Abstract
Chlorophyll fluorescence induction is widely applied to investigate plant growth conditions by calculating the ratio of its intensity at oxidized and reduced states. We examined the applicability of a time-resolved profile of chlorophyll fluorescence induction with the aid of multivariate analysis to monitor the leaf water stress. Principal component (PC) analysis of time-resolved images of chlorophyll fluorescence induction and their score images were reconstructed. Control leaves (non-stressed leaves) and water-stressed leaves could be classified by normalized PC3 score images. This technique has the potential to monitor the water stress condition of plants by using a simple device.
Keywords
INTRODUCTION
Water uptake is one of the most important factors for plant growth. Lack of water or highly osmotic objects within the soil disturb water uptake, thereby inducing a decrease in the moisture content of the leaves. Development of a fast, nondestructive system for monitoring growing conditions is desirable.
Many researchers have reported the nondestructive evaluation of water conditions in leaves by using visible-near infrared (Vis-NIR) reflectance spectra. Vis-NIR spectra at both the leaf and canopy level could be used as an expression of the relative water content (RWC) of plant tissue by calculating the reflectance ratio calculated from the maximum and minimum value between 1500 and 1750 nm. 1 Tian et al. 2 investigated the effectiveness of the ratio of the short wavelength reflectance value at 680 nm and 810 nm as a photosynthesis index. Moreover, RWC could be estimated using the Landsat Thematic Mapper Bands TM4 (750–900 nm) and TM5 (1550–1750 nm). 3 Seelig et al. 4 examined the correlation between RWC and the reflectance ratio of various wavelengths of Vis-NIR energy from leaves. In the NIR region, there are strong absorption bands due to water. For example, the typical absorption band around the wavelength Λ = 1450 nm is assigned to the first overtone of the hydroxyl stretching vibration. They reported that the reflectance at λ = 1450 nm showed the strongest correlation with RWC. We also reported the utility of nondestructive measurement of the water content in plant leaves by using cumulative histograms for luminance brightness at λ = 1450 nm. 5
Chlorophyll fluorescence is also a useful tool for evaluating plant growth conditions under water stress. Chlorophyll fluorescence induction (CFI) is a phenomenon related to changes in fluorescence intensity after a period of exposure to dark conditions. It is a sensitive indicator of photosynthesis, especially related to the electron transport system. Chlorophyll fluorescence intensity is determined by the ratio of the oxidized and reduced states of photosystem II (PSII) reaction centers within leaves. 1 Relationships between some types of stress and chlorophyll fluorescence have been reported, and they indicate that CFI would be a good event for the evaluation of plant stress status. Today, pulse amplitude modulation (PAM) chlorophyll fluorometry is widely used to evaluate the growth status of plants, including carbon dioxide assimilation, 6 water stress detection,7,8 and fungal and virus attack detection.9–12 Using PAM easily enables measurement of the fluorescence quantum yield. A PAM imaging device can obtain the spatial distribution of the quantum yield of fluorescence not only at the leaf level but also at the microscopic level.13,14 Although PAM is very useful to measure the quantum yield, it needs three different light sources (for measuring light, flash, and actinic light). To simplify the measurement procedure, continuous light irradiation also is used.
Some researchers have reported the evaluation of plant growth status by using continuous light irradiation.15–17 Particularly, Omasa et al. 15 examined the effect of sulfur dioxide on the photosynthetic system using fluorescence imaging with continuous light irradiation. These reports mainly mentioned the ratio of characteristic peaks in time profiles of CFI to calculate the quantum yield of fluorescence. In these studies, the ratio of some characteristic peaks is used to evaluate the stress. However, capturing the change of fluorescence intensity during constant time could replace the conventional approach. Capturing the change of fluorescence intensity during constant time under continuous light conditions is easier to measure, and it is easier to construct a measuring system by using a general charge-coupled device (CCD) camera with band-path filter. Moreover, application of multivariate analysis for the time dependency of fluorescence intensity could extract the fine information of stress conditions.
In this study, we examined time-resolved CFI images to monitor water stress conditions in leaves where a three-dimensional array matrix was used. Specifically, the XY surface contained the spatial information, and the Z-axis expressed the time dependency of fluorescence intensity. Hyperspectral imagery is also a three-dimensional array matrix whose XY surface has spatial information, and the Z-axis has spectral information. 18 In the case of hyperspectral imaging, multivariate analysis is often performed in the Z-axis direction to compress and extract useful information. CFI also could be regarded as a kind of time-resolved spectral information. Time-resolved CFI images could be processed like hyperspectral images. We applied principal component analysis 19 (PCA) to time-resolved CFI images. PCA is capable of compressing the possibility-correlated multivariate data into low numbers of uncorrelated data and of magnifying differences among samples. PCA, in terms of the time-resolved profile of CFI, could be potentially used to differentiate water-stressed and water sufficient leaves.
MATERIALS AND METHODS
The aim of this study is to develop a new monitoring technique of water-stressed leaves by using fluorescence images with the aid of PCA. We observed the time dependency of CFI after cutting the leaves as follows.

Outline of experimental device.
Four hydroponically cultivated Epipremnum aureum plants were used in this study. We selected six leaves from two potted E. aureum for fluorescence image acquisition and fixed them on a steel fence. The same experiments were repeated twice. First experiment was labeled as Ex-A and is applied for the calculation of principle component (PC) loading vector. Second experiment was labeled as Ex-B and was used as PC score images.
Figure 2 shows the representative chlorophyll fluorescence image of E. aureum leaves. As shown in Fig. 2, the region of interest (ROI) for six leaves (from a to f) was determined manually. The size of each ROI was 900 pixels, and their average was calculated, i.e., six time dependencies of CFI were extracted from each I. Finally, 90 time dependencies of CFI were acquired from Ex-A. Time dependency was subjected to matrix X, 90 × 53 in size, to obtain PC loading vector.

Representative chlorophyll fluorescence image of Epipremnum aureum leaves acquired at Tir = 3 s and Twst = 1 day of Ex-A. Coordinate values in the figure express the upper left pixel locations of each ROI.
For PCA, explanatory variable X is expressed as the product of the score matrix T and the transposed loading matrix P′.
Note that tA and pA are the Ath score vector and loading vector of the PC, respectively. Matrix P actually contains the eigenvectors of X0X. In this experiment, PCA was conducted without pretreatment.

Outline of image analysis for Ex-A and Ex-B.
where p is loading vector calculated from X of Ex-A.
RESULTS

Averaged time profile of CFI at each ROI for control leaves (a–c) and stressed leaves (d-f). The gray level of each line represents Twst; the darkest line is Twst = 1 day and the lightest line is Twst = 17 days.
There are several differences between control and stressed leaves (Fig. 4). After the maximum florescence intensity at Tir = 3 s, fluorescence intensity moderately decreases until Tir = 7–15 s. As the Twst increases (stressed), the decrement occurs faster, so that the fluorescence intensity of stressed leaves at Tir = 6–8 s is lower than that of control leaves. Another difference could be observed at Tir = 10–20 s. Fluorescence intensity of stressed leaves at Tir = 10–20 s slightly increases, i.e., fluorescence intensity of stressed leaves at Tir = 20 s is higher than that of the control. The Tir at maximum fluorescence intensity (approximately Tir = 3 s) of stressed leaves slightly shifts to the shorter time as Twst increases. The peak shift causes the increase of fluorescence intensity at Tir = 2.5 s.
The time profile of CFI for control leaves a and b did not change greatly during measurement. However for leaf c, fluorescence intensity increased at the beginning of Twst, which seems baseline drift and completely differs from that of stressed leaves. However, such drift phenomena could not be clarified due to unexpected leaf movement, as the angle between the camera and the leaf surface changed.
Figure 5 shows the PC1 to PC4 loading vectors calculated from Ex-A. As described above, the differences in fluorescence intensity between stressed and control leaves appear at Tir = 2.5 s, 6–8 s, and 20 s, respectively. Fluorescence intensity at Tir = 2.5 and 20 s increases for water stress and that at Tir = 6–8 s decreases for water stress. PC3 is better for expressing the differences between stressed and control leaves (Fig. 5); PC3 loading at Tir = 2.5 s and 20 s shows negative values and Tir = 7.5 s shows a positive peak. These peaks could enhance the difference between stressed and control leaves. Every PC scores were compared, and then we used the PC3 to detect water stress.

PC loading vectors of CFI. PC3 loading accurately described the CFI of stressed leaves.

Time dependency of PC3 score of each ROI. Plots show water-supplied leaves (a–c) and stressed leaves (d–f).
Then, PC3 loading vector of calibration was applied to Ex-B. Figure 7 shows score image of Ex-B. To clarify the difference between control and stressed leaves, score images were binarized at the threshold of −4. These images are calculated from matrices I of Ex-B, and the PC3 vector was calculated from Ex-A. At Twst = 12 days and 17 days, stressed leaves could be detected clearly. As shown in Fig. 6, PC3 scores of stressed leaves first increase and then decrease. The water stress condition of leaves could be estimated by tracing the changes of PC3 scores.

PC3 score image for Ex-B.
DISCUSSION
The PC3 score of each of the leaves examined in this study was somewhat different, even at the initiation of the experiment (see Fig. 6). This difference was not induced by water stress but was due to individual variability in CFI affected by various factors, such as another type of environmental stress or the concentration of chlorophyll within the leaf (see Fig. 3). Thus, it was difficult to detect the water stress condition at the initial stage. However, in a longer-term water-stressed condition, PC3 scores of the stressed leaves were higher than that of the control. These results allowed us to conclude that the score image is useful to detect long-term water-stressed leaves. Moreover, the PC3 score of control leaves showed a constant value, whereas the score of stressed leaves showed characteristic variation with the time of exposure to water stress. This result suggests that continuous monitoring of PC3 score of CFI could be applied for the routine monitoring of water-stressed status.
CONCLUSIONS
Time-resolved principal component imaging analysis was performed to evaluate the water stress condition of plant leaves. We examined the time dependency of CFI and applied PCA loading to distinguish stressed leaves from control (non-stressed) leaves. PC3 score images made it possible to reliably compare the differences between control and stressed leaves. PC3 score images could detect long-term water-stressed leaves. In this experiment, the difference between control and stressed leaves was due only to water supply conditions, and any other parameters, such as temperature, humidity, and irradiation, were stable. However, these parameters should effect CFI. When this technique is applied to monitor the water-stress condition of plants grown outside the laboratory, various parameters should be considered. For the application of this technique, additional experiments are required to distinguish the various kinds of stress. This technique could be used for in vivo stress monitoring of plants.
Footnotes
ACKNOWLEDGMENTS
This study was supported by the Japan Society for the Promotion of Sciences (JSPS 7335).
