resolution aerial photography and satellite image data. These applications have
focused on mapping benthic communities and coral reef benthic cover types, such
as live and dead coral, where high levels of detail and local context are available to
identify specific reef features (Cuevas-Jimenez and Ardisson 2002; Knudby et al.
2007; Scopélitis et al. 2009). In some cases, regionally and globally applicable
mapping programs using broad levels of detail have used manual digitizing to
produce reef maps, such as the Millennium Coral Reef Mapping Project, which
utilizes the global archive of Landsat Thematic Mapper and Landsat Enhanced
Thematic Mapper data with 30 9 30 m pixels (Andréfouët et al. 2005b; Andréfouët 2008).
More recent developments have seen image processing systems provide semiautomatic processes that replicate manual interpretation, in the form of geographic
object-based image analysis (GEOBIA). These approaches enable hierarchical
segmentation of images into pre-set features or objects at specific spatial scales
(e.g., reef/non-reef, geomorphic zones and benthic community zones and patches)
(Benfield et al. 2007). After segmentation the image objects or features are then
labeled manually or automatically.
Image classification is the most common algorithmic approach to producing
thematic maps from multispectral and hyperspectral data sets. Image classification
is used to assign a pre-defined thematic class label to each pixel in an image. The
classification algorithms are based on two assumptions: (1) each image pixel
contains only one type of coral reef benthic feature (i.e., that a pixel is smaller than
the feature to be mapped); and (2) all image pixels containing that type of coral
reef feature have a similar spectral reflectance signature. Since hyperspectral
images produce spectral signatures with higher degree of detail and precision than
multispectral and photographic images (e.g., Fig. 1.5), classification algorithms
using hyperspectral data can discriminate more coral reef benthic cover types.
Increased thematic detail can also be achieved by adding contextual information
into the process, including measures such as image texture or roughness and other
forms of image and spatial information. Image classification routines can further
include post-classification manual editing to increase the level of thematic detail
and accuracy of coral reef maps.
The final stage in the mapping process should always be some form of validation, where the output coral reef map is compared to a suitable form of reference
data, either from field survey or other spatial data, so that the overall and individual class mapping accuracies are known (Andrefouet 2008; Mumby et al. 1998;
Roelfsema and Phinn 2010).
1.3.4 Biophysical or Continuous Variable Mapping
Production of maps quantifying biophysical properties or processes on coral reefs
and their surrounding environments can only be done from fully corrected airborne
or satellite images. This type of processing applies one or more equations to each
1 Visible and Infrared Overview
21
focused on mapping benthic communities and coral reef benthic cover types, such
as live and dead coral, where high levels of detail and local context are available to
identify specific reef features (Cuevas-Jimenez and Ardisson 2002; Knudby et al.
2007; Scopélitis et al. 2009). In some cases, regionally and globally applicable
mapping programs using broad levels of detail have used manual digitizing to
produce reef maps, such as the Millennium Coral Reef Mapping Project, which
utilizes the global archive of Landsat Thematic Mapper and Landsat Enhanced
Thematic Mapper data with 30 9 30 m pixels (Andréfouët et al. 2005b; Andréfouët 2008).
More recent developments have seen image processing systems provide semiautomatic processes that replicate manual interpretation, in the form of geographic
object-based image analysis (GEOBIA). These approaches enable hierarchical
segmentation of images into pre-set features or objects at specific spatial scales
(e.g., reef/non-reef, geomorphic zones and benthic community zones and patches)
(Benfield et al. 2007). After segmentation the image objects or features are then
labeled manually or automatically.
Image classification is the most common algorithmic approach to producing
thematic maps from multispectral and hyperspectral data sets. Image classification
is used to assign a pre-defined thematic class label to each pixel in an image. The
classification algorithms are based on two assumptions: (1) each image pixel
contains only one type of coral reef benthic feature (i.e., that a pixel is smaller than
the feature to be mapped); and (2) all image pixels containing that type of coral
reef feature have a similar spectral reflectance signature. Since hyperspectral
images produce spectral signatures with higher degree of detail and precision than
multispectral and photographic images (e.g., Fig. 1.5), classification algorithms
using hyperspectral data can discriminate more coral reef benthic cover types.
Increased thematic detail can also be achieved by adding contextual information
into the process, including measures such as image texture or roughness and other
forms of image and spatial information. Image classification routines can further
include post-classification manual editing to increase the level of thematic detail
and accuracy of coral reef maps.
The final stage in the mapping process should always be some form of validation, where the output coral reef map is compared to a suitable form of reference
data, either from field survey or other spatial data, so that the overall and individual class mapping accuracies are known (Andrefouet 2008; Mumby et al. 1998;
Roelfsema and Phinn 2010).
1.3.4 Biophysical or Continuous Variable Mapping
Production of maps quantifying biophysical properties or processes on coral reefs
and their surrounding environments can only be done from fully corrected airborne
or satellite images. This type of processing applies one or more equations to each
1 Visible and Infrared Overview
21