152 
the Food Early Warning System (FEWS) of the US Agency for International Development showed unusually 
low values of the NDVI for areas where crops were reported to be in good condition. Multidate compositing 
(Holben, 1986) did not appear to reduce this effect. 
This aerosol effect can be clearly understood from our experience in atmospheric effects on satellite 
data. The stratospheric aerosol layer is composed largely of small particles (0.5pm) located at 20km high in the 
atmosphere and therefore not attenuated by water vapor or molecular scattering. A sensitivity analysis was 
performed using the 5S (Tanré et al,1990) radiative code which enables a simulation of the signal observed by a 
satellite in the solar spectrum. By considering a stratospheric aerosol model composed of small particles, the 
NDVI of a typical vegetation cover decreases from 0.5 to 0.4 when the optical thickness at 0.5pm increases 
from 0. to 0.5. This value of optical thickness of 0.5 is in the range of expected values for stratospheric 
aerosols. The simulated decrease in the NDVI was comparable to that observed by the FEWS project. 
In this paper we report an approach to correct NDVI for the stratospheric aerosol effect by using 
AVHRR data itself to derive a quantity proportional to the stratospheric aerosol optical thickness at 0.65pm and 
0.85pm which permits the computation of transmission and intrinsic atmospheric reflectance correction terms. 
These correction terms are then applied to the single channel data prior to computation of the NDVI. 
2. DETERMINATION OF THE LATITUDINAL PROFILE OF STRATOSPHERIC 
OPTICAL DEPTH. 
It has been observed by Stowe (Stowe, 1992) that a few months after volcanic eruption, stratospheric winds 
produce a longitudinally homogeneous aerosol layer. It is however necessary to determine the latitudinal profile 
of the stratospheric aerosols. To determine the latitudinal profile we used the AVHRR data itself. The technique 
involved identifying an area over the ocean where tropospheric aerosol influence is low and where the 
stratospheric aerosol content could be derived using statistics. The area chosen is located in the Pacific Ocean and 
extends from 175° West to 135° West and from 60° South to 60° North. Each day, one AVHRR GAC orbit was 
processed, screened for clouds using CLAVR (Stowe et al,1991), calibrated and composited with 9 other orbits 
by selecting the minimum value in channel 1. The composite technique enable us to better screen cloud as well 
as short term tropospheric aerosol event. The way of compositing orbits with no respect to the longitude of the 
ascending node is justified by the fact that we are looking at longitudinally homogeneous and slowly varying 
optical depth. It's also possible after compositing to interpret physically the signal observed because the NOAA 
satellite has a sun-synchronous orbit so that on a period of 9 days the only variation in geometrical condition is 
due to the precession of the earth. 
After compositing, stratospheric optical thicknesses can be computed in both AVHRR channels by 
substracting the part of the signal due to Rayleigh and background tropospheric aerosol scattering and inverting 
the remaining radiance using a stratospheric aerosol model (King et al, 1984). The consistency of the inversion 
can be checked for different view angles by looking at the signal observed in both channels. Figure 1 compares 
for a view angle of 35° toward back scattering, the signal observed in both channels after correction of a small 
tropospheric background to King's model computation for a optical thickness of 0.4 at 0.55pm for different 
latitudes. The same comparison is showed on Figure 2 but for a view angle of 60°. The radiative transfer 
computation agrees with the observation for two reasons: (a) the optical depth of 0.4 is obtained in both channel 
at the same latitude for the two selected angles, (b) the match between modeled and measured radiance is 
obtained at the same latitude for both angles. It is then possible to confidently invert the observed radiance for 
each composite. Figure 3 shows the optical thickness inverted from the 9 days geometrical composite before and 
after the stratospheric aerosol layer formation. 
For additional quantitative validation we compare the deduced optical depth to the values measured at Mauna Loa 
observatory during the 1991-1992 period. For each 9 nine days composite between 1989 and 1992, we compute 
an average optical depth at 0.55pm for latitude between 19° North and 21° North. For each month, we select the 
minimum value observed out of 4 composites. The period 1989-1990 is used to compute the "average" 
tropospheric background, present in our data, but not observed at Mauna Loa because the observatory is located 
on a mountain of Hawaii at 3000m altitude. The minimum monthly optical depth deduced from our process and 
corrected for 1989-1990 tropospheric background is compared for the 1991-1992 period to the values measured 
daily at Mauna Loa at 0.5pm (Figure 4). 
Before July, stratospheric aerosol background at this latitude is around 0.02. Some variation can be seen in both 
data sets. The variations of Mauna Loa optical depth measurements can be due to some resiludals tropospheric 
aerosol. The variations of AVHRR retrieved optical depth is due to an improper correction of the tropospheric 
aerosol or to noise in the retrieval process. 
The stratospheric aerosol produced by the June 1991 Mt. Pinatubo eruption can been clearly seen in both data 
sets starting in July 1991, the decrease after the maximum that occurs in August 1991 is also shown by both 
data sets and the e folding time is of the order of a year and a half. There are differences between the two that can