1 Introduction
Forestry has been undergoing a budge from conventional forest management techniques, and using time consuming and laborious field work and aerial photography to the use of remotely sensed data and analysis in Geographic Information System (GIS) for last few decades. With the improvement spatial resolution of remotely sensed data (IKONOS (1 to 4 m2), Quick Bird (0.6 to 2.8 m2) etc) and application of new technologies like Light Detection and Ranging (LiDAR) have started to change the way of thinking and direction of how forester approaches forest management questions. Very high resolution (VRH) satellite imageries provide a wide variety of spatial and spectral information about earth surface, on other hand LiDAR provides the opportunity to study the terrain and earth surfaces for several years. With the passage of time, the costs of field work increases accompanied by the demands of in time and detailed information about forest resources, which are compelling forest managers to consider significant changes in their approaches to manage forest resources (Evans, Roberts & Parker 2006). It for these reasons, that the used of these newly available remotely sensed data has gained high popularity among the natural resource managers.
Remote sensing has important role in forestry for several decades, especially when used for extracting information about the location, extent, composition, and structure of the forest resource as part of industrial forest inventories (Chubey, Franklin & Wulder 2006). The forest stand delineation from VRH imageries has been the main focus of the research in forest management for the few years. Some of important studies are (Chubey, Franklin & Wulder 2006, Förster, Kleinschmit 2008, Giannetti, Gottero & Terzuolo 2003, Lamonaca, Corona & Barbati 2008, Shiba, Itaya 2006, Leckie et al. 2003b, Wulder et al. 2008). In these studies object-based analysis approach of VRH multispectral imageries has been used instead of traditional pixel-based analysis. As for given forest stand the spectral response in VRH resolution imageries presented as a series of disjoin pixels covering a wide range of spectral values, on other hand, for forest inventory purposes, forest stands are interpreted as homogenous polygons (Hall 2003). Image segmentation technique can offer possible solution to this, which partition image into spatially continuous, disjoint and homogenous regions (Blaschke, Burnett & Pekkarinen 2004).
Likewise VRH imageries, the use of LiDAR for forest inventory and management purposes has gained much attention in recent years. LiDAR has been used for forest stand level (Hyyppä, Hyyppä 1999, Holmgren 2004, Næsset 1997a, Næsset 2004, Næsset 2004) and single tree(Maltamo et al. 2004b, Persson, Holmgren & SÖDERMAN 2002, Suárez et al. 2005) level estimation of different parameters, especially height. The used of both VRH multispectral imageries and LiDAR can greatly improve the identification of forest stands, as both spectral and height information will be made available, which are important considerations for forest stand delineation.
The central focus of this research is to propose a method for identification of forest stands from different remotely sensed data (VRH multispectral imagery and LiDAR) and analysis. In addition the spatial relationship of pixels in identified stand will also be investigated with help of spatial statistics such as local indicator of spatial association (LISA).
2 Background
2.1 Forest stands: Importance and meaning
Forest stands are the fundamental spatial unit used by foresters in day-to-day practice (Leckie et al. 2003b) and for inventory, economic analyses and planning; thus, their delineation is a vital process in forest management and forest inventory. Generally, the forest stand is defined as a “contiguous group of trees sufficiently uniform in species composition, arrangement of age classes, site quality and condition to be a distinguishable unit” (Smith et al., 1997). More specifically, this also implies uniform height, age, stem density, crown closure, (Leckie et al. 2003b). Basal area (BA), average height, percent cover, trees per acre, stand volume, age and species composition have also been considered (Smith & Anson 1968, Smelser & Patterson 1975, Avery 1978). In many other contexts forest stands are further defined by forestry activities, for example as operational units in forest planning and management (Holmstrom, 2002; Leckie et al., 2003; (Maltamo et al. 2005).
Given this variation in stand delineation parameters, it is unsurprising that determining the nature of a forest stand can be been viewed a subjective process, mainly ruled by the needs and desires of the company, agency or organization that is managing the land (Smith et al. 1997). As Franklin (2001) notes, “it seems increasingly obvious that the rules of forest mapping as practiced over the past few decades are not particularly logical at all, but are strongly dependant on the skill of the analyst, the local nature of the forest condition, and cultural tradition in the particular jurisdiction responsible for fulfilling demands for forest information.” Even when using an “objective” automated segmentation method to delineate stands, it is important to note that these cultural differences and definitions sit behind the process and will therefore impact on methodology and approach.
However the forest stand is defined, it is clear that the collection of large amounts of field data plus considerable office processing time is needed to effect the definition. Natural resources management of large-scale areas can be costly in terms of time, labour and other resources; effective and efficient means of gathering and processing data over large areas are required.
2.1.1 Traditional methods of delineating forest stands
In forest management practices, stands have traditionally been delineated on aerial photographs with the help of human pattern recognition. Photogrammetric interpretations are subsequently made, which are in turn supplemented by local field knowledge and observation (Franklin 2001). Stand delineation using aerial photographs requires the formal training of the analyst in photogrammetry, traditionally necessitated when classifying an extensive coverage of forest with aerial photographs.
Stand delineation from aerial photography is the conventional practice, but it is also extensively accepted that there are significant limitations to this approach. Some researchers have reported their concerns about the accuracies of the stand delineation with this approach; these inaccuracies can potentially result in severe financial impact for the landowner (Næset 1999). This process can also be expensive in terms time and resources (Franklin 2001, Skidmore 1989). It also can provide inconsistent results, which may not be reproducible. Further, classification using aerial photography often does not provide the detail or accuracy of results required for management (Chubey, Franklin & Wulder 2006). Ultimately, as Franklin (2001) reports, “classification and mapping are always done for some purpose; it is that purpose, and the skill of the analyst, which exert perhaps the strongest influence on the accuracy and utility of the final products.”
2.2 LiDAR in forestry activities
New applications of remote sensing technology such as Airborne Laser Scanning (ALS) also known as Light Detection and Ranging (LiDAR) have become an effective tool for the management of natural resources in wide-scale areas. This remote sensing technique developed rapidly in early and middle 1970's in North America, largely through bathymetric and hydrographic applications (Hyyppä et al. 2004). Experiments with laser scanning in connection with forest inventory and management started in the early 1980s (Nelson et al., 1984; Aldred and Bonnor, 1985).
In recent years, the use of LiDAR for forest inventory and management purposes has gained much attention. Key papers include those by (Næsset 2004, Hilker, Wulder & Coops 2008, Hyyppä, Inkinen 1999a, Hyyppae et al. 2000, Mustonen, Packalén & Kangas 2008, Pascual et al. 2008). Methods and techniques for determining forest attributes and structure measurements both at an individual tree level and plot level have been investigated, and are continuing to be improved and developed (Reutebuch et al. 2005).
2.2.1 LiDAR approaches: Stand data
Recent research focusing on the use of LiDAR for plot and stands level measurements of forest has concluded that LiDAR can provide comparable or better accuracy for certain measurements than field or photo interpretation (Maltamo et al. 2004a), (Næsset 1997a, Næsset 1997b).
Estimation of stand height has been particularly successful using LiDAR. Naesset (1997, p. 55) for example reports that “the current study has shown that laser scanner data may be used to obtain estimates of forest stand heights with an accuracy equal to, or even higher than, those provided by present inventory methods based on aerial photo interpretation.” This finding is confirmed by Mean et al. (1999) in their study Scanning LiDAR of Canopies by Echo Recovery (SLICER), which measured the structure of forests in the Pacific Northwest. Mean heights derived from SLICER was observed to be a good predictor of mean canopy height.
More recent work has found considerable success measuring or deriving a wider range of forest parameters. Mean et al.'s (1999) SLICER project, in which ground measurements of forest structure were also made from 26 plots, found that basal area and SLICER-derive height are closely related. The authors also found that there is a strong relationship between the square of SLICER-derived height and total stand biomass. Principal stand characteristics were also estimated by (Næsset 2004)) with higher accuracy using LiDAR data than applying traditional methods in forest inventory, leading the authors to conclude that area-based approaches to estimate forest stand variables from laser scanner data have matured and are now implemented in operational projects in Scandinavian countries.
Holmgren (2004) experimented with the grid-based approach for predicting forest variables on a stand level using. In this experiment regression models were developed that provide relationship between laser data derived variables and mean tree height, basal area and stand volume. The results revealed that the accuracy for all variables were high, both on a plot level and for the validation data. Some researchers compared LiDAR estimates of mean tree height and stand volumes with estimate measure with the help of other airborne and satellite remote sensing data (Hyyppä, Hyyppä 1999). As result of testing these different data types, laser scanner data were found to be provide similar or higher accuracies than traditional forest inventory methods.
2.2.2 LiDAR approaches: Individual tree features
Other research has investigated the use of LiDAR for estimation of individual tree features in a forest. According to (Næsset et al. 2004), p. 492) “the basic idea of single-tree-based forest inventory is that the calculation of the stand attributes for an individual stand is based on measurements of the position, tree height, species and crown area for individually detected trees. All other stand variables are derived from these basic characteristics, possibly also in combination with field data. The position, tree height and tree crown areas can be obtained from laser scanner data, whereas the tree species is obtained from image data, from laser data, or from a combination of laser and image data”.
It has been reported that single trees can be detected within High-pulse-rate laser data (Hyyppä, Inkinen 1999b, Hyyppa et al. 2000). Different methods have been developed for the single trees detection and measuring. For example, (Persson, Holmgren & SÖDERMAN 2002)) proposed a method in which, firstly, CHM is created by an active surface algorithm and then with different scales CHM is smoothened, lastly, parabolic surface is used to determine appropriate scale in different parts of the image. On the validation of the method it was observed that more than 70% of the trees were detected. Other method for single tree detection from laser measurements involved finding of local maxima in a low-pass filtered CHM after wards segmentation procedure was used for edge detection. (Leckie et al. 2003a)) made use of a valley-following approach for the isolation individual tree from high resolution CHM and digital frame camera imagery. It was revealed that in dense stands optical imagery may provide better results in delineating crowns.
(Suárez et al. 2005) used high resolution CHM and aerial photograph to estimate the height of individual tree. A segmentation procedure available in eCognition was used to combine pixels which are similar in terms of elevation and reflectance. The research was able to predict 73% of all the heights within 1 m; 91% within 1.5 m and 96% within 2 m.
In many forest and inventory and management practices, tree species is of particular significance. Conventionally, tree species information is extracted from high-spatial-resolution colour infrared aerial photographs (Brandtberg 2002). Presently, both optical/near-infrared and laser data can be used for classification of tree species. (Holmgren, Persson & Sodermann 2006) conducted a study to identify individual tree species by combing features of high resolution multi-spectral images with high density LiDAR data. It was the observed that the classification accuracy of 95% can be achieved when the combination of LiDAR-derived structure and spectral characteristics are used, in a forest dominated by Norway spruce (Picea abies), Scots pine (Pinus sylvestris), and deciduous trees, mainly birch (Betula spp).
In the light of above discussion, it can be concluded that height both at stand level and individual tree level, are important parameters for the management of forest, which can be best estimated with help of LiDAR with high enough accuracy. Moreover, the CHM derived from LiDAR provide the opportunity to discriminate forest into different stand on the basis of height classes.
2.3 Image Segmentation
Before going into the details of the image segmentation we first define image objects. Image-objects are defined by (Hay et al. 2001) as basic entities, located within an image that are perceptually generated from pixel groups, where each pixel group is composed of similar digital values, and possesses an intrinsic size, shape, and geographic relationship with the real-world scene component it models. (Schneider, Steinwendner 1999) suggest a simpler definition for image-objects, ‘groups of pixels with a meaning in the real world'.
With the improvement of the spatial resolution of remote sensors, the possibilities for identifying image objects increases. The data contain considerably greater volumes of information regarding the relationship between adjacent pixels, including texture and shape information; this allows for identification of individual objects as opposed to single pixels. However:
• The enormous amounts of data created a strong need for new methods to exploit these data efficiently.
• In addition, the complexity of the relationship of pixel and object make it essential to develop additional methods of classification (Blaschke, Burnett & Pekkarinen 2004).
• Further, Very High Resolution (VHR) satellite images (IKONOS, Quick Bird) and aerial images can create classification problems owing to greater spectral variation than older satellites within a class, and their greater degree of shadow (Laliberte et al. 2004).
• Moreover, in nature real world objects are not always separated by hard boundaries and sometimes boundaries are not available readily.
Overall, as image ground instantaneous field of view (GIFOV), or pixel (picture element) size, decreases we are face new challenges. We can potentially resolve a wide variety of real world objects, since the heterogeneity within the object increases and the spectral separability between image object decreases. In order to be able to fully utilise the improving spatial resolution, need a way to combined pixels into suitable spatial units (image objects) for the image analysis. This can be achieved by subject image to segmentation.
Image segmentation is the dividing of an array of measurements on the basis of basis on homogeneity. To be more precise, segmentation is the partitioning of an image into spatially continuous, disjoint and homogenous regions (Blaschke, Burnett & Pekkarinen 2004). Inevitably, this type of image analysis leads to meaningful objects only when the image is segmented in ‘homogenous' areas (Gorte 1998, Molenaar 1998, Baatz, Schäpe 2000). Where these conditions apply, segmentation is intuitively appealing; it provides the opportunity to divide an image into meaningful objects relating to the land surface, just as in human vision.
Image segmentation methods can be classified into three approaches: pixel-, edge and region based segmentation methods.
The pixel based methods include image thresholding and the segmentation in feature space. The results which meet the requirements and definition of segmentation may not be necessarily obtained by pixel base methods, and therefore the resulting outputs needs to be clustered together. In other words, a unique label must be assigned to each spatially continuous unit.
Edge based segmentation is based on locating edges between the image objects and determining the segments as image object within these edges. In this context, edges are considered as boundaries between image objects and they are located where values changes.
Region based segmentation algorithms aggregate pixels with seed pixels and growing into regions or image object until a certain threshold is achieved. The threshold is usually a homogeneity criterion or a combination of size and homogeneity. New seed pixels are placed when a region grow until no more pixels are allocated to any of the segments and the process is repeated. This process continues until the whole image is segmented.
In remote sensing, a single sensor correlates with range of scales rather than a single scale. The ability of resolving an object can be considered relative to the resolution of sensor (Blaschke, Burnett & Pekkarinen 2004). A rough rule of thumb is that the scale of image objects to be interpreted must be significantly higher than the scale of image noise relative to texture (Haralick, Shapiro 1985). This ensures that the subsequent object oriented image processing is based on meaningful objects. An important characteristic of any segmentation process is the homogeneity of the objects. Only if contrasts are treated consistently are good results are expected (Baatz & Schäpe, 2000).
In addition, the resulting segmentation should be reproducible and universal which allows the application to a large variety of data. Baatz & Schäpe argue that multiresolution image processing based on texture and utilising fractal algorithms can fulfil all the main requirements at once.
In conclusion, with the improvement in the spatial resolution of satellite and aerial images, spectral variation in image also increase which can create problems in extracting useful and relevant information. Moreover, forest inventory purposes the forest stands are interpreted as homogenous polygons (objects), which will be difficult to consider under traditional pixel based image analysis. The solution to these problems lies in aggregating pixels into appropriate spatial unit (segment) which can be obtained by image segmentation. Image segmentations are divided into pixel-, edge and region based segmentation methods.
2.3.1 Automatic Segmentation as applied to forestry applications
Object-based analysis and image segmentation techniques have been increasingly applied in fine resolution, multispectral imagery as an alternative to overcome the difficulties of conventional procedures of spectral and texture image analysis for various forestry applications (Hu, Tao & Prenzel 2005).
Some researchers have experimented with the use these techniques for the estimation of individual tree features in a forest. (Wang, Gong & Biging 2004) utilized a combination of spectral classification techniques and segmentation methods for tree-top detection and tree classification in a forested area in British Columbia, Canada.
They estimated a number of 1211 trees per hectare with accuracy of 85% when the results were validated to a manual method of tree counting by visual image interpretations.
(Suárez et al. 2005) used high resolution CHM and aerial photograph to estimate the height of individual tree. A segmentation procedure available in eCognition was used to combine pixels which are similar in terms of elevation and reflectance. The research was able to predict 73% of all the heights within 1 m; 91% within 1.5 m and 96% within 2 m.
(Leckie et al. 2003b)) experimented with High-resolution (60 cm) multispectral airborne imagery and automated tree isolation algorithms in order to delineate tree crowns or clusters of crowns in forest and plantation test area, dominated by young conifer, on the west coast of Canada. An object-oriented single tree classification was conducted using a maximum likelihood classifier. Stands were classified on the basis of similar species composition, closure, and stem density. Species classification was better, with average composition error over all 16 test stands being 7.25%.
The use of remotely sensed data for deriving forest inventory and delineation of stand is not a new area of research. The automated and semi automated segmentation techniques have been increasingly used to delineate forest inventory polygons or stands, in recent years. Automated and computer-assisted interpretation of digital imagery offers a possible solution to extracting more information, reducing time and costs, and increasing consistency. (Wulder et al. 2008) conducted a study with the aim to investigate of an automated segmentation approach for delineating homogeneous forest stands on high spatial resolution satellite imagery, which could subsequently be used to support manual delineation and/or photo interpretation. In this study the automated delineation tool, Size Constrained Region Merging (SCRM), was used to delineate forest stand from IKNOS 1-m panchromatic data. The automatic delineation was then evaluated and compared with manual delineation. However, the SCRM automatic segmentation performed well in most of the situation but in complex areas where there was amalgamation of forest and non forest area the results were not satisfactory. Moreover, coniferous stands of pine and mixed coniferous of pine and spruces were not distinguished satisfactorily, and stands of aspen were frequently merged with non-tree cut block.
Another research presented, a stand delineation method integrating wavelet analysis into image segmentation (Van Coillie, Verbeke & De Wulf 2006). In this study wavelet coefficient and derived statistic, e.g. mean absolute value and standard deviation, were used to discriminate between forest compartments that differ in the above mentioned attributes. This approach was developed using simulated forest stands and was subsequently applied to digital aerial photographs of a forest site (representing a mixture of soft and hardwood stands) in Flanders, Belgium. For evaluation of the method, segmentation base on the image's spectral information was used and it observed that the proposed method was better than the traditional image segmentation method.
In addition to the optical imagery, airborne laser scanning (ALS) data have been increasingly used to delineate forest stand using image segmentation techniques. (Pascual et al. 2008) presented a three-step methodological approach for forest structure characterization. In the first step the laser scanner DCHM (digital canopy height model) was segmented in forest stands; the second step was to cluster these stands into forest structure types based on the LiDAR height summaries; and the final step was to validate the procedure with field data and hypsographs. It was concluded that the best variables for the definition and characterization of forest structure in these forests are the median and standard deviation (S.D.), both derived from LiDAR data.
(Diedershagen, Koch & Weinacker 2004) presented a method of automatically detecting forest stand boundaries. The study dealt with automatic segmentation and characterisation of forest stand units using 3-D information from a combined laser-line scanner system. A normalised DSM (digital surface model) was derived by subtracting DTM (digital terrain model) from DSM to extract stands according to their heights. Mean tree height and crown cover density was also calculated for each stand unit. It was observed that in some cases algorithm worked well but in many cases the algorithm did not divide the stand in the same way a human interpreter would.
However, the quality and accuracy of automatic forest stand delineation greatly depend on the structure of the forest under consideration. Heterogeneous forests are most likely susceptible to errors and inaccuracies. Forests with same tree species but different ranges of vertical canopy structure may result in inaccurate stand delineation. Similarly neighbouring stands of different tree species but same height may be not delineated accurately. The algorithms used for automatic stand delineation are less capable of dividing forest stands on the basis of their heights (Diedershagen, Koch & Weinacker 2004). Similarly, Wulder et al. 2008 experimented with automated segmentation of high resolution IKONOS 1-m panchromatic imagery to delineate forest stands. The results were satisfactory in homogenous forest but in complex area, where there were amalgamation forest and non forest area, the results were not promising (Wulder et al. 2008).
The results of forest stand delineation can be improved by considering heights and as well as spectral characteristics of the forest stands as well. There is a high synergy between the VRH multispectral data and laser scanner data for the pulling out of information about forest. Laser data supplies precise height information, which hand is absent in single optical image. It also facilitates to provide details on crown shape. On other hand multispectral images provide information about spatial geometry and colour information of tree species (Hyyppä et al. 2004). The use of very high resolution (VHR) multispectral data in combination with LiDAR will also help in greatly improving the classification of stand on the basis of tree species for stand delineation.
Baatz, M. & Schäpe, A. 2000, "Multiresolution Segmentation-an optimization approach for high quality multi-scale image segmentation", Angewandte geographische informationsverarbeitung, vol. 12, pp. 12-23.
Blaschke, T., Burnett, C. & Pekkarinen, A. 2004, "Image segmentation methods for object-based analysis and classification", Edited by SM de Jong,.and FD van der Meer.Remote Sensing Image Analysis, , pp. 211-234.
Brandtberg, T. 2002, "Individual tree-based species classification in high spatial resolution aerial images of forests using fuzzy sets", Fuzzy Sets and Systems, vol. 132, no. 3, pp. 371-387.
Chubey, M.S., Franklin, S.E. & Wulder, M.A. 2006, "Object-based analysis of lkonos-2 imagery for extraction of forest inventory parameters", Photogrammetric Engineering and Remote Sensing, vol. 72, no. 4, pp. 383-394.
Diedershagen, O., Koch, B. & Weinacker, H. 2004, "Automatic segmentation and characterisation of forest stand parameters using airborne lidar data, multispectral and fogis data", Laser-Scanners for Forest and Landscape Assessment.Institute for Forest Growth Institute for Remote Sensing and Landscape Information Systems.Albert Ludwigs University Freiburg, Germany, , pp. 208-212.
Evans, D.L., Roberts, S.D. & Parker, R.C. 2006, "LiDAR-A new tool for forest measurements?", Forestry Chronicle, vol. 82, no. 2, pp. 211-218.
Förster, M. & Kleinschmit, B. 2008, "Object-based classification of QuickBird data using ancillary information for the detection of forest types and NATURA 2000 habitats", Object-Based Image Analysis.Springer.Berlin, , pp. 285-300.
Franklin, S.E. 2001, Remote sensing for sustainable forest management, CRC Press.
Giannetti, F., Gottero, F. & Terzuolo, P.G. 2003, "Use of high resolution satellite images in the forest inventory and mapping of Piemonte region (Italy) ", Advances in Forest Inventory for Sustainable Forest Management and Biodiversity Monitoring, .
Gorte, B. 1998, "Probabilistic segmentation of remotely sensed images", ITC, publication, vol. 63.
Hall, R.J. 2003, "The roles of aerial photographs in forestry remote sensing image analysis", Remote sensing of forest environments: concepts and case studies, , pp. 47-75.
Haralick, R. & Shapiro, L. 1985, "Image segmentation techniques.", APPLICATIONS OF ARTIFICIAL INTELLIGENCE II., 1985, vol. 548, pp. 2-9.
Hay, G., Marceau, D., Dube, P. & Bouchard, A. 2001, "A multiscale framework for landscape analysis: Object-specific analysis and upscaling", Landscape Ecology, vol. 16, no. 6, pp. 471-490.
Hilker, T., Wulder, M.A. & Coops, N.C. 2008, "Update of forest inventory data with lidar and high spatial resolution satellite imagery", Can.J.Remote Sensing, vol. 34, no. 1, pp. 5-12.
Holmgren, J., Persson, A. & Sodermann, U. 2006, "Identification of tree species of individual trees by combining very high resolution laser data with multi-spectral images", Workshop on 3D Remote Sensing in Forestry, Vienna.
Holmgren, J. 2004, "Prediction of tree height, basal area and stem volume in forest stands using airborne laser scanning", Scandinavian Journal of Forest Research, vol. 19, no. 6, pp. 543-553.
Hu, X., Tao, C.V. & Prenzel, B. 2005, "Automatic segmentation of high-resolution satellite imagery by integrating texture, intensity, and color features", Photogrammetric Engineering and Remote Sensing, vol. 71, no. 12, pp. 1399-1406.
Hyyppä, H. & Hyyppä, J. 1999, "Comparing the accuracy of laser scanner with other optical remote sensing data sources for stand attribute retrieval", The Photogrammetric Journal of Finland, vol. 16, no. 2, pp. 5-15.
Hyyppä, J., Hyyppä, H., Litkey, P., Yu, X., Haggrén, H., Rönnholm, P., Pyysalo, U., Pitkänen, J. & Maltamo, M. 2004, "Algorithms and methods of airborne laser-scanning for forest measurements", International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. 36, no. Part 8, pp. 1682-1750.
Hyyppä, J. & Inkinen, M. 1999a, "Detecting and estimating attributes for single trees using laser scanner", The photogrammetric journal of Finland, vol. 16, no. 2, pp. 27-42.
Hyyppä, J. & Inkinen, M. 1999b, "Detecting and estimating attributes for single trees using laser scanner", The photogrammetric journal of Finland, vol. 16, no. 2, pp. 27-42.
Hyyppa, J., Hyyppa, H., Inkinen, M., Schardt, M. & Ziegler, M. 2000, "Forest inventory based on laser scanning and aerial photography", , pp. 106.
Hyyppae, J.M., Hyyppae, H., Inkinen, M., Schardt, M. & Ziegler, M. 2000, "Forest inventory based on laser scanning and aerial photography", Proceedings of SPIE, pp. 106.
Laliberte, A.S., Rango, A., Havstad, K.M., Paris, J.F., Beck, R.F., McNeely, R. & Gonzalez, A.L. 2004, "Object-oriented image analysis for mapping shrub encroachment from 1937 to 2003 in southern New Mexico", Remote Sensing of Environment, vol. 93, no. 1-2, pp. 198-210.
Lamonaca, A., Corona, P. & Barbati, A. 2008, "Exploring forest structural complexity by multi-scale segmentation of VHR imagery", Remote Sensing of Environment, .
Leckie, D., Gougeon, F., Hill, D., Quinn, R., Armstrong, L. & Shreenan, R. 2003a, "Combined high-density lidar and multispectral imagery for individual tree crown analysis", Canadian Journal of Remote Sensing, vol. 29, no. 5, pp. 633-649.
Leckie, D., Gougeon, F., Walsworth, N. & Paradine, D. 2003b, "Stand delineation and composition estimation using semi-automated individual tree crown analysis", Remote Sensing of Environment, vol. 85, no. 3, pp. 355-369.
Maltamo, M., Eerikäinen, K., Pitkänen, J., Hyyppä, J. & Vehmas, M. 2004a, "Estimation of timber volume and stem density based on scanning laser altimetry and expected tree size distribution functions", Remote Sensing of Environment, vol. 90, no. 3, pp. 319-330.
Maltamo, M., Mustonen, K., Hyyppa, J., Pitkanen, J. & Yu, X. 2004b, "The accuracy of estimating individual tree variables with airborne laser scanning in a boreal nature reserve", Canadian Journal of Forest Research, vol. 34, no. 9, pp. 1791-1801.
Maltamo, M., Packalén, P., Yu, X., Eerikäinen, K., Hyyppä, J. & Pitkänen, J. 2005, "Identifying and quantifying structural characteristics of heterogeneous boreal forests using laser scanner data", Forest Ecology and Management, vol. 216, no. 1-3, pp. 41-50.
Molenaar, M. 1998, An introduction to the theory of spatial object modelling for GIS, CRC.
Mustonen, J., Packalén, P. & Kangas, A. 2008, "Automatic segmentation of forest stands using a canopy height model and aerial photography", Scandinavian Journal of Forest Research, vol. 23, no. 6, pp. 534-545.
Næset, E. 1999, "Effects of delineation errors in forest stand boundaries on estimated area and timber volumes", Scandinavian Journal of Forest Research, vol. 14, no. 6, pp. 558-566.
Næsset, E. 2004, "Practical large-scale forest stand inventory using a small-footprint airborne scanning laser", Scandinavian Journal of Forest Research, vol. 19, no. 2, pp. 164-179.
Næsset, E., Gobakken, T., Holmgren, J., Hyyppä, H., Hyyppa, J., Maltamo, M., Nilsson, M., Olsson, H., Persson, A. & Söderman, U. 2004, "Laser Scanning of Forest Resources: The Nordic Experience", Scandinavian Journal of Forest Research, vol. 19, no. 6, pp. 482-499.
Næsset, E. 1997a, "Determination of mean tree height of forest stands using airborne laser scanner data", ISPRS Journal of Photogrammetry and Remote Sensing, vol. 52, no. 2, pp. 49-56.
Næsset, E. 1997b, "Estimating timber volume of forest stands using airborne laser scanner data", Remote Sensing of Environment, vol. 61, no. 2, pp. 246-253.
Pascual, C., García-Abril, A., García-Montero, L., Martín-Fernández, S. & Cohen, W. 2008, "Object-based semi-automatic approach for forest structure characterization using lidar data in heterogeneous Pinus sylvestris stands", Forest Ecology and Management, .
Persson, A., Holmgren, J. & SÖDERMAN, U. 2002, "Detecting and measuring individual trees using an airborne laser scanner", Photogrammetric Engineering and Remote Sensing, vol. 68, no. 9, pp. 925-932.
Schneider, W. & Steinwendner, J. 1999, "Landcover mapping by interrelated segmentation and classification of satellite images", International Archives of Photogrammetry and Remote Sensing, vol. 32, no. Part 7, pp. 4-3.
Shiba, M. & Itaya, A. 2006, "Using eCognition for Improved forest Management and monitoring Systems in Precision Forestry", Ackerman, P.A, Längin, DW and Antonides, MC (eds.), .
Skidmore, A. 1989, "An expert system classifies eucalypt forest types using thematic mapper data and a digital terrain model", Photogrammetric engineering and remote sensing (USA), .
Suárez, J.C., Ontiveros, C., Smith, S. & Snape, S. 2005, "Use of airborne LiDAR and aerial photography in the estimation of individual tree heights in forestry", Computers and Geosciences, vol. 31, no. 2, pp. 253-262.
Van Coillie, F.M.B., Verbeke, L.P.C. & De Wulf, R.R. 2006, "SEMI-AUTOMATED FOREST STAND DELINEATION USING WAVELET-BASED SEGMENTATION OF VERY HIGH RESOLUTION OPTICAL IMAGERY IN FLANDERS, BELGIUM", 1st International Conference on Objectbased Image Analysis (OBIA 2006).
Wang, L., Gong, P. & Biging, G. 2004, "Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery.", PE&RS, Photogrammetric Engineering & Remote Sensing, vol. 70, no. 3, pp. 351-357.
Wulder, M., White, J., Hay, G. & Castilla, G. 2008, "Towards automated segmentation of forest inventory polygons on high spatial resolution satellite imagery", Forestry Chronicle, vol. 84, no. 2, pp. 221-230.
