for monitoring purposes (Gitelson
et al
., 2004; Sun
et al
.,
2018). Traditional techniques for measuring the plant relative
chlorophyll content require destructive point sampling, which
is costly, laborious, and time consuming and provides limited
spatial coverage (Herrmann
et al
., 2013). Remote estimation of
the relative chlorophyll content has the potential to overcome
these problems. Currently, the use of remote sensing for crop
monitoring is very relevant (Gajjar
et al
., 2005). It is necessary
to rapidly assess the physiological state of a plant during the
growing season (Sims et al., 2002; Krenchinski
et al
., 2017).
Remote sensing uses a variety of methods and indices
based on the spectral characteristics of plants or crops (Zhang
et al.
, 2010; Gitelson
et al
., 2005). For example, chlorophyll
a
and
b
absorb the largest proportion of photosynthetically
active radiation (
PAR
), which provides energy for the photo-
synthesis reaction (Ustin
et al
., 1999). As auxiliary pigments
of chlorophyll
a
(Blackburn
et al
., 1998), carotenoids protect
the reaction centers from excess light and help intercept
PAR
.
Therefore, photosynthetic pigments are strongly related to
the physiological condition of the plant and its productivity.
Based on this notion, numerous hyperspectral chlorophyll
indices have been developed to estimate the relative chloro-
phyll content using remote sensing data in different environ-
ments and for different crop types (Baret
et al
., 1988; Black-
burn
et al
., 1998; Carter
et al
., 1994; Chappelle
et al
., 1992;
Dash
et al
., 2004; Daughtry
et al
., 2000; Filella
et al
., 1995;
Gamon
et al
., 1992; Gitelson
et al
., 1997; Haboudane
et al
.,
2002; Kim
et al
., 1994).
In recent years, numerous studies have estimated the rela-
tive chlorophyll content. Shibayama and Akiyama (Shibaya-
ma
et al
., 1986) reported a good correlation between the
reflection of rice leaves and their relative chlorophyll content
at a wavelength of 550 nm. Ulissi
et al
., (2011) demonstrated
that the chlorophyll spectrum range of 496~694 nm was
highly correlated with the analyzed leaf N concentration and
reported a portable spectrophotometer for the N concentra-
tion of tomato leaves based on visible and near-infrared (
VNIR
)
spectroscopy. The use of spectrometry sensors for crop nutri-
tion measures has been extensively studied (Peng
et al
., 2014;
Yao
et al
., 2013; Jiménez
et al
., 2012). The
MERIS
(medium-
resolution imaging spectrometer) terrestrial chlorophyll index
(
MTCI
) was evaluated using model spectra, field spectra, and
MERIS
data (Dash
et al
., 2004; Frampton et
al
., 2013). Holer
et
al
. (1983) studied the relationship between the value of the
spectrum and the chlorophyll concentrations and proposed
the role of the red edge position for the vegetation chlorophyll
concentration estimation. Another group of researchers used
a vegetation index to determine the reflectance in different
spectral regions, namely, red and near-infrared and green and
red (Gitelson
et al
., 2005). Fang
et al
., (2007) used samples of
rape leaves and proposed a model to predict the relative leaf
relative chlorophyll content using two parameters of red edge
position and the peak position of the green spectral band.
Predicting the relative leaf relative chlorophyll content by
spectral analysis is feasible.
The traditional hyperspectral vegetation index is still
vulnerable to many factors and still has many limitations. It
is difficult to achieve the best universal spectral parameters
that will lead to the model established by the hyperspectral
vegetation index for crop cultivars under different nutrient
conditions. There is a great difference in prediction abilities
of relative chlorophyll content, and the accuracy cannot meet
the needs of practical applications.
Ratio spectral indices (
RSIs
), chlorophyll indices (
CIs
) and
simple normalized difference spectral indices (
NDSIs
) using all
possible combinations of two bands outperformed previously
published spectral indices for the prediction of biochemical,
biophysical and structural plant parameters (Ashourloo
et al
.,
2014; Delalieux
et al
., 2009; Inoue
et al
., 2012; Marshall
et al
.,
2016; Pôças
et al
., 2015; Stagakis
et al
., 2010; Stratoulias
et al
.,
2015). In addition, two-dimensional visualization of the cor-
relation between optimized spectral indices and physiological
parameters provides a clear overview of sensitive wavebands
and spectral regions for determining indices to predict the
various parameters under study (Inoue
et al
., 2008).
The primary objectives of this study were (1) to investigate
the ability of field spectroscopy to characterize the physiolog-
ical status of spring wheat with different relative chlorophyll
content based on
in situ
measurements; and (2) and identify
the most effective indices and predictive models for the early
detection of plant response to relative chlorophyll content
using the
NDSIs
,
RSIs
,
CIs
and partial least squares regression
(
PLSR
) approaches. With the goal of deepening the hyperspec-
tral data, the hyperspectral estimation accuracy of the spring
wheat relative chlorophyll content can be further improved,
providing scientific support and application reference for
the design of hyperspectral sensors and the development of
regional precision agriculture.
Materials and Methods
Study Site
Field data were collected from farmland situated in Fukang
City, Xinjiang, China (44°23
′
12
″
to 44°23
′
15
″
N, 87°34
′
5
″
to
88°34
′
10
″
E), at an altitude of 577 m during the growing season
in 2017. The study sites are presented in Figure 1. The region
has a temperate continental dry climate with an average annual
temperature of 6.7
°C
and mean annual rainfall of 205 mm. The
region is extremely hot in summer and has a large temperature
difference between day and night (Nijat
et al
., 2017).
The spring wheat sowing time in this study began in 20
April 2017 (day of year (
DOY
) 96). The sowing seed per acre
was 30 kg and the row spacing was 17 cm. The fertilizer ap-
plied at the same time with the seeds was diammonium phos-
phate 10 kg/acre, Ammonium sulfate 10 kg/acre, and potash
fertilizer 7 kg/acre. The three leaf period combined with rain-
fall and fertilized urea 20 kg, irrigation by drip irrigation and
irrigation 50-60 m
3
/acre. The weeds in the field were removed
by chemical means. 20% Bromoxynil octanoate E.C. was
sprayed 100 ml. before the 2 leaf stage to the jointing stage of
spring wheat. For measurement purposes, the same treat-
ments were maintained during the growing season, and both
the timing and amount of water were determined based on a
drip irrigation system. Daily average precipitation, maximum
and minimum air temperature, and precipitation are provided
in Figure 1b with field data collection dates for the one-year
study. The research area is designed with 165 sampling plots.
Each sampling plot is 1
×
1 m in size. The data measurements
were obtained at nine different points in the form of Figure
1c, in order to reduce the test error.
Field Data Collection
In 2017, we collected the relative chlorophyll content of
spring wheat and its leaf spectra using field spectroscopy.
The data were collected during the heading stage on 04 June
2017. On that measurement day, the sky was not cloudy.
The day was sunny and suitable for field data collection. No
clear damage caused by weather factors was identified on the
spring wheat leaves.
Plant Physiological Measurements
In this study, the relative chlorophyll content of spring wheat
leaves was non-destructively measured with the use of the
USA Minolta
SPAD
-502 chlorophyll meter. It calculates the
SPAD
value by measuring the absorptivity of the leaf in the red
and near-infrared regions, and makes the value proportional
to the chlorophyll concentration inside the leaf (Uddling
802
December 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
