613 
catch the seasonal cycles of the four sites, and that vegetation types can be easily distinguished. On the degraded 
bushland, the lack of data has to be related with hard environmental conditions in association with the typology of the 
site (Fig. Id). Still on that site, the NIR interception efficiency signal is on the order of the noise signal and then not 
reported. For the bush/grassland, interception efficiencies start around 0.15 and 0.08 in the PAR and NIR, 
respectively, then increase monotically until October (Fig. la). On the grassland site, the time profiles are different 
since vegetation only appeared the first week of August (Fig. lb). From this date, the herbs developed rapidly, the 
interception efficiencies associated increasing more than for the bush/grassland. The reason is that the branches of the 
bushes act as a buffer, masking partially the leaves growth. Bush/grassland and grassland stop similarly their 
photosynthetic activity the first week of October (Fig. la,b). The interception efficiencies are maximum at this time 
with values of 0.6 and 0.48 for the bush/grassland against 0.5 and 038 for the grassland, in the PAR and NIR, 
respectively. For the millet crop, interception efficiencies show a bowl shape (Fig. lc). The maximum values are 
rather low, 0.23 in the PAR against 0.15 in the NIR, and are reached relatively early (one month before the natural 
vegetations). For the bushland, the PAR interception efficiency does not go beyond 0.1 and is phased in time with the 
variability of the one of the bush/grassland (Fig. Id). Over all canopies, the interception efficiency is larger in PAR 
than in NIR, and this difference increases slightly with the interception. This phenomenon is caused by the optical 
properties of the leaves since the leaves are more opaque to the PAR than to NIR. 
In this study, interception and absorption efficiencies remain always close which is a characteristic of sparse canopies 
over bright bare soils. The loss of radiation by reflectance is balanced by the gain of the radiation transmitted to 
canopy then reflected by the soil background of the canopy toward the vegetation (see Eq. (2)). Finally in that case, 
interception efficiency can replace the absorption efficiency. 
Canopies albedo variabilities show much less dynamic than the time profiles of interception efficiencies. In fact, the 
soil albedos obtained are high and occult most of the vegetation changes. For instance, bush/grassland and grassland 
albedo values are rather similar and oscillate around 0.18 and 0.3, in the PAR and NIR, respectively. Note also that the 
NIR albedo is high for the bushland (Fig. 2d) and millet crop (Fig. 2c), about 0.4 and 0.5 respectively. 
More information seems available for the NDVI for the bush/grassland and grassland, with variations between 0.25 to 
0.5 (Fig. 2a and 3b). On the contrary, the NDVI looks like stationary over the millet crop (value around 0.4) and 
bushland (value around 0.35) sites where the soil brightness effect is more perturbing (Fig. 2c,d). The LAJ curves 
fitted to field measurements using logistic type functions show similar amplitude and phase with the time profiles of 
the interception efficiencies (Fig. la,b,c,d). As expected, interception efficiency offers the possibility to retrieve the 
structure parameter LAI. 
4. MODELING PAR INTERCEPTION FROM LOCAL TO SITE SCALE 
Canopy scale - In the Sahelian region, the vegetation can be schemed by two distinct layers. The upper layer 
represents the perennial canopy of trees or shrubs while the lower layer represents the annual grassland. These layers 
may combine or not together (see Figure 3). A different radiative transfer model is applied for each layer. The SAIL 
model (Verhoef, 1984) is considered for the herb layer, here approximated by a vegetation continuum, while the 
cylinders model (Begue, 1992; Begue et al., 1994) simulates the woody layer and the millet crop, which are schemed, 
respectively, as random and regular distributions of vegetation clumps. The inputs of the SAIL code are the LAI and 
the structural/optical properties of the leaves, the albedo of the soil background, and the geometry of the incoming 
radiation. The cylinders model is based on the geometrical optics theory applied to porous cylinders. The porosity is a 
function of the LAI, the foliage geometry, the mean transmittance of the leaves, and the dimensions of the cylinders. 
Both models have already been validated in various conditions. A coupling of these models allows to simulate a large 
range of canopies. It assumes simply that the incoming PAR for the grass layer comes from the transmitted PAR by 
the woody layer. Here, the SAIL code is validated over the grassland, and the cylinders model on the degraded 
bushland, on the bush/grassland before the herbs developed, and on the millet crop. The conjonction of both model, 
so-called mixture model, will be tested against the bush/grassland. 
In addition to the radiative and biological measurements described before, optical properties of the phytoelements 
were measured with a SE590, equipped with an integrating sphere (see van Leeuwen et al., 1993). The porosity of the 
bushes decreases along the season from 0.7 to 0.2 while the porosity of the millet clump is quite constant, around a 
value of 0.45. The porosity of the wood alone is about 0.9. Over a regularly clumped canopy as millet crop, the 
cylinders model works during the growing season and overestimates lightly the measurements during the senescent 
phase (Figure 4a) (the lack of frequent biological measurements at this time may explain this feature). Over randomly 
clumped canopies, the models fits the data well (Figure 4b). 
SAIL model simulations are carried out for 3 scenarii which are a grass canopy only, a forb only, and a canopy 
mixture of grass and forb (43% and 57% of dry matter respectively). To each case are associated a 'specific area' and a 
LIDF (Leaf Inclined Distribution Function) of the herbs. The results of comparison with measurements are shown in 
Figure 4c with dotted curves for pure herbs and symbols for mixture. The canopy mixture model fits the data well 
until the 1st week od September. From this date, the LAI is overestimated and by consequent the specific area of the