10 Remote Sensing of Surface Water
221
reconfigured. Such changes have effects on water quality, transportation, fishing and
vegetation distribution (Mouchot et aI., 1991). The brightness values of Landsat band
5 (mid-infrared) were initially set to a threshold to separate water from land features
and morphological analysis was then used to defme further the shapes of the water
bodies and the channel connections. In the same area, Roberts et aI. (1994) used
airborne multi-spectral video imagery to provide data for understanding the frequency, timing and duration of flooding events in the more than 25,000 lakes of the
delta during the annual ice breakup. In the study transect, the flooding times of 3 ,216
lakes were determined from lake areas.
Pietroniro and Prowse (1997) used a time series of Landsat MSS bands 4-7, Landsat TM bands 1-7 and SPOT panchromatic data for the Peace-Athabasca Delta (PAD)
in northern Alberta, Canada to document changes in areal water extent over a 15 year
period within one of the world's largest freshwater deltas. This area is particularly
well suited to remote sensing studies because of its remoteness, size and lack of
accessibility. This dynamic ecosystem is affected by seasonal, annual and longer-term
changes in water levels and extents that are rarely documented. In their study, a time
series of satellite images was geometrically corrected to a UTM projection and then
classified as water/no water using a parallelepiped scheme with training areas of
known water bodies. The time-series of images highlights the extent of surface water
change that occurred in this ecologically sensitive region during a significant drying
trend (Colour Plate 10.A). This trend has been attributed to a combination of anthropogenic forcing through the construction of an upstream reservoir and climatic
factors, both of which are believed to have affected the frequency of dynamic ice-jam
flooding (Prowse and Lalonde, 1996). To define hydrologically connected areas, an
algorithm was applied to find all pixels satisfying a four-point connectivity. The
images were then converted to vector polygons and overlain on the original scenes.
Colour Plate 10.B shows an example of the resulting image. The vectors could then
be compared to existing lake level estimates and a simple relationship between lake
areal extent and lake elevation determined. Both these applications provided vital
hydrologic information that was otherwise unavailable for this remote site.
In a study of the Great Lakes of the Mackenzie Basin, Birkett and Kite (1997) used
NOAA A VHRR images in the visible, near- and thermal-infrared bands. The data
were obtained from the NOAA Satellite Active Archive System (SAA) using 2048
pixel-wide swath width LAC images. It was found that channel 2 (near-infrared)
showed maximum contrast between water and land, and channels I (visible) and 2
were used to detect ice. Due to their location and size, the Mackenzie lakes are often
cloud-covered; for example, during the month of August 1994, only six images out
of fifty were cloud-free with another fourteen having partial cloud cover. In winter,
the situation was even worse; Great Slave Lake had no cloud-free images from
October to December 1994. Estimates oflake surface area were made using a local
iso-luminance contour (LIC) routine (see next paragraph). This semi-automatic
routine had been previously used for small lakes (Harris, 1994) but it had difficulty
dealing with the complex coastlines of the Mackenzie lakes and often failed to close
the lake perimeter. Instead ofLIC, a manual system of identifying the lake perimeter
was used. Lake surface areas in pixels were converted to square kilometres using
221
reconfigured. Such changes have effects on water quality, transportation, fishing and
vegetation distribution (Mouchot et aI., 1991). The brightness values of Landsat band
5 (mid-infrared) were initially set to a threshold to separate water from land features
and morphological analysis was then used to defme further the shapes of the water
bodies and the channel connections. In the same area, Roberts et aI. (1994) used
airborne multi-spectral video imagery to provide data for understanding the frequency, timing and duration of flooding events in the more than 25,000 lakes of the
delta during the annual ice breakup. In the study transect, the flooding times of 3 ,216
lakes were determined from lake areas.
Pietroniro and Prowse (1997) used a time series of Landsat MSS bands 4-7, Landsat TM bands 1-7 and SPOT panchromatic data for the Peace-Athabasca Delta (PAD)
in northern Alberta, Canada to document changes in areal water extent over a 15 year
period within one of the world's largest freshwater deltas. This area is particularly
well suited to remote sensing studies because of its remoteness, size and lack of
accessibility. This dynamic ecosystem is affected by seasonal, annual and longer-term
changes in water levels and extents that are rarely documented. In their study, a time
series of satellite images was geometrically corrected to a UTM projection and then
classified as water/no water using a parallelepiped scheme with training areas of
known water bodies. The time-series of images highlights the extent of surface water
change that occurred in this ecologically sensitive region during a significant drying
trend (Colour Plate 10.A). This trend has been attributed to a combination of anthropogenic forcing through the construction of an upstream reservoir and climatic
factors, both of which are believed to have affected the frequency of dynamic ice-jam
flooding (Prowse and Lalonde, 1996). To define hydrologically connected areas, an
algorithm was applied to find all pixels satisfying a four-point connectivity. The
images were then converted to vector polygons and overlain on the original scenes.
Colour Plate 10.B shows an example of the resulting image. The vectors could then
be compared to existing lake level estimates and a simple relationship between lake
areal extent and lake elevation determined. Both these applications provided vital
hydrologic information that was otherwise unavailable for this remote site.
In a study of the Great Lakes of the Mackenzie Basin, Birkett and Kite (1997) used
NOAA A VHRR images in the visible, near- and thermal-infrared bands. The data
were obtained from the NOAA Satellite Active Archive System (SAA) using 2048
pixel-wide swath width LAC images. It was found that channel 2 (near-infrared)
showed maximum contrast between water and land, and channels I (visible) and 2
were used to detect ice. Due to their location and size, the Mackenzie lakes are often
cloud-covered; for example, during the month of August 1994, only six images out
of fifty were cloud-free with another fourteen having partial cloud cover. In winter,
the situation was even worse; Great Slave Lake had no cloud-free images from
October to December 1994. Estimates oflake surface area were made using a local
iso-luminance contour (LIC) routine (see next paragraph). This semi-automatic
routine had been previously used for small lakes (Harris, 1994) but it had difficulty
dealing with the complex coastlines of the Mackenzie lakes and often failed to close
the lake perimeter. Instead ofLIC, a manual system of identifying the lake perimeter
was used. Lake surface areas in pixels were converted to square kilometres using