acquisitions were scheduled on a monthly or fortnight basis. The satellite images used
for the classification of L. majuscula blooms were subsets of the map-oriented, dark
pixel corrected Landsat 7 ETM+ scenes (path 89, row 79) recorded at 9:45 am (AEST)
on dates when field data collection and cloud free imagery coincided.
Mapping of L. majuscula patches followed a multistage process which ensured that
only those areas in which L. majuscula could be reliably mapped were extracted from
the image and used for mapping. The blue, green and red bands were selected for use in
image classification because of the comparatively limited depth-related light
attenuation effects and maximum signal from submerged features. Variations in light
attenuation are particularly acute in this environment, due to the mixture of oceanic and
coastal/estuarine water bodies (Morel, 1977). ERDAS Imagine software was used to
process the imagery. Once geometrically corrected, the subset was corrected for
additive path radiance by applying dark pixel subtraction (Jensen, 1996a).
Image classification to map L. majuscula. The field data collected by Marine Park
authorities together with substrate coverage maps produced from previous studies (e.g.
(Dennison, 1998)) were used to apply a supervised classification to map L. majuscula
coverage. The substrate information used to select training pixels for classifying the
Landsat TM or ETM+ image data were collected as close to the field survey date as
possible. Statistics on substrate reflectance values were extracted from the image for
field sites known to correspond to different L. majuscula density levels. The masked
image of Eastern Banks was then subject to a “minimum distance to means” clustering
routine to group pixels with similar reflectance values into three classes of varying
L. majuscula cover and all other substrate. The final map for presentation to QPWS and
for inclusion in their GIS, was made in Arcview 3.2, presenting the classification image
results overlaid on the original image.
A pseudo “error matrix” was used to assess the accuracy of the classification and
quantify the level of agreement between the classes identified from the image
classification and the field data. The “pseudo” label, if applied as a true reference set,
would have consisted of independently selected sites where L. majuscula cover had
been measured and not used to train the image classification process. Hence, the error
matrix is only a measure of how well the classification correctly identified the training
data, not the whole study area.
6.2.5 Maintaining the Mapping Program
Currently the field component of this program is implemented on a regular basis,
coinciding with Landsat 7 ETM+ over flights. The remote sensing component consists
of an L. majuscula bloom contingency plan, which is initiated when the results of the
field monitoring show medium to high levels of Lyngbya. At this time, a cloud free
image of the study area will be purchased if available, and a classification using QWPS
and community field data will be conducted. The results (in map and report format) of
field and/or remote sensing monitoring are present on a website to be accessible for the
community (Figure 3). The data itself are still analysed on a yearly basis and will be
used as one of the parameters for a report card presenting the health of the local coastal
areas. The presence, size and duration of a bloom are regarded as key indicators of
coastal ecosystem health in Moreton Bay.
243
Integrated Information Acquisition and Management
for the classification of L. majuscula blooms were subsets of the map-oriented, dark
pixel corrected Landsat 7 ETM+ scenes (path 89, row 79) recorded at 9:45 am (AEST)
on dates when field data collection and cloud free imagery coincided.
Mapping of L. majuscula patches followed a multistage process which ensured that
only those areas in which L. majuscula could be reliably mapped were extracted from
the image and used for mapping. The blue, green and red bands were selected for use in
image classification because of the comparatively limited depth-related light
attenuation effects and maximum signal from submerged features. Variations in light
attenuation are particularly acute in this environment, due to the mixture of oceanic and
coastal/estuarine water bodies (Morel, 1977). ERDAS Imagine software was used to
process the imagery. Once geometrically corrected, the subset was corrected for
additive path radiance by applying dark pixel subtraction (Jensen, 1996a).
Image classification to map L. majuscula. The field data collected by Marine Park
authorities together with substrate coverage maps produced from previous studies (e.g.
(Dennison, 1998)) were used to apply a supervised classification to map L. majuscula
coverage. The substrate information used to select training pixels for classifying the
Landsat TM or ETM+ image data were collected as close to the field survey date as
possible. Statistics on substrate reflectance values were extracted from the image for
field sites known to correspond to different L. majuscula density levels. The masked
image of Eastern Banks was then subject to a “minimum distance to means” clustering
routine to group pixels with similar reflectance values into three classes of varying
L. majuscula cover and all other substrate. The final map for presentation to QPWS and
for inclusion in their GIS, was made in Arcview 3.2, presenting the classification image
results overlaid on the original image.
A pseudo “error matrix” was used to assess the accuracy of the classification and
quantify the level of agreement between the classes identified from the image
classification and the field data. The “pseudo” label, if applied as a true reference set,
would have consisted of independently selected sites where L. majuscula cover had
been measured and not used to train the image classification process. Hence, the error
matrix is only a measure of how well the classification correctly identified the training
data, not the whole study area.
6.2.5 Maintaining the Mapping Program
Currently the field component of this program is implemented on a regular basis,
coinciding with Landsat 7 ETM+ over flights. The remote sensing component consists
of an L. majuscula bloom contingency plan, which is initiated when the results of the
field monitoring show medium to high levels of Lyngbya. At this time, a cloud free
image of the study area will be purchased if available, and a classification using QWPS
and community field data will be conducted. The results (in map and report format) of
field and/or remote sensing monitoring are present on a website to be accessible for the
community (Figure 3). The data itself are still analysed on a yearly basis and will be
used as one of the parameters for a report card presenting the health of the local coastal
areas. The presence, size and duration of a bloom are regarded as key indicators of
coastal ecosystem health in Moreton Bay.
243
Integrated Information Acquisition and Management