 %, Table 3.1). Factors affecting the classification accuracy are the spectral
band arrangement of the sensor, spatial resolution of the sensor, and characteristics
of the benthic features to be classified. The broad width of these sensor’s spectral
bands limits accurate discrimination of more specific benthic features, since many
biotic features (coral, seagrass and macroalgae) have similar reflectance characteristics, and it is suggested that hyperspectral sensors can better differentiate such
features (Hochberg et al. 2003). Instruments lacking blue bands, which penetrate
water to see the bottom features, results in significant decrease in classification
accuracy (Table 3.1). In addition, because of the high heterogeneity and spatial
scale of reef features, spectral signatures in one pixel are often mixed, meaning
that studies using medium spatial resolution sensors typically produce lower
classification accuracy (Hochberg and Atkinson 2003; Fig. 3.1). Increasing the
number of habitat categories also results in decreased classification accuracy
(Mumby et al. 1997; Andréfouët et al. 2003). A compilation of results suggests a
predictive relationship between overall accuracy versus number of classes
(r
2 = 0.63) for Landsat ETM+ and for IKONOS (r
2 = 0.82) (Andréfouët et al.
2003).
Bleached corals show significantly higher reflectance than healthy corals in all
wavelength regions (Holden and LeDrew 1998; Hochberg et al. 2003; Yamano and
Tamura 2004), and thus occurrence of bleaching is expected to be detectable using
multispectral remote sensing. In order to avoid spectral mixing in one pixel,
Andréfouët et al. (2002a) suggested the use of high (*2 m) spatial resolution
sensors to detect bleaching in one image. No studies using multispectral satellite
remote sensing have succeeded in detecting coral mass spawning yet, though
observation by airplanes and by Synthetic Aperture Radar (SAR) produced
potentially promising results (Willis and Oliver 1990; Jones et al. 2006).
Morphologic features provide the basis for examining habitat distributions
within and between coral reefs. The location of shallow reef areas is easily
detected by multispectral satellite sensors, and many coral reefs exhibit a common,
54
H. Yamano
3.2.1 Types of Analysis
Remote sensing of coral reefs can be divided into two categories: direct and
indirect (Andréfouët and Riegl 2004; Andréfouët et al. 2005; Table 3.2). Direct
remote sensing is where the coral reef itself is the target of remote sensing, while
indirect remote sensing refers to studies that focus on the oceanic and atmospheric
environment surrounding the reef. Targets that can be measured by multispectral
remote sensing are described in Mumby et al. (2004).
Direct remote sensing. The objectives of direct remote sensing are the biotic and
morphologic features (Table 3.2), where benthic features such as habitats may be
of primary interest to management. Multispectral satellite sensors with blue bands
can differentiate three to six habitat classes with reasonable overall accuracy
(60–75 %, Table 3.1). Factors affecting the classification accuracy are the spectral
band arrangement of the sensor, spatial resolution of the sensor, and characteristics
of the benthic features to be classified. The broad width of these sensor’s spectral
bands limits accurate discrimination of more specific benthic features, since many
biotic features (coral, seagrass and macroalgae) have similar reflectance characteristics, and it is suggested that hyperspectral sensors can better differentiate such
features (Hochberg et al. 2003). Instruments lacking blue bands, which penetrate
water to see the bottom features, results in significant decrease in classification
accuracy (Table 3.1). In addition, because of the high heterogeneity and spatial
scale of reef features, spectral signatures in one pixel are often mixed, meaning
that studies using medium spatial resolution sensors typically produce lower
classification accuracy (Hochberg and Atkinson 2003; Fig. 3.1). Increasing the
number of habitat categories also results in decreased classification accuracy
(Mumby et al. 1997; Andréfouët et al. 2003). A compilation of results suggests a
predictive relationship between overall accuracy versus number of classes
(r
2 = 0.63) for Landsat ETM+ and for IKONOS (r
2 = 0.82) (Andréfouët et al.
2003).
Bleached corals show significantly higher reflectance than healthy corals in all
wavelength regions (Holden and LeDrew 1998; Hochberg et al. 2003; Yamano and
Tamura 2004), and thus occurrence of bleaching is expected to be detectable using
multispectral remote sensing. In order to avoid spectral mixing in one pixel,
Andréfouët et al. (2002a) suggested the use of high (*2 m) spatial resolution
sensors to detect bleaching in one image. No studies using multispectral satellite
remote sensing have succeeded in detecting coral mass spawning yet, though
observation by airplanes and by Synthetic Aperture Radar (SAR) produced
potentially promising results (Willis and Oliver 1990; Jones et al. 2006).
Morphologic features provide the basis for examining habitat distributions
within and between coral reefs. The location of shallow reef areas is easily
detected by multispectral satellite sensors, and many coral reefs exhibit a common,
54
H. Yamano