compared in this study. Compared with previously published
indices, the predictive models with optimized spectral indi-
ces (
NDSIs
,
RSIs
and
CIs
) better estimated the relative chloro-
phyll content in spring wheat. The optimized spectral index
(
CIs
) revealed the best ability to estimate the relative chloro-
phyll content in spring wheat.
Discussion
Chlorophyll is the most important organic molecule in photo-
synthesis, and chlorophyll concentration has the most obvi-
ous influence on the growth of crops (Bannari
et al
., 2007).
The synthesis of nitrogen in crop leaves is related to the in-
ternal structure of chlorophyll, which is an effective method
for estimating the nutrition and physiology of crop based on
chlorophyll levels. The nutritional status of crops is closely
related to spectral characteristics. In particular, research
on field crops has been a hot topic in precision agriculture
(Elarab
et al
., 2015). Spectral reflectivity exhibits a different
spectral response with different levels of chlorophyll. Regard-
ing hyperspectral data with rich information, it is imperative
to illustrate the potential of hyperspectral data in monitoring
chlorophyll levels of spring wheat (Croft
et al
., 2014).
Relative chlorophyll content of crops is estimated by
remote sensing in this study. To eliminate the effects of
background field soil and weather conditions on crop canopy
structure, spectral indices are most commonly used. These
are a simple and reliable method and can extract useful
information from complex spectral reflectance. However,
the traditional
NDVI
based on the red band and near-infrared
band, and some spectral vegetation indices derived from
NDVI
are susceptible to many factors and have certain limitations.
It is difficult to achieve a set of the best universal spectral
parameters. To solve this problem, new vegetation indices and
algorithms have been developed. In this study, band optimiza-
tion was performed on the spectral indices
NDSIs
,
RSIs
, and
CIs
using the band optimization algorithm proposed by Gitelson
et
al
. Optimized spectral indices were calculated to measure leaf
hyperspectral reflectance factor spectra using all possible com-
binations of available bands in the spectral region 400-1300
nm. Hyperspectral data contain more information, and the
most suitable combination of the two bands was chosen for all
possible pairwise band combinations, increasing the correla-
tion between spectral indices and relative chlorophyll content.
(a)
(b)
(c)
(d)
Figure 4. Variable importance in the projection (
VIP
) of the partial least squares regression (
PLSR
) predictive model for relative
chlorophyll content: (a) Variable importance for the
α
-model; (b) Variable importance for the
β
-model, (c) Variable importance
for the
γ
-model, and (d) Variable importance for the
δ
-model.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2018
807
