154 
Table 1: Error budget for Mt. Pinatubo aerosol correction 
Case 
NDVI 
NDVI (before 
Pinatubo) 
top of the 
atmosphere 
NDVI T.O.A. 
x s = 0.3 
NDVI T.O.A. 
t s = 0.6 
Bare Soil 
pi =0.19 , p2=0.22 
0.073 
-0.043 
-0.048 
-0.051 
Deciduous forest 
pl=0.02 , p2=0.36 
0.846 
0.600 
0.547 
0.493 
Non Lambertianity 
- 
- 
±0.01 
±0.02 
Bare Soil 
At s = ±0.05 
- 
“ 
±0.0015 
±0.001 
Deciduous forest 
At s = ±0.05 
’ 
±0.009 
±0.008 
Bare Soil 
Wav. Exp.= 0. 
" 
" 
-0.041 
" 
Deciduous forest 
Wav. Exp.= 0. 
0.550 
5. RESULTS AND DISCUSSION 
A preliminary validation has been undertaken, comparing corrected NDVI with the uncorrected corrected 
for a sample area of 20 by 20 degrees over east Africa centered on the equator. A nine days composite of the 
NDVI has been computed for 1989-1992 before and after the stratospheric aerosol layer formation. Figure 8 
shows the histogram of NDVI before the effect is detectable. The distribution of NDVI for 1991 is comparable 
to the one observed in 1989/1990 especially in the higher range of NDVI [0.4-0.6], Figure 9 gives the same 
distribution for a 9 days composite at the beginning of August. The uncorrected data of 1991 are lower by 0.10- 
0.15 NDVI units than the 1989/1990 values, the corrected values are falling exactly in the range of the 
previously observed values. The same comparison is conducted in 1992, Figure 10, in that case the effect of 
stratospheric aerosol is lower, 0.05 NDVI unit, and the corrected values fit better the previous year for higher 
NDVI. The agreement between corrected and previous year NDVI is not so good in the intermediate range of 
NDVI values [0.15-0.4]. This can be due to a problem with the correction or due to a change in NDVI values. 
As shown by Figure 8 , only the higher range and the lower range of NDVI is expected to be stable over the year 
because it covers area where man activities is less important (Desert, Evergreen Forest). 
Comprehensive validation of the correction is problematic in that it requires both an independent 
assessment of stratospheric and tropospheric aerosols as well as vegetation state. We are highly dependent on the 
quality of the aerosol retrieval particularly with respect to the spatial and temporal coverage. However, we are 
confident in the radiative transfer computation that have been made. 
The effects of the Pinatubo eruption on the NDVI will need correction until the magnitude of the effect 
becomes part of the background atmospheric noise in the NDVI signal. Given the rate of decay we estimate this 
to be approximately towards the begin of 1993. Clearly in the long term, the research community must move 
towards an operational and comprehensive atmospheric correction including all aerosol effects (Vermote et al, 
1993). Until this has been accomplished, we are obliged to apply corrections as presented here. The present 
study must be seen in this context and has raised several issues relevant to the operational atmospheric 
correction planned for the EOS generation sensors. 
The objective of this study was to develop a quick and effective correction procedure that could be 
applied operationally to the AVHRR data for the generation of NDVI. Within the limits of our validation 
exercise this has been achieved. However, a complete characterization of stratospheric aerosols is an area for 
future research and is beyond the scope of this present activity. Two "corrected" global vegetation index data set 
are or will be available soon, the 1 degree x 1 degree FAZIR data set (Los et al, 1994) produced using GIMMS 
Pinatubo corrected data set, and the 8 km NOAA GVI (Vermote and Kaufman, 1993).