3.4 On atmospheric corrections 
One should emphasize that the definition of a reliable method of atmospheric correction and a careful 
assessment of associated accuracies are prerequisites for an operational use on either global or regional scales 
of the methods described in the preceding paragraphs. Molecular scattering and ozone absorption effects may 
be corrected relatively easily with climatologic data. However, water vapor and aerosol amounts are highly 
variable in space and time, and, consequently, more difficult to account for (see Tanré et al., 1992, for a 
detailed review of atmospheric correction methods of AVHRR data). Water vapor absorption is of major 
importance, affecting the apparent reflectance in the near infrared by a reduction of 10-30 % in relative value, 
depending on the amount of water vapor and the direction of illumination and observation. Global maps of 
atmospheric water vapor content at sufficient spatial and temporal resolutions are thus necessary. They can be 
obtained from General Circulation Models (GCM) using as an input radiosounding measurements. However, 
the lack of data over some regions like Sahel or tropical forests results sometimes in poor accuracy. The effect 
of aerosols is also quite substancial. In the visible channel, the effect of moderate haze is of the same 
magnitude as the reflectance of vegetated surfaces and the effect of dense haze, dust outbreaks, or smoke 
episodes can mask completely the surface properties. The required inputs for the correction of aerosol effects 
are often unavailable. This situation is, however, expected to evolve favorably in the near future, since new 
measurements of aerosol characteristics from ground-based Sun-photometer network in selected climatic zones 
(Holben et al., 1994), and from satellite instruments such as POLDER/ADEOS will become available soon. 
4 - APPLICATION OF A BIDIRECTIONAL REFLECTANCE MODEL TO A TOA REFLECTANCE 
TIME SERIES 
Ba et al. (1994) have applied an alternative approach, still based on a reduction of directional effects through 
modelling, but which consists in a normalization (or correction) of directional effects on the Top Of 
Atmosphere reflectances, rather than on atmospherically corrected reflectances. The parameters of a temporal 
and directional TOA reflectance model, of the form 
*2d№.0v.«-WOG(0,.e v .<to, (6) 
are adjusted against the observed TOA reflectance time series over the whole annual vegetation cycle. The 
basic assumption implied by Eq. (6) is that the temporal and directional behaviour of TOA reflectances are 
uncoupled. Rq is the time averaged TOA reflectance of the site. F(t) describes the temporal evolution of the 
reflectances corrected from directional effects, and is represented by a 10-parameter Fourier time series. 
G(0^ ,9 V ,<p) is a 2-parameter empirical analytical function and represents the directional behaviour of the 
coupled surface-atmosphere system. 
Ba et al. (1994) have applied this scheme to AVHRR data acquired in 1987 over the Konza Prairie 
during the FIFE experiment In Figure 7 are shown the observed TOA reflectances, the modelled R(t), and 
observed reflectance values corrected from directional effects (that is, observed R divided by G(9,,0 y ,<|>)), in 
the visible and near infrared. Similar quantities are also shown for the NDVT. Figure 6 shows that the day-to- 
day fluctuations in the original dataset (represented by crosses) are very much reduced when R is normalized 
by the bidirectional function G(9j,0 v ,<{)). The standard deviation 8 of the residual errors is rather small, 0.007 
in the visible, and 0.014 in the near infrared, and anyway smaller than the values of 8 obtained in § 3.2 and 3.3 
(Table 1). This may perhaps be explained by the fact that the number of free parameters is rather high in this 
method. In the near infrared, RqF clearly presents a relative maximum during spring and summer, a hardly 
detectable feature in the "raw" time series. In the visible, the RoF variations indicate that the contrast between 
summer and winter is artificially enhanced by directional effects; when these effects are eliminated, the 
reflectance minimum between Julian days 140 and 250, although still present, is not prominent anymore. The 
resulting NDVI, computed as the ratio of the difference and the sum of RoF (near infrared) and RoF (visible), 
exhibits a relative minimum around Julian day 225. This minimum, which cannot be objectively identified in 
the "raw" NDVI data, may be real and associated with vegetation stress, a conjecture supported by local 
rainfall data (not shown here), which indicate that day 225 has been preceded by a period of lack of rain. 
29