A Study on Health Diagnosis of Zelkova serrata (Thunb.) Makino Using Hyperspectral Technique
Article information
Abstract
Background and objective
Health diagnosis is essential for large, old trees that are designated and protected by the Act on the Preservation and Utilization of Natural Heritage and the Forest Protection Act. However, in the field, the overall health of trees is diagnosed by measuring the photosynthetic efficiency (PE) of some leaves. Therefore, this study aimed to examine the possibility of using hyper-spectral technology to replace the PE measurement as an approach for diagnosing the health of Zelkova serrata (Thunb.) Makino, which has a high rate of natural monument designation among South Korea’s natural heritage.
Methods
The PE and hyper-spectral values were acquired from Zelkova leaves, and a Monte Carlo simulation was performed based on the spectral bands used to calculate vegetation indices (VI) in previous studies to select bands that have a high correlation with the PE. The explanatory power of these values on tree health was verified through a regression analysis of the VI and PE calculated from the selected bands, and the results of the study are as follows.
Results
First, a variance analysis of the PE and VI of leaves found that there were differences depending on direction, and that the hyper-spectral values of leaves collected from the west were the most effective in measuring tree health. Second, it was found that it was effective to use 520 and 570 nm in the green region, 684 nm in the red region, 860 and 890 nm in the NIR region, 500 nm in the carotenoid region, and 531 nm in the xanthophyll region to derive the VI that replace PE metrics from the hyper-spectral values of leaves collected from the west. Third, PRI showed a very high effect with an explanatory power of 73.2% for the maximum quantum yield (Fv/Fm) regression equation. In other words, it was found that PRI can be used as an index derived from hyperspectral values to diagnose and record the healthy parts of Zelkova tree.
Conclusion
By expanding the scope of this study to include Pinus densiflora and Ginkgo biloba, it is expected that if the utility and field applicability of this approach for diagnosing the health of large old trees are generally verified, it can be used for regular surveys of such trees designated as natural monuments in the future. Furthermore, it is expected that it can be expanded to diagnose park trees, street trees and the like, in addition to large, old trees.
Introduction
Since large old trees have great historical, scenic, and academic value, maintaining their health is essential to protecting them, but involves significant economic costs and manpower. This includes natural monuments (plants) designated by the Act on Preservation and Utilization of Natural Heritage, veteran trees designated by the Forest Protection Act, and beautiful trees designated by local government ordinances. There are 177 natural monument trees nationwide, with 13,859 veteran trees managed by the Korea Forest Service (as of October 2023).
Various attempts have been made to diagnose the state of health of large old trees with heights and crowns tens of meters high using different scientific and technological approaches. The health of such trees is usually diagnosed through the trunks and leaves. Physical decay is measured using increment borers or ultrasonic tomography equipment for trunks, and photosynthetic efficiency (PE) for leaves (Jeon and Kim, 2017 ). PE is the most accurate and efficient method known to date, but has limitations when it comes to diagnosing the overall condition of a tree by closely imaging a single spot on a collected leaf, so it is used in limited ways for crops (Kang, 2021).
In contrast, hyperspectral techniques (or hyperspectral imaging) are increasingly being used as a non-contact method of measuring the delicate growth status of plants by splitting light into more than 150 bands. Notably, technologies that produce hyperspectral data in the form of images have great potential as a way to diagnose the overall health of trees by photographing a crown of many leaves, at once. The diagnosis of plant growth status using hyperspectral data employs vegetation indices (VI), which is an arithmetic combination of specific bands. The efficacy of VI has been suggested in many previous studies (Kim and Chung, 2021; Bae et al., 2018).
Park et al. (2020) estimated VI suitable for monitoring crop growth status from hyperspectral imagery, including Normalized Difference Red Edge (NDRE), Green Normalized Difference Vegetation Index (GNDVI), and Green Chlorophyll Index (GCI). Kang et al. (2022) reported that NDVI, GNDVI, GCI, and Photochemical Reflectance Index (PRI) can be used to diagnose plant health. Kwan and Cho (2015) found that the moisture content of leaves can be measured based on the reflectance in the shortwave infrared range, 1426–1583 nm, using hyperspectral imaging. Na et al. (2019) and Kang et al. (2015) also found that plant health and moisture content can be estimated using VI, such as PRI and NDVI, calculated from hyperspectral imagery. Lee et al. (2013) investigated the possibility of early detection of leaf wilting by analyzing the changes in NDVI, PRI, and Anthocyanin Reflectance Index 2 (ARI2) estimated from hyperspectral images.
As such, plant spectral data have bands that are sensitive to chlorophyll, moisture, and the like depending on the tree species, so VI derived from hyperspectral data can be used to determine plant health. However, as many previous studies used previously proposed bands when estimating VI, or did not disclose the band selection process even when new bands were used, research on VI that are appropriate for tree species attributes remains inadequate.
This study aimed to examine the potential of using hyperspectral techniques as an alternative to the conventional method of measuring PE when diagnosing the health status of Zelkova serrata (Thunb.) Makino (hereinafter referred to as “Zelkova”). This included presenting an appropriate measurement direction according to the light requirement when acquiring hyperspectral data of Zelkova, selecting VI that effectively explain the PE of trees, and deriving the bands used to estimate the VI.
Research Methods
Zelkova serrata (Thunb.) Makino is the third most common large old tree designated as a natural monument by the Korea Heritage Service, after Pinus densiflora and Ginkgo biloba, with the highest proportion of 7,278 trees among the 154 species and 13,859 trees designated as veteran trees by the Korea Forest Service. Accordingly, this study was conducted on a Zelkova tree more than 20 years old located at 367 Baekjemun-ro, Gyuam-myeon, Buyeo, Chungcheongnam-do, in the central region (Fig. 1).
Location (left) and research object (right).
The study was performed through the following steps: measurement of PE and hyperspectral values, calculation of VI, one-way ANOVA on samples by direction, correlation analysis between PE and VI, Monte Carlo simulation of bands used to calculate VI, selection of VI and spectral bands suitable for explaining PE, and selection of VI with explanatory power for PE (Fig. 2).
Research flow.
To measure PE and hyperspectral values, 24 leaves were collected from each side of the Zelkova tree based on the four cardinal directions, from which a total of 85 valid leaves were used and analyzed: 21, 16, 24, and 24 from the east, west, south, and north, respectively (Fig. 3). The leaves were collected and measured between 10:00 am and 2:00 pm in July, when tree growth is most active and light is sufficient. Using the Walz MINI-PAM-II Photosynthesis Yield Analyzer, PE was measured when photochemically used energy was 0 after a dark-adapted clip was attached to the leaves and the light source was blocked for 20 minutes. The steady-state fluorescence (Ft) for measuring the minimum fluorescence (Fo), maximum fluorescence (Fm), and maximum quantum yield (Fv/Fm) were set in the range of 300 to 700. PE metrics were analyzed from the chlorophyll fluorescence induction curve (IC) generated after the measurement, including Fv/Fm, effective quantum yield of photochemical reactions (Y(II)), non-photochemical quenching (NPQ), fraction of non-photochemical energy dissipated in the form of heat and fluorescence (Y(NO)), and fraction of energy dissipated through regulated non-photochemical quenching (Y(NPQ)).
Data acquisition.
Hyperspectral data were acquired using Spectral Evolution’s PSR-1100F, which can measure wavelengths from 320 to 1100 nm spectral range. To reduce scatter or radiation from sunlight, the leaves were placed on a black cloth background before a reading was taken. The white balance (WB) used for light compensation was taken at a distance of 2–3 cm from the hyperspectral sensor. To minimize the variance caused by various environmental factors, including light intensity, temperature, and humidity, a light compensation procedure was performed every 10–15 minutes, and the mean of 15 repeated measurements was used for each leaf (Park et al., 2011).
A total of eight VI were compared and analyzed, including NDVI, PRI, ARI1, ARI2, CRI1 and CRI2, which are used to analyze plant vitality, and GNDVI and GCI, which are related to PE (Table 1). Notably, GNDVI, a vegetation index that uses the near-infrared (NIR) and green band (GREEN), is better than NDVI at determining the difference in the ratio of chlorophyll related to nitrogen in trees (Jung and Kim, 2020). GCI was added as it is used to identify plant physiological health, such as stress, by measuring the chlorophyll content of leaves (Park et al., 2022). The bands used to calculate the VI were classified into Green, Red, NIR, Carotenoid, and Xanthophyll, and the VI were estimated by selecting spectral regions with unique reflectance characteristics (Lee, 2021; John, 2016; Gitelson et al., 2006; Genty et al., 1989) and dividing the bands into 2-nm intervals.
Vegetation index and formula selected for research
To identify the relationship between PE and VI, statistical analyses were performed using SPSS Statistics, including one-way ANOVA, correlation analysis, and regression analysis. Based on the bands used in previous studies to calculate VI, Monte Carlo simulations were performed to vary the bands within ± 30 nm and select those with the highest correlation to PE. A regression analysis of the VI calculated with the selected bands and PE was performed to determine whether the VI could explain tree health.
Results and Discussion
Direction of leaf collection suitable for health diagnosis
By measuring the PE of the leaves collected in the four cardinal directions, it was found that the Fv/Fm was consistently 0.8 or higher, indicating that they were not exposed to stress and were healthy. The fractions of Y(II), Y(NPQ), and Y(NO) also showed normal ranges in all directions.1), The collected leaves exhibited comparable characteristics in the east and south and in the west and north, which can be attributed to a slight discrepancy in the daily or seasonal variations of sunlight. The leaves in the west and north, which were exposed to relatively less light, exhibited higher values of Fv/Fm than those in the east and south, accompanied by high values of Y(NO), which suggests a considerable loss. The leaves in the north had lower Y(NPQ) and higher Y(NO) values compared to those in the west, resulting in a somewhat lower loss than estimated in the west (Table 2).
Photosynthetic efficiency of leaves
The results of calculating VI by direction showed that all VI except ARI had values within the normal range. The PRI and CRI exhibited comparable characteristics in the leaves in the west and north, as well as those in the east and south (Table 3).
Mean vegetation index of leaves by direction
A one-way ANOVA on the PE metrics by direction revealed significant differences at the 5% level of significance probability (Table 4). NPQ and Y(NPQ), which violated the homogeneity of variance, were subjected to post hoc testing with Dunnett’s T3, while other indices were tested with Scheffe’s method. The test results showed differences in PE: for Y(II), in the east and west and in the west and south; for NPQ, in the east and north and in the south and north; for Y(NO), in the east and west, in the east and north, and in the south and north; for Y(NPQ), in the west and north; and for Fv/Fm, in the east and south. Notably, there were significant differences in the east and north (Table 5).
Results of one-way analysis of variance (ANOVA) on photosynthetic efficiency of leaves
Results of post-hoc test on photosynthetic efficiency of leaves by direction
A one-way ANOVA on the VI by direction also showed significant differences at the 5% level of significance probability. Based on the Scheffe post-hoc test, Dunnett’s T3 test was performed on the PRI where the homogeneity of variance was violated (Table 6). The test results showed differences in VI: for NDVI, in the north and south; for GNDVI, in the east and south; for PRI, in the east and west, north and south; for CRI1 and CRI2, in the east and north, and in the south and north; and for GCI, in the east and south. A significant difference was found in the north, where the frequency of such differences was highest (Table 7).
Results of one-way analysis of variance (ANOVA) on vegetation index of leaves
Results of post-hoc test on vegetation index of leaves by direction
The one-way ANOVA showed differences in PE and VI of leaves depending on the direction. Thus, a correlation analysis between PE and VI was conducted to select the direction suitable for leaf collection. The analysis showed that CRI, which is related to the photoprotection mechanism, had a positive correlation with NPQ but a negative correlation with Y(NO), indicating non-photochemical energy dissipation (Table 8).
Results of correlation analysis between photosynthetic efficiency and vegetation index
A negative correlation between GCI and NPQ was observed, which was attributed to the reduction in chlorophyll content in plants under stress. However, this correlation did not reach a statistically significant level. The analysis revealed 8 items with positive correlations in the east, 12 in the west, 6 in the south, and 3 in the north. The west exhibited the highest number of items with high correlations, as indicated by a correlation coefficient of 0.6 or higher. For leaves from the east, GNDVI and PRI exhibited significant correlations with NPQ, Y(NO), and Y(NPQ). For leaves from the west, NDVI, PRI, CRI1, and CRI2 showed high correlations with the PE metrics, NPQ, Y(NO), and Y(NPQ). Meanwhile, for leaves from the south, there was an overall correlation between Fv/Fm and the VI. In particular, PRI showed a high correlation with Fv/Fm, the maximum quantum yield. Leaves from the north generally did not show a high correlation for all the items, and the only index that showed a significant correlation was PRI. This means that leaves collected from the west are more effective in diagnosing tree health than those collected from other directions. However, even for leaves from the west, GNDVI and GCI were found to have no correlation with PE metrics. Except for leaves from the east, GNDVI and GCI appeared to be unsuitable indices for diagnosing tree health, as they showed no correlation in the remaining three directions. However, NDVI, PRI, CRI1, and CRI2, which are expected to be used to diagnose tree health, can explain NPQ, Y(NO), and Y(NPQ), but have limitations in explaining Y(II) and Fv/Fm, indicating that follow-up research is required.
Vegetation index and spectral band suitable for health diagnosis
Hyperspectral values measured at 1 nm intervals have sensitive differences in the resulting data depending on the band used to calculate VI. When calculating VI from images captured with multispectral equipment, the number of reflectance of the red band representing 600–690 nm is 1; however, in hyperspectral images, it is 90. As such, it is very important to change the bands utilized to calculate VI, and to select VI that demonstrate a high correlation with PE. To identify the bands that are sensitive to PE, the spectral bands to be analyzed were selected based on the leaf components that affect photosynthesis and the wavelengths employed for calculating VI in previous studies.
The green band is closely related to anthocyanin and was set between 520 nm, where the reflectance increases, and 570 nm, where it decreases sharply. The carotenoid band was set from 400 nm, where reflectance increases markedly, to 510 nm, where the difference is no longer evident, based on the spectral value of 450 nm, which represents the maximum absorbance of green plants (Byon et al., 2019). Both CRI and ARI use 700 nm of the red band to eliminate the influence of chlorophyll, but based on the spectral graph, they were set to 660–690 nm of the red band, where chlorophyll shows the maximum absorbance.
The 800 nm of NIR band is used for the ARI2 calculation; however, the NIR band was used since it exhibits the highest reflectance at 860–890 nm. PRI distinguished 531 nm, which corresponds to xanthophyll within the green band (Yoon et al., 2021; Park et al., 2020). The spectral bands ultimately utilized for the analysis were set as follows: 400–510 nm (carotenoid), 531 nm (xanthophyll), 520–570 nm (green), 660–690 nm (red), and 860–890 nm (NIR; Fig. 4).
Bands selection based on spectral characteristics of leaf components for vegetation index.
From the green, red, NIR, carotenoid, and xanthophyll spectral bands, 26, 16, 16, 56, and 1 bands were selected, respectively. A total of 256 pairs of NDVI, 416 pairs of GNDVI, 26 pairs of PRI, 1,326 pairs of CRI1, 816 pairs of CRI2, and 416 pairs of GCI were generated by combining the respective spectral values. A Monte Carlo simulation was performed to select the bands used to calculate the VI that exhibited the highest correlation with PE (Table 9).
Spectral bands used when the correlation is high
NDVI was found to have a correlation coefficient of 0.4 or higher with NPQ and Y(NPQ) among the PE metrics, while GNDVI had a correlation coefficient of 0.5 or higher with Y(NPQ). PRI showed a correlation coefficient of 0.4 or higher with all PE metrics; and the green band was found to be effective at 531 nm or higher. Both CRI1 and CRI2 exhibited a high correlation (0.5 or higher) with all PE metrics except Y(II). The bands that showed the highest correlation in situations meeting these conditions were as follows: 684 and 890 nm for NDVI, 520 and 860 nm for GNDVI, 531 and 562 nm for PRI, 410 and 520 nm for CRI1, 500 and 672 nm for CRI2, and 520 and 860 nm for GCI.
The occurrence frequencies of the bands utilized in the VI that demonstrated a significant correlation with the PE metrics were examined. The highest occurrence frequency was observed for 520 nm in the green spectral region (eight times), followed by 531 nm in the xanthophyll region (five times). An occurrence frequency three times each was observed for 410 nm and 500 nm in the carotenoid region, and for 860 nm and 890 nm in the NIR region. Additionally, 672 nm in the red region and 562 nm in the green region, which is suitable for PRI, occurred two times each (Table 10).
Occurrence frequency of spectral bands used in the derivation of vegetation indices highly correlated with photosynthetic efficiency
A stepwise linear regression analysis was conducted to determine the relationship between the PE and the VI when the selected spectral bands were used; in this analysis, the PE metrics and VI were set as dependent and independent variables, respectively. The analysis found that CRI1, CIR2, and PRI can be used to diagnose tree health in place of Y(II), Y(NPQ), NPQ, and Fv/Fm (Table 11).
Linear regression analysis results of vegetation indices for photosynthetic efficiency
CRI1 exhibited a moderate negative effect on Y(II), as indicated by a standardized coefficient of 0.389. The regression model showed an acceptable fit with F = 5.353 (p < .05), but had a relatively low explanatory power of 15.1%. It seems that CRI1, which measures the decrease in spectral reflectance in the carotenoid band, can explain the variance in Y(II). In contrast, CRI1 was found to be a more effective predictor of Y(NPQ). This is because it demonstrated a high effect of 0.649 and a high explanatory power of 42.1% on the variance in Y(NPQ), which is the fraction of NPQ related to the photoprotective mechanism. As such, CRI1 seems to be an effective index for diagnosing the health of Zelkova.
CRI2 showed a moderate negative effect, as indicated by a standardized coefficient of −0.336 on NPQ which is related to the photoprotection mechanism, but its explanatory power was not high at 11.3%. PRI exhibited a substantial effect on Fv/Fm, the maximum quantum yield, as indicated by a standardized coefficient of 0.856, with a high explanatory power of the regression equation reaching 73.2%. The regression analysis found that CRI1 and CRI2, which showed low effect or explanatory power, had limitations when it came to replacing PE metrics. Conversely, the high effect of PRI, a VI, on Fv/Fm, exhibited a robust correlation with its high explanatory power, indicating its potential as an alternative index for diagnosing leaf health in trees.
Conclusion
The findings of this study, which was conducted on a Zelkova tree deemed appropriate for evaluating the diagnostic efficacy of hyperspectral technology, are as follows. First, it was found that samples collected from the west for diagnosing tree health had a high correlation with PE. Second, spectral bands were suggested that can be used for each VI in analyzing the correlation with PE using hyperspectral images. The following spectral values were found to be effective: 520 and 570 nm for the green region, 684 nm for the red region, 860 and 890 nm for the NIR region, 500 nm for the carotenoid region, and 531 nm for the xanthophyll region. Third, it was found that PRI can be used as a VI that can replace the maximum quantum yield, Fv/Fm, which is an indicator of tree health.
This study is of significance in that it presents a method for measuring and recording the health status of large old trees by suggesting bands for measuring tree health using hyperspectral techniques and VI. However, as the experiment was limited to Zelkova, further experiments and analyses are required in the future. These should be expanded to include Pinus densiflora and Ginkgo biloba, which account for a significant proportion of natural monument trees. Should such supplementary research be conducted, it is anticipated that it will have a beneficial effect on reducing the budget and input currently spent by the Korea Heritage Service on the health diagnosis of large, old trees designated as natural monuments. Furthermore, it can be expanded to tree management in relevant fields, including protected trees, park trees, and street trees, as well as the fostering of high-quality forests.
Notes
The sum of Y(II), Y(NPQ), and Y(NO) should be 1. If Y(NO) is greater than Y(NPQ), it may indicate an issue with plant health. Fv/Fm is a parameter included in Y(II), and a value approaching 0.8 indicates a normal photochemical reaction and no stress to the plant. (Genty et al., 1989)