Use of a green channel in remote sensing of global vegetation from EOS-MODIS

Use of a green channel in remote sensing of global vegetation from EOS-MODIS

ELSEVIER Use of a Green Channel in Remote Sensing of Global Vegetation from EOS-MODIS Anatoly A. Gitelson,* Yoram J. Kaufman, + and Mark N. Merzlyak ...

994KB Sizes 0 Downloads 4 Views


Use of a Green Channel in Remote Sensing of Global Vegetation from EOS-MODIS Anatoly A. Gitelson,* Yoram J. Kaufman, + and Mark N. Merzlyak M o s t animals use a "green" spectral range to remotely sense the presence and vitality of vegetation. While humans possess the same, ability in their eyes, man-made space-borne sensors that sense evolution of global vegetation, have so far used a combination of the red and near infrared channels instead. In this article we challenge this approach, using measurements of reflectance spectra from 400 nm to 750 nm with spectral resolution of 2 nm, with simultaneous determination of pigment concentrations of mature and autumn senescing leaves. We show that, for a wide range of leaf greenness, the maximum sensitivity of reflectance coincides with the red absorption maximum of chlorophyll-a (Chl-a) at 670 nm. However, for yellow-green to green leaves (with Chl-a more than 3-512g/cm2), the reflectance near 670 nm is not sensitive to chlorophyll concentration because of saturation of the relationship of absorptions versus chlorophyll concentration. Maximum sensitivity of Chl-a concentration for a wide range of its variation (0.3-45 lag/cm 2) was found, not surprisingly so, around the green band from 520 nm to 630 nm and also near 700 nm. We found that the inverse of the reflectance in the green band was proportional to Chl-a concentration with correlation r2> 0.95. This band will be present on several future satellite sensors with a global view of vegetation (SeaWiFS to be launched in 1996, Polder on ADEOS-1 also in 1996, and MODIS on EOS in 1998 and 2000). New indexes that use the green channel and are resistant to atmospheric effects are developed. A green NDVI = (pnir--Pgreen)/ (Pnir"+-Pgreet0 w a s tested for a range of Chl-a from 0.3 I~g/ cm 2 to 45 I~g/cm 2, and found to have an error in the

*J. Blaustein Institute for Desert Research, Ben-Gurion University of the Negev, Sede-Boker Campus, Israel *NASA Goddard Space Flight Center, Greenbelt, Maryland *Department of Cell Physiology and Immunology, Faculty of Biology, Moscow State University, Moscow, Russia Address correspondence to Anatoly A. Gitelson, J. Blaustein Inst. for Desert Research, Ben-Gurion Univ. of the Negev, Sede-Boker Campus, 84993 Israel. Received 8 September 1995; revised 23 March 1996. REMOTE SENS. ENVIRON. 58:289-298 (1996) ©Elsevier Science Inc., 1996 655 Avenue of the Americas, New York, NY 10010

chlorophyll a derivation at leaf level of less than 3 llg / cm 2. The new index has wider dynamic range than the NDVI and is, on average, at least five times more sensitive to Chl-a concentration. A green atmospherically resistant vegetation index (GARI), tailored on the concept of AR VI (Kaufman and TanrC 1992), is developed and is expected to be as resistant to atmospheric effects as ARVI but more sensitive to a wide range of Chl-a concentrations. While ND VI and AR VI are sensitive to vegetation fraction and to rate of absorption of photosynthetic solar radiation, a green vegetation index like GARI should be added to sense the concentration of chlorophyll, to measure the rate of photosynthesis and to monitor plant stress. © Elsevier Science Inc., 1996


The normalized difference vegetation index (NDVI) is widely used to estimate changes in vegetation state. It uses the normalized difference between the nearinfrared (NIR) and red channels, ignoring millions of years of experience by our ancestors on this planet to sense vegetation state with a very high precision using a variety of the "green" channel. NDVI was originally used as a measure of green biomass (Tucker, 1979). It got a solid theoretical basis as a measure of the solar photosynthetically active radiation absorbed by the canopy (Sellers, 1985; 1987). Its application is limited, though, by a complexity of interacting factors involved in the formation of the reflectance response (see, for review, Andeieu and Baret, 1993; Baret and Guyot, 1991; Curran et al., 1991; Horler et al., 1983; Huete et al., 1994). Not enough is yet understood about the peculiarities, specific features, and optical properties of a leaf (e.g., Horler et al., 1983; Fukshansky, 1981; Vogelman and Bjorn, 1986). The NDVI involves relating the reflectance in the red range (near 675 nm) and NIR to vegetation variables such as leaf area index, canopy cover, and the concentration of the total chlorophyll. 0034-4257 / 96 / $15.00 PII S0034-4257(96)00072-7

290 Gitelson et al.

Vegetation has a low level of reflectance in this spectral region and the relationship of NDVI vs. chlorophyll (Chl) saturates for very low Chl concentration (higher than 8 / ~ g / c m 2 for intact bean leaves (Buschmann and Nagel, 1993), 3-5 / t m / c m 2 for maple and chestnut leaves (Gitelson and Merzlyak, 1994 a,b), and even less than 2 g g / c m 2 for sugar maple leaves (Vogelmann et al., 1993). Therefore, NDVI is sensitive to low chlorophyll concentrations, to the fraction of vegetation cover and, as a result, to the absorbed photosynthetically active solar radiation (Yoder and Waring, 1994). But it is not sensitive at higher chlorophyll concentrations or to rate of photosynthesis for large vegetation coverage. Thomas and Gaussman (1977) have found a better correlation between the reflectance at 550 nm and the chlorophyll concentration than that using the reflectance at 675 nm. Tanner and Eller (1986) demonstrated a monotonous (nonsaturated) relation between absorption at 550 nm and chlorophyll concentration for European beach leaves. Buschmann and Nagel (1993), for intact bean leaves, and Gitelson and Merzlyak (1994 a,b) for maple and chestnut leaves, found that the reflectance over a wide range near 550 nm is more sensitive to Chl concentration that that in the main absorption bands of photosynthetic pigments, including 675 nm. Moss and Rock (1991) had shown that ratio P734-747/P715-726 was excellent for assessing chlorophyll in red spruce. Vogelmann et al., (1993) demonstrated this ratio to be applicable for total chlorophyll determination in sugar maple. Baret et al. (1992) suggested using three red edge domains (705-715 nm, 732-737 nm, and 772-780 nm) to evaluate the red edge inflection point shift from space observations. Carter (1993; 1994) found that reflectance was most sensitive to plant stress in the 535-640 nm and 685-700 nm wavelength ranges. Gitelson and Merzlyak (1994 a,b; 1996) have observed high sensitivity of reflectance both in the green and red (near 700 nm) regions to chlorophyll concentrations and have found that the relationships between Chl-a and p550, as well as pT00, are hyperbolic with high degree of accuracy. They used these relations and the observed very low sensitivity of NIR reflectance to chlorophyll level to construct the vegetation indexes p750/pss0 and p7~0/p700. Those were found to be directly proportional to Chl concentration. It allows to achieve an error in total Chl estimation at leaf level of less than 1.3/.tg/cm 2. These algorithms are based on the specific spectral features of absorption of the pigments, and, therefore, it was believed by the authors that these algorithms should be applicable to the estimation of the photosynthetic pigments of all higher plants. Independently, Yoder and Waring (1994) also used the green channel (500-600 nm or 565-575 nm in a vegetation index and found a better correlation (r2=0.83) with photosynthetic activity of miniature Douglas-fir trees than with a "red" channel. In order to develop a vegetation index that is suit-

able for application to space observations, it is necessary to combine these new vegetation bands with spectral bands used to decrease the effect of atmospheric scattering on remote sensing of vegetation. Kaufman and Tanr6 (1992) first suggested the use ofa"blue" band to develop an atmospherically resistant vegetation index (ARVI), where the strong atmospheric effect in the blue channel would correct the vegetation index for the atmospheric effect. Huete et al. (1994) and Huete and Liu (1994) combined this concept with the need to make vegetation indexes that are less sensitive to effects of spectral properties of soil (Huete, 1988). The first objective of this study is to investigate in more detail the specific spectral features of both mature and senescing leaves, covering a wide range of pigment concentration in order to find a wide spectral range where reflectance is maximum sensitive to pigment concentration. Specific wavelengths sensitive to pigment variation are ascertained and the algorithms for Chl assessment at leaf level are developed using the reflectances in the MODIS channels near 550 nm, and channels in the near infra-red region. The algorithms are tested by independent data sets for a range of Chl-a from 0.3/~g/cm 2 to 4 5 / t g / c m 2 and an estimation error of Chl-a concentration of less than 3 / t g / c m 2 is achieved. The second objective is to create vegetation indexes minimally sensitive to atmospheric effects, while still sensitive to a wide range of Chl-a concentrations. Leaf reflectance below 500 nm and at 670 nm were found to be highly correlated over a wide range of leaf greenness (in yellow-green to dark green leaves). This effect is used here, as in the case of ARVI, in a self-correction process for the atmospheric effect on the green channel, using the difference in the radiance between the blue and the red channels. MATERIALS A N D M E T H O D S

The experiments were performed in October 1991 and 1993 on horse chestnut leaves (Aesculus hilypocastanum L.), and in October 1992-1993 on Norway maple leaves (Acer platanoides L.). Leaves of both trees were collected in the Botanical Garden of the Moscow State University, as described previously (Gitelson and Merzlyak, 1994a; Merzlyak and Gitelson, i995). In addition to the senescing samples, the mature green leaves of both species collected in July of 1994 were examined (Gitelson and Merzlyak, 1996). The sampling scheme was intended to cover as high a variation of pigment concentrations as possible. Only leaves having homogeneous dark green, green, green-yellow, yellow-green, and yellow color without anthocyanin pigmentation were selected. Hemispherical reflectance spectra were recorded for the upper surface of the leaves with a Hitachi 15020 spectrophotometer, equipped with an integrating

Use of a Green Channel in Remote Sensing 291

sphere attachment at the rate of 100 nm/min. The spectra were determined for the sections of the leaves between main veins (maple) or with a removed main vein (chestnut). The reflectance spectra were measured with spectral resolution of 2 nm against barium sulfate as a reference standard with a light trap to eliminate the specular reflected component of the radiance, and black velvet was used as a background in order to absorb the light passing through the leaf (Gitelson and Merzlyak, 1994a). Reflectance was expressed as a ratio of the radiance of the leaf to that of the standard. Chl-a, -b, and total carotenoid (Car) concentrations in the leaves were determined in acetone extracts and calculated using equations and specific extinction coefficients as reported by Lichtenthaler (1987). The radiances in the channels of EOS-MODIS were simulated by integrating the reflectance spectra obtained by a spectrophotometer, over the ranges corresponding to channels of MODIS--blue 460--480 nm; green 530-570 nm; red 650-690 nm. To calculate the vegetation indexes, we used the reflectance at 750 nm to simulate the reflectance MODIS channel at 860 nm. Yoder and Waring (1994) showed that the reflectance in the NIR is not sensitive to the chlorophyll concentration, but is sensitive to the leaf area index (LAI). As a result we do not expect that the use of 750 nm rather than 860 nm will make a difference in the dependence of the vegetation indices calculated for the reflectance spectra of the single leaves on the chlorophyll concentration. The atmospheric effects were simulated for four 6S models: no aerosol, continental aerosol with a visibility of 25 km (average) and 5 km (very hazy), and maritime with a visibility 25 km. The original, normalized difference vegetation index (NDVI) is defined by

p,* = nL, I Fauo,


where F0 is the extraterrestrial solar flux, go is cosine of the solar zenith angle, and i stands for the red or NIR channels. These reflectances were then corrected for the molecular effects (scattering and absorption): p , ' = (P,* -- Pm) / Tm


where p,, is the reflectance of sunlight by the atmospheric molecular scattering and Tm is the transmission of light through an aerosol free atmosphere. The selfcorrection for the aerosol effect is introduced by replacing the red reflectance Ptred with P% = P%a - 7(P'bh,~ -- P'.~d).


ARVI = (p'N,a - P%) / (P'Nm + p'rb),


forming ARVI,

where ~ is an empirical parameter that was shown to be optimized for y = 1. We shall use this concept to introduce the atmospheric resistant green vegetation index (GARI).



Pigment Content in the Leaves The dominant pigment was Chl-a that ranged from 0.3 p g / c m 2 to 44.8 g g / c m 2 in maple leaves and from 0.5 g g / c m 2 to 42.4 g g / c m 2 in chestnut leaves. The green leaves collected both in summer and autumn (Chla + b > 20 gg / cm 2) contained approximately equal proportions of pigments (Chl-a, -b, and carotenoids). Although the leaves of both species lost Chl-a and -b during the progression of autumn senescence, relatively high concentrations of carotenoids were present (see also Gitelson and Merzlyak, 1994a; Merzlyak and Gitelson, 1995). Only trace amounts of chlorophylls were detected in completely yellow leaves.

where LNIRand Lred are the radiances of reflected sunlight from the surface as observed from space in the NIR and red channels, respectively. Scattering in the atmosphere by aerosol particles and molecules increases substantially the reflectance in the red channel, thus decreasing the vegetation index (Holben, 1986). In the NIR the atmospheric effect is significantly smaller due to the partial cancellation of an increase in the surface reflectance due to aerosol scattering and a decrease in the reflectance due to aerosol absorption, so that for reflectance at 0.2-0.4 range the net effect is very small (Fraser and Kaufman, 1985). As a result the atmospherically resistant vegetation indices aim to correct mainly the effect of the atmosphere on the red channel. Since ARVI is the basis for the development of the new index, we shall review its mathematical basis here. To define ARVI, Kaufman and Tanrd (1992) replaced LNlR and Lrea with radiances expressed in reflectance units, pi*:

Refectance Spectral Changes in the Leaves The representative reflectance spectra of the leaves which changed color from dark green to completely yellow contained decreasing amounts of Chl-a (Fig. 1). The reflectance spectral features were found to be similar to both species studied. Although the spectra obtained with A. platanoides leaves will be considered later, the results were found to be very close to those obtained in the experiments with A. hippocastanum leaves. Maximum reflectance (about 40-45%) was found at 750 nm and was essentially independent of pigment concentration and the stage of leaf senescence. The lowest reflectance was observed in the blue range of the spectrum from 400 nm to 500 nm. Three carotenoid absorption bands were clearly seen when the background of the Chl was extremely low (Chl-a = 0.3 gg/ cm2). Even a very small increase in Chl-a and -b concen-

NDVI = (LN,. - L.~a)/(LN,. + L,-ed)


Gitelson et al.



Minimum Chl.a = 0.3 1.2




1 0.3 0.8 0.2









oi 400


0.2 450








Wavelength, nm


trations from 0.3/tg / em 2 to 1/tg / cm 2 induced a significant decrease in reflectance and shifted the "green edge" at the reflectance spectra toward the longer wavelengths. An increase in pigment concentration from 3 / t g / c m 2 to m o r e than 4 0 / ~ g / c m 2 did not lead to any variation in the reflectance in this spectral range. An increase in reflectance occurred near 500 nm for all leaves studied. For a completely yellow leaf (Chl-a = 0.3 /,tg/cm2), a wide plateau up to 750 nm followed this increase. In the leaves with Chl-a > 3 / t g / cm 2, the prominent maximum of reflectance occurred in the green range of the spectrum. In yellow-green leaves, this peak was wide and reached 3 0 - 4 0 % , while for green-yellow to green leaves (Chl-a > 3/,tg / cm2), it b e c a m e narrow and was of decreased magnitude. A decrease in reflectance followed by "green" peak. In yellow-green to green leaves (Chl-a > 3.6 ~g / cm2), the reflectance near 670 nm (i.e., the red maximum of Chl-a absorption) was low and remained virtually the same when Chl-a increased. Only when the Chl-a concentration fell to a level of less than 3 / l g / cm 2, a considerable increase in reflectance was observed. A minimum near 670 nm was followed by a sharp increase in reflectance toward longer wavelengths. The slope of the reflectance increase (the "red edge") fluctuated widely, decreasing when the Chl-a concentration increased. To find spectral bands with maximum sensitivity to variation in pigment concentrations, the coefficient of variation of the reflectance (determined as ratio of standard deviation of the reflectance to average reflectance value) was studied (Fig. 2). The coefficient of variation was calculated for different groups of leaves. The first group contained leaves with a Chl-a concentration from 0.3 (the minimum concentration found in our experiments) to 4 4 . 8 / ~ g / c m 2 (the maximum concentration). The second group included leaves with Chl-a from 1.1 ~ g / c m 2 to 4 4 . 8 / ~ g / c m 2. In the third group, the minim u m Chl-a concentration was 3.6 ~ g / c m 2. In other



0 400

Figure 1. The representative reflectance spectra of maple leaves containing different concentrations of pigments. Chlorophyll a concentrations in/lg / cm 2 are indicated.

7.1 11.4








Wavelength, n m

Figure 2. The coefficient of variation of reflectance for groups of maple leaves selected from 25 samples. The groups had different minimal chlorophyll a concentrations indicated in the figure in/lg / cm 2. Maximum Chl-a concentration was 44.8 ~ g / c m 2 for all leaf groups. The first group contained leaves with a Chl-a concentration from 0.3/,tg/cm 2 (the minimum concentration found in our experiments) to 44.8 /tg/cm 2 (the maximum concentration). The second group included leaves with Chl-a from 1.1 to 44.8/~g/cm 2. In the third group, the minimum Chl-a concentration was 3.6/tg/cm 2. Therefore, the groups had different minimum Chl-a concentrations, while the maximum concentration remained the same.

words, the groups had different minimum Chl-a concentrations, while the maximum concentration remained the same. The first group of leaves can be considered to represent a wide-ranging process of senescence or stress, when the color of the leaves turns from completely yellow to dark green. The following groups, in order, corresponded to different stages of senescence ( o r / a n d stress). This range ended with leaves in the very early stages of stress and senescence, when they were still green, but the suppression of biosynthesis and / or increased degradation of green pigments already begun. The obtained coefficient of variation spectra showed several notable features. The spectra indicated different spectral behavior for groups of leaves containing "yellow to green" leaves (with minimum Chl-a = 0.3/,tg / cm 2) as c o m p a r e d to "yellow-green to green" (with minimum Chl-a > 3.6 ~g / cm 2) leaves. For the group of "yellow to green" leaves, several minima due to variation in absorption of pigments near 425 nm, 450 nm, and 490 nm, 520 nm, and 670 nm were found (Fig. 2). In leaves with 44.8 > C h l - a > 1 g g / c m 2, the spectral behavior of the coefficient was different. In the blue range, the coefficient decreased two- to threefold, and spectral features were not detected. In the green and the red

Use of a Green Channel in Remote Sensing




0.07 I 0.065







650-690 +


460-480 •

0.2 D


530-570 •

,~ 0.°4l













Reflectance at 460-490 n m 0

~ 10

i i 20 30 Chlorophyll,

I 40 lag/cm2

i 50


Figure 3. The reflectance in MODIS spectral bands versus Chl-a concentrations for maple and chestnut leaves. A hyperbolic fit relationship between Chl-a and p530-570was best. The correlation coefficient r 2 for the equation Chla = - 7.2 + 4.01"(R530_570)- 1 was more than 0.95.

Figure 4. The reflectance at 650-690 nm versus that at 460-490 nm.

concentration could be recognized in differences between them at the 500 nm and 670 nm. The index [(R670/Rs00)- 1] will be useful to counteract the effects of background reflectance.

Construction of Vegetation Indexes ranges the coefficient of variation decreased and the m a x i m u m near 670 nm now b e c a m e a small minimum. For yellow-green to green leaves (minimum Chla > 3 /~g/cm2), the broad maximum between 580 nm and 690 nm was transformed to a narrow gap centered at 670 nm. Two spectral bands occurred where variation of reflectance was found to be much higher than at 670 nm: one, quite wide near 600 nm and the other, narrow, near 700 nm. Increase in m i n i m u m Chl concentration in the groups of leaves above 3/~g / cm 2 led to a decrease in the coefficient of variation in the whole visible range of the spectrum, but the above-mentioned spectral features remained. Sensitivity of the reflectance, integrated in spectral bands of MODIS, to Chl-a is demonstrated in Figure 3. In the near-infrared as well as in the blue ranges of the spectrum, reflectance remained essentially the same in the very wide region of Chl-a concentration. In the red band near 670 nm, reflectance decreased sharply when Chl-a increased from 0 . 3 / ~ g / c m 2 to 3 / J g / c m z. After this reflectance did not change and remained virtually the same for leaves from yellow-green to dark green. The reflectance in the green band showed maximal sensitivity to Chl-a; it changed m o r e than fourfold for Chl-a ranging from 0.3/~g / cm 2 to 44 g g / cm z. T h e r e is a hyperbolic relationship b e t w e e n Chl-a and pgr.... For the function Chl-a versus (pgreen)-1 the correlation was found to be very high (r 2 > 0.95), with an error of Chl-a estimation of less than 2.8 g g / cm 2. Another notable feature of the reflectance spectra was a high correlation b e t w e e n the reflectances in the red, p650-690, and blue, p460-490, ranges of the spectrum for yellow-green to green both chestnut and maple leaves (Fig. 4). Therefore, the variation in background of nonphotosynthetic reflectance for the same chlorophyll

The index for chlorophyll estimation should be invariant with respect to pigments other than Chl, and should not be influenced by other factors. Therefore, it would be useful to find spectral bands where only one dominant factor (Chl-a) influences variation in reflectance. The coefficient of variation spectra (Fig. 2) clearly indicates specific spectral bands with maximum and minim u m sensitivity of the reflectance to Chl-a. Maximum sensitivity takes place from 520 nm to 630 nm and near 700 nm. Reflectance near 670 nm was almost pigment-concentration-independent for Chl-a ranging from 3-5/~g / cm 2 to more than 40 g g / cm 2. The lowest variation of reflectance took place in the near-infrared (above 750 nm) and in the blue (shorter than 500 nm) parts of the spectrum. The reflectance in the near infra-red can he taken as a term insensitive to Chl-a concentration. Among the spectral bands of the MODIS sensor there are no bands near 700 nm but a spectral band centered at 550 nm does exist. Therefore, we could consider this band pg. . . . instead of Pred, as the sensitive term in NDVI: "Green" N D V I

= LOnir - - P g r e e n ] / ~Onir "~- P g r e e n ] .

This index was found to be m u c h more sensitive to the Chl concentration in a wide range of Chl variations (Fig. 5) than the original "red" NDVI, and enabled precise estimation of pigment concentration. Figure 6 demonstrates a fairly linear relationship between (Chl) °5 and "green" NDVI. The coefficient of correlation for this relationship was r ~ > 0.96, with an estimation error for Chl-a of less than 2 / ~ g / c m 2. In order to create indexes both resistant to atmospheric effects and sensitive to pigment concentration, our findings (Figs. 5 and 6) and the ARVI approach should be combined. The increase in sensitivity to Chl


Gitelson et al.


,ml~m ~


0.8 ) •




n |





:.'('- i


,~0.6 , ,..,.


+0.5 A




~o.4 ]

~A A

m• m ml •

~0.4 Z




• •

• m

0.2 -

g 0



15 20 Chlorophyll,

NDkl A


"Green" ND VI


25 30 tag/cm2



m Ii :--





r 2 > 0.96, n = 3 6 -



: 3





. 5

. 6




Figure 5. NDVI and "green" NDVI, determined as [P,,i,-P~r,,.,d/ LO,,i,+ Pgr~.d verSUS chlorophyll-a concentration. The only difference between them is the use of R~r,,,,,, instead of Rr,,d.

concentration (in comparison to the NDVI and ARVI) was accomplished by employing the reflectance in the green channel of MODIS, pz...... instead of the red channel, Pred. To keep, in the same time, the selfcorrection property for atmospheric effects of the ARVI vegetation index, the difference in the reflectance between the blue and the red channels should be used. For the yellow-green to dark green vegetation, in the absence of atmospheric effect, the difference (Pbh,e-pred) is virtually equal to zero (Fig. 7). Therefore, this difference can be used to correct the green channel in the same way it was used to correct the red channel in ARVI. The following indexes were examined:

Atmospheric resistant green index (GARI): GARI = {P'.ir- [P2. . . . . / p t n i r + [ p i g ......








Figure 6. "Green" NDVI versus total chlorophyll concentration. An estimation error of (Chl) °~5, was found to be as low as 2 ttg/em 2 for Chl-a, and 3 ttg/cm 2 for total chlorophyll. both indexes was identical. But fbr Chl-a > 3 - 5 ttg / cm 2 (for GARI near 0.3), ARVI saturated and did not change any further with an increase in Chl. At the same time GARI increased varying from 0.30 to 0.8 for yellowgreen to green vegetation. GRARI has a similar relationship, like GARI, with Chl concentration. For yellowgreen to green vegetation, GARI and GRARI have a much higher dynamic range than NDVI and ARVI. For Chl-a > 5 t t g / c m 2 the coefficient of variation of ARVI is, on average, at least three times smaller than that of GARI (Fig. 9). For our data sets, GARI permits assessment of Chl-a, with an estimation error of less than 2 / t g / c m 2, and of total Chl within 3 / t g / cm 2. Sensitivity to A t m o s p h e r i c Effects

In order to test the sensitivity of the new indexes to atmospheric effects, a simulation of the radiances oh-

Atmospheric resistant green-red index: GRARI = [P'.ir -- [r/p'~...... + (1 - r/)p%a - 2(p'hh,~ -- P%d)]l /

[P',dr + [rlP~,.,,,,,, + (1 - r/)p%a


2(p',,h,~ --P',-~d)]J,

where P'i is the apparent reflectance (or radiance expressed in reflectance units) after correcting for the molecular scattering and absorption. 2 is a p a r a m e t e r that controls the atmospheric correction, r/is a mix in GRARI of green and red reflectance in order to get properties that are between ARVI and GARI. To analyze the sensitivity of the new indexes to pigment concentration, we c o m p a r e d their dynamic range for Chl-a variations from 0.3 big / cm 2 to more than 40 a g / cm 2. NDVI has a minimal dynamic range (at about 0.8), while ARVI, GARI, and GRARI have a similar and wider dynamic range (more than 1.3). The difference between ARVI and both GARI and GRARI is in their sensitivity to pigment concentration for the yellow-green to dark green vegetation (Chl-a > 3 - 5 ag/cm2). This can be seen from Figure 8, where ARVI is plotted against GARI. For Chl-a less than 3 H g / c m 2, the increase in

Figure 7. Difference of the reflectances (Pbh,,,-P,-~.d)versus chlorophyll-a concentration. For yellow-green to dark green leaves with Chl-a more than 5/*g/cm 2, the difference between them remained essentially the same and was close to zero. 0.1 I

e~ ~.~ -o.2 7 ,~.


g-o.3 ~ /




15 20 25 30 Chlorophyll-a,/ag/cm2




Use o f a Green Channel in Remote Sensing



Table 1. Parameters Used in the Atmospheric

Simulations of the Radiance at the Top of the Atmosphere as Expressed in Reflectance Units, p*: p = po + Tp / ( 1 - sp)

0.8 ]


0.6 -


0.4 ~

M O D I S Spectral Band


0 -0.2"0.4 "0.6 -0.4

470 n m


0.07 0.816 0.14

550 n m

670 n m

860 n m

0.017 0.923 0.04

0.006 0.982 0.01

0.027 0.846 0.08

0.013 0.915 0.05

0.057 0.654 0.15

0.033 0.744 0.11

0.032 0.884 0.09

0.02 0.943 0.06

No Aerosol

I i r\



I 0.2


T s 0.4



Continental 25 km


0.085 0.714 0.18

po Figure 8. Atmospherically resistant vege-

tation index (ARVI) versus atmospherically resistant green index (GARI). For the Chl-a concentration as low as 3-5 p g / c m 2, ARVI saturates while GARI grows larger with an increase in Chl.

0.037 0.846 0.08

T s

0.049 0.761 0.12 Continental 5 km

0.126 0.501 0.23


T s

0.085 0.533 0.19 Maritime 25 km

0.085 0.777 0.18


s e r v e d at t h e t o p of t h e a t m o s p h e r e was p e r f o r m e d u s i n g t h e 6S c o d e ( V e r m o t e et al., 1995; Tanr~ et al., 1990). F o u r a t m o s p h e r i c m o d e l s w e r e used. T h e r e f l e c t a n c e at t h e t o p of t h e a t m o s p h e r e was c a l c u l a t e d for a given surface r e f l e c t a n c e p b y

P*=Po+ Tpl(1 -sp), w h e r e po is the a t m o s p h e r i c p a t h r a d i a n c e in r e f l e c t a n c e units ( r a d i a n c e at t h e top o f t h e a t m o s p h e r e for a b l a c k surface), T is t h e total t r a n s m i s s i o n (diffuse + d i r e c t ) of s u n l i g h t t w i c e t h r o u g h t h e a t m o s p h e r e , to t h e g r o u n d a n d b a c k to space, p is t h e surface r e f l e c t a n c e , a s s u m e d to b e e q u a l to t h e r e f l e c t a n c e of t h e leaf, a n d s is t h e b a c k s c a t t e r i n g of r e f l e c t e d sunlight b y t h e a t m o s p h e r e b a c k to t h e surface. T h e p a r a m e t e r s P0, T a n d s a r e

T s

0.052 0.804 0.12

po is the atmospheric path radiance in reflectance units (radiance at the top of the atmosphere for a black surface), T is the total transmission (diffuse +direct) of sunlight twice through the atmosphere, to the ground and back to space, p is the surface reflectance, and s is the backscattering of reflected sunlight by the atmosphere back to the surface.

given in T a b l e 1 for each m o d e l a n d s p e c t r a l b a n d . T h e s e a t m o s p h e r i c m o d e l s w e r e a p p l i e d to t h e m a p l e and chestnut spectral reflectances and the vegetation i n d e x e s N D V I a n d G A R I w e r e calculated. T h e results a r e p l o t t e d in F i g u r e 10. T h e v e g e t a t i o n i n d e x e s N D V I a n d G A R I a r e p l o t t e d as a function of t h e c h l o r o p h y l l

Figure 10. Plot of the vegetation indexes Figure 9. The coefficient of variation for

various indexes (determined as a ratio of the standard deviation of the index to its average value) for maple and chestnut leaves with Chl-a concentration ranged from 5 to 44.8/~g / cm 2. The sensitivity of GARI and GRARI to Chl-a was found to be at least threefold higher than for ARVI, and flvefold higher than for NDVI.

NDVI and GARI as a function of the chlorophyll for the four ahnospheres given in Table 1. Open syanbols are for chestnut and closed symbols for maple. Dashed lines are for maple only for the four atmospheres. Note the higher sensitivity of GARI to chlorophyll concentration and the smaller sensitivity to atmospheric effects. 1.00





i. ~. 0.15

0.40 Q




@ r~ O.OS

0.20 0.00

;;=" -o.2o -0.40 0.0










Chlorophyll, gg/cm2




Gitelson et al.

Table 2. Results for Chestnut Leaves: the Errors in the Vegetation Indexes Due to Atmospheric Effects for the Three Aerosol Conditions and Several Ranges of Chlorophyll Concentration~ Continental 5 km and 25 km and Maritime 25 km

Chlorophyll 0-10/.tg/cm 2 10-25/~g/cm 2 25-40/lg/cm 2 40-57 j~g/cm 2 Optimized value of y Average error Range E r r o r . range Correlation VI - C h l Correlation VI - Chl °'5





0.004 0.063 0.071 0.075 -0.053 0.82 0.065 0.71 0.84

0.021 0.017 0.015 0.017 1.9 0.018 1.3 0.014 0.76 0.88

0.026 0.021 O.011 0.019 1.6 0.019 1.1 0.017 0.90 0.97

0.025 0.021 0.012 0.018 1.7 0.019 1.16 0.016 0.87 0.96

For each range of concentration the absolute error io the vegetation index is given: ONDVI, OARVI, diGARI, and OAttGRVI, as well as range of the index for the whole scale of Chl and the ratio of the error to range. The optimization of the indices was done for y values larger than those of Kaufinan and Tanr6 (1992).

concentration for the four atmospheres given in Table 1. GARI shows a much higher sensitivity to chlorophyll concentration than NDVI and a smaller sensitivity to atmospheric effects. While the NDVI saturates already fi)r Chl-a < 5 - 7 / , t g / c m 2, GARI does not saturate even at Chl > 40/2g / cm 2. The sensitivity to the atmospheric effect is also much smaller. It is given by the width of the plots in Figure 10, which is 3-4 times wider for the NDVI than for GARI. Table 2 summarizes the sensitivity of NDVI, ARVI, GARI, and GRARI to Chl concentrations and to the atmospheric effects fi)r several ranges of the Chl concentrations. Results are given for chestnut leaves. Except for very small Chl concentrations, where the NDVI is most sensitive, due to the sensitivity of the red channel, the errors in the other vegetation indices due to the atmospheric effects are 3 times smaller, and the range is 1.5 times larger. This gives an advantage of factor 4 over NDVI, similar to the results of Kaufman and Tanr6 (1992). In the present ease, though, the optimization of the indices was achieved for y values larger than those of Kaufman and Tanr6 (1992), y - 1.7. The reason for it is the higher reflectance of the leaves from that of the full canopy (see Fig. 1). Higher reflectance in the red and blue decreases the difference in the atmospheric effect on them, thus requiring a higher value of y to compensate for it. Similar results were obtained for maple leaves. The value of y or 2 in the GARI expression needs to be defined empirically using simulations with spectral data taken of verify of full canopy. DISCUSSION

The mechanisms responsible for the revealed spectral signatures have to be better understood in order to

ascertain whether the parameters of the above indexes will be stable over a wide range of pigment concentrations in the leaves and can be applied to a number of plant species. The leaves studied cover a very wide range of pigment concentrations (leaf color changed from completely yellow to dark green). It could be considered to be a model of different physiological states of a plant, which reveals the following important relationships: 1. Relationship between reflectance near 670 nm (as well as from 400 nm to 500 nm) and Chl-a saturated for a Chl-a concentration near 3 / t g / cm '2, whereas in a wide spectral range from 520 nm to 630 nm and near 700 nm this relation was monotonous and the reflectance remained sensitive to pigment concentrations up to Chla > 40/*g / cm 2 (Figs. 1-3). 2. For green-yellow to green leaves (Chl-a > 3-5 /2g / cm2), a strong correlation (r2 > 0.85) between the red (650-690 nm) and blue (460-480 nm) reflectances was found (Figs. 4 and 7). The spectral range from 530 nm to 570 nm is located between two wide bands of strong pigment absorption. The "green" channel is above the "green edge" in the reflectance spectrum, around 520 nm, on the long wavelength side of the blue Chl-a absorption band. The spectral behavior of this edge was found to be very similar to that of the red edge (Horler et al., 1983). The difference between them is that the green edge is primarily determined by Chl-a, Chl-b and Car absorption, while the red edge is governed by Chl-a and Chl-b. At longer wavelengths this spectral band is located before the large range of absorption by chlorophylls. Thus, in the range 530-570 nm, two strong absorption processes reach their minimum, producing the monotonous relationship of Chl-a versus green reflectance with a high sensitivity to Chl-a concentration. In the range 530-570 nm, as well as at longer wavelengths, Chl-a and -b play a major, even a dominant, role ill light absorption. The contribution of carotenoids is probably much less, as indicated by the reflectance spectra of yellow leaves (Fig. 1). In the presence of trace amounts of both chlorophylls (< 0.3 pg / cm 2) and considerable quantities of carotenoids (> 3/2g / era2), no evidence fi)r the contribution of carotenoids to reflectance in the range 530-570 nm exists (upper curve in Fig. 1). However, an increase in Chl-a up to 3 / l g / c m 2 on a background of approximately the same amounts of earotenoids led to a significant decrease in the "green" reflectance. Therefore, the earotenoids did not contribute significantly to p~...... and the reflectance in the range 530-570 nm can be used for estimating of total Chl and Chl-a. Considering spectral resolution of MODIS green channel, we can conclude it is quite optimal for this application. Higher spectral resolution does not lead to decrease in an estimation error. It should be

Use o f a Green Channel in Remote Sensing

noted also that the AVHRR red channel is more sensitive than the narrower MODIS red channel to variation in Chl concentration in dense vegetation. The data in Figure 2 show that the MODIS red channel will stop to be sensitive to Chl for smaller Chl concentrations than the wider AVHRR channel. The parameters of the relationships between reflectance and pigment concentration depend on many factors; the primary ones are species, pigment composition, and developmental stage. High sensitivity P550 to Chl-a concentration was also demonstrated by Tanner and Eller (1986) for European beach leaves, and by Buschmann and Nagel (1993) for intact bean leaves. This is consistent with the observations of Horler et al., (1983) where, again, a strong correlation between P~40 and the position of the red edge (that is very sensitive to Chl concentration) was found. Considering the sources of "noise" in the algorithms, it is indeed remarkable that the error of Chl-a estimation in the range 0.34 4 . 8 / t g / c m 2 for maple and chestnut leaves was as low as 2/,tg / cm 2. Measurements of additional reflectance spectra of six species (Gitelson et al., 1996) showed that the relationship of GARI to chlorophyll concentration remained practically the same. Therefore, the Chl content at leaf level can be estimated with an accuracy that is not smaller than that for the two species mentioned earlier in the article. Lichtenthaller et al. (1996) showed the application of green band for Chl assessment in tobacco (green and mutant) leaves. An estimation error of total Chl content of less than 2 / t g / cm 2 was achieved. Thus, it shows that the estimate of Chl concentration at the leaf level using GARI can be expected to be not very sensitive to the vegetation species. A high correlation between reflectances /~olue and Prod was found for both plant species with Chl-a > 3/,tg / cm 2 (Fig. 4). This means that absorbance by Chl-a, -b, and Car near 500 nm and by both chlorophylls near 670 nm was almost similar over a wide range of pigment variation (Gitelson and Merzlyak, 1996). The closest correlation between Pblue and Pr~a took place when a certain proportion of green pigments and carotenoids existed. Apparently this phenomenon is unique for yellow-green to green vegetation, where a decrease in green pigment during senescence or disease is followed by a proportional decrease in carotenoid concentration. These spectral features were useful in construction of indexes resistant to atmospheric effects. The similarity in the value of the refectance in the blue and red bands (for yellow-green to green vegetation) over a wide range of pigment concentrations (Fig. 7) allows correction of the green reflectance for atmospheric effects using the difference of the reflectances (phlue- Pred). Even though the optimization of the atmospherically resistant indexes occurred for 7 values much higher than that of Kaufman and Tanr6 (1992), probably due to the much higher reflectance of leaves than that of the whole canopy, the opti-


mized vegetation indices GARI and GRARI were 4 times less sensitive to the atmospheric effect than the NDVI and at the same time they were sensitive to Chl concentration of more than 4 0 / ~ g / c m 2 instead of less than 10 / t g / c m 2 for the NDVI, or ARVI. While the current study demonstrates that developed indexes are accurate in quantifying leaf-level Chl content, we do not know how well these algorithms will hold at the canopy level. The study will also have to be extended to whole canopy spectra, before final conclusions for space application can be made. Our preliminary results of reflectance measurements of different kinds of vegetation (several species of potatoes, tobacco, and so on) showed that "green" NDVI was also very accurate for assessing Chl content at canopy level. Nevertheless, we recognize that differences in species, illumination, canopy architecture, and other factors may potentially decrease correlation between developed indexes and Chl content. Although the relationships between canopy chlorophyll content and developed indexes need to be examined further, this was beyond the scope of the current study.


Andrieu, B., and Baret, F. (1993), Indirect methods of estimating crop structure from optical measurements, In Crop Structure and Light Microclimate. Characterization and Applications (C. Varlet-Grancher, R. Bonhomme, H. Sinoquet,

Eds.), INRA Edition, Paris, pp. 285-322. Baret, F., and Guyot, G. (1991), Potential and limits of vegetation indexes for LAI and APAR assessment. Remote Sens. Environ. 35:161-173. Baret, F., Jacquemoud, S., and Guyot, G. (1992), Modeled analysis of the biophysical nature of spectral shift and comparison with information content of broad bands. Remote Sens. Environ. 41:133-142. Buschmann, C., and Nagel, E. (1993), In vivo spectroscopy and internal optics of leaves as basis for remote sensing of vegetation. Int. J. Remote Sens. 14(4):711-722. Carter, G. A. (1993), Responses of leaf spectral reflectance to plant stress. Am. J. Bot. 80:239-243. Carter, G. A. (1994), Ratios of leaf reflectances in narrow wavebands as indicators of plant stress. Int. J. Remote Sens. 15:697-703. Curran, P. J., Dungan, J. L., Macler, B. A., and Plummer, S. E. (1991), The effect of a red leaf pigment on the relationship between red edge and chlorophyll concentration. Remote Sens. Environ. 35:69-76. Fraser, R. S., and Kaufman, Y. J. (1985), The relative importance of scattering and absorption in remote sensing, IEEE Trans. Geosci. Remote Sens. 23:625-633. Fukshansky, L. (1981), Optical properties of plants, in Plants and the Daylight Spectrum (H. Smith, Ed.), Academic, London, pp. 21-39. Gitelson, A., and Merzlyak, M. N. (1994a), Spectral reflectance changes associated with autumn senescence of Aesculus hipppocastanum L. and Acer platanoides L. leaves. Spectral


Gitelson et al.

features and relation to chlorophyll estimation. J. Plant Physiol. 143:286-292. Gitelson, A., and Merzlyak, M. N. (1994b), Quantitative estimation of chlorophyll-a using reflectance spectra: experiments with autumn chestnut and maple leaves. J. Photochem. Photobiol. (13) 22:247-252. Gitelson, A., and Merzlyak, M. N. (1996), Signature analysis of leaf reflectance spectra: algorithm development for remote sensing of chlorophyll. J. Plant Physiol. 148:494-500. Gitelson, A., Merzlyak, M. N., and Grits, Y. (1996), Novel algorithms for remote sensing of chlorophyll content in higher plant leaves, in International Geoscience and Remote Sensing Symposium, IGARSS'96, Lincoln, NE, 27-31 May, Vol. IV, pp. 1143-1146. Horler, D. N., Dockray, M., and Barber, J. (1983), The red edge of plant leaf reflectance. Int. J. Remote Sens. 4(2): 273-288. Holben, B. N. (1986), Characteristics of maximum value composite images for temporal AVHRR data, INt. J. Remote Sens. 7:1417-1437. Huete, A. R. (1988) A soil-adjusted vegetation index (SAVI), Remote Sens. Environ. 25:295-309. Huete, A. R., and Liu, H. Q. (1994), An error and sensitivity analysis of the atmospheric and soil correcting variants of the NDVI for the MODIS-EOS, IEEE Trans. Geosci. Remote Sens. 32:897-905. 11uete, A., Justice, C., and Liu, H. (1994), Development of vegetation and soil indices for MODIS-EOS, Remote Sens. Environ. 49:224-234. Kaufman, Y. J., and Tanr6, D. (1992), Atmospherically resistance vegetation index (ARVI) for EOS-MODIS, IEEE Trans. Geosci. Remote Sens. 30:261-270. Lichtenthaler, H. K. (1987), Chlorophyll and carotenoids: pigments of photosynethtic biomembranes. Meth. Enzym. 148:331-382. Lichtenthaler, H. K., Gitelson, A. A., and Lang, M. (1996), Non-destructive determination of chlorophyll content of leaves of a green and an aurea mutant of tobacco by reflectance measurements. J. Plant Physiol. 148:501-508. Merzlyak, M. N., and Gitelson, A. (1995), Why and what for

the leaves are yellow in autumn? On the interpretation of optical spectra of senescing leaves (Acer platanoides L.). J. Plant Physiol. 145:315-320. Moss, D. M., and Rock, B. N. (1991), Analysis of red edge spectral characteristics and total chlorophyll values for red spruce (Picea rubens) branch segments from Mt. Moosilauke, NH, U.S.A., in l l t h Annual Internationa Geoscience and Remote Sensing Symposium (IGARSS '91), 3-6 June, Helsinki, Finland, IEEE, New York, Vol. III, pp. 15291532. Sellers, P. J. (1985), Canopy reflectance, photosynthesis and transpiration. Int. J. Remote Sens. 6:1335-1372. Sellers, P. J. (1987), Canopy reflectance, photosynthesis and transpiration. II. The role of biophysics in the linearity of their interdependence. Remote Sens. Environ. 21:143-183. Tanner, V., and Eller, B. M. (1986), Veranderungen der spectralen Eigenschaften der Blatter der Buche (Fagus silvatica L.) von Laubaustrieb bis Laubfall. AUg. Forst. Jagdztg. 157: 108-117. Tanr~, D., Deroo, C., Duhaut, P., et al. (1990), Description of a computer code to simulate the satellite signal in the solar spectrum: 5S code. Int. J. Remote Sens. 11:659-668. Thomas, J. R., and Gaussman, H. W. (1977), Leaf reflectance vs. leaf chlorophyll and cartoenoid concentration for eight crops, Agron. J. 69:799-802. Tucker, J. C. (1979), Red and photographic infrared linear combination for monitoring vegetation. Remote Sens. Environ. 8:127-150. Vermote, E. F., Tanrd, D,, Deuze, J. L., Herman, M., and Morcrette, J. J. (1995), Second simulation of the satellite signal in the solar spectrum: an overview. IEEE Trans. Geosci. Remote Sens., forthcoming. Vogelman, T. C., and Bjorn, L. O. (1986), Plants as light traps. Physiol. Plant. 68:704-708. Vogelman, T. C., Rock, B. N., and Moss, D. M. (1993), Red edge spectral measurements from sugar maple leaves. Int. J. Remote Sens. 14:1563-1575. Yoder, B. J., and Waring, R. H. (1994), The normalized difference vegetation index of small Douglas-fir canopies with varying chlorophyll concentrations. Remote Sens. Environ. 49:81-91.