IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002
Soil moisture conditions in the study area
( 10-Dec-2001)
70
560
A
u D
5
d © E0-5cm
9 30 L15- 10cm
=
= 20
a
= 0
0
154. 7 10:13 458 49 22 25 28 31 34 37
Sample #
Soil moisture conditions in the study area
( 3-Jan-2002) :
70
m8
=
v 90
z
^ © 3 mO-Scm
9 30 T + 05-10cm
= :
8 2 len a HI} ^
a 40 4 j 1 d Li
0 : T ; T T
1 13 16 19
Sample #
Figure 1. Soil moisture conditions in the study area during the
study period at two depths
2.3 PROCEDURE
Two components make up the procedure adopted in the study:
near real time ground truth collection in the study area and
processing and analysis of satellite data.
2.3.1 Groundtruth Collection: On all the days of
RADARSAT SAR data acquisition information on relevant
ground parameters was collected in real time. Initially, suitable
sampling locations have been identified with the following
criteria: (i) the sampling sites should be homogeneous and large
enough for easy identification on satellite data, (ii) easily
approachable, and (iii) provide a range of soil moisture, surface
roughness and crop cover conditions. The information collected
from the fields include (i) sampling site locations in terms of
geographical coordinates using hand held GPS receiver, (ii)
soil moisture samples for gravimetric analysis in the 0-5 and 0-
10cm soil depths, (iii) soil surface roughness and (iv) crop
related information viz, type, height, density, age etc in the
sampling sites. These information collected were carefully
processed and analyzed subsequently.
2.3.2 Preliminary Processing of Satellite Data: In the first
step, the 16 bit unsigned path image plus RADARSAT SAR
data sets procured from RSI, Canada for the study has been
converted to the float valued radar backscatter coefficient (0?)
and geocoded by image to image registration with the geocoded
IRS-1D LISS-III data of the study area acquired on 9 December
2001. To reduce the speckle in the SAR data, the data were
processed by iteratively applying the 3x3 multiplicative Lee
filter. Selection of a smaller convolving window of 3x3 ensured
the retention of edges and application of the filter in iterative
fashion helped suppression of speckle. Based on the sampling
site coordinates obtained during the field work, their
712
identification on false colour composite image of IRS-1D LISS-
III image, sampling site statistics like mean radar backscatter
coefficient, standard deviation, minimum and maximum o? and
mean grey level values of the fields in the four channels of
LISS-III data have been extracted. Following the same
procedure, statistics for all the sampling sites could be extracted
successfully from the four scenes of RADARSAT SAR and
two scenes of IRS LISS-III
2.3.3 Data analysis: Three approaches have been followed to
data analysis to retrieve soil moisture from the satellite data
used in the study. These are (i) develop soil moisture and radar
backscatter nomograms, (ii) develop regression relationships
between soil moisture and radar backscatter coefficient at 16°
and 45° look angles and (iii) combined use of radar and optical
data from RADARSAT SAR and IRS LISS III sensors.
2.3.3a. Soil moisture and c? nomograms: In a simple
approach, an attempt was made to develop nomograms
describing the relationship between soil moisture at different
depths and RADARSAT SAR backscatter data acquired at 16?
look angle with HH polarization in C-band. This approach was
based on the conclusions drawn by Ulaby and Batlivala (1976)
and Dobson et al (1981). They observed that sensitivity of radar
to soil moisture variations is maximum with radar observations
made at 7-17" look angle, with HH or VV polarization and C-
band, while minimizing the effect of surface roughness and
crop cover. Thus nomograms (Figure 3) have been developed
based on the observed minimum and maximum soil moisture
conditions for two depths and their corresponding radar
backscatter coefficient values, while ignoring the influence of
surface roughness crop cover conditions. The observed
minimum and maximum soil moisture conditions were,
respectively, 9.3g/g and 58.16g/g for th e0-5cm depth and
9.9g/g and 53.83g/g for the 0- 10cm soil depth. Corresponding
minimum and maximum values of backscatter coefficient at 16°
look angle were —7.14dB and —1.41dB. À comparison of
soil moisture at sampling sites observed and estimated using
these nomograms shows that for depths 0-5cm and 0-10cm
respectively, the standard error of estimate is 8.8 g/g and 8.1
g/g. The error in soil moisture estimation was found to be
greater than 10.0 g/g for fields covered with dense crop or with
rough surface conditions. It clearly showed that it is difficult to
ignore the effect of surface roughness and Crop cover
conditions on the relationship between soil moisture and radar
backscatter coefficient even at the ideal radar configuration of
C-band, HH polarization and 16° look angle.
2.3.3b. Regression Approach: In the correlation studies
carried out by regressing the RADARSAT SAR backscatter
coefficient data acquired at two look angle mentioned above,
the soil moisture samples collected in the fields have been
separated on bare and crop covered fields. Linear regression by
least squares principle has been carried out separately between
radar backscatter coefficient at 16° and 45° taking (i) each of
them independently and (ii) collectively and the soil moisture at
0-5 and 0-10cm soil depths. To follow such approach, initially
the correlation between the two sets of radar observations at 16°
and 45° look angles has been tested for any redundancy. It
showed, a correlation of only 0.44 between the SAR data sets.
It indicates that data acquired at near vertical look angle
responds more to the soil conditions with significant
penetration, while the oblique view data responds mostly to the
surface cover conditions.
