224
G. Kite and A. Pietroniro
three distinct classes: open water, flooded willow, and non-flooded areas. SupeIVised
classification results of the original images were not acceptable due to the large local
variance introduced by speckle. To improve the classification results they utilized
image tone and texture. Initially, speckle was minimized with a 7x7 Gamma MAP
(Maximum A Posteriori) filter applied to each image. The .resulting classifications
were significantly improved with a Kappa coefficient of over 90%. Textural information is further added to the classification, via a coefficient of variation scene. With
this additional information Kappa was slightly reduced, however, visual inspection
showed that some channel and lake edges were more accurately classified as flooded
willows.
There have been some studies exploring using multi-temporal radar imagery for
wetland mapping and monitoring. Kasischke and Bourgeau-Chavez (1997) evaluated
the utility of SAR imagery for monitoring the hydrology of wetland ecosystems using
two-dates ERS-l imagery. Wang et aI., 1998 also investigated the use of multi date
ERS-l data for wetland classification in Southern Ontario, Canada and used scenes
of ERS-l imagery from nine different months. The results are compared with Landsat
TM imagery of the same area. In their case, accuracy was measured by comparing the
results derived from digital image classification with the ground truth obtained from
airphoto interpretation, map data, and field investigation. They found that although
a cloud free Landsat TM performed better for wetland identification, an accuracy of
over 80% could be achieved using more than 3-date ERS-l data.
10.5 Lake Levels
The advent of satellites with onboard radar altimeters such as Geosat, Seasat, ERS-1
and the recent U.S.AlFrance TOPEX/Poseidon has made possible the measurement
of lake and (large) river levels to centimetre accuracy with frequent repeat cycles.
Remote measurement of lake levels from space using radar altimetry has been demonstrated experimentally using data from the Geosat and Seasat missions (Birkett,
1994). Birkett (1994) has described the use of Geosat data to measure levels of Lake
Ontario and the Caspian Sea to within 10cm accuracy. These satellites are no longer
in seIVice but the ERS-l and TOPEX/Poseidon satellites are operational and carry
advanced radar altimeters capable of accurately measuring surface height changes,
with the most advanced being the TOPEX/Poseidon altimeter. The TOPEX/Poseidon
mission is designed to obtain a global view of earth's ocean topography with sufficient
accuracy to improve models designed to forecast global ocean circulation. The
payload consists of a dual frequency radar operating at the C-band (5.3 GHz) and Kuband (13.6 GHz). The Ku-band secondary channel allows for correction ofpropagation delays in the ionosphere, reducing a significant error source in the measurement.
The satellite sends out radar pulses and measures the time of the return reflection from
the nearest surface directly below the satellite (Fig. lOA). A 10 km footprint contributes to the measured return, and the accuracy of an individual height measurement is
estimated to be about 13 cm. Using the onboard satellite navigation and knowledge
of the orbit, the average ocean height is then estimated (Zieger et aI. 1991). Sources
of error from altimeter measurements can be attributed to instrument errors, atmos-
G. Kite and A. Pietroniro
three distinct classes: open water, flooded willow, and non-flooded areas. SupeIVised
classification results of the original images were not acceptable due to the large local
variance introduced by speckle. To improve the classification results they utilized
image tone and texture. Initially, speckle was minimized with a 7x7 Gamma MAP
(Maximum A Posteriori) filter applied to each image. The .resulting classifications
were significantly improved with a Kappa coefficient of over 90%. Textural information is further added to the classification, via a coefficient of variation scene. With
this additional information Kappa was slightly reduced, however, visual inspection
showed that some channel and lake edges were more accurately classified as flooded
willows.
There have been some studies exploring using multi-temporal radar imagery for
wetland mapping and monitoring. Kasischke and Bourgeau-Chavez (1997) evaluated
the utility of SAR imagery for monitoring the hydrology of wetland ecosystems using
two-dates ERS-l imagery. Wang et aI., 1998 also investigated the use of multi date
ERS-l data for wetland classification in Southern Ontario, Canada and used scenes
of ERS-l imagery from nine different months. The results are compared with Landsat
TM imagery of the same area. In their case, accuracy was measured by comparing the
results derived from digital image classification with the ground truth obtained from
airphoto interpretation, map data, and field investigation. They found that although
a cloud free Landsat TM performed better for wetland identification, an accuracy of
over 80% could be achieved using more than 3-date ERS-l data.
10.5 Lake Levels
The advent of satellites with onboard radar altimeters such as Geosat, Seasat, ERS-1
and the recent U.S.AlFrance TOPEX/Poseidon has made possible the measurement
of lake and (large) river levels to centimetre accuracy with frequent repeat cycles.
Remote measurement of lake levels from space using radar altimetry has been demonstrated experimentally using data from the Geosat and Seasat missions (Birkett,
1994). Birkett (1994) has described the use of Geosat data to measure levels of Lake
Ontario and the Caspian Sea to within 10cm accuracy. These satellites are no longer
in seIVice but the ERS-l and TOPEX/Poseidon satellites are operational and carry
advanced radar altimeters capable of accurately measuring surface height changes,
with the most advanced being the TOPEX/Poseidon altimeter. The TOPEX/Poseidon
mission is designed to obtain a global view of earth's ocean topography with sufficient
accuracy to improve models designed to forecast global ocean circulation. The
payload consists of a dual frequency radar operating at the C-band (5.3 GHz) and Kuband (13.6 GHz). The Ku-band secondary channel allows for correction ofpropagation delays in the ionosphere, reducing a significant error source in the measurement.
The satellite sends out radar pulses and measures the time of the return reflection from
the nearest surface directly below the satellite (Fig. lOA). A 10 km footprint contributes to the measured return, and the accuracy of an individual height measurement is
estimated to be about 13 cm. Using the onboard satellite navigation and knowledge
of the orbit, the average ocean height is then estimated (Zieger et aI. 1991). Sources
of error from altimeter measurements can be attributed to instrument errors, atmos-