As shown in Table 6, spectral reflectance has different
degrees of response to chlorophyll in the visible and near-in-
frared range. The near-infrared (820~940 nm) spectral reflec-
tance has a strong reflection characteristics, and it has a high
reflectivity values. Our study group estimated the chlorophyll
content of spring wheat by hyperspectral features parameters,
studies have shown that establishing a
PLSR
estimation model
based on wavelength (820～940 nm) worked best. The optimal
band combination range for this study is centered on 820~940
nm, which is similar to our previous studies (Nijat
et al
., 2017).
Table 6. Characteristic parameters of reflectance spectrum.
Spectral
characteristic
parameter
Definition
Characteristic
436~495 nm
Blue edge
Wavelength (436~495 nm)
Reflectivity
Chlorophyl0l and
carotene strong
absorption bands
495~566 nm
Green edge
Wavelength (495~566 nm)
Reflectivity
Strong chlorophyll
reflection peak area
566~589 nm
Yellow edge
Wavelength (566~589 nm)
Reflectivity
Chlorophyll strong
absorption band
627~780 nm
Red edge
Wavelength (627~780 nm)
Reflectivity
The slope of the
reflectance curve is
related to the content
of chlorophyll (a+b) per
unit area of the plant
820~940 nm
near-infrared
Wavelength (820~940 nm)
Reflectivity
With strong reflection
characteristics, it has a
high reflectivity value
Figure 6 shows the correlation between the relative chlo-
rophyll content of spring wheat and the 400～1300 nm band.
There is a positive correlation in the 400-1300 nm band. The
correlation coefficient range is 0
≤
r
≤
0.3. The band passing the
significance test (
p
=0.01) was mainly concentrated in the
visible band (400-520 nm). In this study, we use the two-band
combination method to construct different optimized spectral
indices, and the correlation analysis with relative chlorophyll
content shows that the correlation coefficient (
r
) fluctuated
within the range of -0.7
≤
r
≤
0.7. The effective spectral index
was determined in the massive hyperspectral data, and the
correlation was significantly improved.
Compared with previously published spectral indices, the
models that based on optimized spectral indices (
CIs
) exhibit-
ed a high determination coefficient (
R
2
Pre
=0.74) and low
RMSE
Pre
(2.72
SPAD
). The
PLSR
model obtained with optimized spectral
indices was superior to the previously published spectral
index-based methods. Chlorophyll index (
CI
) is an optimized
selection of the original spectral band of the spectral index
by the conceptual chlorophyll model proposed by Gitelson
(2003). The studies have shown that the developed
CI
al-
lowed accurate estimation of total Chlorophyll in both crops,
explaining more than 92% of Chlorophyll variation (Gitelson
et al
., 2005). In the results of this study,
CI
showed the best
inversion ability in the estimation of leaf relative chlorophyll
content, which is consistent with the above findings.
Algorithm fitting for optimizing the spectral index band
combination and the relative chlorophyll content estimation
model are all important factors that influence the predictive
ability of the model. According to the established model and
the relationship between predicted and measured relative
chlorophyll content, the
CI
was the best improved spectral
index among all the optimized spectral indices that combined
two bands.
The optimized spectral parameters obtained in this paper
can provide a basis for rapidly and accurately searching for
the best wavelength band for monitoring relative chloro-
phyll content of spring wheat in satellite sensors. For ex-
ample,
MODIS
can be selected from two (841~876 nm), four
(545~565nm), or 12 (546~556 nm) channels to achieve the
purpose of monitoring the content of chlorophyll in spring
wheat canopy on a regional scale. In addition, the optimi-
zation of the band can also provide a theoretical basis for
designing active sensors in specific bands, further reduce the
workload of hyperspectral mass data processing, and provide
services for real-time requirements of precision agriculture.
Simultaneously, the stability of the optimized spectral index
was demonstrated once again by verification of the measured
data. This index represents a useful tool for estimating and
predicting relative chlorophyll content in spring wheat during
the growth period. For decision-making, researchers should
consider the optimized spectral index for measuring relative
chlorophyll content in spring wheat.
Finally, the study area belonged to a typical arid and
semi-arid region: China has regional heterogeneity, and spring
wheat is widely planted in China including a wide variety of
plant species. These factors will inevitably affect the results
of this study. Optimization of spectral parameters for rela-
tive chlorophyll content in the spring wheat canopy still has
certain regional limitations. In addition, the heading period is
only a short part of the growing period of spring wheat, and
the time-optimal spectral parameters need to be improved.
Therefore, comparing the optimal parameters of the spectral
index optimization algorithm in different regions and at dif-
ferent times, as well as the application in other crops, will be
worth studying.
Conclusions
The chief objectives of this study were to investigate the po-
tential of field spectroscopy to characterize the physiological
status of spring wheat with different relative chlorophyll con-
tents and identify the most effective spectral indices. Simul-
taneously, we established predictive models for evaluating
the potential of the 400~1300 nm band combinations in the
early detection of plant response to the relative chlorophyll
content using the
NDSIs
,
RSIs
,
CIs
, and
PLSR
approaches during
the heading stage.
In this study, we utilized the leaf reflectance spectra and
relative chlorophyll content measured at the experiment base
during the heading stage. The previously published spectral
indices exhibited weak correlations with relative chlorophyll
content. Strong correlations between optimized spectral
indices, and this physiological parameter demonstrated the
potential of ground-based spectroscopy in determining the
relative chlorophyll content of spring wheat during the head-
ing stage. In particular, in combination with the correlation
coefficient (
R
2
) and
RMSEs
, the
δ
-model showed the highest
R
2
Pre
(0.74) and lowest
RMSE
Pre
(2.72
SPAD
); these values were
identified with four optimized indices (
CI
(
R
849 nm
,
R
850 nm
),
CI
(
R
539 nm
,
R
553 nm
),
CI
(
R
540 nm
,
R
553 nm
),
CI
(
R
536 nm
,
R
553 nm
)). These
optimized spectral indices are spectrally wide and applied to
broad satellite bands.
Variable selection for the
PLSR
model illustrates that the
VIP
method provides insights into the importance of variables,
confirming that the selected variables (optimized spectral
indices) improve the ability of hyperspectral reflectance data
for predicting relative chlorophyll content during the head-
ing stage. Nevertheless, we only employed ground-based
field spectroscopy in this study. Therefore, these results are
applicable to estimating and predicting relative chlorophyll
content indicators in spring wheat during the heading stage
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2018
809
