Scattering-Mechanism-Based Investigation
of Optimal Combinations of Polarimetric SAR
Frequency Bands for Land Cover Classification
Zhixin Qi, Anthony Gar-On Yeh, and Xia Li
Abstract
Aiming at steering the selection of optimal combinations of
polarimetric
SAR
(
PolSAR
) frequency bands for different land
cover classification schemes, this study investigates the land
cover classification capabilities of all the possible combi-
nations of L-band
ALOS
PALSAR
fully
PolSAR
data, C-band
RADARSAT-2
fully
PolSAR
data, and X-band
TerraSAR-X HH
SAR
data. A method that integrates polarimetric decomposi-
tion, object-based image analysis, decision tree algorithms,
and support vector machines is used for the classification.
Polarimetric decomposition theorems are used to interpret
the scattering mechanisms at the different frequency bands
to reveal the effect mechanisms of
PolSAR
frequency varia-
tion on the classification capability. This study finds that
(1) X-band
HH
SAR
is not necessary for classifying the land
cover types involved in this study when C- or L-band fully
PolSAR
are used; (2) C-band fully
PolSAR
alone is adequate for
classifying primitive land cover types, namely, water, bare
land, vegetation, and built-up areas; and (3) L-band fully
PolSAR
alone is adequate for distinguishing between various
vegetation types, such as crops, banana trees, and forests.
Introduction
Land cover information is recognized as a key input to a vari-
ety of environmental models (Li
et al.
2017; Liu
et al.
2017).
SAR
remote sensing has been widely used in land cover inves-
tigation since the launch of orbital
SAR
systems, such as
,
ERS-1
, and
RADARSAT-1
, for routine data c
al.
1996; Saatchi
et al.
2000; Shao
et al.
orbital
SAR
systems operate at a single fr
provide limited spectral information, which can obscure the
separation of different land cover types (Qi
et al.
2012). Po-
larimetric
SAR
(
PolSAR
) can compensate for the limited spectral
information by providing the measurements on the changes in
the polarization state of electromagnetic waves reflected from
the earth’s surface (Lee and Pottier 2009).
PolSAR
performs
much better than traditional
SAR
in land cover classifica-
tion because it characterizes different scattering mechanisms
of ground targets using polarimetric information (Lee
et al.
2001).
PolSAR
data have been increasingly used for land cover
mapping since they were obtained by orbital
PolSAR
systems,
such as
ALOS
PALSAR
,
RADARSAT-2
, and
TerraSAR-X
(Li
et al.
2012; Gumma
et al.
2015; Mandianpari
et al.
2017).
Frequency is one of the most important
SAR
system param-
eters that influences the intensity and patterns of the radar
returns from ground targets (Henderson and Lewis 1998).
SAR
data acquired at different frequency bands have been com-
pared in a variety of land surface applications. Kasischke
et
al.
(1997) summarized that L- and P-band data are optimal for
biomass estimation and flooded forest monitoring and that X-
or C-band data are suitable for monitoring coastal/low-stature
wetlands, tundra inundation, and fire-disturbed boreal for-
ests. Lee
et al.
(2001) found that L-band data are best for crop
classification and that P-band data are optimal for forest age
classification based on the use of airborne multi-frequency
PolSAR
data. McNairn
et al.
(2009) compared the crop clas-
sification results obtained with L-band
ALOS
PALSAR
, C-band
ENVISAT ASAR
, and C-band
RADARSAT-1
images and concluded
that large-biomass crops are appropriately classified using
PALSAR
data, whereas C-band data are required to accurately
classify low-biomass crops. Turkar
et al.
(2012) stated that L-
band
PolSAR
works better than C-band
PolSAR
for the classifica-
tion of various land covers. Naidoo
et al.
(2015) indicated that
L-band dual-polarized
SAR
is more effective in the modeling
of woody structure than X- or C-band dual-polarized
SAR
.
Many studies have found that land cover classification ac-
curacy can be improved by combining
PolSAR
images acquired
bands. Dobson
et al.
(1996) combined
nd
JERS-1
data for land cover classifica-
t the results for the composite image
were superior to those obtained from either of the two sensors
alone. Similar conclusions have been drawn with
PolSAR
data.
Chen
et al.
(1996) examined airborne P-, L-, and C-band
Pol-
SAR
data in the classification of forests, water, bare soil, grass,
and eight other types of crops. The examination showed that
the multi-band attained the best discrimination capability and
that the P-band image produced better classification accura-
cies when a single-band image was used. Pierce
et al.
(1998)
used
SIR-C/X-SAR
PolSAR
imagery for land cover classification
and discovered that the single-scene classification accuracy
obtained with the L or C band was better than 90%, and with
the addition of X-band data, the accuracy improved to 98%.
Skriver (2012) found that the combination of L and C bands
yielded lower classification error rate than any of the single
frequency bands by using airborne L- and C-band
PolSAR
data for crop classification. Hagensieker and Waske (2018)
Zhixin Qi is with the Guangdong Provincial Key Laboratory
of Urbanization and Geo-Simulation, School of Geography
and Planning, Sun Yat-sen University, Guangzhou, China.
Anthony Gar-On Yeh is with the Department of Urban Planning
and Design, University of Hong Kong, Hong Kong
SAR
, China.
Xia Li is with the School of Geographic Sciences, Key
Laboratory of Geographic Information Science (Ministry
of Education), East China Normal University, Shanghai,
China; and the Guangdong Provincial Key Laboratory of
Urbanization and Geo-Simulation, School of Geography and
Planning, Sun Yat-sen University, Guangzhou, China (lixia@
geo.ecnu.edu.cn or
). He serves as the
corresponding author for this article.
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 11, November 2019, pp. 799–813.
0099-1112/19/799–813
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.11.799
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
November 2019
799
