the different radar polarization measurements has the different 
sensitivities to the different surface properties. Especially, the 
radar cross-polarization measurements are very sensitive to the 
vegetation volume information since the surface backscattering 
does not generate significant cross-polarization signal. As 
shown in [8], the direct volume backscattering component of 
the co-polarization in (5) can be directly estimated from cross 
polarization signal, that is 
^vV)^;(0)=3-*1(0) (6) 
Where pp can be HH or VV. Then, using HV polarization of 
PALS, volume term can be removed from total scatter. 
Surface- Volume Interaction Contribution 
Ulaby 191 set up an empirical model for <j° sv (6), the expressions 
are 
dependent on incidence angle. When two acquisitions is 
obtained, a formula can be written as 
A($„SR)T BW 
In SMEX02, PALS sensor has a constant incidence angle (~ 
45 ° ), so (9) can be expressed as 
__0 l ' 
= (io) 
The relationship between volumetric soil moisture contents and 
T can be obtained from [13], namely: 
ln(T) = -0.2312 *ln(mv) + 0.6874 (11) 
<Jp P (&) = 1.924 o)-cos#-(l + 0.924-iy-r + 0.398-(iyr) 2 ) 
[l-exp(-1.925rsec#)]-exp(-1.372r' 12 sec#) (7) 
• exp[-0.84(fcs) 2 cos#] ■ r pp 
Where r is optical thickness, co is single scatter albedo, ks is 
normalised roughness factor, k has the value of 2n! a and s is 
surface RMS height, r is Fresnel reflectivity, which is a 
PP 
function of incidence angle and dielectric constant, therefore, a 
function of soil moisture. 
Combining (10) and (11), the relative soil moisture change can 
be expressed as: 
m. 
m. 
V2_ _ (_il.) 0.2312-5(6») 
cr„ 
The parameter B(0) can be obtained from AIEM simulation 
when incidence angle is 45° . In this paper, £(#) is 1.08, 0.90, 
1.10 and 0.88 for L-W, L-HH, S-VV and S-HH, respectively. 
Because cr° (#) is quite smaller than cr° (Q) and too many 
unknows in (7), we have to do some assumptions for the above 
parameters for the computation of olfQ), we believe these 
assumptions will not bring too much errors in final results, 
especially for VV polarization 191 . co is set a constant 0.06 for S- 
band and 0.09 for L-band 1101 (These are just assumptions from 
the literature), s is set 0.9 cm (which is field average), r is 
PP 
computed by setting soil moisture 0.15m 3 m' 3 (Approximately 
field average). Combining with the r from part 3, <j°(#) can 
be estimated for both HH and VV polarization. 
Finally, can be obtained with (1) (5) (6) (7). In this 
process, we must point out, the negative cr°(#)s are removed 
from the database. Negatives can be the reason of wrong 
estimation of r, volume contribution, or the wrong constant 
setting in (7). 
5. SOIL MOISTURE CHANGE RETRIEVAL AND 
VALIDATION 
In many studies 1 " 11121 , AIEM model is simulated with possible 
combinations of dielectric constant and a wide range of surface 
roughness, a simplified relation between the radar backscatter 
and the reflectivity is developed for bare soil, the expression is 
a°(0) = A(0,SR)- T B(9) (8) 
Where the parameter A is dependent on incidence angle and 
roughness, r is surface reflectivity at normal incidence. B is 
In SMEX02, three depths soil moisture are recorded, 0~lcm, 
0~6cm and l~6cm. Obviously, the toppest layer must be 
detected by radar signal. The linear correlation between radar 
backscatter at W polarization and soil moisture is computed, 
the coefficient is shown in Table 2, from which we can see, the 
correlation coefficient at 0~lcm for all four polarization is 
higher than 0~6cm. Therefore in the validation, 0~lcm soil 
moisture is used. 
SM 
ComLW 
ComSVV 
SoyLW 
SoySW 
0~ 1cm 
0.788 
0.691 
0.858 
0.821 
0~6cm 
0.621 
0.513 
0.793 
0.734 
Table.2 Linear correlation coefficient between radar and soil 
moisture at two depths 
The results are shown in Fig.2 for both soybean field and com 
field, and at S-HH, S-VV, L-HH and L-VV polarization. 
Generally, L-band has better results than S-band, and soybean 
field has better results than com field, both are quite consistent 
with our prediction. Compared with Table.2, for L-band, the 
algorithm can improve the inversion accuracy (from 0.86, 0.79 
to 0.91, 0.85, respectively), proved its usefulness. But for S- 
band data, the correlation coefficients from the algorithm are 
lower than linear regression (from 0.69, 0.82 to 0.58, 0.77, 
repectively), this shows that the algorithm is incapable of being 
used in dense vegetation area when wavelength is lower (than 
L-band). 
6. CONCLUSION 
An algorithm is developed to estimate soil moisture change 
combining optical and radar data from SMEX02. Canopy