Preventive Veterinary Medicine, 11 ( 1991 ) 185-200 185

Elsevier Science Publishers B.V., Amsterdam

Interpreting vegetation indices

Ray D. Jackson a and Alfredo R. H u e t e b

aUSDA-ARS. US Water Conservation Laboratory, Phoenix, AZ 85040, USA
bDepartment of Soi/ and Water Science. Universityof Arizona, Tucson, AZ 85721, USA

ABSTRACT

Jackson, R.D. and Huete, A.R., 1991. Interpreting vegetation indices. Prey. Vet. Med., 11: 185-200.

Remotely sensed spectral vegetation indices are widely used and have benefited numerous disci-
plines interested in the assessment of biomass, water use, plant stress, plant health and crop produc-
tion. The successful use of these indices requires knowledge of the units of the input variables used to
form the indices, and an understanding of the manner in which the external environment and the
architectural aspects of a vegetation canopy influence and alter the computed index values. Although
vegetation indices were developed to extract the plant signal only, the soil background, moisture con-
dition, solar zenith angle, view angle, as well as the atmosphere, alter the index values in complex
ways. The nature of these problems are explored both in an empirical and in a theoretical sense, and
suggestions are offered for the effective use and interpretation of vegetation indices.

INTRODUCTION

A p r i m a r y goal o f m a n y r e m o t e sensing p r o j e c t s is to c h a r a c t e r i z e t h e t y p e ,
a m o u n t a n d c o n d i t i o n o f v e g e t a t i o n p r e s e n t w i t h i n a scene. T h e a m o u n t o f
light r e f l e c t e d f r o m a s u r f a c e is d e t e r m i n e d b y t h e a m o u n t a n d c o m p o s i t i o n
o f solar i r r a d i a n c e t h a t strikes the surface, a n d the r e f l e c t a n c e p r o p e r t i e s o f
t h e surface. B e c a u s e solar i r r a d i a n c e varies w i t h t i m e a n d a t m o s p h e r i c c o n -
d i t i o n s , a s i m p l e m e a s u r e o f light r e f l e c t e d f r o m a surface is n o t s u f f i c i e n t to
c h a r a c t e r i z e the surface in a r e p e a t a b l e m a n n e r . T h i s p r o b l e m can be c i r c u m -
v e n t e d s o m e w h a t b y c o m b i n i n g d a t a f r o m t w o o r m o r e spectral b a n d s to f o r m
w h a t is c o m m o n l y k n o w n as a v e g e t a t i o n i n d e x ( V I ) . A VI c a n be c a l c u l a t e d
b y ratioing, d i f f e r e n c i n g , r a t i o i n g d i f f e r e n c e s a n d sums, a n d b y f o r m i n g l i n e a r
c o m b i n a t i o n s o f s p e c t r a l b a n d data. VI are i n t e n d e d to e n h a n c e the vegeta-
t i o n signal, while m i n i m i z i n g solar i r r a d i a n c e a n d soil b a c k g r o u n d effects.
T h e u b i q u i t o u s use o f VI w o u l d suggest t h a t t h e y are c a l c u l a t e d in a uni-
f o r m m a n n e r , a n d t h a t t h e y are c o m p a r a b l e o v e r t i m e a n d l o c a t i o n . U n f o r -
t u n a t e l y , this is n o t t h e case. VI c a n b e c a l c u l a t e d f r o m s e n s o r voltage o u t p u t s
( V ) , r a d i a n c e v a l u e s ( L ) , r e f l e c t a n c e v a l u e s (p) a n d satellite digital n u m b e r s
( D N ) . E a c h is c o r r e c t , b u t e a c h will y i e l d a d i f f e r e n t VI v a l u e for the s a m e

© 1991 Elsevier Science Publishers B.V. All rights reserved 0167-5877/91/$03.50

186 R.D. JACKSON AND A.R. HUETE

surface conditions. View and solar angles affect data from each spectral band
differently, and thus may affect VI values. Soil background exerts a major
influence on VI. VI calculated from data obtained from aircraft- or space-
craft-based sensors are affected by the intervening atmosphere. VI calculated
from data obtained over the same target, but with different instruments, may
not be the same because of detector and filter characteristics of the instru-
ments. It is the purpose of this report to demonstrate the magnitude of differ-
ences that can occur, and to provide suggestions for the effective use and
interpretation of VI.

THE CONCEPT OF VI

When light strikes a surface, part is reflected, part is transmitted and the
remainder is absorbed. The relative amounts of reflected, transmitted and
absorbed light are a function of the surface and vary with the wavelength of
the light. For example, the majority of light striking soils is either reflected or
absorbed, with very little being transmitted and relatively little change with
wavelength. With vegetation, however, most of the light in the near-infrared
( N I R ) wavelengths is transmitted and reflected, with little absorbed, in con-
trast to the visible wavelengths where absorption is predominant, with some
reflected and little transmitted.
Reflectance spectra for bare dry soil, bare wet soil and a full-cover wheat
canopy are depicted in Fig. 1. The vertical dashed lines labeled 'red' and 'near-
infrared' delineate the wavelength intervals representative of Bands 3 and 4
of the Thematic Mapper ( T M ) on LANDSATS 4 and 5, and Bands 2 and 3
of the high resolution visible ( H R V ) sensors on the French satellites SPOT 1
and 2. Horizontal solid lines labeled A - F indicate the average reflectance
within the waveband for the soil and wheat targets. If a wheat field is to be
monitored, early in the season only bare soil will be observed by the sensor.

0.6
Near i n f r o ~ d
D:
0.5 Y ~ a t
0.4-
0 I i ~ . ~ " Dry Soil
(~ A t I

0.2 / ~ i ~ ' - w~tso,

o., ~ ci ! , i /o I :!
0.0 . i , , ~ , i
0.3 0.5 0.7 0.9 1.1

Wavelength (/~)
Fig. 1. Reflectance spectra for wheat, dry bare soil, and wet bare soil. Vertical dashed lines
indicate the approximate band widths of the red and NIR bands of the LANDSATTM.
I N T E R P R E T I N G VEGETATION INDICES 187

0.7

~6 0.6
(3 0.5

0.4
C
(3 0.3

0.2

0.1
r~
0.0
100 12o 14o 16o lao 200 220 240 260 2ao
Day of Yeor
Fig. 2. Reflectancefactor data for the red (open circles) and NIR bands (solid circles) mea-
sured over cotton. The solid lines representsmootheddata.

As the plants develop, the output in the red band decreases from A (if the soil
is dry) or B (if wet), reaching C when the plants fully cover the soil. In the
NIR, the output increases from Point E (if the soil is dry) or F (if wet),
toward Point D. In general, the wavebands used to calculate VI are chosen
such that one decreases and the other increases with increasing vegetation
cover.
Seasonal trends of NIR and red reflectances for cotton are shown in Fig. 2.
The reflectance measurements were made with a hand-held radiometer with
band filters similar to those in Bands 2 and 3 of the HRV sensors aboard
SPOT- 1. The field was subirrigated, thus the soil surface remained essentially
dry throughout the season. In Fig. 2, the red reflectance decreased and the
NIR increased as the cotton plants grew and covered the soil. The solid line
drawn through the points for both bands represent smoothed data that will be
used in a subsequent graph.

CLASSESOF Vl

There are two general classes of VI: ratios and linear combinations, both of
which exploit the surface-dependent a n d / o r wavelength-dependent features
of the data shown in Figs. l and 2. Ratio VI may be the simple ratio of any
two spectral bands, or the ratio of sums, differences or products of any num-
ber of bands. Linear combinations are orthogonal sets of n linear equations
calculated using data from n spectral bands.

Ratios

The ratio vegetation index (RVI), formed by dividing the NIR radiance by
the red radiance, was probably the first index to be defined and is the most
commonly used. Tucker (1979) reported that Jordan ( 1969 ) used a radiance
188 R.D. JACKSON AND A.R. HUETE

ratio of 0.800/0.675/lm, measured at the forest floor, to derive a leaf area

index for a forest canopy. In equation form, the RVI is

NIR
RVI- (1)
red

RVI values for the data shown in Fig. 1 are 12.9 for wheat, 1.21 for bare dry
soil and 1.33 for bare wet soil. For the cotton data shown in Fig. 2, the RVI
was 1.43 on Day 130 (bare soil) and 20.2 on Day 225 (near-maximum green
vegetation ).
For dense green vegetation, the amount of red light reflected from the can-
opy is very small (open circles, Fig. 2). As can be seen in eqn. ( 1 ), as red
band reflectance approaches zero, the ratio increases without bound. Thus,
the amount of reflected red light must be measured with considerable accu-
racy if reasonable values of the ratio are to be expected. If the red band is
measured with sufficient precision, the RVI is quite sensitive to vegetation
changes during the time of peak growth. It is not very sensitive when the veg-
etative cover is sparse.
Deering (1978 ) found that the low dynamic range of the N I R / r e d ratio
over sparse vegetation could be enhanced by ratioing the difference between
the NIR and the red bands to the sum of the two bands. This VI was subse-
quently named the normalized difference vegetation index (NDVI). Thus

N I R - red
NDVI = N - - - ~ r ~ (2)

The upper bound of the NDVI approaches one, while the lower bound is usu-
ally close to zero. The lower bound may be slightly positive or slightly nega-
tive, depending on sensor characteristics and the units of the input variables
(volts, radiance, reflectance, digital numbers, etc. ). NDVI values for the wheat
data in Fig. 1 are 0.86, 0.09 and 0.14 for wheat, dry soil and wet soil, respec-
tively. NDVI for cotton (Fig. 2) are 0.18 and 0.96 for Days of Year 130 and
225, respectively.
The relative sensitivity of the NDVI and the RVI to vegetation density is
demonstrated in Fig. 3. The values used for the lines were calculated from the
smoothed data in Fig. 2 and normalized to range from zero to one. The nor-
malization was required to facilitate comparison of the two indices. When
plotted in this manner, it is evident that the NDVI (dashed line) is more
sensitive to sparse vegetation densities than is the RVI, but is less sensitive to
high vegetation densities.
From the mathematical point of view, these two indices are functionally
equivalent and thus contain the same information. Perry and Lautenschlager
(1984) addressed this point and showed that one index can be readily trans-
INTERPRETING VEGETATION INDICES 189
%, 1.0
0)

iiII//
N
0.8

E
t._
O 0.6
o-

0.4
n
Z
0.2

0.0 i
100 120 14o 18o 18o 2oo 220 240 260 280

Day of Year

Fig. 3. NDVI (dashed line) and RVI (solid line) values calculated from the smoothed data of
Fig. 2 and normalized to one to facilitate comparison.

formed into the other. Dividing both the numerator and the denominator of
eqn. (2) by 'red', we have
RVI- 1
N D V I - RVI
+----- i (3)

A prime reason for using mathematical transformations is to improve the

interpretation and visualization of the included information. Consider the
period from Day 120 to 180 of Fig. 3. The normalized values of NDVI and
RVI increased from 0 to about 0.8, and from 0 to about 0.2, respectively, and
thus it is graphically easier to assess vegetation amounts with the NDVI than
the RVI at low vegetation densities. However, after about Day 200, the NDVI
graph is essentially flat, showing little visual change as the season progresses.
The RVI graph increases and then decreases during this period, with a dis-
cernible visual change.

Linear combinations

VI, which are not functionally equivalent to ratios, can also be calculated
from linear combinations of two or more spectral bands. A graphical exam-
ple, using red and NIR bands, is depicted in Fig. 4. Point A in Fig. 4 corre-
sponds to the values of red and NIR reflectances of the dry soil shown in Fig.
1. Point B corresponds to the reflectance values for the wet soil and Point C
corresponds to vegetation. Thus, values of red and NIR data pairs represent-
ing any soil water content for bare soils would fall on the line connecting Points
A and B of Fig. 4. Point C represents full vegetative cover in r e d - N I R space.
As vegetation emerges from the soil, a r e d - N I R data pair would move toward
Point C, keeping within the bounds indicated by the dotted lines connecting
Points B - C and A-C. The shape formed by the dotted and solid lines is that
of a 'tasseled cap', as first noted by Kauth and T h o m a s (1976). Point D rep-
190 R.D. JACKSONAND A.R. HUETE

0.7

0.6

o 0.5 oc
C
o
o 0.4
i-i"".............
4-~ 0.3

~ o.2
Z
0.1

0.0 ' ' ' '

0.0 0.1 0.2 0.3 0.4 0.5

Red reflectonce
Fig. 4. Graphical representation o f the 'tasseled cap' and the PVI.

resents intermediate vegetative growth with the soil at an intermediate water

content. The perpendicular distance from Point D to the line A-B represents
the perpendicular vegetation index (PVI) of Richardson and Wiegand
(1977).
Kauth and Thomas (1976) applied the Gram-Schmidt technique (Frei-
berger, 1960) and, later, principle components analysis, to obtain four or-
thogonal linear combination indices (LCI) of the four LANDSAT multi
spectral scanner (MSS) bands. Any index of this type can be written in the
form
n

LCI= ~ aiXi (4)

i=l

where a is a coefficient, X is the band data, n is the number of bands and i is

the respective band number.
Kauth and Thomas selected clusters of pixels identified as soil and other
clusters identified as vegetation, and calculated unit vectors that form the
coefficients of four indices: brightness (BR), greenness (GN), yellowness
(YE) and nonsuch (NS). Their three-dimensional (3D) scatter plots of
BR-, GN- and YE-transformed data resembled a 3D tasseled cap, the name
they applied to their spectral-temporal description of agricultural crops. Bas-
ically, brightness represented a vector depicting the magnitude of reflected
energy, particularly that of soils (soil line), greenness represented an orthog-
onal plane containing all of the vegetation information, while yellowness
formed a plane orthogonal to both soils and vegetation. Yellowness has been
related to atmospheric haze and senescent vegetation. Nonsuch contains most
of the noise in the data (Lambeck et al., 1978; Kauth et al., 1979). Jackson
( 1983 ) reviewed the Kauth-Thomas procedure and presented a detailed ex-
INTERPRETING VEGETATION INDICES 191

ample of the calculation of four indices using the Gram-Schmidt technique

on data for four bands similar to LANDSAT MSS Bands 4-7. Figure 4 shows
a two-dimensional representation of the tasseled cap.
Concurrent with the development of the tasseled cap, Richardson and Wie-
gand (1977) proposed a PVI that utilized the red and NIR MSS bands of
LANDSAT. This two-dimensional index required the specification of a soil
line, with the index being the perpendicular distance from this line to the
vegetation point. The PVI is the simplest linear combination index to calcu-
late. An example of a PVI can be generated from data shown in Fig. 4. The
coordinates for Points A and B are used to calculate a line passing through the
two points. Next, a line perpendicular to the soil line that passes through any
vegetation point, say Point D, is calculated. The distance from Point D to the
intersection of the two orthogonal lines is the PVI. A detailed discussion of
the algebraic basis for the PVI calculation was given by Jackson et al. (1980).

C A L C U L A T I N G VI W I T H D I F F E R E N T I N P U T V A R I A B L E S

VI can be calculated using raw detector output voltages (V) or processed

data such as reflectances (p), radiances (L) or digital number ( D N ) . This is
an obvious, but sometimes overlooked factor when values of VI are com-
pared. Essentially, all satellite data are reported in digital numbers (these are
integers that generally range from 0 to 255 ), whereas ground-based radiome-
ter data are usually first recorded as voltage outputs.
Transforming raw data into radiance requires knowledge of the calibration
factors for each band on each instrument. Reflectance values can be calcu-
lated from either voltage output or radiance if a calibrated reference panel is
also measured. Calculation of surface reflectance factors from satellite data
requires the conversion of digital numbers to radiance, then the use of a ra-
diative transfer model to account for atmospheric scattering and absorption.
The above discussion becomes evident when expressed in equation form.
Thus, the RVI can be calculated from a simple voltage ratio
V, ir
RVI= (5)

or from a radiance ratio

RVI=~ VnirCnir (6)

or from a reflectance ratio

RVI = P.ir !target ! _ [ V.ir (target) / V.i~(panel) ]P.ir (panel)
(7)
P~ed(target) [Vred(target)/V~ed(panel)]Pred(panel)
192 R.D. JACKSON AND A.R. HUETE

Unless the calibration coefficients (C) for the two bands are equal (this is
seldom the case), it is obvious that eqns. (5) and (6) will yield different
values of RVI. Equation (7) does not require knowledge of C values, but does
require the reference panel calibration factors pa (panel).
Equations ( 5 ), ( 6 ) and ( 7 ) are most useful for ground-based instruments.
For satellite data, the RVI can be calculated from a simple ratio of digital
numbers, i.e.

RVI = DNn~r (8)

DNred
which does not account for sensor calibration or atmospheric effects. Price
( 1987 ) presented a detailed discussion concerning how the calibration fac-
tors and other sensor characteristics of different satellites affect the resulting
values of VI.
To obtain values of surface reflectance from digital numbers, the sensor
calibration constants must be known, and the optical depth of the atmosphere
must be measured at the time of data acquisition and used in a radiative
transfer model. With these data, the RVI can be calculated from the equation
RVI = OtnirEnir DNnir Cnir -k-finir (9)
O/redEred DN redfred + fired
where a and fi are constants obtained from a radiative transfer code, E is a
factor that includes the exoatmospheric irradiance in the waveband and the
solar zenith angle at the time of acquisition (Moran et al., 1990). It is appar-
ent that the calculation of surface reflectance from satellite data requires data
that are usually not readily available.
Although the above equations yield the RVI, all VI would be affected by
the use of different input data.

C O M P A R I S O N O F VI U S I N G DATA F R O M D I F F E R E N T SENSORS

It is frequently desirable to compare the same VI over time and space using
data from different sensors. A case in point is experiments in which ground-
based data are compared with aircraft- and satellite-based data. Care must be
exercised in making such comparisons because the band-response functions
for each instrument are different, the fields of view are usually different and
VI can be calculated using either raw or transformed data. The literature is
replete with relationships between various VI and biomass (or leaf area in-
dex) that were developed using data from hand-held or boom-mounted radi-
ometers. A comparison of these relationships with satellite-derived VI should
be made with caution. It is necessary to ascertain how the VI were calculated,
what sensor was used, and what the environmental and atmospheric condi-
tions were at the time of measurement.
INTERPRETING VEGETATION INDICES 193

The detectors and filters that are used to measure light within a particular
wavelength interval are unique to each instrument. The response function for
a particular band is a combination of the wavelength-dependent detector and
filter response. For example, LANDSAT-TM, SPOT-HRV and National
Oceanic and Atmospheric Administration (NOAA)-advanced very high res-
olution radiometer (AVHRR) sensors each have a red and a NIR band, and
the response of each is different for both bands (Fig. 5 ). In Fig. 5, the dashed
lines refer to the LANDSAT-TM response, the dash-dot-dash lines refer to
SPOT-HRV and the dotted lines to NOAA-AVHRR. Solid lines in the figure
refer to wheat and bare soil spectra (extracts of spectra shown in Fig. 1 ). The
response functions for LANDSAT-TM and SPOT-HRV are similar, but the
NOAA-AVHRR extends over a wider wavelength interval and is of a some-
what different shape.
The effect of the different response functions on VI values can be assessed
by calculating average reflectance factors for each band and each instrument.
This was accomplished by integrating the product of the response function
and the spectra values over the wavelengths for which the response functions
were non-zero, and dividing by the integral of the response function over the
same wavelength interval (Slater, 1980). The results are given in Table 1,
along with calculated VI.
For this theoretical example, wheat reflectance values were similar for the

1.0

" ; i .......~.~....
o.B ; li !~ .....
0.6 ..: (! |1 " Wheot

O 0.4 :! . ' I,-.,.-,'-'x,~":_,-'z------"~..-

t-

~'
(D ____J/i ~ ... o*'
k.. 0.0 ,. , ,,, , .. . . . . . . . -.... r'-
"0 0.8 0.7 O.B 0.9 1.0 1.1 .~0

0 Red .:" i " "'l. !. ~ ,¢~--

E o.s ......'t'l ~." ~ £x
o : iI ~'1
z 0.6 ": it !ii
? "I l: Wh~t
I . : ~ ~
o.4 it li I
0.2 ~ : - i/ -..
0.0 ~ ~ / - ~ J" ~"
0.5 0.6 0.7 0.8
Wovelength (/2)
Fig. 5. Nominalred and NIR responsefunctionsfor the TM (dashedlines), SPOT-HRV (dash-
dot-dash lines) and the AVHRR (dotted lines). Wheat and bare soil spectra are depicted by
solid lines.
194 R.D. JACKSONAND A.R.HUETE

TABLE 1

Reflectance values obtained by integrating the response functions and spectra in Fig. 5 (as described
in the text), and the resulting VI

LANDSAT-TM Spot-HRV NOAA-AVHRR

Wheat
Red reflectance 0.040 0.041 0.047
NIR reflectance 0.517 0.519 0.471
NIR/red 12.9 12.7 10.0
NDVI 0.856 0.854 0.819

Bare soil
Red reflectance 0.274 0.269 0.252
N1R reflectance 0.333 0.334 0.334
NIR/red 1.21 1.24 1.32
NDVI 0.097 0.108 0.134

LANDSAT-TM and SPOT-HRV, but were different for the AVHRR. The
AVHRR value for the N I R / r e d ratio was about 22% less than that for the
other two sensors. The difference was much less for bare soil. If spectra are
relatively flat over a wide range of wavelengths (as is soil in Fig. 5), the re-
sponse difference is small resulting in little difference between the VI. A word
of caution: a comparison of VI calculated from actual satellite data would be
different from the values shown in Table 1. Other factors, such as spatial res-
olution, sensor calibration and atmospheric effects, were not included in the
theoretical example.

EFFECTS OF SOIL B A C K G R O U N D ON VI

Until the soil is fully covered by vegetation, the soil background will influ-
ence the VI. For incomplete canopies, the wetting of a previously dry soil (or
vice versa) can cause a change in VI. The change is further complicated by
the fact that the transmission of light through vegetation is considerably greater
in the NIR than in the red band.
Both the ratio and the linear combination (orthogonal) classes of VI rely
on the existence of the soil baseline in red and NIR wavelength space for soil
normalization (Huete, 1988). The intercept of this line is close to, but does
not pass through, the origin (Fig. 6) and there is usually some scatter of soil
points away from the principal soil line. Such secondary soil influences are
most noticeable with red- and yellow-colored soils (Kauth and Thomas,
1976). These two factors affect the discrimination of low amounts of vegeta-
tion from bare soil, and are significant in arid regions and in the early stages
of vegetation growth (Huete et al., 1984).
Figure 6 depicts lines of constant vegetation amounts (isolines) as pre-
INTERPRETINGVEGETATIONINDICES 195
5 . 0 (0.661

,, 3.0
/ (0.50)
NIR I '
,' 2.0 ( 0 . 3 3 )
, ,,' ..-" ,,,
,' / /" 1.5 ( 0 . 2 0 )
,' ..,~'/ //.." / .-"
t / i / / / /
s'// ,' // ,'// -'*// I (0)
i [
i / /
_
o'~ \ /
.."1
L /
~f
,
,c'~,
" " D ,~
~/,-'/;./." DRY SOIL
• 12;'"
/ L / , f / . ,.
~'~ 11 l~'/.'." ~.
0 / I ~¢, ,',~.,'-'~
'~ .~I,¢'7,,',',~;..".,,I
~ _ *".- WET SOIL

; J ~"r
, / . / I Red
J2: / ~ . / I
......... I
E h I

Fig. 6. Spectra isolines of equal vegetation amounts in NIR-red wavelength space as predicted
by the RVI, NDVI and the PVI (adapted from Huete, 1988).

dicted by the RVI, NDVI and PVI. The ratio indices are graphically displayed
by vegetation isolines of increasing slopes diverging out from the origin, while
PVI isolines remain parallel to the soil line. To show soil influences on VI in
partially vegetated areas, consider a pixel representing a partial canopy over
a dry soil background (Point A, Fig. 6 ). If the soil background were to be-
come wet, a vegetation isoline bounded by dry and wet soil conditions would
be formed. In order for the ratio indices to effectively normalize such a back-
ground change, the pixel values would have to shift directly toward the origin
(B), following an isoline of constant RVI or NDVI values. The PVI, however,
would require the pixel values to shift along an isoline parallel to the soil line
(C), so that both the wet- and dry-soil partially vegetated pixels would main-
tain a constant PVI value (equidistant to the soil line). As the soil darkens on
wetting (Fig. 6), the actual pixel shifts to D and thus the PVI value of the
partially vegetated pixel (D) with wet soil is lower than that of the dry soil
(A), while the wet pixel RVI and NDVI values exceed those of the dry soil.
Huete ( 1988 ) developed a means to minimize the effect of soil background
on the ratio VI. He extended the line A - D (Fig. 6 ) to the point of intersection
with the soil line (E), and showed that by adding the distances l~ and/2 to the
red and NIR values, respectively, the resulting ratio indices would be only
minimally affected by soil background. Assuming that Ii =/2 and using the
format for NDVI, Huete (1988 ) defined the soil-adjusted vegetation index
(SAVI) as
N I R - red
S A V I - N I R + r e d + L ( 1+ L ) (10)
196 R.D. JACKSON AND A.R. HUETE

where L = l, + 12= 2l. The multiplicative factor ( 1 + L) was necessary to main-

tain the same bounds as the NDVI. Huete (1988) showed that optimal L
values were different for different vegetation amounts, but concluded that
L = 0 . 5 was optimal for a wide range of conditions. Although the SAVI was
developed using ground-based data, Huete and Warrick (1990) demon-
strated that it successfully minimized soil background effects using satellite
data.
The development of improved VI is an ongoing process. Major et al. (1990)
used modeled data (in contrast to Huete's experimental data) to adjust the
RVI for soil background. Baret and Guyot (1991 ) discussed the potentials
and limits of several VI for estimating leaf area index.

EFFECTS OF VIEW AND SOLAR ANGLES ON VI

Light is reflected uniformly in all directions from a lambertian surface, a

condition that rarely exists in nature. Vegetated surfaces are markedly non-
lambertian and light reflected from such a surface is highly dependent on view
and solar angles. This can be visualized by considering a row crop (e.g. cot-
ton) that has not reached full cover. If the rows are north-south, the soil be-
tween the rows will be shaded early and late in the day, and sunlit at midday.
If an instrument views the scene from directly overhead (nadir), the red re-
flectance would be low in the morning and increase as the sun rises because
of the increasing amount of sunlit soil. The NIR, however, would not change
nearly as much because roughly half of this light is transmitted through the

ENI . . . . . . . . .
o 51.4"
a 34.5"
50
0 025.5. o o o o o
o
P o
40 o & A a a °ao
"Io a
¢0 o
p,,
o oo . . . .
3o Oooo a o ~o a
Z =o " ' ~ , 6 " `= ~ ='
o o
10
1.00

0.95 o °°O°°°°A AA.2II a

o a u
0
:>
o A
° O a o o . 2 o ~ o a a
o oa°°°
Z AA~SeA o Sl.4"
0.90 o
o A 34.5"
o o 25.5"

0.85 ! , , , , , , , , ,
-50 -40 -30 -20 -10 0 10 20 30 40 50

View Angle

Fig. 7. Effect of view angle on the NDVI and the RVI at three solar zenith angles over a wheat
canopy.
INTERPRETING VEGETATION INDICES J 97

40

B
30
r~
np
Z

20 i i
1.00

0.95
~>
m

Z
o.go

0.85
20 30 40 50 80

Solar Zenith Angle

Fig. 8. Effect of solar zenith angle on the NDVI and the RVI at one view angle (nadir) over a
wheat canopy.

canopy and can be reflected from the shaded soil. This explanation is greatly
simplified. The actual situation is extremely complex. Nevertheless, it be-
comes obvious that different view and solar angles will cause different values
of VI to be obtained for the same surface.
View angle and solar angle effects on radiation reflected on the surface have
been discussed at length in the literature (see Pinter et al. (1983) for an ex-
tensive list of references). Their effects on VI have been reported by Pinter et
al. (1983, 1987, 1990) and Jackson et al. (1979, 1990). Examples of these
effects are shown in Figs. 7 and 8. Figure 7 shows changes in the N I R / r e d
ratio and the NDVI as caused by view angle changes (for three solar angles)
over a wheat canopy. At the largest solar zenith angle (51.4 ° ), the N I R / r e d
ratio more than doubled as view angles changed from - 4 5 ° (viewing toward
the sun) to + 45 o (viewing away from the sun), whereas the NDVI changed
from about 0.88 to 0.96. Figure 8 shows changes in the two indices with
changes in the solar zenith angle, at a constant view angle (nadir). Again, the
N I R / r e d ratio nearly doubled for a sun angle change of less than 30 °, and the
NDVI changed from about 0.92 to 0.94 for the same range of sun angles. The
NDVI appears to be less sensitive to view and solar angle changes than is the
N I R / r e d ratio; however, the data were obtained for a high vegetation density
where the NDVI is least sensitive.

ATMOSPHERIC EFFECTSON VI

Obtaining quantitative results from satellite data requires an accurate ac-

counting for atmospheric effects. VI are not i m m u n e to these effects. Switzer
198 R.D. JACKSON AND A.R. HUETE

et al. ( 1981 ) pointed out that atmospheric path radiance contributes to falsely
low values of the N I R / r e d ratio in LANDSAT data. Dave ( 1980, 1981 ) sim-
ulated satellite data and demonstrated the effect of different atmospheres on
the N I R / r e d ratio and the tasseled cap transformation of Kauth and Thomas
(1976). Jackson et al. (1983) examined the effect of the atmosphere on a
number of vegetation indices by simulating LANDSAT MSS data for a data
set comprising season-long reflectance measurements over wheat. In general,
both ratio and linear combination indices decrease with increasing atmos-
pheric turbidity.

EFFECT OF CANOPY ARCHITECTURE ON VI

The architecture of a vegetation canopy determines the directions that ra-

diation will be reflected from plant surfaces. The vertical elements of an er-
ectophile canopy trap reflected radiation within the canopy, with a corre-
sponding reduction in the amount reflected vertically towards a nadir-oriented
radiometer. The opposite is true for a planophile canopy. The horizontal leaves
reflect more in the vertical direction and less is trapped within the canopy
(Suits, 1972; Bunnik, 1978). A nadir-pointing sensor can receive 20-30%
more reflected radiation from a planophile canopy than from an erectophile
canopy.
Canopy architectural effects on VI were examined by Jackson and Pinter
( 1986 ). They showed that during the period of peak green vegetation densi-
ties, the RVI was about 30% higher for an erectophile canopy than for a plan-
ophile canopy. On the other hand, the PVI was about 30% higher for the plan-
ophile canopy than for the erectophile canopy. Thus, within a scene that
contains similar amounts of vegetation, but different geometries, different VI
will respond differently to canopy architecture. At first glance, this compli-
cating feature would diminish the value of VI. In fact, it adds a new perspec-
tive that could possibly be exploited to provide otherwise unobtainable infor-
mation on canopy architectural features within a scene.

CONCLUDING REMARKS

VI are useful for assessing the amount and condition of vegetation using
data from ground-, aircraft- and satellite-based sensors. However, atmos-
pheric constraints, sensor view and solar zenith angles, as well as soil back-
ground and canopy architecture, must be accounted for if quantitative com-
parisons of biomass and plant health are to be expected. The effective use of
VI requires that the complex interplay of all the above influences be simulta-
neously considered. It is important that the VI is able to respond to subtle
changes in plant health status amidst variable view, illumination and atmos-
pheric conditions.
INTERPRETING VEGETATION INDICES 199

ACKNOWLEDGMENT

We thank Paul J. Pinter, Jr. for providing the reflectance data for cotton
used in Figs. 2 and 3.

REFERENCES

Baret, F. and Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR
assessment. Remote Sensing Environ., 35:161-173.
Bunnik, N.J.J., 1978. The Multispectral Reflectance of Shortwave Radiation by Agricultural
Crops in Relation with their Morphological and Optical Properties. Veenman and Zonen,
Wageningen, 175 pp.
Dave, J.V., 1980. Effect of atmospheric conditions on remote sensing of vegetation parameters.
Remote Sensing Environ., 10: 87-99.
Dave, J.V., 1981. Influence of illumination and viewing geometry and atmospheric composition
of the "'tasseled cap" transformation of Landsat MSS data. Remote Sensing Environ., 11:
37-55.
Deering, D.W., 1978. Rangeland reflectance characteristics measured by aircraft and spacecraft
sensors. Ph.D. Dissertation, Texas A & M University, College Station, TX, 338 pp.
Freiberger, W.F. (Editor), 1960. The International Dictionary of Applied Mathematics. Van
Nostrand, Princeton, N J, p. 412.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing Environ., 25:
295-309.
Huete, A.R. and Warrick, A.W., 1990. Assessment of vegetation and soil water regimes in par-
tial canopies with optical remotely sensed data. Remote Sensing Environ., 32: 155-167.
Huete, A.R., Post, D.F. and Jackson, R.D., 1984. Soil spectral effects on 4-space vegetation
discrimination. Remote Sensing Environ., ! 5:155-165.
Jackson, R.D., 1983. Spectral indices in n-space. Remote Sensing Environ., 13: 409-421.
Jackson, R.D. and Pinter, Jr., P.J., 1986. Spectral response of architecturally different wheat
canopies. Remote Sensing Environ., 20: 43-56.
Jackson, R.D., Pinter, Jr., P.J., ldso, S.B. and Reginato, R.J., 1979. Wheat spectral reflectance:
interactions between crop configuration, sun elevation, and azimuth angle. Appl. Optics, 18:
3730-3732.
Jackson, R.D., Pinter, Jr., P.J., Reginato, R.J. and ldso, S.B., 1980. Hand-held radiometry.
USDA-SEA-AR Agricultural Reviews and Manuals, US Department of Agriculture, Oak-
land, CA, W-19, 66 pp.
Jackson, R.D., Slater, P.N. and Pinter, Jr., P.J., ! 983. Discrimination of growth and water stress
in wheat by various vegetation indices through clear and turbid atmospheres. Remote Sen-
sing Environ., 13: 187-208.
Jackson, R.D., Teillet, P.M., Slater, P.N., Fedosejevs, G., Jasinski, M.F., Aase, J.K. and Moran,
M.S., 1990. Bidirectional measurements of surface reflectance for view angle corrections of
oblique imagery. Remote Sensing Environ., 32: 189-202.
Jordan, C.F., 1969. Derivation of leaf area index from quality of light on the forest floor. Ecol-
ogy, 50: 663-666.
Kauth, R.J. and Thomas, G.S., 1976. The tasseled cap---A graphic description of the spectral-
temporal development of agricultural crops as seen by Landsat. Proceedings of Symposium
on Machine Processing of Remotely Sensed Data, Purdue University, West Lafayette, IN,
pp. 41-51.
Kauth, R.J., Lambeck, P.F., Richardson, W., Thomas, G.S. and Pentland, A.P., 1979. Feature
200 R.D.JACKSONANDA.R.HUETE

extraction applied to agricultural crops as seen by Landsat. Proceedings of LACIE Sympo-

sium, NASA/Johnson Space Center, Houston, TX, pp. 705-721.
Lambeck, P.F., Kauth, R.J. and Thomas, G.S., 1978. Data screening and preprocessing for
Landsat MSS data. Proceedings of 12th International Symposium on Remote Sensing of
Environment, ERIM, Ann Arbor, MI, pp. 999-1008.
Major, D.J., Baret, F. and Guyot, G., 1990. A ratio vegetation index adjusted for soil brightness.
Int. J. Remote Sensing, 11: 727-740.
Moran, M.S., Jackson, R.D., Hart, G.F., Slater, P.N., Bartell, R.J., Biggar, S.F., Gellman, D.I.
and Santer, R.P., 1990. Obtaining surface reflectance factors from atmospheric and view
angle corrected SPOT-1 HRV data. Remote Sensing Environ., 32:203-214.
Perry, C.R. and Lautenschlager, L.F., 1984. Functional equivalence of spectral vegetation in-
dices. Remote Sensing Environ., 14: 169-182.
Pinter, Jr., P.J., Jackson, R.D., Idso, S.B. and Reginato, R.J., 1983. Diurnal patterns of wheat
spectral reflectances. IEEE Trans. Geosci. Remote Sensing, GE-21: 156-163.
Pinter, Jr., P.J., Zipoli, G., Maracchi, G. and Reginato, R.J., 1987. Influence of topography and
sensor view angles on NIR/red ratio and greenness vegetation indices of wheat. Int. J. Re-
mote Sensing, 8: 953-957.
Pinter, Jr., P.J., Jackson, R.D. and Moran, M.S., 1990. Bidirectional reflectance factors of ag-
ricultural targets: A comparison of ground-, aircraft-, and satellite-based observations. Re-
mote Sensing Environ., 32:215-228.
Price, J.C., 1987. Calibration of satellite radiometers and the comparison of vegetation indices.
Remote Sensing Environ., 21: 15-27.
Richardson, A.J. and Wiegand, C.L., 1977. Distinguishing vegetation from soil background in-
formation. Photogramm. Eng. Remote Sensing, 43:1541-1552.
Slater, P.N., 1980. Remote Sensing: Optics and Optical Systems. Addison-Wesley, Reading,
MA, 575 pp.
Suits, G.H., 1972. The calculation of the directional reflectance of a vegetative canopy. Remote
Sensing Environ., 2:117-125.
Switzer, P., Kowalik, W.S. and Lyon, R.J.P., 1981. Estimation of atmospheric path-radiance by
the covariance matrix method. Photogramm. Eng. Remote Sensing, 47:1469-1476.
Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegeta-
tion. Remote Sensing Environ., 8:127-150.