Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Early detection of pine wilt disease in Pinus tabuliformis in North China using a field portable spectrometer and UAV-based hyperspectral imagery

Early detection of pine wilt disease in Pinus tabuliformis in North China using a field portable... Background: Pine wilt disease (PWD) is a major ecological concern in China that has caused severe damage to millions of Chinese pines (Pinus tabulaeformis). To control the spread of PWD, it is necessary to develop an effective approach to detect its presence in the early stage of infection. One potential solution is the use of Unmanned Airborne Vehicle (UAV) based hyperspectral images (HIs). UAV-based HIs have high spatial and spectral resolution and can gather data rapidly, potentially enabling the effective monitoring of large forests. Despite this, few studies examine the feasibility of HI data use in assessing the stage and severity of PWD infection in Chinese pine. Method: To fill this gap, we used a Random Forest (RF) algorithm to estimate the stage of PWD infection of trees sampled using UAV-based HI data and ground-based data (data directly collected from trees in the field). We compared relative accuracy of each of these data collection methods. We built our RF model using vegetation indices (VIs), red edge parameters (REPs), moisture indices (MIs), and their combination. Results: We report several key results. For ground data, the model that combined all parameters (OA: 80.17%, Kappa: 0.73) performed better than VIs (OA: 75.21%, Kappa: 0.66), REPs (OA: 79.34%, Kappa: 0.67), and MIs (OA: 74.38%, Kappa: 0.65) in predicting the PWD stage of individual pine tree infection. REPs had the highest accuracy (OA: 80.33%, Kappa: 0.58) in distinguishing trees at the early stage of PWD from healthy trees. UAV-based HI data yielded similar results: the model combined VIs, REPs and MIs (OA: 74.38%, Kappa: 0.66) exhibited the highest accuracy in estimating the PWD stage of sampled trees, and REPs performed best in distinguishing healthy trees from trees at early stage of PWD (OA: 71.67%, Kappa: 0.40). Conclusion: Overall, our results confirm the validity of using HI data to identify pine trees infected with PWD in its early stage, although its accuracy must be improved before widespread use is practical. We also show UAV-based data PWD classifications are less accurate but comparable to those of ground-based data. We believe that these results can be used to improve preventative measures in the control of PWD. Keywords: Pine wilt disease, Remote sensing, Spectrometer, Hyperspectral imaging, Random forest, Classification * Correspondence: youqingluo@126.com Key Laboratory for Forest Pest Control, College of Forestry, 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/. Yu et al. Forest Ecosystems (2021) 8:44 Page 2 of 19 Background content. Therefore, water and chlorophyll content could The pine wood nematode (PWN; Bursaphelenchus xylo- be used as indicators of PWD (Huang 2020). This is im- philus) is a hazardous invasive species that infests mul- portant because RS and HRS methods can be used to es- tiple species of pine (Vicente et al. 2012; Douda et al. timate water and chlorophyll content. For example, 2015). Pine wilt disease (PWD), caused by the PWN, is using a field portable spectrometer to measure the spec- widespread throughout East Asia (Mamiya 1988; Hyun tral characteristics of P. thunbergii and P. massoniana at et al. 2007;Ye 2019). Previously isolated to southern different stages of PWN infection, Xu et al. (2011) found China, PWD is now found throughout the country, in- the reflectance spectrum curve in the mid-infrared band cluding Northeast China (Pan et al. 2019; Yu et al. may indicate the early stage of PWD with the analysis of 2019). In 2016, PWD first appeared in Dalian, Liaoning the spectral characteristic parameters and changes in Province, then in May 2017, it happened in Dandong chlorophyll levels. Similarly, Xiang et al. (2018) used a City, Fushun City, Benxi City and other places (National field portable spectrometer, analyzing the relationship Forestry Administration 2018). In addition, Monocha- between spectral properties and chlorophyll, showing mus saltuarius was identified as a new vector of PWD in that the chlorophyll content of pine decreases with the Liaoning Province of China (Yu et al. 2018). In the stage of PWD (later, more severe stages are associated process of spreading northwards, PWD has infected and with lower chlorophyll content). In addition, the position caused severe damage to the Chinese pine (Pinus tabu- of red edge, the wavelength of red edge, the height of laeformis), Korean pine (P. koraiensis), and larch (Larix green peak, and the depth of red band absorption all spp.) populations. This has resulted in significant eco- strongly correlate with chlorophyll content (e.g., Xiang nomic losses and ecological damage to Chinese pine for- et al. 2018). Correspondingly, the area surrounded by ests (e.g., Li et al. 2011; Lin 2015; Hui 2018). the first-order differential spectrum in the 490–530 nm To effectively control PWD, it is necessary to identify range and that in the 680–760 nm range was found to infected trees in the early stage of infection. This is a dif- be a significant hyperspectral feature indicating the oc- ficult task because most trees progress from initial infec- currence of PWD (e.g., Huang et al. 2012). These studies tion to the serious infection stage within 5 weeks all used a field portable spectrometer, which cannot be (Umebayashi et al. 2017). Consequently, current PWD applied in a large-scale area. management strategies emphasize the control of infected Past studies used satellite imagery such as Landsat, trees after the onset of an outbreak by means of fumiga- IKONOS, Quick Bird, and GF-2 images to detect forest tion, burning, and tree felling (Shin 2008; Kim et al. pest disease (e.g., Franklin et al. 2003; White et al. 2005; 2018). What is lacking is a methodology that monitors Hicke and Logan 2009; Zhan et al. 2020). However, due pine populations that can quickly and efficiently detect to limitations in spatial, temporal, spectral resolution as the early signs of PWD (Ma et al. 2011). In addition, well as weather complications, satellite imagery cannot many efforts have been made in early detection of PWD obtain real-time data (Santoso et al. 2016). Because of (Kim et al. 2018; Syifa et al. 2020; Tao et al. 2020), but these limitations, the detection scale of forest pest dis- not in Chinese pine. In this paper, we present a method ease has shifted to Unmanned Airborne Vehicle (UAV) aimed at detecting PWD in Chinese pine in early stage. remote sensing, which offers the advantages of low con- A major obstacle in the management of pines infected sumables and operating costs, high ground resolution by PWD is that the forests they persist in are very large data collection, and more precise accuracy (Tang et al. communities. This can make classical ground identifica- 2015). For example, Huang et al. (2018) used a fixed- tion and sampling methods impractical. To solve this wing unmanned aerial vehicle to monitor dead pine problem, recent studies have used remote sensing (RS) trees caused by PWD, successfully monitoring pine tree to examine the impact of PWD on the physiological and mortality with over 80% accuracy. Li et al. (2020) used biochemical changes after infection (e.g., Shen et al. UAVs to acquire remote sensing images of forest areas 2001; Li et al. 2004; Wang et al. 2007). Advancements in to assess the presence of PWD, successfully recognizing RS technology increasingly support the prediction effi- infection with 90.4% accuracy. Huang (2020) used UAV ciency by reducing inherent spatial and temporal con- multispectral data to draw a conclusion that the first de- straints (Ahmed et al. 2020). Similarly, hyperspectral rivative of healthy and infected P. thunbergii changed remote sensing (HRS) can obtain continuous spectral in- markedly at 710 nm. Except RGB and multispectral cam- formation of objects – this has been used to detect era, hyperspectral imagery was also applied in detecting changes in the spectral characteristics of needles on forest pest diseases. Abdel-Rahman et al. (2014) used infected-trees in the process of discoloration caused by airborne hyperspectral data, random forest and support PWN infection (Pan 2011;Kuai 2012). vector machines classifiers to distinguish amongst Previous studies show the presence of PWD is signifi- healthy, Sirex noctilio grey-attacked and lightning- cantly linearly correlated with water and chlorophyll damaged pine trees. Zhang et al. (2018) utilized the Yu et al. Forest Ecosystems (2021) 8:44 Page 3 of 19 ISIC-SPA-P-PLSR framework based on UAV-based and serious stages) of PWD infection in Chinese pine hyperspectral image to identify the degree of damage trees. Additionally, analyses of PWD simultaneously trees caused by Dendrolimus tabulaeformis. Iordache considering ground and UAV-based hyperspectral data et al. (2020) acquired airborne multispectral and hyper- have not been widely conducted. spectral data, and used Random Forest algorithms to Therefore, to fill this gap, in this study, our objective is compare the classification accuracies of the two datasets to explore the capacity of ground and airborne hyper- in detecting PWD, finding that both datasets performed spectral data using VIs, REPs, and MIs to classify the well in identifying the infected, suspicious, and healthy stage of PWD infection in Chinese pine at the tree level. trees. Importantly, however, in detecting the PWD, most Furthermore, we also aim to provide a useful and fairly studies focus on distinguishing between healthy and in- accurate method of distinguishing between trees in the fected trees using RGB (red, green and blue bands) cam- early stage of PWD infection from healthy trees. era, multispectral data, and ground hyperspectral data, but UAV-based hyperspectral data were not widely stud- Materials and methods ied. In addition, few studies emphasized the identifica- Study area and ground survey tion of trees in each stage of PWD infection in Chinese We conducted our study in Cangshi Village, located in pine, which we focus on in this study. In our study, we Fushun County, Liaoning Province, in northeastern systematically divided the infection stage into four China (124°21′–124°24′ E, 41°53′–41°57′ N; Fig. 1). In stages, making the detection more accurate. Because the study area, the species of plantation forests is domi- high spatial and spectral resolutions, and feasibility of nated by Chinese pine (P. tabulaeformis), and the age of large-scale area application are needed to distinguish the them is approximately 40–50 years. The total area of for- subtle difference between healthy and the early stage of est cover in Fushun County is approximately 12.43 × 10 infected trees, we consider UAV-based hyperspectral ha, of which P. tabulaeformis makes up > 30%. In imagery. addition, the broadleaf tree species and understory vege- Spectral indices, such as Vegetation indices (VIs), red tation in the study site mainly include, Quercus acutis- edge parameters (REPs), and moisture indices (MIs), can sima, Quercus mongolica, grass, et al. The area is reflect the infection condition of PWD (e.g., Kim et al. situated in the Middle Temperate Zone. It has a contin- 2018; Huang 2020). VIs is a combination of different re- ental monsoon climate and experiences approximately mote sensing spectral bands, which can be regarded as a 804.2 mm of precipitation per year. The mean annual air sign of relative abundance and activity of green vegeta- temperature is approximately 6.6 °C. tion (Jones and Vaughan 2010). Over the past years, VIs According to local Forestry Administration records, had been widely applied to extract sensitive estimates of PWD has resulted in the death of tens of thousands of plant biochemical characteristics (e.g., He et al. 2015;De pine trees since the onset of outbreaks in 2016 in Liaoning Klerk and Buchanan, 2017), such as the normalized dif- Province (National Forestry Administration 2018). ference vegetation index (NDVI) that decreases with in- Field measurements were conducted in 12–18 August creasing tree PWD stage severity (Kim et al. 2018), and 2019. We established three 30 m × 30 m plots located the presence of PWD can be detected by calculating the northeast of Cangshi Village (Fig. 1). The coordinates of VIs based on ground, aerial, and satellite data (e.g., the plot boundary and the location of each tree were re- White et al. 2007; Pan et al. 2014; Jung and Park 2019; corded using a handheld differential global positioning Iordache et al. 2020). The REPs are derived from Red system (DGPS, Version S760) with sub-meter accuracy. edge (680–780 nm), which is the most obvious feature of In each plot, we recorded tree growth state including plant spectral curve. As an indicator of plant stress and tree height (H), diameter at breast height (DBH), crown often used to study the growth and health of plants diameter (CD), and PWD infection stage. In addition, we (Boochs et al. 1990; Dawson and Curran 1998), it also measured biochemical parameters: the leaf chlorophyll had been well studied in detecting the PWD (Du et al. content (Cab) and water content (WC) of each tree. Cab 2009; Huang et al. 2012). Additionally, pine trees killed was derived by averaging the Cab of needles from four by PWD by blocking transmission of water (Yang 2002), different directions using a calibrated CCM-300 Chloro- and the water content of pine needles decreased with in- phyll Content Meter. The Cab of seriously damaged creasing PWD infection severity (Chen, 2005). Thus, trees was 0 measured by the CCM-300. Meanwhile, WC changes in MIs (also derived from radiometric data) can of each tree was determined by the fresh weight (FW) be used to detect the presence of trees infected with minus the dry weight (DW) divided by the FW: (WC = PWD (Xu et al. 2012; Song et al. 2018). (FW – DW)/FW × 100%). Finally, a total of 218 pine Although spectral indices were widely used to detect trees (healthy: 76; early stage: 54; middle stage: 47; ser- the PWD, there is no study that yet provides good pa- ious stage: 41) were measured. Summary statistics of rameters to predict each stage (healthy, early, middle, three plots are given in Table 1. Yu et al. Forest Ecosystems (2021) 8:44 Page 4 of 19 Fig. 1 (a) The map of China; (b) The map of Liaoning Province; (c) The location of the study area. The purple and red rectangles represent three field plots locations and UAV hyperspectral flight areas, respectively Additionally, we randomly selected 20 discolored pine were defined as having dark green needles, normal resin trees as samples from each plot, and took them back to secretion, and vigorous growth. “Early stage” trees were the laboratory for testing by Behrman funnel method. defined by slightly yellowed needles, with decreased resin The result showed that they all carried pine wood secretion and grow rates. “Middle stage” trees were de- nematode. fined by yellow-brown needles, wilt, and weak growth. Dry trees with reddish-brown needles were defined as the “Serious stage”. Infected stage division On the basis of previous studies (Xu et al. 2011;Santos and de Vasconcelos, 2012), we combined needle, ground Remote sensing data acquisition and preprocessing tree, and UAV images to categorize PWD infection into Ground spectrum acquisition four stages: (1) Healthy, (2) Early stage, (3) Middle stage, From the ground (physically measuring trees in the field), and (4) Serious stage (Fig. 2). Stages were defined by color we measured the spectrum of sampled trees using ASD of needles, growth vigor, and resin secretion (Table 2). We Field Spec 4 Hi-Res NG (Analytical Spectral Devices, had four people classified each tree, and took the major- Boulder, CO, USA). The spectral range is 350–2500 nm ity’s opinion as final results to reduce subjective errors. Fi- and the spectral resolution is 3 nm in the 350–1000 nm nally, we used the following definitions: “Healthy” trees wavelength range and 6 nm in the 1001–2500 nm Table 1 Statistics of three plots variables (tree numbers = 218) Mean Standard deviation Maximum Minimum Range H (m) 12.66 1.12 14.50 9.60 4.90 DBH (cm) 21.79 5.40 34.06 7.83 26.23 CD (m) 2.13 1.22 7.80 0.80 7.00 −2 Cab (g∙m ) 299.30 165.93 560.00 0.00 560.00 WC (%) 42.13 14.45 68.32 17.25 51.07 H Tree height, DBH Diameter at breast height, CD Crown diameter, Cab Leaf chlorophyll content, WC Water content Yu et al. Forest Ecosystems (2021) 8:44 Page 5 of 19 Fig. 2 Images of needles, ground trees, and unmanned aerial vehicle (UAV) images of pine trees at different PWD infection stages wavelength range. We selected and measured branches 30, on 18 August 2019. The weather was sunny during roughly representative of the average spectrum of each the flight. Standard white board and white tarp were tree. The selected branches were cut from the east, south, placed on the ground within the flying area. The flying west, and north directions from the upper, middle, and height was set at 120 m, the image forward and side − 1 lower layers (Zhang et al. 2018). We calculated the overlaps were set to 50%, and the flight speed is 2 m∙s . spectrum of each sampled tree by averaging the spectrum The imagery consisted of 281 spectral channels (spectral of the selected branches. The ground spectrums were resolution of 2.1 nm) from visible to near infrared (NIR) gathered from 10:00 to 14:00 every day, from August 12 regions (400–1000 nm). Reflectance correction and to August 17. We obtained the ground spectrum for com- radiometric calibration were performed using 3 m car- parison with UAV-based data and auxiliary radiometric pet reference (standard white board) and the Spectronon correction. software. Image geometric corrections were performed using 4 ground control points (GCPs). The positions of UAV-based hyperspectral imagery GCPs were recorded by a DGPS device with sub-meter Hyperspectral Imagery (HI) data were obtained by using accuracy. The ground resolution of HI was produced to a DJI Matrice 600 UAV (DJI, Shenzhen, China) equipped be 0.4 m. a Pika L hyperspectral camera (Resonon, USA). The main parameters of the Pika L are listed in Table 3. Tree crowns extraction from hyperspectral imagery GNSS (Global Navigation Satellite System) and IMU (In- We conducted tree crown segmentations from HI by ertial Measurement Unit) modules are integrated into combining the object-based segmentation method with UAV, and its horizontal and vertical position errors are manually drawing ROIs (regions of interest). First, by approximately 2.0 and 5.0 m, respectively, with an orien- use of ENVI 5.3, we used the object-based segmentation tation precision of approximately 1 degree. The overall method on the HIs using combined spectral and texture UAV-based system is shown in Fig. 3. features to separate trees crowns from the grass back- UAV-based hyperspectral data acquisition was carried ground and shadows (e.g., Yuan et al. 2013). The object- out in the test areas of Cangshi Village from 12:00–12: based segmentation method successfully separated tree Table 2 Classification of infected stages Age of Classification standard Stages of infected trees the stand Healthy Early stage Middle stage Serious stage 40–50 years Color of needles; Growth Trees grow vigorously Needles begin to turn yellow; Most needles turn yellow Trees are dry, and vigor; Resin secretion with dark green needles; resin secretion decreases and brown and wilt, and the needles are all reddish resin secretes normally growth slows down growth is obviously weak brown, but do not fall off Yu et al. Forest Ecosystems (2021) 8:44 Page 6 of 19 Table 3 Main parameters of the Pika L imaging spectrometer Feature selection and prediction model for cab and WC (provided by manufacturer) According to previous study (De Klerk and Buchanan, Parameters Values Parameters Values 2017; Kim et al. 2018; Lin et al. 2019), trees health was highly correlated with biochemical properties (e.g., Cab), Weight 0.6 kg Sampling interval 1.07 nm which also can be precisely estimated using fitting models Digitization 12 bits Spectral resolution 2.10 nm based on spectral indices (Inoue et al. 2012;Schlemmer Wavelength range 400–1000 nm Spectral channels 281 et al. 2013; Xie et al. 2014). In this study, we firstly calcu- lated the Pearson’s correlation coefficient between a num- crowns from the grass background and shadow compo- ber of spectral indices (features in Tables 4, 5 and 6)and nents. However, it was difficult to separate overlapping Cab and WC, PWD infection stages of each tree, respect- crowns. Second, based on the result of object-based seg- ively. In addition, before these variables were selected for mentation, we drew the ROIs manually. We determined constructing regression and classification model to predict the location of every individual sampled trees by use of Cab, WC, and PWD infection stages, we used a stepwise the DGPS information. The ROIs of each tree were regression method to test the multicollinearity between shaped by manually drawing the crown range on the features, eliminating redundant variables. RGB image. Then, the ROIs were added to the prepro- Finally, we selected 5 VIs (NDVI, NDVI [810, 680], cessed HIs, and the spectrum of an individual tree was NDVI [560, 680], RVI, PRI), 5 REPs (λo, Sg, Kg, GH, RD), calculated by averaging the reflectance of the corre- and 5 moisture indices (MSI, WI1, WI2, NDWI, NSII) sponding ROI extracted by ENVI 5.3. The average based on ground spectrum (350–2500 nm). Based on spectrum information of each ROI was used in the sub- UAV hyperspectral data (400–1000 nm), we selected 5 VIs sequent analysis (Fig. 4). Finally, the shadow components (PSI, RVSI1, NDVI, NDVI [810, 450], RVI), 5 REPs (dλb, and overlapped crowns were discarded. Overall, 121 SDr, SDb, SDr-SDb, RD) and 2 MIs with (WI1, WI2). trees (healthy: 39; early stage: 27; middle stage: 29; ser- We estimated the Cab and WC using a RF (Breiman ious stage: 26) were segmented from HI hyperspectral 2001) regression using a bagging method based on the imagery. CART regression tree model. In the regression applica- tion, each tree was built by choosing a random sample and a random set of variables from the training dataset Features extraction by a deterministic algorithm (Mutanga et al. 2012). All In order to eliminate instrument errors and noises, while 121 samples were used for model training, and we then maintaining the original spectral characteristics, a used a 10-fold cross-validation method (Waske et al. Savitzky-Golay filter with 7 points (we tested 3–15 2009) to assess model accuracy. The process of regres- points and finally chose 7 points) was used to smooth sion was conducted using the R package “randomFor- the original spectrums of ground and UAV-based hyper- est”. The coefficient of determination (R ), RMSE (Root spectral data (Mullen 2016). Based on previous research, Mean Square Error), and RRMSE (Relative RMSE) be- we calculated 37 spectral variables including 12 VIs tween measured and estimated values were used to com- (Table 4), 20 REPs showed in Table 5 (Horler et al. pare different indices in predicting the accuracy of Cab 1980; Curran et al. 1990; Yao et al. 2009; Liu et al. 2010), and WC. After selecting the variables which performed and 5 MIs (Table 6) from spectral data. best in predicting the Cab and WC, we used the Cab or Fig. 3 The unmanned aerial vehicle (UAV)-based hyperspectral system with POS, Pika L imaging spectrometer Yu et al. Forest Ecosystems (2021) 8:44 Page 7 of 19 Fig. 4 (a) The original hyperspectral imaging (HI); (b) digital photo of the test area (upper) and hyperspectral image of one sampling plot of the corresponding region (lower); (c) the result of crowns segmentation and (d) the ROIs formed by manual drawing WC estimated by the optimum variables to classify the the higher the MDA value of a variable is, the more im- PWD infection stages directly. portant it is (e.g., Liu et al. 2017; Shi et al. 2018). The selected VIs, REPs, MIs and combining all vari- Classification based on VIs, REPs, and MIs ables were separately input into RF classification model, We then used the selected VIs, REPs, and MIs correlated and the MDA of all selected variables were determined. with Cab and WC to classify trees based on PWD infec- All 121 samples were used for model training. We then tion. We used a RF classification model to assess the infec- used a 10-fold cross-validation method to estimate tion stage of sampled trees. In a RF algorithm, the variable model accuracy. The process of classification was carried importance is a metric of how much the “out-of-bag” out using the R package “randomForest”. The overall ac- (OOB) error of estimate increases due to the removal of a curacy (OA), producer’s accuracy (PA), user’s accuracy single variable from the data (Prasad et al. 2006;Verikas (UA), and Kappa coefficient resulting from confusion et al. 2011). The mean decrease accuracy (MDA) index of matrices (Congalton 1991) were used to evaluate classifi- each variable is obtained when calculating the OOB error: cation accuracy. Kappa coefficient is a popular statistic Table 4 Vegetation indices extracted from spectral data Variables Description Formula Reference SUM Rð760:900Þ=141 - SUM Rð630:900Þ=271 NDVI Normalized difference vegetation index Richardson and Wiegand (1977) SUM Rð760:900ÞþSUM Rð630:900Þ=271 ðR - R Þ 810 450 NDVI (810,450) Normalized difference vegetation index Richardson and Wiegand (1977) NDVI ð810; 450Þ¼ ðR þR Þ 810 450 ðR - R Þ NDVI (810,680) Normalized difference vegetation index 810 680 Richardson and Wiegand (1977) NDVI ð810; 680Þ¼ ðR þR Þ 810 680 ðR - R Þ 560 680 NDVI (560,680) Normalized difference vegetation index Richardson and Wiegand (1977) NDVI ð560; 680Þ¼ ðR þR Þ 560 680 SUM Rð760:900Þ=141 RVI Ratio vegetation index Wu and Niu (2008) SUM Rð630:900Þ=271 DVI Deferent vegetable index DVI = SUM R(760:900)/141 – SUM R(630:900)/271 Chen (1996) R - R 570 531 PRI Photochemical reflectance Index Carter and Miller (1994) PRI ¼ R þR 570 531 MSR Modified simple ratio -1 Blackburn (1998) MSR ¼ sqrtð þ1Þ R695 PSI Plant stress index Hunt and Rock (1989) PSI ¼ RVSI1 Ratio vegetation stress index Hunt and Rock (1989) PVSI1 ¼ RVSI2 Ratio vegetation stress index Hunt and Rock (1989) PVSI2 ¼ PSSR Pigment specific simple ratio Penuelas et al. (1997) PSSR ¼ 635 Yu et al. Forest Ecosystems (2021) 8:44 Page 8 of 19 Table 5 Red Edge parameters extracted from hyperspectral data Type Parameters Description Type Parameters Description Based on original Rg The maximum reflectance in the Based on first dλb The first order differential value spectrum wavelength range of 510–560 nm derivative corresponding to λb Ro The minimum reflectance in the dλr The first order differential value wavelength range of 640–680 nm corresponding to λr λg Wavelength position of Rg SDb The area surrounded by the first-order differential spectrum in the range of 490–530 nm λo Wavelength position of Ro SDr The area surrounded by the first-order differential spectrum in the range of 680–760 nm Sg Skewness of reflectance in the SDr–SDb SDr–SDb wavelength range of 510–560 nm Kg Kurtosis of reflectance in the Others Lwidth Width at half depth of red band wavelength range of 510–560 nm absorption Sr Skewness of reflectance in the Depth672 Absorption depth at 672 nm range of 680–760 nm Kr Kurtosis of reflectance in the Depth560 Absorption depth at 560 nm wavelength range of 680–760 nm Based on first derivative λb Wavelength position of the maximum GH Height of green peak first derivative of reflectance between 490 and 530 nm λr Wavelength position of the maximum RD Depth of red band absorption first derivative of reflectance between 680 and 760 nm for measuring agreement (Meddens et al. 2011). A input. We examined the performance of Cab and WC es- Kappa value from < 0.4 indicates a “poor” agreement, timation of the input parameters using both ground Kappa 0.4–0.8 is defined as having moderate agreement, spectrum data and UAV-based spectral data (Figs. 6 and and Kappa > 0.80 indicates a “strong” agreement. 7). Cab estimation accuracy was slightly greater when Using the overall and individual accuracies for all four using REPs than using VIs for both ground data (REPs: 2 − 2 PWD infection stages, we examined the paired accur- R =0.78, RMSE =82.34 g∙m , RRMSE = 27.44%; VIs: 2 − 2 acies of Healthy, Early stage, Middle stage, and Serious R = 0.74, RMSE = 89.80 g∙m , RRMSE = 29.92%) and 2 − 2 stage pine trees to examine the feasibility of discriminat- UAV-based data (REPs: R = 0.75, RMSE = 87.34 g∙m , 2 − 2 ing between different stages. RRMSE = 29.11%; VIs: R = 0.72, RMSE = 94.11 g∙m , RRMSE = 31.36%). For WC predictions in which MIs were Results used as input parameters, the predictions from ground Estimation of cab and WC data were considerably more accurate than UAV-based Leaf Cab and WC decreased with the severity of PWD data. The results summarized in Table 7. infection (Fig. 5). We estimated the Cab and WC of all It showed that the model tended to overestimate Cab − 2 121 sampled trees using the RF regression model with the below 200 g∙m and underestimate Cab above 300 − 2 three input parameters (VIs, REPs, and MIs) separately g∙m (Fig. 6a and b; Fig. 7a and b), the RF regression Table 6 Moisture indices extracted from spectral data Variables Description Formula Reference R1600 MSI Moisture stress index Gao (1996) MSI ¼ R820 WI1 Water index Hardisky et al. (1983) WI1 ¼ WI2 Water index Hardisky et al. (1983) WI2 ¼ ðR - R Þ 860 1240 NDWI Normalized difference water index Prasad et al. (2006) NDWI ¼ ðR þR Þ 860 1600 ðR - R Þ 860 1600 NDII Normalized difference infrared index Verikas et al. (2011) NDII ¼ ðR þR Þ 860 1600 Yu et al. Forest Ecosystems (2021) 8:44 Page 9 of 19 Fig. 5 The difference of chlorophyll content (Cab) and water content (WC) of all samples at different infected stages model provided unsatisfactory predictions for Cab and stage of PWD. Generally, the spectral variables were sen- WC in pine trees when VIs, REPs, and MIs were taken sitive to changes in biochemical characteristic. as input parameters. In addition, Cab estimated by REPs derived from Comparisons of classifications using different variables ground data (the optimum variables) were chosen to as- from ground and UAV-based data sess the PWD infection stages directly. Finally, the re- The MDA index for the ground data and UAV-based sults showed that using Cab estimated by RF based on data strongly differed among variables. Importance rank- the optimum variables did not perform well in classify- ings indicated REPs to be more important than most VIs ing the PWD infection stages (OA = 47.11%, Kappa = and MIs (Fig. 11). The most important variables were 0.29; Table 8). It means that estimated Cab cannot be REPs, and VIs were generally more important than MIs. directly used to accurately the PWD infection stages. GH was the most important variable for ground data and SDR was the most important variable for UAV- based data. Feature analysis OA (overall accuracy) assessment using the 10-fold The spectral reflectance of trees declined as a function cross-validation method indicated that REPs performed of PWD stage severity (Fig. 8). The difference of spectral best. For ground data REPs yielded an OA of 79.34%, reflectance was obvious near the green peak (500–600 VIs 75.21%, and MIs 74.38%. Combined all variables, it nm), red edge (680–760 nm), and NIR (750–950 nm; yielded an accuracy of 80.17% (Tables 9 and 10). UAV- Fig. 9). VIs, REPs, and MIs exhibited differing responses based data provided less accurate results for all variables: to the severity of infection. While some variables such as 72.73% for REPs, 70.25% for VIs, 63.64% for MIs, and NDVI (810, 680), Kg, and NDWI decreased with the in- 74.38% for combined all variables (Tables 9 and 10). creasing infection stage, others (e.g. MSI, PRI and Sg) Kappa values yielded similar qualitative results for both significantly increased with the increasing of the infec- ground data and UAV-based data. For ground data, tion stage (Fig. 10). Therefore, almost all the selected Kappa was calculated to be 0.67 for REPs, 0.66 for VIs, variables exhibited statistically significant responses to and 0.65 for MIs. For combining all variables, Kappa im- PWD severity, indicating their potential for detecting the proved to 0.73. For UAV-based data, the values of Kappa Fig. 6 Measured vs. estimated chlorophyll content (Cab) and water content (WC) based on ground spectrum using different input parameters: (a) Vegetation Indices; (b) Red edge parameters, and (c) Moisture Indices Yu et al. Forest Ecosystems (2021) 8:44 Page 10 of 19 Fig. 7 Measured vs. estimated chlorophyll content (Cab) and water content (WC) using UAV-based spectrum with different input parameters: (a) Vegetation Indices; (b) Red edge parameters, and (c) Moisture Indices for REPs, VIs, and MIs were 0.63, 0.60, and 0.51, re- (OA: 80.33%, Kappa: 0.58) based on ground data per- spectively. For combining all variables, Kappa again im- formed best when healthy pine trees and early stage of proved (0.66). Therefore, for each data type (ground infected pine trees were compared. REPs and combining data and UAV-based data), REPs yielded the most accur- all variables performed equally well in terms of OA ate results, followed by VIs and MIs respectively. Add- (71.67%) and Kappa (0.40) for UAV-based data. itionally, ground data provided more accurate results than UAV-based data in all cases. Discussion PA (producer’s accuracy) values were high for the In this paper, we employed VIs, REPs, MIs, and combin- middle and serious stage of infection regardless of the ing all variables, to examine the capacity of ground and data source and the parameters used. UA (user’s accur- UAV-based hyperspectral data in PWD infection stages acy) was relatively high for middle and serious stage of estimation at individual tree level. The results reveal that infection, while healthy and early stages had lower UAs combining all variables performed best and yielded a (Table 11). considerably accurate classification with OA of 80.17% Pairwise comparisons of healthy, early stage, middle for ground data and 74.38% for UAV-based data (Tables stage, and serious stage indicated the OAs of all stage 9 and 10). pairs to be considerably greater than 80% in most cases When we look at the capacity of identifying pine trees (Figs. 12 and 13). Lower accuracies resulted when in the early infected stage of PWD, the REPs exhibited healthy pine trees and early stage of infected pine trees the best performance with OA of 80.33% and 71.67% were compared based on the VIs (75.41%), REPs from ground data and UAV-based data, respectively (80.33%), MIs (70.97%), and combined all variables (Figs. 12 and 13). (79.03%) from ground data, as well as VIs (68.33%), REPs Overall, it is understandable that: (1) the REPs are (71.67%), MIs (66.67%), and combined all variables more responsive to stage changes of PWD infection than (71.67%) from UAV-based data. High values of Kappa VIs and MIs, indicating that REPs may be more sensitive were obtained by most pairwise comparisons (Figs. 12 to the biochemical conditions; (2) UAV-based data per- and 13), but not for comparisons between healthy pine formed considerable accuracy in monitoring the PWD trees and pine trees in the early stage of infection for stage at individual tree level, especially REPs, showing its both ground data (VIs: 0.55, REPs: 0.58, MIs: 0.39, com- good accuracy, which were slightly lower than ground bined all variables: 055) and UAV-based data (VIs: 0.35, data and can be applied in a large-scale forest area. REPs: 0.40, MIs: 0.31, combined all variables: 0.40). REPs Table 7 The results of Cab and WC estimation using RF regression based on VIs, REPs, and MIs from ground and UAV hyperspectral data Variable type Cab and WC estimation Ground UAV 2 2 R RMSE RRMSE (%) R RMSE (%) RRMSE (%) VIs (estimating Cab) 0.74 89.80 29.92 0.72 94.11 31.36 REPs (estimating Cab) 0.78 82.34 27.44 0.75 87.34 29.11 MIs (estimating WC) 0.74 0.07 15.78 0.45 0.11 23.69 Yu et al. Forest Ecosystems (2021) 8:44 Page 11 of 19 Table 8 Estimation of PWD infection stages using Cab the whole crown. (4) We collected Cab and WC data on estimated by RF based on the REPs derived from ground data 12–18 August 2019, while we acquired the UAV-based Stage H E M S Total data on 18 August 2019. During the interval, the bio- chemical conditions may have changed. Because it only H25 11 0 0 36 took 30–60 min for the drone to complete the data col- E 12 460 22 lection, but the artificial ground survey took at least 1 M 1 12 14 12 39 week. In this study, we cut each tree branch and then S 1 0914 24 measured the spectrum, Cab, and WC of each tree. Total 39 27 29 26 121 Therefore, the workload is relatively heavy, the ground OA (%) 47.11 survey cannot be synchronized with the drone data col- lection, and we can only keep the time as close as pos- Kappa 0.29 sible. (5) The results of our study may be affected by small sample size. Error analysis Previous studies show hyperspectral data to be effective Possible application of UAV-based hyperspectral data in in examining forest health (e.g., Pontius et al. 2008; Näsi detecting PWD et al., 2015). However, we encountered several difficul- Overall, the PWD infection stage classification of ground ties, obstacles, and sources of error in precisely estimat- data was more accurate than that of UAV-based air- ing leaf Cab, WC, and the stage of PWD in pine trees. borne hyperspectral data (Tables 9 and 10). There are (1) The stage of PWD of each sampled pine tree was several possible sources of this discrepancy. Firstly, judged by visual observation. These measurements were ground data consisted of samples from the entire tree fairly subjective and possibly inaccurate. (2) The acquisi- while the airborne data only measured canopy spectral tion of ground and UAV-based hyperspectral data are data. Therefore, ground data samples may more accur- both easily affected by the weather, especially light. Be- ately reflect the tree condition. Additionally, airborne cause data were collected during light hours, this may data acquisition is easily affected by weather – this may have biased results. (3) The results of individual tree have induced measurement errors. PA and UA of the crown segmentation using UAV-based hyperspectral four PWD infection classes using RF based on VIs, REPs, data were somewhat inaccurate. This increased the un- MIs, and combining all variables also suggest ground data certainly of extracting tree hyperspectral features and, performed better than airborne data (Table 11). However, consequently, it was difficult to distinguish pine trees when the RERs and combining all variables were used from understory trees and separate overlapping crowns from UAV-based data, predictions were comparably ac- from HIs using the image classification algorithm. curate to those of ground data (Tables 9 and 10). Manually drawing and visual interpretation can reduce Importantly, the acquisition of airborne data is simple, the interference of mixed pixels, but there was a prob- convenient, and much faster than ground data acquisi- lem that it cannot be efficiently applied when the sample tion. Therefore, there is a trade-off between the accuracy size was large. Nevertheless, in the actual situation, we and efficiency of data acquisition: ground data acquisi- can hardly meet two requirements at the same time: tion is accurate but time consuming to obtain while obtaining pure pixels and those that completely cover UAV-based airborne data is less accurate but much Fig. 8 The mean reflectance values of different disease stages at 350–2500 nm and 400–1000 nm from ground data (a) and UAV-based data (b) Yu et al. Forest Ecosystems (2021) 8:44 Page 12 of 19 Fig. 9 The mean reflectance values of different disease stages in Green, Red Edge, and NIR Fig. 10 Comparison of 11 Vegetation Indices (VIs), Red Edge Parameters (REPs), and Moisture Indices (MIs) at different disease degrees Yu et al. Forest Ecosystems (2021) 8:44 Page 13 of 19 Fig. 11 The mean decrease accuracy (MDA) of each selected variable from ground data (a) and UAV-based (b) data for estimating the disease stage of pine trees easier to obtain. Because PWD potentially affects trees characteristics (e.g. leaf Cab) of healthy trees and trees at in many large forest areas, ground data acquisition is not early stage of PWD are very different from those of trees a feasible management strategy. UAV-based data pro- in middle and serious stage (Fig. 5). In contrast, it is dif- vides only slightly less accurate classifications than ficult to distinguish healthy trees from trees in early ground-based data and is thus a more practical candi- stage of PWD because the difference in their spectral re- date for future large-scale forest management. sponses cannot be detected easily. REPs performed rela- tively well in distinguishing trees in early stage of PWD The potential of identifying trees in the early stage of infection from healthy trees (ground data OA: 80.33%, PWD Kappa: 0.58; and airborne data OA: 71.67%, Kappa: Our results show that it is relatively simple to distin- 0.40); however, overall, UAV-based data yielded moder- guish healthy trees and trees in early stage of PWD in- ately low accuracy (Fig. 13). Therefore, in practical appli- fection from trees in the middle and serious stage of cation, especially in a large-scale forest area, it is still a PWD infection. This is because the biochemical challenge to use UAV-based hyperspectral data to Yu et al. Forest Ecosystems (2021) 8:44 Page 14 of 19 Table 9 Classification confusion matrix of random forest (RF) classifier using vegetation indices, red edge parameters, moisture indices, and combined all variables based on ground spectral data Stage Vegetation Indices Red Edge Parameters H E M S Total UA (%) H E M S Total UA (%) H 29 7 0 0 36 80.56 32 7 0 0 39 82.05 E 8 17 3 0 28 60.71 5 17 1 0 23 73.91 M 1 2 22 3 28 78.57 1 2 24 3 30 80.00 S 11423 29 79.31 1 1 4 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 74.36 62.96 75.86 88.46 OA (%) 75.21 82.05 62.96 82.76 88.46 OA (%) 79.34 Kappa 0.66 Kappa 0.67 Stage Moisture Indices Combined H E M S Total UA (%) H E M S Total UA (%) H 29 9 1 0 39 74.36 32 8 0 0 40 80.00 E 9 15 1 1 26 57.69 5 17 2 0 24 70.83 M 0 3 24 3 30 80.00 2 2 24 2 30 80.00 S 10322 26 84.62 0 0 3 24 27 88.89 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 74.36 55.56 82.76 84.62 OA (%) 74.38 82.05 62.96 82.76 92.31 OA (%) 80.17 Kappa 0.65 Kappa 0.73 precisely identify trees at early infected stage of PWD. In wood borer in pine forest (Abdel-Rahman et al. 2014;Lin conclusion, the main focus of our next study is to im- et al. 2019;Iordache et al. 2020). prove the accuracy by some effective approaches (e.g., Currently, deep learning algorithms, such as convolu- using multi-temporal UAV hyperspectral data). tional neural network (CNN), have been showing its great potential in plant health monitoring (Yuan et al. 2017; Nagasubramanian et al. 2019; Wu et al. 2021). Classification algorithms However, it still has some dependencies. Firstly, when Machine learning algorithms, such as Random forest the data is small, deep learning algorithms do not per- (RF), support vector machine (SVM), Classification and form well. Furthermore, deep learning is like a black Regression Tree (CART), have been widely conducted in box, it does not reveal why it given the result, so it is classifying damaged trees by forest pest in previous stud- lack of interpretability (Ling et al. 2018; Silaparasetty ies (Abdel-Rahman et al. 2014; Iordache et al. 2020; Syifa 2020). On the other side, with its rigorous calculations et al. 2020; Zhan et al. 2020). In our study, RF algorithm and great flexibility (Schmidhuber 2015; Hao et al. was used. 2016), it could improve our classification accuracy. In In RF algorithm, the mean decrease accuracy (MDA) our next study, deep learning algorithms will be index of each variable is determined when calculating employed on PWD diagnose using multi-temporal the out-of-bag (OOB) error, which measures the import- UAV-based hyperspectral data. ance of the variables by comparing how much OOB error of estimate value increases when excluding one vari- The possible application of Lidar able and keeping others unchanged (Archer and Kimes, In this study, the classification model, predictions for 2008; Verikas et al. 2011;Abdel-Rahman et al. 2013). Cab and WC, and the results of individual tree crown Thus, the higher the MDA values of a variable, the greater segmentation were obtained based on hyperspectral data its importance (Immitzer et al. 2012; Liu et al. 2017), we alone. However, the results were not satisfactory, espe- can thereby determine the most important variable. Add- cially the tree crown segmentation (only delineated 121 itionally, compared with other algorithms, RF is more in- from 218). Another potential method of data collection sensitive to multicollinearity, and its results are relatively is Lidar (light detection and ranging). Lidar can directly, robust to missing and unbalanced data, and it can well quickly, and accurately obtain three-dimensional geo- predict the effect of thousands of explanatory variables graphic coordinates of objects (Vierling et al. 2008). (Breiman 2001). Therefore, RF have been widely used in Much progress has been made in the application of monitoring forest disturbance, especially for detecting Lidar technology in the fields of geology, forestry and Yu et al. Forest Ecosystems (2021) 8:44 Page 15 of 19 Table 10 Classification confusion matrix of random forest (RF) classifier using vegetation indices, red edge parameters, moisture indices, and combined all variables based on UAV hyperspectral data Stage Vegetation Indices Red Edge Parameters H E M S Total UA (%) H E M S Total UA (%) H 25 9 1 0 35 71.43 28 8 1 0 37 75.68 E 10 16 2 0 28 57.14 9 15 3 0 27 55.56 M 3 1 22 4 30 73.33 1 2 22 3 28 78.57 S 11422 28 78.57 1 2 3 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 64.10 59.26 75.86 84.62 OA (%) 70.25 71.79 55.56 75.86 88.46 OA (%) 72.73 Kappa 0.60 Kappa 0.63 Stage Moisture Indices Combined H E M S Total UA (%) H E M S Total UA (%) H 24 9 2 1 36 66.67 28 8 0 0 36 77.78 E 10 14 3 2 29 48.28 9 15 2 0 26 57.69 M 3 2 20 4 29 68.97 1 2 24 3 30 80.00 S 22419 27 70.37 1 2 3 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 61.54 51.85 68.97 73.08 OA (%) 63.64 71.79 55.56 82.76 88.46 OA (%) 74.38 kappa 0.51 Kappa 0.66 ecology, such as the establishment of digital elevation tree segmentation could improve accuracy (e.g., Junttila model (DEM), the extraction of forest structure parame- et al. 2019; Lin et al. 2019). Furthermore, crown struc- ters, and the inversion of forest ecosystem function pa- ture and other tree structural information are likely to rameters (e.g., Watt et al. 2014; Huang and Lian 2015; change throughout PWD infection. Therefore, variables Saarela et al. 2020; Xie et al. 2020). based on the return intensity information from Lidar This makes Lidar a possible candidate to improve data might be useful in estimating the stage of PWD in measurement accuracy. Although Lidar data failed to ac- pine trees, and it will be our next study. curately reflect the biochemical condition of tree crowns (e.g., Liu et al. 2017; Shi et al. 2018), it can be used as Conclusion measure auxiliary data that produces three-dimensional In this paper, we compared the relatively accuracies of tree canopy structures (e.g., Shendryk et al. 2016). Thus, using ground-based data and UAV-based hyperspectral combining Lidar with hyperspectral data for individual data in predicting the stage of PWD infection in pine Table 11 Producer’s accuracy (%) and user’s accuracy (%) of the four stages using RF based on vegetation indices, red edge parameters, moisture indices, and combined all variables from ground and UAV-based data Stage Vegetation Indices Red Edge Parameters Moisture Indices Combined Producer’s User’s Producer’s User’s Producer’s User’s Producer’s User’s accuracy accuracy accuracy accuracy accuracy accuracy accuracy accuracy Based on ground spectrum data H 74.36 80.56 82.05 82.05 74.36 74.36 82.05 80.00 E 62.96 60.71 62.96 73.91 55.56 57.69 62.96 70.83 M 75.86 78.57 82.76 80.00 82.76 80.00 82.76 80.00 S 88.46 79.31 88.46 79.31 84.62 84.62 92.31 88.89 UAV-based hyperspectral data H 64.10 71.43 71.79 75.68 61.54 66.67 71.79 77.78 E 59.26 57.14 55.56 55.56 51.85 48.28 55.56 57.69 M 75.86 73.33 75.86 78.57 68.97 68.97 82.76 80.00 S 84.62 78.57 88.46 79.31 73.08 70.37 88.46 79.31 Yu et al. Forest Ecosystems (2021) 8:44 Page 16 of 19 Fig. 12 Producer’s and user’s accuracies for healthy (H), early (E), middle stage (M), and serious stage (S) of disease of pine tree pairs comparison achieved by random forest when the vegetation indices (a), red edge parameters (b), moisture indices (c), and combined all variables (d) based on ground spectrum data trees. To do this, we selected VIs, REPs, MIs, and com- highest accuracy in distinguishing between the healthy bining all variables as input parameters in a RF classifi- trees and trees in early stage of PWD infection. The clas- cation model. We found that combining all variables sification accuracy of REPs based on UAV (airborne) generally perform the best for estimating the stage of data had slightly poorer performance in distinguishing PWD infection of pine trees, and that REPs exhibit the trees at early stage of PWD and healthy trees (OA: Fig. 13 Producer’s and user’s accuracies for healthy (H), early (E), middle stage (M), and serious stage (S) of disease of pine trees pairs comparison achieved by random forest when the vegetation indices (a), red edge parameters (b), moisture indices (c), and combined all variables (d) when the UAV-based hyperspectral data were employed Yu et al. Forest Ecosystems (2021) 8:44 Page 17 of 19 71.67%, Kappa: 0.40), but is still a feasible method. Boochs F, Kupfer G, Dockter K, Kühbauch W (1990) Shape of the red edge as vitality indicator for plants. Int J Remote Sens 11(10):1741–1753. https://doi. Therefore, UAV-based hyperspectral imaging is a prom- org/10.1080/01431169008955127 ising candidate for measuring forest health. Relative to Breiman L (2001) Random forests. Mach Learn 45(1):5–32. https://doi.org/10.1 methods that use ground data, UAV-based hyperspectral 023/a:1010933404324 Carter GA, Miller RL (1994) Early detection of plant stress by digital imaging imaging has the potential to substantially reduce labor within narrow stress-sensitive wavebands. Remote Sens Environ 50(3):295– and time costs. Future studies should aim to improve 302. https://doi.org/10.1016/0034-4257(94)90079-5 the accuracy of UAV-based data. One possible direction Chen JM (1996) Evaluation of vegetation indices and a modified simple ratio for boreal applications. Can J Remote Sens 22(3):229–242. https://doi.org/10.1 is the use of supplemental data acquisition practices 080/07038992.1996.10855178 such as UAV-based Lidar data to improve classification Chen YH (2005) Effects of pine wood nematode (PWN) infection on water accuracy. regime and metabolism of related to hosts. Acta Phytopathol Sin 35(3):201– 207. https://doi.org/10.13926/j.cnki.apps.2005.03.003 (in Chinese) Acknowledgments Congalton RG (1991) A review of assessing the accuracy of classifications of We thank Klaus v. Gadow for research insights and editorial assistance, and remotely sensed data. Remote Sens Environ 37(1):35–46. https://doi.org/10.1 we also thank the anonymous reviewers for helpful suggestions. We also 016/0034-4257(91)90048-B thank all people involved in the assignments within the scope of the project, Curran PJ, Dungan JL, Gholz HL (1990) Exploring the relationship between related to the issues presented in this article. The authors would like to thank reflectance red edge and chlorophyll content in slash pine. Tree Physiol 7(1- TopEdit (www.topeditsci.com) for its linguistic assistance during the 2-3-4):33–48. https://doi.org/10.1093/treephys/7.1-2-3-4.33 preparation of this manuscript. Dawson TP, Curran PJ (1998) A new technique for interpolating the reflectance red edge position. Int J Remote Sens 19(11):2133–2139. https://doi.org/10.1 Authors’ contributions 080/014311698214910 Y.R. designed and conducted this research, analyzed the results, and wrote De Klerk HM, Buchanan G (2017) Remote sensing training in African conservation. the manuscript; L.R. reviewed the manuscript. All authors gave comments Remote Sens Ecol Conserv 3(1):7–20. https://doi.org/10.1002/rse2.36 and approved the final manuscript. Douda O, Zouhar M, Maňasová M, Dlouhý M, Lišková J, Ryšánek P (2015) Hydrogen cyanide for treating wood against pine wood nematode Funding (Bursaphelenchus xylophilus): results of a model study. J Wood Sci 61(2):204– This research was funded by the National Key Research & Development 210. https://doi.org/10.1007/s10086-014-1452-9 Program of China (2018YFD0600200), Beijing’s Science and Technology Du HQ, Ge HL, Fan YW, Jin W, Li J (2009) Application of fractal theory in Planning Project (Z191100008519004) and Major emergency science and hyperspectral detecting the early stage of pine wood nematode disease technology projects of National Forestry and Grassland Administration (Bursaphelenchus xylophilus)of Pinus massoniana with Hyperspectrum. Sci Silv (ZD202001–05). Sin 45(6):68–76. https://doi.org/10.3321/j.issn:1001-7488.2009.06.012 (in Chinese) Availability of data and materials Franklin SE, Wulder MA, Skakun RS, Carroll AL (2003) Mountain pine beetle red- The data are available upon a reasonable request to the Authors. attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogr Eng Remote Sens 69:283–288. https://doi.org/1 0.14358/pers.69.3.283 Declarations Gao B-C (1996) NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens Environ 58(3):257–266. Ethics approval and consent to participate https://doi.org/10.1016/s0034-4257(96)00067-3 Not applicable. Hao X, Zhang G, Ma S (2016) Deep learning. Int J Semantic Comput 10(3):417– Consent for publication 439. https://doi.org/10.1142/S1793351X16500045 Not applicable. Hardisky MA, Smart RM, Klemas V (1983) Seasonal spectral characteristics and aboveground biomass of the tidal marsh plant, Spartina alterniflora. Photogramm Eng Remote Sens 49:85–92 Competing interests He KS, Bradley BA, Cord AF, Rocchini D, Tuanmu M-N, Schmidtlein S, Turner W, The authors declare that they have no competing interests. Wegmann M, Pettorelli N (2015) Will remote sensing shape the next generation of species distribution models? Remote Sens Ecol Conserv 1(1):4– Received: 24 December 2020 Accepted: 5 April 2021 18. https://doi.org/10.1002/rse2.7 Hicke JA, Logan J (2009) Mapping whitebark pine mortality caused by a mountain pine beetle outbreak with high spatial resolution satellite imagery. References Int J Remote Sens 30(17):4427–4441. https://doi.org/10.1080/01431160802 Abdel-Rahman EM, Ahmed FB, Ismail R (2013) Random forest regression and spectral band selection for estimating sugarcane leaf nitrogen concentration Horler DNH, Barber J, Barringer AR (1980) Effects of heavy metals on the using EO-1 Hyperion hyperspectral data. Int J Remote Sens 34(2):712–728. absorbance and reflectance spectra of plants. Int J Remote Sens 1(2):121– https://doi.org/10.1080/01431161.2012.713142 136. https://doi.org/10.1080/01431160108559256 Abdel-Rahman EM, Mutanga O, Adam E, Ismail R (2014) Detecting Sirex noctilio Huang BH (2020) Monitoring Bursaphelenchus xylophilus with multispectrum grey-attacked and lightning-struck pine trees using airborne hyperspectral camera in UAV. Guangxi Forest Sci 49(3):380–384. https://doi.org/10.19692/j. data, random forest and support vector machines classifiers. ISPRS J cnki.gfs.2020.03.012 (in Chinese) Photogrammd Remote Sens 88:48–59. https://doi.org/10.1016/j.isprsjprs.2 Huang H, Lian J (2015) A 3D approach to reconstruct continuous optical images 013.11.013 using lidar and MODIS. Forest Ecosyst 2(1):20. https://doi.org/10.1186/s40663- Ahmed N, Atzberger C, Zewdie W (2020) Integration of remote sensing and 015-0044-5 bioclimatic data for prediction of invasive species distribution in data-poor regions: a review on challenges and opportunities. Environ Syst Res 9(1):32. Huang HH, Ma XH, Huang HY, Zhou YF, Zhang W, Huang YH (2018) A https://doi.org/10.1186/s40068-020-00195-0 preliminary study on monitoring of dead pine trees caused by pine wilt Archer KJ, Kimes RV (2008) Empirical characterization of random forest variable disease with fixed-wing unmanned aerial vehicle. J Environ Entomol 40(2): importance measures. Comput Stat Data Ann 52(4):2249–2260. https://doi. 306–313. https://doi.org/10.3969/j.issn.1674-0858.2018.02.9 (in Chinese) org/10.1016/j.csda.2007.08.015 Huang MX, Gong JH, Li S, Zhang B, Hao QT (2012) Study on pine wilt disease Blackburn GA (1998) Quantifying chlorophylls and caroteniods at leaf and canopy hyper-spectral time series and sensitive features. Remote Sens Technol Appl scales: an evaluation of some hyperspectral approaches. Remote Sens 27(6):954–960. https://doi.org/10.11873/j.issn.1004-0323.2012.6.954 (in Environ 66(3):273–285. https://doi.org/10.1016/S0034-4257(98)00059-5 Chinese) Yu et al. Forest Ecosystems (2021) 8:44 Page 18 of 19 Hui J (2018) Damage and control measures of pine wilt disease. China South Mullen KE (2016) Early detection of mountain pine beetle damage in ponderosa Agric Machine 49(4):100. https://doi.org/10.3969/j.issn.1672-3872.2018.04.087 pine forests of the black hills using hyperspectral and WorldView-2 data. (in Chinese) Minnesota State University, Mankato, pp 58–60 Hunt ER, Rock BN (1989) Detection of changes in leaf water content using near- Mutanga O, Adam E, Cho MA (2012) High density biomass estimation for and middle-infrared reflectances. Remote Sens Environ 30(1):43–54. https:// wetland vegetation using WorldView-2 imagery and random forest doi.org/10.1016/0034-4257(89)90046-1 regression algorithm. Int J Appl Earth Observ Geoinform 18:399–406. https:// doi.org/10.1016/j.jag.2012.03.012 Hyun MW, Kim JH, Suh DY, Lee SK, Kim SH (2007) Fungi isolated from pine wood nematode, its vector Japanese pine sawyer, and the nematode-infected Nagasubramanian K, Jones S, Singh AK, Sarkar S, Singh A, Ganapathysubramanian Japanese black pine wood in Korea. Mycobiology 35(3):159–161. https://doi. B (2019) Plant disease identification using explainable 3D deep learning on org/10.4489/MYCO.2007.35.3.159 hyperspectral images. Plant Methods 15(1):98. https://doi.org/10.1186/s13007- Immitzer M, Atzberger C, Koukal T (2012) Tree species classification with random 019-0479-8 forest using very high spatial resolution 8-band worldview-2 satellite data. Näsi R, Honkavaara E, Lyytikäinen-Saarenmaa P, Blomqvist M, Litkey P, Hakala T, Remote Sens 4(9):2661–2693. https://doi.org/10.3390/rs4092661 Viljanen N, Kantola T, Tanhuanpää T, Holopainen M (2015) Using UAV-based Inoue Y, Sakaiya E, Zhu Y, Takahashi W (2012) Diagnostic mapping of canopy photogrammetry and hyperspectral imaging for mapping bark beetle nitrogen content in rice based on hyperspectral measurements. Remote damage at tree-level. Remote Sens 7(11):15467–15493. https://doi.org/10.33 Sens Environ 126:210–221. https://doi.org/10.1016/j.rse.2012.08.026 90/rs71115467 Iordache M-D, Mantas V, Baltazar E, Pauly K, Lewyckyj N (2020) A machine National Forestry Administration (2018) National Forestry Administration learning approach to detecting pine wilt disease using airborne spectral Announcement No.1 in 2018. http://www.forestry.gov.cn/main/3457/201802 imagery. Remote Sens 12(14):2280. https://doi.org/10.3390/rs12142280 07/1074330.html. Accessed 16 Oct 2020 Pan CS (2011) Development of studies on pinewood nematodes diseases. J Jones HG, Vaughan RA (2010) Remote sensing of vegetation. Oxford University Xiamen Univ 50(2):476–483 (in Chinese) Press, New York, pp 169–171 Jung KY, Park JK (2019) Analysis of vegetation infection information using Pan J, Ju YW, Wang XT, Zhang H (2014) Detection of Bursaphelenchus xylophilus unmanned aerial vehicle with optical sensor. Sensor Material 31:3319–3326. infection in Pinus massoniana from hyperspectral data. Nematology 16(10): https://doi.org/10.18494/sam.2019.2465 1197–1207. https://doi.org/10.1163/15685411-00002846 Junttila S, Holopainen M, Vastaranta M, Lyytikäinen-Saarenmaa P, Kaartinen H, Pan L, Li YX, Liu ZK, Meng FL, Chen J, Zhang XY (2019) Isolation and Hyyppä J, Hyyppä H (2019) The potential of dual-wavelength terrestrial lidar identification of pine wood nematode in Pinus koraiensis in Fengcheng, in early detection of Ips typographus (L.) infestation – leaf water content as a Liaoning Province. Forest Pest Dis 38(1):1–4. https://doi.org/10.19688/j.cnki. proxy. Remote Sens Environ 231:111264. https://doi.org/10.1016/j.rse.201 issn1671-0886.20180021 (in Chinese) 9.111264 Penuelas J, Pinol J, Ogaya R, Filella I (1997) Estimation of plant water concentration by the reflectance water index WI (R900/R970). Int J Remote Kim S-R, Lee W-K, Lim C-H, Kim H, Kafatos MC, Lee S-H, Lee S-S (2018) Sens 18(13):2869–2875. https://doi.org/10.1080/014311697217396 Hyperspectral analysis of pine wilt disease to determine an optimal detection index. Forests 9(3):115. https://doi.org/10.3390/f9030115 Pontius J, Martin M, Plourde L, Hallett R (2008) Ash decline assessment in Kuai CL (2012) Occurrence and control of pine wilt disease. Modern Agric Sci emerald ash borer-infested regions: a test of tree-level, hyperspectral Technol 18:123–123. https://doi.org/10.3969/j.issn.1007-5739.2012.18.079 (in technologies. Remote Sens Environ 112(5):2665–2676. https://doi.org/10.101 Chinese) 6/j.rse.2007.12.011 Li H, Xu HH, Zheng HY, Chen XY (2020) Monitoring technology of pine wilt Prasad AM, Iverson LR, Liaw A (2006) Newer classification and regression tree disease based on UAV remote sensing image. J Chin Agric Mechan 41(9): techniques: bagging and random forests for ecological prediction. 170–175. https://doi.org/10.13733/j.jcam.issn.2095-5553.2020.09.027 (in Ecosystems 9(2):181–199. https://doi.org/10.1007/s10021-005-0054-1 Chinese) Richardson AJ, Wiegand CL (1977) Distinguishing vegetation from soil background information. Photogram Eng Remote Sens 43(12):1541–1552 Li H, Zhou GY, Liu JN, Zhang HY (2011) Study on pine wilt disease and its control situation. Appl Mechd Mater 55–57:567–572. https://doi.org/10.4028/www. Saarela S, Wästlund A, Holmström E, Mensah AA, Holm S, Nilsson M, Fridman J, scientific.net/amm.55-57.567 Ståhl G (2020) Mapping aboveground biomass and its prediction uncertainty Li ZX, Zhang XL, Zhang SC, Yan XS (2004) Application of RS and GIS in using LiDAR and field data, accounting for tree-level allometric and LiDAR monitoring forest disease and insect pests. Hebei J For Orchard Res 19(4): model errors. Forest Ecosyst 7(1):43. https://doi.org/10.1186/s40663-020-0024 377–380. https://doi.org/10.3969/j.issn.1007-4961.2004.04.017 (in Chinese) 5-0 Lin Q, Huang H, Wang J, Huang K, Liu Y (2019) Detection of pine shoot beetle Santos CS, de Vasconcelos MW (2012) Identification of genes differentially (PSB) stress on pine forests at individual tree level using UAV-based expressed in Pinus pinaster and Pinus pinea after infection with the pine hyperspectral imagery and lidar. Remote Sens 11(21):2540. https://doi.org/1 wood nematode. Eur J Plant Pathol 132(3):407–418. https://doi.org/10.1007/ 0.3390/rs11212540 s10658-011-9886-z Lin XZ (2015) Review on damage and control measures of pine wilt disease. East Santoso H, Tani H, Wang X (2016) A simple method for detection and counting China Forest Manag 29(3):28–30. https://doi.org/10.3969/j.issn.1004-7743.201 of oil palm trees using high-resolution multispectral satellite imagery. Int J 5.03.008 (in Chinese) Remote Sens 37(21):5122–5134. https://doi.org/10.1080/01431161.2016.122 Ling G, Lei G, Elmadany NEI, Liang C (2018) Statistical machine learning vs deep Schlemmer M, Gitelson A, Schepers J, Ferguson R, Peng Y, Shanahan J, Rundquist learning in information fusion: competition or collaboration? In: 2018 IEEE D (2013) Remote estimation of nitrogen and chlorophyll contents in maize conference on multimedia information processing and retrieval. MIPR, Miami, at leaf and canopy levels. Int J Appl Earth Observ Geoinform 25:47–54. pp 251–256. https://doi.org/10.1109/MIPR.2018.00059 https://doi.org/10.1016/j.jag.2013.04.003 Liu L, Coops NC, Aven NW, Pang Y (2017) Mapping urban tree species using integrated airborne hyperspectral and LiDAR remote sensing data. Remote Schmidhuber J (2015) Deep learning in neural networks: an overview. Neural Sens Environ 200:170–182. https://doi.org/10.1016/j.rse.2017.08.010 Netw 61:85–117. https://doi.org/10.1016/j.neunet.2014.09.003 Liu W, Chang QR, Guo M, Xing DX, Yuan YS (2010) Monitoring of leaf nitrogen Shen GR, Wang RC (2001) Review of the application of vegetation remote content in summer corn with first derivative of spectrum based on modified sensing. J Zhejiang Univ (Agric Life Sci) 27(6):682–690. https://doi.org/1 red edge. J Northwest A F Univ 38(4):91–98. https://doi.org/10.13207/j.cnki. 0.3321/j.issn:1008-9209.2001.06.025 (in Chinese) jnwafu.2010.04.018 (in Chinese) Shendryk I, Broich M, Tulbure MG, McGrath A, Keith D, Alexandrov SV (2016) Mapping individual tree health using full-waveform airborne laser scans and Ma J, Liu W, Zhang XL (2011) Remote sensing research on early monitoring and imaging spectroscopy: a case study for a floodplain eucalypt forest. Remote prediction of pine wilt disease. Forest Invent Plan 36(5):75–80. https://doi. Sens Environ 187:202–217. https://doi.org/10.1016/j.rse.2016.10.014 org/10.3969/j.issn.1671-3168.2011.05.018 (in Chinese) Mamiya Y (1988) History of pine wilt disease in Japan. J Nematol 20(2):219–226 Shi Y, Skidmore AK, Wang T, Holzwarth S, Heiden U, Pinnel N, Zhu X, Heurich M Meddens AJH, Hicke JA, Vierling LA (2011) Evaluating the potential of (2018) Tree species classification using plant functional traits from LiDAR and multispectral imagery to map multiple stages of tree mortality. Remote Sens hyperspectral data. Int J Appl Earth Observ Geoinform 73:207–219. https:// Environ 115(7):1632–1642. https://doi.org/10.1016/j.rse.2011.02.018 doi.org/10.1016/j.jag.2018.06.018 Yu et al. Forest Ecosystems (2021) 8:44 Page 19 of 19 Shin S-C (2008) Pine wilt disease in Korea. In: Zhao BG, Futai K, Sutherland JR, Spectrosc Spectr Anal 31(5):1352–1356. https://doi.org/10.3964/j.issn.1000- Takeuchi Y (eds) Pine wilt disease. Springer, Tokyo. https://doi.org/10.1007/ 0593(2011)05-1352-05 (in Chinese) 978-4-431-75655-2_5 Yang BJ (2002) Advance in research of pathogenetic mechanism of pine wood Silaparasetty N (2020) Machine learning vs. deep learning. In: Silaparasetty N (ed) nematode. Forest Pest Disease 1:27–31 (in Chinese) Machine learning concepts with Python and the Jupyter notebook Yao FQ, Zhang ZH, Yang RY, Sun JW, Cui SF (2009) Hyperspectral models for environment. Using Tensorflow 2.0. Apress, Berkeley, pp 57–65. https://doi. estimating vegetation chlorophyll content based on red edge parameter. org/10.1007/978-1-4842-5967-2_4 Trans Chin Soc Agric Eng 25(Sup2):123–129 (in Chinese) Ye JR (2019) Epidemic status of pine wilt disease in China and its prevention and Song QL, Xiang R, Qing L, Yuan L, Tian H, Zhu PB, Liu W, Yang W, Qu YX, Zhou control techniques and counter measures. Sci Silv Sin 55(9):1–10 (in Chinese) JW (2018) Relationship between spectral reflectance characteristic Yu HY, Wu H (2018) Discovery of new host plants and new vector insects of parameters and water content of pine leaves. Sci Technol Innov 03:26–28. Bursaphelenchus xylophilus in Liaoning Province. Forest Pest Dis 37(5):61 (in https://doi.org/10.15913/j.cnki.kjycx.2018.03.026 (in Chinese) Chinese) Syifa M, Park S-J, Lee C-W (2020) Detection of the pine wilt disease tree Yu HY, Wu H, Zhang XD, Wang LM, Zhang XF, Song YS (2019) Preliminary study candidates for drone remote sensing using artificial intelligence techniques. on Larix spp. infected by Bursaphelenchus xylophilus in natural environment. Engineering 6(8):919–926. https://doi.org/10.1016/j.eng.2020.07.001 Forest Pest Dis 38(4):7–10. https://doi.org/10.19688/j.cnki.issn1671-0886.201 Tang L, Shao G (2015) Drone remote sensing for forestry research and practices. J 80024 (in Chinese) For Res 26(4):791–797. https://doi.org/10.1007/s11676-015-0088-y Yuan H, Yang G, Li C, Wang Y, Liu J, Yu H, Feng H, Xu B, Zhao X, Yang X (2017) Tao H, Li C, Zhao D, Deng S, Hu H, Xu X, Jing W (2020) Deep learning-based Retrieving soybean leaf area index from unmanned aerial vehicle dead pine trees detection from unmanned aerial vehicle images. Int J Remote hyperspectral remote sensing: analysis of RF, ANN, and SVM regression Sens 41(21):8238–8255. https://doi.org/10.1080/01431161.2020.1766145 models. Remote Sens 9(4):309. https://doi.org/10.3390/rs9040309 Umebayashi T, Yamada T, Fukuhara K, Endo R, Kusumoto D, Fukuda K (2017) In Yuan J, Wang DL, Li R (2013) Remote sensing image segmentation by combining situ observation of pinewood nematode in wood. Eur J Plant Pathol 147(2): spectral and texture features. IEEE T Geosci Remote 52(1):16–24. https://doi. 463–467. https://doi.org/10.1007/s10658-016-1013-8 org/10.1109/TGRS.2012.2234755 Verikas A, Gelzinis A, Bacauskiene M (2011) Mining data with random forests: a Zhan Z, Yu L, Li Z, Rem L, Gao B, Wang L, Luo Y (2020) Combining GF-2 and survey and results of new tests. Pattern Recogn 44(2):330–349. https://doi. Sentinel-2 images to detect tree mortality caused by red turpentine beetle org/10.1016/j.patcog.2010.08.011 during the early outbreak stage in North China. Forests 11(2):172. https://doi. Vicente C, Espada M, Vieira P, Mota M (2012) Erratum to: pine wilt disease: a org/10.3390/f11020172 threat to European forestry. Eur J Plant Pathol 133(2):497. https://doi.org/10.1 Zhang N, Zhang X, Yang G, Zhu C, Huo L, Feng H (2018) Assessment of 007/s10658-012-9979-3 defoliation during the Dendrolimus tabulaeformis Tsai et Liu disaster outbreak Vierling KT, Vierling LA, Gould WA, Martinuzzi S, Clawges RM (2008) Lidar: using UAV-based hyperspectral images. Remote Sens Environ 217:323–339. shedding new light on habitat characterization and modeling. Front Ecol https://doi.org/10.1016/j.rse.2018.08.024 Environ 6(2):90–98. https://doi.org/10.1890/070001 Wang Z, Zhang X, An S (2007) Spectral characteristics analysis of Pinus Massoniana suffered by Bursaphelenchus Xylophilus. Remote Sens Technol Appl 22(3):367–370. https://doi.org/10.3969/j.issn.1004-0323.2007.03.012 (in Chinese) Waske B, Benediktsson JA, Árnason K, Sveinsson JR (2009) Mapping of hyperspectral AVIRIS data using machine-learning algorithms. Can J Remote Sens 35(Sup1):S106–S116. https://doi.org/10.5589/m09-018 Watt MS, Meredith A, Watt P, Gunn A (2014) The influence of LiDAR pulse density on the precision of inventory metrics in young unthinned Douglas-fir stands during initial and subsequent LiDAR acquisitions. NZ J Forestry Sci 44(1):18. https://doi.org/10.1186/s40490-014-0018-3 White JC, Coops NC, Hilker T, Wulder MA, Carroll AL (2007) Detecting mountain pine beetle red attack damage with EO-1 Hyperion moisture indices. Int J Remote Sens 28(10):2111–2121. https://doi.org/10.1080/01431160600944028 White JC, Wulder MA, Brooks D, Reich R, Wheate RD (2005) Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sens Environ 96(3–4):340–351. https://doi.org/10.1 016/j.rse.2005.03.007 Wu B, Liang A, Zhang H, Zhu T, Zou Z, Yang D, Tang W, Li J, Su J (2021) Application of conventional UAV-based high-throughput object detection to the early diagnosis of pine wilt disease by deep learning. Forest Ecol Manag 486:118986. https://doi.org/10.1016/j.foreco.2021.118986 Wu CY, Niu Z (2008) Sensitivity study of photochemical reflectance index to leaf biochemical components. J Univ Chin Acad Sci 3:346–354 (in Chinese) Xiang R, Yuan L, Qin L, Song Q, Zhang J, Qu Y, Zhou J (2018) Correlation analysis between spectral characteristic parameters and chlorophyll content of pine needles. Sci Technol Innov 35:5–7 (in Chinese) Xie Q, Huang W, Liang D, Chen P, Wu C, Yang G, Zhang J, Huang L, Zhang D (2014) Leaf area index estimation using vegetation indices derived from airborne hyperspectral images in winter wheat. IEEE J-STARS 7(8):3586–3594. https://doi.org/10.1109/JSTARS.2014.2342291 Xie Y, Zhang J, Chen X, Pang S, Zeng H, Shen Z (2020) Accuracy assessment and error analysis for diameter at breast height measurement of trees obtained using a novel backpack LiDAR system. Forest Ecosyst 7(1):33. https://doi.org/1 0.1186/s40663-020-00237-0 Xu HC, Luo YQ, Zhang Q (2012) Changes in water content, pigments and antioxidant enzyme activities in pine needles of Pinus thunbergii and Pinus massoniana affected by pine wood nematode. Sci Silv Sin 11:140–143 (in Chinese) Xu HC, Luo YQ, Zhang TT, Shi YJ (2011) Changes of reflectance spectra of pine needles in different stage after being infected by pine wood nematode. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png "Forest Ecosystems" Springer Journals

Early detection of pine wilt disease in Pinus tabuliformis in North China using a field portable spectrometer and UAV-based hyperspectral imagery

Yu, Run; Ren, Lili; Luo, Youqing
"Forest Ecosystems" , Volume 8 (1) – Jul 5, 2021
Download PDF
19 pages

Loading next page...
 
/lp/springer-journals/early-detection-of-pine-wilt-disease-in-pinus-tabuliformis-in-north-PIHoTDtNQC

References (96)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2021
eISSN
2197-5620
DOI
10.1186/s40663-021-00328-6
Publisher site
See Article on Publisher Site

Abstract

Background: Pine wilt disease (PWD) is a major ecological concern in China that has caused severe damage to millions of Chinese pines (Pinus tabulaeformis). To control the spread of PWD, it is necessary to develop an effective approach to detect its presence in the early stage of infection. One potential solution is the use of Unmanned Airborne Vehicle (UAV) based hyperspectral images (HIs). UAV-based HIs have high spatial and spectral resolution and can gather data rapidly, potentially enabling the effective monitoring of large forests. Despite this, few studies examine the feasibility of HI data use in assessing the stage and severity of PWD infection in Chinese pine. Method: To fill this gap, we used a Random Forest (RF) algorithm to estimate the stage of PWD infection of trees sampled using UAV-based HI data and ground-based data (data directly collected from trees in the field). We compared relative accuracy of each of these data collection methods. We built our RF model using vegetation indices (VIs), red edge parameters (REPs), moisture indices (MIs), and their combination. Results: We report several key results. For ground data, the model that combined all parameters (OA: 80.17%, Kappa: 0.73) performed better than VIs (OA: 75.21%, Kappa: 0.66), REPs (OA: 79.34%, Kappa: 0.67), and MIs (OA: 74.38%, Kappa: 0.65) in predicting the PWD stage of individual pine tree infection. REPs had the highest accuracy (OA: 80.33%, Kappa: 0.58) in distinguishing trees at the early stage of PWD from healthy trees. UAV-based HI data yielded similar results: the model combined VIs, REPs and MIs (OA: 74.38%, Kappa: 0.66) exhibited the highest accuracy in estimating the PWD stage of sampled trees, and REPs performed best in distinguishing healthy trees from trees at early stage of PWD (OA: 71.67%, Kappa: 0.40). Conclusion: Overall, our results confirm the validity of using HI data to identify pine trees infected with PWD in its early stage, although its accuracy must be improved before widespread use is practical. We also show UAV-based data PWD classifications are less accurate but comparable to those of ground-based data. We believe that these results can be used to improve preventative measures in the control of PWD. Keywords: Pine wilt disease, Remote sensing, Spectrometer, Hyperspectral imaging, Random forest, Classification * Correspondence: youqingluo@126.com Key Laboratory for Forest Pest Control, College of Forestry, 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/. Yu et al. Forest Ecosystems (2021) 8:44 Page 2 of 19 Background content. Therefore, water and chlorophyll content could The pine wood nematode (PWN; Bursaphelenchus xylo- be used as indicators of PWD (Huang 2020). This is im- philus) is a hazardous invasive species that infests mul- portant because RS and HRS methods can be used to es- tiple species of pine (Vicente et al. 2012; Douda et al. timate water and chlorophyll content. For example, 2015). Pine wilt disease (PWD), caused by the PWN, is using a field portable spectrometer to measure the spec- widespread throughout East Asia (Mamiya 1988; Hyun tral characteristics of P. thunbergii and P. massoniana at et al. 2007;Ye 2019). Previously isolated to southern different stages of PWN infection, Xu et al. (2011) found China, PWD is now found throughout the country, in- the reflectance spectrum curve in the mid-infrared band cluding Northeast China (Pan et al. 2019; Yu et al. may indicate the early stage of PWD with the analysis of 2019). In 2016, PWD first appeared in Dalian, Liaoning the spectral characteristic parameters and changes in Province, then in May 2017, it happened in Dandong chlorophyll levels. Similarly, Xiang et al. (2018) used a City, Fushun City, Benxi City and other places (National field portable spectrometer, analyzing the relationship Forestry Administration 2018). In addition, Monocha- between spectral properties and chlorophyll, showing mus saltuarius was identified as a new vector of PWD in that the chlorophyll content of pine decreases with the Liaoning Province of China (Yu et al. 2018). In the stage of PWD (later, more severe stages are associated process of spreading northwards, PWD has infected and with lower chlorophyll content). In addition, the position caused severe damage to the Chinese pine (Pinus tabu- of red edge, the wavelength of red edge, the height of laeformis), Korean pine (P. koraiensis), and larch (Larix green peak, and the depth of red band absorption all spp.) populations. This has resulted in significant eco- strongly correlate with chlorophyll content (e.g., Xiang nomic losses and ecological damage to Chinese pine for- et al. 2018). Correspondingly, the area surrounded by ests (e.g., Li et al. 2011; Lin 2015; Hui 2018). the first-order differential spectrum in the 490–530 nm To effectively control PWD, it is necessary to identify range and that in the 680–760 nm range was found to infected trees in the early stage of infection. This is a dif- be a significant hyperspectral feature indicating the oc- ficult task because most trees progress from initial infec- currence of PWD (e.g., Huang et al. 2012). These studies tion to the serious infection stage within 5 weeks all used a field portable spectrometer, which cannot be (Umebayashi et al. 2017). Consequently, current PWD applied in a large-scale area. management strategies emphasize the control of infected Past studies used satellite imagery such as Landsat, trees after the onset of an outbreak by means of fumiga- IKONOS, Quick Bird, and GF-2 images to detect forest tion, burning, and tree felling (Shin 2008; Kim et al. pest disease (e.g., Franklin et al. 2003; White et al. 2005; 2018). What is lacking is a methodology that monitors Hicke and Logan 2009; Zhan et al. 2020). However, due pine populations that can quickly and efficiently detect to limitations in spatial, temporal, spectral resolution as the early signs of PWD (Ma et al. 2011). In addition, well as weather complications, satellite imagery cannot many efforts have been made in early detection of PWD obtain real-time data (Santoso et al. 2016). Because of (Kim et al. 2018; Syifa et al. 2020; Tao et al. 2020), but these limitations, the detection scale of forest pest dis- not in Chinese pine. In this paper, we present a method ease has shifted to Unmanned Airborne Vehicle (UAV) aimed at detecting PWD in Chinese pine in early stage. remote sensing, which offers the advantages of low con- A major obstacle in the management of pines infected sumables and operating costs, high ground resolution by PWD is that the forests they persist in are very large data collection, and more precise accuracy (Tang et al. communities. This can make classical ground identifica- 2015). For example, Huang et al. (2018) used a fixed- tion and sampling methods impractical. To solve this wing unmanned aerial vehicle to monitor dead pine problem, recent studies have used remote sensing (RS) trees caused by PWD, successfully monitoring pine tree to examine the impact of PWD on the physiological and mortality with over 80% accuracy. Li et al. (2020) used biochemical changes after infection (e.g., Shen et al. UAVs to acquire remote sensing images of forest areas 2001; Li et al. 2004; Wang et al. 2007). Advancements in to assess the presence of PWD, successfully recognizing RS technology increasingly support the prediction effi- infection with 90.4% accuracy. Huang (2020) used UAV ciency by reducing inherent spatial and temporal con- multispectral data to draw a conclusion that the first de- straints (Ahmed et al. 2020). Similarly, hyperspectral rivative of healthy and infected P. thunbergii changed remote sensing (HRS) can obtain continuous spectral in- markedly at 710 nm. Except RGB and multispectral cam- formation of objects – this has been used to detect era, hyperspectral imagery was also applied in detecting changes in the spectral characteristics of needles on forest pest diseases. Abdel-Rahman et al. (2014) used infected-trees in the process of discoloration caused by airborne hyperspectral data, random forest and support PWN infection (Pan 2011;Kuai 2012). vector machines classifiers to distinguish amongst Previous studies show the presence of PWD is signifi- healthy, Sirex noctilio grey-attacked and lightning- cantly linearly correlated with water and chlorophyll damaged pine trees. Zhang et al. (2018) utilized the Yu et al. Forest Ecosystems (2021) 8:44 Page 3 of 19 ISIC-SPA-P-PLSR framework based on UAV-based and serious stages) of PWD infection in Chinese pine hyperspectral image to identify the degree of damage trees. Additionally, analyses of PWD simultaneously trees caused by Dendrolimus tabulaeformis. Iordache considering ground and UAV-based hyperspectral data et al. (2020) acquired airborne multispectral and hyper- have not been widely conducted. spectral data, and used Random Forest algorithms to Therefore, to fill this gap, in this study, our objective is compare the classification accuracies of the two datasets to explore the capacity of ground and airborne hyper- in detecting PWD, finding that both datasets performed spectral data using VIs, REPs, and MIs to classify the well in identifying the infected, suspicious, and healthy stage of PWD infection in Chinese pine at the tree level. trees. Importantly, however, in detecting the PWD, most Furthermore, we also aim to provide a useful and fairly studies focus on distinguishing between healthy and in- accurate method of distinguishing between trees in the fected trees using RGB (red, green and blue bands) cam- early stage of PWD infection from healthy trees. era, multispectral data, and ground hyperspectral data, but UAV-based hyperspectral data were not widely stud- Materials and methods ied. In addition, few studies emphasized the identifica- Study area and ground survey tion of trees in each stage of PWD infection in Chinese We conducted our study in Cangshi Village, located in pine, which we focus on in this study. In our study, we Fushun County, Liaoning Province, in northeastern systematically divided the infection stage into four China (124°21′–124°24′ E, 41°53′–41°57′ N; Fig. 1). In stages, making the detection more accurate. Because the study area, the species of plantation forests is domi- high spatial and spectral resolutions, and feasibility of nated by Chinese pine (P. tabulaeformis), and the age of large-scale area application are needed to distinguish the them is approximately 40–50 years. The total area of for- subtle difference between healthy and the early stage of est cover in Fushun County is approximately 12.43 × 10 infected trees, we consider UAV-based hyperspectral ha, of which P. tabulaeformis makes up > 30%. In imagery. addition, the broadleaf tree species and understory vege- Spectral indices, such as Vegetation indices (VIs), red tation in the study site mainly include, Quercus acutis- edge parameters (REPs), and moisture indices (MIs), can sima, Quercus mongolica, grass, et al. The area is reflect the infection condition of PWD (e.g., Kim et al. situated in the Middle Temperate Zone. It has a contin- 2018; Huang 2020). VIs is a combination of different re- ental monsoon climate and experiences approximately mote sensing spectral bands, which can be regarded as a 804.2 mm of precipitation per year. The mean annual air sign of relative abundance and activity of green vegeta- temperature is approximately 6.6 °C. tion (Jones and Vaughan 2010). Over the past years, VIs According to local Forestry Administration records, had been widely applied to extract sensitive estimates of PWD has resulted in the death of tens of thousands of plant biochemical characteristics (e.g., He et al. 2015;De pine trees since the onset of outbreaks in 2016 in Liaoning Klerk and Buchanan, 2017), such as the normalized dif- Province (National Forestry Administration 2018). ference vegetation index (NDVI) that decreases with in- Field measurements were conducted in 12–18 August creasing tree PWD stage severity (Kim et al. 2018), and 2019. We established three 30 m × 30 m plots located the presence of PWD can be detected by calculating the northeast of Cangshi Village (Fig. 1). The coordinates of VIs based on ground, aerial, and satellite data (e.g., the plot boundary and the location of each tree were re- White et al. 2007; Pan et al. 2014; Jung and Park 2019; corded using a handheld differential global positioning Iordache et al. 2020). The REPs are derived from Red system (DGPS, Version S760) with sub-meter accuracy. edge (680–780 nm), which is the most obvious feature of In each plot, we recorded tree growth state including plant spectral curve. As an indicator of plant stress and tree height (H), diameter at breast height (DBH), crown often used to study the growth and health of plants diameter (CD), and PWD infection stage. In addition, we (Boochs et al. 1990; Dawson and Curran 1998), it also measured biochemical parameters: the leaf chlorophyll had been well studied in detecting the PWD (Du et al. content (Cab) and water content (WC) of each tree. Cab 2009; Huang et al. 2012). Additionally, pine trees killed was derived by averaging the Cab of needles from four by PWD by blocking transmission of water (Yang 2002), different directions using a calibrated CCM-300 Chloro- and the water content of pine needles decreased with in- phyll Content Meter. The Cab of seriously damaged creasing PWD infection severity (Chen, 2005). Thus, trees was 0 measured by the CCM-300. Meanwhile, WC changes in MIs (also derived from radiometric data) can of each tree was determined by the fresh weight (FW) be used to detect the presence of trees infected with minus the dry weight (DW) divided by the FW: (WC = PWD (Xu et al. 2012; Song et al. 2018). (FW – DW)/FW × 100%). Finally, a total of 218 pine Although spectral indices were widely used to detect trees (healthy: 76; early stage: 54; middle stage: 47; ser- the PWD, there is no study that yet provides good pa- ious stage: 41) were measured. Summary statistics of rameters to predict each stage (healthy, early, middle, three plots are given in Table 1. Yu et al. Forest Ecosystems (2021) 8:44 Page 4 of 19 Fig. 1 (a) The map of China; (b) The map of Liaoning Province; (c) The location of the study area. The purple and red rectangles represent three field plots locations and UAV hyperspectral flight areas, respectively Additionally, we randomly selected 20 discolored pine were defined as having dark green needles, normal resin trees as samples from each plot, and took them back to secretion, and vigorous growth. “Early stage” trees were the laboratory for testing by Behrman funnel method. defined by slightly yellowed needles, with decreased resin The result showed that they all carried pine wood secretion and grow rates. “Middle stage” trees were de- nematode. fined by yellow-brown needles, wilt, and weak growth. Dry trees with reddish-brown needles were defined as the “Serious stage”. Infected stage division On the basis of previous studies (Xu et al. 2011;Santos and de Vasconcelos, 2012), we combined needle, ground Remote sensing data acquisition and preprocessing tree, and UAV images to categorize PWD infection into Ground spectrum acquisition four stages: (1) Healthy, (2) Early stage, (3) Middle stage, From the ground (physically measuring trees in the field), and (4) Serious stage (Fig. 2). Stages were defined by color we measured the spectrum of sampled trees using ASD of needles, growth vigor, and resin secretion (Table 2). We Field Spec 4 Hi-Res NG (Analytical Spectral Devices, had four people classified each tree, and took the major- Boulder, CO, USA). The spectral range is 350–2500 nm ity’s opinion as final results to reduce subjective errors. Fi- and the spectral resolution is 3 nm in the 350–1000 nm nally, we used the following definitions: “Healthy” trees wavelength range and 6 nm in the 1001–2500 nm Table 1 Statistics of three plots variables (tree numbers = 218) Mean Standard deviation Maximum Minimum Range H (m) 12.66 1.12 14.50 9.60 4.90 DBH (cm) 21.79 5.40 34.06 7.83 26.23 CD (m) 2.13 1.22 7.80 0.80 7.00 −2 Cab (g∙m ) 299.30 165.93 560.00 0.00 560.00 WC (%) 42.13 14.45 68.32 17.25 51.07 H Tree height, DBH Diameter at breast height, CD Crown diameter, Cab Leaf chlorophyll content, WC Water content Yu et al. Forest Ecosystems (2021) 8:44 Page 5 of 19 Fig. 2 Images of needles, ground trees, and unmanned aerial vehicle (UAV) images of pine trees at different PWD infection stages wavelength range. We selected and measured branches 30, on 18 August 2019. The weather was sunny during roughly representative of the average spectrum of each the flight. Standard white board and white tarp were tree. The selected branches were cut from the east, south, placed on the ground within the flying area. The flying west, and north directions from the upper, middle, and height was set at 120 m, the image forward and side − 1 lower layers (Zhang et al. 2018). We calculated the overlaps were set to 50%, and the flight speed is 2 m∙s . spectrum of each sampled tree by averaging the spectrum The imagery consisted of 281 spectral channels (spectral of the selected branches. The ground spectrums were resolution of 2.1 nm) from visible to near infrared (NIR) gathered from 10:00 to 14:00 every day, from August 12 regions (400–1000 nm). Reflectance correction and to August 17. We obtained the ground spectrum for com- radiometric calibration were performed using 3 m car- parison with UAV-based data and auxiliary radiometric pet reference (standard white board) and the Spectronon correction. software. Image geometric corrections were performed using 4 ground control points (GCPs). The positions of UAV-based hyperspectral imagery GCPs were recorded by a DGPS device with sub-meter Hyperspectral Imagery (HI) data were obtained by using accuracy. The ground resolution of HI was produced to a DJI Matrice 600 UAV (DJI, Shenzhen, China) equipped be 0.4 m. a Pika L hyperspectral camera (Resonon, USA). The main parameters of the Pika L are listed in Table 3. Tree crowns extraction from hyperspectral imagery GNSS (Global Navigation Satellite System) and IMU (In- We conducted tree crown segmentations from HI by ertial Measurement Unit) modules are integrated into combining the object-based segmentation method with UAV, and its horizontal and vertical position errors are manually drawing ROIs (regions of interest). First, by approximately 2.0 and 5.0 m, respectively, with an orien- use of ENVI 5.3, we used the object-based segmentation tation precision of approximately 1 degree. The overall method on the HIs using combined spectral and texture UAV-based system is shown in Fig. 3. features to separate trees crowns from the grass back- UAV-based hyperspectral data acquisition was carried ground and shadows (e.g., Yuan et al. 2013). The object- out in the test areas of Cangshi Village from 12:00–12: based segmentation method successfully separated tree Table 2 Classification of infected stages Age of Classification standard Stages of infected trees the stand Healthy Early stage Middle stage Serious stage 40–50 years Color of needles; Growth Trees grow vigorously Needles begin to turn yellow; Most needles turn yellow Trees are dry, and vigor; Resin secretion with dark green needles; resin secretion decreases and brown and wilt, and the needles are all reddish resin secretes normally growth slows down growth is obviously weak brown, but do not fall off Yu et al. Forest Ecosystems (2021) 8:44 Page 6 of 19 Table 3 Main parameters of the Pika L imaging spectrometer Feature selection and prediction model for cab and WC (provided by manufacturer) According to previous study (De Klerk and Buchanan, Parameters Values Parameters Values 2017; Kim et al. 2018; Lin et al. 2019), trees health was highly correlated with biochemical properties (e.g., Cab), Weight 0.6 kg Sampling interval 1.07 nm which also can be precisely estimated using fitting models Digitization 12 bits Spectral resolution 2.10 nm based on spectral indices (Inoue et al. 2012;Schlemmer Wavelength range 400–1000 nm Spectral channels 281 et al. 2013; Xie et al. 2014). In this study, we firstly calcu- lated the Pearson’s correlation coefficient between a num- crowns from the grass background and shadow compo- ber of spectral indices (features in Tables 4, 5 and 6)and nents. However, it was difficult to separate overlapping Cab and WC, PWD infection stages of each tree, respect- crowns. Second, based on the result of object-based seg- ively. In addition, before these variables were selected for mentation, we drew the ROIs manually. We determined constructing regression and classification model to predict the location of every individual sampled trees by use of Cab, WC, and PWD infection stages, we used a stepwise the DGPS information. The ROIs of each tree were regression method to test the multicollinearity between shaped by manually drawing the crown range on the features, eliminating redundant variables. RGB image. Then, the ROIs were added to the prepro- Finally, we selected 5 VIs (NDVI, NDVI [810, 680], cessed HIs, and the spectrum of an individual tree was NDVI [560, 680], RVI, PRI), 5 REPs (λo, Sg, Kg, GH, RD), calculated by averaging the reflectance of the corre- and 5 moisture indices (MSI, WI1, WI2, NDWI, NSII) sponding ROI extracted by ENVI 5.3. The average based on ground spectrum (350–2500 nm). Based on spectrum information of each ROI was used in the sub- UAV hyperspectral data (400–1000 nm), we selected 5 VIs sequent analysis (Fig. 4). Finally, the shadow components (PSI, RVSI1, NDVI, NDVI [810, 450], RVI), 5 REPs (dλb, and overlapped crowns were discarded. Overall, 121 SDr, SDb, SDr-SDb, RD) and 2 MIs with (WI1, WI2). trees (healthy: 39; early stage: 27; middle stage: 29; ser- We estimated the Cab and WC using a RF (Breiman ious stage: 26) were segmented from HI hyperspectral 2001) regression using a bagging method based on the imagery. CART regression tree model. In the regression applica- tion, each tree was built by choosing a random sample and a random set of variables from the training dataset Features extraction by a deterministic algorithm (Mutanga et al. 2012). All In order to eliminate instrument errors and noises, while 121 samples were used for model training, and we then maintaining the original spectral characteristics, a used a 10-fold cross-validation method (Waske et al. Savitzky-Golay filter with 7 points (we tested 3–15 2009) to assess model accuracy. The process of regres- points and finally chose 7 points) was used to smooth sion was conducted using the R package “randomFor- the original spectrums of ground and UAV-based hyper- est”. The coefficient of determination (R ), RMSE (Root spectral data (Mullen 2016). Based on previous research, Mean Square Error), and RRMSE (Relative RMSE) be- we calculated 37 spectral variables including 12 VIs tween measured and estimated values were used to com- (Table 4), 20 REPs showed in Table 5 (Horler et al. pare different indices in predicting the accuracy of Cab 1980; Curran et al. 1990; Yao et al. 2009; Liu et al. 2010), and WC. After selecting the variables which performed and 5 MIs (Table 6) from spectral data. best in predicting the Cab and WC, we used the Cab or Fig. 3 The unmanned aerial vehicle (UAV)-based hyperspectral system with POS, Pika L imaging spectrometer Yu et al. Forest Ecosystems (2021) 8:44 Page 7 of 19 Fig. 4 (a) The original hyperspectral imaging (HI); (b) digital photo of the test area (upper) and hyperspectral image of one sampling plot of the corresponding region (lower); (c) the result of crowns segmentation and (d) the ROIs formed by manual drawing WC estimated by the optimum variables to classify the the higher the MDA value of a variable is, the more im- PWD infection stages directly. portant it is (e.g., Liu et al. 2017; Shi et al. 2018). The selected VIs, REPs, MIs and combining all vari- Classification based on VIs, REPs, and MIs ables were separately input into RF classification model, We then used the selected VIs, REPs, and MIs correlated and the MDA of all selected variables were determined. with Cab and WC to classify trees based on PWD infec- All 121 samples were used for model training. We then tion. We used a RF classification model to assess the infec- used a 10-fold cross-validation method to estimate tion stage of sampled trees. In a RF algorithm, the variable model accuracy. The process of classification was carried importance is a metric of how much the “out-of-bag” out using the R package “randomForest”. The overall ac- (OOB) error of estimate increases due to the removal of a curacy (OA), producer’s accuracy (PA), user’s accuracy single variable from the data (Prasad et al. 2006;Verikas (UA), and Kappa coefficient resulting from confusion et al. 2011). The mean decrease accuracy (MDA) index of matrices (Congalton 1991) were used to evaluate classifi- each variable is obtained when calculating the OOB error: cation accuracy. Kappa coefficient is a popular statistic Table 4 Vegetation indices extracted from spectral data Variables Description Formula Reference SUM Rð760:900Þ=141 - SUM Rð630:900Þ=271 NDVI Normalized difference vegetation index Richardson and Wiegand (1977) SUM Rð760:900ÞþSUM Rð630:900Þ=271 ðR - R Þ 810 450 NDVI (810,450) Normalized difference vegetation index Richardson and Wiegand (1977) NDVI ð810; 450Þ¼ ðR þR Þ 810 450 ðR - R Þ NDVI (810,680) Normalized difference vegetation index 810 680 Richardson and Wiegand (1977) NDVI ð810; 680Þ¼ ðR þR Þ 810 680 ðR - R Þ 560 680 NDVI (560,680) Normalized difference vegetation index Richardson and Wiegand (1977) NDVI ð560; 680Þ¼ ðR þR Þ 560 680 SUM Rð760:900Þ=141 RVI Ratio vegetation index Wu and Niu (2008) SUM Rð630:900Þ=271 DVI Deferent vegetable index DVI = SUM R(760:900)/141 – SUM R(630:900)/271 Chen (1996) R - R 570 531 PRI Photochemical reflectance Index Carter and Miller (1994) PRI ¼ R þR 570 531 MSR Modified simple ratio -1 Blackburn (1998) MSR ¼ sqrtð þ1Þ R695 PSI Plant stress index Hunt and Rock (1989) PSI ¼ RVSI1 Ratio vegetation stress index Hunt and Rock (1989) PVSI1 ¼ RVSI2 Ratio vegetation stress index Hunt and Rock (1989) PVSI2 ¼ PSSR Pigment specific simple ratio Penuelas et al. (1997) PSSR ¼ 635 Yu et al. Forest Ecosystems (2021) 8:44 Page 8 of 19 Table 5 Red Edge parameters extracted from hyperspectral data Type Parameters Description Type Parameters Description Based on original Rg The maximum reflectance in the Based on first dλb The first order differential value spectrum wavelength range of 510–560 nm derivative corresponding to λb Ro The minimum reflectance in the dλr The first order differential value wavelength range of 640–680 nm corresponding to λr λg Wavelength position of Rg SDb The area surrounded by the first-order differential spectrum in the range of 490–530 nm λo Wavelength position of Ro SDr The area surrounded by the first-order differential spectrum in the range of 680–760 nm Sg Skewness of reflectance in the SDr–SDb SDr–SDb wavelength range of 510–560 nm Kg Kurtosis of reflectance in the Others Lwidth Width at half depth of red band wavelength range of 510–560 nm absorption Sr Skewness of reflectance in the Depth672 Absorption depth at 672 nm range of 680–760 nm Kr Kurtosis of reflectance in the Depth560 Absorption depth at 560 nm wavelength range of 680–760 nm Based on first derivative λb Wavelength position of the maximum GH Height of green peak first derivative of reflectance between 490 and 530 nm λr Wavelength position of the maximum RD Depth of red band absorption first derivative of reflectance between 680 and 760 nm for measuring agreement (Meddens et al. 2011). A input. We examined the performance of Cab and WC es- Kappa value from < 0.4 indicates a “poor” agreement, timation of the input parameters using both ground Kappa 0.4–0.8 is defined as having moderate agreement, spectrum data and UAV-based spectral data (Figs. 6 and and Kappa > 0.80 indicates a “strong” agreement. 7). Cab estimation accuracy was slightly greater when Using the overall and individual accuracies for all four using REPs than using VIs for both ground data (REPs: 2 − 2 PWD infection stages, we examined the paired accur- R =0.78, RMSE =82.34 g∙m , RRMSE = 27.44%; VIs: 2 − 2 acies of Healthy, Early stage, Middle stage, and Serious R = 0.74, RMSE = 89.80 g∙m , RRMSE = 29.92%) and 2 − 2 stage pine trees to examine the feasibility of discriminat- UAV-based data (REPs: R = 0.75, RMSE = 87.34 g∙m , 2 − 2 ing between different stages. RRMSE = 29.11%; VIs: R = 0.72, RMSE = 94.11 g∙m , RRMSE = 31.36%). For WC predictions in which MIs were Results used as input parameters, the predictions from ground Estimation of cab and WC data were considerably more accurate than UAV-based Leaf Cab and WC decreased with the severity of PWD data. The results summarized in Table 7. infection (Fig. 5). We estimated the Cab and WC of all It showed that the model tended to overestimate Cab − 2 121 sampled trees using the RF regression model with the below 200 g∙m and underestimate Cab above 300 − 2 three input parameters (VIs, REPs, and MIs) separately g∙m (Fig. 6a and b; Fig. 7a and b), the RF regression Table 6 Moisture indices extracted from spectral data Variables Description Formula Reference R1600 MSI Moisture stress index Gao (1996) MSI ¼ R820 WI1 Water index Hardisky et al. (1983) WI1 ¼ WI2 Water index Hardisky et al. (1983) WI2 ¼ ðR - R Þ 860 1240 NDWI Normalized difference water index Prasad et al. (2006) NDWI ¼ ðR þR Þ 860 1600 ðR - R Þ 860 1600 NDII Normalized difference infrared index Verikas et al. (2011) NDII ¼ ðR þR Þ 860 1600 Yu et al. Forest Ecosystems (2021) 8:44 Page 9 of 19 Fig. 5 The difference of chlorophyll content (Cab) and water content (WC) of all samples at different infected stages model provided unsatisfactory predictions for Cab and stage of PWD. Generally, the spectral variables were sen- WC in pine trees when VIs, REPs, and MIs were taken sitive to changes in biochemical characteristic. as input parameters. In addition, Cab estimated by REPs derived from Comparisons of classifications using different variables ground data (the optimum variables) were chosen to as- from ground and UAV-based data sess the PWD infection stages directly. Finally, the re- The MDA index for the ground data and UAV-based sults showed that using Cab estimated by RF based on data strongly differed among variables. Importance rank- the optimum variables did not perform well in classify- ings indicated REPs to be more important than most VIs ing the PWD infection stages (OA = 47.11%, Kappa = and MIs (Fig. 11). The most important variables were 0.29; Table 8). It means that estimated Cab cannot be REPs, and VIs were generally more important than MIs. directly used to accurately the PWD infection stages. GH was the most important variable for ground data and SDR was the most important variable for UAV- based data. Feature analysis OA (overall accuracy) assessment using the 10-fold The spectral reflectance of trees declined as a function cross-validation method indicated that REPs performed of PWD stage severity (Fig. 8). The difference of spectral best. For ground data REPs yielded an OA of 79.34%, reflectance was obvious near the green peak (500–600 VIs 75.21%, and MIs 74.38%. Combined all variables, it nm), red edge (680–760 nm), and NIR (750–950 nm; yielded an accuracy of 80.17% (Tables 9 and 10). UAV- Fig. 9). VIs, REPs, and MIs exhibited differing responses based data provided less accurate results for all variables: to the severity of infection. While some variables such as 72.73% for REPs, 70.25% for VIs, 63.64% for MIs, and NDVI (810, 680), Kg, and NDWI decreased with the in- 74.38% for combined all variables (Tables 9 and 10). creasing infection stage, others (e.g. MSI, PRI and Sg) Kappa values yielded similar qualitative results for both significantly increased with the increasing of the infec- ground data and UAV-based data. For ground data, tion stage (Fig. 10). Therefore, almost all the selected Kappa was calculated to be 0.67 for REPs, 0.66 for VIs, variables exhibited statistically significant responses to and 0.65 for MIs. For combining all variables, Kappa im- PWD severity, indicating their potential for detecting the proved to 0.73. For UAV-based data, the values of Kappa Fig. 6 Measured vs. estimated chlorophyll content (Cab) and water content (WC) based on ground spectrum using different input parameters: (a) Vegetation Indices; (b) Red edge parameters, and (c) Moisture Indices Yu et al. Forest Ecosystems (2021) 8:44 Page 10 of 19 Fig. 7 Measured vs. estimated chlorophyll content (Cab) and water content (WC) using UAV-based spectrum with different input parameters: (a) Vegetation Indices; (b) Red edge parameters, and (c) Moisture Indices for REPs, VIs, and MIs were 0.63, 0.60, and 0.51, re- (OA: 80.33%, Kappa: 0.58) based on ground data per- spectively. For combining all variables, Kappa again im- formed best when healthy pine trees and early stage of proved (0.66). Therefore, for each data type (ground infected pine trees were compared. REPs and combining data and UAV-based data), REPs yielded the most accur- all variables performed equally well in terms of OA ate results, followed by VIs and MIs respectively. Add- (71.67%) and Kappa (0.40) for UAV-based data. itionally, ground data provided more accurate results than UAV-based data in all cases. Discussion PA (producer’s accuracy) values were high for the In this paper, we employed VIs, REPs, MIs, and combin- middle and serious stage of infection regardless of the ing all variables, to examine the capacity of ground and data source and the parameters used. UA (user’s accur- UAV-based hyperspectral data in PWD infection stages acy) was relatively high for middle and serious stage of estimation at individual tree level. The results reveal that infection, while healthy and early stages had lower UAs combining all variables performed best and yielded a (Table 11). considerably accurate classification with OA of 80.17% Pairwise comparisons of healthy, early stage, middle for ground data and 74.38% for UAV-based data (Tables stage, and serious stage indicated the OAs of all stage 9 and 10). pairs to be considerably greater than 80% in most cases When we look at the capacity of identifying pine trees (Figs. 12 and 13). Lower accuracies resulted when in the early infected stage of PWD, the REPs exhibited healthy pine trees and early stage of infected pine trees the best performance with OA of 80.33% and 71.67% were compared based on the VIs (75.41%), REPs from ground data and UAV-based data, respectively (80.33%), MIs (70.97%), and combined all variables (Figs. 12 and 13). (79.03%) from ground data, as well as VIs (68.33%), REPs Overall, it is understandable that: (1) the REPs are (71.67%), MIs (66.67%), and combined all variables more responsive to stage changes of PWD infection than (71.67%) from UAV-based data. High values of Kappa VIs and MIs, indicating that REPs may be more sensitive were obtained by most pairwise comparisons (Figs. 12 to the biochemical conditions; (2) UAV-based data per- and 13), but not for comparisons between healthy pine formed considerable accuracy in monitoring the PWD trees and pine trees in the early stage of infection for stage at individual tree level, especially REPs, showing its both ground data (VIs: 0.55, REPs: 0.58, MIs: 0.39, com- good accuracy, which were slightly lower than ground bined all variables: 055) and UAV-based data (VIs: 0.35, data and can be applied in a large-scale forest area. REPs: 0.40, MIs: 0.31, combined all variables: 0.40). REPs Table 7 The results of Cab and WC estimation using RF regression based on VIs, REPs, and MIs from ground and UAV hyperspectral data Variable type Cab and WC estimation Ground UAV 2 2 R RMSE RRMSE (%) R RMSE (%) RRMSE (%) VIs (estimating Cab) 0.74 89.80 29.92 0.72 94.11 31.36 REPs (estimating Cab) 0.78 82.34 27.44 0.75 87.34 29.11 MIs (estimating WC) 0.74 0.07 15.78 0.45 0.11 23.69 Yu et al. Forest Ecosystems (2021) 8:44 Page 11 of 19 Table 8 Estimation of PWD infection stages using Cab the whole crown. (4) We collected Cab and WC data on estimated by RF based on the REPs derived from ground data 12–18 August 2019, while we acquired the UAV-based Stage H E M S Total data on 18 August 2019. During the interval, the bio- chemical conditions may have changed. Because it only H25 11 0 0 36 took 30–60 min for the drone to complete the data col- E 12 460 22 lection, but the artificial ground survey took at least 1 M 1 12 14 12 39 week. In this study, we cut each tree branch and then S 1 0914 24 measured the spectrum, Cab, and WC of each tree. Total 39 27 29 26 121 Therefore, the workload is relatively heavy, the ground OA (%) 47.11 survey cannot be synchronized with the drone data col- lection, and we can only keep the time as close as pos- Kappa 0.29 sible. (5) The results of our study may be affected by small sample size. Error analysis Previous studies show hyperspectral data to be effective Possible application of UAV-based hyperspectral data in in examining forest health (e.g., Pontius et al. 2008; Näsi detecting PWD et al., 2015). However, we encountered several difficul- Overall, the PWD infection stage classification of ground ties, obstacles, and sources of error in precisely estimat- data was more accurate than that of UAV-based air- ing leaf Cab, WC, and the stage of PWD in pine trees. borne hyperspectral data (Tables 9 and 10). There are (1) The stage of PWD of each sampled pine tree was several possible sources of this discrepancy. Firstly, judged by visual observation. These measurements were ground data consisted of samples from the entire tree fairly subjective and possibly inaccurate. (2) The acquisi- while the airborne data only measured canopy spectral tion of ground and UAV-based hyperspectral data are data. Therefore, ground data samples may more accur- both easily affected by the weather, especially light. Be- ately reflect the tree condition. Additionally, airborne cause data were collected during light hours, this may data acquisition is easily affected by weather – this may have biased results. (3) The results of individual tree have induced measurement errors. PA and UA of the crown segmentation using UAV-based hyperspectral four PWD infection classes using RF based on VIs, REPs, data were somewhat inaccurate. This increased the un- MIs, and combining all variables also suggest ground data certainly of extracting tree hyperspectral features and, performed better than airborne data (Table 11). However, consequently, it was difficult to distinguish pine trees when the RERs and combining all variables were used from understory trees and separate overlapping crowns from UAV-based data, predictions were comparably ac- from HIs using the image classification algorithm. curate to those of ground data (Tables 9 and 10). Manually drawing and visual interpretation can reduce Importantly, the acquisition of airborne data is simple, the interference of mixed pixels, but there was a prob- convenient, and much faster than ground data acquisi- lem that it cannot be efficiently applied when the sample tion. Therefore, there is a trade-off between the accuracy size was large. Nevertheless, in the actual situation, we and efficiency of data acquisition: ground data acquisi- can hardly meet two requirements at the same time: tion is accurate but time consuming to obtain while obtaining pure pixels and those that completely cover UAV-based airborne data is less accurate but much Fig. 8 The mean reflectance values of different disease stages at 350–2500 nm and 400–1000 nm from ground data (a) and UAV-based data (b) Yu et al. Forest Ecosystems (2021) 8:44 Page 12 of 19 Fig. 9 The mean reflectance values of different disease stages in Green, Red Edge, and NIR Fig. 10 Comparison of 11 Vegetation Indices (VIs), Red Edge Parameters (REPs), and Moisture Indices (MIs) at different disease degrees Yu et al. Forest Ecosystems (2021) 8:44 Page 13 of 19 Fig. 11 The mean decrease accuracy (MDA) of each selected variable from ground data (a) and UAV-based (b) data for estimating the disease stage of pine trees easier to obtain. Because PWD potentially affects trees characteristics (e.g. leaf Cab) of healthy trees and trees at in many large forest areas, ground data acquisition is not early stage of PWD are very different from those of trees a feasible management strategy. UAV-based data pro- in middle and serious stage (Fig. 5). In contrast, it is dif- vides only slightly less accurate classifications than ficult to distinguish healthy trees from trees in early ground-based data and is thus a more practical candi- stage of PWD because the difference in their spectral re- date for future large-scale forest management. sponses cannot be detected easily. REPs performed rela- tively well in distinguishing trees in early stage of PWD The potential of identifying trees in the early stage of infection from healthy trees (ground data OA: 80.33%, PWD Kappa: 0.58; and airborne data OA: 71.67%, Kappa: Our results show that it is relatively simple to distin- 0.40); however, overall, UAV-based data yielded moder- guish healthy trees and trees in early stage of PWD in- ately low accuracy (Fig. 13). Therefore, in practical appli- fection from trees in the middle and serious stage of cation, especially in a large-scale forest area, it is still a PWD infection. This is because the biochemical challenge to use UAV-based hyperspectral data to Yu et al. Forest Ecosystems (2021) 8:44 Page 14 of 19 Table 9 Classification confusion matrix of random forest (RF) classifier using vegetation indices, red edge parameters, moisture indices, and combined all variables based on ground spectral data Stage Vegetation Indices Red Edge Parameters H E M S Total UA (%) H E M S Total UA (%) H 29 7 0 0 36 80.56 32 7 0 0 39 82.05 E 8 17 3 0 28 60.71 5 17 1 0 23 73.91 M 1 2 22 3 28 78.57 1 2 24 3 30 80.00 S 11423 29 79.31 1 1 4 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 74.36 62.96 75.86 88.46 OA (%) 75.21 82.05 62.96 82.76 88.46 OA (%) 79.34 Kappa 0.66 Kappa 0.67 Stage Moisture Indices Combined H E M S Total UA (%) H E M S Total UA (%) H 29 9 1 0 39 74.36 32 8 0 0 40 80.00 E 9 15 1 1 26 57.69 5 17 2 0 24 70.83 M 0 3 24 3 30 80.00 2 2 24 2 30 80.00 S 10322 26 84.62 0 0 3 24 27 88.89 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 74.36 55.56 82.76 84.62 OA (%) 74.38 82.05 62.96 82.76 92.31 OA (%) 80.17 Kappa 0.65 Kappa 0.73 precisely identify trees at early infected stage of PWD. In wood borer in pine forest (Abdel-Rahman et al. 2014;Lin conclusion, the main focus of our next study is to im- et al. 2019;Iordache et al. 2020). prove the accuracy by some effective approaches (e.g., Currently, deep learning algorithms, such as convolu- using multi-temporal UAV hyperspectral data). tional neural network (CNN), have been showing its great potential in plant health monitoring (Yuan et al. 2017; Nagasubramanian et al. 2019; Wu et al. 2021). Classification algorithms However, it still has some dependencies. Firstly, when Machine learning algorithms, such as Random forest the data is small, deep learning algorithms do not per- (RF), support vector machine (SVM), Classification and form well. Furthermore, deep learning is like a black Regression Tree (CART), have been widely conducted in box, it does not reveal why it given the result, so it is classifying damaged trees by forest pest in previous stud- lack of interpretability (Ling et al. 2018; Silaparasetty ies (Abdel-Rahman et al. 2014; Iordache et al. 2020; Syifa 2020). On the other side, with its rigorous calculations et al. 2020; Zhan et al. 2020). In our study, RF algorithm and great flexibility (Schmidhuber 2015; Hao et al. was used. 2016), it could improve our classification accuracy. In In RF algorithm, the mean decrease accuracy (MDA) our next study, deep learning algorithms will be index of each variable is determined when calculating employed on PWD diagnose using multi-temporal the out-of-bag (OOB) error, which measures the import- UAV-based hyperspectral data. ance of the variables by comparing how much OOB error of estimate value increases when excluding one vari- The possible application of Lidar able and keeping others unchanged (Archer and Kimes, In this study, the classification model, predictions for 2008; Verikas et al. 2011;Abdel-Rahman et al. 2013). Cab and WC, and the results of individual tree crown Thus, the higher the MDA values of a variable, the greater segmentation were obtained based on hyperspectral data its importance (Immitzer et al. 2012; Liu et al. 2017), we alone. However, the results were not satisfactory, espe- can thereby determine the most important variable. Add- cially the tree crown segmentation (only delineated 121 itionally, compared with other algorithms, RF is more in- from 218). Another potential method of data collection sensitive to multicollinearity, and its results are relatively is Lidar (light detection and ranging). Lidar can directly, robust to missing and unbalanced data, and it can well quickly, and accurately obtain three-dimensional geo- predict the effect of thousands of explanatory variables graphic coordinates of objects (Vierling et al. 2008). (Breiman 2001). Therefore, RF have been widely used in Much progress has been made in the application of monitoring forest disturbance, especially for detecting Lidar technology in the fields of geology, forestry and Yu et al. Forest Ecosystems (2021) 8:44 Page 15 of 19 Table 10 Classification confusion matrix of random forest (RF) classifier using vegetation indices, red edge parameters, moisture indices, and combined all variables based on UAV hyperspectral data Stage Vegetation Indices Red Edge Parameters H E M S Total UA (%) H E M S Total UA (%) H 25 9 1 0 35 71.43 28 8 1 0 37 75.68 E 10 16 2 0 28 57.14 9 15 3 0 27 55.56 M 3 1 22 4 30 73.33 1 2 22 3 28 78.57 S 11422 28 78.57 1 2 3 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 64.10 59.26 75.86 84.62 OA (%) 70.25 71.79 55.56 75.86 88.46 OA (%) 72.73 Kappa 0.60 Kappa 0.63 Stage Moisture Indices Combined H E M S Total UA (%) H E M S Total UA (%) H 24 9 2 1 36 66.67 28 8 0 0 36 77.78 E 10 14 3 2 29 48.28 9 15 2 0 26 57.69 M 3 2 20 4 29 68.97 1 2 24 3 30 80.00 S 22419 27 70.37 1 2 3 23 29 79.31 Total 39 27 29 26 121 39 27 29 26 121 PA (%) 61.54 51.85 68.97 73.08 OA (%) 63.64 71.79 55.56 82.76 88.46 OA (%) 74.38 kappa 0.51 Kappa 0.66 ecology, such as the establishment of digital elevation tree segmentation could improve accuracy (e.g., Junttila model (DEM), the extraction of forest structure parame- et al. 2019; Lin et al. 2019). Furthermore, crown struc- ters, and the inversion of forest ecosystem function pa- ture and other tree structural information are likely to rameters (e.g., Watt et al. 2014; Huang and Lian 2015; change throughout PWD infection. Therefore, variables Saarela et al. 2020; Xie et al. 2020). based on the return intensity information from Lidar This makes Lidar a possible candidate to improve data might be useful in estimating the stage of PWD in measurement accuracy. Although Lidar data failed to ac- pine trees, and it will be our next study. curately reflect the biochemical condition of tree crowns (e.g., Liu et al. 2017; Shi et al. 2018), it can be used as Conclusion measure auxiliary data that produces three-dimensional In this paper, we compared the relatively accuracies of tree canopy structures (e.g., Shendryk et al. 2016). Thus, using ground-based data and UAV-based hyperspectral combining Lidar with hyperspectral data for individual data in predicting the stage of PWD infection in pine Table 11 Producer’s accuracy (%) and user’s accuracy (%) of the four stages using RF based on vegetation indices, red edge parameters, moisture indices, and combined all variables from ground and UAV-based data Stage Vegetation Indices Red Edge Parameters Moisture Indices Combined Producer’s User’s Producer’s User’s Producer’s User’s Producer’s User’s accuracy accuracy accuracy accuracy accuracy accuracy accuracy accuracy Based on ground spectrum data H 74.36 80.56 82.05 82.05 74.36 74.36 82.05 80.00 E 62.96 60.71 62.96 73.91 55.56 57.69 62.96 70.83 M 75.86 78.57 82.76 80.00 82.76 80.00 82.76 80.00 S 88.46 79.31 88.46 79.31 84.62 84.62 92.31 88.89 UAV-based hyperspectral data H 64.10 71.43 71.79 75.68 61.54 66.67 71.79 77.78 E 59.26 57.14 55.56 55.56 51.85 48.28 55.56 57.69 M 75.86 73.33 75.86 78.57 68.97 68.97 82.76 80.00 S 84.62 78.57 88.46 79.31 73.08 70.37 88.46 79.31 Yu et al. Forest Ecosystems (2021) 8:44 Page 16 of 19 Fig. 12 Producer’s and user’s accuracies for healthy (H), early (E), middle stage (M), and serious stage (S) of disease of pine tree pairs comparison achieved by random forest when the vegetation indices (a), red edge parameters (b), moisture indices (c), and combined all variables (d) based on ground spectrum data trees. To do this, we selected VIs, REPs, MIs, and com- highest accuracy in distinguishing between the healthy bining all variables as input parameters in a RF classifi- trees and trees in early stage of PWD infection. The clas- cation model. We found that combining all variables sification accuracy of REPs based on UAV (airborne) generally perform the best for estimating the stage of data had slightly poorer performance in distinguishing PWD infection of pine trees, and that REPs exhibit the trees at early stage of PWD and healthy trees (OA: Fig. 13 Producer’s and user’s accuracies for healthy (H), early (E), middle stage (M), and serious stage (S) of disease of pine trees pairs comparison achieved by random forest when the vegetation indices (a), red edge parameters (b), moisture indices (c), and combined all variables (d) when the UAV-based hyperspectral data were employed Yu et al. Forest Ecosystems (2021) 8:44 Page 17 of 19 71.67%, Kappa: 0.40), but is still a feasible method. Boochs F, Kupfer G, Dockter K, Kühbauch W (1990) Shape of the red edge as vitality indicator for plants. Int J Remote Sens 11(10):1741–1753. https://doi. Therefore, UAV-based hyperspectral imaging is a prom- org/10.1080/01431169008955127 ising candidate for measuring forest health. Relative to Breiman L (2001) Random forests. Mach Learn 45(1):5–32. https://doi.org/10.1 methods that use ground data, UAV-based hyperspectral 023/a:1010933404324 Carter GA, Miller RL (1994) Early detection of plant stress by digital imaging imaging has the potential to substantially reduce labor within narrow stress-sensitive wavebands. Remote Sens Environ 50(3):295– and time costs. Future studies should aim to improve 302. https://doi.org/10.1016/0034-4257(94)90079-5 the accuracy of UAV-based data. One possible direction Chen JM (1996) Evaluation of vegetation indices and a modified simple ratio for boreal applications. Can J Remote Sens 22(3):229–242. https://doi.org/10.1 is the use of supplemental data acquisition practices 080/07038992.1996.10855178 such as UAV-based Lidar data to improve classification Chen YH (2005) Effects of pine wood nematode (PWN) infection on water accuracy. regime and metabolism of related to hosts. Acta Phytopathol Sin 35(3):201– 207. https://doi.org/10.13926/j.cnki.apps.2005.03.003 (in Chinese) Acknowledgments Congalton RG (1991) A review of assessing the accuracy of classifications of We thank Klaus v. Gadow for research insights and editorial assistance, and remotely sensed data. Remote Sens Environ 37(1):35–46. https://doi.org/10.1 we also thank the anonymous reviewers for helpful suggestions. We also 016/0034-4257(91)90048-B thank all people involved in the assignments within the scope of the project, Curran PJ, Dungan JL, Gholz HL (1990) Exploring the relationship between related to the issues presented in this article. The authors would like to thank reflectance red edge and chlorophyll content in slash pine. Tree Physiol 7(1- TopEdit (www.topeditsci.com) for its linguistic assistance during the 2-3-4):33–48. https://doi.org/10.1093/treephys/7.1-2-3-4.33 preparation of this manuscript. Dawson TP, Curran PJ (1998) A new technique for interpolating the reflectance red edge position. Int J Remote Sens 19(11):2133–2139. https://doi.org/10.1 Authors’ contributions 080/014311698214910 Y.R. designed and conducted this research, analyzed the results, and wrote De Klerk HM, Buchanan G (2017) Remote sensing training in African conservation. the manuscript; L.R. reviewed the manuscript. All authors gave comments Remote Sens Ecol Conserv 3(1):7–20. https://doi.org/10.1002/rse2.36 and approved the final manuscript. Douda O, Zouhar M, Maňasová M, Dlouhý M, Lišková J, Ryšánek P (2015) Hydrogen cyanide for treating wood against pine wood nematode Funding (Bursaphelenchus xylophilus): results of a model study. J Wood Sci 61(2):204– This research was funded by the National Key Research & Development 210. https://doi.org/10.1007/s10086-014-1452-9 Program of China (2018YFD0600200), Beijing’s Science and Technology Du HQ, Ge HL, Fan YW, Jin W, Li J (2009) Application of fractal theory in Planning Project (Z191100008519004) and Major emergency science and hyperspectral detecting the early stage of pine wood nematode disease technology projects of National Forestry and Grassland Administration (Bursaphelenchus xylophilus)of Pinus massoniana with Hyperspectrum. Sci Silv (ZD202001–05). Sin 45(6):68–76. https://doi.org/10.3321/j.issn:1001-7488.2009.06.012 (in Chinese) Availability of data and materials Franklin SE, Wulder MA, Skakun RS, Carroll AL (2003) Mountain pine beetle red- The data are available upon a reasonable request to the Authors. attack forest damage classification using stratified Landsat TM data in British Columbia, Canada. Photogr Eng Remote Sens 69:283–288. https://doi.org/1 0.14358/pers.69.3.283 Declarations Gao B-C (1996) NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens Environ 58(3):257–266. Ethics approval and consent to participate https://doi.org/10.1016/s0034-4257(96)00067-3 Not applicable. Hao X, Zhang G, Ma S (2016) Deep learning. Int J Semantic Comput 10(3):417– Consent for publication 439. https://doi.org/10.1142/S1793351X16500045 Not applicable. Hardisky MA, Smart RM, Klemas V (1983) Seasonal spectral characteristics and aboveground biomass of the tidal marsh plant, Spartina alterniflora. Photogramm Eng Remote Sens 49:85–92 Competing interests He KS, Bradley BA, Cord AF, Rocchini D, Tuanmu M-N, Schmidtlein S, Turner W, The authors declare that they have no competing interests. Wegmann M, Pettorelli N (2015) Will remote sensing shape the next generation of species distribution models? Remote Sens Ecol Conserv 1(1):4– Received: 24 December 2020 Accepted: 5 April 2021 18. https://doi.org/10.1002/rse2.7 Hicke JA, Logan J (2009) Mapping whitebark pine mortality caused by a mountain pine beetle outbreak with high spatial resolution satellite imagery. References Int J Remote Sens 30(17):4427–4441. https://doi.org/10.1080/01431160802 Abdel-Rahman EM, Ahmed FB, Ismail R (2013) Random forest regression and spectral band selection for estimating sugarcane leaf nitrogen concentration Horler DNH, Barber J, Barringer AR (1980) Effects of heavy metals on the using EO-1 Hyperion hyperspectral data. Int J Remote Sens 34(2):712–728. absorbance and reflectance spectra of plants. Int J Remote Sens 1(2):121– https://doi.org/10.1080/01431161.2012.713142 136. https://doi.org/10.1080/01431160108559256 Abdel-Rahman EM, Mutanga O, Adam E, Ismail R (2014) Detecting Sirex noctilio Huang BH (2020) Monitoring Bursaphelenchus xylophilus with multispectrum grey-attacked and lightning-struck pine trees using airborne hyperspectral camera in UAV. Guangxi Forest Sci 49(3):380–384. https://doi.org/10.19692/j. data, random forest and support vector machines classifiers. ISPRS J cnki.gfs.2020.03.012 (in Chinese) Photogrammd Remote Sens 88:48–59. https://doi.org/10.1016/j.isprsjprs.2 Huang H, Lian J (2015) A 3D approach to reconstruct continuous optical images 013.11.013 using lidar and MODIS. Forest Ecosyst 2(1):20. https://doi.org/10.1186/s40663- Ahmed N, Atzberger C, Zewdie W (2020) Integration of remote sensing and 015-0044-5 bioclimatic data for prediction of invasive species distribution in data-poor regions: a review on challenges and opportunities. Environ Syst Res 9(1):32. Huang HH, Ma XH, Huang HY, Zhou YF, Zhang W, Huang YH (2018) A https://doi.org/10.1186/s40068-020-00195-0 preliminary study on monitoring of dead pine trees caused by pine wilt Archer KJ, Kimes RV (2008) Empirical characterization of random forest variable disease with fixed-wing unmanned aerial vehicle. J Environ Entomol 40(2): importance measures. Comput Stat Data Ann 52(4):2249–2260. https://doi. 306–313. https://doi.org/10.3969/j.issn.1674-0858.2018.02.9 (in Chinese) org/10.1016/j.csda.2007.08.015 Huang MX, Gong JH, Li S, Zhang B, Hao QT (2012) Study on pine wilt disease Blackburn GA (1998) Quantifying chlorophylls and caroteniods at leaf and canopy hyper-spectral time series and sensitive features. Remote Sens Technol Appl scales: an evaluation of some hyperspectral approaches. Remote Sens 27(6):954–960. https://doi.org/10.11873/j.issn.1004-0323.2012.6.954 (in Environ 66(3):273–285. https://doi.org/10.1016/S0034-4257(98)00059-5 Chinese) Yu et al. Forest Ecosystems (2021) 8:44 Page 18 of 19 Hui J (2018) Damage and control measures of pine wilt disease. China South Mullen KE (2016) Early detection of mountain pine beetle damage in ponderosa Agric Machine 49(4):100. https://doi.org/10.3969/j.issn.1672-3872.2018.04.087 pine forests of the black hills using hyperspectral and WorldView-2 data. (in Chinese) Minnesota State University, Mankato, pp 58–60 Hunt ER, Rock BN (1989) Detection of changes in leaf water content using near- Mutanga O, Adam E, Cho MA (2012) High density biomass estimation for and middle-infrared reflectances. Remote Sens Environ 30(1):43–54. https:// wetland vegetation using WorldView-2 imagery and random forest doi.org/10.1016/0034-4257(89)90046-1 regression algorithm. Int J Appl Earth Observ Geoinform 18:399–406. https:// doi.org/10.1016/j.jag.2012.03.012 Hyun MW, Kim JH, Suh DY, Lee SK, Kim SH (2007) Fungi isolated from pine wood nematode, its vector Japanese pine sawyer, and the nematode-infected Nagasubramanian K, Jones S, Singh AK, Sarkar S, Singh A, Ganapathysubramanian Japanese black pine wood in Korea. Mycobiology 35(3):159–161. https://doi. B (2019) Plant disease identification using explainable 3D deep learning on org/10.4489/MYCO.2007.35.3.159 hyperspectral images. Plant Methods 15(1):98. https://doi.org/10.1186/s13007- Immitzer M, Atzberger C, Koukal T (2012) Tree species classification with random 019-0479-8 forest using very high spatial resolution 8-band worldview-2 satellite data. Näsi R, Honkavaara E, Lyytikäinen-Saarenmaa P, Blomqvist M, Litkey P, Hakala T, Remote Sens 4(9):2661–2693. https://doi.org/10.3390/rs4092661 Viljanen N, Kantola T, Tanhuanpää T, Holopainen M (2015) Using UAV-based Inoue Y, Sakaiya E, Zhu Y, Takahashi W (2012) Diagnostic mapping of canopy photogrammetry and hyperspectral imaging for mapping bark beetle nitrogen content in rice based on hyperspectral measurements. Remote damage at tree-level. Remote Sens 7(11):15467–15493. https://doi.org/10.33 Sens Environ 126:210–221. https://doi.org/10.1016/j.rse.2012.08.026 90/rs71115467 Iordache M-D, Mantas V, Baltazar E, Pauly K, Lewyckyj N (2020) A machine National Forestry Administration (2018) National Forestry Administration learning approach to detecting pine wilt disease using airborne spectral Announcement No.1 in 2018. http://www.forestry.gov.cn/main/3457/201802 imagery. Remote Sens 12(14):2280. https://doi.org/10.3390/rs12142280 07/1074330.html. Accessed 16 Oct 2020 Pan CS (2011) Development of studies on pinewood nematodes diseases. J Jones HG, Vaughan RA (2010) Remote sensing of vegetation. Oxford University Xiamen Univ 50(2):476–483 (in Chinese) Press, New York, pp 169–171 Jung KY, Park JK (2019) Analysis of vegetation infection information using Pan J, Ju YW, Wang XT, Zhang H (2014) Detection of Bursaphelenchus xylophilus unmanned aerial vehicle with optical sensor. Sensor Material 31:3319–3326. infection in Pinus massoniana from hyperspectral data. Nematology 16(10): https://doi.org/10.18494/sam.2019.2465 1197–1207. https://doi.org/10.1163/15685411-00002846 Junttila S, Holopainen M, Vastaranta M, Lyytikäinen-Saarenmaa P, Kaartinen H, Pan L, Li YX, Liu ZK, Meng FL, Chen J, Zhang XY (2019) Isolation and Hyyppä J, Hyyppä H (2019) The potential of dual-wavelength terrestrial lidar identification of pine wood nematode in Pinus koraiensis in Fengcheng, in early detection of Ips typographus (L.) infestation – leaf water content as a Liaoning Province. Forest Pest Dis 38(1):1–4. https://doi.org/10.19688/j.cnki. proxy. Remote Sens Environ 231:111264. https://doi.org/10.1016/j.rse.201 issn1671-0886.20180021 (in Chinese) 9.111264 Penuelas J, Pinol J, Ogaya R, Filella I (1997) Estimation of plant water concentration by the reflectance water index WI (R900/R970). Int J Remote Kim S-R, Lee W-K, Lim C-H, Kim H, Kafatos MC, Lee S-H, Lee S-S (2018) Sens 18(13):2869–2875. https://doi.org/10.1080/014311697217396 Hyperspectral analysis of pine wilt disease to determine an optimal detection index. Forests 9(3):115. https://doi.org/10.3390/f9030115 Pontius J, Martin M, Plourde L, Hallett R (2008) Ash decline assessment in Kuai CL (2012) Occurrence and control of pine wilt disease. Modern Agric Sci emerald ash borer-infested regions: a test of tree-level, hyperspectral Technol 18:123–123. https://doi.org/10.3969/j.issn.1007-5739.2012.18.079 (in technologies. Remote Sens Environ 112(5):2665–2676. https://doi.org/10.101 Chinese) 6/j.rse.2007.12.011 Li H, Xu HH, Zheng HY, Chen XY (2020) Monitoring technology of pine wilt Prasad AM, Iverson LR, Liaw A (2006) Newer classification and regression tree disease based on UAV remote sensing image. J Chin Agric Mechan 41(9): techniques: bagging and random forests for ecological prediction. 170–175. https://doi.org/10.13733/j.jcam.issn.2095-5553.2020.09.027 (in Ecosystems 9(2):181–199. https://doi.org/10.1007/s10021-005-0054-1 Chinese) Richardson AJ, Wiegand CL (1977) Distinguishing vegetation from soil background information. Photogram Eng Remote Sens 43(12):1541–1552 Li H, Zhou GY, Liu JN, Zhang HY (2011) Study on pine wilt disease and its control situation. Appl Mechd Mater 55–57:567–572. https://doi.org/10.4028/www. Saarela S, Wästlund A, Holmström E, Mensah AA, Holm S, Nilsson M, Fridman J, scientific.net/amm.55-57.567 Ståhl G (2020) Mapping aboveground biomass and its prediction uncertainty Li ZX, Zhang XL, Zhang SC, Yan XS (2004) Application of RS and GIS in using LiDAR and field data, accounting for tree-level allometric and LiDAR monitoring forest disease and insect pests. Hebei J For Orchard Res 19(4): model errors. Forest Ecosyst 7(1):43. https://doi.org/10.1186/s40663-020-0024 377–380. https://doi.org/10.3969/j.issn.1007-4961.2004.04.017 (in Chinese) 5-0 Lin Q, Huang H, Wang J, Huang K, Liu Y (2019) Detection of pine shoot beetle Santos CS, de Vasconcelos MW (2012) Identification of genes differentially (PSB) stress on pine forests at individual tree level using UAV-based expressed in Pinus pinaster and Pinus pinea after infection with the pine hyperspectral imagery and lidar. Remote Sens 11(21):2540. https://doi.org/1 wood nematode. Eur J Plant Pathol 132(3):407–418. https://doi.org/10.1007/ 0.3390/rs11212540 s10658-011-9886-z Lin XZ (2015) Review on damage and control measures of pine wilt disease. East Santoso H, Tani H, Wang X (2016) A simple method for detection and counting China Forest Manag 29(3):28–30. https://doi.org/10.3969/j.issn.1004-7743.201 of oil palm trees using high-resolution multispectral satellite imagery. Int J 5.03.008 (in Chinese) Remote Sens 37(21):5122–5134. https://doi.org/10.1080/01431161.2016.122 Ling G, Lei G, Elmadany NEI, Liang C (2018) Statistical machine learning vs deep Schlemmer M, Gitelson A, Schepers J, Ferguson R, Peng Y, Shanahan J, Rundquist learning in information fusion: competition or collaboration? In: 2018 IEEE D (2013) Remote estimation of nitrogen and chlorophyll contents in maize conference on multimedia information processing and retrieval. MIPR, Miami, at leaf and canopy levels. Int J Appl Earth Observ Geoinform 25:47–54. pp 251–256. https://doi.org/10.1109/MIPR.2018.00059 https://doi.org/10.1016/j.jag.2013.04.003 Liu L, Coops NC, Aven NW, Pang Y (2017) Mapping urban tree species using integrated airborne hyperspectral and LiDAR remote sensing data. Remote Schmidhuber J (2015) Deep learning in neural networks: an overview. Neural Sens Environ 200:170–182. https://doi.org/10.1016/j.rse.2017.08.010 Netw 61:85–117. https://doi.org/10.1016/j.neunet.2014.09.003 Liu W, Chang QR, Guo M, Xing DX, Yuan YS (2010) Monitoring of leaf nitrogen Shen GR, Wang RC (2001) Review of the application of vegetation remote content in summer corn with first derivative of spectrum based on modified sensing. J Zhejiang Univ (Agric Life Sci) 27(6):682–690. https://doi.org/1 red edge. J Northwest A F Univ 38(4):91–98. https://doi.org/10.13207/j.cnki. 0.3321/j.issn:1008-9209.2001.06.025 (in Chinese) jnwafu.2010.04.018 (in Chinese) Shendryk I, Broich M, Tulbure MG, McGrath A, Keith D, Alexandrov SV (2016) Mapping individual tree health using full-waveform airborne laser scans and Ma J, Liu W, Zhang XL (2011) Remote sensing research on early monitoring and imaging spectroscopy: a case study for a floodplain eucalypt forest. Remote prediction of pine wilt disease. Forest Invent Plan 36(5):75–80. https://doi. Sens Environ 187:202–217. https://doi.org/10.1016/j.rse.2016.10.014 org/10.3969/j.issn.1671-3168.2011.05.018 (in Chinese) Mamiya Y (1988) History of pine wilt disease in Japan. J Nematol 20(2):219–226 Shi Y, Skidmore AK, Wang T, Holzwarth S, Heiden U, Pinnel N, Zhu X, Heurich M Meddens AJH, Hicke JA, Vierling LA (2011) Evaluating the potential of (2018) Tree species classification using plant functional traits from LiDAR and multispectral imagery to map multiple stages of tree mortality. Remote Sens hyperspectral data. Int J Appl Earth Observ Geoinform 73:207–219. https:// Environ 115(7):1632–1642. https://doi.org/10.1016/j.rse.2011.02.018 doi.org/10.1016/j.jag.2018.06.018 Yu et al. Forest Ecosystems (2021) 8:44 Page 19 of 19 Shin S-C (2008) Pine wilt disease in Korea. In: Zhao BG, Futai K, Sutherland JR, Spectrosc Spectr Anal 31(5):1352–1356. https://doi.org/10.3964/j.issn.1000- Takeuchi Y (eds) Pine wilt disease. Springer, Tokyo. https://doi.org/10.1007/ 0593(2011)05-1352-05 (in Chinese) 978-4-431-75655-2_5 Yang BJ (2002) Advance in research of pathogenetic mechanism of pine wood Silaparasetty N (2020) Machine learning vs. deep learning. In: Silaparasetty N (ed) nematode. Forest Pest Disease 1:27–31 (in Chinese) Machine learning concepts with Python and the Jupyter notebook Yao FQ, Zhang ZH, Yang RY, Sun JW, Cui SF (2009) Hyperspectral models for environment. Using Tensorflow 2.0. Apress, Berkeley, pp 57–65. https://doi. estimating vegetation chlorophyll content based on red edge parameter. org/10.1007/978-1-4842-5967-2_4 Trans Chin Soc Agric Eng 25(Sup2):123–129 (in Chinese) Ye JR (2019) Epidemic status of pine wilt disease in China and its prevention and Song QL, Xiang R, Qing L, Yuan L, Tian H, Zhu PB, Liu W, Yang W, Qu YX, Zhou control techniques and counter measures. Sci Silv Sin 55(9):1–10 (in Chinese) JW (2018) Relationship between spectral reflectance characteristic Yu HY, Wu H (2018) Discovery of new host plants and new vector insects of parameters and water content of pine leaves. Sci Technol Innov 03:26–28. Bursaphelenchus xylophilus in Liaoning Province. Forest Pest Dis 37(5):61 (in https://doi.org/10.15913/j.cnki.kjycx.2018.03.026 (in Chinese) Chinese) Syifa M, Park S-J, Lee C-W (2020) Detection of the pine wilt disease tree Yu HY, Wu H, Zhang XD, Wang LM, Zhang XF, Song YS (2019) Preliminary study candidates for drone remote sensing using artificial intelligence techniques. on Larix spp. infected by Bursaphelenchus xylophilus in natural environment. Engineering 6(8):919–926. https://doi.org/10.1016/j.eng.2020.07.001 Forest Pest Dis 38(4):7–10. https://doi.org/10.19688/j.cnki.issn1671-0886.201 Tang L, Shao G (2015) Drone remote sensing for forestry research and practices. J 80024 (in Chinese) For Res 26(4):791–797. https://doi.org/10.1007/s11676-015-0088-y Yuan H, Yang G, Li C, Wang Y, Liu J, Yu H, Feng H, Xu B, Zhao X, Yang X (2017) Tao H, Li C, Zhao D, Deng S, Hu H, Xu X, Jing W (2020) Deep learning-based Retrieving soybean leaf area index from unmanned aerial vehicle dead pine trees detection from unmanned aerial vehicle images. Int J Remote hyperspectral remote sensing: analysis of RF, ANN, and SVM regression Sens 41(21):8238–8255. https://doi.org/10.1080/01431161.2020.1766145 models. Remote Sens 9(4):309. https://doi.org/10.3390/rs9040309 Umebayashi T, Yamada T, Fukuhara K, Endo R, Kusumoto D, Fukuda K (2017) In Yuan J, Wang DL, Li R (2013) Remote sensing image segmentation by combining situ observation of pinewood nematode in wood. Eur J Plant Pathol 147(2): spectral and texture features. IEEE T Geosci Remote 52(1):16–24. https://doi. 463–467. https://doi.org/10.1007/s10658-016-1013-8 org/10.1109/TGRS.2012.2234755 Verikas A, Gelzinis A, Bacauskiene M (2011) Mining data with random forests: a Zhan Z, Yu L, Li Z, Rem L, Gao B, Wang L, Luo Y (2020) Combining GF-2 and survey and results of new tests. Pattern Recogn 44(2):330–349. https://doi. Sentinel-2 images to detect tree mortality caused by red turpentine beetle org/10.1016/j.patcog.2010.08.011 during the early outbreak stage in North China. Forests 11(2):172. https://doi. Vicente C, Espada M, Vieira P, Mota M (2012) Erratum to: pine wilt disease: a org/10.3390/f11020172 threat to European forestry. Eur J Plant Pathol 133(2):497. https://doi.org/10.1 Zhang N, Zhang X, Yang G, Zhu C, Huo L, Feng H (2018) Assessment of 007/s10658-012-9979-3 defoliation during the Dendrolimus tabulaeformis Tsai et Liu disaster outbreak Vierling KT, Vierling LA, Gould WA, Martinuzzi S, Clawges RM (2008) Lidar: using UAV-based hyperspectral images. Remote Sens Environ 217:323–339. shedding new light on habitat characterization and modeling. Front Ecol https://doi.org/10.1016/j.rse.2018.08.024 Environ 6(2):90–98. https://doi.org/10.1890/070001 Wang Z, Zhang X, An S (2007) Spectral characteristics analysis of Pinus Massoniana suffered by Bursaphelenchus Xylophilus. Remote Sens Technol Appl 22(3):367–370. https://doi.org/10.3969/j.issn.1004-0323.2007.03.012 (in Chinese) Waske B, Benediktsson JA, Árnason K, Sveinsson JR (2009) Mapping of hyperspectral AVIRIS data using machine-learning algorithms. Can J Remote Sens 35(Sup1):S106–S116. https://doi.org/10.5589/m09-018 Watt MS, Meredith A, Watt P, Gunn A (2014) The influence of LiDAR pulse density on the precision of inventory metrics in young unthinned Douglas-fir stands during initial and subsequent LiDAR acquisitions. NZ J Forestry Sci 44(1):18. https://doi.org/10.1186/s40490-014-0018-3 White JC, Coops NC, Hilker T, Wulder MA, Carroll AL (2007) Detecting mountain pine beetle red attack damage with EO-1 Hyperion moisture indices. Int J Remote Sens 28(10):2111–2121. https://doi.org/10.1080/01431160600944028 White JC, Wulder MA, Brooks D, Reich R, Wheate RD (2005) Detection of red attack stage mountain pine beetle infestation with high spatial resolution satellite imagery. Remote Sens Environ 96(3–4):340–351. https://doi.org/10.1 016/j.rse.2005.03.007 Wu B, Liang A, Zhang H, Zhu T, Zou Z, Yang D, Tang W, Li J, Su J (2021) Application of conventional UAV-based high-throughput object detection to the early diagnosis of pine wilt disease by deep learning. Forest Ecol Manag 486:118986. https://doi.org/10.1016/j.foreco.2021.118986 Wu CY, Niu Z (2008) Sensitivity study of photochemical reflectance index to leaf biochemical components. J Univ Chin Acad Sci 3:346–354 (in Chinese) Xiang R, Yuan L, Qin L, Song Q, Zhang J, Qu Y, Zhou J (2018) Correlation analysis between spectral characteristic parameters and chlorophyll content of pine needles. Sci Technol Innov 35:5–7 (in Chinese) Xie Q, Huang W, Liang D, Chen P, Wu C, Yang G, Zhang J, Huang L, Zhang D (2014) Leaf area index estimation using vegetation indices derived from airborne hyperspectral images in winter wheat. IEEE J-STARS 7(8):3586–3594. https://doi.org/10.1109/JSTARS.2014.2342291 Xie Y, Zhang J, Chen X, Pang S, Zeng H, Shen Z (2020) Accuracy assessment and error analysis for diameter at breast height measurement of trees obtained using a novel backpack LiDAR system. Forest Ecosyst 7(1):33. https://doi.org/1 0.1186/s40663-020-00237-0 Xu HC, Luo YQ, Zhang Q (2012) Changes in water content, pigments and antioxidant enzyme activities in pine needles of Pinus thunbergii and Pinus massoniana affected by pine wood nematode. Sci Silv Sin 11:140–143 (in Chinese) Xu HC, Luo YQ, Zhang TT, Shi YJ (2011) Changes of reflectance spectra of pine needles in different stage after being infected by pine wood nematode.

Journal

"Forest Ecosystems" – Springer Journals

Published: Jul 5, 2021

Keywords: Pine wilt disease; Remote sensing; Spectrometer; Hyperspectral imaging; Random forest; Classification

There are no references for this article.