The photochemical reflectance index (PRI) has emerged to be a pre-visual indicator of water stress. However, whether the varying shaded-leaf fractions, which may be caused by multiple view angles or the changing crop density in the field, affect the performance of PRI in detecting water stress of crops is still uncertain. This study evaluated the impact of the varying shaded-leaf fractions on estimating relative water content (RWC) across growth stages of winter wheat using seven formulations of PRI. Results demonstrated that for the control treatment the mean PRI of sunlit leaves was slightly higher than those of shaded leaves, but the difference between PRI of sunlit and shaded leaves increased as water resources became more limiting. Despite the difference between PRI of sunlit and shaded leaves, the significance of the linear relationship between RWC and most studied formulations of PRI did not show obvious variations with shadow fractions, except for the 100 % shaded-leaf condition. Among the studied formulations of PRI, PRI3 based on reflectance at 512 nm as the reference band provided the most accurate estimates of RWC with varying shaded-leaf fractions, except for the 100 % shaded-leaf condition. The slope and the intercept of linear regression models with PRI3 also showed minimized variations with shaded-leaf fractions. We then applied a uniform RWC prediction model to the data of varying shaded-leaf fractions and found that the accuracy of RWC predictions was not significantly affected in the mixture of sunlit and shaded leaves. However, RWC estimated with PRI of the 100 % shaded-leaf condition had the highest root mean square error (RMSE), implying that PRI of the pure shaded leaves may yield inaccurate estimates of plant water status.
Agriculture consumes about 80 %–90 % of freshwater worldwide (Gonzalez-Dugo et al., 2010). Water stress is one of the most critical abiotic stressors limiting plant growth and crop production (Chaves et al., 2003). Climate change, increasing worldwide shortages of water, and frequent droughts are exacerbating the agricultural water crisis and putting global food security at risk (Hirich et al., 2016; Lei et al., 2016). The assessment of water status in crops is critical for precision irrigation practices, balancing crop production with water supply and sustainable farming.
Remote sensing provides a unique tool to unobtrusively, efficiently, and quantitatively assess water status in crops. Water stress induces plants' stomatal closure, leading to the increasing leaf temperature due to the decreasing evaporative cooling. Remotely monitoring the change in canopy temperature provides information on instantaneous transpiration status, and hence thermal remote sensing has served as an effective tool in detecting water stress for decades (Idso et al., 1981; Sayago et al., 2017). However, thermal remote sensing of water stress has limitations in both physiological and operational aspects. The physiological relationship between canopy temperature and stress is not clear for some crops (Villalobos et al., 2009). Due to the technical reasons, the spatial resolution of thermal imaging sensors is generally coarser than the visible and infrared sensors, limiting its applications at local scales.
In a recent decade, the photochemical reflectance index (PRI) has emerged to be a pre-visual indicator of water stress. PRI is a normalized difference of reflectance at 531 nm and reflectance at a reference band (e.g., 570 nm) in the visible domain. It was initially proposed as an indicator of the de-epoxidation state of xanthophyll pigments, which is related to photosynthesis (Gamon et al., 1992). When the light absorbed by the plants exceeds the photosynthetic demand, de-epoxidation of xanthophyll cycle pigments occurs, leading to the downregulation of photosynthesis (Gamon et al., 1992). Multiple abiotic stressors, including nutrient deficiency (Shrestha et al., 2012; Magney et al., 2016), excessive heat (Dobrowski et al., 2005), and water deficit (Muller, 2001; Sun et al., 2008; Sarlikioti et al., 2010; Zarco-Tejada et al., 2013; Magney et al., 2016), have been shown to trigger the xanthophyll cycle, resulting in the apparent drop in reflectance at 531 nm.
As a promising alternative to thermal remote sensing for monitoring plant water stress, several previous studies have investigated the feasibility of assessing plant water status at leaf level and canopy level using PRI. At leaf level, a number of studies demonstrate a close relationship between PRI and physiological indicators of water stress (Thenot et al., 2002; Shahenshah et al., 2010), but some other studies report a poorer relationship due to the confounding environmental factors (Sarlikioti et al., 2010) or the changes in pigment pools (Sun et al., 2008). At canopy level, studies show stronger correlations between changes in physiological indicators of water stress and PRI, in comparisons with the other indices (e.g., normalized difference vegetation index, NDVI) (Suárez et al., 2008; Rossini et al., 2013; Zarco-Tejada et al., 2013). However, the performance of canopy PRI in the water stress detection is affected by canopy structure, canopy cover, and viewing geometry (Rossini et al., 2013; Panigada et al., 2014). Particularly, at seasonal and interannual timescales, physiological changes, such as relative water content and pigment pools, concurrently occur with structural changes, such as leaf area index (LAI). Canopy PRI is sensitive to the structural changes during the growth season (Gitelson et al., 2017). To minimize the impact of canopy structures on PRI, transformations of PRI are developed using the band insensitive to the canopy structure (Hernández-Clemente et al., 2011), the structural vegetation indices for the normalization (Zarco-Tejada et al., 2013; Gitelson et al., 2017), or the radiative transfer modeling results (Hernández-Clemente et al., 2011).
PRI is primarily driven by the xanthophyll cycle at a short timescale
(e.g., a few hours, 2–3 d), but shaded leaves may not experience
de-epoxidation of the xanthophyll cycle as the sunlit leaves do. As PRI is
expected to be applied to monitoring water stress at a large scale, canopy PRI
derived from satellite data includes contributions from both the sunlit
leaves and shaded leaves. Hall et al. (2008) and Hilker
et al. (2010) found that canopy PRI was strongly dependent
on canopy shadow fractions, because the xanthophyll cycle status was
affected by incident PAR, which was in turn affected by the level of
self-shading within a canopy. Cheng et al. (2009) examined the
contributions of variable sunlit
Previous studies have shown that within-canopy shadowing effects directly affect PRI of a canopy, but whether the proportion of shaded leaves further influences the performance of detecting water stress in the growth season of a crop using PRI is still uncertain. The objective of this study is to analyze the impact of varying shaded-leaf fractions on the performance of canopy PRI in detecting water stress during the growth season of winter wheat using a hyperspectral imager. To accomplish this objective, we conducted water stress experiments on winter wheat for 2 consecutive years. Reflectance of shaded and sunlit leaves derived from hyperspectral imagery was mixed with varying fractions to quantify the impacts of shaded leaves on different formulations of PRI in detecting water stress during the growth season.
During the growth seasons of 2016 and 2017, two water stress experiments
were conducted in the facilities at Huazhong Agricultural University, China
(30
Seedlings were grown outdoor under the natural condition before the water stress experiments started. Soil water content was measured every 4–5 d using time domain reflectometry (TDR300, Spectrum Technology Inc., USA), and tap water was supplied if soil water content was 70 % off field capacity. Water stress treatments started at the end of February, which was during the tiller initiation stage. Pots were moved to a rain-out shelter to prevent external water supply. In 2015–2016, 28 pots were divided into five groups. A group of four pots was used as the control, which had sufficient water supplies throughout the experiment. The other four groups (with six pots for each group) stopped watering on 24 February, 6, 28 March, and 8 April, respectively. In 2016–2017, 15 pots were divided into five groups. A group of three pots was used as the control, which had sufficient water supplies throughout the experiment. The other four groups (with three pots for each group) stopped irrigation on 15, 22, 29 March, and 12 April, respectively. After irrigation stopped, soils of the treated pots were left to dry as analogs for the natural drought condition. In 2016, measurements were taken every 2–5 d depending on the weather conditions until immature senescence occurred. For the water treatment group, three pots of winter wheat were used for capturing hyperspectral images, and the other three pots were used to collect samples. In 2017, measurements were taken every 4–6 d until immature senescence occurred. For the water treatment groups, one pot of winter wheat was used for capturing hyperspectral images, and the other two pots were used to collect samples. In both years, physiological and spectral measurements were taken in control groups during the whole experiment.
In this study, we used relative water content (RWC) as the indicator of
water stress, because RWC was recommended by previous studies as an
effective physiological indicator of water status (Hewitt
et al., 1985; Siddique et al., 2000). We randomly chose three
plants in the sampled pot, and top three leaves of the sampled plants were
cut from the stem. Leaves were cut into 10 small round pieces with a
puncher and put into a zip-lock bag. Leaf samples were enclosed in a cooler
and brought to the laboratory to measure RWC. In the laboratory, fresh
weight was measured with an electronic balance. The leaf samples were
immersed in distilled water for 16–18 h. We dried the surface moisture
and weighed the turgid weight. Afterward, all samples were put into aluminum
boxes to dry in the oven at 105
Hyperspectral images were recorded in situ using the SOC710VP portable
hyperspectral imager (Surface Optics Corporation, SOC, USA). The imager has
We manually selected regions of interest (ROIs) of the most deeply shaded leaves and the brightest sunlit leaves in each image using ENVI 5.1 (the Environment for Visualizing Images) (Fig. 1). Reflectance within ROIs were averaged and used as reflectance of sunlit and shaded leaves, respectively. Based on the assumption of the linear mixture of shadow and sunlit leaves, we mixed different fractions of shaded-leaf reflectance with sunlit-leaf reflectance to evaluate the impact of shaded leaves on detecting water stress with PRI.
The derived spectral data were interpolated to 1 nm band width using the
cubic spline interpolation function in MATLAB (R2011a) software. Overall, seven
formulations of PRI were calculated for both sunlit leaves and shaded leaves
(Table 1). In addition, we calculated the difference (
The original hyperspectral image shown as an RGB image. Region of interest (ROI) A is the sunlit leaves, ROI B is the shaded leaves, and ROI C is the reference spectral panel.
Seven PRI formulations used in this study.
Measurements taken from pots that had the same water treatments were
averaged and used in the analysis. The maximum, minimum, coefficient of
variation (CV), and standard deviation were used to describe the range and
the variation of observations. To analyze the variations of PRI in sunlit
and shaded leaves during the water stress treatment, we divided all the data
into seven groups according to RWC values (RWC ranges of 0.2–0.3, 0.3–0.4,
0.4–0.5, 0.5–0.6, 0.6–0.7, 0.7–0.8, and > 0.8). The mean and the
standard deviation were calculated to evaluate the variations of PRI against
RWC. The least-square linear regression model was established to estimate
RWC with PRI of sunlit leaves and shaded leaves. The quadratic function was
used to describe the relationships between shaded-leaf fractions and the
slope/intercept of the linear regression model between PRI and RWC.
The spectra of sunlit leaves and shaded leaves are presented in Fig. 2. The reflectance of the shaded leaves was lower than that of the sunlit leaves. Unlike the spectra of sunlit leaves, the green peak of shaded leaves was not obvious.
We calculated the difference (
Spectra of sunlit and shaded leaves. The solid lines are the mean reflectance of the samples and the shadings are the standard deviation.
The maximum, minimum, mean, CV and range of the difference
(
The mean and the standard deviation of PRI570 in sunlit leaves and shaded leaves against RWC. Data were divided into seven groups according to RWC values (RWC ranges of 0.2–0.3, 0.3–0.4, 0.4–0.5, 0.5–0.6, 0.6–0.7, 0.7–0.8, and > 0.8).
To assess the impact of shaded leaves on detecting water stress with PRI, we
mixed different fractions of shaded-leaf reflectance with sunlit-leaf
reflectance and analyzed the relationship between RWC and PRI calculated
from the mixed reflectance of shaded and sunlit leaves. Different
formulations of PRI were all positively correlated with RWC (Table 3). Among
the studied formulations, PRI2, PRI3, PRI4, and PRI6 showed significant
correlations with RWC in winter wheat with the varying shaded-leaf
fractions, except for the 100 % shaded-leaf condition.
We further analyzed the impact of shaded-leaf fractions on the slope and intercept of the linear regression model between PRI and RWC. The slope and intercept of the linear regression models between different formulations of PRI and RWC were strongly correlated with fractions of shaded leaves (Table 4). The quadratic function was applied to describe the non-linear relationship between shaded-leaf fractions and the slope/intercept. Examples of non-linear relationships for PRI570 and PRI3 are shown in Fig. 5. For all the studied formulations of PRI, the intercept remained relatively stable under different shaded-leaf fractions, except for the 100 % shaded-leaf fraction. The slope increased non-linearly with shaded-leaf fractions for most of the studied PRI, but PRI3 did not show obvious variations in the slope under the shaded-leaf fractions below 70 % (Fig. 5d).
To evaluate if these changes in the values of linear regression parameters
affected the accuracy of RWC estimates, we applied the linear regression
model derived from the PRI of the generally applicable sunlit
Examples of the linear relationship between PRI of sunlit
leaves (PRI2,
Relationships between shaded-leaf fractions and the slope
of the linear regression models of RWC and PRI570 (
RMSE of RWC estimated with PRI570 (
Theoretically, sunlit leaves are more likely to experience high light-induced environmental stress than shaded leaves (Hilker et al., 2008; Middleton et al., 2009; Cheng et al., 2012). Data from previous field samplings and model simulations, although limited, confirmed the impact of shaded-leaf fractions on PRI values (Middleton et al., 2009; Cheng et al., 2012; Takala and Mõttus, 2016). While interests of detecting plant water stress with PRI are increasing, studies rarely analyzed the impact of shaded leaves on the performance of PRI in water stress detection. This study quantified the differences between PRI of sunlit and shaded leaves in winter wheat under control and water stress conditions, and investigated the impact of varying shaded-leaf fractions on water stress detection during the growth season, using different formulations of PRI derived from hyperspectral images.
Our results showed that for the control treatment the mean PRI of sunlit
leaves was slightly higher than that of shaded leaves. Take PRI570, for
example.
Interestingly, our results showed that
Although the PRI of shaded leaves was different from PRI of sunlit leaves under both control and water stress conditions, the effect of the varying fractions of shaded leaves did not lead to the substantial change in the significance of the relationship between PRI and RWC. We hypothesized it was because the shallow soil in the pot experiment caused the quick wilting during the water stress treatment, and thus the changes in leaf area and pigment content intertwined with physiological responses. Among the studied formulations of PRI proven to minimize the effect of structural change in canopies in previous studies (Hernández-Clemente et al., 2011; Zarco-Tejada et al., 2013), PRI3 that used reflectance at 512 nm as the reference band provided the most accurate estimates of RWC with varying shaded-leaf fractions, except for the 100 % shaded-leaf fraction. PRI3 was originally developed for the needle tree based on the evidence that reflectance at 512 nm was not responsive to the change in xanthophyll epoxidation state (Hernández-Clemente et al., 2011). In their study, PRI3 showed the highest correlation with the stomatal conductance and water potential at the canopy level and the lowest sensitivity to canopy structure, in comparison with PRI570 and NDVI. Our results also showed the superior performance of PRI3 to the other formulations of PRI in estimating RWC, implying that for winter wheat the 512 nm band might be a better reference band that could maximize the physiological responses of the 531 nm band. Unfortunately, we could not provide direct evidence of PRI3's superior sensitivity to the change in xanthophyll epoxidation state due to the lack of measurements of the xanthophyll epoxidation state and leaf area index.
Magney et al. (2016) used the difference between the
midday PRI and early morning PRI (PRI
This study evaluated the impact of the varying shaded-leaf fractions on
seasonal water stress detection in winter wheat using different formulations
of PRI. Results demonstrated that for the control treatment the mean PRI of
sunlit leaves was slightly higher than those of shaded leaves, but the
difference between PRI of sunlit and shaded leaves increased as water
resources became more limiting. Despite the difference between
PRI_shadow and PRI_leaf, the significance of
the linear relationship between RWC and different formulations of PRI did
not show obvious variations with shadow fractions, except for the 100 %
shaded-leaf condition. Among the studied formulations of PRI, PRI3 based on
reflectance at 512 nm as the reference band provided the most accurate
estimates of RWC with varying shaded-leaf fractions, except for the 100 %
shaded-leaf condition. Furthermore, we applied the linear regression model
derived from the PRI of the generally applicable sunlit
All data included in this study are available upon request by contact with the corresponding author (carol.shishi@gmail.com).
XY participated in the data collection, conducted data preprocessing and statistical analysis, and wrote methods and results sections. SL designed the experiment with XR, participated in the data collection, and wrote introduction, discussion, and conclusion sections. YL, XR, and HS participated in the data collection and provided valuable suggestions to the manuscript.
The authors declare that they have no conflict of interest.
We greatly appreciate comments and suggestions from the anonymous reviewers.
This research has been supported by the National Natural Science Foundation of China (grant no. 41501367), National Key Research and Development Program of China (grant no. 2017YFD0100802), and the Fundamental Research Funds for the Central Universities (grant no. 2662018YJ018).
This paper was edited by Christopher Still and reviewed by two anonymous referees.