7 Land-use and Catchment Characteristics
153
average the number of pixels from the same cover type, or by multi-look averaging
of the data, at the expense of loss of texture information.
An advantage of imaging radars is that they are insensitive to local weather conditions. The orbital SAR systems (ERS 1-2, JERS-l, Radarsat, SIR-CIX-SAR) provide multi-date data, which is valuable for monitoring the effects of cover on hydrology, particularly in cloudy weather types.
The backscatter signal of terrain, which is used to process an image, depends
on the wavelength, cover species related (probably morphologic) features, surface
roughness of the ground and moisture content, which leads to ambiguity of single
radar data. Closed canopies of more or less uniform height cause attenuation of the
signal, especially in the C and X band. The shorter wavelengths (C band and shorter)
display a strong relation between the backscatter and species dependent properties
and can be used to detect stress or diseases and the forest development (Hoekman,
1990). Natural forests with mixed species compositions may show a very limited dynamic range of radar backscatter. A high degree of canopy transparency was found
for the longer wavelength L-band, which is useful for differentiation of broad forest classes and for detecting ground cover conditions, or flooding below the canopy
(see also Figs. 10.8 and 1O.9). It may be noted that multi-spectral data pertains to
the canopy only.
An interesting specific application of radar was the detection of water below a
woody forest canopy by the L and P bands. Presence of water increases the backscatter by ground-trunk two-bounce scattering(Hess et aI., 1995}. However, in nonwoody wetlands, a decrease of the backscatter may occur because of the forward
specular scattering of the pulse signal. -
Many studies have used the maximum likelihood approaches or a decision tree
method, based on training sets containing ground truth or local knowledge. It was
found that single date, single frequency/polarization data has limitations, but multidate and multichannel SAR leads to enhanced classification (Kasischke et aI., 1997).
As is the case with multi-spectral data, inclusion of textural features in the classification increases the accuracy of identification of cover types (Soares et al., 1997).
Rignot et al. (1997) found that, at L band, multiple polarizations were required to obtain a reliable classification of deforestation and secondary re-growth in Rondonia,
Brazil. The double reflection of emergent vegetation and water surface can produce
a 3-6-dB increase in L band backscatter magnitude over non-flooded condition, and
a simple model has been presented (Engheta and Elachi, 1982). Use was made of
such differences for the monitoring of flooding of marsh lands using multi temporal,
multi band and polarization radar data (Pope et aI., 1 997). The use of radar for cover
identification was reviewed by Kasischke et al. (1997).
To decrease the dependency on empiricism, explicit relationships between radar
backscatter and structural attributes have been studied (Hoekman, 1990),
(Dobson et aI., 1995), but more work is required before the results can be used at
practical implementation level.
153
average the number of pixels from the same cover type, or by multi-look averaging
of the data, at the expense of loss of texture information.
An advantage of imaging radars is that they are insensitive to local weather conditions. The orbital SAR systems (ERS 1-2, JERS-l, Radarsat, SIR-CIX-SAR) provide multi-date data, which is valuable for monitoring the effects of cover on hydrology, particularly in cloudy weather types.
The backscatter signal of terrain, which is used to process an image, depends
on the wavelength, cover species related (probably morphologic) features, surface
roughness of the ground and moisture content, which leads to ambiguity of single
radar data. Closed canopies of more or less uniform height cause attenuation of the
signal, especially in the C and X band. The shorter wavelengths (C band and shorter)
display a strong relation between the backscatter and species dependent properties
and can be used to detect stress or diseases and the forest development (Hoekman,
1990). Natural forests with mixed species compositions may show a very limited dynamic range of radar backscatter. A high degree of canopy transparency was found
for the longer wavelength L-band, which is useful for differentiation of broad forest classes and for detecting ground cover conditions, or flooding below the canopy
(see also Figs. 10.8 and 1O.9). It may be noted that multi-spectral data pertains to
the canopy only.
An interesting specific application of radar was the detection of water below a
woody forest canopy by the L and P bands. Presence of water increases the backscatter by ground-trunk two-bounce scattering(Hess et aI., 1995}. However, in nonwoody wetlands, a decrease of the backscatter may occur because of the forward
specular scattering of the pulse signal. -
Many studies have used the maximum likelihood approaches or a decision tree
method, based on training sets containing ground truth or local knowledge. It was
found that single date, single frequency/polarization data has limitations, but multidate and multichannel SAR leads to enhanced classification (Kasischke et aI., 1997).
As is the case with multi-spectral data, inclusion of textural features in the classification increases the accuracy of identification of cover types (Soares et al., 1997).
Rignot et al. (1997) found that, at L band, multiple polarizations were required to obtain a reliable classification of deforestation and secondary re-growth in Rondonia,
Brazil. The double reflection of emergent vegetation and water surface can produce
a 3-6-dB increase in L band backscatter magnitude over non-flooded condition, and
a simple model has been presented (Engheta and Elachi, 1982). Use was made of
such differences for the monitoring of flooding of marsh lands using multi temporal,
multi band and polarization radar data (Pope et aI., 1 997). The use of radar for cover
identification was reviewed by Kasischke et al. (1997).
To decrease the dependency on empiricism, explicit relationships between radar
backscatter and structural attributes have been studied (Hoekman, 1990),
(Dobson et aI., 1995), but more work is required before the results can be used at
practical implementation level.