Identifying species of individual trees using airborne laser scanner

doi:10.1016/S0034-4257(03)00140-8

Remote Sensing of Environment

Volume 90, Issue 4, 30 April 2004, Pages 415-423

https://doi.org/10.1016/S0034-4257(03)00140-8 Get rights and content

Abstract

Individual trees can be detected using high-density airborne laser scanner data. Also, variables characterizing the detected trees such as tree height, crown area, and crown base height can be measured. The Scandinavian boreal forest mainly consists of Norway spruce (Picea abies L. Karst.), Scots pine (Pinus sylvestris L.), and deciduous trees. It is possible to separate coniferous from deciduous trees using near-infrared images, but pine and spruce give similar spectral signals. Airborne laser scanning, measuring structure and shape of tree crowns could be used for discriminating between spruce and pine. The aim of this study was to test classification of Scots pine versus Norway spruce on an individual tree level using features extracted from airborne laser scanning data. Field measurements were used for training and validation of the classification. The position of all trees on 12 rectangular plots (50×20 m²) were measured in field and tree species was recorded. The dominating species (>80%) was Norway spruce for six of the plots and Scots pine for six plots. The field-measured trees were automatically linked to the laser-measured trees. The laser-detected trees on each plot were classified into species classes using all laser-detected trees on the other plots as training data. The portion correctly classified trees on all plots was 95%. Crown base height estimations of individual trees were also evaluated (r=0.84). The classification results in this study demonstrate the ability to discriminate between pine and spruce using laser data. This method could be applied in an operational context. In the first step, a segmentation of individual tree crowns is performed using laser data. In the second step, tree species classification is performed based on the segments. Methods could be developed in the future that combine laser data with digital near-infrared photographs for classification with the three classes: Norway spruce, Scots pine, and deciduous trees.

Introduction

Automatic measurements with high precision of position, height, and crown diameter of individual trees have been performed using airborne laser scanning in the forests of Europe (e.g., Hyyppä et al., 2001, Persson et al., 2002, Schardt et al., 2002). High-resolution laser scanning data is typically used to automatically generate a digital canopy model that describes the outer contour of the tree crowns. The airborne laser scanning technique can supply forest monitoring and management planning with information of most trees (e.g., position and tree size), which earlier was impossible to achieve with the same efficiency and precision. In many forest applications, it is important to know the tree species. Because the Swedish forests consist of 42% Norway spruce (Picea abies L. Karst.), 39% Scots pine (Pinus sylvestris L.), and 19% deciduous (Anon, 2002), classification into these three species classes would be useful for several applications related to planning of forest management activities. Knowing the species of individual trees is also useful for three-dimensional (3D) visualization of the forest landscape and for monitoring ecosystem functions.

Recent development of the Global Positioning System (GPS) and Inertial Navigation Systems (INS) now makes it possible to determine the orientation of a sensor with high precision without using any ground control points. Several types of airborne sensors, e.g., digital frame cameras, airborne laser scanners, and multi-spectral scanners, are therefore becoming more operational for identification and classification of objects on the ground. Some researchers have been developing algorithms for tree detection and classification from high-resolution images. To automatically find individual trees in aerial images, several methods have been developed (e.g., Brandtberg & Walter, 1998, Dralle, 1997, Erikson, 2001, Gougeon, 1999, Pollock, 1996). For tree species classification, features describing branch structure, crown shape, and color have been extracted from low-altitude aerial images (e.g., Brandtberg, 1999, Brandtberg, 2002).

Airborne laser scanners have been tested for estimation of tree species-proportions in forest stands. Törmä (2000) concluded that only using 3D-coordinates was not enough for estimating tree species-proportions with high accuracy in forest stands. Törmä proposed that better results would probably be achieved with detection of single trees and by using intensity data from the laser. In Finland, a vector model of individual trees based on laser data has been developed (Pyysalo & Hyyppä, 2002). This vector model could be used for tree species classification.

In this study, the objectives were to (1) find features in laser data useful for discriminating between Scots pine (P. sylvestris L.) and Norway spruce (P. abies L. Karst.) and; (2) validate the classification in different forest types. Discrimination between the two tree species used the following strategy. First, segmentation of individual trees was done using the digital canopy model generated from laser data. Second, tree height and crown area were derived. Third, a number of variables were extracted from the laser data to separate between segments with pine and spruce trees by finding typical characteristics of the crown shape and structure. For operational classification of Norway spruce, Scots pine and deciduous trees, it would be an advantage to combine the laser scanner with a sensor which measures the reflectance of near-infrared light. Near-infrared images are useful for separating between conifers and deciduous trees while conifers usually reflect a similar amount of near-infrared light (Lillesand & Kiefer, 1994). For example, Meyer, Staenz, and Itten (1996) classified tree species of individual trees using near-infrared photos where segments with individual tree crowns had been manually digitized. They report classification accuracies of 100% for both beech (Fagus sylvatica L.) and Silver fir (Abies alba Mill.) but lower classification accuracies for spruce and pine. Thus, laser-generated segments with deciduous trees could be separated from conifers using values of the pixels within the segment from near-infrared imagery and laser data, capturing tree shape and branch structure, could be used to separate between Scots pine and Norway spruce.

Section snippets

Study area

The test area Remningstorp located in Sweden (lat. 58°30′N, long. 13°40′E) was used. The most common tree species were Norway spruce (P. abies L. Karst.), Scots pine (P. sylvestris L.) and birch (Betula spp.). The area had a variation in elevation of 120–145 m above sea level.

Laser data

Laser data used in this study was a subset of the laser data used in an earlier study and is described in Persson et al. (2002): The laser data acquisition was performed on 13 September 2000 using TopEye, an airborne laser

Analysis methods

In an earlier study, a method to extract individual trees from laser data was developed and validated (Persson et al., 2002). The tree crown segmentation from this earlier study was used as the first step for the tree species classification in this study. All laser points within each crown segment were grouped together to form the point cloud belonging to each tree. To separate spruce and pine trees, different variables were derived from the point clouds to capture variations in the crown

Results

There was a high correlation (r=0.84) between field-measured crown base height and laser-measured crown base height for 135 sample trees (Fig. 4). The crown base height was on average overestimated by 0.75 m. The root-mean-square-error was 2.82 m. The crown base height was usually more overestimated with laser data for trees with a low crown base height compared with trees with a higher crown base height.

The classification was performed for all possible combinations of the eight selected

Discussion

The overestimation of the crown base height could be explained by the use of a too low measuring density to cover a long crown. On the other hand, crown base height was overestimated in Finland (Pyysalo & Hyyppä, 2002) with 3.0 m despite a higher measurement density (10 points/m²). The authors suggested that echoes from the lowest branches might not be registered because the system only measures one echo from each emitted pulse. In this study, a different airborne laser scanner system was used.

Conclusions

In this study, promising results, with an overall classification accuracy of 95%, are reported for classification of Scots pine and Norway spruce. In general, pine trees were more often misclassified compared with spruce trees. The crown base height was measured in field and laser scanning estimations of this variable could therefore be compared with field measurements (r=0.84). The crown base height was usually overestimated for trees with long tree crowns. The classification on a specific

Acknowledgements

The authors would like to thank Dr. Ulf Söderman, Dr. Kenneth Olofsson, Professor Håkan Olsson, Dr. Mats Nilsson, and Heather Reese for their advice and comments on the manuscript. We would like to thank Håkan Sterner and the staff at TopEye AB for delivering a high-quality laser data set. We would also like to thank Magnus Elmqvist for making it possible for us to use the active contour algorithm he developed. The high-accuracy field measurements were performed by Johan Dammström and Bernt

References (20)

T. Brandtberg
Individual tree-based species classification in high spatial resolution aerial images of forests using fuzzy sets
Fuzzy Sets and Systems
(2002)
L.D. Cohen
On active contour models and balloons
Computer Vision, Graphics, and Image Processing
(1991)
Anon
Forestry statistics 2002. Official statistics of Sweden
(2002)
Brandtberg, T. (1999). Automatic individual tree-based analysis of high spatial resolution remotely sensed data....
T. Brandtberg et al.
Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis
Machine Vision and Applications
(1998)
L.D. Cohen et al.
Finite element methods for active contour models and balloons for 2D and 3D images
IEEE Transactions on Pattern Analysis and Machine Intelligence
(1993)
Dralle, K. (1997). Locating trees by digital image processing of aerial photos. PhD thesis. Royal Veterinary and...
M. Elmqvist
Ground surface estimation from airborne laser scanner data using active shape models
M. Erikson
Structure-keeping colour segmentation of tree crowns in aerial images
F.A. Gougeon
Automatic individual tree crown delineation using a valley-following algorithm and a rule-based system

There are more references available in the full text version of this article.

Cited by (432)

A study of annual tree-wise LiDAR intensity patterns of boreal species observed using a hyper-temporal laser scanning time series
2024, Remote Sensing of Environment
This study introduces the annual tree-wise intensity patterns of three boreal tree species, silver birch (Betula pendula Roth.), Scots pine (Pinus sylvestris L.), and Norway spruce (P,icea abies H. Karst.), observed from a long-term hyper-temporal point cloud dataset collected with a permanent laser scanning (LiDAR) station. An annual LiDAR intensity pattern refers to the trend of variations of tree-wise calibrated LiDAR intensity values over the course of a year, primarily linked to species-specific phenological characteristics. Such pattern was discovered from hyper-temporal (76 observations between April 2020 and April 2021) high resolution (0.01 m 3D point spacing at 100 m distance) point cloud observations acquired using a single wavelength (1550 nm) static LiDAR system installed on a climate monitoring tower at the Hyytiälä forest research station in central Finland. A set of experiments was designed to explore the interactions among tree-wise calibrated LiDAR intensity, species, data quality, and observation dates. Thus, to provide practical suggestions for expectable accuracies of tree species classification using diverse types of LiDAR systems and data acquisition strategies. It was revealed that by combining high spatial resolution (up to 100,000 points/m²) with bi-weekly observations (two scans per week except winter period), the average tree species classification accuracy reaches 96.8% for the three boreal tree species in the study area, where the few misclassified cases are attributed to inaccurate individual tree segmentation. This suggests that the annual tree-wise LiDAR intensity patterns, which reflects the tree-wise phenological dynamics, are robust and species-specific. When the spatial resolution was reduced to 5 points/ m², the average accuracy of species classification among the three studied species reached 85% when using bi-temporal data from spring and autumn. Our analysis further indicates that arranging data acquisition with a two-week variability in late autumn and late spring could yield higher accuracy in species classification. Furthermore, our study revealed the importance of subcanopy information for species classification, specifically in distinguishing different coniferous species. When limited to upper canopy data (as in simulated airborne LiDAR scenarios), achieving a tree species classification accuracy of over 80% required either high point density (e.g. 8000 points/m²) combined with bi-temporal collection in a year, or low point density (5 points/ m²) with a dense time series (e.g., 76 time points within a year), which showcased the possibility to compensate low spatial resolution with high temporal resolution, and vice versa, in capturing species-specific phenological characteristics of trees.
Airborne data and machine learning for urban tree species mapping: Enhancing the legend design to improve the map applicability for city greenery management
2024, International Journal of Applied Earth Observation and Geoinformation
Presently remote sensing appears to be the only technology that makes it possible to conduct an inventory of tree taxa throughout the whole city. One of the main challenges during a remote sensing implementation project is the constructing of a map legend for tree taxa. The article presents the first comprehensive investigation into the construction of a legend for urban tree species mapping in the entire city, using machine learning and airborne data. The analyses were based on the CatBoost algorithm, which is relatively new in environmental research, as well as hyperspectral and LiDAR data acquired by instrument fusion. The analysis covered the entire city of Warsaw (517.24 km²). The main aim of the study was to compare three approaches to creating map legends of taxonomic diversity in city trees. This research sought to answer which of these three approaches to creating a map legend produces a result with the highest application potential for use by urban greenery managers. Two scenarios are based on different levels of taxonomic organization. The first one, based on a generic level, contained 42 classes and obtained an overall accuracy score (OA) of 0.809. The second one, on the species level, consisted of 81 classes, and the OA reached an accuracy of 0.728. The third scenario presents a mixed approach by grouping classes as different combinations of genus, species and variety. It contained 60 classes and its OA totaled to 0.798. The obtained results indicate that while in the classification process the generic scenario achieved the highest accuracy and the species scenario had the most complex legend, the mixed scenario emerged as the most useful for the greenery managers because it presented a compromise between the complexity of the legend and the high accuracy of the map. In the mixed scenario, accurate classification was achieved by separating high-accuracy individual taxa at the species level and grouping the rest at different taxonomic levels. One of the most significant novelties of our research is proving that remote sensing currently makes it possible to develop a map of tree species containing as many as 60 classes of legends and to achieve an accuracy that allows for practical application in urban greenery management. The study lists suggestions of practical strategies for enhancing map accuracy such as aggregating similar species into species groups, creating distinct groups for ornamental cultivars with unique leaf characteristics or combining species or genera with similar traits.
Detection of woody species Schinopsis haenkeana using phenological spectral differences and NDVI texture measures in subtropical forests
2024, Remote Sensing Applications: Society and Environment
Schinopsis haenkeana is a native tree species of great importance of South America. Different tree species diverge in their vegetation phenology, providing the opportunity to map their presence based on the seasonal dynamics of vegetation indices. Currently, spatially explicit information on tree species composition provides valuable insights for biodiversity conservation. The objective of this study was to detect the presence of S. haenkeana and its forest status in subtropical forests in central Argentina. We used a combination of RGB-NIR bands and indices (NDVI, EVI, GLI, RGR) derived from Sentinel-2 images. The analyses were processed using the Earth Engine platform and the random forest algorithm was used to discriminate S. haenkeana from other plant species. NDVI texture indices were also used to discriminate different forest states where the species is present. A fruiting period and a leaf color change were detected in July, and spectral differences between fruiting and preceding (May) or subsequent (October) months proved to be highly suitable for discriminating S. haenkeana. The final species presence map achieved an overall accuracy of 91%. Only 0.76% of the total area corresponds to S. haenkeana dense forests. This analysis demonstrated the value of the proposed approach for regularly detecting and mapping S. haenkeana using RGB-NIR spectral information, vegetation indices, and phenological spectral differences. Additionally, it highlighted the importance of using texture indices to differentiate between forests and scrublands, providing suitable data for forest management.
Recreating structurally realistic tree maps with airborne laser scanning and ground measurements
2023, Remote Sensing of Environment
The estimation of forest attributes representing aspects of structural biodiversity from vast areas is difficult not only because of the lack of agreement on what constitutes a structurally diverse forest, but also due to difficulties in the detection of small trees. However, a structurally diverse forest should have large variation in tree size (i.e., height and diameter), as well as some clustering of trees, as regular tree pattern is typical for managed forests. In this work we developed a framework for building structurally representative tree maps using airborne laser scanning data and a limited number of ground measurements using data from two locations in Finland. In the proposed method, an individual tree detection algorithm is optimized so that the number of undetected trees and false detections is minimized. The ground measurements are then used to train models for predicting the number and location of undetected trees and false detections by resampling the attributes of undetected trees in the training data. This model can then be applied to other areas in the proximity of the ground measurements to build tree maps, with the location, height, and diameter for each tree. The methodology was shown to reproduce the number of trees, the height distribution, and the spatial pattern of the trees with sufficient accuracy for practical use in large-scale mapping of forest attributes. For example, the individual tree detection could find at best about 54% of the trees with an additional 7% of false detections, resulting in a bias of approximately −30%, while with the new method the bias was reduced to $\pm$ 10% of the stem density. Further research is needed on how to account for tree species when building tree maps.
Integrated wildfire danger models and factors: A review
2023, Science of the Total Environment
Wildfires have been systematically studied from the early 1950s, with significant progress in the applied computational methodologies during the 21st century. However, modern methods are barely adopted by administrative authorities, globally, especially those considering probabilistic models concerning human-caused fires. An exhaustive review on wildfire danger studies has not yet been performed. Therefore, the present review aims at collecting and analyzing integrated modeling approaches in estimating forest fire danger, examining the driving factors, and evaluating their influence on fire occurrence. The main objective is to propose the top performing methods and the most important risk factors for the development of an Integrated Wildfire Danger Risk System (IWDRS). Studies were classified based on the applied technique, i.e., geographic information systems, remote sensing, statistics, machine learning, simulation modeling and miscellaneous techniques. The conclusions of each study concerning the relative importance of model input variables are also reported. Online search engines such as ‘Scopus’, ‘Google Scholar’, ‘WorldWideScience’, ‘ScienceDirect’ and ‘ResearchGate’ were used in relevant literature searches published in scientific journals, manuals and technical documentation. A total of 230 studies were gathered with a selected subset being evaluated in a meta-analysis process. Machine learning techniques outperform average classic statistics, although their predictability relies heavily on the quantity and the quality of the input data. Geographic information systems and remote sensing are considered valuable yet supplementary tools. Modeling techniques apply best to fire behavior prediction, while other techniques referenced in the current review are potentially useful but further investigation is needed. In conclusion, wildfire danger is a function of seven thematic groups of variables: meteorology, vegetation, topography, hydrology, socio-economy, land use and climate. Ninety-five explanatory drivers are proposed.
Individual tree segmentation and species classification using high-density close-range multispectral laser scanning data
2023, ISPRS Open Journal of Photogrammetry and Remote Sensing
Tree species characterise biodiversity, health, economic potential, and resilience of an ecosystem, for example. Tree species classification based on remote sensing data, however, is known to be a challenging task. In this paper, we study for the first time the feasibility of tree species classification using high-density point clouds collected with an airborne close-range multispectral laser scanning system – a technique that has previously proved to be capable of providing stem curve and volume accurately and rapidly for standing trees. To this end, we carried out laser scanning measurements from a helicopter on 53 forest sample plots, each with a size of 32 m × 32 m. The plots covered approximately 5500 trees in total, including Scots pine (Pinus sylvestris L.), Norway spruce (Picea abies (L.) H.Karst.), and deciduous trees such as Downy birch (Betula pubescens Ehrh.) and Silverbirch (Betula pendula Roth). The multispectral laser scanning system consisted of integrated Riegl VUX-1HA, miniVUX-3UAV, and VQ-840-G scanners (Riegl GmbH, Austria) operating at wavelengths of 1550 nm, 905 nm, and 532 nm, respectively. A new approach, layer-by-layer segmentation, was developed for individual tree detection and segmentation from the dense point cloud data. After individual tree segmentation, 249 features were computed for tree species classification, which was tested with approximately 3000 trees. The features described the point cloud geometry as well as single-channel and multi-channel reflectance metrics. Both feature selection and the tree species classification were conducted using the random forest method. Using the layer-by-layer segmentation algorithm, trees in the dominant and co-dominant categories were found with detection rates of 89.5% and 77.9%, respectively, whereas suppressed trees were detected with a success rate of 15.2%–42.3%, clearly improving upon the standard watershed segmentation. The overall accuracy of the tree species classification was 73.1% when using geometric features from the 1550 nm scanner data and 86.6% when combining the geometric features with reflectance information of the 1550 nm data. The use of multispectral reflectance and geometric features improved the overall classification accuracy up to 90.8%. Classification accuracies were as high as 92.7% and 93.7% for dominant and co-dominant trees, respectively.

View all citing articles on Scopus

¹: Fax: +46-13-378287.

View full text

Identifying species of individual trees using airborne laser scanner

Abstract

Introduction

Section snippets

Study area

Laser data

Analysis methods

Results

Discussion

Conclusions

Acknowledgements

Fuzzy Sets and Systems

Computer Vision, Graphics, and Image Processing

Forestry statistics 2002. Official statistics of Sweden

Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis

Machine Vision and Applications

Finite element methods for active contour models and balloons for 2D and 3D images

IEEE Transactions on Pattern Analysis and Machine Intelligence

Ground surface estimation from airborne laser scanner data using active shape models

Structure-keeping colour segmentation of tree crowns in aerial images

Automatic individual tree crown delineation using a valley-following algorithm and a rule-based system