leaf area index of seagrasses (Dierssen et al., 2003). Supervised and unsupervised
classification approaches have also been successful in mapping benthos at various
spatial and temporal scales. Landsat TM and ETM+ have proved valuable for mapping
reef characteristics, including morphological and ecological zonation and cover types
(Neil et al., 2000) and for mapping change in coral, sand and algae cover (Palandro
et al., 2001).
Historically, interpreted aerial photography has been used for fine scale mapping of
change in coral reef communities, with this being supplemented in recent years by the
advent of higher spatial resolution satellite sensors such as IKONOS and QuickBird
(Palandro et al., 2003a). Mapping using these sensors can be quite accurate, with
accuracies of 89% reported in a study to map sand, coral reef and seagrass features
(Maeder et al., 2002). However, substrate mapping, particularly in coral reef
ecosystems, is complicated by geometry and scale of reef feature variation, especially
in relation to the vertical orientation and location of photosynthetic and productive
components (Phinn et al., 2000c), and thus affects the spectral resolution requirements
of sensor systems that can be used (Hochberg and Atkinson, 2003).
5.2 IMAGE PRE-PROCESSING REQUIREMENTS FOR CHANGE AND TREND
DETECTION IN COASTAL ECOSYSTEMS
Change and trend detection in coastal ecosystems requires rigorous image preprocessing to ensure that the variable of interest is detected with sufficient signal to
noise ratio. At a minimum, multi-temporal analysis requires sub-pixel precision
georeferencing, atmospheric correction, multi-date normalization, and ground-truthing
for accuracy assessment (Palandro et al., 2001). These processes need to be explicitly
considered within the framework used to select the remotely sensed data for the specific
monitoring requirement (Phinn, 1998b).
There has been little research on the effects of image mis-registration in aquatic
environments on change detection accuracy, although research in terrestrial regions has
found that significant and serious classification errors can be induced by a misregistration of only one pixel (Townshend et al., 1992; Phinn and Rowland, 2001).
These registration errors will become increasingly significant in the move towards
higher spatial resolution imagery.
Correction of atmospheric effects is dependant on the analytical methods used in
the change analysis. In many cases involving classification and change detection,
atmospheric correction is unnecessary, as long as the training data and the data to be
classified are in the same relative scale. Atmospheric correction is often unnecessary
when using atmospherically resistant indices developed for the application of interest
(Jensen, 1996b). Often atmospheric correction alone will not be adequate in images of
aquatic environments due to whitecaps and/or sun glint, with the corrected images
requiring additional empirical adjustment. Therefore, there is often no substantial
benefit in performing an atmospheric correction compared to an empirical correction
alone (Collins and Woodcock, 1996; Andréfouët et al., 2001).
When the purpose of processing is to derive change using semi-analytical
modelling, corrections to a common radiometric scale are essential. (Song et al., 2001).
Multi-date normalisation is used to minimise radiometric differences among images
caused by changes in acquisition conditions, and require the use of reference and
subject image pairs along with selected sample points. Normalisation methods include
image regression, pseudo-invariant features, histogram matching, and radiometric
235
Integrated Information Acquisition and Management
classification approaches have also been successful in mapping benthos at various
spatial and temporal scales. Landsat TM and ETM+ have proved valuable for mapping
reef characteristics, including morphological and ecological zonation and cover types
(Neil et al., 2000) and for mapping change in coral, sand and algae cover (Palandro
et al., 2001).
Historically, interpreted aerial photography has been used for fine scale mapping of
change in coral reef communities, with this being supplemented in recent years by the
advent of higher spatial resolution satellite sensors such as IKONOS and QuickBird
(Palandro et al., 2003a). Mapping using these sensors can be quite accurate, with
accuracies of 89% reported in a study to map sand, coral reef and seagrass features
(Maeder et al., 2002). However, substrate mapping, particularly in coral reef
ecosystems, is complicated by geometry and scale of reef feature variation, especially
in relation to the vertical orientation and location of photosynthetic and productive
components (Phinn et al., 2000c), and thus affects the spectral resolution requirements
of sensor systems that can be used (Hochberg and Atkinson, 2003).
5.2 IMAGE PRE-PROCESSING REQUIREMENTS FOR CHANGE AND TREND
DETECTION IN COASTAL ECOSYSTEMS
Change and trend detection in coastal ecosystems requires rigorous image preprocessing to ensure that the variable of interest is detected with sufficient signal to
noise ratio. At a minimum, multi-temporal analysis requires sub-pixel precision
georeferencing, atmospheric correction, multi-date normalization, and ground-truthing
for accuracy assessment (Palandro et al., 2001). These processes need to be explicitly
considered within the framework used to select the remotely sensed data for the specific
monitoring requirement (Phinn, 1998b).
There has been little research on the effects of image mis-registration in aquatic
environments on change detection accuracy, although research in terrestrial regions has
found that significant and serious classification errors can be induced by a misregistration of only one pixel (Townshend et al., 1992; Phinn and Rowland, 2001).
These registration errors will become increasingly significant in the move towards
higher spatial resolution imagery.
Correction of atmospheric effects is dependant on the analytical methods used in
the change analysis. In many cases involving classification and change detection,
atmospheric correction is unnecessary, as long as the training data and the data to be
classified are in the same relative scale. Atmospheric correction is often unnecessary
when using atmospherically resistant indices developed for the application of interest
(Jensen, 1996b). Often atmospheric correction alone will not be adequate in images of
aquatic environments due to whitecaps and/or sun glint, with the corrected images
requiring additional empirical adjustment. Therefore, there is often no substantial
benefit in performing an atmospheric correction compared to an empirical correction
alone (Collins and Woodcock, 1996; Andréfouët et al., 2001).
When the purpose of processing is to derive change using semi-analytical
modelling, corrections to a common radiometric scale are essential. (Song et al., 2001).
Multi-date normalisation is used to minimise radiometric differences among images
caused by changes in acquisition conditions, and require the use of reference and
subject image pairs along with selected sample points. Normalisation methods include
image regression, pseudo-invariant features, histogram matching, and radiometric
235
Integrated Information Acquisition and Management