METHODOLOGY 
(b) 
nsities. 
fting the converging 
(3) 
which is not the case 
1 (TSAVI) by taking 
1. (4) 
the TSAVI over the 
ed them to be 1 and 
e ratio b/a as the soil 
version of the SAVI: 
(5) 
:af area index (LAI) 
note sensing studies. 
TSAVI) or contained 
>94) further modified 
red and NIR regions 
( 6 ) 
ized the reflectance 
tion index (ARVI): 
( 7 ) 
( 8 ) 
ctances in blue band, 
simultaneously, they 
n index (SARVI): 
(9) 
may not compensate 
lue to the constant L 
adapt the p, b function 
2 ( 10 ) 
>ise and atmospheric 
global environmental 
(ID 
+ 0.5). (12) 
a to be investigated, 
were selected for the 
nd GEMI. 
Data Descriptions 
Spectral reflectances of forty soils were obtained from results of Condit (1970) at the wavelengths centered at 480nm, 
540nm. 65Qnm, and 81Qnm, which corresponded to the blue, green, red, and NIR spectral regions, respectively. The 
bidirectional reflectances were simulated with SAIL (Scattering by Arbitrarily Inclined Leaves) model by Verhoef 
(1984). The leaf area index (LAI) increased from 0 to 4. The solar zenith angle was set at 30° while the relative azimuth 
was 0°. The 5S (Tanre et al., 1990) code was used to simulate the atmosphere. The atmospheric condition was US 
standard with continental aerosol model. 
Sensitivities of Vegetation Indices 
To evaluate vegetation indices, one needs to consider several aspects: vegetation changes, soil background variadons, 
atmospheric conditions, as well as the sensor viewing geometries (Qi, 1993) : 
Vegetation sensitivity : The vegetation sensitivity deserves the highest priority because a vegetation index was 
meant to enhance the vegetation signals not found in individual spectral bands. Consequently, its sensitivity to subtle 
variations in vegetation density should be used in evaluation of vegetation indices. 
Soil background noises : Reduction on soil noises by vegetation indices deserves the second priority in 
vegetation index evaluations. This is especially important in global change studies. In regions such as arid and semiarid 
areas, where vegetation cover is sparse, the soil background is the major contributor to remote sensing measurement. 
Therefore, a vegetation index should reduce the soil noise to a minimum. 
Atmosphere conditions : Atmospheric fluctuation would alter the remotely sensed signals reflected from the 
earth surface and passed through the atmosphere before received by a sensor. The variation in atmospheric conditions 
is rapid and. therefore, causes severe problems for high temporal frequency measurements. Consequently, a vegetation 
index should be as insensitive as possible to the atmospheric condition variations in order to obtain a consistent measure 
of vegetation status. 
View angles : Some vegetation indices result in a symmetric pattern about the nadir, while others tend to bias 
toward the forward directions. Although it is not necessary to have an index independent of view angles, it is desirable 
that the index has a symmetric behavior about the nadir, so that only the view magnitude needs to be corrected. 
General Criteria 
Practically, all of these external factors (vegetation, soils, and atmosphere) co-exist; therefore, they should be considered 
together in evaluating vegetation indices. Let vegetation signal be v, soil noise be s, and the atmospheric noise be a. 
Then the total observed quantity ,Y (vegetation index in this study), by a remote sensor would be a function of v, 5, 
and a: 
Y = f ( v, s, a, x) , (13) 
where x is related to other external factors such as radiometric calibration and geometric registrations. Any variation 
in vegetation (v), or soils (s), or atmospheric conditions (a) would result in the change in Y, which could be 
mathematically expressed as: 
d Y = dY/dv dv + dY/ds dr + BY /da da + dY/d.t dx . (14) 
We are interested in the signal changes due to vegetation variations rather than due to other three terms. Therefore, a 
’good’ vegetation index should result in a maximum dY/dv, a minimum dY/ds, dY/da. and dY/dx, because the later three 
are regarded as noise while the first one is the vegetation signal to be enhanced. In this study, we only consider the fust 
two sources of noise, which are the major external influencing factors in space remote sensing. The noise from soil 
background (dY/ds) and atmosphere (dY/da ) are normally additive and, therefore, the total noise N should be the sum 
of the two when considering soil and atmosphere only: 
N = dY/ds + dY/da , (15) 
and the signal (S) should be: 
S = dY/dv. (16) 
A general criterion to evaluate a vegetation index is to examine its signal to noise ratio (S/N). Decreases in S or 
increases in N would depress the S/N value. Therefore, based on the absolute S/N value, vegetation indices can be 
compared and evaluated. For easy analysis, the fust three terms in equation (14) are compared for different vegetation 
“dices instead of only S/N since the dynamic ranges of the selected vegetation indices differ. 
725