Combining WV-2 images and tree physiological factors to detect damage stages of Populus gansuensis by Asian longhorned beetle (Anoplophora glabripennis) at the tree level


 
 Anoplophora glabripennis (Motschulsky), commonly known as Asian longhorned beetle (ALB), is a wood-boring insect that can cause lethal infestation to multiple borer leaf trees. In Gansu Province, northwest China, ALB has caused a large number of deaths of a local tree species Populus gansuensis. The damaged area belongs to Gobi desert where every single tree is artificially planted and is extremely difficult to cultivate. Therefore, the monitoring of the ALB infestation at the individual tree level in the landscape is necessary. Moreover, the determination of an abnormal phenotype that can be obtained directly from remote-sensing images to predict the damage degree can greatly reduce the cost of field investigation and management.
 
 
 Multispectral WorldView-2 (WV-2) images and 5 tree physiological factors were collected as experimental materials. One-way ANOVA of the tree’s physiological factors helped in determining the phenotype to predict damage degrees. The original bands of WV-2 and derived vegetation indices were used as reference data to construct the dataset of a prediction model. Variance inflation factor and stepwise regression analyses were used to eliminate collinearity and redundancy. Finally, three machine learning algorithms, i.e., Random Forest (RF), Support Vector Machine (SVM), Classification And Regression Tree (CART), were applied and compared to find the best classifier for predicting the damage stage of individual P. gansuensis.
 
 
 The confusion matrix of RF achieved the highest overall classification accuracy (86.2%) and the highest Kappa index value (0.804), indicating the potential of using WV-2 imaging to accurately detect damage stages of individual trees. In addition, the canopy color was found to be positively correlated with P. gansuensis’ damage stages.
 
 
 A novel method was developed by combining WV-2 and tree physiological index for semi-automatic classification of three damage stages of P. gansuensis infested with ALB. The canopy color was determined as an abnormal phenotype that could be directly assessed using remote-sensing images at the tree level to predict the damage degree. These tools are highly applicable for driving quick and effective measures to reduce damage to pure poplar forests in Gansu Province, China.



Page 3/17
Background Anoplophora glabripennis (ALB; Motschulsky; Coleoptera: Cerambycidae: Lamiinae), which is native to Asia regions, is the most dangerous invasive pest of forests and has demonstrated an accelerating growth trend in recent years (Lingafelter and Hoebeke, 2002). It has also become an extremely invasive and destructive pest in Europe and North America (Favaro et al. 2015). Host plants of ALB are distributed among at least 15 species, mainly including poplar, willow, elm, macro, and birch (Haack et al. 2010; Sjöman et al. 2014). The Three-North Shelter Forest Program in Northwest China aiming to improve forest productivity, enhance soil fertility, and achieve better ecological bene ts in arid and semi-arid areas was initiated decades ago (Han Gao et al. 2020). Poplar and willow are the primary tree species established through this program. Northwest China has a temperate continental climate with water shortage, and only single tree species such as Astragalus glabra is planted most often in the region that is extremely vulnerable to diseases and pests. Invasive pests, including ALB, have nearly destroyed the rst phase of the Three-North Shelterbelt reforestation, resulting in serious consequences to the local ecology.
ALB has caused great damage to the local poplar species Populus gansuensis of Jiuquan city, Gansu Province, China, since 2004. The city is located in the Gobi desert, and each tree in the area has an important ecological and economic value. P. gansuensis, as a primary shelter forest tree species in the Hexi Corridor of China, possesses advantages of high afforestation survival rate, fast growth, and drought and cold tolerance (Qiu M-X et al. 1985). The existing monitoring method is that forestry experts walk into the forest to observe with naked eyes, but because of the tree height, large area and long distance, the monitoring work is di cult with low accuracy and e ciency. Therefore, considering that the damaged area is vast and sparsely populated, it is necessary to use remote sensing technology to monitor individual trees from landscape.
Early detection of infestation by the wood-boring ALB is highly di cult (Dean G. Thompson et al. 2018).
The infested trees gradually demonstrate symptoms of damage from top to bottom. Leaves on the twigs become sparse and eventually wither in the initial stage of infection, followed by the appearance of exit holes and death of the branches due to serious damage. Canopy color is an important indicator of a tree's health and is commonly used to detect single-tree diseases of conifers (Wulder M et al.2006). Damage monitoring of broad-leaved trees often involves assessing the amount of foliage loss (Jennifer Pontius et al. 2008), however it has a weak correlation with the physiological characteristics of trees or cannot be assessed directly from remote-sensing images. In this study, we used a damage monitoring factor having a strong correlation with tree vigor that can be assessed directly from remote-sensing images.
Traditional manual surveys to monitor large areas of a forest are highly time consuming, expensive, labor-intensive, and even impossible. Therefore, the development of a cost-effective, e cient, and rapid warning system is urgently needed to respond to pests and diseases affecting trees in a forest area in a timely manner. The application of large-scale remote-sensing methods to obtain relevant images is essential. Use of remote-sensing technology is recommended to monitor and assess forest health. Previous studies have reported that changes in spectral responses can re ect foliage reduction or greening (Ismail et al. 2008). Vegetation condition can be assessed using both broad and narrow vegetation indices (Franklin et al. 2007).
Compared with traditional satellite imaging, the high 8-band 0.5-m resolution of WorldView-2 (WV-2) imaging can greatly improve the geometric and spectral accuracy of the analysis. For example, the recent spectral bands of coastal blue, yellow, RedEdge, and NIR2 of WV-2 allow for bridging the gaps to detect various vegetation types such as tree species or tree mortality (Marchisio G. 2010;Gwata B. 2012).
Literature on the damage of single wood has reported promising results with the use of WV-2 data (Lars T. Waser et al. 2014). However, none of the studies have used WV-2 data to predict the degree of damage to borer leaf trees in forests under speci c conditions in Northwest China.
Recent studies on using of satellite remote-sensing imaging to assess damage to single trees have mainly focused on coniferous trees, with only limited research on broad-leaved trees. The use of remotesensing imaging to monitor ALB-induced damage to single trees in a large area has not been studied. In this research, we aimed to combine WV-2 images and tree physiological factors to detect the stages of damage to P. gansuensis infested by ALB at the single-tree level and to semi-automatically classify the degree of damage to poplars in the Gansu research area. In our study, we determined a phenotype that can be directly assessed from remote-sensing images for damage judgment of P. gansuensis; evaluated the feasibility of using the original WV-2 image band and the derived vegetation index to detect different stages of damage to P. gansuensis; explored whether explanatory predictors contribute the most to the classi cation accuracy of tree damage period; and developed a detailed high-resolution map of the stages of damage to trees, which provides an important support for the remote sensing monitoring method of Anoplophora glabripennis.

Study area
Sanhe National Forest Farm ( Fig.1), with a total area of 247.6 hectares, located in Qingshui town of Jiuquan city, Gansu Province, China, was selected as the experimental plot. The region was the rst to be invaded by ALB in Jiuquan city and, therefore, it contains abundant samples demonstrating different damage stages. Because it is an intensively managed forest farm, the tree species is relatively single, which is conducive to our research. The local altitude is 1480 m, with the following geographical coordinates: east longitude of 98°20′-99°18′ and north latitude of 39°10′-39°59′. The annual average precipitation is 80 mm, but the annual average evaporation reaches 2000 mm. The annual average temperature is 4°C-6°C, and the monthly average temperature is -15℃-29℃.

Ground truth data collection
Ground truth data were collected in September 2020. The hazardous area of ALB infestation was determined through extensive inspection. After selecting the most suitable area for the study, 229 sample trees with similar breast diameters and different damage degrees were randomly selected. Dead trees were not considered for the following reasons: (1) Dead trees do not contain ALBs and identifying them for a study on infection prevention is meaningless. (2) The canopy of a dead tree is small and typically short, which presents challenges in distinguish them using WV-2 images or dead trees are completely hidden by taller trees, hindering visualization from a higher point. Forest pest management often requires damage quanti cation. To determine the damage stages of P. gansuensis, we clicked photos in four directions and measured the diameter at breast height (DBH), global positioning system (GPS) location, exit hole level, dead twig level, damaged ratio, and leaf area index (LAI) of each inspected tree to determine the damage stages at the individual level. We also obtained orthographic images of each inspected P. gansuensis by using unmanned aerial vehicles (UAVs).

Exit hole level
Exit hole is an important symptom of ALB infestation. Exit holes from larvae are the most obvious and direct evidence of ALB infestation (Y. Fragnière et al. 2017). We examined the main holes of each tree, and telescopically counted the number of exit holes as an estimate of ALB activity. Exit holes are divided into level 1-5 from less to more according to quantity.

Dead twig level
Dead twig level is based on the visual assessment of the overall condition of individual trees. Level values range from 1 to 5, according to a technical scheme of Chinese forest pest survey (Chinese National Forestry and Grassland Administration, 2014). The level of dead branches refers to the ratio of dry branches to the total number of branches, with 1 indicating a healthy branch (no major twig mortality), 2 indicating a slight decline in the tree's health (1%-30% crown damage), 3 indicating a moderate decline in the tree's health (30%-60% crown damage),4 indicating severe decline in the tree's health (>60% crown damage), and 5 indicating a dead twig.

Damage shoot ratio
Shoot damage index is a widely used and an important indicator of the extent of injury to a tree (Qinan Lin et al. 2019). It refers to the ratio of large shoots with symptoms of damage to the total number of large shoots. Shoot damage data are also obtained through visual assessment by forestry experts according to the technical scheme of Chinese forest pest survey (Chinese National Forestry and Grassland Administration, 2014). The index value ranges from 0% (indicating a completely healthy tree) to 100% (indicating a dead tree).
Image acquisition using UAV Images were acquired using Yu Mavic2 professional DJI , China . The orthographic images of single inspected trees were clicked from a height of 50-100 m. The 50-m-high orthographic images were printed to map the inspected individual tree for determining the exact location of each inspected tree. UAV images were acquired to determine the canopy color of individual trees and to serve as a supplement submeter GPS information for analyzing the accurate position of the inspected tree in the forest.

Decline rating summary
To determine the damage stages of P. gansuensis and the reference basis for remote-sensing images, we standardized and integrated the canopy color and aforementioned individual physiological indices according to the experience of practical investigation and previous research on coniferous trees and explored their relevance. Instead of using a numerical value to indicate the damage stage of a broadleaved tree according to previous studies, we used canopy color, which has a strong correlation with a tree's health and can be assessed directly from remote-sensing data, as a criterion.
The following veri cation was performed to determine the canopy color: First, the corresponding canopy color of inspected trees was determined from the UAV image, followed by comparison of the average and standard deviation of the tree's physiological factors corresponding to tree canopy colors. Second, the canopy color was used as an independent variable to perform a one-way ANOVA involving the tree's physiological factors. A p value of <0.05 was considered statistically signi cant. Statistical analyses were performed using R (R Development Core Team, 2018).

Satellite image acquisition
Considering the accuracy requirements for distinguishing single trees and actually available satellite data sources, we obtained the corrected WV-2 commercial 8-band very high resolution (VHR) satellite images on September 27, 2020. The WV-2 images of the study area were ordered as multispectral and panchromatic, which contains 0% cloud cover and spatial resolutions of 2 m (multispectral) and 0.5 m (panchromatic). Main sensor speci cations are described in Table 1.
The satellite scene was coded in units of numbers (DN). Calibration and atmospheric correction models (FLAASH in ENVI 5.3) were applied to the multispectral image to convert the digital (DN) value to the sensor radiation and re ectance values. We used the Gram-Schmidt Pan-sharpening technology to pansharpen multispectral images through the full color band. Finally, a 5-m digital terrain model was used to orthorectify the 0.5-m fully sharpened multispectral image. Tree crown segmentation Methodology schema with all steps applied in classi cation of P. gansuensis damage stages is showed in Fig.3. We used pan-sharpened WV-2 images and compared their accuracy to detect damaged branches of the affected trees at different levels according to the object-based method. We used eCognition Developer 9.0 (Trimble Geospatial, USA) to subdivide the pan-sharpened WV-2 images into image objects through multiresolution segmentation. To keep the canopy consistent with the line segment polygons, we iteratively used multiple subdivision levels of detail to adapt to the shape and tightness parameters. The multiresolution classi cation method was used to categorize the images. The segmentation steps were as follows: initial segmentation of a single tree: scale parameter, 1; shape, 0.7; compactness, 0.9; deshading, 5.6e-34 < intensity < 7.6e-34; forest area classi cation: NDVI > 0.26; single tree segmentation: scale parameter, 4; shape, 0.6; compactness, 0.6.

Reference data modeling and predicting
In order to ensure that the sample size of different degrees of damage is roughly the same, except for the sample trees in the eld survey, we also visually inspected the canopy color according to the UAV image. We manually extracted data on tree crowns to construct a spectral diagnostic model, followed by applying the model to the entire satellite image to distinguish the damage stages of individual trees. A total of 139, 139, and 121 trees were in the green stage, yellow stage, and gray stage, respectively. Further, the re ectance of each canopy of the eight bands was extracted to calculate various vegetation indices (VIs). Finally, the re ectance of 8 bands and 17 types of VIs were extracted to construct the database, as well as the training and validation sets at a 7:3 ratio. The extracted VIs are presented in Table 2. On the basis of WV-2 data, the classi cation model demonstrated the highest classi cation accuracy, and the Kappa index was adopted to generate a prediction map at the individual tree level.

Variable selection and data modeling
In total, data of 399 individual trees at three damage stages were used to construct a classi cation model among WV-2 images. First, to select the best explanatory variable among the 8 original bands and 18 VIs, a variance in ation factor (VIF) analysis to check for multicollinearity was performed. A VIF value of ≥10 indicated the need to eliminate serious collinearity. Second, a stepwise regression analysis was performed. The analysis was based on the Akaike information criterion (AIC) information statistics as the criterion by selecting the smallest AIC information statistics to achieve a purpose of deleting or adding variables. Lastly, we applied three machine learning algorithms to classify P. gansuensis into three damage stages: Random Forest RF , Support Vector Machine SVM and Classi cation And Regression Tree CART . RF and SVM have been widely used in single wood damage classi cation and have shown good performance, whereas CART can be easily implemented and explained by certain rules (

Results
Damage stage determination Figure 4 shows three different canopy colors of P. gansuensis damaged by ALB (green, yellow and grey). Figure 5 shows difference of four physiological factors on three canopy colors under the 95% con dence interval. Clear thresholds about tree physiological factors between different canopy colors can be observed. Table 3 shows Physiological factors values corresponding to different canopy colors.
The LAI decreased with changes in the canopy color from green to yellow to gray, whereas the number of exit holes, damaged branches, and dead branches increased with changes in the canopy color. The four tree physiological factors corresponding to different canopy colors demonstrated a considerably signi cant difference. The gradual damage progression of single P. gansuensis trees was re ected by a change in the canopy color from green to yellow to gray. As such, the color change in the remote-sensing images is an important indicator for the assessment of the degree of damage to individual trees. The canopy color is an excellent parameter to judge the degree of damage to individual poplars with advantages of a strong correlation with tree physiology and acquisition from remote-sensing images.

Classi cation results
For 399 segmented trees, there were 27 explanatory predictors, which were further categorized into 14 predictors based on VIF and stepwise regression analyses (Fig. 6). The User's accuracies (UA), Overall accuracies (OA) and Kappa (K) statistics of each classi er are presented in Table 4. The RF analysis of explanatory variables helped in obtaining the optimum OA and Kappa values. The mean decrease in the RF accuracy (Fig. 6) revealed that blue ratio (BR), plant senescence re ectance index (PSRI), NIR-redyellow ratio (NIRRY), and NIR2 greatly affect the classi cation precision. The NIR2 bands were unique to eight-band satellite imaging, indicating the importance of using WV-2. The OA rate was > 80%, which can meet the requirements of actual production. Regarding the classi cation results of the three stages of green, yellow, and gray, the separation accuracy of yellow and green stages was higher. Considering RF as an example, the classi cation accuracy values of green and yellow stages reached 0.897 and 0.911, respectively, whereas that of the gray stage was only 0.750.  Table.4 Accuracy for different stages of damaged Populus gansuensis. The best models are marked in bold. Predictive mapping A predictive map of the damage stages based on the RF classi cation model was generated, and the highest OA area was selected as a display (Fig. 7). The RF classi er using all spectral variables was used to generate a predictive map. In this study, we selected a moderately infested forest for RF mapping. , as these factors cannot be directly assessed from remote-sensing images. Therefore, we aimed to use high-resolution UAV images to assess an abnormal phenotype at the tree level (e.g., canopy color) for directly determining the degree of damage to a single tree.

Classi cation approach
Canopy color has not been used to determine the damage level in studies on disease monitoring of broad-leaved trees. In our study, we obtained relevant data of 229 damaged P. gansuensis trees and determined tree physiological factors, including LAI, damaged shoot ratio, and exit holes. The analysis results of these factors were highly signi cant, suggesting that these parameters can strongly indicate the damage stages of individual trees. If the degree of damage can be predicted by an abnormal growth phenotype, infestations can be managed based on remote-sensing images in the growth cycle of the year.
This approach greatly reduces time and expenses related to growth management and eld investigation. Thus, our results indicate the importance of determining the canopy color to judge the extent of damage from infestation. Finally, the time interval between ground data collection and satellite image collection must be minimal for achieving consistent and accurate results. In our study, the ground truth data collection and WV-2 imaging were performed in September, 2020, for ensuring data consistency.
Therefore, our results accurately re ected the relationship between canopy color and damage degrees.

WV-2 data and single wood segmentation
In this study, the applicability of WV-2 data was thoroughly tested, through which a high OA was achieved. WV-2 is suitable for classifying different damage stages of trees, and it was found to have high accuracy and Kappa value in our research.
Masking of the shadowed tree elements to minimize the high in-class variance of tree crowns has been recommended by many researchers (Pu, R et al. 2012). Thus, focusing mainly on sunlit tree crowns in the present study greatly improved the accuracy of damage levels and tree species identi cation. In addition, because the performance of object-based classi cation largely depends on the quality of image segments, segmentation was optimized iteratively by approximating them to the shape of the tree crown based on assessors' experience.
The dense canopy of P. gansuensis poses a challenge to segment single trees by using 0.5-m resolution WV-2 imaging. In this study, object-based supervised classi cation methods were used to detect different damage stages. Canopy was automatically extracted by eCognition Developer (Trimble Geospatial, USA). Considering the error of automatic segmentation by the software, the WV-2-sharpened multispectral canopy image automatically drawn from a 0.5-m resolution is not always consistent with the actual object. We, therefore, chose to use the eCognition's multiresolutional segmentation method for single wood segmentation (Linfeng yu et al. 2020). Based on UAV images, we adjusted the parameters several times to obtain an accurate individual tree segmentation map. Then, we used the previously established inversion model to obtain a prediction map of the damage level on a single wood scale. Although WV-2 images have a high spatial resolution, the canopy drill used to segment the object will inevitably be combined with other pixels, including bare soil, fallen leaves, and other tree canopies. This will affect the classi cation accuracy of the trees based on different damage levels. Future research must use more advanced techniques to extract data of tree crowns to improve the classi cation accuracy. The effectiveness of the object-based classi cation depends on the segmentation quality. The automatic canopy demarcation that has been successfully applied in many studies should also be further studied.
To distinguish different stages of damage, other sensors, such as Lidar or hyperspectral data, must be tested.

Feature variables and classi cation accuracy
To the best of our knowledge, remote-sensing technology has not been applied to monitor ALB-infested trees in a large area, let alone at a single tree level. Therefore, we expanded the scope and referred to previous studies using satellite imaging to detect damage to broad-leaved trees. In our study, eld survey data and UAV images were used, in addition to categorization of the damage levels of P. gansuensis into green, yellow, and gray canopy stages. The results indicated the overall classi cation accuracy of 82-87%. The RF model using original spectrum and VI was found to be the best, with an OA of 86.2% and a Kappa coe cient value of 0.804. For RF, the classi cation accuracy values of green and yellow stages reached 0.897 and 0.911, respectively, whereas that for the gray stage was only 0.750. This result is also consistent with the actual situation because for the single wood at the gray stage, the top of the canopy is gray and the lower part of the canopy is still green or yellow, which may have a greater impact on the classi cation accuracy. This can also be seen from the UAV images, which are consistent with the actual scenario. Several studies have shown that VIs can better indicate tree pressure than single-band re ectance because they combine information from multiple bands (Immitzer, M.2014). As for original bands, the red edge and NIR band re ectivity decreased with increasing degrees of damage, and the re ectivity difference was most obvious in the NIR band. This research con rmed that the four most important categorical variables for the classi cation of P. gansuensis by RF are BR, PSRI, NIRRY, and NIR2. VIS can signi cantly improve the accuracy of damage degree discrimination. Meanwhile, NIR2 is unique to eight-band WV-2 satellite images, implying the importance of using WV-2 to study degrees of tree damage from pest infestation.

Conclusion
In summary, we developed a novel approach of combining WV-2 imaging data and tree physiological factors for semi-automatic classi cation of three stages of P. gansuensis damage from ALB infestation. The approach was also used to determine the canopy color, an abnormal phenotype, which could be directly assessed on remote-sensing images at the tree level to predict the degree of tree damage. The OA of detecting the damage degrees of P. gansuensis is promising. An accurate and up-to-date information about location and health information at a single wood scale with a high spatial resolution (0.5 m) can be provided, which is highly important for managing tree damage due to ALB infestation in Jiuquan city, Northwest China. Meanwhile, different physiological factors behind the damage stages (green, yellow, and gray) can also be understood, which can reduce the cost of eld data collection and increase the accuracy and applicability of management measures. The generated maps represent the spatial and single tree damage data required to implement prevention and control measures, thereby reducing the large-scale harm from ALB in the shelter forest in Northwest China.

Consent for publication
Not applicable.
Availability of data and materials The data are available upon a reasonable request to the Authors.

Competing interests
The authors declare that they have no competing interests. Funding