PE&RS December 2018 Full - page 804

spectroradiometer (ASD ) (Analytical Spectral Devices, Inc.,
USA). The spectroradiometer recorded spectral information
of the target in a full wavelength range (350~2500 nm) with
a resolution of 1.4 nm in the 350-1000 nm range and 2 nm in
the 1000~2500 nm range. The field measurements were im-
plemented at 10:00 a.m.-12:00 p.m. local time, and ten mea-
surements were obtained at different points in the plot. The
average leaf spectral reflectance was used for further analysis.
The ViewSpec™ application was used for post-processing of
spectrum files that were saved using the ASD instrument. We
used ViewSpec™ software to manipulate the spectroradiom-
eter and collect data efficiently in the field. The spectroradi-
ometer was configured to average ten spectra automatically
per plot, and the raw spectrum bandwidth was interpolated to
1 nm, resulting in 2,151 individual spectral bands.
Methods
Selection of Spectral Indices and Sensitive Bands
One of the most effective approaches to explore significant
relationships between the plant physiological index and hy-
perspectral data is conducting a comparative analysis of RSIs,
CIs, and simple NDSIs, which are calculated from narrow-band
reflectance factor spectra. We identified the wavelengths or
spectral indices (Table 2) that were able to estimate physiolog-
ical parameters, such as relative chlorophyll content. The RSIs,
CIs, and NDSIs were applied to identify optimal wavelengths or
indices. The spectral indices are defined as follows:
Table 2. Optimized spectral indices selected in this study
where R is the spectral reflectance value and the subscripts
(inm and j nm) are wavelengths in nanometers (nm).
Optimized
spectral index
Abbreviation Equation Reference
Ratio Spectral Index RSI R i/Rj (1) Inoue Y 2008
Chlorophyll Index CI (Ri –1 –Rj –1)Rj (2) Gitelson 2003
Simple Normalized
Difference Spectral
Index
NDSI (Ri –R j)/( Ri +R j) (3) Inoue Y 2008
The spectral indices (RSIs, CIs, and NDSIs) were calculated
for the measured leaf hyperspectral reflectance using all pos-
sible combinations of available bands (inm and j nm) in the full
spectral region (350~2500 nm), excluding the 1301~1500 nm,
1801~2000 nm, and 2401~2500 nm regions due to strong
atmospheric H2O and CO 2 absorption. We examined the cor-
relations between in situ relative chlorophyll content and
spectral indices (NDSIs, RSIs, and CIs) and generated 2-dimen-
sional maps of correlation coefficients (r). The r maps allowed
for the evaluation of the different band combinations and the
selection of sensitive spectral indices (RSIs, CIs, and NDSIs) for
relative chlorophyll content (Stagakis et al., 2010; Stratoulias
et al., 2015).
In this paper, 17 kinds of previously published indices
were selected by leaf pigment and greenness (as provided in
Table 3), which have certain correlation characteristics with
relative chlorophyll content of leaves (Maimaitiyiming et al.,
2017). The results of the identified spectral indices were then
compared with previously published indices from relevant
literature. The most effective NDSIs, RSIs, and CIs and sensi-
tive indices from the literature were determined using the
significance test at the 0.01 level and the highest correlation
coefficient (r) with chlorophyll concentration.
Partial Least-Squares Regression
Partial least-squares regression (PLSR) is a multivariate regres-
sion method that specifies a linear relationship between a
set of dependent response variables, Y, and a set of predictor
variables, X (Haaland et al., 1988). PLSR is a popular model-
ing technique applied in chemo-metrics and is commonly
used for quantitative spectral analyses. To select the optimum
number of factors and avoid overfitting, we calibrated the
model by an iterative leave-one-out cross-validation criterion
called the “minimum predicted residual sum of squares”.
Root mean square error (RMSE ) was minimized by iteratively
leaving one sample out of the calibration dataset and calibrat-
ing the model from the remaining dataset (Maimaitiyiming
et al., 2017). Geladi and Wold described the PLSR method in
detail (Geladi et al., 1986; Wold et al., 2001). The PLSR process
was performed using DPS® (Version 16.05). The predictive
ability of the best selected PLSR model was assessed using the
R2 and RMSE on the independent validation dataset.
Table 3. Previously published spectral indices used in this study.
Spectral Index Acronym Equation References
Leaf pigment
Anthocyanin (Gitelson) AntGitelson AntGitelson=(1/R550−1/R 700)×R780 Gitelson 2003
Carotenoid Reflectance Index 1 CRI1 CRI1=1/R 510−1/ R550 Gitelson 2002
Carotenoid Reflectance Index 2 CRI2 CRI2=1/R 510−1/ R700 Gitelson 2002
G-M G-M G-M=(R750−R 705)/(R750+R705) Gitelson 1994
Optimized Soil-Adjusted Vegetation Index OSAVI OSAVI=(1+0.16)×(R800−R670)/(R800+R670+0.16) Rondeaux 1996
Red–Green Index RGI RGI=R690/R550 Zarco 2005
Structure Insensitive Gigment Index SIPI SIPI=(R800−R450)/(R800+R650) Penuelas 1995
Transformed Chlorophyll Absorption Reflectance Index TCARI TCARI=3×((R 700−R 670)−0.2×( R700−R550)×(R700/R670)) Haboudane 2002
TCARI/OSAVI TCARI/OSAVI Haboudane 2002
Normalized Pigment Chlorophyll index NPCI NPCI=(R680−R430)/(R680+R430) Peñuelas 1994
Greenness
Enhanced Vegetation Index EVI EVI=(2.5×(R 782−R 675)/(R 782+6×R 675−7.5×R 445+1)) Huete 2002
Normalized Difference Vegetation Index NDVI NDVI=(R800−R670)/(R 800+R 670) Rouse 1973
Greenness Index GI GI=R 554/ R677 Zarco 2005
Green NDVI GNDVI GNDVI=(R750−R540+R570)/(R 750+ R 540−R 570) Gitelson 1996
Red Edge Inflection Point REIP REIP=700+40×{[(R670+R780)/2−R700]/(R 740−R 700)} Guyot 1988
Simple Ratio SR SR=R900/R 680 Rouse 1973
Triangular Vegetation Index TVI TVI=0.5×(120×(R750−R550)−200×( R670−R550)) Haboudane 2004
804 December 2018 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING