You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Mesures physiques et signatures en télédétection

J. Qi 1 , Y. Kerr 2 , and A. Chehbouni 3
1. USDA-ARS Water Conservation Laboratory, 4331 E. Broadway, Phoenix 85040, Arizona, USA;
2. LERTS-CNES-CNRS, 18 Avenue Edouard-Belin, 31055 Toulouse Cedex, France;
3. Jet Propulsion Laboratory, 4800 Oak Grove Dr., Pasadena, California 91109, USA.
Several vegetation indices have been developed by ratioing or linear combinations of different spectral bands to
characterize vegetation status with remote sensing measurements. These indices are primarily sensitive, as they are
meant to be, to vegetation, but they are also sensitive to non-vegetation related factors such as soil background
variations, atmospheric conditions, and sensor viewing geometries. Consequently, there is a need to quantify these effects
when utilizing these vegetation indices. In this paper, a sensitivity analysis of existing vegetation indices was made using
both modeled spectral data and ground-based remote sensing measurements. The analysis was made with respect to soil
background, atmosphere, and sensor viewing geometries. General criteria to evaluate vegetation indices were established,
and based on these criteria, a new index was developed. The new index was developed by adapting the reflectance in
blue region for the atmospheric corrections in the modified soil adjusted vegetation index (MSAVT). The resultant
atmosphere-soil-vegetation index (ASVI) was compared with other indices and it increased vegetation sensitivity while
further reduced soil and atmospheric effects. However, differences found among these vegetation indices may not be
statistically significant, and further studies are, therefore, needed with satellite- and ground based data.
KEY WORDS: Vegetation index, remote sensing, soil, atmosphere, view angle, sensitivity
To enhance vegetation signals with remote sensing measurements, several vegetation indices (VI) have been developed.
Most of them are ratio or linear combination of reflectances in the near-infrared (NIR) and red spectral regions. These
indices vary primarily with the vegetation changes. However, they are also sensitive to non-vegetation related external
factors such as soil background variations (Huete,1989), atmosphere conditions (Kaufman, 1989), and sensor viewing
geometry (Deering, 1989). To minimize these external effects, efforts have been made to either incorporate an
adjustment factor (Huete, 1988) or adapt an additional spectral band (Kaufman and Tanre, 1992) into the normalized
difference vegetation index (NDVI), or develop a non-linear vegetation index equation (Pinty and Verstraete, 1992).
Although these works are plausible with respect to soil or atmosphere noise reduction, some general questions also arise
regarding the consequences of introducing an adjustment factor or an additional spectral band;
1) What are the consequences of using an adjustment factor for soil noise reduction ? Would it jeopardize the
vegetation sensitivity or worsen the atmospheric effects ?;
2) What are the consequences of using an additional spectral band ? Would it increase the sensitivity to soils ?;
3) Which vegetation index is more appropriate to use when practically all external factors co-exist? Can these
effects be taken into account in a single vegetation index equation?
In this study, a sensitivity analysis was conducted, considering all possible external factors (vegetation, soil, atmosphere,
and sensor viewing geometry) to address these questions, and then, an index was developed that minimized both soil
and atmosphere effects while op timizin g its vegetation sensitivity.
Two of the most commonly used Vis are the normalized difference vegetation index (NDVT):
NDVI = ( P NIR - P red ) / ( Pnir + Pred ) (D
and perpendicular vegetation index (PVI):
PVI = a p NIR - P p rcd , (2)
where p is reflectances in NIR or red band. The a and (5 are soil line parameters. The NDVI is a measure of the slopes
of the vegetated isolines (Fig. la) while the PVI is a measure of the distances of the vegetation pixel to the soil line.
The NDVI was based on the concept that all isolines converge at the origin, while PVI assumed these isolines are
parallel to the soil line (Fig. la). However, in practice, the isolines are neither converging at the origin nor parallel to