pressions
equation
que. The
letermine
> surface
(2)
(3)
(4)
of the di-
e, k : the
is, at fre-
IS heights
c0 « 70°.
ue, (Oh et
es of sur-
nique and
soil mois-
his study.
ate of soil
(excluding
m the avail-
ve derived
dielectric
» dielectric
ire values
allikainen,
A. and Wu,
| over each
ed values.
and L are
e RMS er-
cm for the
yr the esti-
of the soil
he C-band
> presence
n an over-
imated soil
data of the
o-polarized
ratio Zhh, which is greater than 1 for 83% of the dataset of
the first campaign. For the dataset of the second campaign,
there are only 37% greater than 1. This situation appears,
although we stay in the validity restrictions (kh « 2.5 and
6 > 30°) for which (Dubois et al., 1995) ensured, that the
ratio is always less than 1. This phenomenon is probably
due to the difference in local incidence angle for each C-
band image. Indeed, for the data of the first campaign we
have 0 C-HH « 8 C-VV, therefore the backscattering in the
C-HH image is greater than the backscattering in the C-VV
image.
In Fig. 4 and Fig. 5 we have also displayed the results
excluding the points for which the co-polarized ratio (also
named p in 4.2) is less than 1.
21.04.97 : Dubois (mv)
# A Cp
9 1
2 e |!
»
E Ac
10 4 i qi
> é EAN RE (0-4cm)
1 : | Trime
0 T 1 $ Y»:
1 2 3 4 5 6 7
-10
fields
Figure 4: Measured and estimated soil moisture.
. 21.04.97 : Dubois (RMS)
30
| : 8
251 !
jd i . : 9 & | — measured
t À : ; é à c
5 o
D. 15 8 4 8 A C p«l
2 à i e |
10 4 ; à 0 |p
0,5 4
0,0 t
1 2 3 4 5 6 7
Figure 5: Measured and estimated RMS height.
A more precise analysis has to be made on each individual
lot. For the field 5, we observe for both campaigns that the
estimated soil moisture is closer to the TDR measurements
than to the gravimetric measurements in the 0—4 cm layer.
This field was more dry on the surface than in a deeper
layer and has been ploughed just a few days before the
first flight. We may assume, that the C-band radar return
penetrates this bare and dry soil and thus it returns an in-
formation about a deeper layer.
The calculation of the RMS height over each pixel in a field
is more stable with the L-band than with the C-band. The
estimation was particularly good for the bare soils like fields
2 and 5. Also remarkable is the RMS height estimation
for the intensive area. If one applys the model over the
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
25 points of the area, one can notice the sensibility of the
model to the local incidence angle.
The roughness from the field 3, which was covered with
winter wheat since October 1996, has not changed be-
tween the two flights. Yet, the algorithm overestimates the
roughness for the first campaign. The only parameter who
has changed for this field between the two flights is again
the local incidence angle. In view of the fact that both,
roughness and soil moisture estimates are correct for the
second date, the local incidence angle plays a significant
role.
In a nutshell, the model from DUBOIS has to be improved
above all for the soil moisture estimation. The roughness
estimation is satisfactory, although the influence of the local
incidence angle should be overridden.
4.3.2 Application of the OH-Model The main disadvan-
tage of this model is the employment of the cross-polarized
backscatter coefficient, which is more sensitive to system
noise and to the presence of vegetation. This narrows our
database down to the three L-bands images (two for the
first flight and one for the second). The P-band images
have only been delivered in HH- and HV- polarizations, be-
cause the others have suffered from cross-talking and a
lack of system sensibility.
Fig. 6 represents the soil moisture estimates for the L-band
during the second flight. In the legend, mi corresponds to
the estimation after the filtering within object boundaries,
l.ep corresponds to the estimation after EPOS-filtering and
ml.p « 1 is the same as ml, excluding the points for which
the co-polarized ratio p is greater than 1.
21.04.97 : Oh (mv)
4500
40,00 4
A m
e lep
- 4.1
f Se
‘ $ (
3 2000 4 IIT
mi
Figure 6: Measured and estimated soil moisture.
For lot 1, the values filtered with the classical filter seem
to fit better the measurements in the 0-4 cm layer than the
values after filtering within object boundaries. Neverthe-
less, the observation from the values inside lot 1 let appear
an extreme overestimation for one of the points, for which
also the ratio p is equal to 1.48. This case is not consid-
ered by (Oh et al., 1992) who affirms that the co-polarized
ratio p — = < 1 for all angles, roughness conditions,
and moisture contents. Thus the algorithm has been ap-
plied excluding the points for which p > 1 and these results
(lp < 1) are the only one to consider.
The averaged soil moisture estimations are at all events
considerably better (figure 6) than with the DUBOIS-model
in L-band (RMS error= 4.05vol% toward 13.9vol%).
553
