52 
The standard deviations of the spatial variability of the different sites are comparable (about 5% - 8 %), 
and have no dépendance with the spectral band except for the site Algeria 5 where we have encountered 
different sand types. 
One may also use these measurements to estimate the spatial variability of the spectral signature, that is, 
the ratio of the surface reflectance at a given wavelength to that at 550 nm for example. Figure 5 shows the 
corresponding results for Algeria 5; the error bars in this figure represent +/- the standard deviation of about 
10-20 measurements of this ratio over a 2 x 2 km 2 area. It is shown on this figure that the spatial variability of 
the spectral signature is quite small, of the order of 1 - 3 % in relative value. 
5. CONCLUSIONS AND PERSPECTIVES 
The Saharian field experiment has been a key step in the assessment of the metrology of the 
reflectances of different Algerian zones to be used as calibration standards for optical satellite sensors. 
Measurements of the surface bidirectional spectral reflectance, p sur f ace (^, 6 S , 6 V , ^)> ^ avc ^ )een performed with 
sufficient sampling of wavelength and sun and view angles. Other experiments, intended to assess the accuracy 
of the former measurements, have also been performed, including various laboratory and field calibration 
experiments (to assess the instrumental accuracy), and a characterization of temporal and spatial variabilities of 
the surface reflectance. The peak-to-peak relative variations of calibration coefficients of the ground radiometer 
and of REFPOL for various calibration experiments is of the order of +/- 7 % (Table 3); although the interband 
calibration accuracy of these instruments is certainly much better, this figure gives an upper bound of the type 
of accuracies involved in our measurements. Space and time variabilities have also been assessed, although the 
involved time scales (a few hours) and length scales (a few hundred meters) are of limited amplitude. Note, 
however, that variabilities at larger scales in time (seasonal time scales) and space (a few tens of kilometers) 
have been measured with satellite data (Cosnefroy et al, 1993). 
Much remains to be done, however, before the objective announced in the introduction, that is, the 
metrological measurements of the spatially-averaged, TOA reflectance PtOA^> 0 s > ®v> ♦) °f a given desert 
zone can indeed be achieved. 
At the surface level, it will be necessary to establish a model of the spectral and directional variations of 
reflectance from our measurements, since one needs to extrapolate the measurements to new spectral or 
directional configurations. As a first step, one could model for each spectral band the directional variations 
with a surface bidirectional reflectance model such as those of Verstraete et al. (1990) or Roujean et al. (1992) 
(see a review of directional models by Goel, 1988) and make interpolation of reflectance between two 
consecutive spectral bands (our data do not support the assumption of uncoupling between spectral and 
directional variations, see 4.2.). It will perhaps turn out to be necessary to include in the model a correction due 
to the fact that the needed metrology is at a scale of a wide field of view sensor, spatially averaged over 1 
kilometer square or more, whereas our measurements are strictly local. This space averaging must take into 
account the effect of dune relief, w hich should result in enhancements of reflectance in the backscattering half 
space. 
The next issue is to transpose the reflectances measured at the surface level to TOA reflectances. As a 
boundary condition, we need the bidirectional reflectances with a suitable angular step to match the atmosphere 
radiation transfer code requirements. This is an additional reason why a BRDF model is necessary. The 
atmospheric corrections are mainly based on the aerosol characterization. The currently used aerosol model is a 
Junge size distribution with a slope which is consistent with measured aerosol optical thicknesses. The 
refractive index is quite arbitrary. We have just verified that such a model is able to retrieve correctly a very 
sensitive parameter such as the polarization. Figure 6 is a comparison of the sky polarization measurements 
(with REFPOL) and radiative transfer simulation based on the optical thickness measurements. Even with this 
crude description of the aerosols, we have an acceptable restitution, which is encouraging. We can also refine 
our aerosol description using the field measurements. 
The obtained TOA bidirectional spectral reflectance model can then be tested with satellite data such as 
those of METEOS AT-4 or AVHRR/NOAA. Also, different situations of atmosphere conditions can be 
simulated to study the corresponding variability on the signal. 
ACKNOWLEDGMENTS 
We gratefully thank Fethi Benhamuda and Abdeljelil Lansari, from CNTS of Arzew (Algeria), for their helpful 
participation to the measurements, and Robert Oliel, from the CAP association (Boussens) for logistic support. 
This study has been funded by CNES.