section of the test area of 30 m x 30 m was moistured by ar-
tificial watering after the first imaging campaign. This spe-
cial intensive measurement location was partitioned into a
set of 5 x 5 pixels; field measurements were taken at each
of these 25 positions.
Concurrent with the imaging campaigns, soil moisture mea-
surements have been taken at 62 positions, the location of
which was determined by GPS. The ground truth measure-
ment positions were regularly distributed over the whole
test area. Two Time-Domain-Reflectometry (TDR) intru-
ments were used for measuring the soil moisture in the
layer 0 to 15cm. Additionally and in order to calibrate the
TDR measurements, gravimetrical measurements of the soil
moisture were taken in layers of 0 to 4cm, 4cm to 8cm,
8cm to 16cm and 0 to 16cm. The correlation of both mea-
surement series was at a remarkable level of 0.92.
Comparing field measurements taken at the date and time
of both imaging campaigns (see Fig. 1), we conclude that
soil moisture was not generally higher during the second
imaging campaign, although short precipitation events had
occurred before. A closer look reveals however, that at least
at some locations in the test area the soil moisture varied; in
the artificially watered section, the measurements definitely
show higher values for the second campaign date.
TDR measurements
45 p
| 1 3
40 il f^ |
Hg i i Pa i
: / h í * i i
j LAN el
Yuva EX Ar Ni
+) > j à x
o
a
mv (Vol-%)
8
25 41
-— 4 M9 f A 0 7 0 IV. A OD In i O6 e 09 ID OQ on ro 2
z2,8,9. T. Tx Noi No^6 00 04 N a n n» Ww o
fields
Figure 1: Soil-moisture measurements at the dates of the
two imaging campaigns.
All ground truth measurements pertain to 6 different, rather
large agricultural fields, two of which were bare, two other
fields were covered by weakly developed winter wheat and
the last two fields were covered by medium developed win-
ter wheat with a maximum height of 22cm. Plant row di-
rection, row distance, ploughing width, plant height and the
estimated vegetation coverage density were notified.
Concurrent with the imaging campaigns, the soil surface
roughness of each field was measured using a special rough-
ness measurement gauge that allow the calculation of height
variance of a local surface section.
2 IMAGE PREPROCESSING
2.1 Generating the images
The raw data acquired by the E-SAR sensor system has
to be processed in several steps to obtain the image data.
In a first step one has to compensate for platform motion
errors due to the irregular flight path of the airplane.
In the next step, the radiometric calibration, the backscatter
values are calculated out of the sensed amplitudes. Here,
550
one has to account for some calibration constants provided
by the DLR and for the incidence angles.
The incidence angle of the wave is needed at each pixel po-
sition to obtain backscatter values which are independent
of the orientation of the sensed surface and also of the dis-
tance between surface and sensor. We have developed a
program which calculates the local incidence angle at each
pixel position function of the surface normal and the flight
path of the airplane. The surface normal is calculated out
of a DTM.
In the next step, the slant-to-ground range transformation
as provided by the EasiPace software is performed with re-
spect to the DTM. Since our DTM is georeferenced, the
final result of the described processing steps is a georefer-
enced ground-range SAR image.
2.2 Speckle removal
2.2.1 Filtering We have obtained good results in speckle
removal with the EPOS-Filter (Hagg and Sties, 1996). This
filter uses an adaptive window in order to smooth the data
in homogeneous areas but to preserve edges and isolated
reflectors.
2.2.2 Smoothing within object boundaries The back-
scatter values depend both on radar-specific parameters
(like polarization, frequency, incidence angle) and on object-
or surface parameters (like vegetation, surface roughness,
soil moisture, soil type etc.). In order to isolate the influ-
ence of these parameters, we partition the image in regions
such that inside a region the soil type, land-use, surface
roughness and incidence angles are constant. Thus, the
backscatter values of such a region are function only of the
soil moisture. :
We placed constraints on the four above mentioned param-
eters such that we distinguished five soil types, according to
their clay contents. We supposed also that the boarders of
each agricultural field represent the limits of the roughness
classes as well as the limits of the land-use classes. Based
on several tests, we considered three degrees classes for
each local incidence angle image.
These sets of adjacent pixels with homogeneous properties
can also be used in simple procedure to remove speckle.
According to (Von Poncét et al., 1995), one can also avoid
problems with speckle when smoothing at least over 200
pixels. In our case, the regions have a size of at least 500
pixels. Thus, smoothing the backscatter values inside the
regions, which are homogeneous with respect to a series
of parameters, should also help removing speckle. The re-
sults of speckle removal with the two methods are shown in
Fig. 2.
SAR image, EPOS-filtered image,
smoothed within object boundaries.
image
By classifying the smoothed backscatter values inside a
region we expect to obtain a relative measure of the soil
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
mois
pres
3.1
In oi
soil
SAR
Asa
mea
ficiel
colui
Table
ture.
In ge
21th
of Ac
band
meas
local
pred:
value
The |
value
layer
this |
VH a
value
than
agair
pene
The «
value
camp
the fi
ence
corre
To ar
the T
the C
ment:
0-4cn
analy
tweer
suren
