PE&RS December 2018 Full - page 806

that were conducted for relative chlorophyll content in spring
wheat. Correlation between leaf reflectance spectra and rela-
tive chlorophyll content is presented. NDSI (R849 nm, R850nm), RSI
(R849 nm, R850 nm), and CI (R849 nm, R850 nm) were the most highly
correlated optimized indices with relative chlorophyll con-
tent (), which is in the NIR band (780~1100 nm). Optimized
indices (NDSIs, RSIs and CIs) between the yellow (570~630 nm)
and green regions (530~580 nm) resulted in significant high
correlations (0.4≤|r|≤0.6) (Table 4).
PLSR Model Analysis and Evaluation
In this paper, the PLSR model with predictive variables
was selected based on the rule (VIP) that the addition of other
variables should increase the accuracy of the model (Jin et al.,
2009; Cho, et al., 2007). The influence of each spectral index
in the PLSR model is illustrated in Figure 4 with correspond-
ing VIP values. The VIP method revealed the importance of the
previously published spectral indices and optimized indices
for relative chlorophyll content. The indices (CRI1, CRI2,
SIPS, G-M, and SR) revealed more influence and maximum
VIP values (value ≥1) in previously published spectral indi-
ces. The local maximum VIP values (value ≥1) were 536 nm,
553 nm, 539 nm, 510 nm, 540 nm, 849 nm, and 850 nm in
optimized spectral indices, whereas 849 and 850 nm exhib-
ited the highest VIP values (value≥1.2). The best predictive
variables for the PLSR model were selected, and four models
(α, β, γ and δ) were developed to estimate relative chlorophyll
content in spring wheat. Selected variables and details of PLSR
model accuracy are described in Table 5. Scatter plots of the
predicted and measured relative chlorophyll content values
for the PLSR predictive models were established to illustrate
the predictive ability of the PLSR models in relative chlo-
rophyll content estimation using the R2 and RMSE Pre for the
independent validation dataset (n = 66).
Predictive model results are presented in Table 5. Based
on the coefficient (R2) and RMSE , the -model showed that the
highest coefficient (R2
Pre=0.74) and lowest RMSE (RMSEPre=2.72
SPAD) was identified with four optimized indices (CI (R849 nm,
R850 nm), CI (R539 nm, R553 nm), CI (R540 nm, R553 nm), and CI (R536 nm, R553
nm)). The models and also exhibited good predictive ability,
with R2
Pre values of 0.67 and 0.68, respectively, and RMSE Pre
values of 3.13 and 3.01 SPAD in optimized indices (NDSI (R849
nm, R850 nm), NDSI (R539 nm, R553 nm), NDSI (R540nm , R553 nm), RSI (R849 nm,
R850 nm), RSI (R539 nm, R553 nm), and RSI (R540nm , R553 nm)). The -model
had low R2
Pre (0.10), high RMSEPre (5.55 SPAD), and weak predic-
tive ability for estimating the relative chlorophyll content in
all the studies.
Scatter plots of the predicted and measured relative
chlorophyll content using predictive models are presented in
Figure 5. Overall, 16 previously published spectral indices
and three optimized spectral indices (NDSIs, RSIs, and CIs) were
(a)
(b)
(c)
Figure 3. Three-dimensional maps (contour diagrams)
demonstrating the correlation (r) between the relative
chlorophyll content and narrow-band RSI, NDSI, and CI
calculated from all possible two-band combinations in the
range of 400-1300 nm.
Table 4. The highest values of correlation (r) between optimized indices and relative chlorophyll content.
Variables |r|≤ 0.7 0.6≤ |r|≤0.7
NDSIs (R849 nm– R850 nm)/(R 849 nm+R850 nm) (R539 nm– R553 nm)/(R 539 nm+R553 nm), (R540 nm– R553 nm)/( R540 nm–R 553 nm)
RSIs (R849 nm/R 850 nm) (R539 nm/R 553 nm), (R 540 nm/R553 nm)
CIs (R-1
849 nm– R-1
850 nm)R850 nm (R-1
539 nm– R-1
553 nm)R553 nm, (R -1
540 nm–R -1
553 nm)R553 nm, (R-1
536 nm–R-1
553 nm)R 553 nm
Table 5. Models for prediction.
Models
Number of
VIP (VIP1.0) Variables R2
Cal
RMSECal
(SPAD) R2
Pre
RMSEPre
(SPAD)
α 5 CRI1, CRI2, SIPS, GI, SR 0.27 4.22 0.10 5.55
β 3 NDSI ( R849nm, R 850nm), NDSI (R539nm, R553nm), NDSI (R 540nm, R553nm) 0.62 3.53 0.67 3.13
γ 3 RSI (R 849nm, R850nm), RSI (R539nm, R 553nm), RSI (R 540nm, R553nm) 0.64 3.47 0.68 3.01
δ 4 CI (R 849nm, R850nm), CI (R539nm, R553nm), CI (R540nm, R 553nm), CI (R 536nm, R553nm) 0.67 3.48 0.74 2.72
806 December 2018 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING