No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Mapping individual tree health using full-waveform airborne laser scans and imaging spectroscopy: A case study for a floodplain eucalypt forest
... Airborne light detection and ranging (lidar) can provide high resolution measurements of forest structure useful for a variety of applications (Lefsky et al., 2002;Marchi et al., 2018). When high-density lidar data are analyzed in a single-tree, object-oriented approach (Reitberger et al., 2009;Li et al., 2012;Dalponte and Coomes, 2016;Lindberg and Holmgren, 2017;Campbell et al., 2020), individual tree attributes such as tree volume (Maltamo et al., 2004), basal area (Silva et al., 2016), canopy fuel characteristics (Popescu and Zhao, 2008;Klauberg et al., 2019), species (Ørka et al., 2009;Kim et al., 2009b;Zhao et al., 2020), and condition (Wing et al., 2015;Shendryk et al., 2016;Klauberg et al., 2019;Karna et al., 2019) can be estimated. The amount of energy reflected to the lidar sensor, often referred to as lidar intensity, has proven to be especially useful for distinguishing live and dead vegetation components (Kim et al., 2009a;Morsdorf et al., 2010;Bright et al., 2013;Wing et al., 2015), as has combining optical imagery and lidar data (Polewski et al., 2015;Shendryk et al., 2016;Kamińska et al., 2018;Campbell et al., 2020). ...
... When high-density lidar data are analyzed in a single-tree, object-oriented approach (Reitberger et al., 2009;Li et al., 2012;Dalponte and Coomes, 2016;Lindberg and Holmgren, 2017;Campbell et al., 2020), individual tree attributes such as tree volume (Maltamo et al., 2004), basal area (Silva et al., 2016), canopy fuel characteristics (Popescu and Zhao, 2008;Klauberg et al., 2019), species (Ørka et al., 2009;Kim et al., 2009b;Zhao et al., 2020), and condition (Wing et al., 2015;Shendryk et al., 2016;Klauberg et al., 2019;Karna et al., 2019) can be estimated. The amount of energy reflected to the lidar sensor, often referred to as lidar intensity, has proven to be especially useful for distinguishing live and dead vegetation components (Kim et al., 2009a;Morsdorf et al., 2010;Bright et al., 2013;Wing et al., 2015), as has combining optical imagery and lidar data (Polewski et al., 2015;Shendryk et al., 2016;Kamińska et al., 2018;Campbell et al., 2020). Studies using lidar have successfully classified individual trees as dead or alive (Yao et al., 2012;Wing et al., 2015;Polewski et al., 2015;Casas et al., 2016;Kamińska et al., 2018;Miltiadou et al., 2020;Briechle et al., 2020); fewer have demonstrated the use of lidar for distinguishing between multiple tree condition or "health" classes (Shendryk et al., 2016;Barnes et al., 2017;Meng et al., 2018a;Lin et al., 2019;Varo-Martínez and Navarro-Cerrillo, 2021). ...
... The amount of energy reflected to the lidar sensor, often referred to as lidar intensity, has proven to be especially useful for distinguishing live and dead vegetation components (Kim et al., 2009a;Morsdorf et al., 2010;Bright et al., 2013;Wing et al., 2015), as has combining optical imagery and lidar data (Polewski et al., 2015;Shendryk et al., 2016;Kamińska et al., 2018;Campbell et al., 2020). Studies using lidar have successfully classified individual trees as dead or alive (Yao et al., 2012;Wing et al., 2015;Polewski et al., 2015;Casas et al., 2016;Kamińska et al., 2018;Miltiadou et al., 2020;Briechle et al., 2020); fewer have demonstrated the use of lidar for distinguishing between multiple tree condition or "health" classes (Shendryk et al., 2016;Barnes et al., 2017;Meng et al., 2018a;Lin et al., 2019;Varo-Martínez and Navarro-Cerrillo, 2021). Although previous studies have used remote sensing to detect and map conifer decline and mortality caused by root disease (e.g., Johnson and Wear, 1975;Williams and Leaphart, 1978;Quinn and Niemann, 2017), to our knowledge, only a few have done so at the individual tree scale (Leckie et al., 2004;Navarro-Cerrillo et al., 2019;Iversen, 2020). ...
Armillaria root disease causes tree stress and mortality worldwide, including in conifer forests of western North America. Armillaria-induced tree mortality is slower than other disturbances, such as fire, but persistent over time, and therefore difficult to detect across large areas. Hence methods to detect Armillaria-affected trees and identify infested stands are of great value to forest managers. We used high-density light detection and ranging (lidar), high-resolution aerial orthoimagery, and associated field observations to map individual tree health across an Armillaria-affected forest dominated by Abies in south central Oregon. The lidar point cloud was segmented into individual tree objects (polygons representing tree crown extents as viewed from nadir), for which lidar metrics and spectral metrics derived from orthoimagery were computed. Lidar-detected tree objects were paired with 150 field-observed trees with corresponding health measurements, and a random forest classifier was developed that separated trees into: 1) asymptomatic; 2) live, Armillaria-infected; 3) recently killed (≥50% of red needles remaining); and 4) dead (<50% of dead needles remaining) classes with 83% accuracy using lidar and spectral metrics. The classifier was applied to map individual tree health status for 290,964 tree objects across the 1257-ha study area. Approximately 20% of trees were classified as unhealthy including live, infected and 4% of trees were classified as recently killed or dead. We created hotspot maps using the Getis-Ord Gi* statistic and analyzed clustering of tree health spatial patterns using Ripley’s L statistic. Hotspot maps effectively identified clusters of live unhealthy trees and tree mortality; unhealthy and dead trees were found to be significantly spatially clustered at distances of 100–2500 m. We created a dead tree density grid, which we coupled with a lidar-derived canopy gap grid to identify sites and stands affected by root disease. Canopy openings were mapped using a canopy height model with a minimum opening area of ≥ 202 m2. Clusters of mapped dead trees (excluding recently killed trees) intersecting canopy gaps were used to detect sites with root disease-induced mortality. Twenty-seven stands containing long-term plots from previous studies were evaluated for the presence of root disease-induced mortality. All 27 stands were correctly identified with conifer mortality induced by root disease. This approach for detecting dead trees intersecting canopy openings induced by root disease can aid in: i) prioritizing subsequent field data collection; ii) planning silvicultural prescriptions; and iii) assessing management expectations for snags and wildlife habitat where root disease-induced mortality is altering stand structure.
... Yet, so far, current efforts for tree health assessment using remote sensing mainly focused on forest ecosystems. High-resolution airborne and satellite imagery have been used for forest tree mortality and tree health assessment on both the individual tree canopy scale [26,29,30] and the image pixel scale [31][32][33][34]. Airborne hyperspectral imagery (HSI), providing optical information in high spectral detail, has been extensively applied to provide both biochemical and structural information for identifying tree health [27,28,[35][36][37][38][39]. ...
... In forest research, a few studies have integrated the information derived from ALS and airborne HSI with field assessments to monitor forest health. Shendryk et al. [30] used this approach to assess tree crown dieback and transparency ratios in a Eucalypt forest and their results showed a higher accuracy with the combined use of the two data sources compared to only using one. A similar finding was also reported in Meng et al. [29], in which the authors mapped forest canopy defoliation caused by herbivorous insects at the individual tree level using a combination of ALS and airborne HSI. ...
... In two separate studies, Näsi et al. [19,38] showed the potential of UAV-based photogrammetry and hyperspectral imaging for mapping bark beetle damage at tree level in an urban forest. Most existing tree health studies focused either on assessing tree crown defoliation [29,30,37] or the overall tree health condition [26,27,38], while attention to tree crown discoloration remained limited. In response to this, Degerickx et al. [46] conducted an object-based tree health classification based on chlorophyll content (discoloration) and leaf area index (defoliation) derived from airborne HSI. ...
Declining urban tree health can affect critical ecosystem services, such as air quality improvement, temperature moderation, carbon storage, and biodiversity conservation. The application of state-of-the-art remote sensing data to characterize tree health has been widely examined in forest ecosystems. However, such application to urban trees has not yet been fully explored—due to the presence of heterogeneous tree species and backgrounds, severely complicating the classification of tree health using remote sensing information. In this study, tree health was represented by a set of field-assessed tree health indicators (defoliation, discoloration, and a combination thereof), which were classified using airborne laser scanning (ALS) and hyperspectral imagery (HSI) with a Random Forest classifier. Different classification scenarios were established aiming at: (i) Comparing the performance of ALS data, HSI and their combination, and (ii) examining to what extent tree species mixtures affect classification accuracy. Our results show that although the predictive power of ALS and HSI indices varied between tree species and tree health indicators, overall ALS indices performed better. The combined use of both ALS and HSI indices results in the highest accuracy, with weighted kappa coefficients (Kc) ranging from 0.53 to 0.79 and overall accuracy ranging from 0.81 to 0.89. Overall, the most informative remote sensing indices indicating urban tree health are ALS indices related to point density, tree size, and shape, and HSI indices associated with chlorophyll absorption. Our results further indicate that a species-specific modelling approach is advisable (Kc points improved by 0.07 on average compared with a mixed species modelling approach). Our study constitutes a basis for future urban tree health monitoring, which will enable managers to guide early remediation management.
... One advantage of tri-dimensional LiDAR data is that they are less subject to changing atmospheric and lightning conditions during the survey than spectral data. Moreover, LiDAR coverage is becoming more frequent at the regional/national scale (Parent et al., 2015;Wasser et al., 2015;Shendryk et al., 2016;Tompalski et al., 2017). When an initial nationwide LiDAR survey is performed, digital aerial photogrammetry (DAP) can be used to further update LiDAR canopy height models (CHMs). ...
... Accuracy thus decreases, making it more complicated to study different types of vegetation separately (Tillack et al., 2014;Cunningham et al., 2018). The health status of vegetation is often studied with higher resolution data, occasionally with a single image (Tillack et al., 2014;Michez et al., 2016b;Bucha and Sl� avik, 2013;Shendryk et al., 2016;Sankey et al., 2016). ...
... For example, full-waveform LiDAR data have shown promising results in forestry applications (e.g. Koenig and H€ ofle, 2016), but there are few examples for riparian vegetation (Shendryk et al., 2016). ...
Riparian vegetation is a central component of the hydrosystem. As such, it is often subject to management practices that aim to influence its ecological, hydraulic or hydrological functions. Remote sensing has the potential to improve knowledge and management of riparian vegetation by providing cost-effective and spatially continuous data over wide extents. The objectives of this review were twofold: to provide an overview of the use of remote sensing in riparian vegetation studies and to discuss the transferability of remote sensing tools from scientists to managers. We systematically reviewed the scientific literature (428 articles) to identify the objectives and remote sensing data used to characterize riparian vegetation. Overall, results highlight a strong relationship between the tools used, the features of riparian vegetation extracted and the mapping extent. Very high-resolution data are rarely used for rivers longer than 100 km, especially when mapping species composition. Multi-temporality is central in remote sensing riparian studies, but authors use only aerial photographs and relatively coarse resolution satellite images for diachronic analyses. Some remote sensing approaches have reached an operational level and are now used for management purposes. Overall, new opportunities will arise with the increased availability of very high-resolution data in understudied or data-scarce regions, for large extents and as time series. To transfer remote sensing approaches to riparian managers, we suggest mutualizing achievements by producting open-access and robust tools. These tools will then have to be adapted to each specific project, in collaboration with managers.
... The traditional mulitspectral satellite images (e.g., Landsat) were difficult to distinguish damaged tree crowns from health tree crowns at around 30 m resolution for the early stage of forest insect attack [27]. Moreover, mid-and low-resolution remote sensing data are not readily applicable for forest health studies at the individual tree level due to limitations in the capacity of satellite imagery to detect suppressed trees that were barely seen [28,29]. Especially, the slightly infected Yunan Pine trees only have several damaged shoots within crowns. ...
... With the fast development of unmanned airborne vehicles (UAV) platforms and light sensors, UAV-based imaging technologies using very high resolution optical imaging spectroscopy (IS) and light detection and ranging (lidar) have great potential in forest health monitoring at individual tree level [29][30][31][32]. IS features such as hyperspectral imagery (HI) provided continuous narrowband spectral information, which greatly enhanced the capability to extract forest health information [11,33]. ...
... High density lidar pulses can accurately measure the 3D geometric (horizontal plus vertical) information of forest canopies. Therefore, the integration of lidar data into HI analysis could provide critical information on the shadow distribution of a tree crown [29]. The lighting conditions in tree crown pixels of HI were determined using a combination of lidar point cloud and solar geometry at the acquisition time of HI [29]. ...
In recent years, the outbreak of the pine shoot beetle (PSB), Tomicus spp., has caused serious shoots damage and the death of millions of trees in Yunnan pine forests in southwestern China. It is urgent to develop a convincing approach to accurately assess the shoot damage ratio (SDR) for monitoring the PSB insects at an early stage. Unmanned airborne vehicles (UAV)-based sensors, including hyperspectral imaging (HI) and lidar, have very high spatial and spectral resolutions, which are very useful to detect forest health. However, very few studies have utilized HI and lidar data to estimate SDRs and compare the predictive power for mapping PSB damage at the individual tree level. Additionally, the data fusion of HI and lidar may improve the detection accuracy, but it has not been well studied. In this study, UAV-based HI and lidar data were fused to detect PSB. We systematically evaluated the potential of a hyperspectral approach (only-HI data), a lidar approach (only-lidar data), and a combined approach (HI plus lidar data) to characterize PSB damage of individual trees using the Random Forest (RF) algorithm, separately. The most innovative point is the proposed new method to extract the three dimensional (3D) shadow distribution of each tree crown based on a lidar point cloud and the 3D radiative transfer model RAPID. The results show that: (1) for the accuracy of estimating the SDR of individual trees, the lidar approach (R2 = 0.69, RMSE = 12.28%) performed better than hyperspectral approach (R2 = 0.67, RMSE = 15.87%), and in addition, it was useful to detect dead trees with an accuracy of 70%; (2) the combined approach has the highest accuracy (R2 = 0.83, RMSE = 9.93%) for mapping PSB damage degrees; and (3) when combining HI and lidar data to predict SDRs, two variables have the most contributions, which are the leaf chlorophyll content (Cab) derived from hyperspectral data and the return intensity of the top of shaded crown (Int_Shd_top) from lidar metrics. This study confirms the high possibility to accurately predict SDRs at individual tree level if combining HI and lidar data. The 3D radiative transfer model can determine the 3D crown shadows from lidar, which is a key information to combine HI and lidar. Therefore, our study provided a guidance to combine the advantages of hyperspectral and lidar data to accurately measure the health of individual trees, enabling us to prioritize areas for forest health promotion. This method may also be used for other 3D land surfaces, like urban areas.
... Evaluating the combination of remote sensing data sources recorded at the same timestep makes this study unique in the sphere of date palm management. Several studies have shown that the combination of remote sensing sources can provide multivariate indicators with improved characterisation of the vegetation health status [20,26]. ...
... By determining the ratio between height and canopy area, one can determine whether a tree has an abnormally small canopy in relation to height-this could be an indication that the tree is experiencing growth limitations. In a study by Shendryk et al. (2016), a width height ratio was included amongst hyperspectral and LiDAR indicators to determine the most "useful" indicator for assessing tree health using Principal Component Analysis (PCA) [20]. Of the indicators derived from the recorded tree height, the width height ratio was the most important in their study. ...
... By determining the ratio between height and canopy area, one can determine whether a tree has an abnormally small canopy in relation to height-this could be an indication that the tree is experiencing growth limitations. In a study by Shendryk et al. (2016), a width height ratio was included amongst hyperspectral and LiDAR indicators to determine the most "useful" indicator for assessing tree health using Principal Component Analysis (PCA) [20]. Of the indicators derived from the recorded tree height, the width height ratio was the most important in their study. ...
Date palms are a valuable crop in areas with limited water availability such as the Middle East and sub-Saharan Africa, due to their hardiness in tough conditions. Increasing soil salinity and the spread of pests including the red palm weevil (RPW) are two examples of growing threats to date palm plantations. Separate studies have shown that thermal, multispectral, and hyperspectral remote sensing imagery can provide insight into the health of date palm plantations, but the added value of combining these datasets has not been investigated. The current study used available thermal, hyperspectral, Light Detection and Ranging (LiDAR) and visual Red-Green-Blue (RGB) images to investigate the possibilities of assessing date palm health at two “levels”; block level and individual tree level. Test blocks were defined into assumed healthy and unhealthy classes, and thermal and height data were extracted and compared. Due to distortions in the hyperspectral imagery, this data was only used for individual tree analysis; methods for identifying individual tree points using Normalized Difference Vegetation Index (NDVI) maps proved accurate. A total of 100 random test trees in one block were selected, and comparisons between hyperspectral, thermal and height data were made. For the vegetation index red-edge position (REP), the R-squared value in correlation with temperature was 0.313 and with height was 0.253. The vegetation index—the Vogelmann Red Edge Index (VOGI)—also has a relatively strong correlation value with both temperature (R2 = 0.227) and height (R2 = 0.213). Despite limited field data, the results of this study suggest that remote sensing data has added value in analyzing date palm plantations and could provide insight for precision agriculture techniques.
... Very high-resolution drone images provide additional means to monitor forest health via spectral indices (Dash et al., 2017). Another widely used method to estimate tree stress is airborne laserscanning in combination with a supervised classification (Yao et al., 2012a;Shendryk et al., 2016). Shendryk et al. (2016) combine fullwaveform airborne laser scans with hyperspectral data and train a random forest classifier to estimate dieback and transparency. ...
... Another widely used method to estimate tree stress is airborne laserscanning in combination with a supervised classification (Yao et al., 2012a;Shendryk et al., 2016). Shendryk et al. (2016) combine fullwaveform airborne laser scans with hyperspectral data and train a random forest classifier to estimate dieback and transparency. Mak and Hu (2014) determine the health status of ash trees with ground-based mobile laserscanning as a function of the point density. ...
... Tree defoliation estimation is a major component of tree stress estimation and especially the laserscanning methods mentioned previously are primarily based on estimating defoliation (Yao et al., 2012a;Lin et al., 2014;Shendryk et al., 2016). A wide variety of research applies remote sensing to defoliation estimation of forest trees (Kantola et al., 2010;Mozgeris and Augustaitis, 2013;Marx and Kleinschmit, 2017;Hawrylo et al., 2018). ...
In this paper, we propose to estimate tree defoliation from ground-level RGB photos with convolutional neural networks (CNN). Tree defoliation is usually assessed with field campaigns, where experts estimate multiple tree health indicators per sample site. Campaigns span entire countries to come up with a holistic, nation-wide picture of forest health. Surveys are very laborous, expensive, time-consuming and need a large number of experts. We aim at making the monitoring process more efficient by casting tree defoliation estimation as an image interpretation problem. What makes this task challenging is strong variance in lighting, viewpoint, scale, tree species, and defoliation types. Instead of accounting for each factor separately through explicit modelling, we learn a joint distribution directly from a large set of annotated training images following the end-to-end learning paradigm of deep learning. We evaluate our supervised method on three data sets with different level of difficulty acquired in Swiss forests and compare them to human performance. Results show that tree defoliation estimation from images with CNNs works well and achieves performance very close to human experts.
... Relationships between remotely sensed and ground-based metrics facilitate classifications of tree crown condition, providing a spatial representation of disease or decline throughout forested environments and hence a useful tool for disease management (Shendryk et al., 2016). The selection of disease severity category boundaries is particularly important, with difficulties previously noted in the differentiation between classes across the spectrum of moderate disease severity for forest pests Leckie et al., 2005). ...
... Two important input parameters for the RF classification include the number of regression trees (n tree ) and the number of input variables at each split in the tree building process (m try ). Following a preliminary grid search cross validation of the training dataset, the n tree and m try were set to 500 and 2 respectively, similar to the values applied in previous studies (Immitzer et al., 2012;Ortiz et al., 2013;Shendryk et al., 2016). All processing for the two classification methods was undertaken in Python using Scikitlearn (Pedregosa et al., 2011). ...
... For example, whilst some overlap perfectly with ITCs, those categorised as satisfactory may incorporate returns from neighbouring vegetation, potentially influencing the calculation of metrics employed as input variables. This effect could be managed with more restrictive criteria, such as 60% or 70% minimum overlap, for determining successfully delineated automated crowns (Shendryk et al., 2016). ...
The invasive phytopathogen Phytophthora ramorum has caused extensive infection of larch forest across areas of the UK, particularly in Southwest England, South Wales and Southwest Scotland. At present, landscape level assessment of the disease in these areas is conducted manually by tree health surveyors during helicopter surveys. Airborne laser scanning (ALS), also known as LiDAR, has previously been applied to the segmentation of larch tree crowns infected by P. ramorum infection and the detection of insect pests in coniferous tree species. This study evaluates metrics from high-density discrete ALS point clouds (24 points/m²) and canopy height models (CHMs) to identify individual trees infected with P. ramorum and to discriminate between four disease severity categories (NI: not infected, 1: light, 2: moderate, 3: heavy). The metrics derived from ALS point clouds include canopy cover, skewness, and bicentiles (B60, B70, B80 and B90) calculated using both a static (1 m) and a variable (50% of tree height) cut-off height. Significant differences are found between all disease severity categories, except in the case of healthy individuals (NI) and those in the early stages of infection (category 1). In addition, fragmentation metrics are shown to identify the increased patchiness and intra-crown height irregularities of CHMs associated with individual trees subject to heavy infection (category 3) of P. ramorum. Classifications using a k-nearest neighbour (k-NN) classifier and ALS point cloud metrics to classify disease presence/absence and severity yielded overall accuracies of 72% and 65% respectively. The results indicate that ALS can be used to identify individual tree crowns subject to moderate and heavy P. ramorum infection in larch forests. This information demonstrates the potential applications of ALS for the development of a targeted phytosanitary approach for the management of P. ramorum.
... Refs. [91,92] are more similar in their methodology but focus on a different response variable (woody cover). Ref. [91] used machine learning with ALS data to study dieback of trees in eucalyptus forests. ...
... Refs. [91,92] are more similar in their methodology but focus on a different response variable (woody cover). Ref. [91] used machine learning with ALS data to study dieback of trees in eucalyptus forests. A grid search was used for hyperparameter tuning and forward feature selection (FFS) for variable selection. ...
This study analyzed highly correlated, feature-rich datasets from hyperspectral remote sensing data using multiple statistical and machine-learning methods. The effect of filter-based feature selection methods on predictive performance was compared. In addition, the effect of multiple expert-based and data-driven feature sets, derived from the reflectance data, was investigated. Defoliation of trees (%), derived from in situ measurements from fall 2016, was modeled as a function of reflectance. Variable importance was assessed using permutation-based feature importance. Overall, the support vector machine (SVM) outperformed other algorithms, such as random forest (RF), extreme gradient boosting (XGBoost), and lasso (L1) and ridge (L2) regressions by at least three percentage points. The combination of certain feature sets showed small increases in predictive performance, while no substantial differences between individual feature sets were observed. For some combinations of learners and feature sets, filter methods achieved better predictive performances than using no feature selection. Ensemble filters did not have a substantial impact on performance. The most important features were located around the red edge. Additional features in the near-infrared region (800–1000 nm) were also essential to achieve the overall best performances. Filter methods have the potential to be helpful in high-dimensional situations and are able to improve the interpretation of feature effects in fitted models, which is an essential constraint in environmental modeling studies. Nevertheless, more training data and replication in similar benchmarking studies are needed to be able to generalize the results.
... ROD affects the vasculature of the native ohi'a (Metrosideros polymorpha), the most abundant tree across the Hawai'ian islands, and the spectral signature of infected ohi'a trees is consistently different than their healthy counterparts [55]. Other case studies using imaging spectroscopy to monitor tree health successfully identified declining floodplain eucalyptus trees [56] and basal stem rot in oil palm plantations [57]. ...
... ROD affects the vasculature of the native ohi'a (Metrosideros polymorpha), the most abundant tree across the Hawai'ian islands, and the spectral signature of infected ohi'a trees is consistently different than their healthy counterparts [55]. Other case studies using imaging spectroscopy to monitor tree health successfully identified declining floodplain eucalyptus trees [56] and basal stem rot in oil palm plantations [57]. . In order to map and monitor Rapid Ohi'a Death (ROD) across the landscape, (a) the brightness-normalized mean spectra of live, brown, and leafless ohi'a trees were extracted from Global Airborne Observatory (GAO) data. ...
As humans continue to alter Earth systems, conservationists look to remote sensing to monitor, inventory, and understand ecosystems and ecosystem processes at large spatial scales. Multispectral remote sensing data are commonly integrated into conservation decision-making frameworks, yet imaging spectroscopy, or hyperspectral remote sensing, is underutilized in conservation. The high spectral resolution of imaging spectrometers captures the chemistry of Earth surfaces, whereas multispectral satellites indirectly represent such surfaces through band ratios. Here, we present case studies wherein imaging spectroscopy was used to inform and improve conservation decision-making and discuss potential future applications. These case studies include a broad array of conservation areas, including forest, dryland, and marine ecosystems, as well as urban applications and methane monitoring. Imaging spectroscopy technology is rapidly developing, especially with regard to satellite-based spectrometers. Improving on and expanding existing applications of imaging spectroscopy to conservation, developing imaging spectroscopy data products for use by other researchers and decision-makers, and pioneering novel uses of imaging spectroscopy will greatly expand the toolset for conservation decision-makers.
... The combined use of airborne LiDAR scans with either multispectral or hyperspectral imagery previously also confirmed to substantially improve the estimates of important forest attributes (e.g. biomass, cover types and health) Shendryk et al., 2016;Meng et al., 2018), and we expect these results to be transferable to crop systems. ...
... Similarly, LiDAR scans, after pre-processing steps described in Sofonia et al. (2019a) and Sofonia et al. (2019b), were normalized with respect to the ground surface and used to extract 48 predictor variables within 2 m × 2 m biomass sampling plots (Fig. 2) in LasTools software (LAStools, 2015, Table 3), which previously confirmed to be useful in forest studies (Shendryk et al., 2016;Meng et al., 2018) and here were exclusively used for predicting sugarcane biomass. ...
Unmanned Aerial Vehicle (UAV) platforms and associated sensing technologies are extensively utilized in precision agriculture. Using LiDAR and imaging sensors mounted on multirotor UAVs, we can observe fine-scale crop variations that can help improve fertilizer management and maximize yields. In this study we used UAV mounted LiDAR and multispectral imaging sensors to monitor two sugarcane field trials with variable nitrogen (N) fertilization inputs in the Wet Tropics region of Australia. From six surveys performed at 42-day intervals, we monitored crop growth in terms of height, density and vegetation indices. In each survey period, we estimated a set of models to predict at-harvest biomass at fine scale (2m×2m plots). We compared the predictive performance of models based on multispectral predictors only, LiDAR predictors only, a fusion of multispectral and LiDAR predictors, and a normalized difference vegetation index (NDVI) benchmark. We found that predictive performance peaked early in the season, at 100–142 days after the previous harvest (DAH), and declined closer to the harvest date. At peak performance (i.e. 142 DAH), the multispectral model performed slightly better (R¯2=0.57) than the LiDAR model (R¯2=0.52), with both outperforming NDVI benchmark (R¯2=0.34). This, however, flipped later in the season, with LiDAR performing slightly better than the multispectral imaging and NDVI benchmark. Interestingly, the fusion model did not perform significantly better than the multispectral model at 100–142 DAH, nor better than LiDAR in subsequent periods. We also estimated a model to predict contemporaneous leaf N content (%) using multispectral imagery, which demonstrated an R¯2 of 0.57. Our results are of particular interest to nutrient management programs aiming to deliver N fertilizer guidelines for sustainable sugarcane production, as both fine-scale biomass and leaf N content predictions are feasible when management interventions are still possible (i.e. as early as at 100 DAH).
... Discrete LiDAR metrics have been used to inspect forest damage caused by such agents as mountain pine beetle (Coops et al., 2009) and pine sawfly (Solberg et al., 2006;Kantola et al., 2013). Potential capabilities of full-waveform LiDAR were shown by Shendryk et al. (2016), who were able to successfully classify tree health in terms of crown dieback in a eucalypt forest. ...
... Analysis based on those metrics yielded an accuracy of 80.9%, indicating a potential for using LiDAR metrics to detect disease-induced defoliation, possibly at a tree-level if spatial resolution allows. Recently Shendryk et al. (2016) performed a tree level health classification in terms of crown dieback in eucalyptus trees using full-waveform LiDAR data; the most successful metrics were based on pulse width and intensity. These metrics could also be transferable to analysis of red band needle blight defoliation, as they were found to be sensitive to crown dieback and transparency. ...
Current trends in research for detection of infections in forests almost exclusively involve the use of a single imaging sensor. However, combining information from a range of sensors could potentially enhance the ability to diagnose and quantify the infection. This study investigated the potential of combining hyperspectral and LiDAR data for red band needle blight detection. A comparative study was performed on the spectral signatures retrieved for two plots established in lodgepole pine stands and on a range of LiDAR metrics retrieved at individual tree-level. Leaf spectroscopy of green and partially chlorotic needles affected by red band needle blight highlighted the green, red and short near-infrared parts of the electromagnetic spectrum as the most promising. A good separation was found between the two pine stands using a number of spectral indices utilising those spectral regions. Similarly, a distinction was found when intra-canopy distribution of LiDAR returns was analysed. The percentage of ground returns within canopy extents and the height-normalised 50th percentile (height normalisation was performed to each tree's canopy extents) were identified as the most useful features among LiDAR metrics for separation of trees between the plots. Analysis based on those metrics yielded an accuracy of 80.9%, indicating a potential for using LiDAR metrics to detect disease-induced defoliation. Stepwise discriminant function analysis identified Enhanced Vegetation Index, Normalised Green Red Difference Index, percentage of ground returns, and the height-normalised 50th percentile to be the best predictors for detection of changes in the canopy resulting from red band needle blight. Using a combination of these variables led to a substantial decrease of unexplained variance within the data and an improvement in discrimination accuracy (96.7%). The results suggest combining information from different sensors can improve the ability to detect red band needle blight.
... Tulbure et al. (2016) was also the only published product with a statistical accuracy assessment that created confidence for its use in follow on applications (e.g. Bishop-Taylor et al., 2015, Bishop-Taylor et al., 2017a, 2017b, Bishop-Taylor et al., 2018Heimhuber et al., 2015Heimhuber et al., , 2017Shendryk et al., 2016b). The statistical accuracy assessment met two key criteria. ...
... We quantified seasonal vegetation dynamics using the EVI, hence change in vegetation chlorophyll content and leaf area index and by using standard anomalies we accounted for different vegetation densities. However, vegetation response to flooding may also vary as a function of vegetation structure, type, composition, health, and land use such as irrigation Broich et al., 2014;Kaptué et al., 2015;Nightingale and Phinn, 2003;Shendryk et al., 2016aShendryk et al., , 2016bWestbrooke and Florentine, 2005). Time series of soil moisture content for the entire MDB with sufficient accuracy and spatial detail were also unavailable (Heimhuber et al., 2017;Heimhuber et al., 2015) as were time series of groundwater depth. ...
... Tulbure et al. (2016) was also the only published product with a statistical accuracy assessment that created confidence for its use in follow on applications (e.g. Bishop-Taylor et al., 2015, Bishop-Taylor et al., 2017a, 2017b, Bishop-Taylor et al., 2018Heimhuber et al., 2015Heimhuber et al., , 2017Shendryk et al., 2016b). The statistical accuracy assessment met two key criteria. ...
... We quantified seasonal vegetation dynamics using the EVI, hence change in vegetation chlorophyll content and leaf area index and by using standard anomalies we accounted for different vegetation densities. However, vegetation response to flooding may also vary as a function of vegetation structure, type, composition, health, and land use such as irrigation Broich et al., 2014;Kaptué et al., 2015;Nightingale and Phinn, 2003;Shendryk et al., 2016aShendryk et al., , 2016bWestbrooke and Florentine, 2005). Time series of soil moisture content for the entire MDB with sufficient accuracy and spatial detail were also unavailable (Heimhuber et al., 2017;Heimhuber et al., 2015) as were time series of groundwater depth. ...
Australia is a continent subject to high rainfall variability, which has major influences on runoff and vegetation dynamics. However, the resulting spatial-temporal pattern of flooding and its influence on riparian vegetation has not been quantified in a spatially explicit way. Here we focused on the floodplains of the entire Murray-Darling Basin (MDB), an area that covers over 1M km<sup>2</sup>, as a case study. The MDB is the country’s primary agricultural area with scarce water resources subject to competing demands and impacted by climate change and more recently by the Millennium Drought (1999–2009). Riparian vegetation in the MDB floodplain suffered extensive decline providing a dramatic degradation of riparian vegetation.
We quantified the spatial-temporal impact of rainfall, temperature and flooding patters on vegetation dynamics at the subcontinental to local scales and across inter to intra-annual time scales based on three decades of Landsat (25k images), Bureau of Meteorology data and one decade of MODIS data.
Vegetation response varied in space and time and with vegetation types, densities and location relative to areas frequently flooded. Vegetation degradation trends were observed over riparian forests and woodlands in areas where flooding regimes have changed to less frequent and smaller inundation extents. Conversely, herbaceous vegetation phenology followed primarily a ‘boom’ and ‘bust’ cycle, related to inter-annual rainfall variability. Spatial patters of vegetation degradation changed along the N-S rainfall gradient but flooding regimes and vegetation degradation patterns also varied at finer scale, highlighting the importance of a spatially explicit, internally consistent analysis and setting the stage for investigating further cross-scale relationships.
Results are of interest for land and water management decisions. The approach developed here can be applied to other areas globally such as the Nile river basin and Okavango River delta in Africa or the Mekong River Basin in Southeast Asia.
... A similar methodology was then applied to identify herbaceous and shrub species on the Vistula River, Poland [57]. A few other studies were able to identify tree species or to map the health of individual trees in mature forest floodplains [58][59][60]. However, the forests analyzed in these studies mostly featured hardwood and evergreen species belonging to later successional stages and were not subject to the mosaics of age and hydrological connectivity that can be found in riparian corridors. ...
Riparian forests are complex ecosystems shaped by their connectivity to a river system, which produces a mosaic of ages and species. Because of increasing anthropic pressure from factors such as damming or climate change, they are often endangered and suffer from a drop in groundwater accessibility and increased water stress. By combining hyperspectral, LiDAR, and forestry datasets along a 20 km corridor of the Ain River, this paper assesses the ability of remote sensing to characterize and monitor such environments. These datasets are used to investigate changes in site conditions and forest characteristics, such as height and canopy water content, along a gradient of ecosystem ages and for reaches under distinct geomorphic conditions (shifting, sediment-starved, incised). The data show that, over time, forest patches aggrade, and the forest grows and becomes more post-pioneer. However, forest patches that are located in the incised reach aggrade more and appear to be less developed in height, more stressed, and feature species compositions reflecting dryer conditions, in comparison with better-connected patches of the same age. Random forest analysis was applied to predict the indicators of forest connectivity with remotely sensed LIDAR and hyperspectral data, in order to identify the spatial trends at the reach scale and compare them with the geomorphic segmentation of the river. The random forest classifications achieved an accuracy between 80% and 90% and resulted in spatial trends that highlighted the differences in hydrological connectivity between differing geomorphic conditions. Overall, remote sensing appears to be a good tool for characterizing the impact of channel incisions and adjustments on riparian forest conditions by identifying the locations of dryer forest patches. In addition, good accuracy was achieved when attempting to classify these forest patches, even when using hyperspectral data alone, which suggests that satellite data could become a powerful tool for monitoring the health of riparian forests, in the context of increasing anthropic pressures.
... The spectral signature of a single species comprising its wide range of structural elements, such as leaves, branches, stem, and bark, is hard to define [12]. As tree crowns are much larger than the pixel size, Object Based Image Analysis (OBIA) using individual tree canopies is often deemed more appropriate than pixel-level classification [13]. Tree species classification therefore relies on grouping pixels into tree canopy objects using tree detection and tree-crown delineation [14]. ...
Information on tree species and changes in forest composition is necessary to understand species-specific responses to change, and to develop conservation strategies. Remote sensing methods have been increasingly used for tree detection and species classification. In mixed species forests, conventional tree detection methods developed with assumptions about uniform tree canopy structure often fail. The main aim of this study is to identify effective methods for tree delineation and species classification in an Australian native forest. Tree canopies were delineated at three different spatial scales of analysis: (i) superpixels representing small elements in the tree canopy, (ii) tree canopy objects generated using a conventional segmentation technique, multiresolution segmentation (MRS), and (iii) individual tree bounding boxes detected using deep learning based on the DeepForest open-source algorithm. Combinations of spectral, texture, and structural measures were tested to assess features relevant for species classification using RandomForest. The highest overall classification accuracies were achieved at the superpixel scale (0.84 with all classes and 0.93 with Eucalyptus classes grouped). The highest accuracies at the individual tree bounding box and object scales were similar (0.77 with Eucalyptus classes grouped), highlighting the potential of tree detection using DeepForest, which uses only RGB, compared to site-specific tuning with MRS using additional layers. This study demonstrates the broad applicability of DeepForest and superpixel approaches for tree delineation and species classification. These methods have the potential to offer transferable solutions that can be applied in other forests.
... Nonetheless, in species-rich floodplain areas, it remains challenging to obtain high-accurate classification models, even using hyperspectral imagery (Richter et al., 2016), mainly due to the mixture of vegetation and other types of land covers , but also due to the variations in atmospheric and light conditions. Combining visible, multi-or hyperspectral images with elevation data has been proven to build accurate models for classification of urban trees (Alonzo et al., 2014), conifer species (Scholl et al., 2021;Weinstein et al.;, plant communities in wetlands (Martínez Prentice et al., 2021), grasslands of river valleys (Demarchi et al., 2020), fens (Szporak-Wasilewska et al., 2021) and riparian vegetation (Shendryk et al., 2016;Peerbhay et al., 2021). ...
The seasonal occurrence of plant communities in rivers’ reaches has a direct impact on the flow resistance and sediment dynamics. Monitoring the occurrence of plant communities and their habitats can be of great importance to understand bio-geomorphological changes and hydrological processes and to assess human-induced changes. This work focuses on combining different remote sensors’ data acquired from an unmanned aerial vehicle (UAV) to feed Random Forest models for the recognition of plant communities in reaches of the Vistula River. Botanical surveys were carried out on more than two thousand field plots along the reaches, each being manually classified into nine different communities. Hyperspectral and RGB (Red-Green-Blue) images were collected with UAV over the botanical plots and merged with a LiDAR-based (Light Detection and Ranging) canopy height model. The modelling strategy consisted of fitting Random Forests using uncorrelated scores of principal components. A novel approach is presented to select discriminant features in the presence of high correlations after applying a ridge regularization on the inverse of the covariance matrix. We show how specific combinations of sensors’ features can impact the model’s accuracy, which reached more than 90% for dominating shrubs and trees such as Salicetum triandro-viminalis, Salicetum albo-fragilis and Chelidonio-Aceretum. On the other hand, the fitted model was not as accurate to classify plant communities such as Agropyretalia and Calamagrostietum.
... The elapsed time between transmission and reception is recorded and combined with positional information, resulting in detailed point clouds containing intensity and elevation measurements [32]. Important forest health indicators such as tree crown density, pattern distribution, or structural changes over time can be derived from LiDAR-based point clouds and indicate defoliation [57][58][59] or changes in crown architecture [60]. Besides FHM, these sensors are used in various other forestry applications using drones. ...
In recent years, technological advances have led to the increasing use of unmanned aerial vehicles (UAVs) for forestry applications. One emerging field for drone application is forest health monitoring (FHM). Common approaches for FHM involve small-scale resource-extensive fieldwork combined with traditional remote sensing platforms. However, the highly dynamic nature of forests requires timely and repetitive data acquisition, often at very high spatial resolution, where conventional remote sensing techniques reach the limits of feasibility. UAVs have shown that they can meet the demands of flexible operation and high spatial resolution. This is also reflected in a rapidly growing number of publications using drones to study forest health. Only a few reviews exist which do not cover the whole research history of UAV-based FHM. Since a comprehensive review is becoming critical to identify research gaps, trends, and drawbacks, we offer a systematic analysis of 99 papers covering the last ten years of research related to UAV-based monitoring of forests threatened by biotic and abiotic stressors. Advances in drone technology are being rapidly adopted and put into practice, further improving the economical use of UAVs. Despite the many advantages of UAVs, such as their flexibility, relatively low costs, and the possibility to fly below cloud cover, we also identified some shortcomings: (1) multitemporal and long-term monitoring of forests is clearly underrepresented; (2) the rare use of hyperspectral and LiDAR sensors must drastically increase; (3) complementary data from other RS sources are not sufficiently being exploited; (4) a lack of standardized workflows poses a problem to ensure data uniformity; (5) complex machine learning algorithms and workflows obscure interpretability and hinders widespread adoption; (6) the data pipeline from acquisition to final analysis often relies on commercial software at the expense of open-source tools.
... Remote sensing data sources are used to manage cropland (Pinter et al. 2003), to investigate urban forest health (Näsi et al. 2018), to measure spatial and temporal changes in natural forest cover, in order to investigate forest loss (deforestation) and forest gain (Borrelli et al. 2016;Hansen et al. 2013;Kuemmerle et al. 2009) and or natural tree health (Lévesque and King, 2003;Shendryk et al. 2016). Satellite image data were used to investigate agriculture-vegetated land (cropland) in Phalke et al. (2020). ...
Forests are important to people, wildlife and climate. Yet, not all forests are healthy throughout time. Unhealthy forests are providing fewer services and productions to people, harbouring less biodiversity and regulating less climate. Here, the preliminary findings are presented in a literature review on remote sensing measuring the changes in forest cover and the health of forest and of trees in Southeastern Europe including Albania, Bosnia and Herzegovina, Croatia, Montenegro and Slovenia. The aim is to assess the publications that applied remote sensing data sources to investigate the changes in forest cover and forest health and tree health in the five Southeastern European countries by searching in Scopus and Google scholar. There is a higher number of studies applied to remote sensing data sources investigating forest cover change (92.4 percent) compared to forest health (6.7 percent) and tree health (0.8 percent) for five countries in Southeastern Europe. There was a disparity of remote sensing data source studies on forest cover change, forest health and tree health amongst five countries. Croatia and Slovenia lead by 68.9 percent and Albania, Bosnia and Herzegovina and Montenegro by 31.1 percent of publications, according to Scopus and Google scholar. There were found no remote sensing data source studies on forest cover, forest health and tree health including all five countries altogether. A way to move forward is to increase cooperation between researchers, academic organisations and policy-makers amongst the five countries.
... This is because the narrow band information of hyperspectral images strongly correlates with the physiological state of plants but not LiDAR signal. [52]. Hence, early monitoring using LiDAR alone is challenging. ...
Beijing’s One Million Mu Plain Afforestation Project involves planting large areas with the exotic North American tree species Fraxinus pennsylvanica Marsh (ash). As an exotic tree species, ash is very vulnerable to infestations by the emerald ash borer (EAB), a native Chinese wood borer pest. In the early stage of an EAB infestation, attacked trees show no obvious sign. Once the stand has reached the late damage stage, death occurs rapidly. Therefore, there is a need for efficient early detection methods of EAB stress over large areas. The combination of unmanned aerial vehicle (UAV)-based hyperspectral imaging (HI) with light detection and ranging (LiDAR) is a promising practical approach for monitoring insect disturbance. In this study, we identified the most useful narrow-band spectral HI data and 3D LiDAR data for the early detection of EAB stress in ash. UAV-HI data of different infested stages (healthy, light, moderate and severe) of EAB in the 400–1000 nm range were collected from ash canopies and were processed by Partial Least Squares–Variable Importance in Projection (PLS-VIP) to identify the maximally sensitive bands. Band R678 nm had the highest PLS-VIP scores and the most robust classification ability. We combined this band with band R776 nm to develop an innovative normalized difference vegetation index (NDVI(776,678)) to estimate EAB stress. LiDAR data were used to segment individual trees and supplement the HI data. The new NDVI(776,678) identified different stages of EAB stress, with a producer’s accuracy of 90% for healthy trees, 76.25% for light infestation, 58.33% for moderate infestation, and 100% for severe infestation, with an overall accuracy of 82.90% when combined with UAV-HI and LiDAR.
... For example, researchers developed maps of surface water and flooding extent dynamics (SWD) using statistically representative measures of accuracy based on the entire Landsat archive across one of the largest agricultural dryland basins in the world . They then used the SWD maps together with time series of vegetation dynamics to quantify space time patterns of vegetation response to flooding (Broich et al., 2018;Shendryk et al., 2016), to assess drivers of flooding dynamics across space and time (Heimhuber et al., 2016(Heimhuber et al., , 2017, and to identify surface water habitats that play central roles as biodiversity hubs for water dependent biota (Bishop-Taylor et al., 2015, 2017. This type of work, while technically possible at global scales, will prove less insightful due to generalizations that will need to be made even if enough budget is available for their global development. ...
Unprecedented amounts of analysis‐ready Earth Observation (EO) data, combined with increasing computational power and new algorithms, offer novel opportunities for analysing ecosystem dynamics across large geographic extents, and to support conservation planning and action. Much research effort has gone into developing global EO‐based land‐cover and land‐use datasets, including tree cover, crop types, and surface water dynamics. Yet there are inherent trade‐offs between regional and global EO products pertaining to class legends, availability of training/validation data, and accuracy. Acknowledging and understanding these trade‐offs is paramount for both developing EO products and for answering science questions relevant for ecology or conservation studies based on these data. Here we provide context on the development of global EO‐based land‐cover and land‐use datasets, and outline advantages and disadvantages of both regional and global datasets. We argue that both types of EO‐derived land‐cover datasets can be preferable, with regional data providing the context‐specificity that is often required for policy making and implementation (e.g., land‐use and management, conservation planning, payment schemes for ecosystem services), making use of regional knowledge, particularly important when moving from land cover to actors. Ensuring that global and regional land‐cover and land‐use products derived based on EO data are compatible and nested, both in terms of class legends and accuracy assessment, should be a key consideration when developing such data. Open access high‐quality training and validation data derived as part of such efforts are of utmost importance. Likewise, global efforts to generate sets of essential variables for climate change, biodiversity, or eventually land use, which often require land‐cover maps as inputs, should consider regionalized, hierarchical approaches to not sacrifice regional context. Global change impacts manifest in regions, and so must the policy and planning responses to these challenges. EO data should embrace that regions matter, perhaps more than ever, in an age of global data availability and processing.
... This makes Lidar a possible candidate to improve measurement accuracy. Although Lidar data failed to accurately reflect the biochemical condition of tree crowns (e.g., Liu et al. 2017;Shi et al. 2018), it can be used as measure auxiliary data that produces three-dimensional tree canopy structures (e.g., Shendryk et al. 2016). Thus, combining Lidar with hyperspectral data for individual tree segmentation could improve accuracy (e.g., Junttila et al. 2019;Lin et al. 2019). ...
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.
... According to Xiaoling Deng et al. [36,37] machine learning has been used to set several benchmarks in the field of agriculture. W. Yao et al. [35] and Shendryk et al. [38] published their prior work on the identification of dead trees is performed by individual tree crown segmentation prior to the health assessment. Meng R. et al. [39], Shendryk et al. [30], López-López M et al. [40], Barnes et al. [41], Fassnacht et al. [42], mentioned that most of the current tree health studies centred either on evaluating the defoliation of the tree crown or the overall health status of the tree, although there was minimal exposure to the discolouration of the tree crown. ...
Urban greenery is an essential characteristic of the urban ecosystem, which offers various advantages, such as improved air quality, human health facilities, storm-water run-off control, carbon reduction, and an increase in property values. Therefore, identification and continuous monitoring of the vegetation (trees) is of vital importance for our urban lifestyle. This paper proposes a deep learning-based network, Siamese convolutional neural network (SCNN), combined with a modified brute-force-based line-of-bearing (LOB) algorithm that evaluates the health of Eucalyptus trees as healthy or unhealthy and identifies their geolocation in real time from Google Street View (GSV) and ground truth images. Our dataset represents Eucalyptus trees’ various details from multiple viewpoints, scales and different shapes to texture. The experiments were carried out in the Wyndham city council area in the state of Victoria, Australia. Our approach obtained an average accuracy of 93.2% in identifying healthy and unhealthy trees after training on around 4500 images and testing on 500 images. This study helps in identifying the Eucalyptus tree with health issues or dead trees in an automated way that can facilitate urban green management and assist the local council to make decisions about plantation and improvements in looking after trees. Overall, this study shows that even in a complex background, most healthy and unhealthy Eucalyptus trees can be detected by our deep learning algorithm in real time.
... Although sufficient studies exist on deriving individual tree attributes (and even tree health) from remote-sensing data [50,68,72,78], to the best of our knowledge, this study is the first to use satellite-based SVIs as objective comparators to urban forest soil health. ...
Increasing recognition of the potential ecosystem services provided by urban forests suggests a need to examine soil quality under urban conditions. Soil quality assessment tools are presently mostly applied in agricultural production, but these approaches must also be evaluated in the urban context. This proof-of-concept exploratory study evaluates whether Worldview-3 spectral vegetation indices (SVIs) generated for individual tree crown (ITC) objects can be correlated to soil health attributes measured in the field in Metro Boston, Massachusetts, USA. While similar studies have completed such analysis for agricultural crops, none have done so for urban trees. The statistical analysis by Pearson correlation and principal component analysis (PCA) showed that SVIs, specifically the Normalized Difference Vegetation Index (NDVI), correlated significantly and positively with bulk density (BD) (r = .536) and soil luminance (r = .562) and negatively with CO2 respiration (r = −.536), active fungi and active bacteria (r = −.401), and total carbon (r = −.548). The negative correlations with parameters commonly considered positive for soil health in agricultural settings may indicate strong perturbation at the urban soil surface level; they also suggest soil health attributes measured at this study’s 0–15 cm sampling depth may not be satisfactorily indicative of tree health as measured by SVIs. This study evidences the ground truthing of satellite-based urban SVIs, including their relationships with soil health attributes at the individual tree level.
... Within the context of forestry activities, several studies have shown the usefulness of ML tools to estimate a variety of response variables, i.e. prediction of basal area, volume, and modeling of radial growth increment (Laganis et al. 2008;Dos Reis et al. 2018), modeling lumber yield based on laser scanning metrics (Görgens et al. 2015;Li et al. 2016), prediction of leaf area index (Omer et al. 2016), and wildfire propagation (Lauer et al. 2017). Additionally, ML techniques have been widely used to study aerial imagery collected using remote sensing techniques (Shendryk et al. 2016;Vargas-Larreta et al. 2017;Diamantopoulou et al. 2018;Dos Reis et al. 2018;Li et al. 2018;Pohjankukka et al. 2018;Urbazaev et al. 2018). However, there are no publications addressing the problem of estimating dry biomass weight from data on green logging residue chips by means of ML algorithms. ...
This work shows how modern machine learning techniques can be used to solve current problems faced by the forestry industry. More specifically, the focus is on comparing the predictive performance of several algorithms on estimating the dry weight, in tons, of chip residues. The dataset contains samples obtained during 22 months from 220 trucks coming from 17 different farms located within the area spanned by the Biobío and Maule regions, Chile. Once the trucks arrived, samples were collected and dried to compute the dry tons carried by each truck, which was set as the dependent variable. Using open-source software implementations of state-of-the-art algorithms it was possible to determine, for our data, that even though the non-parametric models Gradient Boosting Machines (GBM) and Neural Networks (NNET) outperformed the linear regression (LM) model, they are not statistically superior to the LASSO regression (GLMNET), an improved version of the LM model. Additionally, it was observed that seasonality affects the ratio of green tons to dry tons a truck can deliver to a power plant during the year. Finally, the continuous variables green tons, elevation, east and north (longitude-latitude) also contribute to explaining the dependent variable.
... There are, however, advances in our capacity to monitor responses to environmental flows, including remote sensing and GIS modeling that increase managers' capacity to generate basin-scale data. For example, remote sensing can now support river morphology (Belletti et al., 2017), organic matter input (Hoffmann et al., 2016) and individual tree condition (Shendryk et al., 2016) assessments. Improvements in monitoring technology are also likely to improve opportunities for extensive on ground monitoring through increased community participation, telemetry and techniques such as environmental DNA. ...
Adaptive management is central to improving outcomes of environmental water delivery. The Australian Government's Murray−Darling Basin (MDB) Plan 2012 explicitly states that adaptive management should be applied in the planning, prioritisation and use of environmental water. A Long Term Intervention Monitoring (LTIM) program was established in 2014 to evaluate responses to environmental water delivery for seven Areas within the MDB, with evaluation also undertaken at the Basin scale. Adaptive management at the Area scale was assessed using two approaches: (a) through a reflective exercise undertaken by researchers, water managers and community members and (b) through an independent review and evaluation of the program, where relevant reports were reviewed and managers and researchers involved in the LTIM program were interviewed. Both assessment approaches revealed that the scale of management actions influenced the extent to which learnings were incorporated into subsequent actions. Although there were many examples where learnings within an Area had been used to adaptively manage subsequent environmental water deliveries within that Area, there was inconsistent documentation of the processes for incorporating learnings into decision making. Although this likely limited the sharing of learnings, there were also examples where learnings from one Area had influenced environmental water management in another, suggesting that sharing between concurrent projects can increase learning. The two assessments identified ways to improve and systematically document the adaptive management learnings. With improved processes to increase reflection, documentation and sharing of learnings across projects, there is an opportunity to improve management of environmental water and ecosystem outcomes.
... High spectral resolution satellite data like Sentinel-2 and WorldView02 should be tested for cost-efficient large-area stress detection in time series. The combination with LiDAR attributes can also improve stress detection, especially for more advanced stress levels with textural and structural changes in the canopy caused by defoliation [131][132][133][134]. The challenging sub-pixel co-location between optical and LiDAR datasets might be overcome with new multispectral waveform LiDAR sensors that allow the direct combination of LiDAR data with multispectral wavelengths in one dataset [135]. ...
The endemic New Zealand kauri trees (Agathis australis) are under threat by the deadly kauri dieback disease (Phytophthora agathidicida (PA)). This study aimed to identify spectral index combinations for characterising visible stress symptoms in the kauri canopy. The analysis is based on an aerial AISA hyperspectral image mosaic and 1258 reference crowns in three study sites in the Waitakere Ranges west of Auckland. A field-based assessment scheme for canopy stress symptoms (classes 1–5) was further optimised for use with RGB aerial images. A combination of four indices with six bands in the spectral range 450–1205 nm resulted in a correlation of 0.93 (mean absolute error 0.27, RMSE 0.48) for all crown sizes. Comparable results were achieved with five indices in the 450–970 nm region. A Random Forest (RF) regression gave the most accurate predictions while a M5P regression tree performed nearly as well and a linear regression resulted in slightly lower correlations. Normalised Difference Vegetation Indices (NDVI) in the near-infrared / red spectral range were the most important index combinations, followed by indices with bands in the near-infrared spectral range from 800 to 1205 nm. A test on different crown sizes revealed that stress symptoms in smaller crowns with denser foliage are best described in combination with pigment-sensitive indices that include bands in the green and blue spectral range. A stratified approach with individual models for pre-segmented low and high forest stands improved the overall performance. The regression models were also tested in a pixel-based analysis. A manual interpretation of the resulting raster map with stress symptom patterns observed in aerial imagery indicated a good match. With bandwidths of 10 nm and a maximum number of six bands, the selected index combinations can be used for large-area monitoring on an airborne multispectral sensor. This study establishes the base for a cost-efficient, objective monitoring method for stress symptoms in kauri canopies, suitable to cover large forest areas with an airborne multispectral sensor.
... Classifications at tree level, while working with a native forest is a challenge since tree delineation is usually performed before health assessment [22,23]. Tree delineation can be achieved by firstly detecting local maxima from the Canopy Height Model (CHM) and then segmenting CHM into individual trees with the watershed algorithm [46]. ...
In southern Australia, many native mammals and birds rely on hollows for sheltering,while hollows are more likely to exist on dead trees. Therefore, detection of dead trees could beuseful in managing biodiversity. Detecting dead standing (snags) versus dead fallen trees (CoarseWoody Debris—CWD) is a very different task from a classification perspective. This study focuseson improving detection of dead standing eucalypt trees from full-waveform LiDAR. Eucalypt treeshave irregular shapes making delineation of them challenging. Additionally, since the study area is anative forest, trees significantly vary in terms of height, density and size. Therefore, we need methodsthat will be resistant to those challenges. Previous study showed that detection of dead standingtrees without tree delineation is possible. This was achieved by using single size 3D-windows toextract structural features from voxelised full-waveform LiDAR and characterise dead (positivesamples) and live (negative samples) trees for training a classifier. This paper adds on by proposingthe usage of multi-scale 3D-windows for tackling height and size variations of trees. Both thesingle 3D-windows approach and the new multi-scale 3D-windows approach were implementedfor comparison purposes. The accuracy of the results was calculated using the precision and recallparameters and it was proven that the multi-scale 3D-windows approach performs better than thesingle size 3D-windows approach. This open ups possibilities for applying the proposed approachon other native forest related applications.
... Multi-temporal satellite images and unmanned aerial data products have been extensively and professionally used in forest disaster analysis at various spatial scales, such as QuickBird (Gaulton et al., 2013), Sentinel-2 (Hawryło et al., 2018), TerraSAR-X (Ortiz et al., 2013), and unmanned aerial vehicle images (Näsi et al., 2015;Sankey et al., 2016). With the development of light detection and ranging (LiDAR) technology, especially airborne laser scanning (ALS), the potential of remote sensing for the monitoring of forest health has been furthered explored and applied (Kim et al., 2009;Kantola et al., 2010;Kantola et al., 2013;Vastaranta et al., 2013;Junttila et al., 2015;Shendryk et al., 2016;Vescovo et al., 2016;Junttila et al., 2017). ...
Forest disruption caused by pest insects is a common disaster occurring in plantations, and is a threatening factor to forest health. Therefore, a precise method for monitoring individual tree health and estimating disturbance severity is urgently needed. Theoretically, terrestrial laser scanning (TLS) is a promising tool in high resolution remote sensing, which can provide information regarding the structural change of the affected trees with millimeter precision. However, few studies have explored the potential of TLS application in this field, especially when using only mono-temporal data. In this study, a single-scan TLS data-based method was developed and validated to classify defoliation at both individual-tree scale and plot scale. The objects were classified into three classes: healthy/slightly defoliated, moderately defoliated and severely defoliated. Sixty features were extracted from TLS data and optimized to six (individual-tree scale) and five (plot scale) explanatory variables by using a Random Forest method to accomplish the classification. By this approach, individual trees can be classified into three defoliation levels with 80% overall accuracy (kappa value 0.70), while plot-scale classification had 94% overall accuracy (kappa value 0.91). Point distribution characteristics proposed in this method were among the most important features for defoliation estimation. Evidently, the methods presented in this study are capable of providing satisfactory estimates of defoliation severity, and supporting a precise inventory and monitoring of forest health.
... The ground features in this study site mainly include: mangroves, low grasses, soil or mud, and water. Because the ITDD algorithms do not distinguish mangroves and non-mangroves, a mask is created to ensure that only the height pixels of mangrove canopies are input to the ITDD algorithms (Shendryk et al., 2016;Zhen et al., 2014). For the LiDAR derived CHM, the mask is created using a height threshold (Ota et al., 2015), because the other features are generally lower than the mangroves. ...
... The MDB has close to 30,000 wetlands, which retain significant biodiversity, with 16 listed as protected wetlands of international importance under the Ramsar convention and 200 wetlands of national significance (Australian Government Department of Environment and Energy, 2012). These wetlands and their riparian vegetation receive water from intermittent river flow, exhibit high variability and change in condition in response to water availability Shendryk et al., 2016) and are important for landscape connectivity (Bishop-Taylor et al., 2018, 2015. The climate of the basin is highly variable, with average annual rainfall of 470 mm, ranging from summer to winter dominant in the northern and southern basin, respectively. ...
Spatiotemporal distribution and systematic quantification of surface water and their drivers of change are critical. However, quantifying this distribution is challenging due to a lack of spatially explicit and temporally dynamic em- pirical data of both surface water and its drivers of change at large spatial scales. We focused on one of the largest dryland basins in the world, Australia's Murray-Darling Basin (MDB), recently identified as a global hotspot of water decline. We used a new remotely sensed time-series of surface water extent dynamics (SWD) data to quantify spatiotemporal patterns in surface water across the entire MDB and catchments and to assess natural and anthropo- genic drivers of SWD, including climate and historical land use change. We show high intra- and inter-annual dy- namics in surface water with a rapid loss during the Millennium Drought, the worst, decade-long drought in SE Australia. We show strong regional and catchment differences in SWD, with the northern basin showing high var- iability compared to the southern basin which shows a steady decline in surface water. Linear mixed effect models including climate and land-use change variables explained up to 70% variability in SWD with climate being more important in catchments of the northwestern MDB, whereas land-use was important primarily in the central MDB. Increase in fraction of dryland agriculture in a catchment and maximum temperature was negatively related to SWD, whereas precipitation and soil moisture were positively related to SWD. The fact that land-use change was an important explanatory variable of SWD in addition to climate is a significant result as land-use can be managed more effectively whereas climate-mitigation actions can be intractable, with global change scenarios predicting drier conditions for the area followed by a further reduction in surface water availability.
... There are, however, advances in our capacity to monitor responses to environmental flows, including remote sensing and GIS modeling that increase managers' capacity to generate basin-scale data. For example, remote sensing can now support river morphology (Belletti et al., 2017), organic matter input (Hoffmann et al., 2016) and individual tree condition (Shendryk et al., 2016) assessments. Improvements in monitoring technology are also likely to improve opportunities for extensive on ground monitoring through increased community participation, telemetry and techniques such as environmental DNA. ...
Environmental flows are a critical tool for addressing ecological degradation of river systems brought about by increasing demand for limited water resources. The importance of basin scale management of environmental flows has long been recognized as necessary if managers are to achieve social, economic, and environmental objectives. The challenges in managing environmental flows are now emerging and include the time taken for changes to become manifest, uncertainty around large-scale responses to environmental flows and that most interventions take place at smaller scales. The purpose of this paper is to describe how conceptual models can be used to inform the development, and subsequent evaluation of ecological objectives for environmental flows at the basin scale. Objective setting is the key initial step in environmental flow planning and subsequently provides a foundation for effective adaptive management. We use the implementation of the Basin Plan in Australia's Murray-Darling Basin (MDB) as an example of the role of conceptual models in the development of environmental flow objectives and subsequent development of intervention monitoring and evaluation, key steps in the adaptive management of environmental flows. The implementation of the Basin Plan was based on the best science available at the time, however, this was focused on ecosystem responses to environmental flows. The monitoring has started to reveal that limitations in our conceptualization of the basin may reduce the likelihood of achieving of basin scale objectives. One of the strengths of the Basin Plan approach was that it included multiple conceptual models informing environmental flow management. The experience in the MDB suggests that the development of multiple conceptual models at the basin scale will help increase the likelihood that basin-scale objectives will be achieved.
... For example, imagery can help with crown detection and stand classification (species, health status), while LiDAR can provide an estimation of structural parameters (e.g., height, DBH, etc.). A recent case study reached an overall accuracy of 81% in the identification of dieback-affected eucalyptus trees within a floodplain forest in Australia [96] through the use of LiDAR data and imaging spectroscopy. The combination/fusion of ALS with TLS data has already demonstrated a positive application for single-tree inventory in Finland [97], but the delineation and quantification of downed logs should be implemented [81]. ...
LiDAR technology is finding uses in the forest sector, not only for surveys in producing forests but also as a tool to gain a deeper understanding of the importance of the three-dimensional component of forest environments. Developments of platforms and sensors in the last decades have highlighted the capacity of this technology to catch relevant details, even at finer scales. This drives its usage towards more ecological topics and applications for forest management. In recent years, nature protection policies have been focusing on deadwood as a key element for the health of forest ecosystems and wide-scale assessments are necessary for the planning process on a landscape scale. Initial studies showed promising results in the identification of bigger deadwood components (e.g., snags, logs, stumps), employing data not specifically collected for the purpose. Nevertheless, many efforts should still be made to transfer the available methodologies to an operational level. Newly available platforms (e.g., Mobile Laser Scanner) and sensors (e.g., Multispectral Laser Scanner) might provide new opportunities for this field of study in the near future.
... LiDAR sensors are a good tool for acquiring dense point cloud data for the purpose of surveying in short ranges. These sensors measure the distance by timing a laser pulse reflected from a target and have been applied in a number of remote sensing applications ranging from mapping (Schwarz, 2010), landslide investigations (Jaboyedoff et al., 2012) to tree inventories (Shendryk et al., 2016b). LiDAR systems are particularly suited to surveying forest canopies due to their active sensors and their ability to penetrate canopies. ...
... As a result, monitoring and estimating forest damage by diseases and insect pests are becoming more precise and timely. However, most researchers have concentrated on the identification, classification, and assessment for forest damage with finer vegetation indexes and models [33][34][35][36][37]. There is still little research on the prediction of damage with the change of climate and other factors. ...
Pine forests (Pinus tabulaeformis) have been in danger of defoliation by a caterpillar in the west Liaoning province of China for more than thirty years. This paper aims to assess and predict the degree of damage to pine forests by using remote sensing and ancillary data. Through regression analysis of the pine foliage remaining ratios of field plots with several vegetation indexes of Landsat data, a feasible inversion model was obtained to detect the degree of damage using the Normalized Difference Infrared Index of 5th band (NDII5). After comparing the inversion result of the degree of damage to the pine in 29 years and the historical damage record, quantized results of damage assessment in a long time-series were accurately obtained. Based on the correlation analysis between meteorological variables and the degree of damage from 1984 to 2015, the average degree of damage was predicted in temporal scale. By adding topographic and other variables, a linear prediction model in spatiotemporal scale was constructed. The spatiotemporal model was based on 5015 public pine points for 24 years and reached 0.6169 in the correlation coefficient. This paper provided a feasible and quantitative method in the spatiotemporal prediction of forest pest occurrence by remote sensing.
... Previous work on dead standing trees detection performs single tree crown delineation before health assessment (Yao et al., 2012;Shendryk et al., 2016b). Tree crown delineation is usually done by detecting local maxima from the canopy height model (CHM) and then segmenting trees using the watershed algorithm (Popescu et al., 2003). ...
In Australia, many birds and arboreal animals use hollows for shelters, but studies predict shortage of hollows in near future. Aged dead trees are more likely to contain hollows and therefore automated detection of them plays a substantial role in preserving biodiversity and consequently maintaining a resilient ecosystem. For this purpose full-waveform LiDAR data were acquired from a native Eucalypt forest in Southern Australia. The structure of the forest significantly varies in terms of tree density, age and height. Additionally, Eucalyptus camaldulensis have multiple trunk splits making tree delineation very challenging. For that reason, this paper investigates automated detection of dead standing Eucalyptus camaldulensis without tree delineation. It also presents the new feature of the open source software DASOS, which extracts features for 3D object detection in voxelised FW LiDAR. A random forest classifier, a weighted-distance KNN algorithm and a seed growth algorithm are used to create a 2D probabilistic field and to then predict potential positions of dead trees. It is shown that tree health assessment is possible without tree delineation but since it is a new research directions there are many improvements to be made.
... The condition of individual trees has been assessed using airborne laser scanning and imaging spectroscopy with promising results, albeit for relatively few (106) individual trees (Shendryk et al., 2016). The storage and processing demands for applications over very large areas (100,000s ha) probably are prohibitive at present, but remote-sensing technology and large data storage and processing capacity continue to evolve quickly (e.g., the Google Earth Engine), so very detailed assessments may be possible relatively soon. ...
Methods that provide rapid assessments of changing ecosystems at multiple scales are needed to inform management to address undesirable change. We developed a remote-sensing method in partnership with, and for use by, natural resource managers to predict annually stand condition of floodplain forests along Australia's longest river, the Murray River. A measure of stand condition, which was developed in collaboration with responsible natural resource managers, is a function of plant area index, crown extent, and the percentage live basal area. We surveyed a broad range of spatial and temporal variation in condition, built predictive stand-condition models using satellite-derived variables, and validated predictions with surveys of new sites. A multiyear model using data from 2 drought years and a year following extensive floods provided better predictions of stand condition than did models on the basis of the data for individual years. The model provided good predictions for data collected after the build for 50 sites and for resurveys of build sites in later years (R2 ≥ 0.86). There was limited, temporary improvement in stand condition after the extensive flooding (2010 to late 2010) that followed a 13-year (1997 to early 2010) drought. Forest condition can be mapped accurately and annually at medium resolution (25 × 25 m) for large areas (100,000s ha) if quantitative ground surveys, satellite imagery, machine learning, and future validation are combined. Regular assessments of forest condition can be related to likely causes of change by using regular, rapid assessments and hence can provide important management information.
With the capability of capturing high-resolution imagery data and the ease of accessing remote areas, aerial robots are becoming increasingly popular for forest health monitoring applications. For example, forestry tasks such as field surveys and foliar sampling which are generally manual and labour intensive can be automated with remotely controlled aerial robots. In this study, we propose two new online frameworks to quantify and rank the severity of individual tree crown loss. The real-time crown loss estimation (RTCLE) model localises and classifies individual trees into their respective crown loss percentage bins. Experiments are conducted to investigate if synthetically generated tree images can be used to train the RTCLE model as real images with diverse viewpoints are generally expensive to collect. Results have shown that synthetic data training helps to achieve a satisfactory baseline mean average precision (mAP) which can be further improved with just some additional real imagery data. We showed that the mAP can be increased approximately from 60% to 78% by mixing the real dataset with the generated synthetic data. For individual tree crown loss ranking, a two-step crown loss ranking (TSCLR) framework is developed to handle the inconsistently labelled crown loss data. The TSCLR framework detects individual trees before ranking them based on some relative crown loss severity measures. The tree detection model is trained with the combined dataset used in the RTCLE model training where we achieved an mAP of approximately 95% suggesting that the model generalises well to unseen datasets. The relative crown loss severity of each tree is estimated, with deep representation learning, by a probabilistic encoder from a fully trained variational autoencoder (VAE) model. The VAE is trained end-to-end to reconstruct tree images in a background agnostic way. Based on a conservative evaluation, the estimated crown loss severity from the probabilistic encoder generally showed moderate agreement with the expert’s estimation across all species of trees present in the dataset. All the software pipelines, the dataset, and the synthetic dataset generation can be found in the GitHub link.
Tree hollow is a semi-enclosed cavity in any kind of tree. The detection of a tree hollow is important not only for a tree but also for the species who use a tree hollow for their survival and settlement. A tree hollow plays a vital role for bird ecology for their survival, growth, and population; therefore, there is a need for detection of a tree hollow. This research paper worked on the same principle of detection of a tree hollow to make people aware of a tree hollow. Here, a feed forward neural network with back propagation of error neural network-based machine learning algorithm is used to automatically detect a tree hollow. The proposed algorithm is implemented using sklearn python-based packages. The implementation shows an accuracy of 82% for the detection of a tree hollow which is good results for detection of a tree hollow.
Forests are increasingly subject to a number of disturbances that can adversely influence their health. Remote sensing offers an efficient alternative for assessing and monitoring forest health. A myriad of methods based upon remotely sensed data have been developed, tailored to the different definitions of forest health considered, and covering a broad range of spatial and temporal scales. The purpose of this review paper is to identify and analyse studies that addressed forest health issues applying remote sensing techniques, in addition to studying the methodological wealth present in these papers. For this matter, we applied the PRISMA protocol to seek and select studies of our interest and subsequently analyse the information contained within them. A final set of 107 journal papers published between 2015 and 2020 was selected for evaluation according to our filter criteria and 20 selected variables. Subsequently, we pair-wise exhaustively read the journal articles and extracted and analysed the information on the variables. We found that (1) the number of papers addressing this issue have consistently increased, (2) that most of the studies placed their study area in North America and Europe and (3) that satellite-borne multispectral sensors are the most commonly used technology, especially from Landsat mission. Finally, most of the studies focused on evaluating the impact of a specific stress or disturbance factor, whereas only a small number of studies approached forest health from an early warning perspective.
Pine wilt disease (PWD) is a global destructive threat to forests, having caused extreme damage in China. Therefore, the establishment of an effective method to accurately monitor and map the infection stage by PWD is imperative. Unmanned aerial vehicle (UAV)-based hyperspectral imaging (HI) and light detection and ranging (LiDAR) technique is an effective approach for forest health monitoring. However, few previous studies have used airborne HI and LiDAR to detect PWD and compared the capability for predicting PWD infection stage at the tree level. In this paper, PWD infection was divided into five stages (green, early, middle, heavy, and grey), and HI and LiDAR data were integrated to detect PWD. We estimated the power of the hyperspectral method (HI data only), LiDAR (LiDAR data only), and their combination (HI plus LiDAR data) to predict the infection stages of PWD using the random forest (RF) algorithm. We obtained the following results: (1) The classification accuracies of HI (OA: 66.86%, Kappa: 0.57) were higher than those of LiDAR (OA: 45.56%, Kappa: 0.27) for predicting PWD infection stages, and their combination had the best accuracies (OA: 73.96%, Kappa: 0.66); (2) LiDAR data had higher ability for dead tree identification than HI data; and (3) The combined use of HI and LiDAR data for estimation of PWD infection stages showed that LiDAR metrics (e.g., crown volume) were essential in the classification model, although the variables derived from HI data contributed more than those extracted from LiDAR. Therefore, we proposed a new approach combining the merits of HI and LiDAR data to precisely predict PWD infection stages at the tree level, allowing better PWD monitoring and control. The approach could also be employed for mapping and monitoring other forest disturbance issues.
This study analyzed highly-correlated, feature-rich datasets from hyperspectral remote sensing data using multiple machine and statistical-learning methods.
The effect of filter-based feature-selection methods on predictive performance was compared.
Also, the effect of multiple expert-based and data-driven feature sets, derived from the reflectance data, was investigated.
Defoliation of trees (%) was modeled as a function of reflectance, and variable importance was assessed using permutation-based feature importance.
Overall support vector machine (SVM) outperformed others such as random forest (RF), extreme gradient boosting (XGBoost), lasso (L1) and ridge (L2) regression by at least three percentage points.
The combination of certain feature sets showed small increases in predictive performance while no substantial differences between individual feature sets were observed.
For some combinations of learners and feature sets, filter methods achieved better predictive performances than the unfiltered feature sets, while ensemble filters did not have a substantial impact on performance.
Permutation-based feature importance estimated features around the red edge to be most important for the models.
However, the presence of features in the near-infrared region (800 nm - 1000 nm) was essential to achieve the best performances.
More training data and replication in similar benchmarking studies is needed for more generalizable conclusions.
Filter methods have the potential to be helpful in high-dimensional situations and are able to improve the interpretation of feature effects in fitted models, which is an essential constraint in environmental modeling studies.
Deep learning provides excellent potentials for hyperspectral images (HSIs) classification, but it is infamous for requiring large amount of labeled samples while the collection of high-quality labels for HSIs is extremely expensive and time-consuming. Furthermore, when the limited training samples are available, deep learning methods may suffer from over-fitting. In this work, we propose a novel collaborative learning framework for semi-supervised HSI classification with joint deep convolutional neural networks and deep clustering. Specifically, a lightweight 3D convolutional neural network (CNN) with much less parameters compared with classical 3D CNNs is designed for deep discriminative feature learning and classification. Then a deep clustering method, that is approximate rank-order clustering (AROC) algorithm, is applied to cluster deep features to generate pseudo labels for abundant unlabeled samples. Finally, we fine-tune the lightweight 3D CNN by minimizing a dual-loss (softmax loss and center loss) using both true and pseudo labels. Experimental results on three challenging HSI datasets demonstrate that the proposed method can achieve better performance than other state-of-the-art deep learning based methods and traditional HSI classification methods methods.
Compared to the Northern Hemisphere, literature concerning remote sensing applications for insect outbreak detection and assessment is scarce in the Southern Hemisphere in general and in Latin America in particular. After a thorough literature review, we found few studies describing insect outbreaks in this part of the world, from which the case of the native moth Ormiscodes amphimone outbreaks in the Argentinian and Chilean Patagonia seems to be most relevant in Latin America. Only in Chile Ormiscodes amphimone disruptions have caused complete defoliation over 164,000 ha between 2000 and 2015 with the largest single continuous event (one growing season) accurately measured with remote sensing of about 25,000 ha. There are indications of other relevant outbreaks in Latin American countries, like the case of Thaumastocoris peregrinus attacks in Eucalyptus plantations in Brazil, but remote sensing assessments still need to be done. Potential causes of this scientific literature shortage could be that (1) there would be ongoing remote sensing applications for detecting and mapping forest pests in commercial plantations, but they would not be publicly available due to restrictions from timber companies; (2) main national and international remote sensing efforts are focused on assessing deforestation and degradation of Latin American forests (a threat especially relevant for tropical forest in the Amazon), while insect outbreaks may not be a main threat; and (3) there may be a lack of remote sensing specialists or existing specialists are not interested in insect outbreaks. We believe there is a research gap on insect outbreak detection and mapping using remote sensing in Latin America and that we have a great opportunity to fill this gap considering the large amount of open access satellite data and software.
In this paper, we propose to estimate tree defoliation from ground-level RGB photos with convolutional neural networks (CNN). Tree defoliation is usually assessed with field campaigns, where experts estimate multiple tree health indicators per sample site. Campaigns span entire countries to come up with a holistic, nationwide picture of forest health. Surveys are very laborious, expensive, time-consuming and need a large number of experts. We aim at making the monitoring process more efficient by casting tree defoliation estimation as an image interpretation problem. What makes this task challenging is the strong variance in lighting, viewpoint, scale, tree species, and defoliation types. Instead of accounting for each factor separately through explicit modeling, we learn a joint distribution directly from a large set of annotated training images following the end-to-end learning paradigm of deep learning.
The proposed workflow works as follows: (i) Human experts visit individual trees in forests distributed all over Switzerland, (ii) acquire one photo per tree with an off-the-shelf, hand-held RGB camera, and (iii) assign a defoliation value. The CNN approach is (iv) trained on a subset of the images with expert defoliation assessments and (v) tested on a hold-out part to check predicted values against ground truth.
We evaluate our supervised method on three data sets with different level of difficulty acquired in Swiss forests and achieve an average mean absolute error (avgMAE) of 7.6% for the joint data set after cross-validation. Comparison to a group of human experts on one of the data sets shows that our CNN approach performs only 0.9% points worse. We show that tree defoliation estimation from ground-level RGB images with a CNN works well and achieves performance close to human experts.
Monoculture plantation woodlands are particularly vulnerable to disturbance events as species uniformity makes such stands highly susceptible to pests and diseases. Red band needle blight (caused by the fungus Dothistroma septosporum) is a disease which has a particularly significant economic impact on pine plantation forests worldwide, affecting diameter and height growth. However, monitoring its spread and intensity is complicated by the fact that the diseased trees are often only visible from aircraft in the advanced stages of the epidemic. Remote sensing could potentially aid in the detection of infected stands and in monitoring disease development and spread. Thermography is one of the techniques that can be used for monitoring changes in the physiological state of plants following infection. However, the use of thermography in forestry has so far been restricted by poor spatial resolution (satellite-based sensors) or high data acquirement costs (airborne sensors). This paper investigates the use of Unmanned Aerial Vehicle (UAV)-borne thermal systems for detecting disease-induced canopy temperature increase and explores the influence of the imaging time and weather conditions on the detected relationship. Furthermore, the potential of a number of airborne LiDAR-derived structural metrics for detection of changes in the canopy structure following the infection are investigated. The study was located in a diseased Scots pine (Pinus sylvestris) stand in Queen Elizabeth II Forest Park (central Scotland, UK), where 60 sample trees were surveyed. The thermal imagery was acquired at six different times of a day from an altitude of 60 m. Statistically significant correlation between canopy temperature depression (CTD) and disease levels was found for most of the flights (R² between 0.27 and 0.41), which may be related to the needle damage symptoms caused by the disease, i.e. loss of cellular integrity, necrosis and eventual desiccation. Furthermore, the standard deviation of the crown temperature exhibited weak but statistically significant correlation (R² between 0.11 and 0.13). The combination of CTD and standard deviation of crown temperature in a partial least squares regression (PLSR) further improved the observed relationship with the estimated disease level. Inclusion of LiDAR structural metrics was also investigated but only provided a slight improvement. A change in environmental conditions altered the magnitude of differences between canopy temperatures; no significant correlation with disease level was found in the morning flight, whilst the strongest relationship was obtained at the time of highest solar radiation, which coincides with the time of maximum photosynthetic activity.
Detailed information on the number and density of trees is important for conservation and sustainable use of forest resources. In this respect, remote sensing technology is a reliable tool for deriving timely and fine-scale information on forest inventory attributes. However, to better predict and understand the functioning of the forest, fine-scale measures of tree number and density must be extrapolated to the forest plot or stand levels through upscaling. In this study, we compared and combined three sources of remotely sensed data, including low point density airborne laser scans (ALS), synthetic aperture radar (SAR) and very-high resolution WorldView-2 imagery to upscale the total number of trees to the plot level in a structurally complex eucalypt forest using random forest regression. We used information on number of trees previously derived from high point density ALS as training data for a random forest regressor and field inventory data for validation. Overall, our modelled estimates resulted in significant fits (p < 0.05) with goodness-of-fit (R 2) of 0.61, but systematically underestimated tree numbers. The ALS predictor variables (e.g. canopy cover and height) were the best for estimating tree numbers (R 2 = 0.48, nRMSE = 61%), as compared to WorldView-2 and SAR predictor variables (R 2 < 0.35). Overall, the combined use of WorldView-2, ALS and SAR predictors for estimating tree numbers showed substantial improvement in R 2 of up to 0.13 as compared to their individual use. Our findings demonstrate the potential of using low point density ALS, SAR and WorldView-2 imagery to upscale high point density ALS derived tree numbers at the plot level in a structurally complex eucalypt forest.
Defoliation by herbivorous insects is a widespread forest disturbance driver, affecting global forest health and ecosystem dynamics. Compared with time- and labor-intensive field surveys, remote sensing provides the only realistic approach to mapping canopy defoliation by herbivorous insects over large spatial and temporal scales. However, the spectral and structural signatures of defoliation by insects at the individual tree level have not been well studied. Additionally, the predictive power of spectral and structural metrics for mapping canopy defoliation has seldom been compared. These critical knowledge gaps prevent us from consistently detecting and mapping canopy defoliation by herbivorous insects across multiple scales. During the peak of a gypsy moth outbreak in Long Island, New York in summer 2016, we leveraged bi-temporal airborne imaging spectroscopy (IS, i.e., hyperspectral imaging) and LiDAR measurements at 1 m spatial resolution to explore the spectral and structural signatures of canopy defoliation in a mixed oak-pine forest. We determined that red edge and near-infrared spectral regions within the IS data were most sensitive to crown-scale defoliation severity. LiDAR measurements including B70 (i.e., 70th bincentile height), intensity skewness, and kurtosis were effectively able to detect structural changes caused by herbivorous insects. In addition to canopy leaf loss, increased exposure of understory and non-photosynthetic materials contributed to the detected spectral and structural signatures. Comparing the ability of individual sensors to map canopy defoliation, the LiDAR-only Ordinary Least-Square (OLS) model performed better than the IS-only model (Adj. R-squared = 0.77, RMSE = 15.37% vs. Adj. R-squared = 0.63, RMSE = 19.11%). The IS + LiDAR model improved on performance of the individual sensors (Adj. R-squared = 0.81, RMSE = 14.46%). Our study improves our understanding of spectral and structural signatures of defoliation by herbivorous insects and presents a novel approach for mapping insect defoliation at the individual tree level. Additionally, with the current and next generation of spaceborne sensors (e.g., WorldView-3, Landsat, Sentinel-2, HyspIRI, and GEDI), higher accuracy and frequent monitoring of insect defoliation may become more feasible across a range of spatial scales, which are critical for ecological research and management of forest resources including the economic consequences of forest insect infestations (e.g., reduced growth and increased mortality), as well as for informing and testing of carbon cycle models.
Understanding post-fire forest recovery is pivotal to the study of forest dynamics and global carbon cycle. Field-based studies indicated a convex response of forest recovery rate to burn severity at the individual tree level, related with fire-induced tree mortality; however, these findings were constrained in spatial/temporal extents, while not detectable by traditional optical remote sensing studies, largely attributing to the contaminated effect from understory recovery. Here, we examined whether the combined use of multi-sensor remote sensing techniques (i.e., 1 m simultaneous airborne imaging spectroscopy and LiDAR and 2 m satellite multi-spectral imagery) to separate canopy recovery from understory recovery would enable to quantify post-fire forest recovery rate spanning a large gradient in burn severity over large-scales. Our study was conducted in a mixed pine-oak forest in Long Island, NY, three years after a top-killing fire. Our studies remotely detected an initial increase and then decline of forest recovery rate to burn severity across the burned area, with a maximum canopy area-based recovery rate of 10% per year at moderate forest burn severity class. More intriguingly, such remotely detected convex relationships also held at species level, with pine trees being more resilient to high burn severity and having a higher maximum recovery rate (12% per year) than oak trees (4% per year). These results are one of the first quantitative evidences showing the effects of fire adaptive strategies on post-fire forest recovery, derived from relatively large spatial-temporal scales. Our study thus provides the methodological advance to link multi-sensor remote sensing techniques to monitor forest dynamics in a spatially explicit manner over large-scales, with important implications for fire-related forest management and constraining/benchmarking fire effect schemes in ecological process models.
The health of the forest vegetation is one of the driving factors and indicator of climate change impacts. Fragmentation of the forests on the other hand brings out the implications of the various stressing factors on the spatial extent of the forests especially the increasing population and industrialization, which has always impacted the forested regions in the form of deforestation for commercial purposes and conversion of forest land for cultivation. Hence there is a demand for spatial assessment and continuous monitoring of the forested regions. With a wide range of advancing technologies, remote sensing methods are increasingly being employed for monitoring a number of remotely measurable properties of different types of vegetation.This paper reviews the utilitarian application of remote sensing as a tool to assess the health of the forest regions. The various multispectral satellite image derived vegetation indices like Broadband and narrowband Greenness, leaf pigments, canopy water content and Light use efficiency can be used as a combined tool to generate a comprehensive health metrics for the forest canopy. These new indices may be useful in remote sensing plant physiology status as well. Various band ratio exist for above factors depending on the bands available in the selected dataset. A detailed discussion of the existing multispectral satellite sensors capable of studying the forest canopy is presented herewith, in addition to briefing on the potential of the latest microwave remote sensing technique for vegetation studies. The estimated vegetation indices can be used to generate the final health map of forest regions. Any change in the spatial extent of the forested areas and their pathway for future change can be assessed by the fragmentation analysis.The application of the discussed techniques in various environmental conditions are elaborated and the constraints in addition to the requirements for effective forest health monitoring is provided in this review.
Purpose of Review
In this review, we highlight recent developments and applications in remote sensing that can improve the accuracy and timeliness of health assessments in plantations managed for timber and pulp production. The detection and mapping of damage extent and severity caused by insect pests and fungal pathogens is a common requirement of foresters managing plantations. The objectives of these surveys can range from early detection for targeted intervention to more strategic aims of predicting stand susceptibility or evaluating the performance of management strategies.
Recent Findings
Recent developments in remote sensing technologies and big data modelling techniques can now provide spatially explicit, quantitative solutions for these management objectives that are more accurate than manual field-based assessments of tree damage or airborne visual mapping. Past studies have identified a large number of spectral, textural and structural metrics that have been used in models to classify specific tree crown damage symptoms. This process requires a detailed understanding of the chronology of crown symptoms for specific damaging agents and the spectral responses to these symptoms. Continuing increases in the spatial and spectral resolution of remote sensors enables crown-level damage classification.
Summary
The development of data processing workflows that fuse spectral information with three-dimensional (3D) data acquired simultaneously from single or different remote platforms promote the opportunities to derive both structural and physiological crown-level attributes that relate to crown damage. The simultaneous acquisition of spectral and 3D point data will enable plantation foresters to derive several spatial products, including the assessment of tree health in a cost-effective manner.
Stage 1 of the Riverina river red gum reserves mapping program,
outlines the methods used and results of the stem density and canopy condition
mapping of two of the largest of the NSW Riverina reserves; the Yanga portion of
the Murrumbidgee Valley National Park, and the Millewa group of the Murray Valley
Regional and National Parks. It also includes the results from mapping of the Barmah
National Park and Murray River Park in Victoria.
This paper discusses a best-practice representation of uncertainty in satellite remote sensing data. An estimate of uncertainty is necessary to make appropriate use of the information conveyed by a measurement. Traditional error propagation quantifies the uncertainty in a measurement due to well-understood perturbations in a measurement and in auxiliary data – known, quantified "unknowns". The under-constrained nature of most satellite remote sensing observations requires the use of various approximations and assumptions that produce non-linear systematic errors that are not readily assessed – known, unquantifiable "unknowns". Additional errors result from the inability to resolve all scales of variation in the measured quantity – unknown "unknowns". The latter two categories of error are dominant in under-constrained remote sensing retrievals, and the difficulty of their quantification limits the utility of existing uncertainty estimates, degrading confidence in such data. This paper proposes the use of ensemble techniques to present multiple self-consistent realisations of a data set as a means of depicting unquantified uncertainties. These are generated using various systems (different algorithms or forward models) believed to be appropriate to the conditions observed. Benefiting from the experience of the climate modelling community, an ensemble provides a user with a more complete representation of the uncertainty as understood by the data producer and greater freedom to consider different realisations of the data.
The work described in this paper is aimed at validating hyperspectral airborne reflectance data collected during the Regional Experiments For Land-atmosphere EXchanges (REFLEX) campaign. Ground reflectance data measured in a vineyard were compared with airborne reflectance data. A sampling strategy and subsequent ground data processing had to be devised so as to capture a representative spectral sample of this complex crop. A linear model between airborne and ground data was tried and statistically tested. Results reveal a sound correspondence between ground and airborne reflectance data (R2 > 0.97), validating the atmospheric correction of the latter.
DOI: 10.1515/acgeo-2015-0036
Humans rely on healthy forests to supply energy, building materials, and food and to provide services such as storing carbon, hosting biodiversity, and regulating climate. Defining forest health integrates utilitarian and ecosystem measures of forest condition and function, implemented across a range of spatial scales. Although native forests are adapted to some level of disturbance, all forests now face novel stresses in the form of climate change, air pollution, and invasive pests. Detecting how intensification of these stresses will affect the trajectory of forests is a major scientific challenge that requires developing systems to assess the health of global forests. It is particularly critical to identify thresholds for rapid forest decline, because it can take many decades for forests to restore the services that they provide.
Copyright © 2015, American Association for the Advancement of Science.
Maintaining the health and condition of the forest plantation estate is critical to ensuring there are no adverse losses in productivity. Within Australian Pinus radiata plantations a diverse range of damaging agents are present. One significant agent is a fungal pathogen Sphaeropsis sapinea. In this research, we detail the development of relationships between a range of individual crown health attributes representing symptoms of Sphaeropsis sapinea infection and high spatial and spectral resolution remotely sensed imagery characteristics. To do this, two methods were used; the first utilized vegetation spectral indices including simple and normalized difference ratios, and the second, linear spectral mixture analysis. Results indicate that spectral indices that utilize either chlorophyll absorption wavelengths at 680 nm with a non-chlorophyll region of the spectrum (such as 710 or 750 nm) or the slope of the upper red-edge between 710 and 740 nm were most significantly related to individual crown damage attributes. Linear unmixing analysis consistently extracted four fraction endmember images (sunlit canopy, soil, shadow, and non-photosynthetic vegetation (NPV)) from the 12 channel imagery. Multiple linear stepwise regression models developed using mixed fractional abundances provided similar results to those derived using spectral indices. The NPV and shadow endmembers, in order, were consistently identified as the most significant in these developed models.
As airborne laser scanning (ALS) gains wider adoption to support forest operations in Canada, the consistency and quality of derivative products that support long-term monitoring and planning are becoming a key issues for managers. The Canadian Consortium for Lidar Environmental
Applications Research (C-CLEAR) has supported almost 200 projects across Canada since 2000, with forest-related studies being a dominant theme. In 2010 and 2011, field operations were mobilized to support 13 ALS projects spanning almost the full longitudinal gradient of Canada’s forests.
This paper presents case studies for seven plus an overview of some best practices and data processing workflow tools that have resulted from these consortium activities. Although the projects and research teams are spread across Canada, the coordination and decade of experience provided through
C-CLEAR have brought common methodological elements to all. It is clear that operational, analytical and reporting guidelines that adhere to community accepted standards are required if the benefits promised by ALS forestry are to be realized. A national Lidar Institute that builds upon the
C-CLEAR model and focuses on developing standards, guidelines, and certified training would address this need.
Abundance and spatial distribution of standing dead trees (snags) are key indicators of forest biodiversity and ecosystem health and represent a critical component of habitat for various wildlife species, including the great grey owl and the black-backed woodpecker. In this work we assess the potential of light detection and ranging (LiDAR) to discriminate snags from the live trees and map their distribution. The study area encompasses the burn perimeter of the Rim Fire, the third largest wildfire in California’s recorded history (~104.000 ha) and represents a heterogeneous mosaic of mixed conifer forests, hardwood, and meadows. The snags mapping procedure is based on a 3D single tree detection using a Watershed algorithm and the extraction of height and intensity metrics within each segment. Variables selected using Gaussian processes form a feature space for a classifier to distinguish between dead trees and live trees. Finally, snag density and snag diameter classes that are relevant for avian species are mapped. This work shows the use of LiDAR metrics to quantify ecological variables related to the vertical heterogeneity of the forest canopy that are important in the identification of snags, for example, fractional cover. We observed that intensity-related variables are critical to the successful identification of snags and their distribution. Our study highlights the importance of high-density LiDAR for characterizing the forest structural variables that contribute for wildlife habitat suitability assessment.
Standing dead trees, known as snags, are an essential factor in maintaining biodiversity in forest ecosystems. Combined with their role as carbon sinks, this makes for a compelling reason to study their spatial distribution. This paper presents an integrated method to detect and delineate individual dead tree crowns from color infrared aerial imagery. Our approach consists of two steps which incorporate statistical information about prior distributions of both the image intensities and the shapes of the target objects. In the first step, we perform a Gaussian Mixture Model clustering in the pixel color space with priors on the cluster means, obtaining up to 3 components corresponding to dead trees, living trees, and shadows. We then refine the dead tree regions using a level set segmentation method enriched with a generative model of the dead trees’ shape distribution as well as a discriminative model of their pixel intensity distribution. The iterative application of the statistical shape template yields the set of delineated dead crowns. The prior information enforces the consistency of the template’s shape variation with the shape manifold defined by manually labeled training examples, which makes it possible to separate crowns located in close proximity and prevents the formation of large crown clusters. Also, the statistical information built into the segmentation gives rise to an implicit detection scheme, because the shape template evolves towards an empty contour if not enough evidence for the object is present in the image. We test our method on 3 sample plots from the Bavarian Forest National Park with reference data obtained by manually marking individual dead tree polygons in the images. Our results are scenario-dependent and range from a correctness/completeness of 0.71/0.81 up to 0.77/1, with an average center-of-gravity displacement of 3-5 pixels between the detected and reference polygons.
Developments in hyperspectral remote sensing have provided new indices or indicators of biochemical and biophysical properties. Most of the studies involving the novel spectral indices have been conducted at the leaf scale and have been rarely investigated for species discrimination. The objectives of the study were to determine hyperspectral indices that (i) are likely to be influenced by change in spectral measurement from the leaf to the canopy scale and (ii) can discriminate species at both scales. Leaf and canopy reflectance measurements were made from six species (3 shrubs, 3 trees) using an ASD spectroradiometer. The two-sample t test was used to evaluate whether significant differences exist between leaf and canopy indices, while differences between species pairs (15 pairs) were evaluated with ANOVA and pair-wise Bonferroni adjusted t tests. The hyperspectral indices evaluated in this study were, in general, sensitive to the change in spectral measurement scale from the leaf to the canopy. However, among the indices studied, red-edge positions (REP) extracted by the linear extrapolation I method were least sensitive to the change in measurement scale as three out of the six species showed no significant differences between the leaf and canopy indices. With respect to species discrimination, the canopy indices were better discriminators than the leaf indices. This is essential for air-or spaceborne remote sensing of species assemblages. The photochemical reflectance index (PRI) showed the highest potential to discriminate species at the canopy scale (all 15 pairs), while the linear extrapolation REPs showed the highest potential to discriminate the same species pairs (10 pairs) at both scales. Hyperspectral indices might provide new possibilities of differentiating plant species.
The amount of scientific literature on (Geographic) Object-based Image Analysis - GEOBIA or OBIA in short - has been and still is sharply increasing. The term OBIA is herein used for generic image processing tasks. OBIA researchers with backgrounds in geology, geography or mineralogy are joining forces with colleagues from cell biology, molecular biology, medicine, pharmacy or nanotechnology. More narrowly, GEOBIA's objective is the generation of geographic information (in GIS-ready format), from which new spatial knowledge can be obtained. GEOBIA is a relatively new field of research based upon two major concepts: A) dissecting images or any (pseudo- )continuous multidimensional fields of data, and b) allowing for multiple scales when organizing and utilizing the resulting objects. Analyses of the amount of scientific literature reveal that not only the number of articles is increasing, but that the rate of growth is also noticeably accelerating. Finally, it is discussed whether or not GEOBIA can be considered as an important paradigm in remote sensing. Copyright © (2013) by the American Society for Photogrammetry & Remote Sensing.
Digital remote sensing is rapidly developing into an operational tool for forest health assessment at a range of scales. A key aspect of this development is the derivation of models relating spectral data contained in the images to the extent and severity of plant stress symptoms. This study assesses the utility of high spatial resolution (0.5m × 0.5m pixel size) airborne digital imagery to assess the presence and severity of defoliation and necrosis from Mycosphaerella leaf disease at the crown scale across a small plantation in north-western Tasmania, Australia. The best model of defoliation included the variance of reflectance at 780nm within the crown (r2 = 0.5; p < 0.0001) and the best model of necrosis included the ratio of reflectance at 680nm/550nm (r2 = 0.3; p < 0.0001). Maps derived from these linear models clearly illustrate the distribution and gradient of disease severity observed in the field. Error matrix analysis indicates moderate map accuracy for classified versions of both defoliation (OA = 71%; Kappa = 0.63) and necrosis (OA = 67%; Kappa = 0.54) but the maps reveal additional information about the distribution and severity of symptoms throughout the plantation. Methods such as those described in this paper may enable managers to more accurately target remedial actions. The transfer of this technology to space-borne platforms will potentially improve map accuracy and image availability, and reduce the cost of image collection. Together, these improvements increase the likelihood of remote sensing being incorporated into forest management strategies in future.
Numerous studies showed the impact of droughts events on the seasonal dynamic of satellite-based vegetation index but validation and relation to ground observations is still lacking. Focusing on the impacts of 2003 and 2011 droughts events, we defined and validated forest health indicators using extensive in-situ damage inventories carried out on all French forests. Analysis of vegetation indices dynamics (EVI, LAI) at the national scale was performed using MODIS and SPOT VEGETATION products (MOD13A2 V5 and CYCLOPES V3.1). Satellite-based indicators of (i) crown condition - which is itself a commonly used forest health indicator - and (ii) leaf phenology anomalies were calibrated and validated. In our study, average of CYCLOPES LAI during the growing season is a good proxi of crown condition (r2=0.79, p
The visual progression of sirex (Sirex noctilio) infestation symptoms has been categorized into three distinct infestation phases, namely the green, red and grey stages. The grey stage is the final stage which leads to almost complete defoliation resulting in dead standing trees or snags. Dead standing pine trees however, could also be due to the lightning damage. Hence, the objective of the present study was to distinguish amongst healthy, sirex grey-attacked and lightning-damaged pine trees using AISA Eagle hyperspectral data, random forest (RF) and support vector machines (SVM) classifiers. Our study also presents an opportunity to look at the possibility of separating amongst the previously mentioned pine trees damage classes and other landscape classes on the study area. The results of the present study revealed the robustness of the two machine learning classifiers with an overall accuracy of 74.50% (total disagreement = 26%) for RF and 73.50% (total disagreement = 27%) for SVM using all the remaining AISA Eagle spectral bands after removing the noisy ones. When the most useful spectral bands as measured by RF were exploited, the overall accuracy was considerably improved; 78% (total disagreement = 22%) for RF and 76.50% (total disagreement = 24%) for SVM. There was no significant difference between the performances of the two classifiers as demonstrated by the results of McNemar’s test (chi-squared; χ2 = 0.14, and 0.03 when all the remaining ASIA Eagle wavebands, after removing the noisy ones and the most important wavebands were used, respectively). This study concludes that AISA Eagle data classified using RF and SVM algorithms provide relatively accurate information that is important to the forest industry for making informed decision regarding pine plantations health protocols.
Accurate mapping of benthic habitats in the Florida Keys is essential in developing effective management strategies for this unique coastal ecosystem. In this study, we evaluated the applicability of hyperspectral imagery collected from Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) for benthic habitat mapping in the Florida Keys. An overall accuracy of 84.3% and 86.7% was achieved respectively for a group-level (3-class) and code-level (12-class) classification by integrating object-based image analysis (OBIA), hyperspectral image processing methods, and machine learning algorithms. Accurate and informative object-based benthic habitat maps were produced. Three commonly used image correction procedures (atmospheric, sun-glint, and water-column corrections) were proved unnecessary for small area mapping in the Florida Keys. Inclusion of bathymetry data in the mapping procedure did not increase the classification accuracy. This study indicates that hyperspectral systems are promising in accurate benthic habitat mapping at a fine detail level.
Bark beetle outbreaks have killed large numbers of trees across North America in recent years. Lidar remote sensing can be used to effectively estimate forest biomass, but prediction of both live and dead standing biomass in beetle-affected forests using lidar alone has not been demonstrated. We developed Random Forest (RF) models predicting total, live, dead, and percent dead basal area (BA) from lidar metrics in five different beetle-affected coniferous forests across western North America. Study areas included the Kenai Peninsula of Alaska, southeastern Arizona, north-central Colorado, central Idaho, and central Oregon, U.S.A. We created RF models with and without intensity metrics as predictor variables and investigated how intensity normalization affected RF models in Idaho. RF models predicting total BA explained the most variation, whereas RF models predicting dead BA explained the least variation, with live and percent dead BA models explaining intermediate levels of variation. Important metrics varied between models depending on the type of BA being predicted. Generally, height and density metrics were important in predicting total BA, intensity and density metrics were important in predicting live BA, and intensity metrics were important in predicting dead and percent dead BA. Several lidar metrics were important across all study areas. Whether needles were on or off beetle-killed trees at the time of lidar acquisition could not be ascertained. Future work, where needle conditions at the time of lidar acquisition are known, could improve upon our analysis and results. Although RF models predicting live, dead, and percent dead BA did not perform as well as models predicting total BA, we concluded that discrete-return lidar can be used to provide reasonable estimations of live and dead BA. Our results also showed which lidar metrics have general utility across different coniferous forest types.
Forests in Flux
Forests worldwide are in a state of flux, with accelerating losses in some regions and gains in others. Hansen et al. (p. 850 ) examined global Landsat data at a 30-meter spatial resolution to characterize forest extent, loss, and gain from 2000 to 2012. Globally, 2.3 million square kilometers of forest were lost during the 12-year study period and 0.8 million square kilometers of new forest were gained. The tropics exhibited both the greatest losses and the greatest gains (through regrowth and plantation), with losses outstripping gains.
Full-waveform laser scanning data acquired with a Riegl LMS-Q560 instrument were used to classify an orange orchard into orange trees, grass and ground using waveform parameters alone. Gaussian decomposition was performed on this data capture from the National Airborne Field Experiment in November 2006 using a custom peak-detection procedure and a trust-region-reflective algorithm for fitting Gauss functions. Calibration was carried out using waveforms returned from a road surface, and the backscattering coefficient γ was derived for every waveform peak. The processed data were then analysed according to the number of returns detected within each waveform and classified into three classes based on pulse width and γ. For single-peak waveforms the scatterplot of γ versus pulse width was used to distinguish between ground, grass and orange trees. In the case of multiple returns, the relationship between first (or first plus middle) and last return γ values was used to separate ground from other targets. Refinement of this classification, and further sub-classification into grass and orange trees was performed using the γ versus pulse width scatterplots of last returns. In all cases the separation was carried out using a decision tree with empirical relationships between the waveform parameters. Ground points were successfully separated from orange tree points. The most difficult class to separate and verify was grass, but those points in general corresponded well with the grass areas identified in the aerial photography. The overall accuracy reached 91%, using photography and relative elevation as ground truth. The overall accuracy for two classes, orange tree and combined class of grass and ground, yielded 95%. Finally, the backscattering coefficient γ of single-peak waveforms was also used to derive reflectance values of the three classes. The reflectance of the orange tree class (0.31) and ground class (0.60) are consistent with published values at the wavelength of the Riegl scanner (1550 nm). The grass class reflectance (0.46) falls in between the other two classes as might be expected, as this class has a mixture of the contributions of both vegetation and ground reflectance properties.