available, an effective method is to perform calibration and normalization of the
images in order to use the training of recent in situ data to assist with classification
for all the images (Palandro et al. 2008). The pre-classification approach involves
the analysis of change in the actual spectral signatures or spectral indices
(Matsunaga et al. 2000; Dustan et al. 2001; Yamano and Tamura 2004). Most
spectral change identification techniques require image calibration or normalization to detect spectral differences between pixels in multi-temporal images. For
example, normalization using the depth invariant index (Lyzenga 1978) is often
insensitive to variations in overall intensity since the index is based on logtransformed values (Fig. 3.2). Direct comparison of spectral values can be
achieved by normalizing the effects of changes in the atmosphere, incident light,
water depth (tide) and sensor response, and can be accomplished using models
and/or known values for pixels of shallow sand, deep-water or the object of
interest (e.g., coral) (Yamano and Tamura 2004).
In order to detect long-term changes ([10 years), it is preferable to select
images taken in the same season, because macroalgae, which is often difficult to be
distinguished from corals due to similar reflectance characteristics, exhibits seasonal changes in distribution and abundance. Additionally, because pixel-based
change detection has high sensitivity to spatial mis-registration of the pixels, the
RMS errors in geometric correction should be smaller than 0.5 pixels.
3.3 Example Applications
For effective management of coral reefs, Phinn et al. (2006) suggests analysis
should follow a progression of knowledge: mapping, monitoring and modeling.
Mapping provides baseline surveys or inventories. Monitoring can be achieved
through comparison of baseline maps against updated information, enabling
changes to be mapped and measured. Modeling includes data integration to make
statistical or physics-based links between environmental variables and coral reef
processes, enabling prediction of system response to certain environmental conditions. In this section, several studies based on this approach are discussed. Table 3.2
presents a summary of other applications relevant to coral reef management.
3.3.1 Reef Mapping
Maps produced using multispectral imagery include: geologic, geomorphologic
and sedimentary features (Rankey 2002; Naseer and Hatcher 2004; Purkis et al.
2010); and ecological habitats (Mumby et al. 1997; Andréfouët et al. 2003). These
maps not only serve as base maps to examine reef structure and resource inventory
but also allow estimation of ecological functions and biodiversity. An overview of
two different categories of reef mapping applications are presented below, reefscale mapping and regional to global-scale mapping.
3 Multispectral Applications
63
images in order to use the training of recent in situ data to assist with classification
for all the images (Palandro et al. 2008). The pre-classification approach involves
the analysis of change in the actual spectral signatures or spectral indices
(Matsunaga et al. 2000; Dustan et al. 2001; Yamano and Tamura 2004). Most
spectral change identification techniques require image calibration or normalization to detect spectral differences between pixels in multi-temporal images. For
example, normalization using the depth invariant index (Lyzenga 1978) is often
insensitive to variations in overall intensity since the index is based on logtransformed values (Fig. 3.2). Direct comparison of spectral values can be
achieved by normalizing the effects of changes in the atmosphere, incident light,
water depth (tide) and sensor response, and can be accomplished using models
and/or known values for pixels of shallow sand, deep-water or the object of
interest (e.g., coral) (Yamano and Tamura 2004).
In order to detect long-term changes ([10 years), it is preferable to select
images taken in the same season, because macroalgae, which is often difficult to be
distinguished from corals due to similar reflectance characteristics, exhibits seasonal changes in distribution and abundance. Additionally, because pixel-based
change detection has high sensitivity to spatial mis-registration of the pixels, the
RMS errors in geometric correction should be smaller than 0.5 pixels.
3.3 Example Applications
For effective management of coral reefs, Phinn et al. (2006) suggests analysis
should follow a progression of knowledge: mapping, monitoring and modeling.
Mapping provides baseline surveys or inventories. Monitoring can be achieved
through comparison of baseline maps against updated information, enabling
changes to be mapped and measured. Modeling includes data integration to make
statistical or physics-based links between environmental variables and coral reef
processes, enabling prediction of system response to certain environmental conditions. In this section, several studies based on this approach are discussed. Table 3.2
presents a summary of other applications relevant to coral reef management.
3.3.1 Reef Mapping
Maps produced using multispectral imagery include: geologic, geomorphologic
and sedimentary features (Rankey 2002; Naseer and Hatcher 2004; Purkis et al.
2010); and ecological habitats (Mumby et al. 1997; Andréfouët et al. 2003). These
maps not only serve as base maps to examine reef structure and resource inventory
but also allow estimation of ecological functions and biodiversity. An overview of
two different categories of reef mapping applications are presented below, reefscale mapping and regional to global-scale mapping.
3 Multispectral Applications
63