study area with second polynomial model with 47
GCP
’s
(ground control points) with a root mean square error of less
than half a pixel. Finally, the decomposition parameters were
used for classification. Training sites corresponding to rice
field, urban land, water bodies, and plantation were selected
based on ground reference data from the field visits. The clas-
sification was performed on single-date, hybrid polarimetric
data and
LISS
-IV data when the rice was in the reproductive
stage and ideal for rice classification. The supervised paral-
lelepiped minimum distance classifier was applied on the
decomposition parameters. A supervised maximum likeli-
hood classifier was applied on the Resourcesat-2
LISS
-IV data.
Due to difference in path orientation of
LISS
-IV and
RISAT
-1
imagery, a subset was selected (Achanta Mandal representing
an administrative unit) for comparison purposes. The accu-
racy was estimated using the field observations and spatial
distribution of both the classified
results was compared over a common
area overlapping the
LISS
-
IV
and
RI-
SAT
-1 imagery (due to differential path
orientation).
Results and Discussion
Response of Rice Crop
The mean and standard deviation of the
backscatter coefficient of various class-
es is shown in Figure 1. The backscatter
coefficient of settlement class showed
high mean value and that of water
showed lowest value. Although, there
is a separation of mean values of vari-
ous classes, it is difficult to classify the
data using backscatter alone because of
the high standard deviation leading to
overlapping of various classes.
The amplitude (
RH
,
RV
), Raney
decomposition and Resourcesat-2 (
LISS
IV) images, covering parts of study area
is shown in Figure 2. In this part of the
study area, the majority area is under
rice crop, followed by plantations, and
other land-cover features. The zoomed
view of rice fields is shown in Figure
2d and 2e for better visualization.
In the Raney decomposition im-
age both the odd and double bounce
scattering mechanism were dominant
over the rice crop at reproductive stage
(Figure 4c). The degree of polarization
(
m
) is a vital parameter for decomposi-
tion (Charbonneau, 2010) with higher
values of
m
indicates the purity of
scattering mechanism. It is observed
that rice crop exhibits high degree of
polarization followed by settlement
and water body (Figure 3) because of
high magnitude of double bounce scat-
tering than other scattering mecha-
nisms. The relative phase, a sensitive
indicator of double bounce, is positive
when odd bounce scattering is domi-
nant and negative if double bounce
scattering is dominant. (Kausika, 2013).
The three decomposition parameters, i.e., even, odd, and
double scattering mechanisms of rice and other classes were
shown in the Figure 4. The scattering plots of the double versus
volume bounce scattering (Figure 4a), odd versus volume
scattering (Figure 4b) for various classes delineated in the study
area show that the rice, plantation, settlement, fallow, and water
were clearly separable. The rice crop at reproductive stage is
likely to have a scattering mechanism similar to settlement.
The settlement class shows high magnitude of double bounce
than rice but in periphery of the settlement, it mixes up with
Figure 1. Mean and standard deviation of
σ
0
(
dB
)
in Hybrid Polari-
metric data.
Figure 2.
fcc
images of
rh
,
rv
Amplitude data, Raney decomposition and Resourcesat-2 (LISS IV).
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2015
559
