Professional Documents
Culture Documents
2, MARCH 1995 48 1
I. INTRODUCTION
Type I: The most widely used index in the processing of where IC = 1 / [2 \rho(X)]. The related Soil and Atmo-
satellite data is the Normalized Difference Vegetation Index spherically Resistant Vegetation Index [ 181 can be similarly
+
(NDVI) defined as [(PN - P R ) / ( P N P R ) ] , where P N and +
expressed (6) with $k = ( l + a ) / [2 p(X) a]. These indexes
P R are spectral bidirectional reflectance factors (ratio of the are therefore grouped in (k d 2 p / d X 2 ) (Fig. 1).
radiance of a target surface to the radiance of a conservative,
lambertian surface) at near-infrared and red wavelengths, 111. THE SPECTRAL DERIVATIVE
respectively [ll]. To obtain a continuous form let NDVI = The reflectance of a vegetated surface depends on the struc-
AV, p~ = p(X + AX) and p~ = p(X). Note that tural and optical properties of the vegetation and underlying
soil. In the case of remote directional measurements, the direc-
p(X+AX) - p ( X ) = - dP
dX
+AX @[(AX)2]
tions of incident solar radiation and observation also determine
surface reflectance. However, only the optical properties of
2 vegetation and soil control the spectral dependence of surface
a.I,
A+AA
p(X+AX) +p(X) = +
dX' p(X') @[(AX)']. reflectance [19].Therefore, p(X) = F [ p s ( X ) ,W L ( X ) ] , where
W L is leaf albedo, ps is soil hemispherical reflectance, F is
(2) canopy reflection function and p is bidirectional reflectance
Here @ ( A X ' ) denotes error of order AX2. In the limit factor of the vegetated surface at wavelength A.
(AX -+ 0) Radiation incident on a leaf may be specularly reflected at
the surface, the magnitude of which is usually assumed to be
dV dp
- = -k ( 3 ) small. If this is not the case, p must be assumed to denote the
dX dX diffuse reflectance of a canopy only. Radiant energy reaching
where IC = [1/2p(X)J. The Soil Adjusted Vegetation Index, the interior of a leaf can be absorbed by the pigments, water
designed to minimize the soil effect in a vegetation signal [ 121, and other constituents. The principal mechanism of scattering,
+
can be similarly expressed (3) with k = (l+a)/[2 p(X) a]; defined here as change in direction of photon travel, is due
a is a constant. The Simple Ratio (SR) is equivalent to NDVI to refractive index discontinuities at cell wall-air interfaces.
because NDVI = (SR - 1) I SR + 1) and therefore IC = Radiation not absorbed inside a leaf emerges diffused on both
[l/P(X)l. sides of the leaf [20]. The leaf albedo WL is thus the sum of
A related class of indexes employs a weighted contrast leaf hemispherical reflectance and transmittance.
in redhear-infrared reflectance to minimize soil effects. The The optical system of a leaf can be modelled as a pile of
continuous forms of these indexes (Weighted Difference-, transparent plates. Each plate represents a hypothetical layer
Perpendicular- and Transformed Soil Adjusted-Vegetation In- of leaf-interior of unit thickness [21]. The number of plates is
dex) can also be expressed in a similar manner (Appendix A). independent of wavelength. The spectral dependence of leaf
Therefore, indexes containing a simple or weighted contrast albedo is governed by the transmittance K of a single such
can be grouped in ( k d p / d X ) (Fig. 1). plate and the refractive index v of the cell walls. The latter
Type 11: Vegetation indexes in this category are non-linear however is a weak function of wavelength and a constant value
because they contain products of reflectance. For example, of -1.4 can be assumed [211. Thus, W L ( X ) = P[K(X)];P is
the Global Environment Monitoring Index (GEMI), designed the leaf albedo function.
to minimize atmospheric effects in Advanced Very High Bare soil reflectance ps is a function of soil moisture
Resolution Radiometer data [16], can be shown (Appendix and physical and optical properties of the soil particles [22].
B) to conform to Of these, only the single scattering albedo of the particles
ws depends on the wavelength and soil moisture content
d V - dp'
-
dX
-k
dX (4) [23]. Therefore, ps(X) = Q [ w s ( X ) ] ; Q is the soil reflection
function.
where k = 2 / [2 \rho(X)+ 0.51. Another index in this category In light of the above discussion, the spectral derivative can
is the Greenness [17], defined as A;pi. The evaluation of be expanded as
dV/dX in the limit (AX + 0 ) is straightforward (Appendix C) d F dP d n
d p % - -d -F+ -d-Q- dws
-
(7)
dV dpp' dX dps d w s dX ~ W L dX'
- K - k.j .
dX dX The functions F, Q, and P describe radiative transfer in
The derivative dp2/dX is shown in Fig. 1. a canopy of leaves layered above a soil surface, a semi-
Type 111: The Atmospherically Resistant Vegetation In- infinite medium of soil particles and the interior of a leaf
dex, developed to minimize atmospheric effects in Moderate modelled as a pile of transparent plates, respectively. The
Resolution Imaging Spectrometer data [18], is defined as governing equations of transfer are linear integro-differential
equations (241. The solutions can be expressed formally as
[(PA' - P R , B ) / ( P W + P R , B ) ] , where PR,B = P R - ?' ( P B - P R )
and p~ is reflectance at blue wavelength (-0.45 pm). The a sum of exponential functions, that is, the photon count
continuous form of this index is (Appendix D) decays exponentially through successive absorption and scat-
tering events in the media. The partial derivatives ( d F / d p s ,
d'V - d 2 p d F / d w L , d Q / d w s and aP/atc) are therefore exponential
- (6)
dX2 Z k functions-smooth and smaller in magnitude than the total
MYNENI et 01.: THE INTERPRETATION OF SPECTRAL VEGETATION INDEXES 483
70.00 -
60.00 -
50.00 -
40.00-
30.00 -
d (Inp)
- 20.00 -
dh
10.00 -
000 -
-
::L
- ’a:
.IO.yJ
-5 00
- - ‘ . I L L
000
- dx+ -
500
d%
1000
1
1500
VARIABLE OF DIFFERENTIATION dh dh
Fig. 2. Partial derivatives of the canopy reflection function F , soil reflection Fig. 4. The relationship between the spectral derivative of canopy reflectance
function Q and leaf albedo function P. W L is leaf albedo, us is single and spectral derivatives of leaf and soil optical properties. Soil, leaf, and
scattering albedo of soil particles, K is plate transmittance and ps is soil canopy spectra were measured in August 1989 at a natural grassland prairie
hemispherical reflectance. Other parameters are as in Fig. 1. (site 916) [30]. The soil reflectance spectrum was inverted using a model [23]
to obtain the spectrum of soil-particle single scattering albedo w s ( X ) . The
plate transmittance spectrum .(A) was estimated by inverting a model [25]
with measured leaf reflectance spectra. Note that the ordinate is equivalent to
the continuous form of NDVI.
-241.0
-__ IV. THE SPECTRALABSOIUTON COEFFICIENT
0.0 1.0 2.0 3.0
WAVELENGTH (MICRO METER) The plate transmittance K. is equivalent to the spectral
Fig. 3. The derivatives of plate transmittance K and single scattering albedo absorption coefficient a of an assembly of absorbers located
of soil particles ws with respect to wavelength A. us data are for slightly in the plate
moist clayey (A), peaty (B) and sandy (C) soils [23]. 6 is for the soybean
leaf discussed in [25]. .(a) = (1 - a ) e x p ( - a ) + a 2 E l ( a ) (10)
5.0 -' I This may be the basis for the observed correlations be-
tween redhear-infrared indexes and vegetation amount [3],
the fraction of absorbed photosynthetically active radiation [4],
4.0 1 unstressed vegetation conductance and photosynthetic capacity
[ 5 ] , and seasonal atmospheric carbon dioxide variations [ 6 ] .
VI. APPENDIXA
The Weighted Difference Vegetation Index is ,defined as
(pN - a p ~ )where
, +
a is a constant [131. Let Vw(X AX) =
+
[p(X AX) - a p ( J ) ] and Vw(X) = [(I - a)p(X)].Thus
AVw/AX = Ap/AX and in the limit ( A x -+ 0)
A 4.0
+
0.0-
0.0
1
20.0
I
40.0
d
60.0 80.0
O.O The related Perpendicular Vegetation Index is defined as
c ( p - R b), where a, b and c are constants [14]. Let
~ U ~ -
LEAF CHLOROPHYLL CONTENT @g1 Cm2 leaf area)
Vp(X+AX) = c[Vw(X+AX)-b]and Vp(X)= c[Vw(X)-b].
Fig. 5. The relationship between the discrete derivative of plate transmit-
tance n and soybean leaf chlorophyll content. A similar relationship at the Thus AVp/AX = cAp/AX and in the limit (AX -+ 0)
canopy scale is also shown; p is modelled canopy reflectance [19]. Other
parameters are as in Fig. 1.
L i p i denotes the concentration of the ith-absorber species per A recent index in this category is the Transformed Soil
unit ground area. Therefore Adjusted Vegetation Index (TSAVI) defined as [15]
N N TSAVI = a(pN-apR-b)
(-43)
+
~ P N PR - c
Let
that is, the spectral derivative is indicative of the abundance
and activity of the absorbers pertaining to radiation absorption.
In fact d p / d X = -dA/dX, where A is the fraction of incident
radiant flux density absorbed by an optically dense canopy of and
horizontal lambertian leaves, because p +
A = 1 from en-
ergy conservation. Therefore, sensing a surface remotely with
measurements of scattered radiation is equivalent to inferring
radiation absorption by the surface. This is the theoretical
Therefore AV,/AX = k Ap/AX and in the limit (AX -+ 0)
basis sought for satellite remote sensing of leaf biochemical
constituents (Appendix G ) . The related forms d 2 p / d X 2 and dVT - d p
d p 2 / d X can be similarly interpreted. It is therefore concluded dX-zk
that spectral vegetation indexes are indicative of the abundance
and activity of the absorbers in the leaves. The indexes saturate where k = a / [ (l+a) \rho(X) - c] (2).
at sufficiently high values of L,pi (1 1).
VII. APPENDIXB
V. F&D/NEAR-INFRARED VEGETATION
INDEXES The Global Environment Monitoring Index is a quadratic
The in vivo absorption spectra of chlorophylls a & b in q where
peak at about 0.69 pm and approach zero at 0.85 pm [25].
Following (9) and (12), An/AX and Ap/AX evaluated at
these wavelengths are indicative of chlorophyll abundance and
energy absorption (Fig. 5) - a result confirmed experimentally Let q, =4 p ( X ) / [2 p(X)+O.5] and AV = (q - q,). In the
[28]. The broad-band rednear-infrared vegetation indexes can limit (AX + 0) (4) results with IC = 2 / [2 p(X)+ O S ] .
MYNENI et al.: THE INTERPRETATION OF SPECTRAL VEGETATION INDEXES 485
Here ps is soil reflectance, k1 = 1/G and k2 = D / B G [ B , where el = exp(p) - exp(-p), e2 = exp(p) exp(-p), +
D and G are defined in [17]. p = WL, ps is soil hemispherical reflectance and L is leaf
area index [W, X and V are defined in [26]. The derivative
dK./dA is (lo), (see (F4), shown at the bottom of the previous
IX. APPENDIXD page) where a, = ~ ~ 6Inserting
% . (F2) and (F4) in (Fl) results
+ +
Let PB = p(X), PR = p(X AX), p~ = p(X 2AA) and in (11).
y = 1 [18]. Then
XII. APPENDIX
G
The apparent reflectance of a vegetated surface p measured
at the top of a cloudless atmosphere is related to surface
+ AX)^] (Dl) reflectance p as: p(X) = T [ 7 a ( A ) , W a ( X ) , g a ( A ) , P ( A ) ] . The
anisotropy parameter ga is a weak function of wavelength
and can be assumed constant. The aerosol optical depth T~
(PN+PR,B) = [A X+2AX
dA’ p(A’) + AX a
dp and single scattering albedo w a vary near-linearly between
0.4 - 2.2km [27]. The function T describes radiative transfer
+ @[(AX)2]. In an aerosol atmosphere. Its partial derivatives are therefore
exponential and smaller in magnitude than (d2p/dX21.Thus
(W
d2p
- d2p
Let ARVI = A 2 V and in the limit (AA + 0) (6) results. dX2 cc dX2
that is, the apparent surface reflectance, absent of gaseous
X. APPENDIX
E and molecular interactions, is proportional to the true surface
The second order spectral derivative reflectance.
ACKNOWLEDGMENT
This research was funded by the Terrestrial Ecology Pro-
can be simplified to gram of N A S A . Dr. Wickland provided much needed Lmpetus
to this topic. The contributions of Drs. Ganapol and Jacque-
d2p d2Ws d2K moud are acknowledged with gratitude.
S0:2F+dX2 (E21
REFERENCES
because ld2ws/dA21 and ld2n/dX21are several orders of mag-
nitude greater ‘than ‘the absolute values of dps/dX, [I] P. J. Sellers and D. S. Schimel, “Remote sensing of the land biosphere
and biogeochemistry in the EOS era: Science priorities, methods and
dwL,dA, dK.ldX, (dwsldA12, (dK.ldA)z and all the first and implementation - EOS land biosphere and biogeochemical cycles
second order partial derivatives of F , Q, and P with respect panels,” Global Planetary Change, vol. 7, pp. 279-297, 1993.
to PS, u s , WL,and K . [2] G. Asrar and D. J. Dokken, Eds., EOS Reference Handbook.Washington,
DC: NASA, 1993.
[3] C. J. Tucker, “Red and photographic infrared linear combination for
monitoring vegetation,” Remote Sensing Environ., vol. 8, pp. 127-150,
XI. APPENDIXF 1919.
[4] G. Asrar, M. Fuchs, E. T. Kanemasu, and J. L. Hatfield, “Estimating
Consider the case of an optically dense canopy of Lamber- absorbed photosynthetic radiation and leaf area index from spectral
tian, horizontal leaves. The spectral derivative is (7) reflectance in wheat,” Agron. J., vol. 76, pp. 300-306, 1984.
[5] P. J. Sellers, J. A. Berry, G. J. Collatz, C. B. Field, and F. G . Hall.
“Canopy reflectance, photosynthesis and transpiration, 111. A reanalysis
(F1) using improved leaf models and a new canopy integration scheme,”
Remote Sensing Environ., vol. 42, pp. 187-216, 1992.
486 IEEE TRANSACl’IONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 33, NO. 2, MARCH 1995
[6] C. J. Tucker, Y. Fung, C. D. Keeling, and R. H. Gammon, “Relationship Ranga B. Myneni received the Ph.D. degree in
between atmospheric COz variations and a satellite-derived vegetation biology from the University of Antwerp, Antwerp,
index,” Nature, vol. 319, pp. 195-199, 1986. Belgium, in 1985.
[7] C. J. Tucker, J. R. G. Townshend, and T.E. Goff, “African land-cover He was previously employed by Kansas State
classification using satellite data,” Sci., vol. 227, pp. 369-375, 1985. University, Manhattan, and Georg-August Univer-
[8] A. F. H. Goetz, G. Vane, I. Solomon, and B. N. Rock, “Imaging sitat Gottingen, Germany, prior to joining Goddard
spectrometry for earth remote sensing,” Sci., vol. 228, pp. 1147-1153, Space Flight Center as an employee of the Uni-
1985. versity of Maryland, College Park. His research
[9] C. A. Wessman, J. D. Aber, D. L. Peterson, and J. A. Melillo, “Remote interests are radiation transport and remote sensing
sensing of canopy chemistry and nitrogen cycling in temperate forest of vegetation.
ecosystems,” Nature, vol. 335, pp. 154-156, 1988.
[IO] B. N. Holben, “Characteristics of maximum value composite images for
temporal AVHRR data,” Int. J. Remote. Sensing, vol. 7, pp. I4 17- 1437,
1986.
[ I l l J. W. Rouse, R. H. Hass, J. A. Schell, and D. W. Deering, ”Monitoring
vegetation systems in the Great Plains with ERTS,” in 3rd ERTS Symp.,
NASA SP-351, U.S. Gov. Printing Office, Washington, DC, vol. I , pp. Forrest G. Hall received the B.S. degree from the
309-317, 1973. University of Texas, Austin, and the M.S. and Ph.D.
[I21 A. R. Huete, ”A soil adjusted vegetation index,“ Remote Sensing degrees in physics from the University of Houston,
Environ., vol. 25, pp. 295-309, 1988. Houston, TX.
[I31 J. G. P. W. Clevers, “The application of a weighted infrared-red He is with NASA Goddard Spaceflight Center’s
vegetation index for estimating leaf area index by correcting for soil Laboratory for Terrestrial Physics, where he cur-
moisture,” Remote Sensing Environ., vol. 29, pp. 25-37, 1989. rently serves as Project Manager for the Boreal
[I41 A. J. Richardson and C. L. Wiegand, “Distinguishing vegetation from Ecosystem-Atmosphere Study. During his 25-year
soil background information,” Photogramm. Eng., vol. 43, pp. 1541- NASA career, he has worked and published in the
1552, 1977. areas of spacecraft design, modeling planetary atmo-
[IS] F. Baret and G. Guyot, “Potentials and limits of vegetation indexes spheres, microwave remote sensing of sea surface
for LA1 and APAR assessment,” Remote Sensing Environ., vol. 35, pp. wind fields, satellite calibration and atmospheric effects removal, image
161-174, 1991. processing, mathematical pattern recognition, statistical sampling, global crop
[ 161 B. Pinty and M. M. Verstraete, “GEMI: A non-linear index to monitor monitoring, climatehiosphere interactions, and forest ecosystem dynamics.
global vegetation from satellites,” Vegetatio, vol. 101, pp. 15-20, 1992.
1171 R. D. Jackson, “Spectral indexes in n-spaces,” Remote Sensing Environ.,
vol. 13, pp. 409-421, 1983.
11 81 Y. J. Kaufman and D. TanrC, “Atmospherically resistant vegetation index
(ARVI) for EOS-MODIS,” IEEE Trans. Geosci. Remote Sensing, vol.
30, 261-270, 1992.
[19] R. B. Myneni, G. Asrar, and S. A. W. Gerstl, “Radiative transfer in Piers J. Sellers received the B.Sc. degree in ecological science from the
three dimensional leaf canopies,” Trans. Theory Stat. Phys., vol. 19, pp. University of Edinburgh, Edinburgh, Scotland, and the Ph.D. degree in
205-250, 1990. vegetation-atmospheric interactions from k e d s University, k e d s , UK, in
[20] L. Grant, “Diffuse and specular characteristics of leaf reflectance,” 1976 and 1981, respectively.
Remote Sensing Environ., vol. 22, pp. 309-322, 1987. Since 1982, he has bee working at the University of Maryland, College Park,
1211 W. A. Allen, H. W. Gausman, A. J. Richardson, and J.R. Thomas, and NASA Goddard Space Flight Center, on projects related to land biosphere-
“Interaction of isotropic light with a compact plant leaf,” J. Opt. SOC. atmosphere interactions including integration of biosphere models and general
Am., vol. 59, pp. 1376.1379, 1969. circulation models, remote sensing of biophysical function, and design and
[22] B. Pinty, M. M. Verstraete, and R. E. Dickinson, “A physical model for execution of field experiments (FIFE and BOREAS). He is currently an
predicting bidirectional reflectances over bare soils,’’ Remote Sensing interdisciplinary science investigator with the Earth Observing System and
Environ., vol. 27, pp. 273-288, 1989. serves as the EOS-AM platform project scientist.
[23] S. Jacquemoud, F. Baret, and J.F. Hanocq, “Modeling spectral and
bidirectional soil reflectance,” Remote Sensing Environ., vol. 41, pp.
123-132, 1992.
124) S. Chandrasekhar, Radiative Transfer. New York Dover, 1960.
[25] S. Jacquemoud and F. Baret, “PROSPECT: A model of leaf optical
properties spectra,” Remote Sensing Environ., vol. 34, pp. 75-91, 1990.
[26] J. A. den Dulk, “The interpretation of remote sensing, a feasibility Alexander L. Marshak received the M.S. degree in
study,” PhD dissertation, Agricultural Univ. Wageningen, Netherlands, applied mathematics from Tartu University, Estonia,
1989, p. 145. and the Ph.D. degree in numerical analysis from
[27] D. Tank, C. Deroo, P. Duhaut, M. Herman, and J. J. Mor- the Siberian Branch of the Russian Academy of
crette,”Simulation of the satellite signal in the solar spectrum (5S),” Sciences, Novosibirsk, Russia, in 1978 and 1983,
Laboratoire d’Optique AtmosphCrique, Universitk des Sciences et respectively.
Techniques de Lille, 59655 Villeneuve d’ Ascq CCdex, France, p. 83-84, From 1978 to 1989, he was a Research Scientist
1987. at the Institute of Astrophysics and Atmospheric
[28] C. Buschmann and E. Nagel, “In vivo spectroscopy and internal optics Physics, Tartu, Estonia. He was an Alexander von
of leaves as basis for remote sensing of vegetation,” Znt. J. Remote Humboldt Fellow from 1989 to 1991, and worked
Sensing, vol. 14, pp. 711-722, 1993. at the University of Gottingen, Germany. He is
[29] K. J. Ranson, L. L. Biehl, and C. S. T. Daughtry, “Soybean canopy currently a Research Scientist in the Climate and Radiation Branch of NASA
modelling data set,” LARS Tech. Rep. 071584, Purdue Univ., W. Goddard Space Flight Center. He has been actively involved in radiative
Lafayette, Indiana, 1984. transfer in atmosphere and vegetation since 1980. Since 1991, his research
[30] P. J. Sellers, F. G. Hall, G. Asrar, D. E. Strebel, and R. E. Murphy, “An interests have been in nonlinear processes in geophysics and scale-invariant
overview of the First International Satellite Land Surface Climatology analysis of geophysical data. Currently, he IS involved in theoretical cloud
Project (ISLSCP) Field Experiment (FIFE),” J. Geophys. Res., vol. 97, radiation studies in support of the Atmospheric Radiation Measurement
pp. 18345-18372, 1992. program.