ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

Contents lists available at ScienceDirect

ISPRS Journal of Photogrammetry and Remote Sensing

journal homepage: www.elsevier.com/locate/isprsjprs

Deriving leaf chlorophyll content of green-leafy vegetables from

hyperspectral reflectance
Lihong Xue, Linzhang Yang ∗
Institute of Soil Science, Chinese Academy of Sciences, Nanjing, China

article info a b s t r a c t

Article history: Different nitrogen (N) treatments of four common green-leafy vegetable varieties with different leaf
Received 7 November 2007 color: lettuce (Lactuca sativa L. var. crispa L.) with yellow green leaves, pakchoi (Brassica chinensis L.) var.
Received in revised form aijiaohuang in Chinese (AJH) with middle green leaves, spinach (Spinacia oleracea L.) with green leaves
24 April 2008
and pakchoi (B. chinensis L.) var. shanghaiqing in Chinese (SHQ) with dark green leaves, were carried out
Accepted 2 June 2008
Available online 11 October 2008
to achieve a wide range of chlorophyll content. The relationship of vegetable leaf hyperspectral response
to its chlorophyll content was examined in this study. Almost all reported successful leaf chlorophyll
Keywords:
indices in the literature were evaluated for their ability to predict the chlorophyll content in vegetable
Hyperspectral reflectance leaves. Some new indices based on the first derivative curve were also developed, and compared with
Vegetation indices the chlorophyll indices published. The results showed that most of the indices showed a strong relation
Leaf chlorophyll content with leaf chlorophyll content. In general, modified indices with the blue or near red edge wavelength
Vegetable performed better than their simple counterpart without modification, ratio indices performed a little
Red edge position better than normalized indices when chlorophyll expressed on area basis and reversed when chlorophyll
expressed on fresh weight basis. A normalized derivative difference ratio (BND: (D722 − D700 )/(D722 +
D700 ) calibrated by Maire et al. [Maire, G., Francois, C., Dufrene, E., 2004. Towards universal broad leaf
chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements.
Remote Sensing of Environment 89 (1), 1–28]) gave the best results among all published indices in this
study (RMSE = 22.1 mg m−2 ), then the mSR-like indices P RMSE between 22.6 and 23.0 mg m .
P with the
−2

The new indices EBAR (ratio of the area of red and blue, dRE/ dB), EBFN (normalized difference of the
amplitude
P of redPand blue, P(dRE −P dB)/(dRE + dB)) and EBAN (normalized difference of the area of red and
blue, ( dRE − dB)/( dRE + dB)) calculated with the derivatives also showed a good performance
with the RMSE of 23.3, 24.15 and 24.33 mg m−2 , respectively. The study suggests that spectral reflectance
measurements hold promise for the assessment of chlorophyll content at the leaf level for green-leafy
vegetables. Further investigation is needed to evaluate the effectiveness of such techniques on other
vegetable varieties or at the canopy level.
© 2008 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by
Elsevier B.V. All rights reserved.

1. Introduction playing a major role in quality evaluation and in the decisions of

the consumer. In recent years, chlorophylls, the most abundant
It is widely known that leaf chlorophyll content is an important pigments in green plants are gaining increasing importance in the
parameter for testing plant status. For example, chlorophyll human diet, not only as food colorants, but also as healthy food
content can be used as an index of the photosynthetic potential ingredients.
as well as of plant productivity (Carter, 1998; Filella et al., 1995). Traditionally, wet chemical methods required chlorophyll
In addition, chlorophyll gives an indirect estimation of the nutrient extraction in a solvent, followed by the spectrophotometric
status because much of leaf nitrogen is incorporated in chlorophyll determination of absorbance by the chlorophyll solution, and
(Filella et al., 1995). Furthermore, leaf chlorophyll content is closely conversion from absorbance to content using standard published
related to various types of plant stresses and senescence (Carter equations (e.g. those of Arnon (1949) and modified thereof).
et al., 1996; Gitelson and Merzlyak, 1994; Penuelas et al., 1995). Although long considered as the standard method for chlorophyll
As for vegetable, color is a fundamental property of food products, determination, extraction requires destructive sampling and
thus does not permit measurement of changes in chlorophyll
over time for a single leaf. In addition, the techniques are
∗ Corresponding author. time-consuming and labor intensive. In contrast, measurement
E-mail addresses: [email protected] (L. Xue), [email protected] (L. Yang). of spectral reflectance is nondestructive, rapid and can be
0924-2716/$ – see front matter © 2008 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.isprsjprs.2008.06.002
98 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

applied across spatial scales. Due to intensive development of Chinese (SHQ) with dark green leaves were selected as materials.
hyperspectral equipment for remote sensing, the use of leaf Five N treatments with 0 (N0), 0.033 (N1), 0.067 (N2), 0.100
reflectance spectra for testing vegetation status is attractive now. (N3), 0.133 (N4) g N per kg soil were designed to create different
These points stimulate active searches for effective methods for chlorophyll content. The pot experiments with a completely
remote chlorophyll estimation. random design and 4 replicates were performed in the greenhouse
Reflectance at visible region of the electromagnetic spectrum is in Nanjing Institute of Soil Science, Chinese Academy of Sciences,
mainly influenced by chlorophyll pigments. Therefore, reflectance from September 2006 to May 2007. Soil was collected from the
spectra in this region can be used to estimate chlorophyll content. suburb garden of Nanjing city. Each soil treatment was distributed
Reflectance at about 675 nm, where the chlorophyll a absorption by weight (2500 g dry soil per pot) into a 20 cm diameter
is maximal, was related to the content of the total chlorophyll plastic pot. Plastic Petri dishes were placed underneath each pot.
(Chappell et al., 1992). As low reflectance at 675 nm reduces Seedlings were thinned to four plants per pot on a week after
its sensitivity to chlorophyll a changes (saturation is reached germination. Plants were watered daily to field capacity. Plants
at medium content), reflectance at 550 nm has been proposed were grown for a total 52 days for spinach, 80 days for pakchoi var.
by others (Thomas and Gausman, 1977). Jacquemoud and Baret AJH, 36 days for lettuce and 42 days for pakchoi var. SHQ.
(1990) have indicated that there should be higher sensitivity at
675 nm for low content of chlorophyll a and at 550 nm for 2.2. Leaf spectra measurements
high content by modeling the effect of chlorophyll a content in
the reflectance spectra. However, the use of reflectance at one Shortly after harvest, fully expanded leaves of 2 plants in each
determined wavelength can be inaccurate in some cases because plot were selected for spectra determination. Reflectance spectra
they are very sensitive to biomass, variable irradiance, background were obtained using an ASD Fieldspec FR spectroradiometer
effects, and geometric arrangement of the scene, sensor, and (Analytical Spectral Devices, Boulder, CO) with a wavelength range
surface (Munden et al., 1994). of 350–2500 nm, a sampling interval of 1.4 nm between 350 and
In order to overcome the sensitivity of single waveband to 1000 nm and 2 nm between 1000 and 2500 nm, and a spectral
other influence factors, several spectral indices based on two resolution of 3 nm between 350 and 1000 nm and 10 nm between
1000 and 2500 nm. A leaf clip was attached to the fore-optic,
or three discrete bands including the sensitive bands or their
achieving several purposes: it shields the sample from ambient
neighborhood to chlorophyll were developed and shown to be well
light, it limits the FOV to a constant 2.3 mm, and it holds the for-
correlated with chlorophyll content (Barnes et al., 1992; Gitelson
optic at a constant 90◦ . Leaf illumination was provided through
and Merzlyak, 1994; Penuelas et al., 1994, 1995; Sims and Gamon,
one side of the bifurcated fiber from an internal 7.0 W halogen
2002). In recent years, emphasis has been on the development
lamp in the spectrometer. Sample reflectance was obtained by
of more sophisticated ways of interpreting spectral signatures,
inserting a leaf into the leaf clip and comparing leaf reflectance to
some of which were based on the shape and relative position
the reflectance of the white spectralon standard, also inserted into
of the spectral reflectance curve. Derivative spectral indices and
the leaf clip. Individual spectral measurements were an average
the characteristic red edge reflectance pattern of vegetation were of 10 scans and each leaf was sampled 5 times and subsequently
shown to be very sensitive to chlorophyll content, while they tend averaged for further analysis; four to seven leaves were sampled
to minimize spectral noise caused by background and atmospheric per plant.
effects (Filella and Penuelas, 1994). To reduce noise, a 5-point Savitzky–Golay filter was applied
So far, most of leaf chlorophyll reflectance indices have been to the spectra. Then, the first derivatives were computed using
developed and tested for herbaceous crops, grasses, or common polynomials of second order and a 9-point moving window. This
deciduous tree and coniferous tree species (Horler et al., 1983; was done with Unscrambler 9.5 software (CAMO PROCESS AS. Oslo,
Carter, 1994; Datt, 1998, 1999; Sims and Gamon, 2002; Kochubey Norway, 2005). As the absorption of chlorophyll mainly occurs at
and Kazantsev, 2007). A limited study was carried out on fruity visible region, only 400–1000 nm region was used for analysis.
vegetables such as bean and capsicum (Buschman and Nagel,
1993; Filella and Penuelas, 1994). Fewer reports were found on 2.3. Leaf chlorophyll content
green-leafy vegetables. Some of these proposed indices failed
when tested using a second, independent dataset. In some cases, Once spectral measurements were completed for each leaf,
the results of such tests are not presented in a way that allows the leaf samples were cut into 1 to 2-cm square pieces, frozen,
meaningful comparison of the indices across different studies. and later analyzed for chlorophyll a + b in the laboratory. Each
Finally, some indices have rarely been tested using data from leaf sample was extracted for 24 h in the dark at 22 ◦ C with
species other than those used in the formulation of the index. 80% acetone. Wavelength absorbance was measured at 662 nm
Extrapolation of these techniques to green-leafy vegetables is more for chlorophyll a and 645 nm for chlorophyll b in a Shimadzu
dubious and should be tested. Thus, our objective in this study UV 160A spectrophotometer (Arnon, 1949). All the analyses were
was to investigate the spectra behavior of the relationship between conducted in three replications, each in two parallel samples. Total
reflectance and chlorophyll content both on area basis (mg m−2 ) leaf chlorophyll content was determined for each sample and
and on fresh weight basis (mg g−1 ) in green leafy vegetables and to expressed both on an area basis (mg m−2 ) and on a fresh weight
check whether it is possible to use the spectral indices proposed in basis (mg g−1 ).
the literatures for quantitative chlorophyll estimation in vegetable
leaves as well as in other plant varieties. 2.4. Spectral indices

2. Materials and methods 2.4.1. Spectral indices based on several discrete bands
Ratio and normalized difference indices of two discrete bands
2.1. Experiment design were widely used to evaluate the nitrogen /chlorophyll status,
and most common band was the NIR (750–870 nm), Green
Four common green-leafy vegetable varieties with different leaf (550 nm or so) and Red (660–680 nm) band. Richardson et al.
color: lettuce (Lactuca sativa L. var. crispa L.) with yellow green (2002) found that some indices based on reflectance in the red
leaves, pakchoi (Brassica chinensis L.) var. aijiaohuang in Chinese edge region (Gitelson and Merzlyak, 1994) were much better
(AJH) with middle green leaves, spinach (Spinacia oleracea) with indicators of chlorophyll content than some of more commonly
green leaves and pakchoi (B. chinensis L.) var. shanghaiqing in used indices. Recently defined structure independent pigment
 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106 99

index (SIPI) that uses a near-infrared waveband (800 nm) as a leaves of various plants. In addition, modified simple ratio (SR)
subtracted and ratioed reference in order to remove any additive and normalized difference (ND) index calculated with derivatives
and multiplicative factors provided a very good semi-empirical were also found to be good estimators for chlorophyll on a larger
estimation of the carotenoids/chlorophyll-a ratio for data from experimental database (Maire et al., 2004).
several species (Penuelas et al., 1995). However, Sims and Gamon All the indices validated in this paper are listed in Table 1. The
(2002) found that previously published spectral indices provided indices may be classified in five categories:
relatively poor correlations with leaf chlorophyll content when 1. Indices using a single reflectance or a difference between two
applied across a wide range of species and plant functional wavelengths (R, DVI)
types, and developed two new spectral indices (mSR index and 2. Simple ratio or modified simple ratio of reflectance or
mND index) to reduce the effect of difference in leaf surface derivatives (SR = Ra /Rb or Da /Db , mSR = (Ra − Rc )/(Rb − Rc )
reflectance by introducing R445 (the reflectance at 445 nm, R or (Da − Dc )/(Db − Dc ))
denotes reflectance, the same as follows) as a reference. Similarly, 3. Normalized difference or modified normalized difference of
a new vegetation index, the inner reflectance (RI , obtained by reflectance or derivatives (ND = (Ra − Rb )/(Ra + Rb ) or
subtracting the external reflectance identified with the blue (Da − Db )/(Da + Db ), mND = (Ra − Rb )/(Ra + Rb − 2Rc ) or
flat reflectance) at green band normalized at a non-absorbing (Da − Db )/(Da + Db − 2Dc ))
wavelength of 750 nm (RI542 /RI750 ), was developed by Maccioni 4. Indices related with red edge derived with derivatives (REIs)
et al. (2001) to reduce the external reflectance influence caused 5. Other forms, e.g. double difference (DD), Blog(1/737)
by the mismatch of the refractive index at the air–epidermis and
In this paper, R represents reflectance, D represents first
epidermis–inner layer boundary. Gitelson et al. (2003) found that
derivative, a, b and c represent wavelengths.
[(Rλ )−1 − (RNIR )−1 ] × (RNIR ) (λ, 695–705 nm) can adjust the
difference of leaf structure, and further increase the accuracy of the
2.5. Statistical analyses
chlorophyll estimation. This new index virtually was identical with
the R750 /R550 and R750 /R700 previously put forward by Gitelson and
In total, 67 leaf spectra for spinach, 43 leaf spectra for lettuce
Merzlyak (1994) and Lichtenthaler and Lang (1996). In addition,
and 81 leaf spectra for Pakchoi var. AJH and 133 leaf samples
first-difference transformation of the apparent absorbance, the
for Pakchoi var. SHQ were acquired. The mean and coefficient of
logarithm of reciprocal reflectance, was found to be the best
variation (CV) were used to characterize the leaf chlorophyll status
predictor for nitrogen and chlorophyll in fresh big leaf maple leaves
under different N treatments for each vegetable variety. ANOVA
(Yoder and Pettigrew-Crosby, 1995). Maire et al. (2004) found that
was used to evaluate the significance of differences between mean
the new index double difference (DD) which was calculated with
values of leaf chlorophyll status under different treatments for
four discrete wavebands gives a good estimation for chlorophyll on
each vegetable. Correlation analysis was conducted separately
larger experimental databases.
for each vegetable variety and collectively for the entire dataset
to evaluate the performance of the indices. Regression analysis
2.4.2. Hyperspectral indices based on first derivative reflectance was used to assess relationships between leaf chlorophyll content
The position of red edge (REP) defined as maximum slope and spectral variables. All statistical analyses were conducted in
in the vegetation reflectance spectra between the red and near SPSS 12.0. Different regression models were evaluated and the
infrared regions, is less sensitive to disturbing factors and has ones giving the highest R2 were selected. RMSE was calculated to
been found to be one of the best estimators of chlorophyll content compare the prediction ability of spectral indices.
(Filella and Penuelas, 1994; Munden et al., 1994). Red edge position
moves to longer wavelengths as chlorophyll content increases. 3. Results and discussion
There are different techniques to parameterize this spectral shift.
In the present study, three methods to derive the REP were 3.1. Effect of N treatment on leaf chlorophyll status
applied: maximum first derivative spectrum, inverted Gaussian
fitting technique (Miller et al., 1990), and linear extrapolation As expected, the fertilization treatments induced broad vari-
technique (Cho and Skidmore, 2006). Other parameters of the red ation in leaf chlorophyll content (Table 2). Range of chloro-
edge derived from the first derivative of reflectance curve, the phyll content in leaves used for spectrometry studies were
amplitude(dRE, amplitude of the main peak in the P first derivatives 72.5–245.3 mg m−2 (or 1.25–2.23 mg g−1 ) for spinach, 23.1–96.9
of the reflectance spectra), and the area ( dR680-780 nm, mg m−2 (0.42–1.14 mg g−1 ) for lettuce, 69.0–156.3 mg m−2
calculated as the sum of the amplitudes between 680 and 780 nm, (1.16–2.04 mg g−1 ) for pakchoi var. AJH and 91.7–242.1 mg m−2
in the first derivative reflectance spectra) were found to be (1.31–2.70 mg g−1 ) for pakchoi var. SHQ, respectively. Leaf chloro-
highly correlated with plant chlorophyll a content (Filella and phyll content increased with N levels and leaf color. Chlorophyll
Penuelas, 1994). In addition, ratio and normalized difference of the content was lowest for lettuce with yellow green leaves and high-
amplitude of red and green (EGFR, EGFN) have also been used to est for pakchoi var. SHQ with dark green leaves when expressed
estimate chlorophyll content (Penuelas et al., 1994). In this paper, on a fresh weight basis. While, when expressed on an area basis,
ratio and normalized difference of the area and the amplitude the highest chlorophyll level was observed on pakchoi var. SHQ
between red (680–780 nm) and blue (490–530 nm) (EBAR, EBAN, without N supply but on spinach under N supply. This may be re-
EBFR, EBFN) were also calculated. The details of the calculation of lated to the bigger specific fresh leaf weight of pakchoi var. SHQ.
these indices are listed in Table 1. It should be noted that intra-treatments variability was low in all
In addition, changes in the ratio of relative heights of cases (that is, CV < 15%) although the leaf age was not consistent,
maxima in the first derivatives of red edge have been revealed whereas inter-treatments variability was high for all four vegeta-
when chlorophyll content varied. The ratio of the heights of bles with the CV from 20.96% to 36.53%, and the variability between
two derivative maxima in red-edge region could be used for four vegetables was also high with the CV from 27.61% to 34.69%
quantitative chlorophyll estimation, and several derivative indices when chlorophyll content expressed on a fresh-weight basis (Ta-
such as D705 /D722 (D means the first derivative reflectance, ble 2). The CV became higher when chlorophyll content was ex-
numbers is wavelength in nm for fixed determination) (Zarco- pressed on an area basis (Table 2).
Tejada et al., 2002), D725 /D702 (Smith et al., 2004), D715 /D705 Differences in the mean leaf chlorophyll level among the five
(Zarco-Tejada et al., 2004), D754 /D704 (Datt, 1999) were used N treatments were statistically significant at 0.01 levels (Table 2).
for semi-quantitative and quantitative chlorophyll estimation in Significant differences (p < 0.01) of leaf chlorophyll were
100 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

Table 1
Summary of the spectral indices used in this study
Index Category References Index Category References

R550 /R800 SR Aoki et al. (1981) REP-FD: λ max(dR/dλ) REIs Horler et al. (1983)
PSSRa: R800 /R680 SR SRPI: R430 /R680 SR
PSSRb: R800 /R650 SR EGFR: dRE/dG REIs
Blackburn (1998a,b) Penuelas et al. (1994)
PSNDa: (R800 − R680 )/(R800 + R680 ) ND NPCI:(R680 − R430 )/(R680 + R430 ) ND
PSNDb: (R800 − R635 )/(R800 + R635 ) ND EGFN: (dRE − dG)/(dRE + dG) REIs
R800 − R550 DVI Buschman and Nagel (R850 − R710 )/(R850 − R680 ) mSR
Datt (1999)
R800 /R550 SR (1993) D754 /D704 SR
(R780 − R710 )/(R780 − R680 ) mSR Maccioni et al. (2001) NPQI: (R415 − R435 )/(R415 + R435 ) ND Barnes et al. (1992)
R750 /R700 SR Gitelson and Merzlyak R672 /(R550 ∗ R708 ) SR
Chappell et al. (1992)
R750 /R550 SR (1994) R657 /R700 SR
ND705 : (R750 − R705 )/(R750 + R705 ) ND R675 /(R700 ∗ R650 ) SR
(R695−705 )−1 − (R750−800 )−1 DVI D715 /D705 SR
R750−800 /R695−705 − 1 SR Gitelson et al. (2003) R740 /R720 SR Vogelman et al. (1993)
(R750−800 − R430−470 )/(R695−705 − R430−470 ) − 1 mSR (R734 − R747 )/(R715 + 726) Others

mSR705 : (R750 − R445 )/(R705 − R445 ) mSR R672 /R550 SR

Sims and Gamon (2002) Datt (1998)
mND705 : (R750 − R705 )/(R750 + R705 − 2R445 ) mND R860 /(R550 ∗ R708 ) SR
REP-Gaus: REP estimated with an inverted Gaussian REIs Miller et al. (1990) REP-LEM: REP estimated with a REIs Cho and Skidmore (2006)
fir on reflectance linear extrapolation method
dRE: REIs Filella and Penuelas OCAR: R630 /R680 SR
Schlemmer et al. (2005)
YCAR: R600 /R680
P
dRE: REIs (1994) SR
SIPI: (R800 − R445 )/(R800 − R680 ) mSR Penuelas et al. (1995) Blog (1/R737 ) Others Yoder and
Pettigrew-Crosby (1995)
R700 /R670 SR Mc Murtey et al. (1994) RIλ /RINIR : mSR Maccioni et al. (2001)
(R542 − Rblue )/(R750 − Rblue )
D705 /D722 SR Zarco-Tejada et al. (2002) D715 /D705 SR Zarco-Tejada et al. (2004)
D725 /D702 dRE/
P P
SR Kochubey and Kazantsev EBAR: dB REIs New indices
(2007)
( PdRE − P
P
EBAN:
dB)/( dRE + dB)
P
EBFR: dRE/dB
EBFN: (dRE − dB)/(dRE + dB)
(R728 − R434 )/(R720 − R434 ) mSR Maire et al. (2004) R695 /R670 SR
(R728 − R720 )/(R728 + R720 − 2R434 ) mND R695 /R760 SR
BND: (D722 − D700 )/(D722 + D700 ) ND R710 /R760 SR
Carter (1994)
BmSR: (D722 − D502 )/(D700 − D502 ) mSR R695 /R420 SR
BmND: (D722 − D700 )/(D722 + D700 − 2D502 ) mND R550 R
DD: (R749 − R720 ) − (R701 − R672 ) Others R605 /R760 SR
R: reflectance; D: the first derivative; REP: red edge position, wavelength of dRE; dRE, dG and dB: maximum Pamplitude of the first derivative on the red-edge region
dRE,
P
(680–780 Pnm, dRE), green region P(530–570 nm, dG) and blue region (490–530 nm, dB), respectively; dB: the sum of first derivative reflectance between red
region ( dRE) or blue region ( dB); Blog: the first derivative of logarithm; SR: Simple Ratios of reflectance or derivatives; mSR: modified Simple Ratios of reflectance
or derivatives; ND: Normalized Difference of reflectance or derivatives; mND: modified Normalized Difference of reflectance; DD: Double Difference; REIs: indices related
with red edge derived from first derivatives; BND: Normalized Difference derivatives; BmSR: modified Simple Ratio derivatives; BmND: modified Normalized Difference
derivatives.

Table 2
Mean and coefficient of variation (CV, % in parenthesis) of leaf chlorophyll content for different vegetable under different N treatments
Varieties Unit N0 N1 N2 N3 N4 Mean

Lettuce mg m−2 23.1(11.6)E* 40.6(10.4)D 60.7(7.2)C 76.0(5.9)B 96.9(6.8)A 59.5(48.7)

mg g−1 0.42(11.56)E* 0.58(10.41)D 0.81(7.17)C 0.95(5.88)B 1.14(6.85)A 0.78(36.53)
Spinach mg m−2 72.5(8.6)E 126.4(7.9)D 162.0(9.4)C 203.0(6.7)B 245.3(8.4)A 161.8(41.3)
mg g−1 1.25(8.56)E 1.58(7.94)D 1.80(9.41)C 2.03(6.70)B 2.23(8.40)A 1.78(21.55)
Pakchoi var.
AJH mg m−2 69.0(5.7)E 85.7(11.2)D 102.4(7.3)C 118.3(10.1)B 156.3(10.1)A 106.3(31.5)
mg g−1 1.16(5.73)E 1.42(11.20)D 1.58(7.32)C 1.77(10.14)B 2.04(10.08)A 1.60(20.96)
SHQ mg m−2 91.7(8.9)E 120.8(6.1)D 146.4(4.8)C 188.7(5.7)B 242.1(10.0)A 157.9(37.4)
mg g−1 1.31(8.94)E 1.61(6.10)D 1.97(4.76)C 2.20(5.74)B 2.70(10.03)A 1.93(25.09)
Mean mg m−2 64.1(45.4)E 93.4(42.3)D 117.9(38.8)C 146.5(40.8)B 185.1(38.8)A
mg g−1 1.04(34.69)E 1.30(32.41)D 1.54(28.80)C 1.74(27.61)B 2.03(27.91)A
*
Different letters in the same row indicate significant differences at the p < 0.01 level.

also observed among four different vegetables under the same reflectance spectra and first derivative spectra. For clarity, only the
N treatment, with the only exception of the difference between mean reflectance and first derivative reflectance with three rates
pakchoi var. SHQ and spinach. of N (N0, N2, and N4) are shown in Figs. 1 and 2. The other rates
were intermediate and were not shown. Leaves with the lowest
3.2. Impact of N treatment on leaf reflectance response chlorophyll content had the highest reflectance in the visible range
and lowest reflectance in the near-infrared, similar to the results of
Leaves of spinach, lettuce and pakchoi var. SHQ and AJH other crops (Daughtry et al., 2000). Increase of N level significantly
responded in a very similar fashion to the N treatments, in terms of diminished the peak at green band around 560 nm, and increased
 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106 101

Fig. 1. Reflectance of spinach, lettuce, pakchoi var. aijiaohuang (AJH) and pakchoi var. shanghaiqing (SHQ) under different N treatment.

Fig. 2. First derivative reflectance of spinach (A), lettuce (B), pakchoi var. aijiaohuang (AJH) (C) and pakchoi var. shanghaiqing (SHQ) (D) under different N treatment.

the reflectance plateau at NIR (Fig. 1). The most pronounced change identifies the inflection points of the red-edge region. Under N
in leaf reflectance spectra due to fertilization was in the visible stress, red edge position shifts to shorter wavelength (blue shift).
region at the chlorophyll peak near 550 nm, the reflectance ranged However, no obvious difference was observed for first difference
from 10.2% for leaves with high chlorophyll levels of N4 treatment reflectance after 760 nm (Fig. 2).
to 17.1% for leaves with low chlorophyll levels of N0 treatment Correlograms provide a clearer picture of the relationships
(Fig. 1). between spectra and leaf chlorophyll, and were almost identical
The first derivative reflectance curves can provide information regardless of the unit of chlorophyll (Fig. 3). The patterns
about rate of change along the curve and at what wavelength of correlation were similar for four vegetable varieties across
these changes occur. Such insight is not apparent with simple all wavelengths for both reflectance (R) and first derivative
reflectance curves (Fig. 2). The spikes in Fig. 2 identify the location reflectance (R0 ) with the exception of lettuce for first derivative
of the inflection points from the reflectance curves and indicate reflectance in the 500–560 nm regions. This may be related to the
the magnitude and direction for the rate of change at those crinkly leaf of lettuce. Absolute values of correlations for R were
inflection points. The magnitude of spikes at 525, 575, 610, and higher in the visible regions, achieving a maximum absolute value
640–660 nm appears to show a strong relationship to the N around 0.80 at wavelengths of 700–705 nm and 540–560 nm. Both
treatments. The location of spikes between 680 and 740 nm the magnitude of correlation with chlorophyll and the spectral
102 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

Fig. 3. Correlogram of leaf chlorophyll content (mg m−2 ) to leaf reflectance (A) and first derivative reflectance (B).

Fig. 4. Relation of BND (normalized difference derivative index, (D722 − D700 )/(D722 + D700 )), BmND (modified normalized difference derivatives, (D722 − D700 )/(D722 +
D700 − 2D502 )) and EGFN (normalized difference of amplitude of red and green in the first derivative curve, (dRE − dG)/(dRE + dG)) to leaf chlorophyll content (mg m−2 ).

region of highest correlation are similar to those of other studies R500 /R800 , R605 /R760 , R695 /R760 , PSNDa, PSNDb shown in Fig. 4B
with other sensors (Schlemmer et al., 2005). Correlations with saturates when chlorophyll content is greater than 200 mg m−2 or
reflectance were positive and relatively low through most of the 2 mg g−1 . EGFN, EGFR, R657 /R700 and R672 /R550 shown in Fig. 4C in
infrared region. With the first-difference transformation, more the third group decreased for chlorophyll content increasing from
areas of high correlation spanned the visible and infrared regions, 0 to 100 mg m−2 (or 1 mg g−1 ) and then increase. Fig. 5 shows
band widths of high correlation were narrower, and absolute how well those spectral indices are related to the chlorophyll
values of correlations were generally higher than for reflectance. content in this study. Most of the spectral indices calculated were
And strong relationships exist with the magnitudes of the first highly related with chlorophyll content for all four vegetable
derivative and chlorophyll content. Strong negative correlations varieties with the correlation coefficient (R) being higher than
with chlorophyll content exist at the leading edge of the red-edge P (n = 324), except for NPQI, SRPI, NPCI, R657 /R700 , dRE and
0.60
region (680–690 nm) for all four vegetables and at 510–540 nm dRE (Fig. 5). The best ones are DD, BND developed by Maire
for spinach and pakchoi var. SHQ and AJH while being slightly et al. (2004), the mSR-like index (R850 − R710 )/(R850 − R680 ),
weaker at 480–500 nm for lettuce. Strong positive correlations (R780 − R710 )/(R780 − R680 ), (D722 − D502 )/(D700 − D502 ) given
exist at 615 nm, 630-670 nm, and at the trailing edge of the red- by Datt (1999), Maccioni et al. (2001) and Maire et al. (2004),
edge region (715–745 nm). Similarly strong correlation between (R734 − R747 )/(R715 + 726) given by Vogelman et al. (1993),
chlorophyll content and first derivative spectra for corn was also with the correlation coefficient more than 0.90 for chlorophyll
observed by Schlemmer et al. (2005). This indicates that changes expressed both on area and fresh weight basis (Fig. 5). It should
in the shape of reflectance curves are largely a function of the leaf be noted that red-edge waveband (700–715 nm) appeared in all
chlorophyll content (Penuelas et al., 1994). these best spectral indices, indicating its importance in chlorophyll
estimation. Furthermore, some spectral indices such as PSSR
and PSND found by Blackburn (1998a,b) and SIPI developed by
3.3. Relation between leaf chlorophyll content and spectral indices Penuelas et al. (1995) showed a high correlation with lettuce of
low chlorophyll level, but a poor correlation with other three
Scatter plots provided a visual relationship between these vegetables with moderate to high chlorophyll level (Fig. 5).
spectral indices and chlorophyll content. The results can be divided Maire et al. (2004) also found that PSNDa became invariant with
into three groups and three figures are shown to represent respect to chlorophyll >100 mol/m2 , and became insensitive
different features (Fig. 4): Normalized Difference derivatives BND to chlorophyll >200 mol/m2 , this indicated that indices using
((D722 − D700 )/(D722 + D700 )), modified simple ratio derivatives reflectance in the red range around 680 nm were sensitive to
BmSR ((D722 − D502 )/(D700 − D502 )), R800 /R550 , R860 /(R550 × R708 ), only very low chlorophyll and non-sensitive to moderate to high
R750 /R550 , R750 /R700 , mSR705 , (R734 − R747 )/(R715 + 726), D725 /D702 , chlorophyll, and more suited for low chlorophyll prediction. While
Blog (1/R737 ), R710 /R760 , R740 /R720 , R550 , R675 /(R700 × R650 ), RIλ /RINIR , some indices such as YCAR, OCAR, R700 /R670 , R672 /(R550 × R708 ),
REP-FD, REP-LEM, Double Difference DD ((R749 − R720 ) − (R701 − R695 /R420 , R695 /R670 , D754 /D704 , EBFR and EBFN developed in this
R672 )), D715 /D705 , mSR, mND, (R780 − R710 )/(R780 − R680 ), (R850 − study were well related with spinach and pakchoi, they were
R710 )/(R850 − R680 ), ND705 , mSR705 , mND705 were the first group poorly related with lettuce, suggesting that these indices are
being linearly related or approximatively linearly related with insensitive to low chlorophyll.
chlorophyll content (Fig. 4A), the second group with modified In order to further evaluate the performance of these indices,
normalized difference derivatives BmND, D705 /D722 , D754 /D704 , the determination coefficient (R2 ) of best regression equation and
 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106 103

Fig. 5. Correlation coefficient of spectral indices to leaf chlorophyll content in mg m−2 (A) and mg g−1 (B) with four vegetable varieties.

RMSE of prediction of these indices with correlation coefficient results than their simple counterpart (SR, ND), identical with
higher than 0.80 for all pulled data were computed and compared the results of Maire et al. (2004). The modified indices use a
(Fig. 6). BND was proved to be the best index for chlorophyll ‘‘base wavelength’’ (often blue, 434 nm), originally introduced
estimation with the RMSE of 22.1 mg m−2 of chlorophyll expressed to avoid the effects of differences in leaf surface reflectance
on area basis and 0.223 mg g−1 of chlorophyll expressed on (specular effect as opposed to internal component effect) (Sims
fresh weight basis, then (R850 − R710 )/(R850 − R680 ), (R780 − and Gamon, 2002). This modification significantly improves the
R710 )/(R780 − R680 ), (D722 − D502 )/(D700 − D502 ), mSR705 , mND705 , results at the leaf level. Ratio indices performed a little better
EBAR, D705 /D722 , (R728 − R434 )/(R720 − R434 ), (R728 − R720 )/(R728 + than when chlorophyll was expressed on area basis, similar with
R720 − 2R434 ) and RIλ /RINIR followed, with the RMSE less than the results of Maire et al. (2004). But when chlorophyll was
both 24.0 mg m−2 and 0.240 mg g−1 . Among the top 11 spectral expressed on a fresh weight basis, normalized indices showed
indices, 6 were mSR-like index, 3 were ND or mND-like indices, superior results compared to the ratio indices; this may be owing to
one was a simple ratio index and one was a new index developed the normalized indices avoiding the effects of specific leaf weight
in this study. The modified indices (mSR, mND) give better between different vegetable species. In this paper, reflectance
104 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

Fig. 6. Evaluation of spectral indices with the correlation coefficient higher than 0.80 for the whole dataset with four vegetable varieties.

derivative (e.g. D740 /D720 , RMSE = 25.03 mg m−2 ) was a little were developed to accurately calculate the REP. It is evident
inferior to their direct reflectance-based counterpart (R740 /R720 , that results of REP calculations are highly dependent upon the
RMSE = 22.94 mg m−2 ), contrary to the results of Maire et al. choice of method. The REP data range was widest for linear
(2004). Among the modified indices, the performance of (R850 − extrapolation method (663.0–729.1 nm), then invert-Gaus method
R710 )/(R850 −R680 ) was superior to the mSR705 and mND705 index, (678.0–713.2 nm), and narrowest for the maximum first derivative
contrary to the results of Sims and Gamon (2002), who found that method (690–720 nm). But the mean of REP was highest for
the index developed by Datt (1999) produces unstable results for the maximum first derivative method and lowest for linear
leaves with very low or no chlorophyll content. This may be due extrapolation method. Furthermore, the distribution of REP-LEM
to the leaf chlorophyll content not being low enough compared data was near normal, similar to the distribution of chlorophyll
with that in the study of Sims and Gamon (2002). It should be data, while the REP-Gaus data skewed to left (skewness =
noticed that six spectral indices developed by Maire et al. (2004) −1.6) and had a distinct peak near the 705 nm, reflected the
using PROSPECT model and calibrated with deciduous tree species overrepresentation of low chlorophyll. Thus, REP-LEM was linearly
also performed very well for the green leafy vegetables in this related with chlorophyll, while REP-Gaus was log-linear related
study; especially for the indices of BND, (R728 − R434 )/(R720 − R434 ), with chlorophyll. Maccioni et al. (2001) also pointed out that the
(R728 − R720 )/(R728 + R720 − 2R434 ) and BmSR, the RMSE was 22.1, relation between REP and chlorophyll content is not generally
22.75, 22.79 and 22.96 mg m−2 respectively. linear, it strongly depends on the kind of plant and for some species
Shifts in the REP to longer or shorter wavelengths have it vanishes. Maire et al. (2004) found that REP was not as good as
been widely used as a means to estimate changes in foliar a simple index at the leaf level, similar results were also found in
chlorophyll content and also as an indicator of vegetation stress this study when chlorophyll was expressed on area basis. But when
(Buschman and Nagel, 1993; Datt, 1999). And several methods chlorophyll expressed on fresh weight basis, the performance
 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106 105

Fig. 7. Relation of measured and predicted leaf chlorophyll content from BND (normalized difference derivative index, (D722 − D700 )/(D722 + D700 )), RIλ /RINIR : (R542 −
Rblue )/(R750 − Rblue ), EBAR (ratio of area of red and blue in the first derivative curve, dRE/ dB) and R740 /R720 for different vegetable varieties.
P P

of REP-Gaus was comparative with the best simple index. One performed a little better than normalized indices when chlorophyll
advantage of the red edge parameters is that they are not affected expressed on area basis and reversed when chlorophyll expressed
by additive constants such as leaf surface reflectance. Thus, it on fresh weight basis. A normalized derivative difference ratio
would be expected to be more reliable when the data includes (BND calibrated by Maire et al. (2004)) gave the best results among
a wide range of leaf structures. Due to the peak jump or multi- all published indices in this study (RMSE = 22.1 mg m−2 ), then
peak features of red edge, the performance of red edge position for the RIλ /RINIR and (R780 − R710 )/(R780 − R680 ) developed by Maccioni
chlorophyll estimation became unstable and strongly dependent et al. (2001), (R850 − R710 )/(R850 − R680 ) developed by Datt (1999)
on not only the choice of calculation method (Fig. 6) but also the and BmSR developed by Maire et al. (2004), with the RMSE less
than both 23.0 mg m−2 and 0.23 mg g−1 for the whole dataset.
plant species (Curran et al., 1995; Datt, 1999; Maccioni et al., 2001);
In addition, the new indices EBAR, EBFN and EBAN also showed a
the estimation error may be as high as 30% (Curran et al., 1995).
good performance with the RMSE of 23.3, 24.15 and 24.33 mg m−2 ,
Furthermore, attention should be paid not only to the red edge
respectively. The performance of REP was strongly dependent on
region but also to the other part of the first derivative curve, the technique adopted.
e.g. blue edge region. The definition of blue edge, similar to the red The experimental database used to calibrate the indices is
edge, was the maximum slope of reflectance curve during blue and composed only of common four green-leafy vegetable species. This
green regions (490–530 nm). The ratio and normalized difference could be a limitation to generality. It is worth noticing, however,
of the area and the amplitude between red (680–780 nm) and blue no fundamental difference exists between leaf reflectance from
(490–530 nm): EBAR, EBAN, EBFR and EBFN depicted the relative different leafy vegetable species, as shown in this study. Therefore,
change of first derivative curve shape in red edge and blue edge successful application of our results to other green-leafy vegetable
region, and were curvilinear related to chlorophyll (Fig. 7). It should leaves is more probable, but should be tested.
be noted that the performance of EBAR, EBFN and EBAN developed
in this study was also very good, with the RMSE of 23.3, 24.15 and Acknowledgments
24.33 mg m−2 , respectively (Fig. 6).
This study was funded by the foundation of Nature and Science
of Jiangsu province (BK2005167) and National Project of Scientific
4. Conclusions and Technical Supporting Programs (2007BAD87B07). The authors
are indebted to Professor Cao Weixing for kindly lending their ASD
All reported successful indices for leaf chlorophyll evaluation Fieldspec FR spectroradiometer, to Mr. Ju Changhua for his help in
we were able to find in literature have been tested with four reflectance measurements and reflectance indices calculation.
green-leafy vegetable varieties using the same protocol to allow a
general comparison of indices. Most of the indices showed a strong References
relation with leaf chlorophyll content. In general, modified indices Aoki, M., Yabuki, K., Totsuka, T., 1981. An evaluation of chlorophyll content of
with the blue or near red edge wavelength performed better leaves based on the spectral reflectivity in several plants. Research Report of
than their simple counterpart without modification, ratio indices the National Institute of Environmental Studies of Japan, 66, pp. 125–130.
106 L. Xue, L. Yang / ISPRS Journal of Photogrammetry and Remote Sensing 64 (2009) 97–106

Arnon, D.L., 1949. Copper enzymes in isolated chloroplasts. Polyphenoloxidae in Jacquemoud, S., Baret, F., 1990. PROSPECT: A model for leaf optical properties
Beta vulgaris. Plant Physiology 24 (1), 1–15. spectra. Remote Sensing of Environment 34 (1), 75–91.
Barnes, J.D., Balaguer, L., Manrique, E., Elvira, S., Davison, A.W., 1992. A reappraisal Kochubey, S.M., Kazantsev, T.A., 2007. Changes in the first derivatives of leaf
of the use of DMSO for the extraction and determination of chlorophylls a and reflectance spectra of various plants induced by variations of chlorophyll
b in lichens and higher plants. Environmental and Experimental Botany 32 (2), content. Journal of Plant Physiology 164 (12), 1648–1655.
85–100. Lichtenthaler, H.K., Lang, M., 1996. Nondestructive determination of chlorophyll
Blackburn, G.A., 1998a. Quantifying chlorophylls and carotenoids at leaf and canopy content of leaves of a green and an aurea mutant of tobacco by reflectance
scales: An evaluation of some hyperspectral approaches. Remote Sensing of measurements. Journal of Plant Physiology 148 (5), 483–493.
Environment 66 (3), 273–285. Maccioni, A., Agati, G., Mazzinghi, P., 2001. New vegetation indices for remote
Blackburn, G.A., 1998b. Spectral indices for estimating photosynthetic pigment measurement of chlorophylls based on leaf directional reflectance spectra.
concentrations: A test using senescent tree leaves. International Journal of Journal of Photochemistry and Photobiology B: Biology 61 (1), 52–61.
Remote Sensing 19 (4), 657–675. Maire, G., Francois, C., Dufrene, E., 2004. Towards universal broad leaf chlorophyll
Buschman, C., Nagel, E., 1993. In vivo spectroscopy and internal optics of leaves as a indices using PROSPECT simulated database and hyperspectral reflectance
basis for remote sensing of vegetation. International Journal of Remote Sensing measurements. Remote Sensing of Environment 89 (1), 1–28.
14 (4), 711–722. Mc Murtey III, J.E., Chappelle, E.W., Kim, M.S., Meisinger, J.J., Corp, L.A., 1994.
Carter, G.A., 1998. Reflectance wavebands and indices for remote estimation of Distinguish nitrogen fertilization levels in field corns (Zea mays L.) with actively
photosynthesis and stomatal conductance in pine canopies. Remote Sensing of induced fluorescence and passive reflectance measurements. Remote Sensing
Environment 63 (1), 61–72. of Environment 47 (1), 36–44.
Carter, G.A., Cibula, W.G., Miller, R.L., 1996. Narrow-band reflectance imagery Miller, J.R., Hare, E.W., Wu, J., 1990. Quantitative characterization of the red edge
compared with thermal imagery for early detection of plant stress. Journal of reflectance 1. An inverted-Gaussian model. International Journal of Remote
Plant Physiology 148 (5), 515–520. Sensing 11 (10), 1755–1773.
Carter, G.A., 1994. Ratios of leaf reflectance in narrow wavebands as indicators of Munden, R., Curran, P.J., Catt, J.A., 1994. The relationship between the red edge
plant stress. International Journal of Remote Sensing 15 (3), 697–703. and chlorophyll concentration in the broadbalk winter wheat experiment at
Chappell, E.W., Kim, M.S., McMurtrey III, J.E., 1992. Ratio analysis of reflectance Rothamsted. International Journal of Remote Sensing 15 (3), 705–709.
spectra (RARS): An algorithm for the remote estimation of the concentrations of Penuelas, J., Filella, I., Baret, F., 1995. Semi-empirical indices to assess
chlorophyll a, chlorophyll b and carotenoids in soybean leaves. Remote Sensing carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica
of Environment 39 (3), 239–247. 31 (2), 221–230.
Cho, M.A., Skidmore, A.K., 2006. A new technique for extracting the red edge Penuelas, J., Gamon, J.A., Fredeen, A.L., Merino, J., Field, C.B., 1994. Reflectance
position from hyperspectral data: The linear extrapolation method. Remote indices associated with physiological changes in nitrogen- and water-limited
Sensing of Environment 101 (2), 181–193. sunflower leaves. Remote Sensing of Environment 48 (2), 135–146.
Curran, P.J., Windham, R.W., Gholz, H.L., 1995. Exploring the relationship between Richardson, A.D., Duigan, S.P., Berlyn, G.P., 2002. An evaluation of noninvasive
reflectance red edge and chlorophyll concentration in slash pine leaves. Tree methods to estimate foliar chlorophyll content. New Phytologist 153 (1),
Physiology 15 (3), 203–206. 185–194.
Datt, B., 1998. Remote sensing of chlorophyll a, chlorophyll b, chlorophyll a + b and Schlemmer, M.R., Francis, D.D., Shanahan, J.F., Schepers, J.S., 2005. Remotely
total carotenoid content in eucalyptus leaves. Remote Sensing of Environment measuring chlorophyll content in corn leaves with differing nitrogen levels and
66 (2), 111–121. relative water content. Agronomy Journal 97 (1), 106–112.
Datt, B., 1999. Visible/near infrared reflectance and chlorophyll content in Sims, D.A., Gamon, J.A., 2002. Relationships between leaf pigment content and
Eucalyptus leaves. International Journal of Remote Sensing 20 (14), 2741–2759. spectral reflectance across a wide range of species, leaf structures and
Daughtry, C.S., Walthal, C.L., Kim, M.S., Brown, de Colstoun E., McMurtrey, J.E., developmental stages. Remote Sensing of Environment 81 (2–3), 337–354.
2000. Estimating corn leaf chlorophyll concentration from leaf and canopy Smith, K.L., Steven, M.D., Colls, J.J., 2004. Use of hyperspectral derivative ratios in the
reflectance. Remote Sensing of Environment 74 (2), 229–239. red edge region to identify plant stress responses to gas leak. Remote Sensing
Filella, L., Penuelas, J., 1994. The red edge position and shape as indicators of plant of Environment 92 (2), 207–217.
chlorophyll content, biomass and hydric status. International Journal of Remote Thomas, J.R., Gausman, H.W., 1977. Leaf reflectance vs. leaf chlorophyll and
Sensing 15 (7), 1459–1470. carotenoid concentration for eight crops. Agronomy Journal 69 (2), 799–802.
Filella, L., Serrano, L., Serra, J., Penuelas, J., 1995. Evaluation of wheat crop nitrogen Vogelman, J.E., Rock, B.N., Moss, D.M., 1993. Red edge spectral measurements
status by remote sensing: Reflectance indices and discriminant analysis. Crop from sugar maple leaves. International Journal of Remote Sensing 14 (8),
Science 35 (5), 1400–1405. 1563–1575.
Gitelson, A.A., Gritz, Y., Merzlyak, M.N., 2003. Relationships between leaf Yoder, B.J., Pettigrew-Crosby, R.E., 1995. Predicting nitrogen and chlorophyll
chlorophyll content and spectral reflectance and algorithms for non-destructive content and concentrations from reflectance spectra (400–2500 nm) at leaf and
chlorophyll assessment in higher plant leaves. Journal of Plant Physiology 160 canopy scales. Remote Sensing of Environment 53 (3), 199–211.
(3), 271–282. Zarco-Tejada, P.J., Miller, J.R., Mohammed, G.H., Noland, T.L., Sampson, P.H.,
Gitelson, A.A., Merzlyak, M.N., 1994. Spectral reflectance changes associate with 2002. Vegetation stress detection through chlorophyll a + b estimation
autumn senescence of Aesculus hippocastanum L. and Acer platanoides L. and fluorescence effects on hyperspectral imagery. Journal of Environmental
leaves. Spectral features and relation to chlorophyll estimation. Journal of Plant Quality 31 (5), 1433–1441.
Physiology 143 (3), 286–292. Zarco-Tejada, P.J., Miller, J.R., Morales, A., Berjon, A., Aguera, J., 2004. Hyperspectral
Horler, D.N.H., Dockray, M., Barber, J., 1983. The red edge of plant leaf reflectance. indices and model simulation for chlorophyll estimation in open-canopy tree
International Journal of Remote Sensing 4 (2), 273–288. crops. Remote Sensing of Environment 90 (4), 463–476.