Vegetation indices may be separated in two categories: the intrinsic indices (such as the simple ratio 
and the NDVI), which do not involve any external factor other than the measured spectral reflectances; and 
the soil-line related indices, which include soil-line parameters, such as the Soil Adjusted Vegetation Index or 
SAVI (Huete, 1988), defined as SAVI= 1.5(NIR-R)/(NIR + R+0.5), and the Transformed SAVI or TSAVI 
(Baret and Guyot, 1991), etc), defined as TSAVI=a(NIR-aR-b)/(R+a(NIR-b)+0.08(l +a 2 )), where NIR and 
R are respectively the reflectances in the near infrared and red channels, and (a,b) the soil line coefficients. 
Figure 1 displays those indices as a function of the "plant cover" parameter [l-exp(K*LAI)], where 
K is an extinction coefficient related to the leaf angle distribution. It has been estimated using an ellipsoidal 
inclination angle distribution (Campbell, 1986), for mean leaf angle varying from a planophile (25 ’) to an 
erectophile (65°) canopy. LAI varies from 0.1 to 8 . Vegetation Indices have been calculated for the 26 
backgrounds in each canopy geometry . 
3.2 Results 
The NDVI (Figure la) demonstrates a very large range of values with varying background, especially 
for low vegetation amount. For LAI<0.1, the reflected signal may be considered as equivalent to the soil 
signature. With an intermediate amount of vegetation (0.5<LAI<3) the large range of possible values would 
produce very large errors when trying to retrieve the plant cover from the NDVI values. At large LAI, the 
NDVI reaches a saturation level, which also makes the retrieval of cover difficult. Clearly NDVI will give poor 
information about a vegetation canopy when no detail about the soil background is known. 
Soil-related indices, SAVI and TSAVI (Figure lb-d), perform better than NDVI. However some 
problems remain, due to the use of a single universal soil line for all soil types. Baret et al. (1993) have shown 
that in fact, when the variations of reflectance with moisture and roughness are considered, each soil has its 
own soil line. If for each soil, the appropriate soil line is used, it is clear that the TSAVI will perform very 
well, but this assumes that the soil and its characteristics are well known. Figure 2 presents red and near- 
infrared reflectances of the 26 soils for nadir view and two solar zenith angles: 30 and 60'. The graph indicates 
that the soils do not all follow the same line, but may be broadly separated into two distinct lines 
corresponding to two groups of soils: first, clayey and sandy soils, (including pozzolana and pebbles), which 
are mineral based soils, and, second, peaty soils. The intercept of both line is very close to the origin, but the 
slope for the peaty soils line is about twice the other, resulting from larger reflectances in the near-infrared 
region. 
When using a single soil line for all soils (averaging all the 26 singles lines), the TSAVI, although 
better than NDVI, does not greatly improve the information about the vegetation given by the index (Figure 
lc). In this case, the SAVI (Figure lb), which uses coefficients previously adjusted from a global soil line 
(Huete, 1988), produces a slightly better result, but still with quite large variations. Better results may be 
obtained when using the two soil lines (Figure Id), where the dispersion of the data is smaller for relatively 
low LAI. But here the problem appears at the larger cover values (LAI >2) where the index produces an 
artificial variation, with a different value of saturation level for each group of soils. This divergence is not 
realistic, although quite good performance is obtained when considering each group of soil separately. 
These results are summarized in Figure 3, which shows the standard deviation in estimated cover with 
each vegetation index for the different soils, indicating the dispersion of values from the average, and therefore 
how precisely the plant characteristics could be retrieved from the indices. 
3.3 Choice of an index 
The previous tested indices do not minimize soil effects in a satisfactory manner. It seems difficult 
to find a global index, in a simple formulation, which would reduce soil influences both at low and high cover. 
For one of the objectives is to keep an "easy-to-use" index formulation, we have chosen the TSAVI, having 
simplified the formulation by assuming a=l and b= 0 , i.e. as: 
S-TSAVI = (NIR-R)/(NIR + R+0.16), where S-TSAVI means Simplified TSAVI.