)ped at the
ystem is a
S, C-band
yyal Danish
12.500°m,
ngles range
calibration
n is being
f the wave
| signal is
HH (H for
vertically
id V, or V
«scattering
1 different
scattering
scatter any
ind part of
tions) will
ight in the
get, e.g.
pends, for
)larimetric
when such
ce number
tudying of
areas have
ether with
reas. The
two years
thematic
using e.g.
cs of SAR
etric SAR
wetlands,
traditional
ferometric
including
ronmental
1encies, at
has to be
quent, and
o separate
n varying
available
ation and
Henning Skriver
characterization of area objects (e.g. agricultural fields and forests), feature detection (e.g. roads and buildings), and
change detection.
3.1 Area objects
The discrimination between different object types and identification of the individual objects are of decisive importance
for production of topographic maps. Different SAR images may be combined for the same area to obtain the optimum
data set for discrimination and identification of objects. Polarimetric SAR images are, as mentioned above, sensitive to
the geometric structure and dielectric properties of an object. SAR images acquired at different wavelengths will
improve the possibility of discrimination between objects, because the SAR images are sensitive to shape and
orientation of geometrical structures at the same size as the wavelength. Consequently, a SAR with different
wavelengths will be sensitive to different geometrical structures. The radar waves penetrate to a certain extent into, for
instance, the vegetation. The penetration depth depends on the wavelength, where a longer wavelength penetrates
deeper into the vegetation. Furthermore, SAR images acquired at different times during the year may be combined. .
This is especially an advantage when discrimination between different types of vegetation is important, because the
geometrical structure of the vegetation will change through the year (for instance, agricultural crops through the
growing season) and hence the polarimetric response will change (Skriver et al., 19992).
Classification of agricultural crops is possible using SAR data, and polarimetric and/or multitemporal acquisitions
ensure a high accuracy (Skriver et al., 1999b). In Fig. 1 are shown classification results for agricultural crops for
different combinations of polarimetric SAR data. The SAR data available are L- and C-band data acquired in both May
(early in the growing season) and in July (in the middle of the growing season). Using the full polarimetric information
in the classification, Fig. 1(a) shows that combinations of both frequencies and/or combinations of multitemporal data
improve the classification accuracy. Furthermore, it is clearly seen from Fig. 1(b) that the use of full polarimetric
information compared to single (VV) or dual polarization (VV--HH) strongly improves the classification accuracy. The
classification errors shown in Fig. 1 are obtained using a large number of test areas for the individual crop types,
making these error estimates realistic.
JOO —
90 pU T
= 80 = i E
5 Wt 5
S 60 5
S 50 2
d 40 ë
S 20 - ; B ue D
10 ce e mn
0+ c - | | rt
C July L+C July . € LE VV VV - HH Polarimetric
July+May May+July
(a) (b)
Fig 1. Classification error for agricultural crops for different combinations of EMISAR polarimetric SAR data. (a) Full
polarimetric classification with different combinations of multitemporal and dual-frequency combinations.
(b)” Classification using combined L- and C-band in May and July for single, dual, and full polarimetric data.
In general, it is possible to discriminate between forest and non-forest areas with a high accuracy. Reasonable results
are normally obtained for discrimination between the different forest types. In Fig. 2 are shown the results of the
above-mentioned classification using full polarimetric information from L- and C-band acquisitions in May and July for
forest and hedge areas with all other classes shown with the same color except the lake area. It is clearly seen from the
comparison with the aerial photographs shown in Fig. 2, that the SAR has a great potential in identifying and
classification of forest and hedge areas, which is of paramount importance in the monitoring of natural vegetation. The
backscattering from water surfaces is normally very low, because these surfaces are rather smooth, except in very windy
condition, where the water surfaces become relatively rough. The SAR will probably also be able to detect and identify
some of the protected natural vegetation types. In this case, the spatial resolution of the SAR could be a limiting factor,
because these areas are relatively small and heterogeneous.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 309
