near infrared (x) and 
ened data in July- 
r zenith (o) and off- 
August 1986. (from 
;es (full curves) and 
» and (b): visible 
bars stand for the 
lospheric correction. 
170 180 190 200 
of the retrieved 
). Moreover, the 
n induce large 
7; Fig. 1). These 
to be that of the 
sm which is the 
signal variations 
die near infrared 
ace bidirectional 
ional reflectance 
i various sources 
tal effects which 
ional reflectance 
' the surface, and 
Tie kind of data 
i unavoidable. A 
iporal variations 
id discuss recent 
of reflectances, 
t position), and a 
then replaced by 
time profiles of reflectance corrected from angular effects, the latter reflectance being in turn derived from the 
retrieved parameters of the regression. 
Data composition techniques have been applied so far to AVHRR data. One should however 
emphasize that appropriate techniques of data composition have also to be prepared for the next generation of 
satellite optical sensors. Even data acquired by future geostationary satellites with visible and near infrared 
channels, such as Meteosat Second Generation, will need to be corrected from the effect of variations of Sun 
elevation during a vegetation cycle. Future heliosynchronous sensors, such as MODIS/EOS, 
MERIS/ENVISAT, VEGETATION/SPOT4, will have the same basic geometric configuration as AVHRR and 
will require similar compositing as AVHRR for vegetation monitoring applications. Other future 
heliosynchronous sensors, such as POLDER/ADEOS, MISR/EOS, and ATSR/ERS2 will benefit of an 
additional capacity of directional measurements, resulting in fine, through the use of data composition 
techniques, in an improved temporal resolution. 
2 - MAXIMUM VALUE COMPOSITE AND OTHER METHODS 
The data compositing method which has been most widely applied is the Maximum Value Composite (MVC) 
technique (Tarpley et al., 1984), whose principle is to select, during a given period of composition, the 
AVHRR data which maximizes the Normalized Differential Vegetation Index (NDVI), defined as the 
difference between near infrared and visible reflectances, divided by the sum of the two. The period of 
composition may vary from 1 week (Tarpley et al., 1984) to 1 month (Justice et al., 1985) depending on cloud 
cover conditions. The MVC technique has been a standard routine for most uses of AVHRR in the context of 
vegetation monitoring (see the references mentioned in the introduction), and has been the basis of a Global 
Vegetation Index (GVI) standard product distributed by the NOAA administration. The physical rationale of 
this method is that the presence of clouds, aerosols, water vapour, and the increase of atmospheric paths along 
the Sun-target or target-sensor trajectories, tend to decrease the NDVI (Holben, 1986). Therefore, the selection 
of the AVHRR data maximizing the NDVI is expected to minimize the influence of all these perturbing 
effects. 
The adequacy of the MVC technique for land surface characterization has been questioned by Thomas 
and Henderson-Sellers (1987), and Gutman (1987, 1991). They first have pointed out that this method selects 
an erroneous measurement if all days are cloudy during the selected compositing period. Second, Gutman 
(1991) has noticed that the MVC technique selects preferentially observations in the forward scattering 
direction, a conclusion supported by the data set of Roujean et al. (1992a) over the Valensole plateau in the 
South-East of France. Figure 3, from Gutman (1991), shows a frequency distribution of AVHRR data viewing 
angles obtained by retaining 10-day maximum of NDVI indices on a site of the US Great Plains in Kansas 
during the month of July. As noted by Gutman (1991), a strong bias towards the forward scanning direction is 
obvious. Moreover, extended periods of composition (20- and 30-day) do not qualitatively modify this result 
(Fig. 3). It is apparent in Fig. 3 that the selection procedure does not select a single viewing angle, which 
results in errors since the surface is generally characterized by significant directional effects. 
Fi gure 3 : Frequency distribution of viewing zenith 
angles after 10-, 20-, and 30-day compositing by 
retaining the day with maximum NDVI. (from 
Gutman, 1991). 
Other methods have also been proposed. Duggin et al. (1982) recommend to restrict the range of 
viewing geometry to about +/- 25-30 °, with a corresponding loss of temporal resolution. Viovy et al. (1992) 
23