1107 
Figure 4 suggests a simple approach to quantitatively evaluate the performance of a vegetation index as a 
predictor of a. This approach is based on the concept of signal to noise ratio (S/N). We propose that the 'signal' 
of interest over a limited range of fractional covers [a m , is given by the difference between the average 
index value for a maximum and a minimum fractional cover. Consequently,, the signal may be evaluated as : 
Sa(VD = VI(a M )-VI(CT m ) 
where a m and represent the minimum and the maximum fractional vegetation covers considered 
respectively. We are obviously interested in selecting an index as insensitive as possible to a hence with a high 
'signal' value, but also as insensitive to the noise as possible. Referring back to Figure 4, it is seen that an ideal 
index would take on identical values irrespective of the soil type, so the maximum and minimum curves would 
superpose. Accordingly, we take the area between the maximum and the minimum curves as our measure of the 
noise: 
Nct(VI) = J [max(VI(oi)) - min (Vl(cri))] do 
°m 
A criterion to evaluate the usefuleness of these vegetation indices as predictors of the a is therefore provided by 
the following signal to noise ratio: 
No(VI) 
Table 2 shows the value of this signal to noise ratio for the indices, for three values of the LAI, and for two range 
of a: 0-50% and 50-100%, since most ecosystems will experience a limited range of variation of theit a. It can 
be seen that GEMI and SAVI/MSAVI are in general both significantly better than NDVI and WDVL Over thin 
canopies (low LAI), the signal to noise ratio remains low and the indices are less differentiated. At high values 
of LAI, the Signal to noise ratio of NDVI is almost 3 to 4 times lower than that of GEMI. 
[0-50%] 
Index 
LAI=1 
LAI=3 
LAI=5 
[50-100%] 
Index 
LAI=1 
LAI=3 
LAI=5 
GEMI 
2.59 
5.41 
6.62 
GEMI 
130 
5.38 
13.82 
NDVI 
1.84 
2.34 
2.38 
NDVI 
2.34 
2.55 
3.69 
SAVI 
3.29 
5.03 
5.51 
SAVI 
3.35 
8.59 
10.82 
MSA VI 
3.38 
4.92 
5.27 
MSA VI 
4.12 
8.66 
8.62 
WDVI 
1.86 
3.77 
4.70 
WDVI 
1.46 
4.58 
938 
Table 2 : Signal to noise ratios for various indices as estimator of a 
4-EVALUATION OF INDICES USING SIMULATED TOA SPECTRAL DATA 
Vegetation indices are most often computed on the basis of satellite remote sensing data outside of the 
atmosphere. Since the reflectance on which they are based are sensitive to atmospheric composition, it is 
important to also evaluate the performance opf the indices at the ‘top of the atmosphere' (TOA). To this end, we 
have used the 5S radiation transfer code (Taint et a!., 1990) to compute the values of the spectral albedos TOA 
for the various heterogeneous targets described above, and for three types of atmosphere : a Rayleigh atmosphere 
and two mid-latitude summer continental atmosphere with aerosol loads corresponding to visibilities of 23 and 5 
km, respectively. It is desirable to define again a signal to noise ratio to see how each index performs with 
respect to atmospheric contamination, in this case, the signal will be defined by the difference between the 
surface value of the index at a given a and its value for a lower a, for a specified type of soil and a given value 
of LAI (5): 
S*atm(VD = VI(a M ) - VI(a m ) 
Since it is uncertainty about the state of the atmosphere which translates as noise in the vegetation index 
values, and since atmospheric aerosols represent the largest source of errors, we have attempted to estimate to 
what extent vegetation indices were sensitive to that particular constituent The noise will accordingly be 
defined as