PERS_September_2018_Flipping_86E2 - page 539

construct the spectral parameters is
more complex; the computation is too
complicated to applied to agricultural
remote sensing practice. However,
the normalized spectral index (NDSI),
difference vegetation index (DVI), and
ratio vegetation index (RSI) are simple,
convenient and effective to reduce soil
noise. They have been widely used in
the spectral study of vegetation (Miller
et al.,1990; Lyon et al., 1998).
Previous studies have shown that
any possible two-band combinations
of vegetation indices (NDSI(Equation
2), DVI(Equation 3), and RSI(Equation 4)
over 380 nm to 2500 nm were built in
the form of a matrix linkage. NDSI, DVI
and RSI were based on the formula:
NDSI R R
R R ( , )λ λ
λ λ
λ λ 1 2
1 2
1 2
= −
+ (2)
DVI R R
( , )λ λ λ λ
1 2 1 2 = − (3)
RSI R
R ( , )λ λ
λ
λ 1 2
1
2
= (4)
where Rλ1, Rλ2 are the reflectance values at wavelengths
λ1, λ2 nm, respectively.
Statistical Analysis
In this study, coefficient of determination (R2), and F-test, root
mean square error (RMSE) and residual predictive deviation
(RPD) were used to examine the model’s precision.
In general, the larger the R2 and RPD , the smaller the RMSE,
the better the model’s robustness and the better predictive
ability. For RPD , when RPD≥2, it shows that the prediction
ability of the model is very good; when 1.4 ≤RPD≤2, the model
can make rough estimates of the sample; when RPD＜1.4, indi-
cating that the model of the samples can’t be estimated (Liu et
al., 2017). The completion of data is processing and modeling
used EXCEL and SPSS software.
The RMSE is the square root of the observed value and the
true value deviation and the n-ratio of the observation times.
The F statistic is the ratio of mean square between groups and
mean square sum; if the different levels of control variables
have significant effects on the observed variables, the F value
is high, on the contrary is small (Gao et al., 2008). During the
actual measurement, RMSE and F-test can evaluate the preci-
sion accurately with the 1:1 plot of the two value’s groups .
The RMSE reflects the degree of deviation from the true value
of the measured data, with a value indicating a higher accu-
racy of the measurement.
RMSE P O
n i i
i
n
= × − ( )
=
∑
1 2
1
(5)
where Pi and Oi are the observed values and simulated values
of the i sample, respectively (Feng et al.,2016).
Results
Spectral Characteristics of Vegetation
Under Different Leaf Water Content
Experts believe that environmental conditions, climate types,
vegetation characteristics and other factors have a certain de-
gree of influence on optical remote sensing. Under steady con-
ditions, changes in the vegetation characteristics, such as the
internal structure of leaves, dry matter content, chlorophyll
content, and leaf water content, play a leading role in changes
in the spectral reflectance of plants. Accordingly, these char-
acteristics strongly influence variations in the VWC and are
important to the extraction of vegetation information and the
quantitative retrieval of vegetation (Figure 2). The absorption
band of Tamarix mainly centered on 380 to 400 nm, 680 to
720 nm, 1420 to 1450 nm, 1900 to 1940 nm, and 2450 to 2500
nm. The strong absorption, bands at 380 to 400 nm and 680 to
720 nm, were caused by chlorophyll; the strong one band in
1420 to 1450 nm, 1900 to 1940 nm, and 2450 to 2500 nm, was
caused by VWC; the reflection peak at 550 nm was the green
peak, which was formed by visible light; the high reflectivity
of 750 to 1300 nm formed a steep slope, because of the cell
structure of vegetation. The spectral reflectance of Tamarix
leaf in the bands at 1500 to 1900 nm and 1950 to 2400 nm
was reduced as the VWC increased. Most of the spectral curves
of different leaf water content were similar; the reflection con-
tent at 30% to 45% was lower than that at 15% to 30%. In the
range of 750 to 1300 nm, the variation of the leaf reflectance
and leaf water content was not linear. Thus, 1500 to 1900 nm
and 1950 to 2400 nm bands could be used for water content
identification of Tamarix leaf. In the first order differential
spectrum, only the range of 750 to 975 nm could distinguish
the water layer (Figure 3).
The Haloxylon ammodendron reflectance curves revealed
that the absorption band was similar to Tamarix , which was
caused by the characteristics of the plant. In the spectral range
of 380 to 550 nm and 1200 to 1425 nm, the highest reflectance
of the leaf water content was 15percent to30percent, the
lowest one was 60 percent to 75 percent, and the spectral
reflectance decreased with increased leaf water content. There
was a sharp rise in the range of 680 to 750 nm, which formed a
phenomenon of a red edge similar to Tamarix . In the range of
550 to 675 nm and 1900 to 2500 nm, the reflection spectrum
was reduced as the VWC increased, except the leaf water
content was 45% to 60%. The spectrum in the range of 750to
975 nm of 60% to 75% was significantly higher than that for
leaves with other water content. In the range of 1450 to 1900
Table 1. Spectral index formula.
Water
index Types Calculation formula Reference
MSI
Ratiotype
MSI=R1600/R 820 (Hunt et al., 1989)
WI WI=R900/R970 (Penuelas et al., 1996)
SRWI SRWI=R860/R 1240
(Zarco-tejada et al., 2001;
Zarco-tejada et al., 2003)
NDII
Normalized type
NDII=(R820/R1600)/(R820+R1600 ) (Hardisk et al., 1983)
NDWI1200 NDWI1200=(R860–R1200)/(R860+R1200) (Wu et al., 2009)
NDWI1240 NDWI1240=(R860–R1240)/(R860+R1240) (Gao, 1996)
NDWI1450 NDWI1450=(R860–R1450)/(R860+R1450) (Wu et al., 2009)
NDWI1640 NDWI1640=(R860–R1640)/(R860+R1640) (Chen et al., 2005)
NDWI1940 NDWI1940=(R860–R1940)/(R860+R1940) (Wu et al., 2009)
NDWI2130 NDWI2130=(R860–R2130)/(R860+R2130) (Chen et al., 2005)
NMDI NMDI=(R860–(R1640 –R2130))/(R860+(R1640–R2130 )) (Wang et al., 2008)
GVMI GVMI=((R820+0.1)-(R1600+0.02))/((R 820+0.1)+(R1600+0.02)) (Ceccato et al., 2002)
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2018 539