BACKWARD ANGLE (deg) FORWARO REFLECTANCE (per cent) and NOVI*100 
Fipiire 1 : Top: Variation of visible (o), near infrared (x) and 
NDVI (*) calculated from cloud-screened data in July- 
August 1986. Bottom: variation of solar zenith (o) and off- 
nadir viewing (*) angles during July-August 1986. (from 
Gutman, 1987) 
Figure 2 : Time series of TOA reflectances (full curves) and 
surface reflectances (dotted curves), (a) and (b): visible 
band; (c) and (d): near infrared band, bars stand for the 
standard deviation of errors of the atmospheric correction, 
(from Roujean el al., 1992a). 
time profiles of ref 
retrieved paramete 
Data com 
emphasize that apf 
satellite optical se 
channels, such as 1 
elevation during 
MERIS/ENVIS AT 
will require simi 
heliosynchronous 
additional capacit; 
techniques, in an ii 
2 - MAXIMUM V 
JULIAN DAY 
erroneous estimates of water vapor and aerosols) can result in extreme conditions in errors of the retrieved 
reflectances commensurable with the actual value of surface reflectances (Tanré et al., 1992). Moreover, the 
variations of Sun-target-sensor configurations between consecutive measurements can induce large 
fluctuations of observed reflectances, in phase with the orbit cycle of AVHRR (Gutman, 1987; Fig. 1). These 
effects may originate from the atmosphere or from the surface. The dominant effect seems to be that of the 
surface, however; atmospheric corrections do not alter significantly the saw-tooth like pattern which is the 
characteristic signature of directional effects (Roujean et al., 1992a; Fig.2). The peak-to-peak signal variations 
may be very large, about 100 % and 50 % of the average reflectance in the visible and in the near infrared 
respectively (Fig. 2). These findings are consistent with ground truth measurements of surface bidirectional 
effects (e.g., Kriebel, 1978; Kimes, 1983). 
There is a qualitative difference between atmospheric corrections and surface bidirectional reflectance 
corrections. While atmospheric corrections are generally performed with ancillary data from various sources 
(climatology, other sensors data), it is difficult to conceive a correction of surface bidirectional effects which 
would not extract the directional information from the reflectance data themselves. The directional reflectance 
signature of a surface depends in a complicated way of the optical and structural properties of the surface, and 
is therefore hardly predictable from external sources. This is the primary reason why some kind of data 
compositing over a given period of time, and associated losses of temporal resolution, seem unavoidable. A 
second advantage of data compositing is that it statistically removes the high frequency temporal variations 
due to inaccuracies in the atmospheric correction procedure. 
The aim of this paper is to briefly review existing methods of data composition, and discuss recent 
work where data composition is conceived as a fitting procedure between a time series of reflectances, 
corrected or not from atmospheric effects, during a given period (called here a period of composition), and a 
parameterized model of bidirectional reflectance. The original time series of reflectances is then replaced by 
The data composit 
technique (Tarple; 
AVHRR data wh 
difference betweei 
composition may \ 
cover conditions. ' 
vegetation monitoi 
Vegetation Index ( 
this method is that 
the Sun-target or tt 
of the AVHRR d: 
effects. 
The adeqi 
and Hendcrson-Se 
an erroneous meat 
(1991) has notice- 
direction, a conclu 
South-East of Fran 
angles obtained bj 
during the month c 
obvious. Moreovei 
(Fig. 3). It is appt 
results in errors sir 
NADIR ANC 
22 - 
20 - 
V7\ 10-QAY 
Other mei 
viewing geometry 
22