index, the linear combination of
DVI
(1712,1382)
,
NDSI
(2201,1870)
and
RSI
(2259,1870)
improved the estimation of
VWC
. Such
improvement may be caused by the employment of more
spectral information.
Verification
In order to examine the stability and universality of the mod-
els, 30 independent samples were used to test the predictive
power of the model. All vegetation indices, which calculated
from the dataset, were used to calibrate univariate regression
models, and the calibrated regression models were validated
using the validation dataset. In this study, four statistical pa-
rameters, namely, R
2
,
RMSE
,
RPD
and F-test were evaluated the
performance of the established models. The results are shown
in Table 4.
As shown in Table 4 and Figures 9 and 10, the performance
and stability of published vegetation indices’ models had
some differences with the newly developed ones. The R
2
of
published vegetation indices models were under 0.5,
RPD
were less than 1.4, indicating that the models were very weak
in predicting samples and basically did not have the ability to
estimate
VWC
. The R
2
of newly developed vegetation indices
model were above 0.75, and optimized spectral indices
model were above 0.8, the
RPD
for
NDSI
(2201,1870)
and
RSI
(2259,1870)
and optimized spectral indices model were greater than 2,
indicating that the models had a good prediction of its
VWC
,
the
RPD
for
DVI
(1712,1382)
was 1.861, indicating the ability of the
model had a rough estimate of its
VWC
. This result showed
that the combination of
DVI
(1712,1382)
,
NDSI
(2201,1870)
and
RSI
(2259,1870)
might be as universal equation to monitor
VWC
in
Halophyte
Leaves, and the previous vegetation water indices may not be
Figure 9. Relationship graphs between estimated and measured values of published spectral indices.
544
September 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
