724 
1 
1 
Figure 1. Concept of NDVI and PVI (a) and isolines (b) of different vegetation densities. 
the soil line (Fig. lb). To minimize these discrepancies, Huete (1988) modified the NDVI by shifting the converging 
origin along the soil line and obtained a soil adjusted vegetation index (SAVT): 
SAVI = ( p NIR - p«d) / ( p NIR + p rcd + L) ( 1+L) , (3) 
where L was fixed at 0.5 as an optimal value. The SAVI assumed that the soil line is a 1:1 line, which is not the case 
in practice (Fig. lb). Baret et al. (1989) and Baret and Guyot (1991) developed a transformed SAVI (TSAVI) by taking 
into account the soil line slope (s) and intercept (i): 
TSAVI = [v ( p MR - j p red - 01 Ks Psir + Pred - si + X(l+s 2 )], (4) 
where X is a factor adjusted so as to minimize the soil background effect. The improvement of the TSAVI over the 
SAVI was to take the soil line slope (s) and intercept (0 into account, whereas the SAVI assumed them to be 1 and 
0, respectively. Major et al. (1990) took a step further to model the vegetation isoline by using the ratio b/a as the soil 
adjustment factor, with 'b’ as the intercept and ’a’ as slope of each isoline. They obtained a second version of the SAVI: 
SAVT = p NIR / (p red + b/a) . (5) 
The SAVT, does not have an empirical adjustment factor for each isoline, but it contains the leaf area index (LAI) 
parameter in the ’a’ and ’b’ modeling, which is usually the target parameter being retrieved in remote sensing studies. 
These versions of the SAVI either utilized a constant soil adjustment factor (SAVI and TSAVI) or contained 
variable (LAI) not directly measurable from remote sensing measurements (SAVT). Qi et al. (1994) further modified 
the SAVI by modeling the soil adjustment factor L for each isoline as a function of reflectances in red and NIR regions 
and obtained a modified SAVI (MSAVI): 
MSAVI = {( 2 p KIR + l) 2 - ft (2 p NIR + l) 2 - 8(p NIR - p r J] } / 2 (6) 
to reduce the soil effect, while restoring the vegetation sensitivity lost by the SAVI . 
To reduce atmospheric effect on vegetation indices, Kaufman and Tanre(1992) utilized the reflectance 
differences between the red and blue bands and developed an atmospherically resistant vegetation index (ARVI): 
ARVI = NDVI = ( p N , IR - p rh ) / ( p N1R + p rb ), (7) 
where 
Prb = Pred - Y ( Pbluc - Pred ) > (8) 
and the optimal value for y is 1.0. The ARVI reduces the atmospheric effect with the use of reflectances in blue band, 
but it did not take soil effect into account. To reduce both atmosphere and soil background effects simultaneously, they 
also suggested to insert p rb function into SAVI to yield a soil and atmosphere resistant vegetation index (SARVI): 
SARVI = ( p NIR - p rb ) / ( p NIR + p rb + L) (1 + L) . № 
As mentioned before, the use of constant L buffers the vegetation sensitivity and, furthermore, it may not compensate 
the soil noise increases induced by the use of blue band. To overcome the buffering problems due to the constant L 
factor in the SARVI and take the advantages of p rb being insensitive to atmosphere conditions, we adapt the p rb function 
in the MSAVI equation (6) and propose an atmosphere-soil-vegetation index (ASVI): 
ASVI = {( 2 p NIR + l) 2 - ft (2 p NIR + l) 2 - 8(p NIR - Prb )] } / 2 (10) 
This index needs to be tested against other indices to explore its vegetation sensitivity, soil noise and atmosphenc 
effects. 
Pinty and Verstraete (1992) took another approach and proposed a non-linear index called global environmental 
monitoring index (GEMI): 
GEMI = p ( 1 - 0.25 n ) - ( Pred - 0.125) / ( 1 - p red ), ( U) 
where 
R = C2( p 2 X!R - p 2 red ) + 1.5 p N1R + 0.5 p red ] / ( p NIR + p red + 0.5). ( 12) 
The GEMI was meant to reduce atmospheric effects, but its vegetation and soil sensitivity remain to be investigated. 
Since some vegetation indices are functionally equivalent, only eight vegetation indices were selected for tb e 
sensitivity analysis in this study. They are NDVI, SAVE PVI, MSAVI, ARVI, SARVI, ASVI, and GEMI.