Classification of image matching point clouds over an urban area

Airborne laser scanning (ALS) and image matching are the two main techniques for generating point clouds for large areas. While the classification of ALS point clouds has been well investigated, there are few studies that are related to image matching point clouds. In this study, point clouds of multiple resolutions from high-resolution aerial images (ground sampling distance, GSD, of 6 cm) over the city of Vienna were generated and investigated with respect to point density and processing time. Three different study sites with various urban structures are selected from a bigger dataset and classified based on two different approaches: machine learning and a traditional operator-based decision tree. Classification accuracy was evaluated and compared with confusion matrices. In general, the machine learning method results in a higher overall accuracy compared to the simple decision tree method, with accuracies of 87% and 84%, respectively, at the highest resolution. At lower-resolution levels (GSDs of 12 cm and 24 cm), the overall accuracy of machine learning drops by 4% and that of the simple decision tree by 7% for each level. Classifying rasterized data instead of the original point cloud resulted in an accuracy drop of 5%. Thus, using machine learning on point clouds at the highest available resolution is suggested for classification of urban areas.     

Imputation of single-tree attributes using airborne laser scanning-based height, intensity, and alpha shape metrics

Forest inventories based on single-tree interpretation of airborne laser scanning (ALS) data often rely on an allometric estimation chain in which inaccuracies in the estimates of the diameter at breast height (DBH) propagate to other characteristics of interest such as the stem volume. Our purpose was to test nearest neighbor imputation by the k-Most Similar Neighbor (k-MSN) and the Random Forest (RF) methods for the simultaneous estimation of species, DBH, height and stem volume using ALS data. The predictors included computational alpha shape metrics and variables based on the height and intensity distributions in the ALS data. Separate data sets covering 1898 and 1249 dominant to intermediate trees in a typical Scandinavian stand structure were used for training and validation, respectively. RF proved to be a flexible method with an ability to handle 1846 predictors with no need for their reduction. Classification of Scots pine, Norway spruce and deciduous trees showed an accuracy of 78%, and the estimates of DBH, height and volume had root mean square errors of 13%, 3%, and 31%, respectively, when evaluated against the validation data. The two selection strategies implemented here reduced the number of candidate variables effectively without any substantial effect on the accuracy relative to the use of all predictors. Differences in k-MSN and RF imputations were marginal when the reduced sets of variables were used. Estimation accuracies could be maintained practically unchanged with only 12.5% of the initial reference data (237 trees), provided the distribution of the observations was similar in the reference and target data. Since we used information collected in the field for extracting the ALS point clouds for individual trees, our results represent an optimal case and should nevertheless be validated against automated tree delineation.     

Estimation of 3D vegetation structure from waveform and discrete return airborne laser scanning data

This study presents and compares new methods to describe the 3D canopy structure with Airborne Laser Scanning (ALS) waveform data as well as ALS point data. The ALS waveform data were analyzed in three different ways; by summing the intensity of the waveforms in height intervals (a); by first normalizing the waveforms with an algorithm based on Beer–Lambert law to compensate for the shielding effect of higher vegetation layers on reflection from lower layers and then summing the intensity (b); and by deriving points from the waveforms (c). As a comparison, conventional, discrete return ALS point data from the laser scanning system were also analyzed (d). The study area was located in hemi-boreal, spruce dominated forest in the southwest of Sweden (Lat. 58° N, Long. 13° E). The vegetation volume profile was defined as the volume of all tree crowns and shrubs in 1 dm height intervals in a field plot and the total vegetation volume as the sum of the vegetation volume profile in the field plot. The total vegetation volume was estimated for 68 field plots with 12 m radius from the proportion between the amount of ALS reflections from the vegetation and the total amount of ALS reflections based on Beer–Lambert law. ALS profiles were derived from the distribution of the ALS data above the ground in 1 dm height intervals. The ALS profiles were rescaled using the estimated total vegetation volume to derive the amount of vegetation at different heights above the ground. The root mean square error (RMSE) for cross validated regression estimates of the total vegetation volume was 31.9% for ALS waveform data (a), 27.6% for normalized waveform data (b), 29.1% for point data derived from the ALS waveforms (c), and 36.5% for ALS point data from the laser scanning system (d). The correspondence between the estimated vegetation volume profiles was also best for the normalized waveform data and the point data derived from the ALS waveforms and worst for ALS point data from the laser scanning system as demonstrated by the Reynolds error index. The results suggest that ALS waveform data describe the volumetric aspects of vertical vegetation structure somewhat more accurately than ALS point data from the laser scanning system and that compensation for the shielding effect of higher vegetation layers is useful. The new methods for estimation of vegetation volume profiles from ALS data could be used in the future to derive 3D models of the vegetation structure in large areas.     

Influence of terrain model smoothing and flight and sensor configurations on detection of small pioneer trees in the boreal-alpine transition zone utilizing height metrics derived from airborne scanning lasers

It has been suggested that airborne laser scanning (ALS) with high point densities could be used to monitor changes in the alpine tree line. The overall goal of this study was to assess the influence of ALS sensor and flight configurations on the ability to detect small trees in the alpine tree line and on the estimation of their heights. The study was conducted in a sub-alpine/alpine environment in southeast Norway. 342 small trees (0.11-5.20 m tall) of Norway spruce, Scots pine, and downy birch were precisely georeferenced and measured in field. ALS data acquired with two different instruments and at different flying altitudes (700-1130 m a.g.l.) with different pulse repetition frequencies (100, 125, and 166 kHz) were collected with a point density of all echoes of 7.7-11.0 m^− 2. For each acquisition, three different terrain models were used to process the ALS point clouds in order to assess the effects of different preprocessing parameters on the ability to detect small trees. Regardless of acquisition and terrain model, positive height values were found for 91% of the taller trees (> 1 m). For smaller trees (< 1 m), 29-61% of the trees displayed positive height values. For the lowest repetition frequencies (100 and 125 kHz) in particular, the portion of trees with positive laser height values increased significantly with increasing terrain smoothing. For the highest repetition frequency there were no differences between smoothing levels, likely because of large ALS measurement errors at low laser pulse energy levels causing a large portion of the laser echoes to be discarded during terrain modeling. Error analysis revealed large commission errors when detecting small trees. The commissions consisted of objects like terrain structures, rocks, and hummocks having positive height values. The magnitude of commissions ranged from 709 to 8948% of the true tree numbers and tended to increase with increasing levels of terrain smoothing and with acquisitions according to increasing point densities. The accuracy of tree height derived from the ALS data indicated a systematic underestimation of true tree height by 0.35 to 1.47 m, depending on acquisition, terrain model, and tree species. The underestimation also increased with increasing tree height. The standard deviation for the differences between laser-derived and field-measured tree heights was 0.16-0.57 m. Because there are significant effects of sensor and flight configurations on tree height estimation, field calibration of tree heights at each point of time is required when using airborne lasers for tree growth monitoring.     

Classification of airborne laser scanning point clouds based on binomial logistic regression analysis

This article presents a newly developed procedure for the classification of airborne laser scanning (ALS) point clouds, based on binomial logistic regression analysis. By using a feature space containing a large number of adaptable geometrical parameters, this new procedure can be applied to point clouds covering different types of topography and variable point densities. Besides, the procedure can be adapted to different user requirements. A binomial logistic model is estimated for all a priori defined classes, using a training set of manually classified points. For each point, a value is calculated defining the probability that this point belongs to a certain class. The class with the highest probability will be used for the final point classification. Besides, the use of statistical methods enables a thorough model evaluation by the implementation of well-founded inference criteria. If necessary, the interpretation of these inference analyses also enables the possible definition of more sub-classes. The use of a large number of geometrical parameters is an important advantage of this procedure in comparison with current classification algorithms. It allows more user modifications for the large variety of types of ALS point clouds, while still achieving comparable classification results. It is indeed possible to evaluate parameters as degrees of freedom and remove or add parameters as a function of the type of study area. The performance of this procedure is successfully demonstrated by classifying two different ALS point sets from an urban and a rural area. Moreover, the potential of the proposed classification procedure is explored for terrestrial data.     

Development of a robust canopy height model derived from ALS point clouds for predicting individual crown attributes at the species level

ABSTRACT

Tree crown attributes are important parameters during the assessment and monitoring of forest ecosystems. Canopy height models (CHMs) derived from airborne laser scanning (ALS) data have proved to be a reliable source for extracting different biophysical characteristics of single trees and at stand level. However, ALS-derived tree measurements (e.g., mean crown diameter) can be negatively affected by pits that appear in the CHMs. Thus, we propose a novel method for generating pit-free CHMs from ALS point clouds for estimating crown attributes (i.e., area and mean diameter) at the species level. The method automatically calculates a threshold for a pixel based on the range of height values within neighbouring pixels; if the pixel falls below the threshold then it is recognized as a pitted pixel. The pit is then filled with the median of the values of the neighbouring pixels. Manually delineated individual tree crowns (ITC) of four deciduous and two coniferous species on Colour Infrared (CIR) stereo images were used as a reference in the analysis. In addition, a variety of different algorithms for constructing CHMs were compared to investigate the performance of different CHMs in similar forest conditions. Comparisons between the estimated and observed crown area (R² = 0.95, RMSE% = 19.12% for all individuals) and mean diameter (R² = 0.92, RMSE% = 12.16% for all individuals) revealed that ITC attributes were correctly estimated by segmentation of the pit-free CHM proposed in this study. The goodness of matching and geometry revealed that the delineated crowns correctly matched up to the reference data and had identical geometry in approximately 70% of cases. The results showed that the proposed method produced a CHM that estimates crown attributes more accurately than the other investigated CHMs. Furthermore, the findings suggest that the proposed algorithm used to fill pits with the median of height observed in surrounding pixels significantly improve the accuracy of the results the species level due to a higher correlation between the estimated and observed crown attributes. Based on these results, we concluded that the proposed pit filling method is capable of providing an automatic and objective solution for constructing pit-free CHMs for assessing individual crown attributes of mixed forest stands.     

The k-MSN method for the prediction of species-specific stand attributes using airborne laser scanning and aerial photographs

Various studies have been presented within the last 10 years on the possibilities for predicting forest variables such as stand volume and mean height by means of airborne laser scanning (ALS) data. These have usually considered tree stock as a whole, even though it is tree species-specific forest information that is of primary interest in Finland, for example. We will therefore concentrate here on prediction of the species-specific forest variables volume, stem number, basal area, basal area median diameter and tree height, applying the non-parametric k-MSN method to a combination of ALS data and aerial photographs in order to predict these stand attributes simultaneously for Scots pine, Norway spruce and deciduous trees as well as total characteristics as sums of the species-specific estimates. The predictor variables derived from the ALS data were based on the height distribution of vegetation hits, whereas spectral values and texture features were employed in the case of the aerial photographs. The data covered 463 sample plots in 67 stands in eastern Finland, and the results showed that this approach can be used to predict species-specific forest variables at least as accurately as from the current stand-level field inventory for Finland. The characteristics of Scots pine and Norway spruce were predicted more accurately than those of deciduous trees.     

Airborne traffic monitoring in large areas using LiDAR data – theory and experiments

This article investigates the theoretical background for airborne LiDAR (light detection and ranging) and ALS (airborne laser scanning) systems that are used to monitor traffic from airborne platforms. An object moving with a velocity deviating from the assumptions incorporated in the scanning process will generally appear both stretched and sheared – motion artefacts. To study the impact of these deformations on the ALS data, the analytic relations between an arbitrarily moving object and its conjugate in the ALS data have been examined and adapted to concrete airborne specifications. Furthermore, a complete scheme is proposed to analyse urban traffic in real-life situations, which combines vehicle detection successively with the motion classification method, which is the main focus of this article. Finally, the velocity of the moving vehicle can be derived with knowledge about the vehicle shape. The experimental results obtained by using real ALS data were assessed with respect to the reference data concurrently acquired by a video camera to validate the theory.     

Per‐field urban land use classification based on tax parcel boundaries

The objective of this study is to test a per‐field approach for classifying detailed urban land use, such as single‐family, multi‐family, industrial and commercial. Tax parcel boundaries are used as the field boundaries for classification. Twelve attributes of parcels, such as parcel sizes, parcel shape, building counts and building heights, are used as the discriminant factors between different land use types. For our study area that consists of 33 025 parcels, we first derived parcel attributes from geographic information system (GIS) and remote sensing data. We then converted the parcel vector data to an image of 12 bands with pixel values from parcel attributes. After that, we performed a standard supervised classification to classify the image into nine land use types. The best classification result with a decision tree classifier had an overall accuracy of 93.53% and a Kappa Coefficient of 0.7023. This study shows the feasibility of applying a per‐field approach based on tax parcel boundaries to classify detailed urban land use.     

Geo-referencing forest field plots by co-registration of terrestrial and airborne laser scanning data

Remote sensing plays an important role within the field of forest inventory. Airborne laser scanning (ALS) has become an effective tool for acquiring forest inventory data. In most ALS-based forest inventories, accurately positioned field plots are used in the process of relating ALS data to field-observed biophysical properties. The geo-referencing of these field plots is typically carried out by means of differential global navigation satellite systems (dGNSS), and often relies on logging times of 15–20 min to ensure adequate accuracy under different forest conditions. Terrestrial laser scanning (TLS) has been proposed as a possible tool for collection of field data in forest inventories and can facilitate rapid acquisition of these data. In the present study, a novel method for co-registration of TLS and ALS data by posterior analysis of remote-sensing data – rather than using dGNSS – was proposed and then tested on 71 plots in a boreal forest. The method relies on an initial position obtained with a recreational-grade GPS receiver, in addition to analysis of the ALS and TLS data. First, individual tree positions were derived from the remote-sensing data. A search algorithm was then used to find the best match for the TLS-derived trees among the ALS-derived trees within a search area, defined relative to the initial position. The accuracy of co-registration was assessed by comparison with an accurately measured reference position. With a search radius of 25 m and using low-density ALS data (0.7 points m^?2), 82% and 51% of the TLS scans were co-registered with positional errors within 1 m and 0.5 m, respectively. By using ALS data of medium density (7.5 points m^?2), 87% and 78% of the scans were co-registered with errors within 1 m and 0.5 m of the reference position, respectively. These results are promising and the method can facilitate rapid acquisition and geo-referencing of field data. Robust methods to identify and handle erroneous matches are, however, required before it is suitable for operational use.     

Constructive Hypervolume Modeling

This paper deals with modeling point sets with attributes. A point set in a geometric space of an arbitrary dimension is a geometric model of a real/abstract object or process under consideration. An attribute is a mathematical model of an object property of arbitrary nature (material, photometric, physical, statistical, etc.) defined at any point of the point set. We provide a brief survey of different modeling techniques related to point sets with attributes. It spans such different areas as solid modeling, heterogeneous objects modeling, scalar fields or “implicit surface” modeling and volume graphics. Then, on the basis of this survey we formulate requirements to a general model of hypervolumes (multidimensional point sets with multiple attributes). A general hypervolume model and its components such as objects, operations, and relations are introduced and discussed. A function representation (FRep) is used as the basic model for the point set geometry and attributes represented independently using real-valued scalar functions of several variables. Each function defining the geometry or an attribute is evaluated at the given point by a procedure traversing a constructive tree structure with primitives in the leaves and operations in the nodes of the tree. This reflects the constructive nature of the symmetric approach to modeling geometry and associated attributes in multidimensional space. To demonstrate a particular application of the proposed general model, we consider in detail the problem of texturing, introduce a model of constructive hypervolume texture, and then discuss its implementation, as well as the special modeling language we used for modeling hypervolume objects.     

Single tree species classification from Terrestrial Laser Scanning data for forest inventory

Due to the increasing use of Terrestrial Laser Scanning (TLS) systems in the forestry domain for forest inventory, the development of software tools for the automatic measurement of forest inventory attributes from TLS data has become a major research field. Numerous research work on the measurement of attributes such as the localization of the trees, the Diameter at Breast Height (DBH), the height of the trees, and the volume of wood has been reported in the literature. However, to the best of our knowledge the problem of tree species recognition from TLS data has received very little attention from the scientific community. Most of the research work uses Airborne Laser Scanning (ALS) data and measures tree species attributes on large scales. In this paper we propose a method for individual tree species classification of five different species based on the analysis of the 3D geometric texture of the bark. The texture features are computed using a combination of the Complex Wavelet Transforms (CWT) and the Contourlet Transform (CT), and classification is done using the Random Forest (RF) classifier. The method has been tested using a dataset composed of 230 samples. The results obtained are very encouraging and promising.     

基于最小生成树的大规模数据分类模型及其MapReduce实现

数据的快速增长,为我们提供了更多的信息,然而,也对传统信息获取技术提出了挑战。这篇论文提出了MCMM算法,它是基于MapReduce的大规模数据分类模型的最小生成树（MST）的算法。它可以看做是介于传统的KNN方法和基于聚类分类方法之间的模型,旨在克服这两种方法的不足并能处理大规模的数据。在这一模型中,训练集作为有权重的无向完全图来处理。顶点是对象,两点之间边的权重是对象间的距离。这一距离,不同于欧几里得距离,它是一个特定的距离度量。这样,可以找到图中最小生成树集,其中,图中每棵树代表一个类。为了降低时间复杂度,提取了每棵树中最具代表性的点来代表该树。这些压缩了的点集,可以通过计算无标签对象和它们之间的距离,来进行分类。MCMM模型基于MapReduce实现并且部署在Hadoop平台。该模型可扩展处理大规模的数据,是因为Hadoop支持数据密集分布应用,并且这些应用可以和数以千计的节点和数据一起运作。另外,MapReduce 和Hadoop能在由商品机组成的集群上很好的运行。MCMM模型使用云平台并且通过使用MapReduce 和Hadoop进行云计算是有益处的。实验采用的数据集包括从UCI数据库得到的真实数据和一些模拟数据,实验使用了4000个集群。实验表明,MCMM模型在精确度和扩展性上优于KNN和其他一些经常使用的基础分类方法。     

Effects of different sensors and leaf-on and leaf-off canopy conditions on echo distributions and individual tree properties derived from airborne laser scanning

The objectives of this study were to quantify and analyze differences in laser height and laser intensity distributions of individual trees obtained from airborne laser scanner (ALS) data for different canopy conditions (leaf-on vs. leaf-off) and sensors. It was also assessed how estimated tree height, stem diameter, and tree species were influenced by these differences. The study was based on 412 trees from a boreal forest reserve in Norway. Three different ALS acquisitions were carried out. Leaf-on and leaf-off data were acquired with the Optech ALTM 3100 sensor, and an additional leaf-on dataset was acquired using the Optech ALTM 1233 sensor. Laser echoes located within the vertical projection of the tree crowns were attributed to different echo categories (“first echoes of many”, “single echoes”, “last echoes of many”) and analyzed. The most pronounced changes in laser height distribution from leaf-on to leaf-off were found for the echo categories denoted as “single” and “last echoes of many” where the distributions were shifted towards the ground under leaf-off conditions. The most pronounced change in the intensity distribution was found for “first echoes of many” where the distribution was extremely skewed towards the lower values under leaf-off conditions compared to leaf-on. Furthermore, the echo height and intensity distributions obtained for the two different sensors also differed significantly. Individual tree properties were estimated fairly accurately in all acquisitions with RMSE ranging from 0.76 to 0.84 m for tree height and from 3.10 to 3.17 cm for stem diameter. It was revealed that tree species was an important model term in both and tree height and stem diameter models. A significantly higher overall accuracy of tree species classification was obtained using the leaf-off acquisition (90 vs. 98%) whereas classification accuracy did not differ much between sensors (90 vs. 93%).     

Airborne laser data for stand delineation and information extraction

A literature review of new publications in the field of 3D data for forest applications shows that the application of airborne laser scanner data (ALS) is in the focus of research today due to its great potential for practical applications. While there is a lot of research carried out to derive forest management parameters based on laser metrics deduced from a single tree assessment or a statistical area based assessment, the delineation of stand or sub‐stand units derived from laser metrics itself is a rather new approach. In order to describe stand characteristics statistical grid cell approaches or single tree approaches have been developed. The LIDAR based segmentation of stand or sub‐stand units is rarely documented. This article provides information on enhanced processes to delineate stand or sub‐stand units and to extract different forest information based on airborne laser derived parameters. For the stand delineation an automatic process was developed which provides a stand or sub‐stand unit delineation which is according to the first results sufficiently uniform within stands and sufficiently different in species, age class, height class, structure and composition between stands in order to be distinguishable from adjacent areas. With a combined method the stand boundaries as they are established by the mapping units today, as well as sub‐stand units which have in common physical characteristics indicating the same management disposition, were assessed. Finally a first validation of the forest stand unit delineation is provided, indicating the high potential of ALS data for separating stand units.     

基于原型抽象和分类价值量的决策树算法*

为了克服现有决策树分类算法在大数据集上的有效性和可伸缩性的局限,提出一种新的基于粗糙集理论的决策树算法。首先提出基于代表性实例的原型抽象方法,该方法从原始数据集中抽取代表性实例组成抽象原型,可缩减实例数目和无关属性,从而使算法可以处理大数据集;然后提出属性分类价值量概念,并作为选择属性的启发式测度,该测度描述了属性对分类的贡献价值量的多少,侧重考虑了属性之间以及实例与分类之间的关系。实验表明,新算法比其他算法生成的决策树规模要小,准确率也有显著提高,在大数据集上尤为明显。     

Boosting the predictive accuracy of urban hedonic house price models through airborne laser scanning

This paper introduces an integrative approach to hedonic house price modeling which utilizes high density 3D airborne laser scanning (ALS) data. In general, it is shown that extracting exploratory variables using 3D analysis – thus explicitly considering high-rise buildings, shadowing effects, etc. – is crucial in complex urban environments and is limited in well-established raster-based modeling. This is fundamental in large-scale urban analyses where essential determinants influencing real estate prices are constantly missing and are not accessible in official and mass appraiser databases. More specifically, the advantages of this methodology are demonstrated by means of a novel and economically important externality, namely incoming solar radiation, derived separately for each flat. Findings from an empirical case study in Vienna, Austria, applying a non-linear generalized additive hedonic model, suggest that solar radiation is significantly capitalized in flat prices. A model comparison clearly proves that the hedonic model accounting for ALS-based solar radiation performs significantly superior. Compared to a model without this externality, it increases the model’s explanatory power by approximately 13% and additionally reduces the prediction error by around 15%. The results provide strong evidence that explanatory variables originating from ALS, explicitly regarding the immediate 3D surroundings, enhance traditional hedonic models in urban environments.     

Object class segmentation of massive 3D point clouds of urban areas using point cloud topology

A large number of remote-sensing techniques and image-based photogrammetric approaches allow an efficient generation of massive 3D point clouds of our physical environment. The efficient processing, analysis, exploration, and visualization of massive 3D point clouds constitute challenging tasks for applications, systems, and workflows in disciplines such as urban planning, environmental monitoring, disaster management, and homeland security. We present an approach to segment massive 3D point clouds according to object classes of virtual urban environments including terrain, building, vegetation, water, and infrastructure. The classification relies on analysing the point cloud topology; it does not require per-point attributes or representative training data. The approach is based on an iterative multi-pass processing scheme, where each pass focuses on different topological features and considers already detected object classes from previous passes. To cope with the massive amount of data, out-of-core spatial data structures and graphics processing unit (GPU)-accelerated algorithms are utilized. Classification results are discussed based on a massive 3D point cloud with almost 5 billion points of a city. The results indicate that object-class-enriched 3D point clouds can substantially improve analysis algorithms and applications as well as enhance visualization techniques.     

混合属性数据点集的特征权重优化方法研究

应用决策树方法来获取混合属性数据点集的“规则聚类区域”,利用“异类子聚类相离,同类子聚类相近”的原则来交替优化有序属性和无序属性的权重,提出了基于决策树划分的特征权重优化方法。该方法在一定程度上解决了有效获取数据子集的子聚类问题和混合属性数据点集的特征权重优化难题。仿真实验表明,该方法在优化混合属性数据点集的特征权重时是有效的。     

Classifying species of individual trees by intensity and structure features derived from airborne laser scanner data

The objective of this study was to identify candidate features derived from airborne laser scanner (ALS) data suitable to discriminate between coniferous and deciduous tree species. Both features related to structure and intensity were considered. The study was conducted on 197 Norway spruce and 180 birch trees (leaves on conditions) in a boreal forest reserve in Norway. The ALS sensor used was capable of recording multiple echoes. The point density was 6.6 m^− 2. Laser echoes located within the vertical projection of the tree crowns, which were assumed to be circular and defined according to field measurements, were attributed to three categories: “first echoes of many”, “single echoes”, or “last echoes of many echoes”. They were denoted FIRST, SINGLE, and LAST, respectively. In tree species classification using ALS data features should be independent of tree heights. We found that many features were dependent on tree height and that this dependency influenced selection of candidate features. When we accounted for this dependency, it was revealed that FIRST and SINGLE echoes were located higher and LAST echoes lower in the birch crowns than in spruce crowns. The intensity features of the FIRST echoes differed more between species than corresponding features of the other echo categories. For the FIRST echoes the intensity values tended to be higher for birch than spruce. When using the various features for species classification, maximum overall classification accuracies of 77% and 73% were obtained for structural and intensity features, respectively. Combining candidate features related to structure and intensity resulted in an overall classification accuracy of 88%.