The
Populus euphratica
reflectance curves revealed that
the spectrum reflectance of 60% to 75%
VWC
in the range of
375 to 680 nm was highest, while the spectral reflectance of
45% to 60% water content was highest at 850 nm. Moreover,
the spectral reflectance decreased with increased
VWC
, so the
spectrum of the two water content layers could be effectively
distinguished. In the first order differential spectrum, 380
to 550 nm and 1500 nm could distinguish water contents of
45% to 60% and 60% to 75% in the leaf effectively, while the
other spectral ranges could not be distinguished (Table 2).
Relationship of VWC Against Two-Band Vegetation Indices
In order to further clarify the sensitive bands of
VWC
in
halo-
phytes
,
the researchers measured the spectral reflectance of
VWC
in May 2016, established the contour maps of
DVI
,
NDSI
and
RSI
with the coefficient of
VWC
by using Matlab R2012a.
We constructed the best spectral index, according to the dif-
ferent color of contour map, and extracted the sensitive band
combination of
VWC
(Figure 8).
The contour maps of R
2
values between
VWC
and two-band
vegetation indices (
DVI
,
NDSI
, and
RSI
) are shown in Figure 8;
this method extracted sensitive region and significant peak-
wavelengths efficiently. The contour map for
NDSI
exhibits
a similar distribution pattern as the map for
RSI
. For
DVI
; the
high significant areas were mainly found at the near-infrared
and mid-infrared regions, X:1400 to 1875 nm combining with
Y:1300 to 1400nm; X:2000 to 2400 nm combining with Y:1860
to 1900 nm; X:2200 to 2300 nm combining with Y:1400 to
1560 nm. For
NDSI
, the high significant areas were found at
the near-infrared and mid-infrared regions, X:1600 to 1750
nm combining with Y:1380 to 1400 nm; X:2100 to 2500 nm
combining with Y:1880 to 1900 nm. For
RSI
, the high significant
areas were found at the near-infrared and mid-infrared regions,
X:1600 to 1800nm combining with Y:1380 to 1400nm; and
X:2140 to 2500nm combining with Y:1870 to 1900nm.
The maximum coefficient of
VWC
and the three newly
developed indices were all above 0.65, and the correlation
was good. The authors selected
DVI
(1712,1382)
,
NDSI
(2201,1870)
and
RSI
(2259,1870)
to construct the optimized spectral indices, the
coefficients of determination were 0.654, 0.671, and 0.665
respectively.
Modeling and Verification
Modeling
The researchers used six types of vegetation (n = 71) for
modeling, randomly selected 30 measured data as the valida-
tion dataset, and constructed the contour map to screen out
the optimal vegetation index for the study. Compared with
the published vegetation water indices, 15 indices are used
as independent variables in the estimation model of
VWC
. We
adopted two regression methods, both linear and nonlinear.
The nonlinear models include quadratic, compound, growth,
logarithmic, cubic, S-curve, exponential, inverse, power, and
logistic. The results of the best regression models between the
calculated vegetation indices and
VWC
are shown in Table 3.
Table 2. The identify band of original reflectance and first order differential spectra of species.
Vegetation
type
Tamarix
Haloxylon
ammodendron
Alhagi
pseudalhagi
Phragmites
australis
Nitraria
tangutorum
Populus
euphratica
Main
absorption
band
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
380~400nm
680~720nm
1420~1450nm
1900~1940nm
2450~2500nm
Original
reflectance
500~1900nm
1950~2400nm
380~550nm
1200~1425nm
375~700nm
(45%-60%)
750~1000nm
(15%-30%)
(30%-60%)
(60%-75%)
1400~2500nm
(15%-30%)
(60%-75%)
550~675nm
1400~2500nm Can not distinguish
375~680nm
(60%-75%)
850~2500nm
(45%-60%)
First
order
differential
750~975nm
680~750nm
1125~1200nm
(60%-75%)
700~1000nm Can not distinguish Can not distinguish
380~550nm
1500nm
(45%-60%)
(60%-75%)
Table 3. Quantitative relationships between
VWC
of leaves and the optimized and published spectral indices.
Spectrum
parameters
Type
Regression equation
R
2
RMSE
F-test
MSI
WI
SRWI
RSI
(2259,1870)
Ratio type
y=703.698x
3
–959.015x
2
+331.929x+39.925
y=228.204x
2
–356.178x+177.591
y=40.667x
2
–56.750x+66.164
y=96.380+175.601Inx
0.431
0.399
0.327
0.667
12.448
12.692
13.438
9.384
16.902
22.612
16.497
138.097
NDII
NDWI
1200
NDWI
1240
NDWI
1450
NDWI
1640
NDWI
1940
NDWI
2130
NMDI
GVMI
NDSI
(2201,1870)
Normalized
type
y=0.164x
3
–3.222x
2
+23.080x+20.618
y=–4015.276x
3
+1500.729x
2
–9.442x+45.599
y=–4122.569x
3
+1375.815x
2
+26.209x+45.659
y=–32.885x
3
+99.170x
2
+31.307
y=–1327.965x
3
+1498.878x
2
–442.307x+85.010
y=76.217x
1.190
y=151.835x
2
–106.003x+64.534
y=265.052x
2
–206.031x+84.703
y=–1471.406x
3
+1870.283x
2
–665.106x+117.767
y=9781.754x
3
+3345.702x
2
+761.073x+103.783
0.417
0.383
0.378
0.379
0.409
0.135
0.294
0.398
0.434
0.674
12.600
12.960
13.014
12.903
12.685
0.299
13.760
12.709
12.414
9.413
15.957
13.863
13.564
20.771
15.449
10.773
14.161
22.458
17.115
46.274
DVI
(1712,1382)
Difference type
y=10618.768x
3
+3549.659x
2
+766.446x+83.355
0.664
9.567
44.088
542
September 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
