533
reflectivity. Figures 4 (c) to 4 (d) concern the PO model, which reproduces well the angular variations for the
H polarization for the two soil moisture conditions. For the V polarization, at lower angles a good agreement is
observed, which is not the case for the other increasing angles, because of the tendancy of the model to keep
the same behaviour as a smooth surface at Brewster angle. Nevertheless, we have to be careful on the value of
the polarization ratio at 5.05 GHz because of problems on vertical channel calibrations. Figures 4 (e) to 4 (0,
for 10.65 GHz, show a relatively poor agreement between simulated data and observed data, both in mean
level, and angular variations. Since the data set is not fully calibrated, it is difficult to comment precisely the
differences between the two channels. But, we can note the difficulty for the model to accurately predict the
angular variations for H and V polarizations. The angular behaviours of the measured data at 10.65 GHz show
a rougher surface than at 5.05 GHz, which is not taken into account in the output model (10.65 GHz not
different from 5.05 GHz). This is perhaps a limitation of the model, which considers explicitly two roughness
parameters, but whose outputs are slope dependent for low angles (Mo et al. 1987, Saatchi and Wegmuller
1992). Figures 4 (g) to 4 (h) are a good illustration of the roughness influence when comparing to figures 4 (e)
to 4 (f). The observed data show the increase in the mean level of the emissivity, the lesser dynamic angular
variations, and the decrease in the polarization ratio. The simulated data can predict a good angular variation
for the V polarization (due to the polarization mixing) which would not be true for the PO model. The shift
between observed and simulated emissivities could be attributed to the difficulty of determining the soil
moisture, because of the heterogeneity due to the rough surface. Figures 5 (a) to 5 (b) are an example of the
improvement which can be reasonably expected, in considering surface scattering models.
figure 5 : Comparison measure /
simulation for SL field, wet condition,
at 10.65 GHz, using (a) Wilheit model,
(b) PO model
CONCLUSIONS
Both Wilheit and Fresnel models provide a good agreement with the measured data. The validation of the
Wilheit model, over a wide range of soil moisture conditions, shows that the model can well predict the
microwave emission for smooth surface conditions. Moreover, in L band the Wilheit model which accounts for
soil moisture and temperature gradients improve the accuracy of the results in comparison with the Fresnel
approach. The statistical results obtained for L band, further support the earlier works done on sampling depth
of microwave radiometers. But, in the statistical study, the moisture gradient effect is not so evident for C and
X bands results. This shows, for higher frequencies, that the Fresnel model, with proper sampling depths of
moisture and temperature is able to predict the TB correctly. Lastly, we can conclude that the Wilheit model
appears useful to study the soil moisture and temperature gradient effects on microwave emission for lower
frequencies but the Fresnel model seems to be sufficient to invert soil moisture from radiometric data.
The second part devoted to surface scattering models, has shown that it is possible to well reproduce
the angular behaviour of the emissivity for the H polarization, with the three models; but problems concerning
the soil moisture sampling depth still remain. This last point could be improved in merging the two previous
approaches ( radiative transfer and surface scattering models). Nevertheless, to account for the roughness
influence, for a slightly rough surface, the derivation of the SPM model to the second order has to be done.
Then, roughness information is also contained in the vertical polarization (as well as in the polarization ratio)
and the models, especially, the PO model, are not able to reproduce well both polarization variations.
Moreover, the GO and PO model are sensitive to the rms slope (not explicitly for the PO model), which fix the
degree of the surface roughness (mean level in emissivity; polarization ratio). Consequently, because of the
great variability of the surface correlation length (1), we can observe large dispersions for the same field, on
the simulated outputs. On the other hand, figures 5, which concern SL field whose roughness characteristics do
