1170 
Second point is roughness effects that proved to be of significant importance on soil backscatter at 
C band (Beaudoin et al., 1990). In view of MF 7 on Fig. 1 (no soil moisture available for Fig. 2) we can point 
out an expected downward shift of o° compared to other MF sites (1- 2 dB), as the roughness is smaller for 
this site due to the lower rock surface cover reported in Tab. 1. Therefore, roughness effects can explain partly 
the level of backscatter temporal curves in Fig. 1 and the dispersion within backscatter values for each date in 
Fig. 2. 
In addition to roughness effects, the vegetation effects on backscatter must be addressed. 
Contribution from volume scattering can be neglected considering low biomass levels and the small size of 
leaves compared to C-band wavelength (5.6 cm). Therefore, only attenuation effects should be accounted for. 
Variable attenuation effects can be obtained due to temporal/spatial changes in water content, biomass and 
structure of grass and bushes. It is expected that attenuation properties in time should be mainly affected by 
water content changes with respect to season. According to this hypothesis, MF 1 to 3, which have similar soil 
roughness and biomass characteristics, are showing on Fig. 2 quite good dependence upon soil moisture 
respectively for drying period (DOY 170 & DOY 240) and wet period (DOY 275 & DOY 310) but significant 
difference between these periods. Therefore, this temporal difference can be attributed to soil backscatter 
attenuation from fully developped vegetation compared to dry one. During the wet season, mean water content 
values of 0.15-0.2 kg/m2 were measured in 90 on these sites, whereas water content can drop down to 0.02 
kg/m2 during dry season. In a first approximation, we can then estimate the resulting attenuation factor (r 2 ) 
according to the cloud model formulation as given by Attema & Ulaby (1978): 
r 2 
-2.B.m v 
cos a inc 
(4) 
where m v is this total amount of water contained in the vegetation layer (kg/m 2 ) and B an unknown constant, 
function of the canopy type/structure for a given radar configuration. At C band, most of the studies (Bertuzzi 
& Bruckler, 1991, Jackson et Schmugge, 1991, Prevot et al., 1993b) reported B values for various crops 
between 0.1 and 0.5. However, Jackson & Schmugge (1991) reported larger B values (up to 2) for short and tall 
grass covers. This difference from crops could be explained by the presence of stubble and detritus matter, in 
addition to structure. As few B values are available for rangeland, we can consider in a first approximation a B 
range from 0.5 to 2, leading an attenuation factor ranging from 0.6 to 2.5 using (4), in which falls observed 
values around 2dB. 
Further investigation will be done to study attenuation properties of the vegetation, taking into 
account the evolution in structure and moisture content. At this step, a statistical approach was designed to 
relate o° to soil moisture. 
3.2. soil moisture estimation from o° 
As it has been shown previously, the period of measurement during the year is greatly responsible for the 
change in backscattering relation to soil moisture because of the evolution of vegetation characteristics, both 
water content and structure. Two sets of data have thus been marked off corresponding to two different 
vegetation conditions. The first set (DOYS 170-240) corresponds to full development and well water supplied 
vegetation during the monsoon period while the second one (DOYS 275-310) concerns the drying season. 
A linear relationship was then deduced for the first set of data (Fig. 2. R=0.98) considering only 
MF 1 to 3 (similar roughness and vegetation characteristics). Since the vegetation attenuation acts as a 
translation of the a° curve when expressed in dB units, the second linear model was adjusted simply by 
changing the constant value. The resulting radar sensitivity to soil moisture (0.23 dB / % sm) is in good 
agreement with other studies (Bertuzzi & Bruckler, 1991; Prevot et al., 1993b) and the direct estimate of soil 
moisture from a° (Fig. 3 ) provides a root mean square error (RMSE) of only 1.31% sm. Even though it gives 
quite accurate estimate of soil moisture on most test sites representative of the watershed, in terms of roughness 
and vegetation conditions, further semi-empirical model development is needed. This model should take into 
account the vegetation attenuation effect through the use of a relevant and simple vegetation parameter. For 
soil moisture inversion, this parameter could be obtained either from other remote sensing data (VIS for 
example) or a priori knowledge of the temporal behavior of this parameter for this type of ecosystem.