Get 20M+ Full-Text Papers For Less Than $1.50/day. Subscribe now for You or Your Team.

Learn More →

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

Combining WV-2 images and tree physiological factors to detect damage stages of Populus... Background: 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. Methods: 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. Results: 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. * Correspondence: youqingluo@126.com Beijing Key Laboratory for Forest Pest Control, Beijing Forestry University, Beijing 100083, China Sino-French Joint Laboratory for Invasive Forest Pests in Eurasia, Beijing Forestry University - French National Research Institute for Agriculture, Food and Environment (INRAE), Beijing 100083, China © The Author(s). 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Zhou et al. Forest Ecosystems (2021) 8:35 Page 2 of 12 Conclusions: 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. Keywords: Worldview-2, Anoplophora glabripennis, Populus gansuensis, Infestation, Degree of damage, Canopy color, Classification Background followed by the appearance of exit holes and death of Anoplophora glabripennis (ALB; Motschulsky; Coleop- the branches due to serious damage. Canopy color is an tera: Cerambycidae: Lamiinae), which is native to Asia important indicator of a tree’s health and is commonly regions, is the most dangerous invasive pest of forests used to detect single-tree diseases of conifers (Wulder and has demonstrated an accelerating growth trend in et al. 2006). Damage monitoring of broad-leaved trees recent years. It has also become an extremely invasive often involves assessing the amount of foliage loss (Pon- and destructive pest in Europe and North America tius et al. 2008), however it has a weak correlation with (Favaro et al. 2015). Host plants of ALB are distributed the physiological characteristics of trees or cannot be among at least 15 species, mainly including poplar, wil- assessed directly from remote-sensing images. In this low, elm, macro, and birch (Haack et al. 2010; Sjöman study, we used a damage monitoring factor having a et al. 2014). The Three-North Shelter Forest Program in strong correlation with tree vigor that can be assessed Northwest China aiming to improve forest productivity, directly from remote-sensing images. enhance soil fertility, and achieve better ecological bene- Traditional manual surveys to monitor large areas of a fits in arid and semi-arid areas was initiated decades ago forest are highly time consuming, expensive, labor- (Gao and Huang 2020). Poplar and willow are the pri- intensive, and even impossible. Therefore, the develop- mary tree species established through this program. ment of a cost-effective, efficient, and rapid warning sys- Northwest China has a temperate continental climate tem is urgently needed to respond to pests and diseases with water shortage, and only single tree species such as affecting trees in a forest area in a timely manner. The Astragalus glabra is planted most often in the region application of large-scale remote-sensing methods to ob- that is extremely vulnerable to diseases and pests. Inva- tain relevant images is essential. Use of remote-sensing sive pests, including ALB, have nearly destroyed the first technology is recommended to monitor and assess forest phase of the Three-North Shelterbelt reforestation, health. Previous studies have reported that changes in resulting in serious consequences to the local ecology. spectral responses can reflect foliage reduction or green- ALB has caused great damage to the local poplar spe- ing (Ismail et al. 2008). Vegetation condition can be cies Populus gansuensis of Jiuquan city, Gansu Province, assessed using both broad and narrow vegetation indices China, since 2004. The city is located in the Gobi desert, (Franklin et al. 2007). and each tree in the area has an important ecological Compared with traditional satellite imaging, the high and economic value. P. gansuensis, as a primary shelter 8-band 0.5-m resolution of WorldView-2 (WV-2) im- forest tree species in the Hexi Corridor of China, pos- aging can greatly improve the geometric and spectral ac- sesses advantages of high afforestation survival rate, fast curacy of the analysis. For example, the recent spectral growth, and drought and cold tolerance (Qiu and Liu bands of coastal blue, yellow, RedEdge, and NIR2 of 1985). The existing monitoring method is that forestry WV-2 allow for bridging the gaps to detect various vege- experts walk into the forest to observe with naked eyes, tation types such as tree species or tree mortality but because of the tree height, large area and long dis- (Marchisio et al. 2010; Gwata 2012). For the extraction tance, the monitoring work is difficult with low accuracy of tree crown, compared with traditional satellite like and efficiency. Therefore, considering that the damaged sentinel-2, the resolution of 0.5 m for WV-2 is better area is vast and sparsely populated, it is necessary to use than that of more than 1 m. At the same time, literature remote sensing technology to monitor individual trees on the damage of single wood has reported promising from landscape. results with the use of WV-2 data (Waser et al. 2014). Early detection of infestation by the wood-boring ALB However, none of the studies have used WV-2 data to is highly difficult (Thompson et al. 2018). The infested predict the degree of damage to borer leaf trees in for- trees gradually demonstrate symptoms of damage from ests under specific conditions in Northwest China. top to bottom. Leaves on the twigs become sparse and Recent studies on using of satellite remote-sensing im- eventually wither in the initial stage of infection, aging to assess damage to single trees have mainly Zhou et al. Forest Ecosystems (2021) 8:35 Page 3 of 12 focused on coniferous trees, with only limited research reaches 2000 mm. The annual average temperature is 4– on broad-leaved trees. The use of remote-sensing im- 6 °C. The lowest monthly average temperature is − 15 °C. aging to monitor ALB-induced damage to single trees in a large area has not been studied. In this research, we Ground truth data collection aimed to combine WV-2 images and tree physiological Ground truth data were collected in September 2020. factors to detect the stages of damage to P. gansuensis The hazardous area of ALB infestation was determined infested by ALB at the single-tree level and to semi- through extensive inspection. After selecting the most automatically classify the degree of damage to poplars in suitable area for the study, 229 sample trees with similar the Gansu research area. In our study, we determined a breast diameters and different damage degrees were ran- phenotype that can be directly assessed from remote- domly selected. Dead trees were not considered for the sensing images for damage judgment of P. gansuensis; following reasons: (1) Dead trees do not contain ALBs evaluated the feasibility of using the original WV-2 and identifying them for a study on infection prevention image band and the derived vegetation index to detect is meaningless; (2) The canopy of a dead tree is small different stages of damage to P. gansuensis; explored and typically short, which presents challenges in distin- whether explanatory predictors contribute the most to guish them using WV-2 images or dead trees are com- the classification accuracy of tree damage period; and pletely hidden by taller trees, hindering visualization developed a detailed high-resolution map of the stages from a higher point. Forest pest management often re- of damage to trees, which provides an important support quires damage quantification. To determine the damage for the remote sensing monitoring method of Anoplo- stages of P. gansuensis, we clicked photos in four direc- phora glabripennis. tions 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) Materials and methods of each inspected tree to determine the damage stages at Study area the individual level. We also obtained orthographic im- Sanhe National Forest Farm (Fig. 1), with a total area of ages of each inspected P. gansuensis by using unmanned 247.6 ha, located in Qingshui town of Jiuquan City, aerial vehicles (UAVs). Figure 2 presents examples of Gansu Province, China, was selected as the experimental different ALB-induced damage stages of P. gansuensis. plot. The region was the first to be invaded by ALB in Jiuquan City and, therefore, it contains abundant sam- Leaf area index ples demonstrating different damage stages. Because it is LAI characterizes the leaf density and canopy structure an intensively managed forest farm, the tree species is to reflect the ability of photosynthesis, respiration, tran- relatively single, which is conducive to our research. The spiration, and other biophysical processes of the vegeta- local altitude is 1480 m, with the following geographical tion (Liu et al. 2013). We selected poplars with similar coordinates: east longitude of 98°20′–99°18′ and north stand structure as sample trees for collection to latitude of 39°10′–39°59′. The annual average precipita- minimize the effect of surrounding trees. The average tion is 80 mm, but the annual average evaporation LAI value is the mean of individual tree measurements Fig. 1 (a) Gansu Province in China; (b) study area in Jiuquan City; (c) Sanhe National Forest Farm in Qingshui town Zhou et al. Forest Ecosystems (2021) 8:35 Page 4 of 12 Fig. 2 Examples of different damage degrees of Populus gansuensis caused by ALB. (a) green-stage; (b) yellow-stage; (c) grey-stage; (d) dead tree from four directions by using LAISmart (Orlando et al. major twig mortality), 2 indicating a slight decline in 2015). the tree’s health (1%–30% crown damage), 3 indicat- ing a moderate decline in the tree’s health (30%–60% Exit hole level crown damage), 4 indicating severe decline in the Exit hole is an important symptom of ALB infestation. tree’s health (> 60% crown damage), and 5 indicating Exit holes from larvae are the most obvious and direct a dead twig. evidence of ALB infestation (Fragnière et al. 2018). We examined the main holes of each tree, and telescopically counted the number of exit holes as an estimate of ALB Damage shoot ratio activity. Exit holes are divided into level 1–5 from less to Shoot damage index is a widely used and an important more according to quantity. indicator of the extent of injury to a tree (Lin et al. 2019). It refers to the ratio of large shoots with symp- Dead twig level toms of damage to the total number of large shoots. Dead twig level is based on the visual assessment of Shoot damage data are also obtained through visual as- the overall condition of individual trees. Level values sessment by forestry experts according to the technical range from 1 to 5. The level of dead branches refers scheme of Chinese forest pest survey. The index value to the ratio of dry branches to the total number of ranges from 0% (indicating a completely healthy tree) to branches, with 1 indicating a healthy branch (no 100% (indicating a dead tree). Zhou et al. Forest Ecosystems (2021) 8:35 Page 5 of 12 Image acquisition using UAV Table 1 Technical specifications of the WV-2 imagery Images were acquired using Yu Mavic2 professional Senor characteristic (DJI, China). The orthographic images of single Spatial Resolution 0.5 m PAN and 2 m MS inspected trees were clicked from a height of 50–100 m. Spectral Resolution (nm) Coastal: 400–450 The 50-m-high orthographic images were printed to Blue: 450–510 map the inspected individual tree for determining the Green: 510–580 exact location of each inspected tree. UAV images were Yellow: 585–625 acquired to determine the canopy color of individual trees and to serve as a supplement sub-meter GPS infor- Red: 630–690 mation for analyzing the accurate position of the Red Edge: 705–745 inspected tree in the forest. NIR1: 770–895 NIR2: 860–1040 Decline rating summary Data acquisition Jiuquan City, 27 Sep 2020 To determine the damage stages of P. gansuensis and the reference basis for remote-sensing images, we stan- a 5-m digital terrain model was used to orthorectify the dardized and integrated the canopy color and aforemen- 0.5-m fully sharpened multispectral image. tioned individual physiological indices according to the experience of practical investigation and previous re- Tree crown segmentation search on coniferous trees and explored their relevance. Methodology schema with all steps applied in classifica- Instead of using a numerical value to indicate the dam- tion of P. gansuensis damage stages is showed in Fig. 3. age stage of a broad-leaved tree according to previous We used pan-sharpened WV-2 images and compared studies, we used canopy color, which has a strong correl- their accuracy to detect damaged branches of the af- ation with a tree’s health and can be assessed directly fected trees at different levels according to the object- from remote-sensing data, as a criterion. based method. We used eCognition Developer 9.0 The following verification was performed to determine (Trimble Geospatial, USA) to subdivide the pan- the canopy color: first, the corresponding canopy color sharpened WV-2 images into image objects through of inspected trees was determined from the UAV image, multiresolution segmentation. To keep the canopy con- followed by comparison of the average and standard de- sistent with the line segment polygons, we iteratively viation of the tree’s physiological factors corresponding used multiple subdivision levels of detail to adapt to the to tree canopy colors; second, the canopy color was used shape and tightness parameters. The multiresolution as a dependent variable to perform a one-way ANOVA classification method was used to categorize the images. (González-Rodríguez et al. 2012) involving the tree’s The segmentation steps were as follows: initial segmen- physiological factors. A p-value of < 0.05 was considered tation of a single tree: scale parameter, 1; shape, 0.7; statistically significant. Statistical analyses were per- compactness, 0.9; de-shading, 5.6e–34 < intensity < 7.6e– formed using R (R Development Core Team 2018). 34; forest area classification: NDVI > 0.26; single tree segmentation: scale parameter, 4; shape, 0.6; compact- Satellite image acquisition ness, 0.6. Considering the accuracy requirements for distinguish- ing single trees and actually available satellite data Reference data modeling and predicting sources, we obtained the corrected WV-2 commercial 8- In order to ensure that the sample size of different de- band very high resolution (VHR) satellite images on grees of damage is roughly the same, except for the sam- September 27, 2020. The WV-2 images of the study area ple trees in the field survey, we also visually inspected were ordered as multispectral and panchromatic, which the canopy color according to the UAV image. We contains 0% cloud cover and spatial resolutions of 2 m manually extracted data on tree crowns to construct a (multispectral) and 0.5 m (panchromatic). Main sensor spectral diagnostic model, followed by applying the specifications are described in Table 1. model to the entire satellite image to distinguish the The satellite scene was coded in units of numbers damage stages of individual trees. A total of 139, 139, (DN). Calibration and atmospheric correction models and 121 trees were in the green stage, yellow stage, and (FLAASH in ENVI 5.3) were applied to the multispectral gray stage, respectively. Further, the reflectance of each image to convert the digital (DN) value to the sensor ra- canopy of the eight bands was extracted to calculate diation and reflectance values. We used the Gram- various vegetation indices (VIs). Finally, the reflectance Schmidt Pan-sharpening technology to pan-sharpen of 8 bands and 17 types of VIs were extracted to con- multispectral images through the full color band. Finally, struct the database, as well as the training and validation Zhou et al. Forest Ecosystems (2021) 8:35 Page 6 of 12 Fig. 3 Methodology schema with all steps applied in classification of Populus gansuensis damage stages using the WorldView-2 and UAV image sets at a 7:3 ratio. The extracted VIs are presented in gansuensis into three damage stages: Random Forest Table 2. On the basis of WV-2 data, the classification (RF), Support Vector Machine (SVM) and Classification model demonstrated the highest classification accuracy, and Regression Tree (CART). RF and SVM have been and the Kappa index was adopted to generate a predic- widely used in single wood damage classification and tion map at the individual tree level. have shown good performance, whereas CART can be easily implemented and explained by certain rules (Jing Variable selection and data modeling et al. 2015; Kaszta et al. 2016). In total, data of 399 modeling sample trees at three dam- RF is an improved algorithm compared to traditional age stages were used to construct a classification model decision trees that generates numerous decision trees. among WV-2 images. First, to select the best explana- Among the classification results of all the constructed tory variable among the 8 original bands and 18 VIs, a decision trees, new data are classified based on the ma- variance inflation factor (VIF) analysis to check for mul- jority of votes (Breiman 2001). The SVM algorithm helps ticollinearity was performed. A VIF value of ≥10 indi- find the best hyperplane as the decision function in the cated the need to eliminate serious collinearity. Second, high-dimensional space and classify the input vector into a stepwise regression analysis was performed. The ana- different classes (Cortes and Vapnik 1995). CART is a lysis was based on the Akaike information criterion binary recursive partitioning algorithm based on tree (AIC) information statistics as the criterion by selecting nodes generated by training data. Finally, the overall ac- the smallest AIC information statistics to achieve a pur- curacy (OA), producer’s accuracy, user’s accuracy (UA), pose of deleting or adding variables. Lastly, we applied and the Kappa coefficient generated by the confusion three machine learning algorithms to classify P. matrix were used to evaluate the accuracy of Zhou et al. Forest Ecosystems (2021) 8:35 Page 7 of 12 Table 2 Remote sensing vegetation indices tested in this study and adapted to the WV-2 Abbreviation Name Formula ref NDVI Normalized Difference (NIR1 − R)/(NIR1 + R) Zhang et al. 2006 Vegetation Index NDVI3,5 Green–red ratio (G − R)/(G + R) Gitelson et al. 1996 NDVI8,4 NIR-yellow ratio (NIR2 − Y)/(NIR2 + Y) Gwata 2012 NIRRY NIR-Red-yellow ratio (NIR1)/(R + Y) Gwata 2012 DD Difference Vegetation Index (2 × NIR1 − R) − (G − B) Le Maire et al. 2004 NORM NIR Normalized NIR NIR1/(NIR1 + R + G) ENVI 2013 PSRI Plant Senescence Reflectance (R − B)/RE Sims and Gamon 2002 Index RVI Ratio vegetation index NIR/RED Hildebrandt 1996 GR Green–red ratio G/R Waser et al. 2014 BR Blue ratio (R/B) × (G/B) × (RE/B) × (NIR/B) Waser et al. 2014 RR Red ratio (NIR1/R) × (G/R) × (NIR1/RE) Waser et al. 2014 REY RedEdge yellow ratio (RE − Y)/(RE + Y) Gwata 2012 VIRE Vegetation Index based on NIR1/RE Chávez Oyanadel RedEdge and Clevers 2012 RGI Red–green index RED/GREEN Miura et al. 2008 EVI1 Enhanced Vegetation Index 1 2.4 × (NIR1 − RED)/(NIR1 + RED + 1) Bezerra et al. 2020 EVI2 Enhanced Vegetation Index 2 2.4 × (NIR2 − RED)/(NIR2 + RED + 1) Jiang et al. 2008 GI2 Greenness Index 2 (B × (−0.2848) + G × (− 0.2434) + R × (− 0.5436) + NIR1 × 0.7243 + Gwata 2012 NIR2 × 0.0840) × 5 identification of the damaged stage of a single tree. physiological factors values corresponding to different Kappa values of < 0.4, 0.4–0.8, and > 0.80 indicate poor canopy colors. agreement, moderate agreement, and strong agreement, The LAI decreased with changes in the canopy color respectively (Meddens et al. 2011). 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 physio- Results logical factors corresponding to different canopy colors Damage stage determination demonstrated a considerably significant difference. The Figure 4 shows three different canopy colors of P. gan- gradual damage progression of single P. gansuensis trees suensis damaged by ALB (green, yellow and grey). Fig- was reflected by a change in the canopy color from ure 5 shows difference of four physiological factors on green to yellow to gray. As such, the color change in the three canopy colors under the 95% confidence interval. remote-sensing images is an important indicator for the Clear thresholds about tree physiological factors between assessment of the degree of damage to individual trees. different canopy colors can be observed. Table 3 shows The canopy color is an excellent parameter to judge the Fig. 4 Three canopy colors of P. gansuensis damaged by ALB Zhou et al. Forest Ecosystems (2021) 8:35 Page 8 of 12 Fig. 5 Under the 95% confidence interval (a) the average value of LAI corresponding to different damaged stages; (b) the average wormhole grade corresponding to different damaged stages; (c) the average dead branch rate corresponding to different damaged stages; (d) the average corresponding to different damaged stages Percentage of injured branches degree of damage to individual poplars with advantages The user’s accuracies (UA), overall accuracies (OA) and of a strong correlation with tree physiology and acquisi- Kappa (K) statistics of each classifier are presented in tion from remote-sensing images. 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 Classification results blue ratio (BR), plant senescence reflectance index (PSRI), For 399 segmented trees, there were 25 explanatory pre- NIR-red-yellow ratio (NIRRY), and NIR2 greatly affect the dictors, which were further categorized into 14 predictors classification precision. The NIR2 bands were unique to based on VIF and stepwise regression analyses (Fig. 6). eight-band satellite imaging, indicating the importance of Table 3 Physiological factors values corresponding to different canopy colors Canopy color Measured variation DBH (cm) LAI Exit hole level Dead twig level Damaged shoot ratio (%) Green Avg 30.207 2.407 1.849 1.644 31.301 stdev 6.341 0.328 0.328 1.059 20.582 range 23 1.575 3.000 2.000 60.000 Yellow Avg 26.982 2.290 2.167 2.181 42.778 stdev 6.616 0.258 0.435 1.012 19.661 range 22 1.475 2.000 3.000 40.000 Grey Avg 27.174 2.122 3.464 3.565 70.942 stdev 6.813 0.435 0.258 1.036 25.786 range 18 2.225 3.000 3.000 60.000 Zhou et al. Forest Ecosystems (2021) 8:35 Page 9 of 12 Fig. 6 Graphs of variable importance in the model for damage stages of Populus gansuensis caused by ALB. (a) Mean decrease in the accuracy of the model if that variable were to be removed; (b) mean decrease in the Gini coefficient. Variables that result in nodes with higher purity have a higher decrease in Gini coefficient using WV-2. The OA rate was > 80%, which can meet the Discussion requirements of actual production. Regarding the classifi- Canopy color and damage periods of individual trees cation results of the three stages of green, yellow, and gray, The damage periods of conifer trees are recognized by the separation accuracy of yellow and green stages was changes in the canopy color from green to yellow to red higher. Considering RF as an example, the classification (Zhan et al. 2020); however, there is no clear indicator accuracy values of green and yellow stages reached 0.897 for the damage period of broad-leaved trees. Previous and 0.911, respectively, whereas that of the gray stage was studies on the damage level of broad-leaved trees at the only 0.750 (Table 5). single wood scale were mostly based on a specific index, such as leaf loss rate, emergence hole number, or dead branch rate (Waser et al. 2014). Some studies have nor- Predictive mapping malized all of these factors into a single damage index to A predictive map of the damage stages based on the RF indicate the damage level on a single wood scale. The classification model was generated, and the highest OA normalized index was then used to perform modeling area was selected as a display (Fig. 7). The RF classifier and inversion (Pontius et al. 2008; Pontius et al. 2017), using all spectral variables was used to generate a pre- as these factors cannot be directly assessed from dictive map. In this study, we selected a moderately remote-sensing images. Therefore, we aimed to use infested forest for RF mapping. Figure 7a represents 0.5- high-resolution UAV images to assess an abnormal m WV-2 pan-sharpened true color images. The results phenotype at the tree level (e.g., canopy color) for dir- of mapping of different damage periods obtained using ectly determining the degree of damage to a single tree. the RF classifier are presented in Fig. 7b. We did not choose to use one single factors to repre- sent the damage level of individual tree because there are possibly subjective, ocular measurements through Table 4 Accuracy for different stages of damaged Populus gansuensis. The best models are marked in bold Classification approach Damage stage UA OA K Table 5 Confusion matrix for the classification of damaged RF green 0.897 Populus gansuensis yellow 0.911 0.862 0.804 Reference data Classified as grey 0.750 damage stages green yellow grey User’s Accuracy SVM green 0.837 green 35 1 3 0.897 yellow 0.846 0.824 0.756 yellow 1 41 3 0.911 grey 0.784 grey 4 4 24 0.750 CART green 0.818 Prod’s accuracy 0.875 0.891 0.8 yellow 0.75 0.775 0.692 OA 0.862 grey 0.774 K 0.804 Zhou et al. Forest Ecosystems (2021) 8:35 Page 10 of 12 Fig. 7 Detailed example of the RF mapping of Populus gansuensis forest. (a) 0.5-m WV-2 pan-sharpened true color image; (b) three damaged stages mapping based on RF classifier field assessments. We collected them all to minimize recommended by many researchers (Pu and Landry subjective, ocular measurements where possible. At the 2012). Thus, focusing mainly on sunlit tree crowns in same time, in the process of investigation, we found that the present study greatly improved the accuracy of dam- canopy color is a good indicator, which well represents age levels and tree species identification. In addition, be- these indicators and is suitable for remote sensing moni- cause the performance of object-based classification toring. Canopy color has not been used to determine the largely depends on the quality of image segments, seg- damage level in studies on disease monitoring of broad- mentation was optimized iteratively by approximating leaved trees. In our study, we obtained relevant data of them to the shape of the tree crown based on assessors’ 229 damaged P. gansuensis trees and determined tree experience. physiological factors, including LAI, damaged shoot ra- The dense canopy of P. gansuensis poses a challenge tio, and exit holes. The analysis results of these factors to segment single trees by using 0.5-m resolution WV-2 were highly significant, suggesting that these parameters imaging. In this study, object-based supervised classifica- can strongly indicate the damage stages of individual tion methods were used to detect different damage trees. If the degree of damage can be predicted by an ab- stages. Canopy was automatically extracted by eCogni- normal growth phenotype, infestations can be managed tion Developer (Trimble Geospatial, USA). Considering based on remote-sensing images in the growth cycle of the error of automatic segmentation by the software, the the year. This approach greatly reduces time and ex- WV-2-sharpened multispectral canopy image automatic- penses related to growth management and field investi- ally drawn from a 0.5-m resolution is not always consist- gation. Thus, our results indicate the importance of ent with the actual object. We, therefore, chose to use determining the canopy color to judge the extent of the eCognition’s multiresolutional segmentation method damage from infestation. Finally, the time interval be- for single wood segmentation (Yu et al. 2020). Based on tween ground data collection and satellite image collec- UAV images, we adjusted the parameters several times tion must be minimal for achieving consistent and to obtain an accurate individual tree segmentation map. accurate results. In our study, the ground truth data col- Then, we used the previously established inversion lection and WV-2 imaging were performed in Septem- model to obtain a prediction map of the damage level ber, 2020, for ensuring data consistency. Therefore, our on a single wood scale. Although WV-2 images have a results accurately reflected the relationship between can- high spatial resolution, the canopy drill used to segment opy color and damage degrees. the object will inevitably be combined with other pixels, including bare soil, fallen leaves, and other tree canopies. WV-2 data and single wood segmentation This will affect the classification accuracy of the trees In this study, the applicability of WV-2 data was thor- based on different damage levels. Future research must oughly tested, through which a high OA was achieved. use more advanced techniques to extract data of tree WV-2 is suitable for classifying different damage stages crowns to improve the classification accuracy. The ef- of trees, and it was found to have high accuracy and fectiveness of the object-based classification depends on Kappa value in our research. the segmentation quality. The automatic canopy demar- Masking of the shadowed tree elements to minimize cation that has been successfully applied in many studies the high in-class variance of tree crowns has been should also be further studied. To distinguish different Zhou et al. Forest Ecosystems (2021) 8:35 Page 11 of 12 stages of damage, other sensors, such as Lidar or hyper- City, Northwest China. Meanwhile, different physio- spectral data, must be tested. logical factors behind the damage stages (green, yellow, and gray) can also be understood, which can reduce the Feature variables and classification accuracy cost of field data collection and increase the accuracy To the best of our knowledge, remote-sensing technol- and applicability of management measures. The gener- ogy has not been applied to monitor ALB-infested trees ated maps represent the spatial and single tree damage in a large area, let alone at a single tree level. Therefore, data required to implement prevention and control mea- we expanded the scope and referred to previous studies sures, thereby reducing the large-scale harm from ALB using satellite imaging to detect damage to broad-leaved in the shelter forest in Northwest China. trees. In our study, field survey data and UAV images Abbreviations were used, in addition to categorization of the damage ALB: Asian longhorned beetle; RF: Random Forest; SVM: Support Vector levels of P. gansuensis into green, yellow, and gray can- Machine; CART: Classification And Regression Tree; WV-2: Worldview-2; VIs: Vegetation indices opy stages. The results indicated the overall classification accuracy of 82%–87%. The RF model using original Authors’ contributions spectrum and VI was found to be the best, with an OA Quan Zhou, Linfeng Yu, Lili Ren and Youqing Luo designed this research. Quan Zhou carried out the fieldwork, analyzed the results, and wrote the of 86.2% and a Kappa coefficient value of 0.804. For RF, paper. Xudong Zhang and Linfeng Yu facilitated in the fieldwork. The the classification accuracy values of green and yellow author(s) read and approved the final manuscript. stages reached 0.897 and 0.911, respectively, whereas Funding that for the gray stage was only 0.750. This result is also This study was supported by National Key Research & Development Program consistent with the actual situation because for the sin- of China “Research on key technologies for prevention and control of major gle wood at the gray stage, the top of the canopy is gray disasters in plantation” (Grant No. 2018YFD0600200). Beijing’s Science and Technology Planning Project “Key technologies for prevention and control of and the lower part of the canopy is still green or yellow, major pests in Beijing ecological public welfare forests” (Grant Nos. which may have a greater impact on the classification Z191100008519004 and Z201100008020001). accuracy. This can also be seen from the UAV images, Availability of data and materials which are consistent with the actual scenario. Several The data are available upon a reasonable request to the Authors. studies have shown that VIs can better indicate tree pressure than single-band reflectance because they com- Declarations bine information from multiple bands (Immitzer and Ethics approval and consent to participate Atzberger 2014). As for original bands, the red edge and Not applicable. NIR band reflectivity decreased with increasing degrees Consent for publication of damage, and the reflectivity difference was most obvi- Not applicable. ous in the NIR band. This research confirmed that the four most important categorical variables for the classifi- Competing interests The authors declare that they have no competing interests. cation of P. gansuensis by RF are BR, PSRI, NIRRY, and NIR2. VIS can significantly improve the accuracy of Received: 9 February 2021 Accepted: 20 May 2021 damage degree discrimination. Meanwhile, NIR2 is unique to eight-band WV-2 satellite images, implying References the importance of using WV-2 to study degrees of tree Bezerra FGS, Aguiar APD, Alvalab RCS, Giarollaa A, Bezerraa KRA, Lima PVPS, do damage from pest infestation. Nascimento FR, Araie E (2020) Analysis of areas undergoing desertification, using EVI2 multi-temporal data based on MODIS imagery as indicator. Ecol Indic 117:106579. https://doi.org/10.1016/j.ecolind.2020.106579 Conclusion Breiman L (2001) Random forests. Mach Learn 45(1):5–32. https://doi.org/10.1 In summary, we developed a novel approach of combin- 023/A:1010933404324 Chávez Oyanadel RO, Clevers JGPW (2012) Object-based analysis of 8-bands ing WV-2 imaging data and tree physiological factors for worldview-2 imagery for assessing health condition of desert trees. semi-automatic classification of three stages of P. gan- Wageningen UR, Wageningen suensis damage from ALB infestation. The approach was Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273–297. https://doi.org/10.1007/BF00994018 also used to determine the canopy color, an abnormal Development Core Team R (2018) R: a language and environment for statistical phenotype, which could be directly assessed on remote- computing. R foundation for statistical computing, Vienna sensing images at the tree level to predict the degree of ENVI (2013) Vegetation Indices. Available online: http://www.exelisvis.com/docs/ VegetationIndices.html. Accessed 10 Apr 2014 tree damage. The OA of detecting the damage degrees Favaro R, Wichmann L, Ravn HP, Faccoli M (2015) Spatial spread and infestation of P. gansuensis is promising. An accurate and up-to- risk assessment in the Asian longhorned beetle, Anoplophora glabripennis. date information about location and health information Entomol Exp Appl 155(2):95–101. https://doi.org/10.1111/eea.12292 Fragnière Y, Forster B, Hölling D, Wermelinger B, Bacher S (2018) A local risk map at a single wood scale with a high spatial resolution (0.5 using field observations of the Asian longhorned beetle to optimize m) can be provided, which is highly important for man- monitoring activities. J Appl Entomol 142(6):578–588. https://doi.org/10.1111/ aging tree damage due to ALB infestation in Jiuquan jen.12491 Zhou et al. Forest Ecosystems (2021) 8:35 Page 12 of 12 Franklin SE, Fan H, Guo X (2007) Relationship between Landsat TM and SPOT Pontius J, Martin M, Plourde L, Hallett R (2008) Ash decline assessment in vegetation indices and cumulative spruce budworm defoliation. Int J Remote emerald ash borer-infested regions: a test of tree-level, hyperspectral Sens 29:1215–1220 technologies. Remote Sens Environ 112(5):2665–2676. https://doi.org/10.101 Gao H, Huang Y (2020) Impacts of the three-north shelter forest program on the 6/j.rse.2007.12.011 main soil nutrients in northern Shanxi China: a meta-analysis. Forest Ecol Pu R, Landry S (2012) A comparative analysis of high spatial resolution IKONOS Manag 458:117808. https://doi.org/10.1016/j.foreco.2019.117808 and WorldView-2 imagery for mapping urban tree species. Remote Sens Environ 124:516–533. https://doi.org/10.1016/j.rse.2012.06.011 Gitelson AA, Kaufman YJ, Merzlyak MN (1996) Use of a green channel in remote Qiu MX, Liu JQ (1985) Some observation on the biological and ecological sensing of global vegetation from EOS-MODIS. Remote Sens Environ 58(3): characters of Populus gansuensis. Chin J Plan Ecol 9(3):202–211 289–298. https://doi.org/10.1016/S0034-4257(96)00072-7 Sims DA, Gamon JA (2002) Relationships between leaf pigment content and González-Rodríguez G, Colubi A, Gil MÁ (2012) Fuzzy data treated as functional spectral reflectance across a wide range of species, leaf structures and data: a one-way ANOVA test approach. Comput Stat Data An 56(4):943–955. developmental stages. Remote Sens Environ 81(2-3):337–354. https://doi. https://doi.org/10.1016/j.csda.2010.06.013 org/10.1016/S0034-4257(02)00010-X Gwata B (2012) Developing high resolution clutter for wireless network Sjöman H, Ôstberg J, Nilsson J (2014) Review of host trees for the wood-boring propagation using WorldView-2 imagery. In: Proc. SPIE 8390. Algorithms and pests Anoplophora glabripennis and Anoplophora chinensis: an urban forest Technologies for Multispectral, Hyperspectral, and Ultraspectral imagery XVIII, perspective. Arboric Urban For 40(3):143–164 83902Q, Baltimore, Maryland, 24 may 2012. https://doi.org/10.1117/12.919091 Thompson DG, Tonon A, Beltran E, Hernandez F (2018) Inhibition of larval growth Haack RA, Hérard F, Sun J, Turgeon JJ (2010) Managing invasive populations of and adult fecundity in Asian long-horned beetle (Anoplophora glabripennis) Asian longhorned beetle and citrus longhorned beetle: a worldwide exposed to azadirachtins under quarantine laboratory conditions. Pest perspective. Annu Rev Entomol 55(1):521–546. https://doi.org/10.1146/a Manag Sci 74(6):1351–1361. https://doi.org/10.1002/ps.4810 nnurev-ento-112408-085427 Waser LT, Küchler M, Jütte K, Stampfer T (2014) Evaluating the potential of Hildebrandt G (1996) Fernerkundung und Luftbildmessung: für Forstwirtschaft, WorldView-2 data to classify tree species and different levels of ash mortality. Vegetationskartierung, und Landschaftsökologie. Herbert Wichmann, Remote Sens 6(5):4515–4545. https://doi.org/10.3390/rs6054515 Heidelberg Wulder MA, Dymond CC, White JC, Leckie DG, Carroll AL (2006) Surveying Immitzer M, Atzberger C (2014) Early detection of bark beetle infestation in mountain pine beetle damage of forests: a review of remote sensing Norway spruce (Picea abies L.) usingWorldView-2 data. Photogramm opportunities. Forest Ecol Manag 221(1-3):27–41. https://doi.org/10.1016/j. Fernerkun 5:351–367 foreco.2005.09.021 Ismail R, Mutanga O, Ahmed F (2008) Discriminating Sirex noctilio attack in pine Yu LF, Zhan ZY, Ren LL, Zong SX, Luo YQ, Huang HG (2020) Evaluating the forest plantations in South Africa using high spectral resolution data. potential of WorldView-3 data to classify different shoot damage ratios of Hyperspectral Remote Sensing of Tropical and Sub-Tropical Forests. Crc Pinus yunnanensis. Forests 11(4):417. https://doi.org/10.3390/f11040417 Press, London, pp 161–175 Zhan ZY, Yu LF, Li Z, Ren LL, Gao BT, Wang LX, Luo YQ (2020) Combining GF-2 Jiang Z, Huete AR, Didan K, Miura T (2008) Development of a two-band and Sentinel-2 images to detect tree mortality caused by red turpentine enhanced vegetation index without a blue band. Remote Sens Environ beetle during the early outbreak stage in North China. Forests 11(2):172. 112(10):3833–3845. https://doi.org/10.1016/j.rse.2008.06.006 https://doi.org/10.3390/f11020172 Jing W, Yang Y, Yue X, Zhao X (2015) Mapping urban areas with integration of Zhang X, Yan G, Li Q, Li ZL, Wan H, Guo Z (2006) Evaluating the fraction of DMSP/OLS nighttime light and MODIS data using machine learning vegetation cover based on NDVI spatial scale correction model. Int J Remote techniques. Remote Sens 7(9):12419–12439. https://doi.org/10.3390/ Sens 27(24):5359–5372. https://doi.org/10.1080/01431160600658107 rs70912419 Kaszta Ż, Van De Kerchove R, Ramoelo A, Cho MA, Madonsela S, Mathieu R, Wolff E (2016) Seasonal separation of African savanna components using WorldView-2 imagery: a comparison of pixel-and object-based approaches and selected classification algorithms. Remote Sens 8(9):763. https://doi.org/1 0.3390/rs8090763 Le Maire G, Francois C, Dufrene E (2004) Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements. Remote Sens Environ 89(1):1–28. https://doi.org/10.1016/j. rse.2003.09.004 Lin QN, Huang HG, Wang JX, Huang K, Liu YY (2019) Detection of pine shoot beetle (PSB) stress on pine forests at individual tree level using UAV-based hyperspectral imagery and lidar. Remote Sens 11(21):2540. https://doi.org/1 0.3390/rs11212540 Liu Y, Liu R, Chen J, Cheng X, Zheng G (2013) Current status and perspectives of leaf area index retrieval from optical remote sensing data. J Geo-infor Sci 15(5):734–743 (in Chinese) Marchisio G, Pacifici F, Padwick C (2010) On the relative predictive value of the new spectral bands in the WorldWiew-2 sensor. 2010 IEEE International Geoscience and Remote Sensing Symposium, pp 2723–2726 Meddens A, Hicke J, Vierling L (2011) Evaluating the potential of multispectral imagery to map multiple stages of tree mortality. Remote Sens Environ 115(7):1632–1642. https://doi.org/10.1016/j.rse.2011.02.018 Miura T, Yoshioka H, Fujiwara K, Yamamoto H (2008) Inter-comparison of ASTER and MODIS surface reflectance and vegetation index products for synergistic applications to natural resource monitoring. Sensors 8(4):2480–2499. https:// doi.org/10.3390/s8042480 Orlando F, Movedi E, Paleari L, Gilardelli C, Foi M, Dell’Oro M, Confalonieri R (2015) Estimating leaf area index in tree species using the PocketLAI smart app. Appl Veg Sci 18(4):716–723. https://doi.org/10.1111/avsc.12181 Pontius J, Hanavan RP, Hallett RA, Cook BD, Corp LA (2017) High spatial resolution spectral unmixing for mapping ash species across a complex urban environment. Remote Sens Environ 199:360–369. https://doi.org/10.101 6/j.rse.2017.07.027 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png "Forest Ecosystems" Springer Journals

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

Loading next page...
 
/lp/springer-journals/combining-wv-2-images-and-tree-physiological-factors-to-detect-damage-pRja6IFtzJ

References (2)

  • L Breiman (2001)

    5

    Mach Learn, 45

  • FGS Bezerra (2020)

    106579

    Ecol Indic, 117

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2021
eISSN
2197-5620
DOI
10.1186/s40663-021-00314-y
Publisher site
See Article on Publisher Site

Abstract

Background: 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. Methods: 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. Results: 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. * Correspondence: youqingluo@126.com Beijing Key Laboratory for Forest Pest Control, Beijing Forestry University, Beijing 100083, China Sino-French Joint Laboratory for Invasive Forest Pests in Eurasia, Beijing Forestry University - French National Research Institute for Agriculture, Food and Environment (INRAE), Beijing 100083, China © The Author(s). 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. Zhou et al. Forest Ecosystems (2021) 8:35 Page 2 of 12 Conclusions: 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. Keywords: Worldview-2, Anoplophora glabripennis, Populus gansuensis, Infestation, Degree of damage, Canopy color, Classification Background followed by the appearance of exit holes and death of Anoplophora glabripennis (ALB; Motschulsky; Coleop- the branches due to serious damage. Canopy color is an tera: Cerambycidae: Lamiinae), which is native to Asia important indicator of a tree’s health and is commonly regions, is the most dangerous invasive pest of forests used to detect single-tree diseases of conifers (Wulder and has demonstrated an accelerating growth trend in et al. 2006). Damage monitoring of broad-leaved trees recent years. It has also become an extremely invasive often involves assessing the amount of foliage loss (Pon- and destructive pest in Europe and North America tius et al. 2008), however it has a weak correlation with (Favaro et al. 2015). Host plants of ALB are distributed the physiological characteristics of trees or cannot be among at least 15 species, mainly including poplar, wil- assessed directly from remote-sensing images. In this low, elm, macro, and birch (Haack et al. 2010; Sjöman study, we used a damage monitoring factor having a et al. 2014). The Three-North Shelter Forest Program in strong correlation with tree vigor that can be assessed Northwest China aiming to improve forest productivity, directly from remote-sensing images. enhance soil fertility, and achieve better ecological bene- Traditional manual surveys to monitor large areas of a fits in arid and semi-arid areas was initiated decades ago forest are highly time consuming, expensive, labor- (Gao and Huang 2020). Poplar and willow are the pri- intensive, and even impossible. Therefore, the develop- mary tree species established through this program. ment of a cost-effective, efficient, and rapid warning sys- Northwest China has a temperate continental climate tem is urgently needed to respond to pests and diseases with water shortage, and only single tree species such as affecting trees in a forest area in a timely manner. The Astragalus glabra is planted most often in the region application of large-scale remote-sensing methods to ob- that is extremely vulnerable to diseases and pests. Inva- tain relevant images is essential. Use of remote-sensing sive pests, including ALB, have nearly destroyed the first technology is recommended to monitor and assess forest phase of the Three-North Shelterbelt reforestation, health. Previous studies have reported that changes in resulting in serious consequences to the local ecology. spectral responses can reflect foliage reduction or green- ALB has caused great damage to the local poplar spe- ing (Ismail et al. 2008). Vegetation condition can be cies Populus gansuensis of Jiuquan city, Gansu Province, assessed using both broad and narrow vegetation indices China, since 2004. The city is located in the Gobi desert, (Franklin et al. 2007). and each tree in the area has an important ecological Compared with traditional satellite imaging, the high and economic value. P. gansuensis, as a primary shelter 8-band 0.5-m resolution of WorldView-2 (WV-2) im- forest tree species in the Hexi Corridor of China, pos- aging can greatly improve the geometric and spectral ac- sesses advantages of high afforestation survival rate, fast curacy of the analysis. For example, the recent spectral growth, and drought and cold tolerance (Qiu and Liu bands of coastal blue, yellow, RedEdge, and NIR2 of 1985). The existing monitoring method is that forestry WV-2 allow for bridging the gaps to detect various vege- experts walk into the forest to observe with naked eyes, tation types such as tree species or tree mortality but because of the tree height, large area and long dis- (Marchisio et al. 2010; Gwata 2012). For the extraction tance, the monitoring work is difficult with low accuracy of tree crown, compared with traditional satellite like and efficiency. Therefore, considering that the damaged sentinel-2, the resolution of 0.5 m for WV-2 is better area is vast and sparsely populated, it is necessary to use than that of more than 1 m. At the same time, literature remote sensing technology to monitor individual trees on the damage of single wood has reported promising from landscape. results with the use of WV-2 data (Waser et al. 2014). Early detection of infestation by the wood-boring ALB However, none of the studies have used WV-2 data to is highly difficult (Thompson et al. 2018). The infested predict the degree of damage to borer leaf trees in for- trees gradually demonstrate symptoms of damage from ests under specific conditions in Northwest China. top to bottom. Leaves on the twigs become sparse and Recent studies on using of satellite remote-sensing im- eventually wither in the initial stage of infection, aging to assess damage to single trees have mainly Zhou et al. Forest Ecosystems (2021) 8:35 Page 3 of 12 focused on coniferous trees, with only limited research reaches 2000 mm. The annual average temperature is 4– on broad-leaved trees. The use of remote-sensing im- 6 °C. The lowest monthly average temperature is − 15 °C. aging to monitor ALB-induced damage to single trees in a large area has not been studied. In this research, we Ground truth data collection aimed to combine WV-2 images and tree physiological Ground truth data were collected in September 2020. factors to detect the stages of damage to P. gansuensis The hazardous area of ALB infestation was determined infested by ALB at the single-tree level and to semi- through extensive inspection. After selecting the most automatically classify the degree of damage to poplars in suitable area for the study, 229 sample trees with similar the Gansu research area. In our study, we determined a breast diameters and different damage degrees were ran- phenotype that can be directly assessed from remote- domly selected. Dead trees were not considered for the sensing images for damage judgment of P. gansuensis; following reasons: (1) Dead trees do not contain ALBs evaluated the feasibility of using the original WV-2 and identifying them for a study on infection prevention image band and the derived vegetation index to detect is meaningless; (2) The canopy of a dead tree is small different stages of damage to P. gansuensis; explored and typically short, which presents challenges in distin- whether explanatory predictors contribute the most to guish them using WV-2 images or dead trees are com- the classification accuracy of tree damage period; and pletely hidden by taller trees, hindering visualization developed a detailed high-resolution map of the stages from a higher point. Forest pest management often re- of damage to trees, which provides an important support quires damage quantification. To determine the damage for the remote sensing monitoring method of Anoplo- stages of P. gansuensis, we clicked photos in four direc- phora glabripennis. tions 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) Materials and methods of each inspected tree to determine the damage stages at Study area the individual level. We also obtained orthographic im- Sanhe National Forest Farm (Fig. 1), with a total area of ages of each inspected P. gansuensis by using unmanned 247.6 ha, located in Qingshui town of Jiuquan City, aerial vehicles (UAVs). Figure 2 presents examples of Gansu Province, China, was selected as the experimental different ALB-induced damage stages of P. gansuensis. plot. The region was the first to be invaded by ALB in Jiuquan City and, therefore, it contains abundant sam- Leaf area index ples demonstrating different damage stages. Because it is LAI characterizes the leaf density and canopy structure an intensively managed forest farm, the tree species is to reflect the ability of photosynthesis, respiration, tran- relatively single, which is conducive to our research. The spiration, and other biophysical processes of the vegeta- local altitude is 1480 m, with the following geographical tion (Liu et al. 2013). We selected poplars with similar coordinates: east longitude of 98°20′–99°18′ and north stand structure as sample trees for collection to latitude of 39°10′–39°59′. The annual average precipita- minimize the effect of surrounding trees. The average tion is 80 mm, but the annual average evaporation LAI value is the mean of individual tree measurements Fig. 1 (a) Gansu Province in China; (b) study area in Jiuquan City; (c) Sanhe National Forest Farm in Qingshui town Zhou et al. Forest Ecosystems (2021) 8:35 Page 4 of 12 Fig. 2 Examples of different damage degrees of Populus gansuensis caused by ALB. (a) green-stage; (b) yellow-stage; (c) grey-stage; (d) dead tree from four directions by using LAISmart (Orlando et al. major twig mortality), 2 indicating a slight decline in 2015). the tree’s health (1%–30% crown damage), 3 indicat- ing a moderate decline in the tree’s health (30%–60% Exit hole level crown damage), 4 indicating severe decline in the Exit hole is an important symptom of ALB infestation. tree’s health (> 60% crown damage), and 5 indicating Exit holes from larvae are the most obvious and direct a dead twig. evidence of ALB infestation (Fragnière et al. 2018). We examined the main holes of each tree, and telescopically counted the number of exit holes as an estimate of ALB Damage shoot ratio activity. Exit holes are divided into level 1–5 from less to Shoot damage index is a widely used and an important more according to quantity. indicator of the extent of injury to a tree (Lin et al. 2019). It refers to the ratio of large shoots with symp- Dead twig level toms of damage to the total number of large shoots. Dead twig level is based on the visual assessment of Shoot damage data are also obtained through visual as- the overall condition of individual trees. Level values sessment by forestry experts according to the technical range from 1 to 5. The level of dead branches refers scheme of Chinese forest pest survey. The index value to the ratio of dry branches to the total number of ranges from 0% (indicating a completely healthy tree) to branches, with 1 indicating a healthy branch (no 100% (indicating a dead tree). Zhou et al. Forest Ecosystems (2021) 8:35 Page 5 of 12 Image acquisition using UAV Table 1 Technical specifications of the WV-2 imagery Images were acquired using Yu Mavic2 professional Senor characteristic (DJI, China). The orthographic images of single Spatial Resolution 0.5 m PAN and 2 m MS inspected trees were clicked from a height of 50–100 m. Spectral Resolution (nm) Coastal: 400–450 The 50-m-high orthographic images were printed to Blue: 450–510 map the inspected individual tree for determining the Green: 510–580 exact location of each inspected tree. UAV images were Yellow: 585–625 acquired to determine the canopy color of individual trees and to serve as a supplement sub-meter GPS infor- Red: 630–690 mation for analyzing the accurate position of the Red Edge: 705–745 inspected tree in the forest. NIR1: 770–895 NIR2: 860–1040 Decline rating summary Data acquisition Jiuquan City, 27 Sep 2020 To determine the damage stages of P. gansuensis and the reference basis for remote-sensing images, we stan- a 5-m digital terrain model was used to orthorectify the dardized and integrated the canopy color and aforemen- 0.5-m fully sharpened multispectral image. tioned individual physiological indices according to the experience of practical investigation and previous re- Tree crown segmentation search on coniferous trees and explored their relevance. Methodology schema with all steps applied in classifica- Instead of using a numerical value to indicate the dam- tion of P. gansuensis damage stages is showed in Fig. 3. age stage of a broad-leaved tree according to previous We used pan-sharpened WV-2 images and compared studies, we used canopy color, which has a strong correl- their accuracy to detect damaged branches of the af- ation with a tree’s health and can be assessed directly fected trees at different levels according to the object- from remote-sensing data, as a criterion. based method. We used eCognition Developer 9.0 The following verification was performed to determine (Trimble Geospatial, USA) to subdivide the pan- the canopy color: first, the corresponding canopy color sharpened WV-2 images into image objects through of inspected trees was determined from the UAV image, multiresolution segmentation. To keep the canopy con- followed by comparison of the average and standard de- sistent with the line segment polygons, we iteratively viation of the tree’s physiological factors corresponding used multiple subdivision levels of detail to adapt to the to tree canopy colors; second, the canopy color was used shape and tightness parameters. The multiresolution as a dependent variable to perform a one-way ANOVA classification method was used to categorize the images. (González-Rodríguez et al. 2012) involving the tree’s The segmentation steps were as follows: initial segmen- physiological factors. A p-value of < 0.05 was considered tation of a single tree: scale parameter, 1; shape, 0.7; statistically significant. Statistical analyses were per- compactness, 0.9; de-shading, 5.6e–34 < intensity < 7.6e– formed using R (R Development Core Team 2018). 34; forest area classification: NDVI > 0.26; single tree segmentation: scale parameter, 4; shape, 0.6; compact- Satellite image acquisition ness, 0.6. Considering the accuracy requirements for distinguish- ing single trees and actually available satellite data Reference data modeling and predicting sources, we obtained the corrected WV-2 commercial 8- In order to ensure that the sample size of different de- band very high resolution (VHR) satellite images on grees of damage is roughly the same, except for the sam- September 27, 2020. The WV-2 images of the study area ple trees in the field survey, we also visually inspected were ordered as multispectral and panchromatic, which the canopy color according to the UAV image. We contains 0% cloud cover and spatial resolutions of 2 m manually extracted data on tree crowns to construct a (multispectral) and 0.5 m (panchromatic). Main sensor spectral diagnostic model, followed by applying the specifications are described in Table 1. model to the entire satellite image to distinguish the The satellite scene was coded in units of numbers damage stages of individual trees. A total of 139, 139, (DN). Calibration and atmospheric correction models and 121 trees were in the green stage, yellow stage, and (FLAASH in ENVI 5.3) were applied to the multispectral gray stage, respectively. Further, the reflectance of each image to convert the digital (DN) value to the sensor ra- canopy of the eight bands was extracted to calculate diation and reflectance values. We used the Gram- various vegetation indices (VIs). Finally, the reflectance Schmidt Pan-sharpening technology to pan-sharpen of 8 bands and 17 types of VIs were extracted to con- multispectral images through the full color band. Finally, struct the database, as well as the training and validation Zhou et al. Forest Ecosystems (2021) 8:35 Page 6 of 12 Fig. 3 Methodology schema with all steps applied in classification of Populus gansuensis damage stages using the WorldView-2 and UAV image sets at a 7:3 ratio. The extracted VIs are presented in gansuensis into three damage stages: Random Forest Table 2. On the basis of WV-2 data, the classification (RF), Support Vector Machine (SVM) and Classification model demonstrated the highest classification accuracy, and Regression Tree (CART). RF and SVM have been and the Kappa index was adopted to generate a predic- widely used in single wood damage classification and tion map at the individual tree level. have shown good performance, whereas CART can be easily implemented and explained by certain rules (Jing Variable selection and data modeling et al. 2015; Kaszta et al. 2016). In total, data of 399 modeling sample trees at three dam- RF is an improved algorithm compared to traditional age stages were used to construct a classification model decision trees that generates numerous decision trees. among WV-2 images. First, to select the best explana- Among the classification results of all the constructed tory variable among the 8 original bands and 18 VIs, a decision trees, new data are classified based on the ma- variance inflation factor (VIF) analysis to check for mul- jority of votes (Breiman 2001). The SVM algorithm helps ticollinearity was performed. A VIF value of ≥10 indi- find the best hyperplane as the decision function in the cated the need to eliminate serious collinearity. Second, high-dimensional space and classify the input vector into a stepwise regression analysis was performed. The ana- different classes (Cortes and Vapnik 1995). CART is a lysis was based on the Akaike information criterion binary recursive partitioning algorithm based on tree (AIC) information statistics as the criterion by selecting nodes generated by training data. Finally, the overall ac- the smallest AIC information statistics to achieve a pur- curacy (OA), producer’s accuracy, user’s accuracy (UA), pose of deleting or adding variables. Lastly, we applied and the Kappa coefficient generated by the confusion three machine learning algorithms to classify P. matrix were used to evaluate the accuracy of Zhou et al. Forest Ecosystems (2021) 8:35 Page 7 of 12 Table 2 Remote sensing vegetation indices tested in this study and adapted to the WV-2 Abbreviation Name Formula ref NDVI Normalized Difference (NIR1 − R)/(NIR1 + R) Zhang et al. 2006 Vegetation Index NDVI3,5 Green–red ratio (G − R)/(G + R) Gitelson et al. 1996 NDVI8,4 NIR-yellow ratio (NIR2 − Y)/(NIR2 + Y) Gwata 2012 NIRRY NIR-Red-yellow ratio (NIR1)/(R + Y) Gwata 2012 DD Difference Vegetation Index (2 × NIR1 − R) − (G − B) Le Maire et al. 2004 NORM NIR Normalized NIR NIR1/(NIR1 + R + G) ENVI 2013 PSRI Plant Senescence Reflectance (R − B)/RE Sims and Gamon 2002 Index RVI Ratio vegetation index NIR/RED Hildebrandt 1996 GR Green–red ratio G/R Waser et al. 2014 BR Blue ratio (R/B) × (G/B) × (RE/B) × (NIR/B) Waser et al. 2014 RR Red ratio (NIR1/R) × (G/R) × (NIR1/RE) Waser et al. 2014 REY RedEdge yellow ratio (RE − Y)/(RE + Y) Gwata 2012 VIRE Vegetation Index based on NIR1/RE Chávez Oyanadel RedEdge and Clevers 2012 RGI Red–green index RED/GREEN Miura et al. 2008 EVI1 Enhanced Vegetation Index 1 2.4 × (NIR1 − RED)/(NIR1 + RED + 1) Bezerra et al. 2020 EVI2 Enhanced Vegetation Index 2 2.4 × (NIR2 − RED)/(NIR2 + RED + 1) Jiang et al. 2008 GI2 Greenness Index 2 (B × (−0.2848) + G × (− 0.2434) + R × (− 0.5436) + NIR1 × 0.7243 + Gwata 2012 NIR2 × 0.0840) × 5 identification of the damaged stage of a single tree. physiological factors values corresponding to different Kappa values of < 0.4, 0.4–0.8, and > 0.80 indicate poor canopy colors. agreement, moderate agreement, and strong agreement, The LAI decreased with changes in the canopy color respectively (Meddens et al. 2011). 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 physio- Results logical factors corresponding to different canopy colors Damage stage determination demonstrated a considerably significant difference. The Figure 4 shows three different canopy colors of P. gan- gradual damage progression of single P. gansuensis trees suensis damaged by ALB (green, yellow and grey). Fig- was reflected by a change in the canopy color from ure 5 shows difference of four physiological factors on green to yellow to gray. As such, the color change in the three canopy colors under the 95% confidence interval. remote-sensing images is an important indicator for the Clear thresholds about tree physiological factors between assessment of the degree of damage to individual trees. different canopy colors can be observed. Table 3 shows The canopy color is an excellent parameter to judge the Fig. 4 Three canopy colors of P. gansuensis damaged by ALB Zhou et al. Forest Ecosystems (2021) 8:35 Page 8 of 12 Fig. 5 Under the 95% confidence interval (a) the average value of LAI corresponding to different damaged stages; (b) the average wormhole grade corresponding to different damaged stages; (c) the average dead branch rate corresponding to different damaged stages; (d) the average corresponding to different damaged stages Percentage of injured branches degree of damage to individual poplars with advantages The user’s accuracies (UA), overall accuracies (OA) and of a strong correlation with tree physiology and acquisi- Kappa (K) statistics of each classifier are presented in tion from remote-sensing images. 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 Classification results blue ratio (BR), plant senescence reflectance index (PSRI), For 399 segmented trees, there were 25 explanatory pre- NIR-red-yellow ratio (NIRRY), and NIR2 greatly affect the dictors, which were further categorized into 14 predictors classification precision. The NIR2 bands were unique to based on VIF and stepwise regression analyses (Fig. 6). eight-band satellite imaging, indicating the importance of Table 3 Physiological factors values corresponding to different canopy colors Canopy color Measured variation DBH (cm) LAI Exit hole level Dead twig level Damaged shoot ratio (%) Green Avg 30.207 2.407 1.849 1.644 31.301 stdev 6.341 0.328 0.328 1.059 20.582 range 23 1.575 3.000 2.000 60.000 Yellow Avg 26.982 2.290 2.167 2.181 42.778 stdev 6.616 0.258 0.435 1.012 19.661 range 22 1.475 2.000 3.000 40.000 Grey Avg 27.174 2.122 3.464 3.565 70.942 stdev 6.813 0.435 0.258 1.036 25.786 range 18 2.225 3.000 3.000 60.000 Zhou et al. Forest Ecosystems (2021) 8:35 Page 9 of 12 Fig. 6 Graphs of variable importance in the model for damage stages of Populus gansuensis caused by ALB. (a) Mean decrease in the accuracy of the model if that variable were to be removed; (b) mean decrease in the Gini coefficient. Variables that result in nodes with higher purity have a higher decrease in Gini coefficient using WV-2. The OA rate was > 80%, which can meet the Discussion requirements of actual production. Regarding the classifi- Canopy color and damage periods of individual trees cation results of the three stages of green, yellow, and gray, The damage periods of conifer trees are recognized by the separation accuracy of yellow and green stages was changes in the canopy color from green to yellow to red higher. Considering RF as an example, the classification (Zhan et al. 2020); however, there is no clear indicator accuracy values of green and yellow stages reached 0.897 for the damage period of broad-leaved trees. Previous and 0.911, respectively, whereas that of the gray stage was studies on the damage level of broad-leaved trees at the only 0.750 (Table 5). single wood scale were mostly based on a specific index, such as leaf loss rate, emergence hole number, or dead branch rate (Waser et al. 2014). Some studies have nor- Predictive mapping malized all of these factors into a single damage index to A predictive map of the damage stages based on the RF indicate the damage level on a single wood scale. The classification model was generated, and the highest OA normalized index was then used to perform modeling area was selected as a display (Fig. 7). The RF classifier and inversion (Pontius et al. 2008; Pontius et al. 2017), using all spectral variables was used to generate a pre- as these factors cannot be directly assessed from dictive map. In this study, we selected a moderately remote-sensing images. Therefore, we aimed to use infested forest for RF mapping. Figure 7a represents 0.5- high-resolution UAV images to assess an abnormal m WV-2 pan-sharpened true color images. The results phenotype at the tree level (e.g., canopy color) for dir- of mapping of different damage periods obtained using ectly determining the degree of damage to a single tree. the RF classifier are presented in Fig. 7b. We did not choose to use one single factors to repre- sent the damage level of individual tree because there are possibly subjective, ocular measurements through Table 4 Accuracy for different stages of damaged Populus gansuensis. The best models are marked in bold Classification approach Damage stage UA OA K Table 5 Confusion matrix for the classification of damaged RF green 0.897 Populus gansuensis yellow 0.911 0.862 0.804 Reference data Classified as grey 0.750 damage stages green yellow grey User’s Accuracy SVM green 0.837 green 35 1 3 0.897 yellow 0.846 0.824 0.756 yellow 1 41 3 0.911 grey 0.784 grey 4 4 24 0.750 CART green 0.818 Prod’s accuracy 0.875 0.891 0.8 yellow 0.75 0.775 0.692 OA 0.862 grey 0.774 K 0.804 Zhou et al. Forest Ecosystems (2021) 8:35 Page 10 of 12 Fig. 7 Detailed example of the RF mapping of Populus gansuensis forest. (a) 0.5-m WV-2 pan-sharpened true color image; (b) three damaged stages mapping based on RF classifier field assessments. We collected them all to minimize recommended by many researchers (Pu and Landry subjective, ocular measurements where possible. At the 2012). Thus, focusing mainly on sunlit tree crowns in same time, in the process of investigation, we found that the present study greatly improved the accuracy of dam- canopy color is a good indicator, which well represents age levels and tree species identification. In addition, be- these indicators and is suitable for remote sensing moni- cause the performance of object-based classification toring. Canopy color has not been used to determine the largely depends on the quality of image segments, seg- damage level in studies on disease monitoring of broad- mentation was optimized iteratively by approximating leaved trees. In our study, we obtained relevant data of them to the shape of the tree crown based on assessors’ 229 damaged P. gansuensis trees and determined tree experience. physiological factors, including LAI, damaged shoot ra- The dense canopy of P. gansuensis poses a challenge tio, and exit holes. The analysis results of these factors to segment single trees by using 0.5-m resolution WV-2 were highly significant, suggesting that these parameters imaging. In this study, object-based supervised classifica- can strongly indicate the damage stages of individual tion methods were used to detect different damage trees. If the degree of damage can be predicted by an ab- stages. Canopy was automatically extracted by eCogni- normal growth phenotype, infestations can be managed tion Developer (Trimble Geospatial, USA). Considering based on remote-sensing images in the growth cycle of the error of automatic segmentation by the software, the the year. This approach greatly reduces time and ex- WV-2-sharpened multispectral canopy image automatic- penses related to growth management and field investi- ally drawn from a 0.5-m resolution is not always consist- gation. Thus, our results indicate the importance of ent with the actual object. We, therefore, chose to use determining the canopy color to judge the extent of the eCognition’s multiresolutional segmentation method damage from infestation. Finally, the time interval be- for single wood segmentation (Yu et al. 2020). Based on tween ground data collection and satellite image collec- UAV images, we adjusted the parameters several times tion must be minimal for achieving consistent and to obtain an accurate individual tree segmentation map. accurate results. In our study, the ground truth data col- Then, we used the previously established inversion lection and WV-2 imaging were performed in Septem- model to obtain a prediction map of the damage level ber, 2020, for ensuring data consistency. Therefore, our on a single wood scale. Although WV-2 images have a results accurately reflected the relationship between can- high spatial resolution, the canopy drill used to segment opy color and damage degrees. the object will inevitably be combined with other pixels, including bare soil, fallen leaves, and other tree canopies. WV-2 data and single wood segmentation This will affect the classification accuracy of the trees In this study, the applicability of WV-2 data was thor- based on different damage levels. Future research must oughly tested, through which a high OA was achieved. use more advanced techniques to extract data of tree WV-2 is suitable for classifying different damage stages crowns to improve the classification accuracy. The ef- of trees, and it was found to have high accuracy and fectiveness of the object-based classification depends on Kappa value in our research. the segmentation quality. The automatic canopy demar- Masking of the shadowed tree elements to minimize cation that has been successfully applied in many studies the high in-class variance of tree crowns has been should also be further studied. To distinguish different Zhou et al. Forest Ecosystems (2021) 8:35 Page 11 of 12 stages of damage, other sensors, such as Lidar or hyper- City, Northwest China. Meanwhile, different physio- spectral data, must be tested. logical factors behind the damage stages (green, yellow, and gray) can also be understood, which can reduce the Feature variables and classification accuracy cost of field data collection and increase the accuracy To the best of our knowledge, remote-sensing technol- and applicability of management measures. The gener- ogy has not been applied to monitor ALB-infested trees ated maps represent the spatial and single tree damage in a large area, let alone at a single tree level. Therefore, data required to implement prevention and control mea- we expanded the scope and referred to previous studies sures, thereby reducing the large-scale harm from ALB using satellite imaging to detect damage to broad-leaved in the shelter forest in Northwest China. trees. In our study, field survey data and UAV images Abbreviations were used, in addition to categorization of the damage ALB: Asian longhorned beetle; RF: Random Forest; SVM: Support Vector levels of P. gansuensis into green, yellow, and gray can- Machine; CART: Classification And Regression Tree; WV-2: Worldview-2; VIs: Vegetation indices opy stages. The results indicated the overall classification accuracy of 82%–87%. The RF model using original Authors’ contributions spectrum and VI was found to be the best, with an OA Quan Zhou, Linfeng Yu, Lili Ren and Youqing Luo designed this research. Quan Zhou carried out the fieldwork, analyzed the results, and wrote the of 86.2% and a Kappa coefficient value of 0.804. For RF, paper. Xudong Zhang and Linfeng Yu facilitated in the fieldwork. The the classification accuracy values of green and yellow author(s) read and approved the final manuscript. stages reached 0.897 and 0.911, respectively, whereas Funding that for the gray stage was only 0.750. This result is also This study was supported by National Key Research & Development Program consistent with the actual situation because for the sin- of China “Research on key technologies for prevention and control of major gle wood at the gray stage, the top of the canopy is gray disasters in plantation” (Grant No. 2018YFD0600200). Beijing’s Science and Technology Planning Project “Key technologies for prevention and control of and the lower part of the canopy is still green or yellow, major pests in Beijing ecological public welfare forests” (Grant Nos. which may have a greater impact on the classification Z191100008519004 and Z201100008020001). accuracy. This can also be seen from the UAV images, Availability of data and materials which are consistent with the actual scenario. Several The data are available upon a reasonable request to the Authors. studies have shown that VIs can better indicate tree pressure than single-band reflectance because they com- Declarations bine information from multiple bands (Immitzer and Ethics approval and consent to participate Atzberger 2014). As for original bands, the red edge and Not applicable. NIR band reflectivity decreased with increasing degrees Consent for publication of damage, and the reflectivity difference was most obvi- Not applicable. ous in the NIR band. This research confirmed that the four most important categorical variables for the classifi- Competing interests The authors declare that they have no competing interests. cation of P. gansuensis by RF are BR, PSRI, NIRRY, and NIR2. VIS can significantly improve the accuracy of Received: 9 February 2021 Accepted: 20 May 2021 damage degree discrimination. Meanwhile, NIR2 is unique to eight-band WV-2 satellite images, implying References the importance of using WV-2 to study degrees of tree Bezerra FGS, Aguiar APD, Alvalab RCS, Giarollaa A, Bezerraa KRA, Lima PVPS, do damage from pest infestation. Nascimento FR, Araie E (2020) Analysis of areas undergoing desertification, using EVI2 multi-temporal data based on MODIS imagery as indicator. Ecol Indic 117:106579. https://doi.org/10.1016/j.ecolind.2020.106579 Conclusion Breiman L (2001) Random forests. Mach Learn 45(1):5–32. https://doi.org/10.1 In summary, we developed a novel approach of combin- 023/A:1010933404324 Chávez Oyanadel RO, Clevers JGPW (2012) Object-based analysis of 8-bands ing WV-2 imaging data and tree physiological factors for worldview-2 imagery for assessing health condition of desert trees. semi-automatic classification of three stages of P. gan- Wageningen UR, Wageningen suensis damage from ALB infestation. The approach was Cortes C, Vapnik V (1995) Support-vector networks. Mach Learn 20(3):273–297. https://doi.org/10.1007/BF00994018 also used to determine the canopy color, an abnormal Development Core Team R (2018) R: a language and environment for statistical phenotype, which could be directly assessed on remote- computing. R foundation for statistical computing, Vienna sensing images at the tree level to predict the degree of ENVI (2013) Vegetation Indices. Available online: http://www.exelisvis.com/docs/ VegetationIndices.html. Accessed 10 Apr 2014 tree damage. The OA of detecting the damage degrees Favaro R, Wichmann L, Ravn HP, Faccoli M (2015) Spatial spread and infestation of P. gansuensis is promising. An accurate and up-to- risk assessment in the Asian longhorned beetle, Anoplophora glabripennis. date information about location and health information Entomol Exp Appl 155(2):95–101. https://doi.org/10.1111/eea.12292 Fragnière Y, Forster B, Hölling D, Wermelinger B, Bacher S (2018) A local risk map at a single wood scale with a high spatial resolution (0.5 using field observations of the Asian longhorned beetle to optimize m) can be provided, which is highly important for man- monitoring activities. J Appl Entomol 142(6):578–588. https://doi.org/10.1111/ aging tree damage due to ALB infestation in Jiuquan jen.12491 Zhou et al. Forest Ecosystems (2021) 8:35 Page 12 of 12 Franklin SE, Fan H, Guo X (2007) Relationship between Landsat TM and SPOT Pontius J, Martin M, Plourde L, Hallett R (2008) Ash decline assessment in vegetation indices and cumulative spruce budworm defoliation. Int J Remote emerald ash borer-infested regions: a test of tree-level, hyperspectral Sens 29:1215–1220 technologies. Remote Sens Environ 112(5):2665–2676. https://doi.org/10.101 Gao H, Huang Y (2020) Impacts of the three-north shelter forest program on the 6/j.rse.2007.12.011 main soil nutrients in northern Shanxi China: a meta-analysis. Forest Ecol Pu R, Landry S (2012) A comparative analysis of high spatial resolution IKONOS Manag 458:117808. https://doi.org/10.1016/j.foreco.2019.117808 and WorldView-2 imagery for mapping urban tree species. Remote Sens Environ 124:516–533. https://doi.org/10.1016/j.rse.2012.06.011 Gitelson AA, Kaufman YJ, Merzlyak MN (1996) Use of a green channel in remote Qiu MX, Liu JQ (1985) Some observation on the biological and ecological sensing of global vegetation from EOS-MODIS. Remote Sens Environ 58(3): characters of Populus gansuensis. Chin J Plan Ecol 9(3):202–211 289–298. https://doi.org/10.1016/S0034-4257(96)00072-7 Sims DA, Gamon JA (2002) Relationships between leaf pigment content and González-Rodríguez G, Colubi A, Gil MÁ (2012) Fuzzy data treated as functional spectral reflectance across a wide range of species, leaf structures and data: a one-way ANOVA test approach. Comput Stat Data An 56(4):943–955. developmental stages. Remote Sens Environ 81(2-3):337–354. https://doi. https://doi.org/10.1016/j.csda.2010.06.013 org/10.1016/S0034-4257(02)00010-X Gwata B (2012) Developing high resolution clutter for wireless network Sjöman H, Ôstberg J, Nilsson J (2014) Review of host trees for the wood-boring propagation using WorldView-2 imagery. In: Proc. SPIE 8390. Algorithms and pests Anoplophora glabripennis and Anoplophora chinensis: an urban forest Technologies for Multispectral, Hyperspectral, and Ultraspectral imagery XVIII, perspective. Arboric Urban For 40(3):143–164 83902Q, Baltimore, Maryland, 24 may 2012. https://doi.org/10.1117/12.919091 Thompson DG, Tonon A, Beltran E, Hernandez F (2018) Inhibition of larval growth Haack RA, Hérard F, Sun J, Turgeon JJ (2010) Managing invasive populations of and adult fecundity in Asian long-horned beetle (Anoplophora glabripennis) Asian longhorned beetle and citrus longhorned beetle: a worldwide exposed to azadirachtins under quarantine laboratory conditions. Pest perspective. Annu Rev Entomol 55(1):521–546. https://doi.org/10.1146/a Manag Sci 74(6):1351–1361. https://doi.org/10.1002/ps.4810 nnurev-ento-112408-085427 Waser LT, Küchler M, Jütte K, Stampfer T (2014) Evaluating the potential of Hildebrandt G (1996) Fernerkundung und Luftbildmessung: für Forstwirtschaft, WorldView-2 data to classify tree species and different levels of ash mortality. Vegetationskartierung, und Landschaftsökologie. Herbert Wichmann, Remote Sens 6(5):4515–4545. https://doi.org/10.3390/rs6054515 Heidelberg Wulder MA, Dymond CC, White JC, Leckie DG, Carroll AL (2006) Surveying Immitzer M, Atzberger C (2014) Early detection of bark beetle infestation in mountain pine beetle damage of forests: a review of remote sensing Norway spruce (Picea abies L.) usingWorldView-2 data. Photogramm opportunities. Forest Ecol Manag 221(1-3):27–41. https://doi.org/10.1016/j. Fernerkun 5:351–367 foreco.2005.09.021 Ismail R, Mutanga O, Ahmed F (2008) Discriminating Sirex noctilio attack in pine Yu LF, Zhan ZY, Ren LL, Zong SX, Luo YQ, Huang HG (2020) Evaluating the forest plantations in South Africa using high spectral resolution data. potential of WorldView-3 data to classify different shoot damage ratios of Hyperspectral Remote Sensing of Tropical and Sub-Tropical Forests. Crc Pinus yunnanensis. Forests 11(4):417. https://doi.org/10.3390/f11040417 Press, London, pp 161–175 Zhan ZY, Yu LF, Li Z, Ren LL, Gao BT, Wang LX, Luo YQ (2020) Combining GF-2 Jiang Z, Huete AR, Didan K, Miura T (2008) Development of a two-band and Sentinel-2 images to detect tree mortality caused by red turpentine enhanced vegetation index without a blue band. Remote Sens Environ beetle during the early outbreak stage in North China. Forests 11(2):172. 112(10):3833–3845. https://doi.org/10.1016/j.rse.2008.06.006 https://doi.org/10.3390/f11020172 Jing W, Yang Y, Yue X, Zhao X (2015) Mapping urban areas with integration of Zhang X, Yan G, Li Q, Li ZL, Wan H, Guo Z (2006) Evaluating the fraction of DMSP/OLS nighttime light and MODIS data using machine learning vegetation cover based on NDVI spatial scale correction model. Int J Remote techniques. Remote Sens 7(9):12419–12439. https://doi.org/10.3390/ Sens 27(24):5359–5372. https://doi.org/10.1080/01431160600658107 rs70912419 Kaszta Ż, Van De Kerchove R, Ramoelo A, Cho MA, Madonsela S, Mathieu R, Wolff E (2016) Seasonal separation of African savanna components using WorldView-2 imagery: a comparison of pixel-and object-based approaches and selected classification algorithms. Remote Sens 8(9):763. https://doi.org/1 0.3390/rs8090763 Le Maire G, Francois C, Dufrene E (2004) Towards universal broad leaf chlorophyll indices using PROSPECT simulated database and hyperspectral reflectance measurements. Remote Sens Environ 89(1):1–28. https://doi.org/10.1016/j. rse.2003.09.004 Lin QN, Huang HG, Wang JX, Huang K, Liu YY (2019) Detection of pine shoot beetle (PSB) stress on pine forests at individual tree level using UAV-based hyperspectral imagery and lidar. Remote Sens 11(21):2540. https://doi.org/1 0.3390/rs11212540 Liu Y, Liu R, Chen J, Cheng X, Zheng G (2013) Current status and perspectives of leaf area index retrieval from optical remote sensing data. J Geo-infor Sci 15(5):734–743 (in Chinese) Marchisio G, Pacifici F, Padwick C (2010) On the relative predictive value of the new spectral bands in the WorldWiew-2 sensor. 2010 IEEE International Geoscience and Remote Sensing Symposium, pp 2723–2726 Meddens A, Hicke J, Vierling L (2011) Evaluating the potential of multispectral imagery to map multiple stages of tree mortality. Remote Sens Environ 115(7):1632–1642. https://doi.org/10.1016/j.rse.2011.02.018 Miura T, Yoshioka H, Fujiwara K, Yamamoto H (2008) Inter-comparison of ASTER and MODIS surface reflectance and vegetation index products for synergistic applications to natural resource monitoring. Sensors 8(4):2480–2499. https:// doi.org/10.3390/s8042480 Orlando F, Movedi E, Paleari L, Gilardelli C, Foi M, Dell’Oro M, Confalonieri R (2015) Estimating leaf area index in tree species using the PocketLAI smart app. Appl Veg Sci 18(4):716–723. https://doi.org/10.1111/avsc.12181 Pontius J, Hanavan RP, Hallett RA, Cook BD, Corp LA (2017) High spatial resolution spectral unmixing for mapping ash species across a complex urban environment. Remote Sens Environ 199:360–369. https://doi.org/10.101 6/j.rse.2017.07.027

Journal

"Forest Ecosystems"Springer Journals

Published: Jun 3, 2021

There are no references for this article.