imagery, 14 papers used high spatial resolution imagery, and 9 papers used aerial
photographs. For seagrass mapping (26 papers), 10 papers used aerial photographs,
7 papers used moderate spatial resolution imagery, and 3 papers used high spatial
resolution imagery.
The choice of sensor type was often based on what type of map product was
required. Some of the trade-offs, and relevant examples, related to this decision
include:
• Spectral resolution: Multispectral imagery such as Landsat has three bands in
visible range versus hyperspectral imagery such as CASI, which can have a
predetermined number of bands up to 512.
• Temporal resolution: Landsat imagery is acquired every 16 days over the last 3
decades, whereas high spatial resolution imagery such as Ikonos needs to be
specifically scheduled and has only been available since 2002.
• Spatial resolution: MODIS imagery is a resolution of 250 9 250 m per pixel,
whereas Ikonos is 4 9 4 m.
• Areal extent: Landsat imagery covers an area of 185 9 185 km, whereas
QuickBird only covers around 12 9 12 km.
The main calibration and validation data sources used in these studies were
primarily spot check or point data (34 papers), followed by transect data without
video (18 papers), video transects (15 papers), and other data sources such as local
knowledge (13 papers). From the coral reef habitat mapping papers (54 papers), 22
papers used spot check methods and 25 papers used video transects. For seagrass
mapping (26 papers), 12 papers used spot check and 8 used video transects. The
tradeoff for selecting the different types of calibration or validation data collection
techniques was most often based on the available means (e.g., people, equipment,
boats, remoteness) and the information required (e.g., live coral presence versus
species composition). In some instances, existing data that was collected for other
purposes was deemed suitable for use in a different calibration and validation
application (Roelfsema et al. 2006; Roelfsema and Phinn 2010).
The most common method for classifying habitats was supervised classification
(31 papers), followed by manual delineation (14 papers), spectral analysis and
radiative transfer modeling (9 papers), and unsupervised classification (7 papers).
The remaining 19 papers represent those that used various combinations of processing techniques, or techniques not frequently applied, such as regression
analysis (2 papers) and object based classification (3 papers). Methods chosen for
classification depended on the available software, equipment and knowledge and
skills of the producers. Unsupervised classification, for example, can be readily
achieved with GIS and remote sensing software packages (e.g., ARCGIS, ENVI,
IMAGINE); however, object based image analysis is mostly limited to costly
additional software that requires specialized training of the operator.
The Millennium Global Coral Reef Mapping Project (Andréfouët et al. 2005)
covered the largest extent of coral reefs, and the largest for seagrass was conducted
by Wabnitz et al. (2008), which mapped seagrass extent throughout the Caribbean.
14 Validation
385
photographs. For seagrass mapping (26 papers), 10 papers used aerial photographs,
7 papers used moderate spatial resolution imagery, and 3 papers used high spatial
resolution imagery.
The choice of sensor type was often based on what type of map product was
required. Some of the trade-offs, and relevant examples, related to this decision
include:
• Spectral resolution: Multispectral imagery such as Landsat has three bands in
visible range versus hyperspectral imagery such as CASI, which can have a
predetermined number of bands up to 512.
• Temporal resolution: Landsat imagery is acquired every 16 days over the last 3
decades, whereas high spatial resolution imagery such as Ikonos needs to be
specifically scheduled and has only been available since 2002.
• Spatial resolution: MODIS imagery is a resolution of 250 9 250 m per pixel,
whereas Ikonos is 4 9 4 m.
• Areal extent: Landsat imagery covers an area of 185 9 185 km, whereas
QuickBird only covers around 12 9 12 km.
The main calibration and validation data sources used in these studies were
primarily spot check or point data (34 papers), followed by transect data without
video (18 papers), video transects (15 papers), and other data sources such as local
knowledge (13 papers). From the coral reef habitat mapping papers (54 papers), 22
papers used spot check methods and 25 papers used video transects. For seagrass
mapping (26 papers), 12 papers used spot check and 8 used video transects. The
tradeoff for selecting the different types of calibration or validation data collection
techniques was most often based on the available means (e.g., people, equipment,
boats, remoteness) and the information required (e.g., live coral presence versus
species composition). In some instances, existing data that was collected for other
purposes was deemed suitable for use in a different calibration and validation
application (Roelfsema et al. 2006; Roelfsema and Phinn 2010).
The most common method for classifying habitats was supervised classification
(31 papers), followed by manual delineation (14 papers), spectral analysis and
radiative transfer modeling (9 papers), and unsupervised classification (7 papers).
The remaining 19 papers represent those that used various combinations of processing techniques, or techniques not frequently applied, such as regression
analysis (2 papers) and object based classification (3 papers). Methods chosen for
classification depended on the available software, equipment and knowledge and
skills of the producers. Unsupervised classification, for example, can be readily
achieved with GIS and remote sensing software packages (e.g., ARCGIS, ENVI,
IMAGINE); however, object based image analysis is mostly limited to costly
additional software that requires specialized training of the operator.
The Millennium Global Coral Reef Mapping Project (Andréfouët et al. 2005)
covered the largest extent of coral reefs, and the largest for seagrass was conducted
by Wabnitz et al. (2008), which mapped seagrass extent throughout the Caribbean.
14 Validation
385