Article

Effect of slope on treetop detection using a LiDAR Canopy Height Model

Authors:
  • rapidlasso GmbH
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Canopy Height Models (CHMs) or normalized Digital Surface Models (nDSM) derived from LiDAR data have been applied to extract relevant forest inventory information. However, generating a CHM by height normalizing the raw LiDAR points is challenging if trees are located on complex terrain. On steep slopes, the raw elevation values located on either the downhill or the uphill part of a tree crown are height-normalized with parts of the digital terrain model that may be much lower or higher than the tree stem base, respectively. In treetop detection, a highest crown return located in the downhill part may prove to be a ''false'' local maximum that is distant from the true treetop. Based on this observation, we theoretically and experimentally quantify the effect of slope on the accuracy of treetop detection. The theoretical model presented a systematic horizontal displacement of treetops that causes tree height to be systematically displaced as a function of terrain slope and tree crown radius. Interestingly, our experimental results showed that the effect of CHM distortion on treetop displacement depends not only on the steep-ness of the slope but more importantly on the crown shape, which is species-dependent. The influence of the systematic error was significant for Scots pine, which has an irregular crown pattern and weak apical dominance, but not for mountain pine, which has a narrow conical crown with a distinct apex. Based on our findings, we suggest that in order to minimize the negative effect of steep slopes on the CHM, especially in heterogeneous forest with multiple species or species which change their morphological characteristics as they mature, it is best to use raw elevation values (i.e., use the un-normalized DSM) and compute the height after treetop detection.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... In treetop detection results obtained through these methods, "false" local maxima may be detected in areas with steeply sloping canopies, and displacement may occur between these maxima and the true treetops. To date, only three previous studies have reported the effects of the above factors on treetop detection considering displacement [31][32][33]. Khosravipour et al. [31] proposed a theoretical model to quantify the effect of the slope gradient on treetop detection accuracy. In their theoretical model, the horizontal and vertical displacements of the treetops were treated as functions of the slope gradient and crown radius. ...
... To date, only three previous studies have reported the effects of the above factors on treetop detection considering displacement [31][32][33]. Khosravipour et al. [31] proposed a theoretical model to quantify the effect of the slope gradient on treetop detection accuracy. In their theoretical model, the horizontal and vertical displacements of the treetops were treated as functions of the slope gradient and crown radius. ...
... Kosravipur et al. [31] and Alexander et al. [32] proposed theoretical treetop displacement models that are applicable to conical and spherical tree crown structures, respectively, and can be used to quantify treetop detection accuracy. Nie et al. [33] extended this approach to scenes with different terrain changes. ...
Article
Full-text available
Normalized point clouds (NPCs) derived from unmanned aerial vehicle-light detection and ranging (UAV-LiDAR) data have been applied to extract relevant forest inventory information. However, detecting treetops from topographically normalized LiDAR points is challenging if the trees are located in steep terrain areas. In this study, a novel point cloud normalization method based on the imitated terrain (NPCIT) method was proposed to reduce the effect of vegetation point cloud normalization on crown deformation in regions with high slope gradients, and the ability of the treetop detection displacement model to quantify treetop displacements and tree height changes was improved, although the model did not consider the crown shape or angle. A forest farm in the mountainous region of south-central China was used as the study area, and the sample data showed that the detected treetop displacement increased rapidly in steep areas. With this work, we made an important contribution to theoretical analyses using the treetop detection displacement model with UAV-LiDAR NPCs at three levels: the method, model, and example levels. Our findings contribute to the development of more accurate treetop position identification and tree height parameter extraction methods involving LiDAR data.
... Nonetheless, this process has limitation. As discussed by Khosravipour et al. (2015), distortions can occur in the point cloud in sites with high slopes. Such distortions are more prone to happen depending on the crown shape of tree species. ...
... Such behaviour could be the result of a considerable difference of terrain complexity. As mentioned in chapter 2 section 11 and further confirmed by Khosravipour et al. (2015), trees tend to tilt proportionally to the terrain slope. Site A slope condition is more complex compared to site B, resulting in more severe tilted trees, which could have affected the metrics calculated. ...
... However, araucaria trees are considerably distorted when normalizing the point cloud. This effect is a combination of the irregular terrain and araucaria's crown unique shape (Khosravipour et al., 2015). Figures 24.A and 24.B demonstrate that effect in more detail. ...
Thesis
Background: The Atlantic Forest is one, if not the most, diverse ecosystem in the planet. The Atlantic Forest contains an estimated 250 species of mammals (55 endemic), 340 amphibians (90 endemic), 1,023 birds (188 endemic), and approximately 20,000 trees, half of them endemic. Unfortunately, several of these species are currently threatened to become extinct. Amongst those, a tree species from the Araucariaceae family called Araucaria angustifolia can be found. Known as Brazilian Pine or just araucaria, this tree occupies the higher forest stratum, characterizing the landscape of the Atlantic Forest highlands, thus being considered a symbol of the Brazilian southern region. Araucaria trees are profitable regarding non-wood utilisation. The species produces a highly nutritious seed (pinhão in Portuguese), appreciated by humans and animals. There is a culturally established market around the seed, which, unfortunately, is not developed enough to allow sufficient economical return. Araucaria trees are also attractive due to its high quality and aesthetically pleasing wood, which led to its intense exploration in the 1960s and 1970s. Nowadays, the species is classified as critically endangered and is protected by law against illegal logging. However, such restrictive laws have resulted in further threat to the species. Knowing the legal difficulties to remove a grown araucaria tree from their property, landowners remove new natural saplings. Practices have been proposed to promote other sustainable uses for the species. Still, lawmakers in consort with researchers and specialists can only legislate upon detailed data regarding the species. In order to collect such paramount information throughout the species occurrence areas (approx. 200,000 km2 ), specialists face an extremely very fragmented environment, which poses operational and financial difficulties to acquire the data. Aims: The aim of this thesis was to introduce a methodology to automatically detect and measure Araucaria angustifolia in complex native forest formations in southern Brazil. The proposed methodology leverages on light detection and ranging (LiDAR) as well as high resolution aerial imagery. Normally, species mapping and measurement is conducted by combining LiDAR and spectral information (e.g. aerial, satellite or drone imagery). In this study, an analysis was performed to determine if there is, in fact, the necessity to add spectral information to map araucaria trees. Lastly, an alternative method was proposed, where no LiDAR data is required. A novel methodology was developed to detect araucarias from unmanned aerial vehicles (UAV), as an alternative to LiDAR-based methodologies. Research questions: 1. Is spectral information imperative to map A. angustifolia or LiDAR data alone is enough? 2. Can A. angustifolia trees be detected and measured in dense forest formation using Remote Sensing data? 3. If araucaria trees are detectable, can tree parameters (e.g. total height, diameter at breast height (DBH) and crown area) be acquired with reasonable accuracy? 4. Can A. angustifolia trees be detected based on their morphology, i.e. branches distribution? Study site and data: The data analysed in this thesis come from a municipality called Lages, located in the state of Santa Catarina in the Brazilian southern region. The two study sites contain fragments of the Atlantic Forest, where the target species A. angustifolia can be found with 38 and 34 ha for study site A and B, respectively. Both study sites are covered with LiDAR data with an average point density of 14 points/m2 and aerial imagery with 0.1 m spatial resolution. The datasets were collected in the same flight performed in June 2019. In addition, field data from March 2016 was available from 10 plots, each with 0.2 ha (total sampled area of 2 ha), located in site A, where all araucaria trees within the plots were measured and georeferenced. Lastly, UAV imagery with ground sample distance of 5 cm was also available for study site A, also collected in March 2016. Methodological approach: The methodology implemented in this thesis consisted mainly of two parts: (1) araucaria tree mapping and forest parameter estimation using LiDAR and aerial imagery; (2) detection of araucarias based on branch recognition from UAV imagery. 1. In order to map araucaria trees employing LiDAR and aerial imagery, a Random Forest classification was conducted. An analysis was performed to determine the efficiency of the classification when using only LiDAR data and when adding spectral information to it. Moreover, the random forest classifier was trained in site A and tested in site B. With the result of the mapping, a clipping mask was generated and used to clip the LiDAR point cloud. The clipped point cloud was assessed in terms of individual tree detection (ITD) as a means to determine the number of stems per hectare, total tree height and crown area, as well as estimate DBH. 2. A new methodology was developed as an alternative for LiDAR-based approaches. The approach consists of recognizing araucaria branches and use their orientation to determine A. angustifolia tree locations. The approach was implemented using a computer vision method called Probabilistic Hough Transform associated with other image processing techniques such as morphological filtering and image segmentation. By employing such techniques, the branches were detected as lines, which then could be used to calculate branch orientation, culminating on tree location. Results and discussion: LiDAR data is commonly used for commercial conifer tree species mapping and have been used for inventory purposes operationally in many countries such as Finland, Sweden, Canada, the United States and others. However, fewer have explored the applications of LiDAR data in complex environments such as the Atlantic Forest. The reason is mainly due to the multi layer structure in native forests and the high occurrence of tree occlusion, which affects the stems count, an important forest parameter for inventory purposes. In this study, such reasons were also noticed, even considering araucarias’ crown size and the fact that the adult individuals of the species usually are located in the upper layer of the forest canopy. Nonetheless, it was possible to determine that the majority of A. angustifolia trees were successfully mapped employing LiDAR data. Moreover, there was enough statistical evidence to state that no difference was found when mapping A. angustifolia employing only LiDAR data and combined spectral information and LiDAR data. When comparing both maps derived from the Random Forest classification, it was possible to observe similar performance from both datasets. Overall accuracies of 90.8% and 89.8% were observed for sites A and B, respectively. Even though adult trees of araucaria are usually visible in the upper forest canopy, a more basic operation such as a height threshold would not be able to separate araucarias from the rest of the species. This happens since there are many other species that occupy the same height level, which could result in commission errors. Hence, a supervised classification such as RF was efficient in removing the remainder of the tree species. When addressing the stems count, after running a ITD approach using local maxima detection an overall accuracy of 73.34% was reached. That resulted in a density of 43 stems/ha, which is below the 61 stems/ha calculated from the field data. This difference in mainly due to tree occlusion, which is often observed in multi layer structure of complex natural environments such as araucaria forests. If a comparison is performed against the actual visible trees, the accuracy would be increased to 87.9%. There is still 12.1% error considering the upper visible forest canopy, which is caused by smaller trees close to dominant ones, resulting in further omissions. One of the challenges of working with LiDAR and araucarias is the species morphology. Araucarias possess a unique crown shape, commonly described in the literature as being similar to a wine glass or an inverted chandelier. However, a combination of uneven terrain and oddly shaped crowns results in a distorted normalized point cloud, which in turn, affects the total tree height generated from it. As a solution, the ITD was performed using the digital surface model (DSM) to detect the local maxima. Once the coordinate of the highest point of a tree was determined, these coordinates were used to retrieve the tree height from the CHM. As a result, the total height and crown diameter measurements reached errors of 1.44 m and 1.72 m, respectively. However, point cloud normalization was not the only probable source of variation. The field measurements were performed in March 2016, while the LiDAR data was acquired in June 2019. This represents a difference of 39 months between measurements. Evidently, adult trees are not expected to grow too much, specially when a slow-growth native species such as araucaria is concerned. Yet, this discrepancy needs to be considered when assessing the results. Hence, it was not possible to determine if the height and crown measurements were affected by the field measurements procedure (field measurement errors), the time difference between LiDAR and field data acquisitions or the methodology proposed in this study. Lastly, considering that the DBH was estimated from the total height and crown diameter, these inconsistencies are carried over to the estimates, yielding a DBH error of 9.89 cm. Araucaria trees are easy to distinguish from other tree species when observed from nadir. Due to the unique format of the species crown, a novel approach based on the branch distribution from an orthogonal view was proposed to automatically detect araucaria trees.After implementing and testing this approach, an overall accuracy of 93% was achieved. During the analysis, a difficulty index was introduced, in which trees easily distinguishable were assigned difficulty level 1 (easy), partially occluded ones received index 2 (medium) and severely occluded but still partially visible ones were assigned index 3 (hard). The highest accuracy was achieved with difficulty index 1 with overall accuracy of 98%, followed by 92% and 89% for difficulty index 2 and 3, respectively. If the difficulty index is ignored and the tree detection is assessed as a whole, an overall accuracy of 93% was reached. The methodology demonstrated to be robust, considering it relies solely on branches to determine the tree location. Nonetheless, limitations were observed with this approach. Considering that branches are the main element of this methodology, if they are not visible, trees are simply not detected. Adult araucaria trees present visible branches, which are commonly very thick (reaching up to 30 cm in diameter based on field observations). However, a high density of secondary and tertiary branches may interfere in the visibility of primary branches. Lastly, adjacent trees with branches similarly oriented also result in omissions, since they seem merged in the image, resulting in only one branch being detected. Conclusions: In this thesis, evidences demonstrated that LiDAR data can be used for A. angustifolia mapping and forest parameter estimation. Moreover, when considering the test sites addressed in this study, the addition of spectral information didn’t significantly improve the mapping, leading to the conclusion that LiDAR data alone is enough for A. angustifolia mapping. Lastly, when working with native species, conventional methods might not be the best practice and approaching issues with different perspectives can yield new solutions. The new proposed method is a clear example of that. By using the species unique morphology as basis, the approach has showed promising results, which could be further improved in future research. This thesis constitutes the first study to provide an in-depth analysis on the use of LiDAR data to automatically map Araucaria angustifolia in natural dense forest formations. Moreover, considering the current situation of the species, this work contributes to a better understanding of the challenges when working with araucaria trees as well as working with complex forest structures. In addition, further work can be developed based on this study, which could provide even more accurate large-scale information to lawmakers, researchers and specialists when developing new strategies to sustainably manage the species.
... In this study, the impact of height normalization on the estimation and extraction of tree height and LVV was negligible because the terrain is relatively flat. Khosravipour et al. [58] discovered that on steep slopes, the raw elevation values located on either the downhill or the uphill part of a tree crown are height-normalized with parts of the digital terrain model that may be much lower or higher than the tree stem base, respectively. They suggest that in order to minimize the negative effect of steep slopes on the CHM, it ...
... In this study, the impact of height normalization on the estimation and extraction of tree height and LVV was negligible because the terrain is relatively flat. Khosravipour et al. [58] discovered that on steep slopes, the raw elevation values located on either the downhill or the uphill part of a tree crown are height-normalized with parts of the digital terrain model that may be much lower or higher than the tree stem base, respectively. They suggest that in order to minimize the negative effect of steep slopes on the CHM, it is best to use raw elevation values (i.e., use the un-normalized DSM) and compute the height after treetop detection. ...
Article
Full-text available
The living vegetation volume (LVV) can accurately describe the spatial structure of greening trees and quantitatively represent the relationship between this greening and its environment. Because of the mostly line shape distribution and the complex species of street trees, as well as interference from artificial objects, current LVV survey methods are normally limited in their efficiency and accuracy. In this study, we propose an improved methodology based on vehicle-mounted LiDAR data to estimate the LVV of urban street trees. First, a point-cloud-based CSP (comparative shortest-path) algorithm was used to segment the individual tree point clouds, and an artificial objects and low shrubs identification algorithm was developed to extract the street trees. Second, a DBSCAN (density-based spatial clustering of applications with noise) algorithm was utilized to remove the branch point clouds, and a bottom-up slicing method combined with the random sampling consistency iterative method algorithm (RANSAC) was employed to calculate the diameters of the tree trunks and obtain the canopy by comparing the variation in trunk diameters in the vertical direction. Finally, an envelope was fitted to the canopy point cloud using the adaptive AlphaShape algorithm to calculate the LVVs and their ecological benefits (e.g., O2 production and CO2 absorption). The results show that the CSP algorithm had a relatively high overall accuracy in segmenting individual trees (overall accuracy = 95.8%). The accuracies of the tree height and DBH extraction based on vehicle-mounted LiDAR point clouds were 1.66~3.92% (rRMSE) and 4.23~15.37% (rRMSE), respectively. For the plots on Zijin Mountain, the LVV contribution by the maple poplar was the highest (1049.667 m³), followed by the sycamore tree species (557.907 m³), and privet’s was the lowest (16.681 m³).
... The performance of ITD approaches also depends on object and point cloud characteristics. The object characteristics include the type of land cover (e.g., tree species, crown shape, and stem density) [14], the site conditions (e.g., topography and type of terrain) [14], and the point cloud characteristics related to the type of LiDAR sensor and mission parameters (e.g., flying height, scanning angle, scanning mode, footprint size, and point density) [15,16]. ...
... The performance of ITD approaches also depends on object and point cloud characteristics. The object characteristics include the type of land cover (e.g., tree species, crown shape, and stem density) [14], the site conditions (e.g., topography and type of terrain) [14], and the point cloud characteristics related to the type of LiDAR sensor and mission parameters (e.g., flying height, scanning angle, scanning mode, footprint size, and point density) [15,16]. ...
Article
Full-text available
Numerous semi- and fully-automatic algorithms have been developed for individual tree detection from airborne laser-scanning data, but different rates of falsely detected treetops also accompany their results. In this paper, we proposed an approach that includes a machine learning-based refinement step to reduce the number of falsely detected treetops. The approach involves the local maxima filtering and segmentation of the canopy height model to extract different segment-level features used for the classification of treetop candidates. The study was conducted in a mixed temperate forest, predominantly deciduous, with a complex topography and an area size of 0.6 km × 4 km. The classification model’s training was performed by five machine learning approaches: Random Forest (RF), Extreme Gradient Boosting, Artificial Neural Network, the Support Vector Machine, and Logistic Regression. The final classification model with optimal hyperparameters was adopted based on the best-performing classifier (RF). The overall accuracy (OA) and kappa coefficient (κ) obtained from the ten-fold cross validation for the training data were 90.4% and 0.808, respectively. The prediction of the test data resulted in an OA = 89.0% and a κ = 0.757. This indicates that the proposed method could be an adequate solution for the reduction of falsely detected treetops before tree crown segmentation, especially in deciduous forests.
... The slope, which is usually at least 20 degrees in cable yarder terrain, can influence the assignment of the detected trees to the reference trees. Depending on the slope and the crown shape of the tree, the tree top may be shifted in relation to the trunk position (Khosravipour et al. 2015). ALS points that are slightly upslope or downslope from the tree top are height-normalized with a digital elevation model pixel value that may be significantly higher or lower than the base of the trunk. ...
... ALS points that are slightly upslope or downslope from the tree top are height-normalized with a digital elevation model pixel value that may be significantly higher or lower than the base of the trunk. Therefore, Khosravipour et al. (2015) suggested performing single tree detection with a digital surface model (DSM) and then reading the tree heights from the CHM. This displacement between the base of the trunk and the top of the tree may occur as a result of factors other than terrain slope and crown shape. ...
Article
Full-text available
For the provision of various ecosystem services in steep terrain, such as protection against natural hazards, a forest must be managed, which often requires the use of cable yarders. The design of a cable road is a complex and demanding task that also includes the search for appropriate support and anchor trees. The aim of this study was to evaluate whether and with what reliability potential support trees for cable yarding can be detected using remote sensing data. The detection of potential support trees was tested using 48 method combinations on 10 test plots of the Experimental Forest Management project in cable yarder terrain in the Swiss Alps in the Canton of Grisons. The most suitable method combinations used a Gaussian filter and a local maxima algorithm. On average, they had an extraction rate of 108.9–124.5% (root mean square, RMS) and a mean commission error of 66.0–67.2% (RMS). The correctly detected trees deviated horizontally by an average of 1.8 to 1.9 m from the position of the reference trees. The difference in tree heights was 1.1 to 1.6 m. However, for the application of single tree detection to support cable road planning in steep and complex terrain, too few potential support trees were detected. Nonetheless, the accuracy of the extracted tree parameters would already be sufficient for cable road planning.
... Tree crowns tend to develop toward the slope direction to maximize their light interception, which results in stem inclination [53], and the DEM-based normalized point cloud might further lead to a systematic overestimation of tree height [54]. Khosravipour et al. [55] reported that the terrain effect on tree height estimation closely links to crown shape, which is species-dependent. The tree species are various in this study, and the effect on tree height estimation may be complex. ...
... The tree species are various in this study, and the effect on tree height estimation may be complex. Understanding how crown morphology affects UAV-LiDAR-based tree estimation would be interesting [55]. Finally, a possible uncertainty is from field measurement of tree height. ...
Article
Full-text available
Accurate individual tree segmentation (ITS) is fundamental to forest management and to the studies of forest ecosystem. Unmanned Aerial Vehicle Light Detection and Ranging (UAV-LiDAR) shows advantages for ITS and tree height estimation at stand and landscape scale. However, dense deciduous forests with tightly interlocked tree crowns challenge the performance for ITS. Available LiDAR points through tree crown and appropriate algorithm are expected to attack the problem. In this study, a new UAV-LiDAR dataset that fused leaf-off and leaf-on point cloud (FULD) was introduced to assess the synergetic benefits for ITS and tree height estimation by comparing different types of segmentation algorithms (i.e., watershed segmentation, point cloud segmentation and layer stacking segmentation) in the dense deciduous forests of Northeast China. Field validation was conducted in the four typical stands, including mixed broadleaved forest (MBF), Mongolian oak forest (MOF), mixed broadleaf-conifer forest (MBCF) and larch plantation forest (LPF). The results showed that the combination of FULD and the layer stacking segmentation (LSS) algorithm produced the highest accuracies across all forest types (F-score: 0.70 to 0.85). The FULD also showed a better performance on tree height estimation, with a root mean square error (RMSE) of 1.54 m at individual level. Compared with using the leaf-on dataset solely, the RMSE of tree height estimation was reduced by 0.22 to 0.27 m, and 12.3% more trees were correctly segmented by the FULD, which are mainly contributed by improved detection rate at nearly all DBH levels and by improved detection accuracy at low DBH levels. The improvements are attributed to abundant points from the bole to the treetop of FULD, as well as each layer point being included for segmentation by LSS algorithm. These findings provide useful insights to guide the application of FULD when more multi-temporal LiDAR data are available in future.
... On the other hand, our measurements were collected in mountain areas, and thus the topographic effect may have further influenced the ALS retrieval. 54 In this line, a previous study found that this effect is higher in trees with complex crown and weak apical dominance, 55 which in turn may partly contribute to the lower performance found in beech, considering that both sampled forest types have similar slope conditions; the same authors also observed that the combined effect of topographic slope and crown structure was not significant in a mountain pine forest, which may be comparable with our sampled mountain pine forest. 55 These differences may explain the different Lidar performance observed in the sampled forest types. ...
... 54 In this line, a previous study found that this effect is higher in trees with complex crown and weak apical dominance, 55 which in turn may partly contribute to the lower performance found in beech, considering that both sampled forest types have similar slope conditions; the same authors also observed that the combined effect of topographic slope and crown structure was not significant in a mountain pine forest, which may be comparable with our sampled mountain pine forest. 55 These differences may explain the different Lidar performance observed in the sampled forest types. ...
... O mapeamento do desvio padrão indicou que as maiores incertezas estão associadas as áreas de maior declividade. Esse fato ocorre porque ao se utilizar sensor LiDAR em áreas declivosas há redução na precisão da detecção das copas das árvores, ocorrendo um deslocamento horizontal dessas copas no processo de normalização, ocasionando assim limitações na estimativa da altura das árvores [10]. ...
Conference Paper
Full-text available
O objetivo do estudo foi estimar a biomassa acima do solo (AGB), em áreas de floresta Amazônica com extração seletiva de madeira, utilizando os algoritmos de aprendizado de máquina RF (Random Forest) e SVM (Support Vector Machine) e dados dos sensores LiDAR e OLI/Landsat 8. Para tal, foram utilizadas 79 parcelas de 50x50m dispostas na Fazenda Cauaxi, Paragominas, Pará. A predição da AGB foi realizada por meio dos algoritmos RF e SVM utilizando dados LiDAR e estes combinados com variáveis espectrais do sensor OLI/Landsat 8. Utilizou-se o RMSE (Root Mean Square Error) para verificar o desempenho dos algortimos. O SVM utilizando apenas LiDAR apresentou o menor RMSE médio (44,99 Mg/ha), contudo a associação das variáveis permitiu aumentar o desempenho do RF. Desse modo, foi possível inferir que os algoritmos de aprendizado de máquina se mostraram eficientes para estimar a AGB.
... The height overestimation for the NFI plots may be related to i) systematic errors of lidar-based heights for sloping terrain and ii) field measurement errors. As previous studies (Khosravipour et al., 2015;Sibona et al., 2017) suggest that only marginal differences (0.5 m-1 m) occurs when estimating tree height from small footprint lidar data for slopes around 20 • degree, as those encountered in the Basque Country, it seems most of the ME may be related to errors of the indirect in-situ height measurements. This assumption seems valid considering the higher ME for the Basque Country where forests are taller (14.5 vs 9.0 m) and denser (65% vs. 55% cover) when compared to the Madrid region. ...
... The Digital Terrain Model (DTM) and Digital Surface Model (DSM) are common numerical elevation models. The Canopy Height Model (CHM), which is synonymous with a normalized digital surface model (nDSM) for forest areas, is often discussed in the literature on forest areas [8][9][10][11]. CHM models are used to perform analyses to extract information about a stand of trees. They are often used in forestry to determine the quantities that characterize a stand, which is important for describing tree growth [12], planning silvicultural activities [13], and inventorying timber volume and stand biomass [12,14,15]. ...
Article
Full-text available
Modern technologies, such as airborne laser scanning (ALS) and advanced data analysis algorithms, allow for the efficient and safe use of resources to protect infrastructure from potential threats. This publication presents a study to identify trees that may fall on highways. The study used free measurement data from airborne laser scanning and wind speed and direction data from the Institute of Meteorology and Water Management in Poland. Two methods were used to determine the crown tops of trees: PyCrown and OPALS. The effect of wind direction on potential hazards was then analyzed. The OPALS method showed the best performance in terms of detecting trees, with an accuracy of 74%. The analysis showed that the most common winds clustered between 260° and 290°. Potential threats, i.e., trees that could fall on the road, were selected. As a result of the analysis, OPALS detected between 140 and 577 trees, depending on the chosen strategy. The presented research shows that combining ALS technology with advanced algorithms and wind data can be an effective tool for identifying potential hazards associated with falling trees on highways.
... The accuracy of seed point detection can have an influence on individual tree segmentation. The traditional methods (e.g., local maximum method) are strongly dependent on the window size [42,43]. If the window is too large, the treetops of small trees may be overlooked, and if it is too small, the tops of elongated branches are likely to be misidentified as seed points. ...
Article
Full-text available
Unmanned aerial vehicle–light detection and ranging (UAV-LiDAR) provides a convenient and economical means of forest data acquisition that can penetrate canopy gaps to obtain abundant ground information, offering huge potential in forest inventory. Individual tree segmentation is a prerequisite to obtain individual tree details but is highly dependent on the accuracy of seed point detection. However, most of the existing methods, such as the local maximum (LM) and CHM-based methods, are strongly dependent on the window size, and, for individual tree segmentation, they can result in over-segmentation and under-segmentation, especially in natural forests. In this paper, we propose an adaptive crown shaped algorithm for individual tree segmentation without consideration of the window size. It was implemented in four plots with different forest types and topographies (i.e., planted coniferous forest with flat terrain, coniferous forest with sloping terrain, mixed forest with flat terrain and broadleaf forest with flat terrain). First, the normalized point clouds were rotated and blocked at multiple angles to extract the surface points of the forest. Then, the crown boundaries were delineated by analyzing the crown profiles to extract the treetops as seed points. Finally, a region growing method based on seed points was applied for individual tree segmentation. Our results showed that the recall, precision and F1-score of seed point detection reached 91.6%, 95.9% and 0.94, respectively, and that the accuracy rates for individual tree segmentation for the four plots were 87.7%, 80.6%, 73.2% and 70.5%, respectively. Our proposed method can effectively detect seed points via the adaptive crown shaped algorithm and reduce the impacts of elongated branches by applying distance thresholds between trees, enhancing the accuracy of seed point detection and subsequently improving the precision of individual tree segmentation. In addition, the proposed algorithm demonstrated superior performance in comparison to LM and CHM-based methods for the calculation of seed points, as well as outperforming PCS in individual tree segmentation. The proposed method demonstrates effectiveness and feasibility in dense forests and natural forests, providing an important reference for future research on seed point detection and individual tree segmentation.
... However, in actual LiDAR datasets, some problems exist that are not present in synthetic datasets. For example, individual-tree segmentation results are substantially affected by surface topography [54,55]. In steep areas, the normalization of the point cloud may change the shape of crowns and would affect the performance of the proposed algorithms. ...
Article
Full-text available
Accurate individual-tree segmentation is essential for precision forestry. In previous studies, the canopy height model-based method was convenient to process, but its performance was limited owing to the loss of 3D information, and point-based methods usually had high computational costs. Although some hybrid methods have been proposed to solve the above problems, most canopy height model-based methods are used to detect subdominant trees in one coarse crown and disregard the over-segmentation and accurate segmentation of the crown boundaries. This study introduces a combined approach, tested for the first time, for treetop detection and tree crown segmentation using UAV–LiDAR data. First, a multiscale adaptive local maximum filter was proposed to detect treetops accurately, and a Dalponte region-growing method was introduced to achieve crown delineation. Then, based on the coarse-crown result, the mean-shift voxelization and supervoxel-weighted fuzzy c-means clustering method were used to identify the constrained region of each tree. Finally, accurate individual-tree point clouds were obtained. The experiment was conducted using a synthetic uncrewed aerial vehicle (UAV)–LiDAR dataset with 21 approximately 30 × 30 m plots and an actual UAV–LiDAR dataset. To evaluate the performance of the proposed method, the accuracy of the remotely sensed biophysical observations and retrieval frameworks was determined using the tree location, tree height, and crown area. The results show that the proposed method was efficient and outperformed other existing methods.
... These measurement from LiDAR, potentially distorting the plant area index (PAI) contours and altering the spatial distribution patterns of plant area density. This effect becomes more pronounced with increasing slope [28,29]. ...
Article
Full-text available
The leaf area index (LAI) serves as a crucial metric in quantifying the structure and density of vegetation canopies, playing an instrumental role in determining vegetation productivity, nutrient and water utilization, and carbon balance dynamics. In subtropical montane forests, the pronounced spatial heterogeneity combined with undulating terrain introduces significant challenges for the optical remote sensing inversion accuracy of LAI, thereby complicating the process of ground validation data collection. The emergence of UAV LiDAR offers an innovative monitoring methodology for canopy LAI inversion in these terrains. This study assesses the implications of altitudinal variations on the attributes of UAV LiDAR point clouds, such as point density, beam footprint, and off-nadir scan angle, and their subsequent ramifications for LAI estimation accuracy. Our findings underscore that with increased altitude, both the average off-nadir scan angle and point density exhibit an ascending trend, while the beam footprint showcases a distinct negative correlation, with a correlation coefficient (R) reaching 0.7. In contrast to parallel flight paths, LAI estimates derived from intersecting flight paths demonstrate superior precision, denoted by R2 = 0.70, RMSE = 0.75, and bias = 0.42. Notably, LAI estimation discrepancies intensify from upper slope positions to middle positions and further to lower ones, amplifying with the steepness of the gradient. Alterations in point cloud attributes induced by the terrain, particularly the off-nadir scan angle and beam footprint, emerge as critical influencers on the precision of LAI estimations. Strategies encompassing refined flight path intervals or multi-directional point cloud data acquisition are proposed to bolster the accuracy of canopy structural parameter estimations in montane landscapes.
... Canopy height models (CHMs) derived from LiDAR point cloud data have been commonly used to generate the boundaries of individual tree crowns [49,118,119], which is obtained through cutting a digital surface model (DSM) with a digital terrain model (DTM) [120][121][122]. To maintain the same spatial resolution as aerial photographs, we processed AHN3 LiDAR point cloud data to generate CHMs with a spatial resolution of 25 cm. ...
Article
Full-text available
Integrating multimodal remote sensing data can optimize the mapping accuracy of individual trees. Yet, one issue that is not trivial but generally overlooked in previous studies is the spatial mismatch of individual trees between remote sensing datasets, especially in different imaging modalities. These offset errors between the same tree on different data that have been geometrically corrected can lead to substantial inaccuracies in applications. In this study, we propose a novel approach to match individual trees between aerial photographs and airborne LiDAR data. To achieve this, we first leveraged the maximum overlap of the tree crowns in a local area to determine the correct and the optimal offset vector, and then used the offset vector to rectify the mismatch on individual tree positions. Finally, we compared our proposed approach with a commonly used automatic image registration method. We used pairing rate (the percentage of correctly paired trees) and matching accuracy (the degree of overlap between the correctly paired trees) to measure the effectiveness of results. We evaluated the performance of our approach across six typical landscapes, including broadleaved forest, coniferous forest, mixed forest, roadside trees, garden trees, and parkland trees. Compared to the conventional method, the average pairing rate of individual trees for all six landscapes increased from 91.13% to 100.00% (p = 0.045, t-test), and the average matching accuracy increased from 0.692 ± 0.175 (standard deviation) to 0.861 ± 0.152 (p = 0.017, t-test). Our study demonstrates that the proposed tree-oriented matching approach significantly improves the registration accuracy of individual trees between aerial photographs and airborne LiDAR data.
... Airborne Light Detection and Ranging (LiDAR) is an established and reliable means of generating data on the physical structure of topography, vegetation and man-made structures that have been used for diverse purposes, including forestry (Naesset and Økland 2002), landscape mapping (Wang et al. 2021), habitat modelling and assessment (Getzin et al. 2021;Hagar et al. 2020), landcover types and habitat classification , natural resources management (Garabedian et al. 2017) and wildfire modelling (Botequim et al. 2019;Rosa and Stow 2014). Airborne LiDAR at resolutions of < 1 m is capable of providing information on the scale of individual trees (Jaskierniak et al. 2021;Lichstein et al. 2010), which can be important for a variety of forestry activities and for environmental modelling (Khosravipour et al. 2015). Such high-resolution spatial data also have the potential to identify landscape features that are important for the survival of individual organisms, whilst highlighting ecological variations at multiple scales, in local areas, regions and countries. ...
Article
Full-text available
The hazel dormouse is predominantly an arboreal species that moves down to the ground to hibernate in the autumn in temperate parts of its distributional ranges at locations not yet well understood. The main objective of this study is to test whether environmental characteristics surrounding hazel dormouse hibernacula can be identified using high-resolution remote sensing and data collected in situ. To achieve this, remotely sensed variables, including canopy height and cover, topographic slope, sky view, solar radiation and cold air drainage, were modelled around 83 dormouse hibernacula in England ( n = 62) and the Netherlands ( n = 21), and environmental characteristics that may be favoured by pre-hibernating dormice were identified. Data on leaf litter depth, temperature, canopy cover and distance to the nearest tree were collected in situ and analysed at hibernaculum locations in England. The findings indicated that remotely sensed data were effective in identifying attributes surrounding the locations of dormouse hibernacula and when compared to in situ information, provided more conclusive results. This study suggests that remotely sensed topographic slope, canopy height and sky view have an influence on hazel dormice choosing suitable locations to hibernate; whilst in situ data suggested that average daily mean temperature at the hibernaculum may also have an effect. Remote sensing proved capable of identifying localised environmental characteristics in the wider landscape that may be important for hibernating dormice. This study proposes that this method can provide a novel progression from habitat modelling to conservation management for the hazel dormouse, as well as other species using habitats where topography and vegetation structure influence fine-resolution favourability.
... For P1, the result is relatively poor with the R 2 = .5. There are three reasons for this: (1) the average slope of this plot is 30°, so the point cloud normalization will cause distortion of the trees (Khosravipour et al., 2015). (2) there is distortion of the trunk of birch due to the natural environment; (3) there is no obvious top of broadleaf trees, which is different from coniferous trees. ...
Article
Full-text available
Light detection and ranging (LiDAR) data can provide 3D structural information of objects and are ideal for extracting individual tree parameters, and individual tree segmentation (ITS) is a vital step for this purpose. Various ITS methods have been emerging from airborne LiDAR scanning (ALS) or unmanned aerial vehicle LiDAR scanning (ULS) data. Here, we propose a new individual tree segmentation method, which couples the classical and efficient watershed algorithm (WS) and the newly developed connection center evolution (CCE) clustering algorithm in pattern recognition. The CCE is first used in ITS and comprehensively optimized by considering tree structure and point cloud characteristics. Firstly, the amount of data is greatly reduced by mean shift voxelization. Then, the optimal clustering scale is automatically determined by the shapes in the projection of three different directions. We select five forest plots in Saihanba, China and 14 public plots in Alpine region, Europe with ULS or ALS point cloud densities from 11 to 3295 pts/m2. Eleven ITS methods were used for comparison. The accuracy of tree top detection and tree height extraction is estimated by five and two metrics, respectively. The results show that the matching rate (R match) of tree tops is up to 0.92, the coefficient of determination (R 2) of tree height estimation is up to .94, and the minimum root mean square error (RMSE) is 0.6 m. Our method outperforms the other methods especially in the broadleaf forests plot on slopes, where the five evaluation metrics for tree top detection outperformed the other algorithms by at least 11% on average. Our ITS method is both robust and efficient and has the potential to be used especially in coniferous forests to extract the structural parameters of individual trees for forest management, carbon stock estimation, and habitat mapping.
... A more subtle case of erroneous terrain-induced high points happens in forested terrain with steep slopes or cliffs. The effect of sloped terrain on deriving forest metrics from LiDAR is known (Khosravipour et al., 2015). Vegetation points of trees can reside far enough out horizontally from the tree stem such that the ground point selected for normalizing is substantially lower compared to the true ground level of the tree. ...
... High terrain slopes often lead to large errors in height metrics derived from LiDAR data (Breidenbach et al., 2008;Hollaus et al., 2006;Khosravipour et al., 2015). In this study, the tree level AGB estimates derived using the LBI-based approach had errors at all slope levels, but the absolute value of the mean error did not show obvious trends when the slope increased from 0 to 30 degrees (Fig. 13), suggesting that AGB estimates derived using the LBI-based approach might not be very sensitive to terrain slopes of up to 30 • . ...
... Since we extracted the NDVI values at the local maxima location, i.e., for one pixel only, this approach may have led to the misclassification of healthy trees as dead if there were dead branches near the apex. In addition, ref. [68] reported a negative effect of the slope on the tree detection, leading to horizontal and/or vertical displacements of the tree apex of more than a meter in extreme cases, and therefore extraction of the wrong pixel to determine the status. An alternative to the extraction of a single pixel could be to average the pixel values within a buffer distance from the location of the local maximum. ...
Article
Full-text available
Deadwood is an important key ecological element for forest ecosystem biodiversity. Its low occurrence, especially in managed forests, makes inventory through field campaigns challenging. Remote sensing can provide a more objective and systematic approach to detect deadwood for large areas. Traditional area-based approaches have, however, shown limitations when it comes to predicting rare objects such as standing dead trees (SDT). To overcome this limitation, this study proposes a tree-based approach that uses a local maxima function to identify trees from airborne laser scanning (ALS) and optical data, and predict their status, i.e., living or dead, from normalized difference vegetation index (NDVI). NDVI was calculated from aerial images (hyperspectral and simulated aerial image) and from satellite images (PlanetScope and Sentinel-2). By comparing the different remotely sensed data sources, we aimed to assess the impact of spatial and spectral resolutions in the prediction of SDT. The presence/absence of SDT was perfectly predicted by combining trees identified using ALS-derived canopy height models with spatial resolutions between 0.75 m and 1 m and a search window size of 3 pixels, and NDVI computed from aerial images to predict their status. The presence/absence of SDT was not predicted as accurately when using NDVI computed from satellite images. A root-mean-square deviation of around 35 trees ha−1 was obtained when predicting the density of SDT with NDVI from aerial images and around 60 trees ha−1 with NDVI from satellite images. The tree-based approach presented in this study shows great potential to predict the presence of SDT over large areas.
... The algorithm required two inputs: a canopy height model (CHM) and tree top points. Although we tried to mirror the analysis steps for both ALS and MLS, we found that the best algorithm for generating CHMs differed between ALS and MLS, with the former benefitting from a simpler triangulated irregular network-based interpolation, and the latter benefitting from the more complex algorithm of Khosravipour et al. (2015). Tree top points came from the field-acquired GNSS points, although three adjustments needed to be made to ensure accurate crown delineation. ...
Article
Full-text available
Piñon-juniper (PJ) woodlands are a widespread dryland ecosystem in the US containing an immense but poorly-constrained amount of aboveground biomass (AGB). Found at the dry end of the climatic range within which trees can persist, PJ faces an uncertain future in a changing climate, giving unique importance to mapping its AGB, past, present, and future. Lidar remote sensing offers great potential towards that end with research in a range of tree-dominant ecosystems demonstrating a strong capacity for mapping AGB. However, studies applying lidar to the task of mapping AGB in PJ are few. Given the unique structural characteristics of PJ trees, which tend to be short in stature (
... As expected, we found that topographic characteristics have effects on the detection accuracy of tree crowns, which is in line with the observations of Khosravipour et al. [55] and Nie et al. [56], who carried out treetop detection using canopy height models derived from LiDAR. Alexander et al. [57] also found that topography influences tree detection and height estimations from LiDAR canopy height models in tropical forests. ...
Article
Full-text available
The automatic detection of tree crowns and estimation of crown areas from remotely sensed information offer a quick approach for grasping the dynamics of forest ecosystems and are of great significance for both biodiversity and ecosystem conservation. Among various types of remote sensing data, unmanned aerial vehicle (UAV)-acquired RGB imagery has been increasingly used for tree crown detection and crown area estimation; the method has efficient advantages and relies heavily on deep learning models. However, the approach has not been thoroughly investigated in deciduous forests with complex crown structures. In this study, we evaluated two widely used, deep-learning-based tree crown detection and delineation approaches (DeepForest and Detectree2) to assess their potential for detecting tree crowns from UAV-acquired RGB imagery in an alpine, temperate deciduous forest with a complicated species composition. A total of 499 digitized crowns, including four dominant species, with corresponding, accurate inventory data in a 1.5 ha study plot were treated as training and validation datasets. We attempted to identify an effective model to delineate tree crowns and to explore the effects of the spatial resolution on the detection performance, as well as the extracted tree crown areas, with a detailed field inventory. The results show that the two deep-learning-based models, of which Detectree2 (F1 score: 0.57) outperformed DeepForest (F1 score: 0.52), could both be transferred to predict tree crowns successfully. However, the spatial resolution had an obvious effect on the estimation accuracy of tree crown detection, especially when the resolution was greater than 0.1 m. Furthermore, Dectree2 could estimate tree crown areas accurately, highlighting its potential and robustness for tree detection and delineation. In addition, the performance of tree crown detection varied among different species. These results indicate that the evaluated approaches could efficiently delineate individual tree crowns in high-resolution optical images, while demonstrating the applicability of Detectree2, and, thus, have the potential to offer transferable strategies that can be applied to other forest ecosystems.
... Among the conventional approaches, the local maxima (LM) and marker-controlled watershed segmentation (MCWS) algorithms are the most common detection methods [12]. The LM algorithm is appropriate for trees like different coniferous species in which a point is visible as the brightest pixel in UAV images or as the highest point in CHM [29,30]. The highest points can be identified by using a moving window to filter the image [13]. ...
Article
Full-text available
Monitoring tree decline in arid and semi-arid zones requires methods that can provide up-to-date and accurate information on the health status of the trees at single-tree and sample plot levels. Unmanned Aerial Vehicles (UAVs) are considered as cost-effective and efficient tools to study tree structure and health at small scale, on which detecting and delineating tree crowns is the first step to extracting varied subsequent information. However, one of the major challenges in broadleaved tree cover is still detecting and delineating tree crowns in images. The frequent dominance of coppice structure in degraded semi-arid vegetation exacerbates this problem. Here, we present a new method based on edge detection for delineating tree crowns based on the features of oak trees in semi-arid coppice structures. The decline severity in individual stands can be analyzed by extracting relevant information such as texture from the crown area. Although the method presented in this study is not fully automated, it returned high performances including an F-score = 0.91. Associating the texture indices calculated in the canopy area with the phenotypic decline index suggested higher correlations of the GLCM texture indices with tree decline at the tree level and hence a high potential to be used for subsequent remote-sensing-assisted tree decline studies.
... Unmanned aerial vehicle (UAV) is a more flexible tool than satellite, providing remote sensing data with extremely high temporal, spatial, and spectral resolutions (Maes & Steppe, 2019;Manfreda et al., 2018). UAV remote sensing systems are able to collect crop growth information from different dimensions, including spectral information (from RGB (Nevavuori et al., 2019), or multispectral (MS) images (Wan et al., 2020)), thermal information (from thermal images (Khanal et al., 2017)), and structure information (from structure from motion (SfM) or LiDAR point clouds (Khosravipour et al., 2015)). By extracting agronomic traits (e.g., vegetation indices, VIs; plant height, PH) from these data (Bendig et al., 2014;Chlingaryan et al., 2018) and constructing the regression relationship between these traits and measured AGB, the crop AGB over a large scale could be well estimated (Casadesús & Villegas, 2014;Jin et al., 2020). ...
Article
Full-text available
Accurate estimation of above-ground biomass (AGB) plays a significant role in characterizing crop growth status. In precision agriculture area, a widely-used method for measuring AGB is to develop regression relationships between AGB and agronomic traits extracted from multi-source remotely sensed images based on unmanned aerial vehicle (UAV) systems. However, such approach requires expert knowledges and causes the information loss of raw images. The objectives of this study are to (i) determine how multi-source images contribute to AGB estimation in single and whole growth stages; (ii) evaluate the robustness and adaptability of deep convolutional neural networks (DCNN) and other machine learning algorithms regarding AGB estimation. To establish multi-source image datasets, this study collected UAV red-green-blue (RGB), multispectral (MS) images and constructed the raster data for crop surface models (CSMs). Agronomic features were derived from the above-mentioned images and interpreted by the multiple linear regression, random forest, and support vector machine models. Then, a DCNN model was developed via an image-fusion architecture. Results show that the DCNN model provides the best estimation of maize AGB when a single type of image is considered, while the performance of DCNN degrades when sufficient agronomic features are used. Besides, the information of above three image datasets changes with various growth stages. The structure information derived from CSM images are more valuable than spectrum information derived from RGB and MS images in the vegetative stage, but less useful in the reproductive stage. Finally, a data fusion strategy was proposed according to the onboard sensors (or cost).
... This finding is very attractive for the application in precision agriculture, since LAI and PH data can be easily collected by a UAV system with multispectral and digital cameras, which saves a large number of labor costs. Also, the monitoring of PH data on a large scale could be conducted by a Light Detection and Ranging (LiDAR) system onboard a helicopter (Khosravipour et al., 2015). Therefore, the potential of PL fs strategy for sugarcane yield estimation in a large scale should be explored further. ...
Article
Accurate crop growth simulations and yield estimation play a crucial role in agricultural development and food security. Incorporating multi-source observations into crop growth models could reduce the prediction uncertainty propagated from input data and model parameters. However, the value of different data sources varies. Incorporating redundant data into models not only increases computational cost, but also introduces additional prediction uncertainties. The objective of this study is to investigate the value of three common agronomy variables (plant height, PH; leaf area index, LAI; and soil moisture, SM) for sugarcane growth simulations and explore which variable(s) have the largest information content and hence should be included in data assimilation system. The measurements of PH, LAI and SM data are collected through a two-year sugarcane experiment (in 2016-2017) at Chongzuo station (Guangxi, China). Results show that the value of SM is the lowest among all three variables for sugarcane yield estimation if the spatial hetero-geneity of water and nutrient both exist. When sugarcane plots have relatively homogeneous cultivation density, it is preferable to incorporate PH data into the model. In contrast , assimilation of LAI might be more suitable when the cultivation density and tiller number contain strong spatial variability. Moreover, compared with traditional LAI & SM fusion strategy, the fusion of LAI and PH data is recommended to obtain more robust sug-arcane simulation results. Furthermore, observations during the elongation period provide the most valuable information for sugarcane growth simulation and yield estimation, while those in the emergence and tillering period are less informative.
... As with most point-cloud-based algorithms [28], [34]- [36], [50], our method applied normalized point cloud as input because it is easy to eliminate the interference of ground points and understory vegetation on individual tree segmentation, and further, this approach can directly obtain the absolute height trees. However, the topographic correction may change the actual shape of the crowns and affect the performance, as discussed in [26], [90], and [91]; this correction should be considered for application in some steep areas. In addition, our study applied the power function of the number of points in regions as the scaling function for the merging cost. ...
Article
Over an extended period, remote-sensing-based individual tree analysis has played a critical role in modern forest inventory and management research. The segmentation of individual trees from aerial point clouds usually depends on the characteristics of peak-like uplift on the crown surface; however, the performance inevitably decreases with increasing visibility of such features in point clouds, especially for high-density forests. Herein, we developed a novel hierarchical region-merging algorithm that first over-segmented the entire forest scene based on local density and then merged the over-segmented partitions into pairs through a stepwise optimal process to produce the final segmentation. In the region-merging method, a global merging cost was introduced to shift from local detection of crown features to utilize the overall compactness of forest point clouds. The experiments were conducted using unmanned aerial vehicle light detection and ranging (UAV-LiDAR) point clouds from three coniferous stands with different densities and a high-density coniferous and broad-leaved mixed stand. A total of 5510 field-measured trees in 36 plots were used to assess the accuracy of the proposed method. Our method achieved F-scores of 0.91, 0.88, 0.84 and 0.80 for low- (~700 stems/ha), medium- (~1000 stems/ha), and high-density (~2000 stems/ha) conifer stands and coniferous and broad-leaved mixed forests (~1800 stems/ha), respectively. Compared to the classical individual tree segmentation methods (marker-controlled watershed segmentation and point cloud region-growing algorithm), our method obtained comparable performance in low-density conifer stands and superior performance in the other stands. Furthermore, the region-merging algorithm could detect 10% more suppressed trees on average, which led to an apparent improvement in detection accuracy. The proposed algorithm provides a flexible segmentation framework that could be further improved by a different design that merges costs or applies multiscale segmentation with different stopping criteria.
... It seems to be a great choice to generate CHMs from point-cloud data and extracted tree height points at a slope ranging from 2 • to 18 • (Hyyppä and Inkinen, 1999;Eysn et al., 2012). However, some studies have demonstrated that as the slope increases, the treetop position displacement increases linearly with the crown radius in the CHM (Khosravipour et al., 2015;Nie et al., 2019). In order to minimize the impact of the pre-processing steps on tree height detection, we only generated the DSM, and hence achieved good results with an accuracy range of 0.22-0.69 ...
Article
Full-text available
Modern laser scanning techniques have been commonly used for forest land studies. Complementary scanning capabilities can be particularly obtained using unmanned aerial vehicle laser scanning (ULS) and terrestrial laser scanning (TLS) technologies. The registration of the ULS and TLS data can thus lead to more comprehensive data acquisition and information extraction in forest lands. However, this registration process is typically hindered by problems of data density and inconsistency of the scanned forest canopy shape. In this paper, we propose a tree-height registration (TR) method for ULS-TLS point-cloud registration, and apply this method to the alpine forest land of the Shangri-La City of the Northwestern Yunnan Province in China. The tree height points are obtained from a digital surface model (DSM), which contains isosurface points of the forest structures. Rotation and translation matrices are then calculated through singular value decomposition (SVD), and rough registration is completed. Finally, fine registration is achieved through nearest-neighbor iterative SVD. The results show that the proposed method can effectively register ULS and TLS forest data samples, with an average accuracy of 0.43 m. The method has an average running time of less than 21 s through 3m-by-3m window size to search tree height point for registration, and also shows good applicability in forest lands with slopes in the range of 2-18°. Moreover, the best outcomes were obtained for a 3m-by-3m window size.
... It is important to keep in mind that it is a raster-based method and allows the identification of trees that are in the upper canopy strata. For more complex forest environments such as areas with steep slopes or dense canopies, alternative algorithmssuch as TM or VF or deep learningand approaches or their combinations should be tested [19,[39][40][41][42][43][44]. Furthermore, it is worth noting that most of the functions used here have optional parameters that might be different depending on the forest structure and must be tested based on inferences drawn from previous studies [2]. ...
Article
Full-text available
Applications of unmanned aerial vehicles (UAVs) have proliferated in the last decade due to the technological advancements on various fronts such as structure-from-motion (SfM), machine learning, and robotics. An important preliminary step with regard to forest inventory and management is individual tree detection (ITD), which is required to calculate forest attributes such as stem volume, forest uniformity, and biomass estimation. However, users may find adopting the UAVs and algorithms for their specific projects challenging due to the plethora of information available. Herein, we provide a step-by-step tutorial for performing ITD using (i) low-cost UAV-derived imagery and (ii) UAV-based high-density lidar (light detection and ranging). Functions from open-source R packages were implemented to develop a canopy height model (CHM) and perform ITD utilizing the local maxima (LM) algorithm. ITD accuracy assessment statistics and validation were derived through manual visual interpretation from high-resolution imagery and field-data-based accuracy assessment. As the intended audience are beginners in remote sensing, we have adopted a very simple methodology and chosen study plots that have relatively open canopies to demonstrate our proposed approach; the respective R codes and sample plot data are available as supplementary materials.
... Its success relies on the correct identification of tree tops and a precise estimation of their heights [44,60,61]. Therefore, the method is sensitive to the quality of the DAP products, the efficiency of the identification method, and the physical properties of the area under analysis [62][63][64][65]. ...
Article
Full-text available
Digital aerial photogrammetry (DAP) data acquired by unmanned aerial vehicles (UAV) have been increasingly used for forest inventory and monitoring. In this study, we evaluated the potential of UAV photogrammetry data to detect individual trees, estimate their heights (ht), and monitor the initial silvicultural quality of a 1.5-year-old Eucalyptus sp. stand in northeastern Brazil. DAP estimates were compared with accurate tree locations obtained with real time kinematic (RTK) positioning and direct height measurements obtained in the field. In addition, we assessed the quality of a DAP-UAV digital terrain model (DTM) derived using an alternative ground classification approach and investigated its performance in the retrieval of individual tree attributes. The DTM built for the stand presented an RMSE of 0.099 m relative to the RTK measurements, showing no bias. The normalized 3D point cloud enabled the identification of over 95% of the stand trees and the estimation of their heights with an RMSE of 0.36 m (11%). However, ht was systematically underestimated, with a bias of 0.22 m (6.7%). A linear regression model, was fitted to estimate tree height from a maximum height metric derived from the point cloud reduced the RMSE by 20%. An assessment of uniformity indices calculated from both field and DAP heights showed no statistical difference. The results suggest that products derived from DAP-UAV may be used to generate accurate DTMs in young Eucalyptus sp. stands, detect individual trees, estimate ht, and determine stand uniformity with the same level of accuracy obtained in traditional forest inventories.
... The spike-free and GPMF differ fundamentally from the others; they used all laser returns to generate CHMs while improving the potential pits. Using all returns avoids the loss of crown information and eliminates some errors caused by using only the first return [7,45]. The difference between them is that the spike-free algorithm prevents spike formation during TIN construction [17], while GPMF excludes non-surface points from all returns in a progressive filtering process [27]. ...
Article
Full-text available
As a common form of light detection and ranging (LiDAR) in forestry applications, the canopy height model (CHM) provides the elevation distribution of aboveground vegetation. A CHM is traditionally generated by interpolating all the first LiDAR echoes. However, the first echo can-not accurately represent the canopy surface, and the resulting large amount of noise (data pits) also reduce the CHM quality. Although previous studies concentrate on many pit-filling meth-ods, the applicability of these methods in high-resolution unmanned aerial vehicle laser scan-ning (UAVLS)-derived CHMs has not been revealed. This study selected eight widely used, re-cently developed, representative pit-filling methods, namely first-echo interpolation, smooth filtering (mean, medium and Gaussian), highest point interpolation, pit-free algorithm, spike-free algorithm and graph-based progressive morphological filtering (GPMF). A compre-hensive evaluation framework was implemented, including a quantitative evaluation using simulation data and an additional application evaluation using UAVLS data. The results indi-cated that the spike-free algorithm and GPMF had excellent visual performances and were clos-est to the real canopy surface (root mean square error (RMSE) of simulated data were 0.1578 m and 0.1093 m, respectively; RMSE of UAVLS data were 0.3179 m and 0.4379 m, respectively). Compared with the first-echo method, the accuracies of the spike-free algorithm and GPMF im-proved by approximately 23% and 22%, respectively. The pit-free algorithm and highest point interpolation method also have advantages in high-resolution CHM generation. The global smooth filter method based on the first-echo CHM reduced the average canopy height by ap-proximately 7.73%. Coniferous forests require more pit-filling than broad-leaved forests and mixed forests. Although the results of individual tree applications indicated that there was no significant difference between these methods except the median filter method, pit-filling is still of great significance for generating high-resolution CHMs. This study provides guidance for us-ing high-resolution UAVLS in forestry applications.
... Finally, the area coverage of the polygons was proportioned to the total plot area. Further information regarding LiDAR-based canopy height models can be found in Khosravipour et al. [65] and Duncanson et al. [64]. ...
Article
Forest inventory (FI) surveys are cumbersome when field measurements are performed by manual means. We propose a semi-automated data collection approach using handheld mobile laser scanning (HMLS) to estimate and map key FI parameters. To this end, machine learning (e.g., random forest classifier for tree detection) and innovative algorithms (e.g., ellipse fitting for diameter estimation of noncircular trees) were used for the first time in FI surveying. After surveying nine plots, we compared HMLS-derived data against the field reference. HMLS-derived tree diameters (DBHs) were strongly correlated with the reference data at the single-tree level (r = 0.93-0.99; p < 0.001). At the plot level, HMLS slightly overestimated DBHs in complex plots due to the influence of undergrowth and creepers on trunks. Yet, no statistically significant difference was found between the two datasets (p > 0.05). Overall, HMLS was concluded as efficient and effective tool for FIs, even if used alone.
... The classical tree crown detection methods comprise of local maximum filter (Wulder et al., 2000;Pouliot et al., 2002;Vastaranta et al., 2012;Khosravipour et al., 2015;Wang et al., 2016;Li et al., 2019b), template matching (Ke and Quackenbush, 2011;Murray et al., 2019), image binarization (Pitkänen, 2001;Daliakopoulos et al., 2009) and image segmentation (Ferraz et al., 2016;Weinmann et al., 2016;Qin et al., 2014;Wagner et al., 2018;Aval et al., 2018), etc. Classical image processing methods are designed by the characteristic of morphology for tree crowns (Gomes et al., 2018;Campbell et al., 2020). In most cases, these methods can be completed using existing tools or software (e.g., eCognition and ArcGIS) without time-consuming and labor-exhausting ground truths collections (except for the template matching method) (Ardila et al., 2012;Zhang et al., 2014;Mongus & Ž alik, 2015). ...
Article
Full-text available
For both the positive economic benefit and the negative ecological impact of the rapid expansion of oil palm plantations in tropical developing countries, it is significant to achieve accurate detection for oil palm trees in large-scale areas. Especially, growing status observation and smart oil palm plantation management enabled by such accurate detections would improve plantation planning, oil palm yield, and reduce manpower and consumption of fertilizer. Although existing studies have already reached a high accuracy in oil palm tree detection, rare attention has been paid to automated observation of each single oil palm tree’s growing status. Nowadays, with its high spatial resolution and low cost, Unmanned Aerial Vehicle (UAV) has become a promising tool for monitoring the growing status of individual oil palms. However, the accuracy is still a challenging issue because of the extreme imbalance and high similarity between different classes. In this paper, we propose a Multi-class Oil PAlm Detection approach (MOPAD) to reap both accurate detection of oil palm trees and accurate monitoring of their growing status. Based on Faster RCNN, MOPAD combines a Refined Pyramid Feature (RPF) module and a hybrid class-balanced loss module to achieve satisfying observation of the growing status for individual oil palms. The former takes advantage of multi-level features to distinguish similar classes and detect small oil palms, and the latter effectively resolves the problem of extremely imbalanced samples. Moreover, we elaborately analyze the distribution of different kinds of oil palms, and propose a practical workflow for detecting oil palm vacancy. We evaluate MOPAD using three large-scale UAV images photographed in two sites in Indonesia (denoted by Site 1 and Site 2), containing 363,877 oil palms of five categories: healthy palms, dead palms, mismanaged palms, smallish palms and yellowish palms. Our proposed MOPAD achieves an F1-score of 87.91% (Site 1) and 99.04% (Site 2) for overall oil palm tree detection, and outperforms other state-of-the-art object detection methods by a remarkable margin of 10.37–17.09% and 8.14%-21.32% with respect to the average F1-score for multi-class oil palm detection in Site 1 and Site 2, respectively. Our method demonstrates excellent potential for individual oil palm tree detection and observation of growing status from UAV images, leading to more precise and efficient management of oil palm plantations.
... The CHM, also known as the normalized Digital Surface Model (DSM), can depict the canopy surface and is the base data for tree height and density assessment (Li et al. 2012;Forzieri et al. 2009;Tu et al. 2019Tu et al. , 2020. Two approaches for CHM generation are raster-based and by using point clouds directly (Khosravipour et al. 2015). The raster-based method for CHM creation is computed by subtracting the Digital Terrain Model (DTM) from a DSM (Lim et al. 2016), while the point clouds are used to classify the ground points and treetop points, and then calculate the height (Van Leeuwen et al. 2010;Khosravipour et al. 2014). ...
Article
Full-text available
Accurate, cost-effective monitoring and management of young forests is important for future stand quality. There is a critical need for a rapid assessment tool for forest monitoring and management. This study uses a low-cost unmanned aerial vehicle (UAV) to complete a tree height and tree density assessment in a newly forested Chinese fir (Cunning-hamia lanceolata (Lamb) Hook) planting (15 sample plots), Shunchang County, Fujian, China (1.12 ha). Images obtained from a Phantom4-Multispectral UAV were used to generate a digital surface model (DSM) with DJI Terra software (0.02 m spatial resolution). Based on the DSM, the individual trees were identified and the height of each corresponding tree was determined. The impacts of factors related to individual tree detection (ITD) and tree height accuracy were also analyzed. For the tree-level, the highest accuracy of ITD for Chinese fir was 98.93% (F-score = 98.93%). Remotely sensed individual tree heights produced an R 2 value of 0.89, RMSE value of 0.22 m when compared to a field survey. At the stand-level, tree height assessment yielded R 2 = 0.95, RMSE = 0.12 m, and tree density assessment yielded R 2 = 0.99, RMSE = 48 tree ha −1. The results highlight that UAVs can successfully monitor forest parameters and hold great potential as a supplement or substitute tool in field inventory.
... Special attention needs to be paid to the selection of filtering algorithms based on the characteristics of the study area. Normalization is designed to remove the influence of terrain elevation on lidar height measurements [77]. A normalized point cloud can be calculated by subtracting the terrain elevation from lidar points. ...
Article
Full-text available
The advent of lidar has revolutionized the way we observe and measure vegetation structure from the ground and from above and represents a major advance toward the quantification of 3D ecological observations. Developments in lidar hardware systems and data processing algorithms have greatly improved the accessibility and ease of use of lidar observations in ecological studies. A wide range of studies has been devoted to accurately measuring and modeling vegetation structural and functional attributes from lidar data across a range of spatial scales (from individual organs to global scales) and ecosystem types (e.g., forest, agricultural, grassland, and urban ecosystems).
... CHM can be derived as the difference between DSM and DEM which can be acquired by reconstruction after classifying and filtering point clouds (Wallace, Lucieer, and Watson 2014). There was also direct detection of individual tree based on the maximum value of DSM without deriving CHM (Khosravipour et al. 2015). Previous studies reported that the method based on CHM had only 40-70% or less than 35% individual tree detection accuracy for poor representativeness in point clouds and overstory occlusion (Nevalainen et al. 2017;Goldbergs et al. 2018). ...
Article
Full-text available
Low consumer-grade cameras attached to small unmanned aerial vehicles (UAV) can easily acquire high spatial resolution images, leading to convenient forest monitoring at small-scales for forest managers. However, most studies were carried out in the low canopy density and flat ground plantations to detect individual trees. We selected overlapping canopy plantation in mountainous area in the eastern of China and acquired high spatial resolution UAV RGB images to detect individual trees. A total of 402 reference trees were located in three rectangle plots (900 m²). To enhance the confidence of the tested individual tree detection method, clear-cutting and Real-Time Kinematic (RTK) were used to obtain the truth values in the plots. A novel method for semi-automatic individual tree detection was proposed based on a local-maximum algorithm and UAV-derived DSM data (LAD) in this study. The detection accuracy of LAD was compared with commonly used methods based on UAV-derived orthophoto images, local-maximum algorithm (LAO), object-oriented feature segmentation (OFS), multiscale segmentation technique (MST) and manual visual interpretation (MVI). The overall accuracy (OA (%) decreased in the order of LAD (84.5%) > MST (69.1%) > OFS (65.1%) > MVI (64.1%) > LAO (59.1%). LAD had only 15.5%s omission errors (OM (%), which was less than half of the other four methods in comparison. It was noteworthy that MVI had 35.9% OM %, which revealed that MVI should be used carefully as the truth value. LAD showed similar repeated detection error (RP (%) and completely wrong detection (CW (%), while the other four methods had obviously higher CW % than the RP %. From our results, it can be concluded that the proposed LAD method may help improving the accuracy of individual tree detection to an acceptable accuracy (>80%) in dense mountain forests, and has practical advantages in future research direction to assess tree attributes from UAV RGB image.
... Digital elevation models (DEMs) are an indispensable input variable for geomorphological, hydrological, and ecological models in the applications, such as landslide detection [1], forest inventory [2] and flood hazard assessment [3]. The data source of DEMs includes field surveys, topographical maps, and remote sensing techniques [4,5]. ...
Article
Full-text available
Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) inherently suffers from various errors. Many previous works employed Geoscience Laser Altimeter System onboard the Ice, Cloud, and land Elevation Satellite (ICESat/GLAS) data to assess and enhance SRTM DEM accuracy. Nevertheless, data coregistration between the two datasets was commonly neglected in their studies. In this paper, an automated and simple three dimensional (3D) coregistration method (3CM) was introduced to align the 3-arc-second SRTM (SRTM3) DEM and ICESat/GLAS data over Jiangxi province, China. Then, accuracy evaluation of the SRTM3 DEM using ICESat/GLAS data with and without data coregistration was performed on different classes of terrain factors and different land uses, with the purpose of evaluating the importance of data coregistration. Results show that after data coregistration, the root mean square error (RMSE) and mean bias of the SRTM3 DEM are reduced by 14.4% and 97.1%, respectively. Without data coregistration, terrain aspects with a sine-like shape are strongly related to SRTM3 DEM errors; nevertheless, this relationship disappears after data coregistration. Among the six land uses, SRTM3 DEM produces the lowest accuracy in forest areas. Finally, by incorporating land uses, terrain factors and ICESat/GLAS data into the correction models, the SRTM3 DEM was enhanced using multiple linear regression (MLR), back propagation neural network (BPNN), generalized regression NN (GRNN), and random forest (RF), respectively. Results exhibit that the four enhancement models with data coregistration obviously outperform themselves without the coregistration. Among the four models, RF produces the best result, and its RMSE is about 3.1%, 2.7% and 11.3% lower than those of MLR, BPNN, and GRNN, respectively. Moreover, 146 Global Navigation Satellite System (GNSS) points over Ganzhou city of Jiangxi province were used to assess the accuracy of the RF-derived SRTM3 DEM. It is found that the DEM quality is improved and has a similar error magnitude to that relative to the ICESat/GLASS data.
... The normalisation process implies a distortion of the point cloud and, therefore, of the sampled above-ground objects, such as trees and shrubs. Because this can be exacerbated in areas of high slope (Fig. 3), some authors have chosen to work with raw point-cloud to preserve the geometry of tree tops (Vega et al., 2014;Khosravipour et al., 2015;Alexander et al., 2018). In lidR, normalisation is easily reversible by switching absolute and relative height coordinates allowing versatile back and forth representations from raw to normalised point clouds if desired. ...
Article
Full-text available
Airborne laser scanning (ALS) is a remote sensing technology known for its applicability in natural resources management. By quantifying the three-dimensional structure of vegetation and underlying terrain using laser technology, ALS has been used extensively for enhancing geospatial knowledge in the fields of forestry and ecology. Structural descriptions of vegetation provide a means of estimating a range of ecologically pertinent attributes, such as height, volume, and above-ground biomass. The efficient processing of large, often technically complex datasets requires dedicated algorithms and software. The continued promise of ALS as a tool for improving ecological understanding is often dependent on user-created tools, methods, and approaches. Due to the proliferation of ALS among academic, governmental, and private-sector communities, paired with requirements to address a growing demand for open and accessible data, the ALS community is recognising the importance of free and open-source software (FOSS) and the importance of user-defined workflows. Herein, we describe the philosophy behind the development of the lidR package. Implemented in the R environment with a C/C++ backend, lidR is free, open-source and cross-platform software created to enable simple and creative processing workflows for forestry and ecology communities using ALS data. We review current algorithms used by the research community, and in doing so raise awareness of current successes and challenges associated with parameterisation and common implementation approaches. Through a detailed description of the package, we address the key considerations and the design philosophy that enables users to implement user-defined tools. We also discuss algorithm choices that make the package representative of the ‘state-of-the-art’ and we highlight some internal limitations through examples of processing time discrepancies. We conclude that the development of applications like lidR are of fundamental importance for developing transparent, flexible and open ALS tools to ensure not only reproducible workflows, but also to offer researchers the creative space required for the progress and development of the discipline.
Preprint
Full-text available
Recent years have seen a rapid surge in the use of Light Detection and Ranging (LiDAR) technology for characterizing the structure of ecosystems. Even though repeated airborne laser scanning (ALS) surveys are increasingly available across several European countries, only few studies have so far derived data products of ecosystem structure at a national scale, possibly due to a lack of free and open source tools and the computational challenges involved in handling the large volumes of data. Nevertheless, high-resolution data products of ecosystem structure generated from multi-temporal country-wide ALS datasets are urgently needed if we are to integrate such information into biodiversity and ecosystem science. By employing a recently developed, open-source, high-throughput workflow (named “Laserfarm”), we processed around 70 TB of raw point clouds collected from four national ALS surveys of the Netherlands (AHN1–AHN4, 1996–2022). This resulted in ~ 59 GB raster layers in GeoTIFF format as ready-to-use multi-temporal data products of ecosystem structure at a national extent. For each AHN dataset, we generated 25 LiDAR-derived vegetation metrics at 10 m spatial resolution, representing vegetation height, vegetation cover, and vegetation structural variability. The data enable an in-depth understanding of ecosystem structure at fine resolution across the Netherlands and provide opportunities for exploring ecosystem structural dynamics over time. To illustrate the utility of these data products, we present ecological use cases that monitor forest structural change and analyse vegetation structure differences across various Natura 2000 habitat types, including dunes, marshes, grasslands, shrublands, and woodlands. The provided data products and the employed workflow can facilitate a wide use and uptake of ecosystem structure information in biodiversity and carbon modelling, conservation science, and ecosystem management. The full data products and source code are publicly available on Zenodo (https://doi.org/10.5281/zenodo.13940846) (Shi and Kissling 2024).
Article
Accurate quantification and mapping of forest productivity are critical to understanding and managing forest ecosystems. Local LiDAR or photogrammetric surveys have been used to obtain reliable estimates of canopy heights, yet these acquisitions can entail substantial expenses. Therefore, we developed models using freely available US Geological survey (USGS) LiDAR data for prediction of dominant height to map site index across loblolly pine (Pinus taeda L.) plantations in the southeastern US. We used 2017–2020 national USGS 3D Elevation Program LiDAR acquisitions and explored how different height percentiles, grid output resolutions, time difference between LiDAR and ground acquisitions, tree height, and dominant height definition affected the proposed model. We built the dominant height models using 1301 ground plots. The final regression model was constructed with the 95th percentile of the height distribution of the first returns above-ground and had values of R2 = 0.89, RMSE = 1.55 m, and RRMSE = 7.66 per cent at a 20-m pixel grid, yet all the examined percentile-resolution combinations were acceptable. No effect evidence was found for time difference when the flight was less than 4 months in advance or after the ground measurement, and it was also found independent of pulse density when this variable was lower than 9.5 pulses m−2. Using the recorded age of the plantations, we assessed the error propagation when translating dominant height to site index in two site index models, obtaining an RRMSE lower than 10 per cent in both. We found that USGS LiDAR acquisitions can be reliably used to map dominant height at a large scale, and consequently used to map forest productivity when age is known. This ability adds more value to a tool proven widely applicable in time and space and offers a great opportunity for stakeholders in different fields of use.
Article
Full-text available
Digital elevation models (DEMs) are used for many geosciences studies; hence, their accuracy is essential. Throughout the world, there are many small islands of various sizes and densities; hence, it is important to assess the DEM accuracy on very small islands since DEMs serve as the major data source for many investigations, particularly in geomorphology, land-use planning, and disaster management. Therefore, this paper aims to validate the accuracy of an open-source Indonesian DEM (DEMNAS) in the very small islands of Karimunjawa–Indonesia. Validation was conducted by comparing elevation values from DEMNAS to the true elevation values in four very small islands in Karimunjawa, namely Cemara Besar, Cemara Kecil, Menjangan Besar, and Menjangan Kecil. The true elevation values were obtained by orthorectification of aerial imagery using a DJI Mavic Air-2 Unmanned Aerial Vehicle (UAV). The orthorectification came from ground control points (GCP) from the geodetic Global Positioning System (GPS). In the study area, fourteen GCP were erected; for more significant coverage, they were placed along the edges of the very small islands. After that, Agisoft software analyzed the images to produce a DEM using GCP orthorectification. Based on 280 sampling points, we applied a root-mean-square error (RMSE) to calculate elevation errors, and we performed the linear error 90% (LE90) calculation to judge the average errors with the 90% threshold of absolute values of discrepancies. The DEMNAS RMSE and LE90 calculation results in the Karimunjawa archipelago were 6.33 m and 10.45 m, respectively. Citing Regulation Number 15 of the Head of the Indonesian Geospatial Information Agency of 2014 concerning Technical Guidelines for Basic Map Accuracy, DEMNAS with 10.45 m LE90 can be utilized for producing geomorphological maps with scales of 1:25,000 or smaller. However, detailed geomorphological mapping of a very small island (less than 100 km2) needs better DEM data that is usually produced using aerial photogrammetry. Using UAVs for DEMs creation may benefit small island developing states (SIDS) worldwide.
Article
Canopy height is an important crop biophysical parameter. It provides information about the crop growth as well as act as an input parameter for biomass and crop yield models. Considering the importance of this parameter, a novel semi-automatic canopy height estimation model has been developed which can work with both georeferenced or non-georeferenced top-of-canopy aerial images. The model employs a Structure-from-Motion algorithm followed by dense point cloud reconstruction and polygon triangulation to obtain polygon meshes which are used for height estimation. The process has been tested on drone-based data collected from a maize crop over the 2018-19 Rabi season from a semi-arid area in central-south India. The ground truth canopy height was measured by manually measuring height of plants using a meter scale. The ground elevation has been modelled using a linear best fit plane and the estimated canopy height was found to have the best R2 value of 0.85 and RMSE values of 14.17 cm.
Article
The forest land carbon storage assessment model proposed in this study belongs to the IPPC Tier-3 method. The model operates through a conceptual model that includes environmental parameters such as Photosynthesis Index (NDVI), Soil Depth Index (TWI) and Canopy Height Model (CHM). The model is verified by the IPPC Tier-2 method, which can effectively construct the distribution of watershed-scale forest carbon stock. The spatial distribution of carbon fluxes can be obtained from the carbon stock information of the two periods. The carbon flux index can be used to describe the distribution of carbon stocks and carbon emissions in the watershed, and then the location of negative carbon flux (vegetation degradation) can be extracted as a value-added application of vegetation restoration hotspots. The accuracy of the model estimation in forest land verification is high, reaching a significant level of 0.1%, which can faithfully reflect the carbon stock of forest land. However, in non-forest lands, carbon stock estimates tend to exhibit carbon storage potential due to land-use disturbances. A combination of conceptual models and environmental indicators is of great help in assessing watershed carbon stocks and fluxes.
Article
Full-text available
Unmanned aerial vehicles (UAVs) are becoming essential tools for surveying and monitoring forest ecosystems. However, most forests are found on steep slopes, where capturing individual tree characteristics might be compromised by the difference in ground sampling distance (GSD) between slopes. Thus, we tested the performance of treetop detection using two algorithms on canopy height models (CHMs) obtained with a commercial UAV (Mavic 2 Pro) using the terrain awareness function (TAF). The area surveyed was on a steep slope covered predominantly by fir (Abies mariesii) trees, where the UAV was flown following (TAF) and not following the terrain (NTAF). Results showed that when the TAF was used, fir trees were clearly delimited, with lower branches clearly visible in the orthomosaic, regardless of the slope position. As a result, the dense point clouds (DPCs) were denser and more homogenously distributed along the slope when using TAF than when using NTAF. Two algorithms were applied for treetop detection: (connected components), and (morphological operators). (connected components) showed a 5% improvement in treetop detection accuracy when using TAF (86.55%), in comparison to NTAF (81.55%), at the minimum matching error of 1 m. In contrast, when using (morphological operators), treetop detection accuracy reached 76.23% when using TAF and 62.06% when using NTAF. Thus, for treetop detection alone, NTAF can be sufficient when using sophisticated algorithms. However, NTAF showed a higher number of repeated points, leading to an overestimation of detected treetop.
Article
Full-text available
Detecting and mapping individual trees accurately and automatically from remote sensing images is of great significance for precision forest management. Many algorithms, including classical methods and deep learning techniques, have been developed and applied for tree crown detection from remote sensing images. However, few studies have evaluated the accuracy of different in-dividual tree detection (ITD) algorithms and their data and processing requirements. This study explored the accuracy of ITD using local maxima (LM) algorithm, marker-controlled watershed segmentation (MCWS), and Mask Region-based Convolutional Neural Networks (Mask R-CNN) in a young plantation forest with different test images. Manually delineated tree crowns from UAV imagery were used for accuracy assessment of the three methods, followed by an evaluation of the data processing and application requirements for three methods to detect individual trees. Overall, Mask R-CNN can best use the information in multi-band input images for detecting individual trees. The results showed that the Mask R-CNN model with the multi-band combination produced higher accuracy than the model with a single-band image, and the RGB band combination achieved the highest accuracy for ITD (F1 score = 94.68%). Moreover, the Mask R-CNN models with multi-band images are capable of providing higher accuracies for ITD than the LM and MCWS algorithms. The LM algorithm and MCWS algorithm also achieved promising accuracies for ITD when the canopy height model (CHM) was used as the test image (F1 score = 87.86% for LM algorithm, F1 score = 85.92% for MCWS algorithm). The LM and MCWS algorithms are easy to use and lower computer computational requirements, but they are unable to identify tree species and are limited by algorithm parameters, which need to be adjusted for each classification. It is highlighted that the application of deep learning with its end-to-end-learning approach is very efficient and capable of deriving the information from multi-layer images, but an additional training set is needed for model training, robust computer resources are required, and a large number of accurate training samples are necessary. This study provides valuable information for forestry practitioners to select an optimal approach for detecting individual trees.
Article
The spatial distribution of soil depth is the key information of watershed planning and design, and also one of the factors that determine the classification of slope utilization. The depth of soil on hillside requires on-site survey, which is labor-intensive. In addition, there are real challenges in updating or supplementing data because surveyed data may be altered by subsequent artificial or natural forces. Based on the concept of risk assessment, this study calculated the external forces (external factors) and internal forces (location factors) on the topsoil’s surface, and analyzed the soil formation relevant environmental indicators and then established a potential soil depth estimating model for the study area. Besides terrain and vegetation factors which are proved to be significant factors concerning soil genesis, plant height was also used to estimate soil depth because it is proportional to the depth of the underground root system and depth of root system is generally positively related to soil depth. Therefore, the information of Canopy Height Model (CHM) to the model can increase the model interpretation rate by 15%. The depth index value (SDI) is used to estimate soil depth by using the conversion equation (Dp = 706.5 × SDI - 43.812, R² = 0.76, p = 0.024). The extracted environmental indicators can dynamically and accurately estimate the spatial distribution of soil depth. The correlation between soil depth and slope gradient was also studied. The results can provide reference for relevant departments in the fields of watershed management, tending and afforestation, and resource conservation.
Article
Full-text available
Although the random forest algorithm has been widely applied to remotely sensed data to predict characteristics of forests, such as tree canopy height, the effect of spatial non-stationarity in the modeling process is oftentimes neglected. Previous studies have proposed methods to address the spatial variance at local scales, but few have explored the spatial autocorrelation pattern of residuals in modeling tree canopy height or investigated the relationship between canopy height and model performance. By combining Light Detection and Ranging (LiDAR) and Landsat datasets, we used spatially-weighted geographical random forest (GRF) and traditional random forest (TRF) methods to predict tree canopy height in a mixed dry forest woodland in complex mountainous terrain. Comparisons between TRF and GRF models show that the latter can lower predefined extreme residuals, and thus make the model performance relatively stronger. Moreover, the relationship between model performance and degree of variation of true canopy height can vary considerably within different height quantiles. Both models are likely to present underestimates and overestimates when the corresponding tree canopy heights are high (>95% quantile) and low (<median), respectively. This study provides a critical insight into the relationship between tree canopy height and predictive abilities of random forest models when taking account of spatial non-stationarity. Conclusions indicate that a trade-off approach based on the actual need of project should be taken when selecting an optimal model integrating both local and global effects in modeling attributes such as canopy height from remotely sensed data.
Article
Full-text available
To deal with the problem of spectral variability in high resolution satellite images, this paper focuses on the analysis and modelling of spatial autocorrelation feature. The semivariograms are used to model spatial variability of typical object classes while Getis statistic is used for the analysis of local spatial autocorrelation within the neighbourhood window determined by the range information of the semivariograms. Two segmentation experiments are conducted via the Fuzzy C-Means (FCM) algorithm which incorporates both spatial autocorrelation features and spectral features, and the experimental results show that spatial autocorrelation features can effectively improve the segmentation quality of high resolution satellite images.
Article
Full-text available
The cost of forest sampling can be reduced substantially by the ability to estimate forest and tree parameters directly from aerial photographs. However, in order to do so it is necessary to be able to accurately identify individual treetops and then to define the region in the vicinity of the treetop that encompasses the crown extent. These two steps commonly have been treated independently. In this paper, we derive individual tree-crown boundaries and treetop locations under a unified framework. We applied a two-stage approach with edge detection followed by marker-controlled watershed segmentation. A Laplacian of Gaussian edge detection method at the smallest effective scale was employed to mask out the background. An eight-connectivity scheme was used to label the remaining tree objects in the edge map. Subsequently, treetops are modeled based on both radiometry and geometry. More specifically, treetops are assumed to be represented by local radiation maxima and also to be located near the center of the tree-crown. As a result, a marker image was created from the derived treetop to guide a watershed segmentation to further differentiate touching and clumping trees and to produce a segmented image comprised of individual tree crowns. Our methods were developed on a 256- by 256-pixel CASI image of a commercially thinned trial forest. A promising agreement between our automatic methods and manual delineation results was achieved in counting the number of trees as well as in delineating tree crowns.
Conference Paper
Full-text available
Airborne laser scanning (ALS) data has revolutionized the landslide assessment in a rugged vegetated terrain. It enables the parameterization of morphology and vegetation of the instability slopes. Vegetation characteristics are by far less investigated because of the currently available accuracy and density ALS data and paucity of field data validation. We utilized a high density ALS (HDALS) data with 170 points m-2 for characterizing disrupted vegetation induced by landslides by means of a variable window filter and the SkelTre-skeletonisation. Tree analyses in landslide areas resulted in relatively low height, small crown and more irregularities, whereas these peculiarities are not so obvious in the healthy forests. The statistical tests unveiled the clear differences between the extracted parameters in landslide and non-landslide zones and supported the field evidences. We concluded that HDALS is a promising tool to geometrically retrieve disrupted woody vegetation structures and can be good bioindicator to landslide activity.
Article
Full-text available
Lidar technology has reached a point where ground and forest canopy elevation models can be produced at high spatial resolution. Individual tree crown isolation and classification methods are developing rapidly for multispectral imagery. Analysis of multispectral imagery, however, does not readily provide tree height information and lidar data alone cannot provide species and health attributes. The combination of lidar and multispectral data at the individual tree level may provide a very useful forest inventory tool. A valley following approach to individual tree isolation was applied to both high resolution digital frame camera imagery and a canopy height model (CHM) created from high-density lidar data over a test site of even aged (55 years old) Douglas-fir plots of varying densities (300, 500, and 725 stems/ha) on the west coast of Canada. Tree height was determined from the laser data within the automated crown delineations. Automated tree isolations of the multispectral imagery achieved 80%-90% good correspondence with the ground reference tree delineations based on ground data. However, for the more open plot there were serious commission errors (false trees isolated) mostly related to sunlit ground vegetation. These were successfully reduced by applying a height filter to the isolations based on the lidar data. Isolations from the lidar data produced good isolations with few commission errors but poorer crown outline delineations especially for the densest plot. There is a complimentarity in the two data sources that will help in tree isolation. Heights of the automated isolations were consistently underestimated versus ground reference trees with an average error of 1.3 m. Further work is needed to test and develop tools and capabilities, but there is an effective synergy of the two high resolution data sources for providing needed forest inventory information.
Article
Full-text available
Airborne laser scanning data and corresponding field data were acquired from boreal forests in Norway and Sweden, coniferous and broadleaved forests in Germany and tropical pulpwood plantations in Brazil. Treetop positions were extracted using six different algorithms developed in Finland, Germany, Norway and Sweden, and the accuracy of tree detection and height estimation was assessed. Furthermore, the weaknesses and strengths of the methods under different types of forest were analyzed. The results showed that forest structure strongly affected the performance of all algorithms. Particularly, the success of tree detection was found to be dependent on tree density and clustering. The differences in performance between methods were more pronounced for tree detection than for height estimation. The algorithms showed a slightly better performance in the conditions for which they were developed, while some could be adapted by different parameterization according to training with local data. The results of this study may help guiding the choice of method under different forest types and may be of great value for future refinement of the single-tree detection algorithms.
Article
Full-text available
We describe crown-extraction (CE) filtering to accurately determine tree apex positions for various coniferous species using an airborne light detection and ranging–derived digital canopy height model (DCHM). This method uses a square mask, with a frame at the edges, that overlaps pixels within the DCHM image; when no pixels touch the frame, the pixel at the center is extracted as a tree-crown pixel. The apex of each tree is determined by choosing the pixel with maximum height from the pixels in the crown. We compared the performance of this method and of two other methods (local-maximum filtering and canopy-segmentation method) for several species. The CE filtering had the most accurate results for most tree species with appropriate mask size selection. The mean omission, commission, and total errors for all tree species were 8.1%, 1.6%, and 9.7%, respectively, for CE filtering. Comparing mask sizes and canopy diameters estimated from the DCHM for each species revealed that the smallest canopy diameter of each species was close to the most appropriate mask size for that species in CE filtering. We also confirmed that the smoothing process used in the DCHM has little effect on the accuracy of CE filtering.
Article
Full-text available
Individual tree detection algorithms can provide accurate measurements of individual tree locations, crown diameters (from aerial photography and light detection and ranging (lidar) data), and tree heights (from lidar data). However, to be useful for forest management goals relating to timber harvest, carbon accounting, and ecological processes, there is a need to assess the performance of these image-based tree detection algorithms across a full range of canopy structure conditions. We evaluated the performance of two fundamentally different automated tree detection and measurement algorithms (spatial wavelet analysis (SWA) and variable window filters (VWF)) across a full range of canopy conditions in a mixed-species, structurally diverse conifer forest in northern Idaho, USA. Each algorithm performed well in low canopy cover conditions (50%) conditions. The results presented herein suggest that future algorithm development is required to improve individual tree detection in structurally complex forests. Furthermore, tree detection algorithms such as SWA and VWF may produce more accurate results when used in conjunction with higher density lidar data.
Article
Full-text available
The objective was to investigate the error sources of the airborne laser scanning based individual tree detection (ITD), and its effects on forest management planning calculations. The investigated error sources were detection of trees (e(td)), error in tree height prediction (e(h)) and error in tree diameter prediction (e(d)). The effects of errors were analyzed with Monte Carlo simulations. e(td) was modeled empirically based on a tree's relative size. A total of five different tree detection scenarios were tested. Effect of e(h) was investigated using 5% and 0% and effect of e(d) using 20%, 15%, 10%, 5%, 0% error levels, respectively. The research material comprised 15 forest stands located in Southern Finland. Measurements of 5,300 trees and their timber assortments were utilized as a starting point for the Monte Carlo simulated ITD inventories. ITD carried out for the same study area provided a starting point (Scenario 1) for e(td). In Scenario 1, 60.2% from stem number and 75.9% from total volume (V-total) were detected. When the only error source was e(td) (tree detection varying from 75.9% to 100% of V-total), root mean square errors (RMSEs) in stand characteristics ranged between the scenarios from 32.4% to 0.6%, 29.0% to 0.5%, 7.8% to 0.2% and 5.4% to 0.1% in stand basal area (BA), V-total, mean height (Hg) and mean diameter (Dg), respectively. Saw wood volume RMSE varied from 25.1% to 0.2%, as pulp wood volume respective varied from 37.8% to 1.0% when errors stemmed only from e(td). The effect of e(d) was most significant for V-total and BA and the decrease in RMSE was from 12.0% to 0.6% (BA) and from 10.9% to 0.5% (V-total) in the most accurate tree detection scenario when e(d) varied from 20% to 0%. The effect of increased accuracy in tree height prediction was minor for all the stand characteristics. The results show that the most important error source in ITD is tree detection. At stand level, unbiased predictions for tree height and diameter are enough, given the present tree detection accuracy.
Article
Full-text available
This study examined the effect of local competition, as determined by the number, size, and spatial distribution of neighboring trees, on crown base height, crown width, and crown asymmetry of trees growing in a natural mature Scotspine (Pinus sylvestris L.) forest in eastern Finland. Local competition was described with competition indices. Crown asymmetry was characterized as the distance and direction from the stem center to the visually estimated crown mass center, Nonspatial and spatial models for crown base height and crown width were derived, as well as a spatial model for the horizontal distance from the crown mass center to stem base position. The effect of neighbors on the direction of crown mass center from stem base position was studied using the center of competition and the widest free angle free of competitors closer than 5 m. The derived spatial models for crown base height and crown width, taking local competition into account, were only slightly better than the nonspatial models. This was because diameter at breast height and tree height, the independent variables used in the nonspatial models, described quite well the conditions of local competition for a tree. The spatial model for crown base height was 2.22% units (adjusted R-2) better than the nonspatial model. Correspondingly, the spatial model for crown width was 1.77% units better than the nonspatial model. Crown asymmetry characteristics were measured fur 57% of the: trees. The spatial model for the distance to the crown mass center could only explain about 10% of the total variation of the dependent variable. The model for the direction of the crown mass center from stem base position showed that the variables describing direction of local competition, i.e., the competition center and the widest free angle, could not adequately explain the direction of asymmetry. One reason for this was that tree crowns showed plastic responses to the prevailing asymmetry of solar radiation geometry. Therefore, the directional distribution of the crown asymmetry was strongly weighted toward southern and southwestern directions, i.e., the directions of most abundant incoming solar radiation. We suggest that the detected developmental plasticity in crown asymmetry was affected both by the directionality or solar radiation geometry and by that of local competition, which is influenced by the number, size, and spatial distribution of close competitors.
Article
Full-text available
This study compared the use of light detection and ranging (lidar) data and X‐band interferometric synthetic aperture radar (InSAR) data for estimating Lorey's height (HL) within inventory plots. At plot level, HL can be estimated with a root mean squared error (RMSE) of 1.8 m or 6.0% using lidar data and 2.7 m or 9.0% using InSAR data. The most effective predictor variables were found to be the 75th percentile of the vegetation heights for lidar (p75lidar) and the 90th percentile for InSAR (p90InSAR). Estimation results can be improved considerably by incorporating ground slope into the models. An increase in slope, given the same field‐measured HL, was associated with an upward shift in height percentiles. We assume that this is a cause of the asymmetrical crown shape on slopes. An interaction between slope and aspect was noted for InSAR data, with a tendency for the influence of slope to be stronger on upward (i.e. for slopes facing the InSAR sensor) than downward slopes. Comparison of the lidar‐ and InSAR‐derived vegetation heights revealed that the p90InSAR is more commonly below the p75lidar. This difference was greater on upward than downward slopes. Estimations of HL can be further improved if additional information, such as the conifer proportion (CP) (as a surrogate for crown shape) or stem density (SD), is included in the models. It was possible to substitute the predictor variable CP by means of density metrics in the statistical models that depend on InSAR data. However, this was not possible for the statistical models depending on lidar data. Density metrics were also not able to explain the variability that was explained by the predictor variable SD (this was the case for models that depend on lidar and on InSAR data).
Article
Full-text available
Crown structure for 9- to 30-year-old loblolly pine (Pinustaeda L.) trees in unthinned plantations was quantified via analysis of branch diameters and location, both along and around the bole, using data from 68 trees. The trees ranged in size from 11.1 to 31.6 cm in diameter at breast height and from 8.30 to 25.67 m in total height, and were growing in Piedmont and Atlantic Coastal Plain stands. Three equations were used to describe the diameter distribution of branches. Circular statistics were used to examine branching patterns around the bole. A system of two equations was developed to predict the total number of branches within a crown. A series of three equations was used to describe the mean of and range in diameters within a whorl. Attempts at modeling the height above ground to branches (whorls) were unsuccessful; therefore, equal height growth per year was assumed and whorls were placed within each year of growth in concordance with the literature. Similarly, predicting the number of branches within a whorl of a certain height was difficult, and overall percentages were employed. Analysis of branch azimuths on a whole-tree basis indicated that a uniform, or regular, distribution was appropriate (and not a "circular normal" distribution). Finally, circular correlation was used to analyze rotational patterns within and between whorls, and a strong positive correlation was found for consecutive whorls of the same number of branches. From this study it was concluded that modeling crown structure is difficult, with much variation occurring among trees.
Article
Full-text available
Light Detection and Ranging (lidar) has been widely applied to characterize the 3-dimensional (3D) structure of forests as it can generate 3D point data with high spatial resolution and accuracy. Individual tree segmentations, usually derived from the canopy height model, are used to derive individual tree structural attributes such as tree height, crown diame-ter, canopy-based height, and others. In this study, we develop a new algorithm to segment individual trees from the small footprint discrete return airborne lidar point cloud. We experimentally applied the new algorithm to segment trees in a mixed conifer forest in the Sierra Nevada Moun-tains in California. The results were evaluated in terms of recall, precision, and F-score, and show that the algorithm detected 86 percent of the trees ("recall"), 94 percent of the segmented trees were correct ("precision"), and the overall F-score is 0.9. Our results indicate that the proposed algorithm has good potential in segmenting individual trees in mixed conifer stands of similar structure using small footprint, discrete return lidar data.
Article
Full-text available
We present two improvements for laser-based forest inventory. The first improvement is based on using last pulse data for tree detection. When trees overlap, the surface model between the trees corresponding to the first pulse stays high, whereas the corresponding model from the last pulse results in a drop in elevation, due to its better penetration between the trees. This drop in elevation can be used for separating trees. In a test carried out in Evo, Southern Finland, we used 292 forests plots consisting of more than 5,500 trees and airborne laser scanning (ALS) data comprised of 12.7 emitted laser pulses per m(2). With last pulse data, an improvement of 6% for individual tree detection was obtained when compared to using first pulse data. The improvement increased with an increasing number of stems per plot and with decreasing diameter breast height (DBH). The results confirm that there is also substantial information for tree detection in last pulse data. The second improvement is based on the use of individual tree-based features in addition to the statistical point height metrics in area-based prediction of forest variables. The commonly-used ALS point height metrics and individual tree-based features were fused into the non-parametric estimation of forest variables. By using only four individual tree-based features, stem volume estimation improved when compared to the use of statistical point height metrics. For DBH estimation, the point height metrics and individual tree-based features complemented each other. Predictions were validated at plot level.
Article
Full-text available
Light detection and ranging (LiDAR) technology provides horizontal and vertical information at high spatial resolutions and vertical accuracies. Forest attributes such as canopy height can be directly retrieved from LiDAR data. Direct retrieval of canopy height provides opportunities to model above-ground biomass and canopy volume. Access to the vertical nature of forest ecosystems also offers new opportunities for enhanced forest monitoring, management and planning.
Conference Paper
Full-text available
Disrupted vegetation is often used as an indicator for landslide activity in forested regions. The extraction of irregular trees from airborne laser scanning data remains problematic because of low quality of observed data and paucity of field data validation. We obtained high density airborne LiDAR (HDAL) data with 180 points m -2 for characterizing tree growth anomalies caused by landslides. HDAL allowed the mapping of a complex landslide and its three kinematic zones. The TreeVaw method delineated trees from the HDAL data and determined their position and height, while the SkelTre-skeletonization method extracted the tree inclination. Tree detection accuracies were measured up to 88.2% in the transport zone with tree height (r 2 =0.77). The coefficients of determination for tree inclination were 0.82, 0.60, and 0.83 for source, transport, and deposition zones, respectively. We found the distribution of LiDAR-derived tree growth anomalies was statistically different for landslide areas as compared to stable areas.
Article
Full-text available
One problem of individual tree detection on aerial images or on raster canopy height models is handling of tree crowns of different sizes. On laser scanner data one size attribute, height, is directly available. This gives possibilities to develop processing methods that adapt to the object size. In this study, three adaptive methods were developed and tested for individual tree detection on canopy height model (CHM). The CHM of 0.5 m pixel size was computed from dense first-pulse point data that was acquired with a small-footprint airborne laser scanner. The field data consisted of 10 tree mapped field plots in Kalkkinen, southern Finland. The plots were mainly on mature, heavily stocked forest stands, many of which had multi-layered canopy structure. In the first method, the CHM was smoothed with canopy height based selection of degree of smoothing and local maxima on the smoothed CHM were considered as tree locations. In the second and third methods, we utilised crown diameter predicted from tree height. The second method used elimination of candidate tree locations based on the predicted crown diameter and distance and valley depth between two locations studied. The third method was modified from scale-space method used for blob detection. Instead of automatic scale selection of the scale-space method, the scale for Laplacian filtering, used in blob detection, was determined according to the predicted crown diameter. The presented three methods are compared based on the accuracy of individual tree detection. Differences of the proposed methods considering tree crown delineation by segmentation methods are discussed.
Article
Full-text available
The objective of the "Tree Extraction" project organized by EuroSDR (European Spatial data Research) and ISPRS (International Society of Photogrammetry and Remote Sensing) was to evaluate the quality, accuracy, and feasibility of automatic tree extraction methods, mainly based on laser scanner data. In the final report of the project, Kaartinen and Hyyppä (2008) reported a high variation in the quality of the published methods under boreal forest conditions and with varying laser point densities. This paper summarizes the findings beyond the final report after analyzing the results obtained in different tree height classes. Omission/Commission statistics as well as neighborhood relations are taken into account. Additionally, four automatic tree detection and extraction techniques were added to the test. Several methods in this experiment were superior to manual processing in the dominant, co-dominant and suppressed tree storeys. In general, as expected, the taller the tree, the better the location accuracy. The accuracy of tree height, after removing gross errors, was better than 0.5 m in all tree height classes with the best methods investigated in this experiment. For forest inventory, minimum curvature-based tree detection accompanied by point cloud-based cluster detection for suppressed trees is a solution that deserves attention in the future.
Article
Full-text available
The main study objective was to develop robust processing and analysis techniques to facilitate the use of small-footprint lidar data for estimating plot-level tree height by measuring individual trees identifiable on the three-dimensional lidar surface. Lidar processing techniques included data fusion with multispectral optical data and local filtering with both square and circular windows of variable size. The lidar system used for this study produced an average footprint of 0.65 m and an average distance between laser shots of 0.7 m. The lidar data set was acquired over deciduous and coniferous stands with settings typical of the southeastern United States. The lidar-derived tree measurements were used with regression models and cross-validation to estimate tree height on 0.017-ha plots. For the pine plots, lidar measurements explained 97 percent of the variance associated with the mean height of dominant trees. For deciduous plots, regression models explained 79 percent of the mean height variance for dominant trees. Filtering for local maximum with circular windows gave better fitting models for pines, while for deciduous trees, filtering with square windows provided a slightly better model fit. Using lidar and optical data fusion to differentiate between forest types provided better results for estimating average plot height for pines. Estimating tree height for deciduous plots gave superior results without calibrating the search window size based on forest type.
Article
Full-text available
Forest structure data derived from lidar is being used in forest science and management for inventory analysis, biomass estimation, and wildlife habitat analysis. Regression analysis dominated previous approaches to the derivation of tree stem and crown parameters from lidar. The regression model for tree parameters is locally applied based on vertical lidar point density, the tree species involved, and stand structure in the specific research area. The results of this approach, therefore, are location-specific, limiting its applicability to other areas. For a more widely applicable approach to derive tree parameters, we developed an innovative method called ‘wrapped surface reconstruction’ that employs radial basis functions and an isosurface. Utilizing computer graphics, we capture the exact shape of an irregular tree crown of various tree species based on the lidar point cloud and visualize their exact crown formation in three-dimensional space. To validate the tree parameters given by our wrapped surface approach, survey-grade equipment (a total station) was used to measure the crown shape. Four vantage points were established for each of 55 trees to capture whole-tree crown profiles georeferenced with post-processed differential GPS points. The observed tree profiles were linearly interpolated to estimate crown volume. These fieldwork-generated profiles were compared with the wrapped surface to assess goodness of fit. For coniferous trees, the following tree crown parameters derived by the wrapped surface method were highly correlated (p < 0.05) with the total station-derived measurements: tree height (R2 = 0.95), crown width (R2 = 0.80), live crown base (R2 = 0.92), height of the lowest branch (R2 = 0.72), and crown volume (R2 = 0.84). For deciduous trees, wrapped surface-derived parameters of tree height (R2 = 0.96), crown width (R2 = 0.75), live crown base (R2 = 0.53), height of the lowest branch (R2 = 0.51), and crown volume (R2 = 0.89) were correlated with the total station-derived measurements. The wrapped surface technique is less susceptible to errors in estimation of tree parameters because of exact interpolation using the radial basis functions. The effect of diminished energy return causes the low correlation for lowest branches in deciduous trees (R2 = 0.51), even though leaf-off lidar data was used. The wrapped surface provides fast and automated detection of micro-scale tree parameters for specific applications in areas such as tree physiology, fire modeling, and forest inventory.
Conference Paper
Full-text available
It is dicult to generate raster Digital Elevation Models (DEMs) from terrain mass point data sets too large to fit into mem- ory, such as those obtained by LIDAR. We describe prototype tools for streaming DEM generation that use memory and disk I/O very e- ciently. From 500 million bare-earth LIDAR double precision points (11.2 GB) our tool can, in just over an hour on a standard laptop with two hard drives, produce a 50,394 ◊ 30,500 raster DEM with 20 foot post spacing in 16 bit binary BIL format (3 GB), using less than 100 MB of main memory and less than 300 MB of temporary disk space.
Article
Full-text available
We show how to greatly accelerate algorithms that compute Delaunay triangulations of huge, well-distributed point sets in 2D and 3D by exploiting the natural spatial coherence in a stream of points. We achieve large performance gains by introducing spatial finalization into point streams: we partition space into regions, and augment a stream of input points with finalization tags that indicate when a point is the last in its region. By extending an incremental algorithm for Delaunay triangulation to use finalization tags and produce streaming mesh output, we compute a billion-triangle terrain representation for the Neuse River system from 11.2 GB of LIDAR data in 48 minutes using only 70 MB of memory on a laptop with two hard drives. This is a factor of twelve faster than the previous fastest out-of-core Delaunay triangulation software.
Article
High-resolution airborne laser scanner data offer the possibility to detect and measure individual trees. In this study, an algorithm which estimated position, height, and crown diameter of individual trees was validated with field measurements. Because all the trees in this study were measured on the ground with high accuracy, their positions could be linked with laser measurements, making validation on an individual tree basis possible. In total, 71 percent of the trees were correctly detected using laser scanner data. Because a large portion of the undetected trees had a small stem diameter, 91 percent of the total stem volume was detected. Height and crown diameter of detected trees could be estimated with a root-mean-square error (RMSE) of 0.63 m and 0.61 m, respectively. Stem diameter was estimated, using laser measured tree height and crown diameter, with an RMSE of 3.8 cm. Different laser beam diameters (0.26 to 3.68 m) were also tested, the smallest beam size showing a better detection rate in dense forest. However, estimates of tree height and crown diameter were not affected much by different beam size.
Article
. We describe crown-extraction (CE) filtering to accurately determine tree apex positions for various coniferous species using an airborne light detection and ranging–derived digital canopy height model (DCHM). This method uses a square mask, with a frame at the edges, that overlaps pixels within the DCHM image; when no pixels touch the frame, the pixel at the center is extracted as a tree-crown pixel. The apex of each tree is determined by choosing the pixel with maximum height from the pixels in the crown. We compared the performance of this method and of two other methods (local-maximum filtering and canopy-segmentation method) for several species. The CE filtering had the most accurate results for most tree species with appropriate mask size selection. The mean omission, commission, and total errors for all tree species were 8.1%, 1.6%, and 9.7%, respectively, for CE filtering. Comparing mask sizes and canopy diameters estimated from the DCHM for each species revealed that the smallest canopy diameter of each species was close to the most appropriate mask size for that species in CE filtering. We also confirmed that the smoothing process used in the DCHM has little effect on the accuracy of CE filtering.