See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222741732

Optimization of Soil-Adjusted Vegetation Indices

Article in Remote Sensing of Environment · February 1996

DOI: 10.1016/0034-4257(95)00186-7

CITATIONS READS

2,259 6,567

3 authors, including:

M. D. Steven Baret Frederic

University of Nottingham French National Institute for Agriculture, Food, and Environment (INRAE)
142 PUBLICATIONS 7,463 CITATIONS 559 PUBLICATIONS 38,826 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by M. D. Steven on 29 August 2017.

The user has requested enhancement of the downloaded file.

 ELSEVIER

Optimization of Soil-Adjusted Vegetation

Indices
Genevibve Rondeaux,* Michael Steven,* and Fr6d6ric Baret t

T h e sensitivity of the normalized difference vegetation are also sensitive to three main external factors: solar
index (NDVI) to soil background and atmospheric effects and viewing geometry, soil background, and atmo-
has generated an increasing interest in the development spheric effects. Responses to all three factors are com-
of new indices, such as the soil-adjusted vegetation index plex, intricately coupled, and dependent on surface
(SA VI), transformed soil-adjusted vegetation index (TSA VI), characteristics (Qi et al., 1993). Several new indices,
atmospherically resistant vegetation index (AR VI), global such as SAVI (soil-adjusted vegetation index; Huete,
environment monitoring index (GEMI), modified soil-ad- 1988) and ARVI (atmospherically resistant vegetation
justed vegetation index (MSA VI), which are less sensitive index; Kaufman and Tanr~, 1992) or combinations of
to these external influences. These indices are theoretically both (SARVI; Kaufman and Tanr6, 1992), have been
more reliable than ND VI, although they are not yet widely developed in an attempt to minimize these influences.
used with satellite data. This article focuses on testing The biophysical explanation of the relations between
and comparing the sensitivity of NDVI, SAVI, TSAVI, vegetation indices and observable vegetation phenom-
MSA VI and GEMI to soil background effects. Indices are ena is still subject to much discussion, however (Baret
simulated with the SAIL model for a large range of soil and Guyot, 1991, Sellers et al., 1992, Clevers and Ver-
reflectances, including sand, clay, and dark peat, with hoef, 1993). Although these indices appear to be more
additional variations induced by moisture and roughness. reliable and less noisy than the NDVI, they are not
The general formulation of the SA VI family of indices with widely used except in theoretical studies. The NDVI
the forvn VI = (NIR - R) / (NIR + R + X) is also reexamined. seems still to be the leading index in remote sensing
The value of the parameter X is critical in the minimiza- applications. The reason for this may be either the other
tion of soil effects. A value of X = 0.16 is found as the indices' more complex formulation or the fact that they
optimized value. Index performances are compared by have not been convincingly demonstrated to improve
means of an analysis of variance. on the NDVI in the assessment of vegetation parame-
ters. This article focuses on testing and comparing the
sensitivity of the different indices to soil background
INTRODUCTION effects.
During the last decade, vegetation indices based on The spectral reflectance of a plant canopy is a
simple combinations of visible and near-infrared reflec- combination of the reflectance spectra of plant and soil
tances, such as the normalized difference vegetation components, governed by the optical properties of these
index (NDVI) and simple ratio (SR), have been widely elements and photon exchanges within the canopy. As
used by the remote sensing community to monitor vege- the vegetation grows, the soil contribution progressively
tation from space, both on regional and global scales. decreases but may still remain significant, depending
These indices correlate well with foliage density, but on plant density, row effects, canopy geometry, wind
effects, and so on. In this article soil reflectance proper-
ties and variations are summarized and the range of
*Geography Department, University of Nottingham, Notting- current vegetation indices are briefly reviewed. Values
ham, UK of the various vegetation indices are then simulated for
tlNRA, Station de Bioclimatologie, Montfavet, France
the growth of a typical healthy green canopy, using
Address correspondence to Dr. Genevieve Rondeaux, Geography
Department, University of Nottingham, Nottingham NG7 2RD, U.K. several soil backgrounds incorporating a wide range of
Received 2 August 1994; revised 27July 1995. soil reflectances.
REMOTE SENS. ENVIRON. 55:95-107 (1996)
©Elsevier Science Inc., 1996 0034-4257 / 96 / $15.00
655 Avenue of the Americas, New York, NY 10010 SSDI 0034-4257(95)00186-5
96 Rondeaux et al.

SOIL COLOR AND SOIL effect of decreasing reflectance because of an increase

REFLECTANCE SPECTRA of multiple scattering and shadowing. Understanding
the interactions of solar radiation with soil has led to
The wide variety of soils encountered on the Earth's
mathematical models to predict the bidirectional reflec-
surface, subject to continuous environmental and tem-
tance over bare soils (Irons et al., 1989, Curran et al.,
poral variations, has never led to an easy classification
1990, Cierniewski and Courault, 1993). Models ade-
of the different soils. The color of the soil, which is
quately describe the multiple scattering according to
closely related to important soil properties, such as soil
the size and characteristics of soil particles (assuming
composition and soil moisture, has appeared to be a
simple geometric shapes), but more difficulty occurs
more practical way of soil identification. Color classifi-
when introducing soil moisture and organic content. A
cation is often done by means of the U.S. "Munsell Soil
more realistic approach to soil reflectance is at present
Color Charts" atlas (Soil Survey Staff, 1975), which is
only possible by introducing semi-empirical parameters
based on visual observation. This concept of color is
into the models (e.g., SOILSPECT; Jacquemoud et al.,
described by the colorimetric system, which character-
izes every color as a combination of the three primary
1992).
Analysis and experimental studies indicate that for
colors: red, green, and blue. These three coordinates
a given type of soil variability, the soil reflectance (p)
may then be rearranged in the three-dimensional (3D)
color space as intensity (or brightness), hue, and satura- at one wavelength is often functionally related to the
tion (Escadafal, 1993). The development of new instru- reflectance in another wavelength (Jasinki and Eagleson,
ments and remote sensing techniques has added a new 1989). In many cases, the relation can be approximated
dimension by providing soil spectra not only in the by a simple linear expression:
visible wavelengths but also in the near- and middle- p(22) = ap(2~) + b (1)
infrared regions, thus obtaining complementary infor-
mation about the soil properties (Stone et al., 1980). where the slope a and intercept b are coefficients depen-
Thus, the use of remote sensing has undeniably given dent on both the wavelengths (21, 3.2) and the type of
a new perspective to soil surveys, but has also created variability. Thus, variation of any one soil parameter can
a need to be precise when using the term "color." lead to a representative "line" in a two-dimensional
In general, soil reflectance is relatively low (<10%) scattergram (3.1 versus 3,2). For red-infrared scatter-
in the blue channel, but increases monotonically with grams this line is called the "soil line," and is used as a
wavelength through the visible and near-infrared re- base or reference line in studies of vegetated areas
gions. These optical properties vary with each soil and (Huete, 1988; Jasinki and Eagleson, 1989; Baret and
its complex composition, however. The major soil com- Guyot, 1991). The problem is that real soil surfaces are
ponents are inorganic solids (minerals); organic matter heterogeneous and contain a composite of several types
(humus, roots, decaying plant residue, living soil organ- of variability, which may be due to soil type, moisture
isms); air, and water. Mineral soils by definition contain content, roughness, and also to view-illumination geom-
less than 20% organic matter and cover over 95% of etry. Jasinki and Eagleson (1989) showed three hypo-
the world land surface (Irons et al., 1989). They may thetical visible-infrared scattergrams in which only one
be classed on a textural basis according to the propor- soil parameter is allowed to vary at a time, resulting in
tions of sand, silt, and clay. Organic matter can strongly three lines: a soil mineral line, a soil moisture line, and
influence soil reflectance. Humus compounds especially a soil shadow line (from roughness). When varying all
give a dark coloration (Curran et al., 1990), resulting in three parameters together, the composite soil line is
a general tendency to decrease reflectance throughout generally linear in the mean although considerable scat-
the spectrum. The form or decomposition stage of organic ter can exist. A unique soil line will exist only if either
material is critical, however, in that less-decomposed one type of soil variability dominates, or the scatter due
organic materials have a much higher reflectance in the to the different types of soil variability act in the same
near-infrared region than highly decomposed materials direction (Jasinki and Eagleson, 1989).
(Stoner et al., 1980). Water is also a very important
factor, and an increase of the moisture content results
in a general decrease of the reflectance. In the visible Vegetation Indices
and near-infrared regions the fraction of reflected radia- Spectral vegetation indices (VIs) use the well-known
tion changes when air-particle interfaces are replaced characteristic shape of the green vegetation speetrmn
by air-water-particle interfaces, whereas in the middle- by combining the low reflectance in the visible part of
infrared region the reflectance is controled by water the spectrum with the high reflectance in the near-
absorption bands. infrared. The combination may be in the form of a ratio,
For remote sensing purposes, soil components are a slope, or some other formulation. Indices may be
generally grouped into three characteristics: color, broadly separated into three categories: 1) intrinsic indi-
roughness, and water content. Roughness also has the ces (such as the simple ratio and the NDVI), which do
 Soil-Adjusted Vegetable Indices 97

not involve any external factor other than the measured SAVI = (1 + L)*(NIR - R) / (NIR + R + L)
spectral reflectances; 2) soil-line related indices, which with L = 0.5 (2)
include soil-line parameters, such as the perpendicular where the constant L = 0.5 has been adjusted to account
vegetation index (PVI), the weighted difference vegeta- for first-order soil background variation. A further devel-
tion index (WDVI, Clevers and Verhoef, 1993), the opment of this concept is the transformed SAVI (TSAVI)
soil-adjusted vegetation index or SAVI (Huete, 1988),
(Baret and Guyot, 1991), defined as:
the transformed SAVI (TSAVI; Baret and Guyot, 1991),
and the modified SAVI (MSAVI; Qi et al., 1994); and TSAVI = a(NIR - aR - b) / [R + a(NIR - b)
3) atmospheric-corrected indices such as ARVI (Kauf- + 0.08(1 + a2)] (3)
man and Tanr6, 1992) and GEMI (Pinty and Verstraete, where a and b are, respectively, the slope and the
1992). The characteristics and use of each group of intercept of the soil line (NIRsoil = aRs~l + b); and the
indices are described below. coefficient value 0.08 has been adjusted to minimize
soil effects. More recently the modified SAVI (MSAVI)
Intrinsic Indices has been defined as:
The simple ratio (SR) = NIR / R, and the normalized differ- MSAVI = (1 + L)*(NIR - R) / (NIR + R + L)
ence vegetation index, NDVI = (NIR - R) / (NIR + R), in with L = 1 - 2a*NDVI*WDVI (4)
which NIR and R are, respectively, the reflectances in
(Qi et al., 1994) where WDVI = NIR-aR is the weighted
the near-infrared and red channels, were developed
difference vegetation index (Clevers and Verhoef, 1993),
from channels 5 and 7 of the Landsat MSS (MultiSpec-
and a is the slope of the soil line.
tral Scanner). Their use was rapidly extended to other
The most recent index of this type is a two-axis
satellite visible-infrared sensors (Landsat TM, NOAA-
vegetation index or TWVI (Xia, 1994), defined as:
AVHRR, SPOT HRV) and also to airborne sensors and
high spectral resolution radiometers. TWVI = (1 + L)*(NIR - R - A)/(NIR + R + L)
Their popularity comes from their ability to monitor with A = ~/2*exp( - K*LAI)*D, (5)
global changes in vegetated areas. These indices are
very well related to vegetation amount until saturation where K is an extinction coefficient (defined below),
at full canopy cover and are therefore also related to D = (NIRsoil- aRsoil- b)/(1 + a2) 1/2, and L is a soil-adjust-
the biophysical properties of plant canopies, such as the ment factor with a value between 0 and 1. According
absorbed photosynthetically active radiation, efficien- to the author, this index certainly diminishes most soil
cies, and productivity. It has also been shown on experi- background influences. The TWVI requires preliminary
mental data (Steven et al., 1992) that any combination knowledge of the studied area however, (such as the
of two thematic mapper bands in an NDVI formulation leaf area index (LAI) and the soil reflectances).
shows a qualitatively similar relationship to the amount These indices considerably reduce the soil influ-
of vegetation, with only a subtle difference between the ence, especially for agricultural crops or homogeneous
color sensitivity in the indices; therefore this concept plant canopies, but because the soil line is not as univer-
can be extended beyond the two spectral bands normally sal as we would wish, a case-to-case study is always
used. necessary, as described in the next section.
The intrinsic indices, however, such as NDVI, are
extremely sensitive to soil optical properties, and are A t m o s p h e r i c a l l y C o r r e c t e d Indices
difficult to interpret with low vegetation cover when In order to reduce the dependence of the NDVI on
the soil is unknown. the atmospheric properties, Kaufman and Tanr6 (1992)
proposed a modification in the formulation of the index,
Indices U s i n g the Soil Line introducing the atmospheric information contained in
the blue channel, defining:
The relationship between near-infrared and visible re-
flectances from bare soils is generally linear, and several ARVI = (NIR - RB) / (NIR + RB), (6)
vegetation indices have been developed using the co- where RB is a combination of the reflectances in the
efficients of this relationship. Such indices attempt to blue (B) and in the red (R) channels:
reduce the influence of the soil by assuming that most
soil spectra follow the same soil line. The PVI (or its RB = n - y(B - R), (7)
successor the WDVI) expresses the distance between and y depends on the aerosol type (a good value is 7 = 1
canopy R and NIR reflectances and the soil line. Al- when the aerosol model is not available). The authors
though better than NDVI at low vegetation densities, emphasize the fact that this concept can be applied to
PVI is still significantly affected by the soil. Significant other indices. For example, the SAVI can be upgraded to
improvements are found with the soil-adjusted vegeta- the resultant SARVI (soil-adjusted and atmospherically
tion index (Huete, 1988), defined as: resistant vegetation index), by changing R to RB in the
98 Rondeaux et al.

formulation of the index; and in the same way TSAVI INTERCOMPARISON OF VEGETATION INDICES
becomes TSARVI. Myneni and Asrar (1994) have shown, VERSUS SOIL EFFECTS
however, that although SAVI and ARVI minimize the
Soil Data, Models, and Simulations
soil and atmospheric effects independently, they fail to
correct for their combined effects when applied simulta- Canopy reflectance is the result of several intricately
neously. coupled physical processes and it is therefore difficult
In a separate approach, Pinty and Verstraete (1992) to estimate the influence of a single parameter from
proposed a new index designed specifically to reduce experimental data. Only mathematical models allow the
both the soil and the atmospheric effects on satellite variation of a single parameter individually. The problem
data. This is a nonlinear index, called GEMI (global of the soil can thus be isolated by inputting different
environment monitoring index), defined as: soil optical properties in a vegetation bidirectional re-
flectance model and examining the sensitivity of the
GEMI = r/(1 - 0.25q) - (R - 0.125) / (1 - R) different indices to soil. The effects of a large range of
where r/= [2(NIR 2 - R 2) + 1.5NIR + 0.5R] / soil reflectances have thus been studied. Soil reflec-
(NIR + R + 0.5). (8) tances were graciously provided by Jacquemoud et al.
(1992) and Baret et al. (1993), who measured the spec-
This index represents plant biological information as
tral reflectances of a total of 26 soil samples. Their
least as well as the NDVI, but is transparent to the
samples consisted of five basic soil types: fine sand, clay,
atmosphere. The problem of such an index is its complex
and peat (each with three levels of moisture and two
formulation, which makes it difficult to use and interpret.
or three levels of roughness), plus pozzolana, and peb-
bles. This set of soils encompasses a very large range of
General SAVI Formulation soil refleetances, varying from about 2% to more than
Most of the soil-line related indices are, in fact, variants 60% in the visible and near-infrared wavelengths (Table
of the same general formulation, which is: 1). The radiative transfer model SOILSPECT (Jacque-
moud et al., 1992) was tested by the authors on the 26
General SAVI = (1 + L ) * ( N I R - R)/(NIR + R + L) (9)
soil samples, and was fitted to provide the bidirectional
where L is an adjusting factor. The SAVI (Huete, 1988) reflectance of these soils. The model was used to gener-
has a value of L = 0.5, which was found in his experi- alize the data and interpolate them for different spectral
ments to be the optimal adjustment factor in reducing bands and view/illumination geometries.
soil noise over the full range of canopy covers. The The bidirectional reflectance of vegetation canopies
multiplication factor (1 +L) in front of the index is is relatively well known and modeled, as a result of
needed only to maintain the dynamic range of the index. nearly half a century of canopy light-microclimate re-
Huete (1988) indicated that L becomes smaller in value search. In this study, we used the SAIL model (Verhoef,
as the vegetation becomes more dense, however, and 1984) enhanced by the hot-spot effect (Kuusk, 1991).
that for more precise studies, three adjustment factors These relatively simple models have been widely used
are preferable: L = 1 for analyzing very low vegetation by other authors and their characteristics are well
densities; L = 0.5 for intermediate vegetation densities, known. The main vegetation parameters required are:
and L = 0.25 for higher densities. Bausch (1993) found the leaf optical properties (reflectance and transmit-
that the optimal L for a corn canopy is not linearly tance), which are affected by different factors such as
correlated with LAI and proposed the following values: plant type, leaf age, leaf water content, mineral defi-
L = 0 . 6 for LAI~<I; L = 0 . 4 for l < L A I ~ < 2 . 5 ; and ciencies or disease; the leaf" area index, defined as the
L = 0.15 for LAI > 2.5. The problem with optimizing L total leaf area per unit ground area; and the canopy
for SAVI is that LAI must be known prior to computing architecture expressed as a leaf angle distribution, which
the SAVI. can play a significant role in directional effects.
Qi et al. (1994) proposed L = 1 - 2*NDVI*WDVI For a green healthy canopy, the most significant
to try to account for the variation of L from one soil to parameters are the LAI and the leaf angle distribution.
another, but then L also varies with the canopy cover In order to combine these parameters, we will use as
from 0 to 1 for very dense to very sparse canopy. the defining vegetation characteristic the "foliage cover"
The value of L appears to be very significant in defined as the fraction of the ground surface covered
minimizing the soil noise. The following section tom- by leaves. This may be calculated as:
pares the responses of the NDVI, SAVI, TSAVI, and
1 - exp( - K*LAI) (10)
MSAVI to different soil backgrounds. It also further
explores the adjustment of the L factor in order to where K is an extinction coefficient related to the leaf
optimize the normalization of soil influence over a wide angle distribution. Vegetation indices are more linearly
range of cover situations. The index GEMI, although and consequently more usefully related to foliage cover
not of the SAVI family, has also been included in the than to LAI, especially at high values of LAI when the
comparison. index values tend to approach an asymptote. Foliage
 Soil-Adjusted Vegetable Indices 99

Table 1. Description of the 26 Soil Samples (from Baret et al., 1993) and their Reflectances in the Red (660 nm)
and Near-Infrared (865 nm) Wavelengths for Nadir Viewing and Two Solar Zenith Angles: 30 ° and 60 °
Soil Reflectance Factor: Reflectance Factor:
Sample Sun = 30, Nadir View Sun = 60, Nadir View
Number Soil Type Moisture Roughness 660 nm 865 nm 660 nm 865 nm
1 Clayey soil Wet Rough 0.1069 0.1286 0.081 0.098
2 (Montfavet) Median 0.1019 0.1244 0.0794 0.0974
3 Smooth 0.1907 0.2209 0.16 0.188
4 Median Rough 0.176 0.2113 0.1348 0.1636
5 Median 0.198 0.2344 0.146 0.1744
6 Smooth 0.3715 0.4234 0.3356 0.384
7 Dry Rough 0.2239 0.2637 0.1545 0.1837
8 Median 0.2408 0.2773 0.1735 0.2013
9 Smooth 0.3847 0.4335 0.3654 0.4117
10 Silica Wet Rough 0.3847 0.4766 0.3823 0.4707
11 Fine sand Smooth 0.4185 0.4997 0.4147 0.4897
12 Median Rough 0.5422 0.6331 0.5001 0.5824
13 Smooth 0.5752 0.6518 0.5463 0.6123
14 Dry Rough 0.4961 0.5842 0.4712 0.5525
15 Smooth 0.5863 0.6597 0.5605 0.6263
16 Peat Wet Rough 0.0313 0.0831 0.0262 0.0709
17 Median 0.0302 0.0843 0.0285 0.0811
18 Smooth 0.0371 0.1043 0.0391 0.1101
19 Median Rough 0.0686 0.1552 0.0511 0.1183
20 Median 0.0691 0.1572 0.0557 0.1299
21 Smooth 0.1132 0.2596 0.1014 0.238
22 Dry Rough 0.0723 0.1606 0.0515 0.1174
23 Median 0.0646 0.1432 0.0521 0.1177
24 Smooth 0.1192 0.2468 0.1008 0.2167
25 Pozzolana Dry Rough 0.2832 0.2894 0.2664 0.2724
26 Pebbles Dry Rough 0.1287 0.1309 0.0991 0.1008

cover is also very closely related to the fraction of present single wavelength simulations, which can then
radiation absorbed b y the c a n o p y (Steven et al., 1986), be generalized to b r o a d bands.
which is the basis of m a n y agricultural and ecophysiolog-
ical p r o d u c t i o n models using r e m o t e sensing. The fo- Soil Line
liage cover has b e e n calculated in this study using Most of the present indices are related to the soil line.
Steven's m e t h o d with an ellipsoidal inclination angle Figure 2 presents red and near-infrared reflectances of
distribution (Campbell, 1986), for m e a n leaf angles vary- the 26 soils for the nadir view and two solar zenith
ing from a planophile (25 °) to an erectophile (65 °) angles: 30 ° and 60 ° . T h e graph indicates that the soils
canopy; L A I varying from 0.1 to 8. do not all follow the same line, but may be broadly
By changing the soil u n d e r n e a t h the c a n o p y in the separated into two distinct lines c o r r e s p o n d i n g to two
SAIL model, we were able to estimate the variations groups of soils: first, the mineral-based soils: clay, sand,
due to soil in the c a n o p y bidirectional reflectance. Re- pozzolana and pebbles; and second, the dark peaty soils.
flectances in the red and near-infrared wavelengths were The characteristics of the lines are as follows:
c o m p u t e d (Fig. 1) for the 26 soils and different c a n o p y Soil line: NIR = aR + b slope"a" intercept "b"
cover parameters. Because of the large range of soil All 26 soil samples 1.447 0.0225
reflectances, the c a n o p y reflectance, for low vegetation Clay and sand, plus
cover, also shows v e r y large variations in both spectral pozzolona, pebbles 1.086 0.0243
Peat soils 2.131 0.0192
channels. T h e vegetation signal is then indistinguishable
from that of the soil. The intercept of both lines is very close to the origin, but
Simulations w e r e p e r f o r m e d for wavelengths of 660 the slope for the peaty soils is about twice that of
n m in the red and 865 nm in the near-infrared. These mineral soils, due to greater reflectances in the near-
two wavelengths are the c e n t e r of the ATSR-2 bands infrared region.
and are also representative of the wavebands of other
satellites (SPOT and Landsat). Test calculations showed Results
no difference in the results obtained using broad spectral Figure 3 shows NDVI, SAVI, TSAVI, MSAVI, and G E M I
bands and single wavelengths, so we have c h o s e n to simulated for each of the 26 soil samples and a range
100 Rondeaux et al.

Soil R e f l e c t a n c e s of 26 soil s a m p l e s :
N a d i r v i e w ; S u n z e n i t h a n g l e s = 30 & 60 °
R e d R e f l e c t a n c e s ( 6 6 0 n m ) ; 26 s o i l s
S o l a r z e n i t h a n g l e = 30 ° ; V i e w = 0 °

f
Ellip~oidal leaf angle distribution /'
0506 I L AT¢M . . t,
LAI= 0.5 f
!
/
65 ° Leaf Angle(°) 05 /
04 J 55 ° /
/
• [q 45 ° LAI= 1 t~ Qa}
- L~ 35° o / O4 /
03 e~ • - D 25 o / / • Sand
"-- • • 55 °
• ½45 ° \ 03 • Peat
~ , . o
LAI=2
Pozzolona
01 LAI=4 & 8 o2
Pebbles

0 I I I I I I Soil line
0 I O2 03 04 05 06 {I 7 0g 09 l Soil iinc peat
cover
I I [ I I [
o I o2 O3 O4
Red Reflectance
N e a r - I n f r a r e d R e f l e c t a n c e s ( 8 6 5 n m ) ; 26 s o i l s
Solar zenith angle=30 ° ; View= 0 °
Figure 2. Relationships between near-infrared and red re-
08 LAI= 4 & 8
flectances for the 26 soil samples, resulting in two soil lines.
LAI= 0.1 LAI = 0.5 LAI= I LAI= 2
07

06 Illl

05
the index (Fig. 3c). In this case, the SAVI (Fig. 3b),
which uses coefficients previously adjusted from a global
04
soil line (Huete, 1988), produces a slightly better result,
~o3
but still with quite large variations. Better results may
02
be obtained when applying coefficients from the two
65° Leaf Angle (°)
OI soil lines separately (Fig. 3d): the dispersion of the data
0 I I I I I I I I I I is smaller at low values of LAI, but here the problem
Ill O2 03 04 1) 5 I) 6 07 O8 O9 I reappears at the larger cover values (LAI >2) where
cover
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 overall
Figure 1. Red and near-infrared spectral reflectances simu-
lated with the SAIL + SOILSPECT model for the range of performance is obtained when considering each group
26 soil samples, as a function of foliage cover (0 = bare soil; of soils separately. MSAVI (Fig. 3e) is quite similar to
1 = 100% cover) represented by a range of leaf area index SAVI (Fig. 3b). GEMI (Fig. 3f) displays a wide spread
and leaf angle distributions. of the data at very low cover (LAI = 0.1), then a much
smaller spread for LAI >i 0.5, but also a reduced dynamic
range of the index.
of canopy covers. The NDVI (Fig. 3a) demonstrates, as The choice of the parameter L in the SAVI-type
expected, a very large range of values with varying indices appears to be quite critical in minimizing the soil
background, especially for low vegetation densities. For background effect. Figure 4 shows normalized standard
LAI~< 0.1, the reflected signal may be considered as deviations, accounting for the full range of soils, for
equivalent to the soil signature. With intermediate several indices and for the general formulation of the
amounts of vegetation (0.5 < LAI < 3) the wide range of "SAVI family" of indices, simplified to:
possible values could produce very large errors when
VI = (NIR - R) / (NIR + R + X) (11)
trying to retrieve the plant cover from the NDVI values.
At large LAI, the NDVI reaches a saturation level. where we prefer to note the adjusting factor "X" instead
Clearly NDVI will give poor information about a vegeta- of"L" because of its variability. The multiplication factor
tion canopy when the soil background is unknown. (1 + L) has been left out, because it only significantly
Soil-adjusted indices, SAVI, TSAVI, and MSAVI affects the index at relatively large values of L (> 0.4).
(Fig. 3b, 3c, and 3e), perform better than NDVI. Some All indices have been normalized between 0 (for the
problems remain, however, because of the use of a minimum value) and 1 (for the maximum at 100%
universal soil line for all soil types. When using a single cover) before calculation of the standard deviation.
soil line for all soils (averaging all the 26 single lines), For low vegetation cover, if no information about
the TSAVI, although better than NDVI, does not greatly the soil type is available, then SAVI and MSAVI have
improve the information about the vegetation given by the lowest standard deviations with soil variation. Their
 Soil-Adjusted Vegetable Indices 101

NDVI ; Sun= 45°; View= 0 ° SAVI ; Sun= 45 ° ; View= 0 °

26 soils 26 soils

1 LAI = 4 & 8 1 -Ellipsoildal Leaf Angle Distribution

LAI = 2
0.9

0.8
LAI = 0 5
LAI=

• ":i gl
I

i
j!ll' 0.9

0.8
LAI = 4 & 8

!
I II
m==.. =
Ill- LAI = 2
I , .==."
.i: a
0.7 LAI = 0.1
I
• :-•
• :aiD=
,= =!
I 0.7 -
LAI = 1
. ii i ,
.":. ...::' 250 0.6 - |=l
0.6
:: :!" 3~ iI
tm
Z
0.5
i
/. i

, , : ==m
,:.
l=
..

I
•
I
= 55
45

Leaf Angle
0.5 - LAI = 0.5

.: !i-
'ii!2° I

.;iJ!
0.4 0.4 - LAI = 0.1
I' 45
•

|= : 55
0.3 0.3 - .ff~ |! "
||"
i" 65 Leaf Angle

t
0.2 0.2 - ~

0.1 0.1 - I
Ellipsoidal Leaf Angle Distribution
#
0 I I ' I I ' I 0 I I ' I I I
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

cover = 1-exp(-K*LAI) cover

(a) (b)

TSAVI;Sun= 45°; View= 0°: 26 soils TSAVI ; Sun= 45°; View= 0°; 24 soils

0.8 LAI = 4 & 8 0.8 LAI= 4 & 8

0.6 LAI = I
LA;,,, r LAI = 2

.i! !

!i;"
I 0.6
i!
LAI = 0.5
I=1
1,1" LAI= 1 .i
.!.i
I
0.4
~i! I

_AI = o.1 ;=
|I
|
I
0.4 L A I = 0.5
!
,ii','
0.2 Soilline:
.li! i
i ,:.
i-
l
a= 1.4475;b= 0.0225
,I
0.2
L A I = 0.1 | Soil lines:
-' l I ' I ' I ' I
C l a y + S a n d : a = 1.086 ; b = 0 . 0 2 4 3
t ! 0.2 0.4 0.6 0.8 1
Peat soils: a= 2.131 ; b= 0 . 0 1 9 2

-0.2 I I I I I
0.2 0.4 0.6 0.8 1
cover cover
(c) (d)
Figure 3. Vegetation indices as a function of foliage cover, for different canopy geometries and the 26 soil backgrounds:
a) NDVI, b) SAVI, c) TSAVI using a common soil line for the 26 soils, d) TSAVI using separate soil lines for peat and
mineral soil, e) MSAVI, and f) GEMI.
102 Rondeaux et al.

MSAVI ; Sun= 45 ° ; View= 0 ° ; 2 6 soils GENII ; Sun=45 ° ; View= 0 ° ; 2 6 soils

LAI=4&8
0.9 1.2
"LAI= 4 & 8

0.8 LA,:2. t I LAI= 2

nil m LAI= 1
0.7 ill n
|;m
LAI=, :iN LAI =0.5 !|.roll'""
0.6
! !u
!,
0.8

LAI= 0. l "" :i"'

.~ 0.5
<

0.4
LAI= 0.5
!!!
L9
0.6
1 '
0.3
LAI= Ol

-
.,.
n;;
"::|i!

..;!i'.
io
i
0.4

0.2
t
Nan
0.2 " ::
°

0.1 0 I I ! I I
Ellipsoidal Leaf Angle Distribution
• 0.2 0.4 0.6 0.8 1
0 I I I I I
-0.2 t
0.2 0.4 0.6 0.8 1
cover cover
(e) (f)

Figure 3. Continued.

standard deviations are quite constant over the range The value X = 0.16 thus appears to be the optimum
of cover, however, decreasing only above 80% cover, adjusting factor for the SAVI family of indices with our
and therefore, for instance, being larger than the devia- data set. W e will refer to this index by:
tion of most other indices (except GEMI and NDVI) at OSAVI = (NIR - R ) / ( N I R + R+0.16), (12)
60% cover. When minimizing the standard deviations
over the full range of cover, the best general index where OSAVI stands for optimized SAVI. The OSAVI
comes with a value of X = 0 . 1 6 or 0.2, which gives corresponds, in fact, to the TSAVI with the parameters
variations a little higher than SAVI for low vegetation a = 1 and b = 0, and is therefore not equivalent to the
covers (< 50%) but still much better than NDVI, while NDVI, as has often been noted erroneously in the litera-
having a very good behavior for vegetation covers ture.
greater than 50%.
Figure 5 is similar to Figure 4 but for the indices Analysis of Variance
applied to a reduced range of soils comprising the Analysis of variance is a powerful tool for isolating the
mineral soils, that is, clay and sand, but excluding peat effects of different nominal variables, separately and in
and also excluding pozzolana and pebbles because they combination, on a dependent variable. This is an analyti-
are quite unlikely to be found as a vegetation back- cal procedure that indicates what proportion of the
ground. All the indices now show lower standard devia- variation in a dependent variable can be accounted for
tions than in Figure 4 (except GEMI for a very sparse by variations in one or more independent variables
canopy), which is expected because of the narrower (Johnston, 1980). It is used here to estimate the impact
range of soil reflectances. With the SAVI family of of the different factors affecting the vegetation index.
indices, however, an increase of the value of X signifi- An analysis of variance has been performed on the
cantly improves the index sensitivity but only up to a indices previously tested. Results are presented in Ta-
value X between 0.1 and 0.2 according to the percentage bles 2 and 3. The analysis was done first by considering
of cover. Higher values of X, such as for the SAVI the full range of soils, and second by considering only
(X = 0.5), result in higher standard deviation. the reduced range of clay and sand samples.
 Soil-Adjusted Vegetable Indices 103

Standard Deviation of "normalized" Vegetation Indices

Sun zenith= 45 ° ; View= 0 ° ; 26 soils

0.2
LAI = 0.1 Spherical Leaf Angle Distribution

0.18
~ LAI= 0.5

0.16
0x " ' ' -
o 0.14

-A-- X=0.2
"o 0.12
-- ×-- X=0.3
~ 0.1
-- x--- X=O. 4

0.08
• SAVI
E
0.06 "~'-~"- " -x " ' ' ' - " ~ ~ ">~q~ - - -- -- - e -- TSAVI
z

0.04 -- + MSAVI

-<3- - GEMI
0.02
, , , , , , , ,-
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0,9 1

cover

Figure 4. Sensitivity of vegetation indices to soil background effects, as a function of the foliage cover. The normalized stan-
dard deviation, which has been calculated after normalizing all indices between 0 and 1, represents the variability caused by
the 26 soil samples.

The three main independent variables affecting a sum of the contributions from LAI, Leaf-Angle and
vegetation index are: the soil reflectance (Soil), the their interactions. The independent contributions of the
amount of vegetation (LAI) and the canopy architecture variables Soil, LAI, Leaf-Angle and Cover are presented
(mean leaf inclination angle: Leaf-Angle). A three-way in Table 2a. Soil/LAI interactions are also included,
analysis of variance provides eight sources of variation, but for clarity the other interactive effects and residuals
respectively, from the following influences: are not included in the table as they do not seem to
make any contribution to interpreting the data: Soil/
1. Soil,
Leaf-Angle interactions are always very small (less than
2. LAI,
0.05%) and unsignificant; LAI / Leaf-Angle interactions
3. Leaf-Angle,
are part of the cover contribution and less than 1%.
4. interaction of Soil and LAI,
The variable, Soil, has initially 26 classes representing
5. interaction of Soil and Leaf-Angle,
the 26 samples. Table 2b shows the analysis where Soil
6. interaction of LAI and Leaf-Angle,
has been divided into the three soil variables from
7. interaction of Soil, LAI and Leaf-Angle,
Table 1: Soil-type (5 classes), moisture (3 classes), and
8. residual (or unexplained).
roughness (3 classes). The influences of LAI, Leaf-Angle,
Each of the three variables (Soil, LAI, Leaf-Angle) or Cover are then unchanged and are the same as in
makes an independent contribution to the explanation Table 2a.
of variation in VI, and in addition all have joint effects LAI or Cover are naturally the main variables ex-
with each other. By combining LAI and Leaf-Angle as plaining the variation of vegetation indices. The aim in
leaf cover, the analysis can be reduced to the two optimizing a soil-adjusted vegetation index is to increase
independent variables: Soil and Cover, resulting in only the response to these variables while decreasing the
four sources of variation, which are: Soil, Cover, interac- variability due to soil, and therefore also decreasing the
tion Soil-Cover, and residual. When this is done, the Soil/LAI interactions. On these criteria, the best index
contribution from Soil is unchanged, and the contribu- from Table 2a seems to be the MSAVI, followed by
tion due to Cover is, as expected, almost equal to the SAVI, then OSAVI, TSAVI and finally NDVI and GEMI.
104 Rondeaux et al.

S t a n d a r d D e v i a t i o n of " n o r m a l i z e d " V e g e t a t i o n Indices

Sun zenith= 45 ° ; View= 0 ° ; 1 5 soils (clay & sand)

0.18
Spherical Leaf Angle Distribution
41 NDVI
0.16
\
*- X-0.1
0.14 \ • X=0.16
o= \\\
0.12 • X -0.2
\
\\\ -- x X=0.3
0.1

\\\ - ~( X=0.4
g$ 0.08
'\ • SAVI

E 0.06 o TSAVI all soils

Z
- <> TSAVI 15 soils
0.04
• MSAVI
0.02
• GEMI
i
0 I I I I I ) I I I "~,h ~
0.1 0.2 0.3 0.4 0,5 0.6 0.7 0.8 0.9 1

cover

Figure 5. As in Figure 4 but considering only the 15 samples of clay and sandy soil.

GEMI has a low direct sensitivity to Soil, but also has When soil samples are separated into soil type,
the second lowest sensitivity to LAI or Cover (after moisture, and roughness (Table 2b), interpretation of
NDVI) and the highest Soil/LAI interactions (as well the analysis is less clear because of the collinearity of
as the highest residual). This result is thought to be due these three variables. Both soil moisture and roughness
to the high sensitivity of this index to soil at very low seem to have a relatively small influence on vegetation
vegetation cover (Fig. 3f and 4). From Figure 4 also, it index, which is to be expected for indices based on the
appears that the good results of SAVI and MSAVI are soil line. Changes in water content or roughness have
in fact due to their almost constant sensitivity to Soil at the effect of moving the soil reflectances up or down
all LAI. Table 3 shows the same analysis of variance as the soil line. The main effect is due to the soil type,
Table 2a, but considers only the reduced range of min- which is more important by about an order of magnitude
eral soils (clay and sand samples), that is, 15 classes for then either moisture or roughness. A surprisingly low
the variable Soil. TSAVI and OSAVI are then the leading sensitivity is obtained with GEMI, which is probably an
indices, followed by SAVI and MSAVI, and then NDVI artifact of the previous finding, that its soil response
and GEMI. was linked to Soil/LAI interaction.

Table 2a. Analyses of Variance Have Been P e r f o r m e d

on the Various Indices to Estimate t h e I m p a c t of the Table 2b. Effects from the 26 Soil Samples Are Separated
Main Factors Affecting Indices into 3 Variables: Soil Type (5 classes), Soil Moisture
Source of Soil/LAI (3 classes), and Soil Roughness (3 classes)
Variation Main Effects (%) Interactions
Source of Main Effects front the
(VI) Soil LAI Leaf Angle Cover (%) 26 Soil Samples (%) Main Effects (%)
Variation
NDVI 7.49 83.77 0.73 85.36 6.93 Wt) Soil type Moisture Roughness LAI Leaf Angle
SAVI 1.14 93.92 3.17 97.59 1.22
TSAVI 2.11 92.44 1.83 94.90 2.93 NDVI 6.89 0.60 0.59 83.77 0.73
MSAVI 1.05 94.68 2.60 98.18 0.69 SAVI 0.83 0.08 0.35 93.92 3.17
CEMI 0.93 88.28 3.22 91.94 6.99 TSAVI 2.03 0.12 0.09 92.44 1.83
OSAVI 1.71 93.40 1.94 95.98 2.27 MSAVI 0.71 0.08 0.37 94.68 2.60
GEM1 0.10 0.04 0.19 88.28 3.22
Independent variables are: Soil with 26 classes (the 26 soil samples), OSAVI 1.63 0.09 0.08 93.40 1.94
LAI with 6 classes (LAI = 0.1, 0.5, 1, 2, 4, 8), Leaf Angle with 5 classes
(25 °, 35 °, 45 °, 55 °, 65°), and Cover with 5 x 6 = 30 classes. Effects from LAI and Leaf Angle are unchanged.
 Soil-Adjusted Vegetable Indices 1 0 5

Table 3. Same as Table 2a, But Considering only 60

15 Classes for the Variable Soil: 15 Clay and

Sand Samples dry peaty soils

50 • w e t peaty soils
Source of Main Effects (%) Soil/LAI
Variation Interactions • w e t soils z

(VI) Soil LAI Leaf Angle Cover (%) d r y soils

NDVI 0.97 96.33 0.89 98.19 0.78 4O soilline

SAVI 0.99 95.37 2.79 98.62 0.34 p o t t i n g soil I

TSAVI 0.10 97.31 1.79 99.76 0.10 m-
2 n d soilline
MSAVI 1.04 95.54 2.23 98.57 0.31 -~ 30
GEMI 0.57 89.66 2.57 92.52 6.78 =

OSAVI 0.06 97.37 1.74 99.76 0.15

2O

DISCUSSION
The soil background is a major surface component con- 10

trolling the spectral behavior of vegetation canopies,

and on which the retrieval of biophysical characteristics
of the canopy depends. The sensitivity of different vege- "' I I I I
tation indices to soil color has been investigated. Al- 5 10 15 2O
Red Reflectance
though vegetation indices, such as the soil-adjusted veg-
etation indices, considerably reduce these soils effects,
estimation of the vegetation characteristics from the Figure 6. Relationships between near-infrared and red re-
flectances of different soil samples measured with the Sin-
indices still suffers from some imprecision, especially at gle-FOV-IRIS spectroradiometer (Rondeaux, 1995).
relatively low cover, if no information about the target
is known.
The first problem was the use of the soil line, which although they are a good addition to a data set to test
cannot be generalized because of its variability. Even and validate models.
though the 26 soil samples from Table 1 represent a In the general formulation of the "SAVI family" of
very large range of spectral values, a question still re- indices, the minimization of the soil effect is done by
mains about their' representativity. The anomalous soil adjustment of the parameter X. In this study, the value
line encountered with the peaty soil has been investi- of X has been reexamined to optimize the index sensitiv-
gated further. The peat soil used for the simulation was ity with our soil data set. The value X = 0 . 1 6 is found
taken from a bag of commercial peat compost (Baret, to give satisfactory reduction of soil noise, both at low
personal communication) and is assumed to have a and high vegetation cover. The resulting index, OSAVI,
high organic-matter content. A small experiment was is attractive because of its simple formulation: It is a
conducted to test the differentiation between peat and simplified version of the TSAVI, which does not require
mineral soil lines seen in Figure 2. Spectral signatures any preliminary knowledge of the soil line parameters.
of an independent set of soil samples were measured This index is currently being applied semi-operationally
with a high resolution spectroradiometer (Rondeaux, for sugar beet yield prediction and management in the
1995). Soil samples were collected in agricultural fields UK (Xu and Steven, 1995).
in East Anglia, UK, including light-brown soils and soils The main objectives of vegetation remote sensing
designated as peat on British Soil Survey maps. An are to analyze, interpret, and monitor temporal vegeta-
organic potting soil from a commercial bag was also tion changes on seasonal or annual time scales, The
tested. The results are shown in Figure 6, where all optimal choice of a vegetation index is still very much
soils taken from fields, including the peaty soils, follow related to the purpose of study and the type of vegetation
the same line with a slope of 1.35, whereas the bagged considered as well as to the amount of prior information
soil shows a line with a much higher slope of about available. In agricultural studies over a single soil type
3.5. Although the absolute values of these slopes are there is a little to choose between the indices studied
somewhat greater than those in Figure 2, this result here, but for the generalization and intercomparison of
seems to confirm the assumption that the second soil results from different places, it is desirable to standard-
line observed in Figures 2 and 6 is due to the high ize on a single, soil-invariant index. No generalization
organic content of the soil (Stoner et al., 1980, NcNairn can easily be made, however. Simulations have been
and Protz, 1993). The peat (as in Table 1) or highly done here by means of the SAIL model, which charac-
organic composts from commercial bags, are not very terizes homogeneous canopies such as grass and agricul-
representative of peat soils found in the real world, tural crops at mid-latitudes. TSAVI or OSAVI seem to
106 Rondeaux et al.

offer the best results for this kind of canopy. In arid Jasinki, M. F., and Eagleson, P. S. (1989), The structure
climates where the vegetation cover is generally less of red-infrared scattergrams of semivegetated landscapes,
than 25%, MSAVI may be better (Fig. 5), whereas both IEEE Trans. Geosci. Remote Sens. 27:441-451-.
SAVI and MSAVI, which have a slightly higher but more Johnston, R. J. (1980), Multivariate Statistical Analysis in Geog-
raphy, Longman Scientific & Technical, in press.
constant sensitivity over the full range of cover, could
Kaufman, Y. J., and Tanr6 D. (1992), Atmospherically resistant
also be quite useful for general-purpose vegetation clas-
vegetation index (ARVI) for EOS-MODIs, IEEE Trans.
sification. To make progress, we suggest, for the reasons Geosci. Remote Sens. 30 (2):261-270.
outlined above, that OSAVI should be adopted for ag- Kuusk, A. (1991), The hot-spot effect in plant canopy reflec-
ricultural applications, whereas MSAVI is r e c o m m e n d e d tance, in Photon-Vegetation Interactions, Application in Op-
for more general purposes. Standardization of vegetation tical Remote Sensing and Plant Ecology. (Myneni and Ross,
indices is important at this stage of progress of the Eds.), Springer Velag, New York, pp. 139-159.
science and further empirical evaluation of indices is McNairn, H., and Protz R. (1993), Mapping corn residue cover
required to establish confidence in any standards to be on agricultural fields in Oxford county, Ontario, using
adopted. Thematic Mapper, Can. Remote Sens. 19:152-159.
Malthus, T. J. (1989), Anglo-French Collaborative Reflectance
Experiment. Experiement 1, Brooms Barn Station, July
This work is part of a research project on "Biophysical Indices, 1989. Internal report, University of Nottingham.
from ATSR-2 ~ supported by a grant from the UK Natural Myneni, R. B., and Asrar G. (1994), Atmospheric effects and
Environment Research Council (NERC). The authors gratefully
spectral vegetation indices, Remote Sens. Environ. 47:390-
acknowledge the anonymous referees for their helpful comments.
402.
Pinty, B., and Verstraete, M. M. (1992), GEMI: A non-linear
index to monitor global vegetation from satellites, Vegeta-
REFERENCES
tion 101:15-20.
Qi, J., Huete, A. R., Moran, M. S., Chehbouni, A., and Jackson,
Baret, F., and Guyot, G. (1991), Potentials and limits of vegeta- R. D. (1993), Interpretation of vegetation indices derived
tion indices for LAI and PAR assessment, Remote Sens. from multi-temporal SPOT images, Remote Sens. Environ.
Environ. 35:161-173. 44:89-101.
Baret, F., Jacquemoud, S., and Hanocq, J. F. (1993), The soil Qi, J., Chehbouni, AI., Huete, A. R., Kerr, Y. H., and Soroos-
line concept in remote sensing, Remote Sens. Rev. 7:65- hian, S. (1994), A modified soil adjusted vegetation index
82. (MSAVI) Remote Sens. Environ. 48:119-126.
Bausch, W. C. (1993), Soil background effects on reflectance- Rondeaux, G. (1995), Analysis of Soil Spectral Properties with
based crop coefficients for corn, Remote Sens. Environ. 46: the GER Single Field-of-View IRIS, Internal report, Geog-
213-222. raphy Department, University of Nottingham, Nottingham,
Campbell, G. S. (1986), Extinction coefficients for radiation UK.
in plant canopies calculated using an ellipsoidal inclination Sellers, P. J., Berry, J. A., Collatz, G. J., Field, C. B., and
angle distribution. Agric. For. Meteorol. 36:317-321. Hall, F. G. (1992), Canopy reflectance, photosynthesis and
Cierniewski, J., and Courault, D. (1993), Bidirectional reflec- transpiration III. A reanalysis using improved leaf models
tance of bare soil surfaces in the visible and near-infrared and a new canopy integration scheme, Remote Sens. Envi-
range, Remote Sens. Rev. 7:321-339. ron. 42:187-216.
Clevers, J. G. P. W., and Verhoef, W. (1993), LAI estimation by Soil Survey Staff (1975), Soil taxonomy: A Basic System of Soil
means of the WDVI: A sensitivity analysis with a combined Classification for Making and Interpreting Soil Survey, Soil
PROSPECT-SAIL model, Remote Sens. Environ. 7:43-64. Conservation Office. U.S. Dept. of Agric. Handbook, no.
Curran, P. J., Foody, G. M., and Kondratyev, K. Ya., Kozod- 436. Washington, D.C.
erov, V. V., and Fedchenko, P. P. (1990), Remote Sensing Steven, M. D., Biscoe, P. V., Jaggard, K. W., and Paruntu J.
of Soils and Vegetation in the USSR, Taylor & Francis. (1986), Foliage cover and radiation interception, Field
Escadafal, R. (1993), Remote sensing of soil colour: Principles Crops Res. 13:75-87.
and applications. Remote Sens. Rev. 7:261-279. Steven, M. D., Malthus, T. J., Danson, F. M., Jaggard, K. W.,
Escadafal, R., and Huete A. (1991), Etude des propr6t6s and Andrieu, B. (1992), Monitoring Response of Vegetation
spectrales des sols arides appliqude h l'am61ioration des to Stress, in Proceedings Remote Sensing Society Annual
indices de vdg6tation obtenus par t616ddtection, C. R. Acad. Conference, Dundee, UK. Remote Sensing from Research to
Sci. Paris, 312:1385-1391. Operation, pp. 369-377.
Huete, A. R. (1988), A soil-adjusted vegetation index (SAVI), Steven, M. D., Malthus, T. J., and Baret, F. (1994), Intercali-
Remote Sens. Environ. 25:295-309. bration of vegetation indices from different sensor systems,
Irons, J. R., Weismiller, R. A., and Petersen, G. W. (1989), Soil In 6th International Symposium Physical Measurements and
reflectance, in Theory and Application of Optical Remote Signatures in Remote Sensing. Val D'Isere, France, January
Sensing, (G. Asrar, Ed.), Wiley Interscience, New York, pp. 94 (In press).
66-106. Stoner, E. R., Baumgardner, M. F., Biehl, L. L., Robinson,
Jacquemoud, S., Baret, F., and Hanocqu, J. F. (1992), Model- B. F. (1980), Atlas of soil reflectance properties. Research
ling spectral and bidirectional soil reflectance, Remote Sens. Bulletin 962, Laboratory for Application of Remote Sensing,
Environ. 41:123-132. Purdue University, West Lafayette, Indiana, 75 pp.
 Soil-Adjusted Vegetable Indices 107

Verhoef, W. (1984), Light scattering by leaf layers with appli- Xu, H., and Steven, M. D. (1995), Intercalibration of airborne
cation to canopy reflectance modelling: the SAIL model, and satellite optical systems. Sugar beet prediction and
Remote Sens. Environ. 16:125-141. management. Technical note no. SBP-UN-TN-008. Univer-
Xia, Li. (1994), A two-axis vegetation index (TWVI), Int. J. sity of Nottingham and Logica UK.
Remote Sens. 15:1447-1458.

View publication stats