Estimating plot-level tree height and volume of

Article

Estimating plot-level tree height and volume of Eucalyptus grandis plantations using small-footprint, discrete return lidar data

Progress in Physical Geography 34(4) 515–540 ª The Author(s) 2010 Reprints and permission: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0309133310365596 ppg.sagepub.com

S.G. Tesfamichael University of KwaZulu-Natal, South Africa

J.A.N. van Aardt Rochester Institute of Technology, USA

F. Ahmed University of KwaZulu-Natal, South Africa

Abstract This study explores the utility of small-footprint, discrete return lidar data in deriving important forest structural attributes with the primary objective of estimating plot-level mean tree height, dominant height, and volume of Eucalyptus grandis plantations. The secondary objectives of the study were related to investigating the effect of lidar point densities (1 point/m2, 3 points/m2, and 5 points/m2) on height and volume estimates. Tree tops were located by applying local maxima (LM) filtering to canopy height surfaces created at each density level, followed by buffering using circular polygons. Maximum and mean height values of the original lidar points falling within each tree polygon were used to generate lidar mean and dominant heights. Lidar mean value was superior to the maximum lidar value approach in estimating mean plot height (R2*0.95; RMSE*7%), while the maximum height approach resulted in superior estimates for dominant plot height (R2*0.95; RMSE*5%). These observations were similar across all lidar point density levels. Plot-level volume was calculated using approaches based on lidar-derived height variables and stems per hectare, as well as stand age. The level of association between estimated and observed volume was relatively high (R2¼0.82–0.94) with non-significant differences among estimates at high lidar point densities and field observation. Nearly all estimates, however, exhibited negative biases and RMSE ranging in the order of 20–43%. Overall, the results of the study demonstrate the potential of lidar-based approaches for forest structural assessment in commercial plantations, even though further research is required on improving stems per hectare (SPHA) estimation. Keywords Eucalyptus grandis, plot-level tree height, plot-level tree volume, small-footprint lidar

I Introduction Forest structural assessment is an important component in timber resources management (von Gadow and Bredenkamp, 1992). In

Corresponding author: S.G. Tesfamichael, School of Environmental Sciences, Howard College Campus, University of KwaZulu-Natal, Durban 4041, South Africa Email: [email protected]

516

addition, it has increasingly become pressing to monitor forest resources with the aim of tackling problems related to ecosystem dynamics (Brown, 2002). There is, for example, the need for sound management of South African forestry sector, which contributes significantly to the hydrological processes of catchments through stream flow reduction (Le Maitre et al., 2002; Louw and Scholes, 2002; Dye and Versfeld, 2007). Traditional assessment methods employ very expensive and time-consuming manual inventories that involve field visits by personnel. These surveys often cover a portion of forest population and are designed according to specific sampling strategies (Schreuder et al., 1993; Avery and Burkhart, 2001). The required accuracy, spatial extent, cost, and time (among others) are, in turn, critical considerations when selecting an appropriate sample size. The introduction of remote sensing technologies in forestry has been crucial in addressing the challenges of manual inventory by providing a synoptic view of often extensive areas with continuous spatial coverage, including places that are difficult to access during field visits (Wulder, 1998; Boyd and Danson, 2005). Optical imagery is the earliest remote sensing technology employed in forestry (Boyd and Danson, 2005). Advances of the technology over the years have resulted in significant improvements in spatial and spectral resolutions that are suitable for detailed information extraction (Wulder, 1998; Boyd and Danson, 2005). This, coupled with developments in sophisticated machine computing capabilities, has enabled the characterization of forest attributes at even the tree level (eg, Gougeon, 1995; Dralle and Rudemo, 1996; Brandtberg and Walter, 1998; Pollock, 1998; Wulder et al., 2000; Pouliot et al., 2002; Wang et al., 2004; Chubey et al., 2006). Spectral, spatial, and textural information are important features of imagery in structural characterizations (Wulder, 1998). Structural attributes derived from these features are mostly related to the horizontal dimension of forests.

Progress in Physical Geography 34(4)

Specific examples of such attributes include forest area holding (Wynne et al., 2000; Chubey et al., 2006), crown cover (Wu and Strahler, 1994; Ozdemir, 2008; Song and Dickinson, 2008), leaf area index (McAllister and Valeo, 2007; Song and Dickinson, 2008), and stand density (Wu and Strahler, 1994; Franco-Lopez et al., 2001). Attributes that require inputs from the vertical dimension, such as height, volume, and biomass, are often inferred empirically from attributes with horizontal features (eg, Ma¨kela¨ and Pekkarinen, 2001; Hall et al., 2006; Chubey et al., 2006; Ozdemir, 2008). Ozdemir (2008), for example, related field-measured stem volume with crown area and shadow area derived from Quickbird imagery in open juniper forest stands. The study illustrated that the approach was unable to account for stem size and tree height, two attributes that are obscured from top view by crown. Chubey et al. (2006), on the other hand, used a decision tree analysis to classify forest areas according to land-cover types, forest species composition, crown closure, stand age, and stand height. Overall accuracies of results for all classifications except height exceeded 80%. Height classification, which focused on pine forests, based on reflectance from red and near-infrared bands returned a rather poor accuracy (49%) and led to the conclusion that height estimation from imagery is unreliable. Another common shortcoming in the application of optical remote sensing to forest inventories relates to the saturation of information embedded in the imagery at high levels of biomass or leaf area index (Sader et al., 1989; Nemani et al., 1993; Foody and Curran, 1994; Chen and Cihlar, 1996; Wulder, 1998; Turner et al., 1999; Steininger, 2000). This is attributed to the fact that optical imagery captures mainly the horizontal features of a canopy and is virtually insensitive to the increase in the vertical structural quantity (and composition) once the canopy is closed (Wulder, 1998). Turner et al.

Tesfamichael et al.

(1999) observed a decrease in sensitivity of spectral vegetation indices derived from Landsat 5 imagery to leaf area index (LAI) exceeding 5 in temperate vegetation covers. Steininger (2000) found that the relationship between stand reflectance derived from Landsat imagery and biomass saturated at approximately 15 kg m2 or over 15 years of stand age. Although high spatial resolution imagery (eg, IKONOS, Quickbird) is better equipped to minimize saturation, the problem may still persist at very high biomass levels (Leboeuf et al., 2007). A relatively new generation of remote sensing technology that has shown great promise to overcome the shortcomings of conventional optical imagery is lidar (light detection and ranging). Lidar sensors acquire three-dimensional (3-D) information of objects by emitting dense laser pulses and recording the reflected signals from the objects (Wehr and Lohr, 1999). This characteristic enables such sensors to capture a large volume of information about the underlying ground surface (Ackermann, 1999) and has attracted a rapidly growing interest within the forestry community (Lefsky et al., 2002; Lim et al., 2003; Reutebuch et al., 2005; Hyyppa¨ et al., 2008). A plethora of lidar applications in forestry has been presented in the literature. Forest classification, for example, is one process that has traditionally been performed by spectral and/or textural information of optical imagery but can also be done using lidar data. Zimble et al. (2003) exploited the vertical distribution of lidar points and classified forests as single-story and multi-story with an accuracy level of 97% compared to the field-based classification. A similar accuracy level was reported by Falkowski et al. (2009) who demonstrated the potential of lidar data and a non-parametric classification algorithm, namely Random Forest, to classify a structurally diverse and mixed-species coniferous forest into six succession stages ranging from open land to old multi-story forest. Classification of species can also be performed at

517

individual tree level, given sufficient information acquired from individual trees (eg, Brandtberg, 2007; Ørka et al., 2009). Jensen et al. (2008) estimated LAI using various lidar distributional statistics such as height percentiles and canopy cover metrics in coniferous forests with accuracies in R2 reaching up to 0.86. The authors found that inclusion of spectral vegetation indices from SPOT 5 did not improve LAI estimation significantly. Thomas et al. (2008) compared hyperspectral and lidar height percentiles in the estimation of chlorophyll and carotenoid concentrations in a boreal forest and found the lidar data to be superior, indicating the direct relationship between canopy structure and the biochemical concentrations. Lidar data can also be extended to characterize forest structural features in support of various management efforts of ecological importance. Andersen et al. (2005) applied lidar data and regression analysis to derive crown fuel weight (R2¼0.86), crown bulk density (R2¼0.84), canopy base height (R2¼0.77), and canopy height (R2¼0.98) for use in fire behavior prediction models. A number of studies were also conducted to explore the potential of lidar in wildlife habitat assessment (eg, Goetz et al., 2007; Graf et al., 2009; Mu¨ller and Brandl, 2009). Accurate assessment of timber and/or biomass resources in a range of forest environments using lidar data is well documented. In this regard, the technology has achieved a high level of accuracies relative to other remote sensing methods (Lefsky et al., 2001; McCombs et al., 2003; Coops et al., 2004) as well as field inventories (eg, Hyyppa¨ et al., 2001; Coops et al., 2004; Holmgren, 2004; Næsset, 2004; Maltamo et al., 2006). Most studies have followed area-based analysis to estimate structural attributes where multiple trees are treated as a unit. This is basically the most appropriate approach for lidar systems that have large-footprint laser returns (eg, Means et al., 1999; Lefsky et al., 1999; Drake et al., 2002). Numerous studies involving

518

small-footprint discrete laser returns have also adopted a similar approach in terms of spatial scale (Lefsky et al., 2001; Næsset, 2002; 2004; 2007; Holmgren, 2004; Hall et al., 2005; Thomas et al., 2006; van Aardt et al., 2006). Typically, this approach assesses structural attributes using lidar distributional (eg, maximum, mean, coefficient of variation, and percentiles of canopy height, as well as canopy density distributions) metrics extracted either from point data or interpolated canopy surface. Næsset (2004) extracted several lidar metrics from lidar data at the plot level and regressed against a number of field inventory structural attributes in coniferous forests. A stepwise regression analysis returned models that were able to explain 60–97% of the variability present in field values of the attributes. Hall et al. (2005) applied an information-theoretic approach, namely Akaike’s Information Criterion (AIC), to relate stand-level structural attributes with lidar-derived metrics and found R2 ranging between 0.57 and 0.87. van Aardt et al. (2006) classified coniferous, deciduous, and mixed forests into homogenous and hierarchical objects using canopy height surface derived from discrete return lidar data. Lidar height and intensity distributional metrics were then extracted for each object and correlated to field volume and biomass using stepwise regression analysis (adjusted R2¼0.58–0.79). Area-based analysis is useful for large area inventories, as is applied routinely in Scandinavia (a summary is given in Næsset, 2007). Other advantages include easy integration with traditional forest inventory methods due mainly to similarity in support scale (Hyyppa¨ et al., 2008) and the insensitivity to lidar point density (Reutebuch et al., 2005; Thomas et al., 2006; Hyyppa¨ et al., 2008). A notable drawback of the area-based approach is the requirement for robust models to serve as prediction tools (Hyyppa¨ et al., 2008). The cost of developing such models is directly related to the spatial coverage of the


forest under consideration due to the necessity for sufficient sample size to represent the variability present in the population. In addition, the high level of accuracy required in commercial forestry may require model development that addresses site-specific characteristics (Kato et al., 2009; Zhao et al., 2009). This, in turn, is compounded when there is a great deal of fragmentation and associated site quality variations, as is prevalent in the context of South African forestry landholdings. Models can also be scale-dependent, as shown by Patenaude et al. (2004), who used a model to predict aboveground biomass and found different results dependent on whether an area was considered as a single unit or an aggregate of subsets. Furthermore, area-based expressions of forest structures often lack information on individual trees and thus do not contribute to management approaches that require assessment of individual trees (Bortolot and Wynne, 2005; Koch et al., 2006). In comparison, individual tree-based analysis is often preferable due to the detail of information extracted and the low amount of field reference data required (Hyyppa¨ et al., 2008). It is also advantageous in that scalability to larger spatial coverage requires simple aggregation of tree-level measurements, provided suitable tree detection accuracy (Zhao et al., 2009). It is therefore logical that increased interest is being shown in the research community to design methods that are capable of extracting lidar information about individual trees (Hyyppa¨ et al., 2001; Persson et al., 2002; Popescu et al., 2002; Brandtberg et al., 2003; Morsdorf et al., 2004; Bortolot and Wynne, 2005; Roberts et al., 2005; Chen et al., 2007; Kato et al., 2009; Koch et al., 2009). Popescu et al. (2003) used a variable window size and local maxima filtering to locate and segment crowns of individual trees in stands of deciduous, coniferous, and mixed species. Stepwise regression was used to estimate field-measured crown diameter from lidar-derived height, diameter of crown

Tesfamichael et al.

segments, and the number of trees, with the best models being similar for both deciduous and coniferous trees (R2¼0.62–0.63). The best models for estimating plot-level tree volume and biomass also included crown diameter as independent variable. The corresponding R2 values for coniferous trees were 0.83 and 0.78, while poorer results were obtained for deciduous species. Holmgren et al. (2003) derived tree height and stem volume from lidar data using different approaches for forests dominated by Norway spruce, Scots pine, and birch. Tree height was derived at the plot and single tree levels (local maxima filtering) with results showing strong similarities between the two methods (R2¼0.89– 0.91; RMSE¼10–11%). Height assessment was followed by computation of volume. The first method utilized height derived using the areabased approach and canopy area, which was assumed to inherit the generally good correlation between total crown area and basal area. The second method used the number of stems and height derived from lidar data as inputs. The first method (area-based approach) returned superior results (R2¼0.90; RMSE¼22%) to the second approach (R2¼0.82; RMSE¼26%). Maltamo et al. (2004) derived stem counts and volume from lidarderived height and diameter at breast height (DBH) for similar species studied by Holmgren et al. (2003). Estimation errors in RMSE were reported to be 25% and 75% for volume and tree count, respectively. The use of a Weibull distribution to account for small trees decreased the corresponding values to 16% and 49%. Heurich (2008) estimated tree-level attributes from laser-derived height and crown radius for coniferous and deciduous trees. Watershed segmentation was used to identify trees before applying regression analysis to estimate tree height (R2¼0.97–0.98, RMSE¼4–6%), crown radius (R2¼0.45–0.56, RMSE¼11-19%), DBH (R2¼0.79–0.92, RMSE¼17–23%), and volume (R2¼0.87–0.95, RMSE¼49–70%). The studies cited here showed that modeling efforts are still incorporated at some stage of the

519

analysis. For example, Maltamo et al. (2004) relied on DBH models that are calibrated using field inventory data, while Holmgren et al. (2003) used regressed field volume with lidarderived variables. Chen et al. (2006) argued that combining area-based and individual tree-based analysis constitutes an acceptable compromise. The manner in which such a compromise is achieved is critical and should focus on the cost-benefit aspect of the approach. This may hint at the need for model development at a lower sampling intensity than the conventional inventory, a notion that has yet to be realized in the forestry industry. In the mean time, more research is needed to further minimize the reliance on models that require continual training or calibration. Additionally, an increased emphasis should be placed on translating current research outputs into operational applications. This in particular has an important significance for commercial forest industries which essentially perform pre-harvest structural assessments for monetary valuation purposes. In a rare example, Peuhkurinen et al. (2007) demonstrated such an assessment using lidar data and fielddeveloped models in forest stands dominated by Norway spruce. The study compared information derived from lidar canopy height with other methods, including systematic plot sampling and inventory by compartments. An actual harvest inventory was used as reference data. Comparisons of the methods based on diameter distributions, volumes, number of trees, and bucking simulations showed superior results for the lidar-based approach. This study extends the efforts of the above studies to retrieve important forest structural variables in tropical plantation environments using lidar data and limited inputs from standard management (field) information. The inputs from field information do not require traditional field visits, but are available in most commercial plantation management as documentation compiled at the time of planting. As such we purposely avoided data inputs that require field-based

520


measurements. Intensive management of the plantations targeted in this study is believed to provide the necessary platform to investigate the success of our proposal. The specific objectives of this study were to: (1)

(2)

(3)

evaluate the utility of discrete return, small-footprint lidar data for derivation of accurate plot-level mean and dominant tree height; assess the usefulness of lidar-derived tree height and stems per hectare (SPHA) for estimation of plot-level volume; and explore the effect of simulated lidar point densities on the estimation of tree height and volume.

Potential trees were first identified using local maxima filtering in order to achieve these objectives. Although not stated as an objective in this study, the local maxima filtering compared two methods of window size determination, namely the semi-variogram range and known tree spacing of the plantations, as recorded at the time of planting. As a result, the effects of these methods on variable estimation also will be presented. This was followed by computing height of individual trees using a suitable estimator. Stand density and estimated height were then used to calculate plot-level volume.

II Methods 1 Study area and field data The study area is located near the town of Richmond in the municipality of Sisonke, KwaZuluNatal, South Africa (Figure 1). The geographical extent of the area coincident with the lidar survey is bounded by 30 110 4100 E to 30 150 4800 E and 29 480 5400 S to 29 510 1100 S. The study area falls within the summer rainfall region of South Africa and experiences cold dry winters and warm wet summers. Mean annual rainfall ranges from 746 to 1100 mm while temperatures vary

between a high of 20 C to below 10 C (Schulze, 1997). Plantation forestry dominates the land use in the area, with species from the Eucalyptus and Pinus genera primarily grown. The dominance of these genera in the area reflects the figure at the regional and national levels (Godsmark, 2009). The increase in areal coverage of the Eucalyptus genus, in particular, has been the largest over the last two decades. Most of this increase is observed in the province of KwaZulu-Natal where the study area is situated (Godsmark, 2009). Eucalyptus grandis is the most widely grown species of the genus due to favorable climatic factors in the area as well as the economic importance of the species (Rietz and Smith, 2004). The present study has therefore focused on E. grandis. The main purpose of Eucalyptus plantations in the country is to provide raw material for pulp production, which accounts for approximately 80% of the total harvest, while timber, sawlogs, poles, and other end products make up the remaining purposes (Godsmark, 2009). The high demand for these products necessitates a shortrotation cropping system of the plantations (Scho¨nau and Boden, 1982; Clarke and Wessels, 1995; Brown et al., 1997; Gonçalves et al., 1997; Louw and Scholes, 2002). Typical harvest age in South Africa ranges between six and 11 years (Coetzee, 1999; Rietz and Smith, 2004). The age of the trees for this study ranged from four to nine years, a range deemed representative of the population. A total of 19 stands were selected. Field data were collected during spring (October) of 2006. Plots were sampled following a standard protocol practiced by the forest industry for inventory of structural attributes (Esler, 2004). The protocol specifies that the sample size in terms of area should be approximately 5% of a given stand. Plots (*10 m radius) are selected systematically as nodes of a 50 m by 100 m grid covering the stand and thereby capturing as much variability within the compartment as possible. Such systematic sampling is

Tesfamichael et al.

521

Figure 1. Map of the study area located in KwaZulu-Natal, South Africa

a standard protocol followed in different regions (eg, Holmgren et al., 2003; Næsset, 2004). In order to avoid subjectivity, we laid the grids on existing maps of stand boundaries before the field visits. A total of 63 nodes representing plots were sampled in 19 selected stands. Differential global positioning system (DGPS) at a submeter accuracy level was used to mark an open space at the edge of a stand and nearest to each sample plot. Distance and bearing was then used to locate the node. The nearest tree to the node was used as the center of a plot. Instead of a 10 m radius, a 15 m radius was specified for this study in order to include more trees. Plot size modification was done to strengthen the statistical reliability of the results by including more trees than would be found in a 10 m radius plot. Furthermore, we believed that a larger plot

would compensate for an offset due to error of locating a plot center. Plot area was then adjusted for slope, since this has an implication on the conversion of tree count to stems per hectare (SPHA) and volume (von Gadow and Bredenkamp, 1992). Spacing between and within rows of the stands were approximately 3 and 2.4 m, respectively. Such spacing is recommended in intensive silvicultural plantations of Eucalyptus grandis (Pallett and Sale, 2004). This, in combination with homogeneity of stands, limits the variation in crown diameter as can be observed visually in the field. An assumption was therefore made that 2.4 m would approximate the average crown diameter of a tree for the study area. Only trees with a DBH greater than 5 cm were considered for enumeration within each plot.

522


Table 1. Summary of field survey data Variable

Mean + standard deviation

Minimum

Maximum

Mean height (m) Dominant height (m) Stand density (trees/ha) DBH (cm) Volume (m3/ha)

21.5 + 4.2 25.2 + 5.8 1024 + 211 16.9 + 3.1 278.5 + 136.7

15.2 14.5 529 12.2 95.1

31.6 40.8 1504 24.8 597.9

This is commensurate to inventory protocol of the industry managing the plantations and has also been followed by other studies (eg, Holmgren, 2004; Maltamo et al., 2004). Trees that appeared to be damaged, dead, or dying were excluded from the counting process for two main reasons. First, the commercial industry rarely includes such trees in an accounting scheme and, second, the number of lidar returns from trees devoid of canopies was deemed less significant. Tree counts in each plot were then converted to SPHA based on the area of the respective plot. Tree height was measured on a subsample of trees selected across the range of DBH values. Relationships between DBH and corresponding height were established at plot level using regression analysis with coefficients of determination (R2) exceeding 80% for the majority of plots. Heights of all trees within a plot were estimated using the resulting regression equations. Subsequently, two area-based definitions of height were derived, namely mean height and dominant height. Mean height was calculated as the arithmetic mean of all trees. Dominant height can be defined as the mean height of a selected number of trees per hectare (Garcıá, 1998). The selection of these trees could be based on DBH or height. Dominant height was derived as the mean of the top 20% tallest trees, as per the forest mensuration protocol of the forestry company that manages the study area. Tree height and DBH are the two most important variables explaining the variability in tree volume (Avery and Burkhart, 2001). Thus, when these variables are known for individual trees,

volume can be calculated for each tree. The Schumacher and Hall (1933) stem volume function is used in this study (equation 1): ln Vt ¼ b0 þ b1 lnðDBHÞ þ b2 ln HT ð1Þ where Vt is volume of the mean tree, DBH and HT are diameter at breast height and mean tree height, respectively, and b0–b2 are coefficients fit for Eucalyptus grandis (Coetzee, 1992). Plot-level volume was derived by adding volume of individual trees. The aggregate volume was then converted to a hectare scale based on the area of a plot. Summaries of the field inventory and analysis data are presented in Table 1.

2 Lidar data A lidar survey of the study area was carried out by Airborne Laser Solutions (ALS, www.alsafrica.com) on 8 November 2006, using the Optech ALTM 3033 sensor. The system uses discretereturn near infrared laser pulses (wavelength 1064 nm) with a pulse rate of 33 kHz and was configured to record first and last returns per pulse. A flying height of 550 m above the ground, coupled with a laser beam divergence of 0.2 mrad and associated footprint size of 0.2 m, resulted in a mean nominal point density of five points per square meter (points/m2). The vertical and horizontal accuracies of the height returns were 0.15 m and 0.28 m, respectively. The data provider classified the points as ground and non-ground using Terrasolid’s TerraScan1 software (Terrasolid Ltd, Helsinki, Finland). The software first searches and

Tesfamichael et al.

removes low and isolated points. Low points (below ground) are identified by routinely comparing the elevation of each point called a center point with its nearest point in the xy plane. If the center point is lower than the other, it is considered as a low point. Isolated points refer to points that do not fall within a neighborhood specified by a 3-D radius. Following filtering, an initial triangulated irregular network (TIN) is built using local low points that are assumed to be ground hits. These triangles fall mostly below the ground with the vertices touching the ground. Points are then iteratively added to shape the triangle upwards resulting in a surface that closely resembles a ground. Non-ground points are classified by selecting points that are at a specified height from a triangulated surface built from the final ground points. The data provider of this study reported that this process involved an initial automated classification using the software, which attained accuracies between 60 and 80% depending on terrain variation. This was followed by a manual editing process aided by 0.15 m resolution orthophoto images. Accuracy of the final digital terrain model (DTM) was assessed using ground control points. Based on this, a mean error for the DTM was reported as þ0.031 m (range: – 0.017 to þ0.078 m).

3 Tree identification at different lidar point densities A parallel study (Tesfamichael et al., 2009) was dedicated to the analysis of local maxima filtering and the estimation of SPHA using the same data set. A brief summary is presented in the following subsections. a Generating the canopy height model. A standlevel digital elevation model (DEM) was created from the lidar ground return layer. Since original lidar points were irregularly spaced, a continuous surface was created by estimating values for non-sampled areas using the ordinary

523

kriging interpolation approach (Oliver, 1990) in Surfer 8.0 (Golden Software, Inc., Golden, CO, USA). A pixel size of 1 m was used in the interpolation, with at most one point used for generating a pixel height value. Whenever two or more points occurred within a pixel, the minimum point was utilized for the interpolation. The selection of these points was based on the assumption that they theoretically were closest to the ‘true’ ground value for that pixel. Canopy return heights subsequently were derived by subtracting DEM values from non-ground lidar returns. It was felt that point-based normalization before interpolation reduces the error levels of the final canopy height model (CHM). These points were then used to generate a CHM using ordinary kriging interpolation. An alternative approach to create the CHM would be generating separate surface models representing the terrain and canopy surfaces and differencing the two surfaces (Maltamo et al., 2004; Hyyppa¨ et al., 2008). Stand-level CHMs were generated at 0.2 m, 0.5 m, and 1 m spatial resolutions, after which a subset was extracted for each plot for further plot-level analysis (Tesfamichael et al., 2009). The maximum return height was selected whenever more than one lidar return fell within a pixel, a method also followed by Hyyppa¨ et al. (2001), Koch et al. (2009), and Zhao et al. (2009), among others. The selection of maximum returns increases the likelihood that the resulting CHM will represent the upper shape of the canopy for a given plot. A CHM is used to locate potential tree locations by identifying a pixel with maximum height within a neighborhood (Popescu et al., 2002; McCombs et al., 2003; Maltamo et al., 2004). Certain studies have shown that the application of a smoothing filter to the CHM, prior to local maxima filtering, improves the distinctiveness of trees (eg, Hyyppa¨ et al., 2001; Persson et al., 2002; Maltamo et al., 2004; Koch et al., 2006). However, smoothing, evaluated for a sample of plots, reduced the detection rate considerably; this was

524


Table 2. Comparison of observed and estimated SPHA using local maxima (LM) filtering. Figures in parentheses refer to percent accuracy relative to field observed Local maxima filtering window Lidar density (points/m2) 1 3 5 Field

Variogram range* a

566 (56%) 744b (70%) 800b (73%)

Tree spacing* 799b (74%) 932c (80%) 997cd (82%) 1024d

*Different characters indicate significant differences at a¼0.05.

attributed to the loss of variability in height for closely spaced trees (Koch et al., 2006; Reitberger et al., 2009). As a result, smoothing was not applied here. b Specifying a local maxima filtering window size. Specification of the filter size is an important step in local maxima filtering (Wulder et al., 2000; Popescu et al., 2002). Filter size was determined using two methods, namely the semi-variogram range and known tree spacing (Tesfamichael et al., 2009). In the first method, an omnidirectional semi-variogram analysis was performed at each spatial resolution. The semi-variogram range has been shown to provide an approximate estimate of tree crown diameter based on multispectral imagery (eg, Woodcock et al., 1988a; 1988b; Wulder et al., 2000). It was proposed that the geometry of the upper CHM would prove useful to semivariogram derivation. Since the target of the study is an even-aged, monoculture plantation, the resulting range value can be used as the average canopy diameter of a tree for a plot and consequently the filter size in the local maxima filtering approach. In the second method, a fixed filter size, equal to within-row tree spacing at the time of planting, was used for local maxima filtering. The number of maxima from each plot

was converted to a SPHA estimate for that plot. Comparison of the spatial resolutions showed that 0.2 m returned the best estimate of SPHA for both approaches (Tesfamichael et al., 2009). This spatial resolution was therefore used for further analysis of CHMs, the results of which are presented in this study. c Simulating lidar point densities and tree identification. Lidar point density is directly related to survey cost (Baltsavias, 1999; Lovell et al., 2005) and its effect on forest structural estimations has been the subject of interest in different studies (eg, Holmgren, 2004; Thomas et al., 2006; Magnusson et al., 2007). Three lidar point densities were compared in this study to assess the effect on SPHA, height, and volume estimations. These densities were 5 points/m2 (the original density), 3 points/m2, and 1 point/m2, with the latter two densities extracted from the original density level. Extraction of points was based on random selection (Anderson et al., 2005). It was deemed important to evaluate the similarity in distributions of the simulated point densities and the original point density. This was confirmed using histogram analysis. CHMs were then generated at a spatial resolution of 0.2 m for each density level. Local maxima filtering was performed using filter sizes determined by the semi-variogram range and tree spacing at CHMs generated at the three lidar point densities. A summary of the results is given in Table 2. The overall accuracies of plot-level SPHA estimates derived from local maxima were distinctly better for the tree spacing approach than for the semi-variogram approach at each density level. In both local maxima filtering approaches, lidar-derived SPHA underestimated fieldenumerated SPHA significantly, with the exception of the approach that utilized tree spacing at the density of 5 points/m2. Thus, although there was no validation using individual trees, an error of omission prevailed over that of commission.

Tesfamichael et al.

The magnitude of underestimation tended to vary directly with SPHA. The net underestimation of SPHA was attributed to the exclusion of small trees that were overtopped by taller, adjacent trees (Persson et al., 2002; Popescu et al., 2002; Heurich, 2008).

4 Estimation of height In order to estimate plot-level mean and dominant heights, individual tree crowns were first delineated using local maxima as centers of crowns. The delineation was based on window sizes specified by the semi-variogram range and tree spacing approaches. It has been reported that gridding introduces error in estimates of canopy surfaces due to the interpolation procedure and the choice of grid size (Smith et al., 2004), which led to our use of points rather than rasterized information. Non-ground points thus were assigned to each of the crown polygons, from which mean and maximum heights were extracted as estimates of individual tree heights. Average values of these heights were calculated and compared with field-measured mean height for each plot. Dominant height of each plot was calculated using the same approach employed to the field data, that is, the average height of the tallest 20% tree-level point heights. Derivation of mean and dominant height was replicated for each set of trees identified at the three lidar point densities (1 point/m2, 3 points/m2, and 5 points/m2).

5 Estimation of volume Volume typically can be calculated for each tree, followed by aggregation to a plot or stand level, given the availability of necessary information at individual tree level. Tree height and DBH are the most common independent variables needed for the estimation of tree volume (Avery and Burkhart, 2001). The intention of this study was to derive both these variables from the lidar data. Although height can be retrieved in a relatively straightforward manner, DBH cannot be

525

measured from airborne lidar scanning directly. Relationships between height and DBH could be used to estimate the latter; however, establishing such a relationship breaches the assumption of variable independence. An alternative approach to DBH derivation is to make use of information on stand stocking, which relates to basal area (von Gadow and Bredenkamp, 1992). In an even-aged plantation, mean basal area can be computed from SPHA, dominant height, and the age of trees. SPHA and dominant height were estimated using the procedures described above while age is documented for each plantation. We assumed that the derivation of basal area, and thus resultant DBH, remained independent from the mean height independent variable in the volume estimation (dependent variable). This assumption is based on the facts that (1) basal area and subsequent DBH are derived from dominant (maximum) heights and (2) basal area estimation furthermore relies on two additional non-height-specific variables (SPHA and age). The basal area function chosen in this study is given in equation 2 (Pienaar and Harrison, 1988): " BA ¼ exp b0 þ

b1 þ b2 lnðSPHAÞ Age

lnðSPHAÞ lnðHDÞ þ b5 þb3 lnðHDÞ þ b4 Age Age

#

ð2Þ where BA is basal area in m2/ha, Age is the age of trees, and HD is dominant height per plot. b0– b5 are site specific coefficients (Kotze, 2000). The resulting BA was then used to derive quadratic mean DBH (QDBH) using equation 3 (Pienaar and Kotze, 1998). Equation 3 is an inversion of the function for estimating BA of a tree (equation 4) while SPHA and 40,000 are scaling factors to a hectare-level computation. QDBH and mean height, which was derived either from the mean or maximum lidar returns,

526


Table 3. Comparison of observed and estimated plot-level mean height* Lidar density (points/m2) Local maxima filtering window Metric 1, 3, 5 1, 3, 5

Variogram range/tree spacing Variogram range/tree spacing

Mean (m)** RMSE (%) R2

Field 21.5a Lidar_mean 21.5–21.8a Lidar_max 23.5–24.2b

7 13–15

0.93–0.94 0.93

*Lidar estimates range for densities and local maxima (LM) filtering techniques. **Different characters indicate significant differences at a¼0.05.

were then used as inputs to equation 1 to calculate volume of the mean tree. Multiplying volume of the mean tree by SPHA of a plot yielded volume per hectare for that plot. Similar to height estimations, volume estimation was replicated for the three lidar point densities. sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi BA=SPHA ð3Þ QDBH ¼ p=40000 BA ¼ pr2

ð4Þ

where r ¼ DBH/2.

6 Accuracy assessment Lidar estimates and field observations were compared among density levels and between local maxima filtering approaches. Comparisons also included estimates of height derived from mean and maximum lidar metrics. Analysis of variance (ANOVA) was used to determine if the estimated means differed significantly from the observed mean and among each other. Residual analysis was used to quantify the difference between observed and estimated plot-level means using RMSE. RMSE was expressed as a percentage of the observed mean. A linear relationship between estimated and observed values was established for each of the variables (mean height, dominant height, and volume). The relationship was evaluated both graphically and based on the coefficient of determination (R2).

III Results 1 Height Table 3 summarizes results of ANOVA, regression analysis, and errors of estimation for mean tree height. The difference between observed and estimated mean height using mean lidar height values was not significant. In contrast, estimates using maximum lidar height values varied significantly from field observations. The difference between estimates of the two metrics was also significant. The local maxima filtering technique and lidar point density did not affect height estimation for a given metric. Error analysis showed higher RMSE values for the maximum lidar than for the mean lidar estimation approach. This observation held true for all lidar point densities and local maxima filtering approaches. The relationship between observed and estimated mean heights, based on mean lidar points, consistently exhibited R2 values exceeding 0.90 for both local maxima filtering approaches. Such accuracies were achieved at all density levels. Similar R2 values were obtained when maximum lidar points within delineated trees were used. Figure 2 provides various graphical representations of the results. The graph relating observed and estimated mean height shows that nearly all plots were overestimated by maximum lidar height values (Figure 2a). In general, the overestimation by both lidar height values increased with height. Errors of estimation were also plotted and exhibited a direct relationship with mean height (Figure 2b). This behavior was in particular more noticeable for maximum lidar height

Tesfamichael et al.

527

Figure 2. Mean tree height estimation: (a) estimated versus observed at 5 points/m2; (b) estimation error at 5 points/m2; (c) comparison of estimates at the three lidar point densities; (d) comparison of estimates between local maxima filtering approaches

values. These characteristics were deemed similar across lidar point densities (Figure 2c) and for both local maxima filtering approaches (Figure 2d). Table 4 summarizes the comparison between observed and estimated dominant height at all density levels and local maxima filtering

approaches. Estimated and observed dominant height did not vary significantly. There were also no significant differences between estimates based on lidar metrics at all density levels and for both local maxima filtering approaches. However, the difference between the mean and maximum estimates and the observed values

528


Table 4. Comparison of observed and estimated dominant height* Lidar density (points/m2) Local maxima filtering window Metric 1, 3, 5 1, 3, 5

Variogram range/tree spacing Variogram range/tree spacing

Mean (m)** RMSE (%) R2

Field 25.3a Lidar_mean 23.5–24.4a Lidar_max 25.3–25.9a

7–9 5–6

0.95 0.94–0.95

*Lidar estimates range for densities and local maxima (LM) filtering techniques. **Different characters indicate significant differences at a¼0.05.

appears to be considerable with approximately 1–2 m for mean and *0.5 m or less for maximum lidar height values. RMSE was also relatively similar across all estimates. Coefficients of determination for the linear relationship between estimated and observed dominant height at all density levels and local maxima filtering approaches were similar for both lidar metrics (R2*0.95). Figure 3 shows the relationship between estimated and observed dominant height as well as the error analysis; the graphs indicate that maximum lidar height input values yielded higher estimates than mean values (Figure 3, a and b). The lidar point density (Figure 3c) and local maxima filtering approaches (Figure 3d) did not have any effect, as can be seen from the similar estimates in the respective graphs.

2 Volume Although the objective was to estimate volume at the plot level, volume of the mean basal area tree had to be computed first. It is therefore useful to explore the derivatives of lidar that will serve as independent variable inputs to the tree volume estimation. These inputs are mean height and QDBH of the tree with mean basal area. Calculation of mean basal area, in turn, requires dominant height, in addition to SPHA and age. Results presented above showed that dominant height computed from maximum lidar height values was marginally better than when mean lidar values were used. Hence maximum lidar height values were chosen for computing QDBH. Mean lidar height values, on the other

hand, were superior for estimating plot-level mean height. These estimates were, therefore, used in the equation to calculate mean tree volume (equation 1). A summary of the results of mean tree volume estimation is presented in Table 5. The mean values of volume estimates ranged between 0.26 and 0.31 m3 compared to 0.26 m3 of field observation. Local maxima filtering using the semi-variogram range resulted in overestimation, while tree spacing-based local maxima resulted in underestimation. The estimates, however, were not significant, compared to the field mean. There was, however, a significant difference between estimates at 1 point/m2 density level for the semi-variogram-based local maxima filtering approach and at 3 points/m2 and 5 points/m2 densities for the tree spacing-based local maxima filtering approach. Comparison based on RMSE shows higher values for trees identified using the semi-variogram approach than the tree spacing-based local maxima filtering approach. Linear model fitness between observed and estimated volume resulted in R2 values ranging between 0.76 and 0.83 across all lidar point densities for both local maxima filtering approaches. Graphical representation of observed-estimated relationships shows increased overestimation in the case of the semi-variogram, as opposed to the tree spacing local maxima filtering approach (Figure 4a). This is confirmed by the graph showing the error distribution (Figure 4b). Figures 4a and 4b also show that only a limited number of samples were considerably underestimated at higher tree volume values. Comparison of estimation among

Tesfamichael et al.

529

Figure 3. Dominant height estimation: (a) estimated versus observed at 5 points/m2; (b) estimation error at 5 points/m2; (c) comparison of estimates at the three lidar point densities; (d) comparison of estimates between local maxima filtering approaches

densities (Figure 4c) exhibited slightly higher values for 1 point/m2 than for 3 points/m2 and 5 points/m2 for most of the samples. Plot-level volume estimates were lower than field-observed values at all point density levels and for trees identified using the two respective local maxima filtering methods (Table 6). Estimates that followed tree spacing-based local

maxima filtering at the density levels of 3 points/m2 and 5 points/m2 did not differ significantly from field mean. Differences among estimates were not significant, except for estimates between the semi-variogram range at 1 point/m2 and the tree spacing approach at 5 points/m2. There also appeared to be relatively strong similarities among estimates within each

530


Table 5. Comparison of observed and estimated mean tree volume Lidar density (points/m2)

Local maxima filtering window

Mean (m3)*

RMSE (%)

R2

1

Field Variogram range Tree spacing

0.26ab 0.31b 0.26ab

42 29

0.76 0.80

3

Variogram range Tree spacing

0.28ab 0.25a

36 29

0.77 0.82

5


0.28ab 0.24a

33 29

0.83 0.81


local maxima filtering approach. It can also be seen that mean volume estimates increased with lidar point density within each local maxima filtering approach (Table 6). Superior accuracies were obtained for trees identified using local maxima filtering based on tree spacing, as opposed to the semi-variogram range approach. Error of estimation (RMSE) and fit statistics (R2) hinted at improved estimation accuracy when the tree spacing approach, rather than semi-variogram range, was used for local maxima filtering (Table 6). Graphical representations of estimations are illustrated in Figure 5. Figure 5a shows a fairly high degree of relationship between observed and estimated values at 5 points/m2. However, field volume was underestimated for most samples for both local maxima filtering approaches, as illustrated in Figure 5b. A comparative evaluation of the graphs indicates less underestimation when tree spacing rather than semi-variogram range was used in the tree identification procedure. It can also be seen from Figures 5a and 5b that error of estimation increased as volume increased. Comparison of the effects of lidar point density on the estimation of volume indicates marginally higher values for nearly all samples in the case of 3 points/m2 and 5 points/m2, as opposed to 1 point/m2 (Figure 5c).

IV Discussion 1 Height The relationship between field-observed and lidar-derived mean height using either mean or maximum lidar height values was regarded as strong (R2¼0.93–0.94) (Table 3). ANOVA and error comparisons between the two estimates in relation to field observations indicated that mean lidar height values were superior to maximum values, which overestimated field height significantly (Table 3). A relatively poor accuracy for the maximum lidar height estimates may contradict the assumption that the maximum point within each tree should approximate the top height of a tree measured on the ground. This view holds, particularly, when comparisons are made based on individual trees (eg, Clark et al., 2004). The observation in this study was attributed to various reasons. Since fewer trees were identified than exist on the ground, it is probable that small trees were omitted due mainly to a lidar shadow effect by taller trees. The absence of such small trees from a plot is likely to inflate the mean height calculated only from the lidar-identified trees. Another possible explanation relates to the field measurement. This was a closed-canopy environment and crowns of the trees were often interleaved, which led to errors in identification of tree tops.

Tesfamichael et al.

531

Figure 4. Mean tree volume estimation: (a) estimated versus observed at 5 points/m2; (b) estimation error at 5 points/m2; (c) comparison of estimates at the three lidar point densities

Falkowski et al. (2008) also noted this issue as a potential reason for overestimation of mean tree height in their study. Furthermore, the tops of Eucalyptus trees are not distinctively pointed, resulting in difficulty when finding the tree apex from the ground (Tickle et al., 2006). It is thus likely that the field measurements generally underestimated actual tree height, exacerbated

by the marginally inflated mean tree height in the case of lidar estimates due to the exclusion of smaller trees. The fact that overestimation was greater for taller trees using both metrics also hinted at the difficulty of measuring such trees in the field. As a result, it is logical to assume that the negative bias of field measurement increased with height. These arguments

532


Table 6. Comparison of observed and estimated plot-level volume Density (points/m2)

Local maxima filtering window

Mean (m3/ha)*

RMSE (%)

R2

1

Field Variogram range Tree spacing

274.7a 170.8b 207.7bc

43 29

0.87 0.91

3


200.3bc 226.9ac

35 24

0.82 0.90

5


205.9bc 233.1ac

33 20

0.84 0.94


also explain why lidar mean height did not underestimate field mean height, as was observed in other studies that followed individual tree height estimation (eg, McCombs et al., 2003; Morsdorf et al., 2004; Sua´rez et al., 2005; Heurich, 2008). Similar to mean height, the goodness-of-fit of the relationship between observed and estimated dominant height was strong for both mean and maximum lidar height metrics (R2¼0.94–0.95; Table 4). Although there was no significant variation between observed and estimated values, a difference in mean values of 1–2 m using mean lidar height values compared to a maximum of 0.5 m using maximum lidar height values (Table 4) could be considered large in commercial forest management contexts. A variation of such magnitude could have an impact on characterizing site quality of a plantation area. In addition, dominant height is also an essential input to the computation of basal area, since it generally relates to the stocking of a plantation. Both mean height and dominant height did not vary across lidar point densities, corroborating results from a concurrent study that utilized an area-based modeling approach (Tesfamichael et al., 2010). This phenomenon is worth noting, given that the current study extracted the height of individual trees as identified from lidar data. An additional advantage of the observations is that height accuracies did not rely on the

accuracy of tree counts. The number of trees extracted using local maxima identification differed significantly between lower (1 point/m2) and higher lidar point density levels (3 points/ m2 or 5 points/m2), as well as between the tree spacing-based and semi-variogram-based local maxima filtering methods (Table 2). It is likely that the variability in point height is comparable across the study area, provided that identified trees in most cases represented the tallest trees. In addition, the relatively high canopy closure of the plantations favors point returns from the very top surface of canopy. A generally rounded or flattened canopy top of Eucalyptus trees further adds to this effect, since the variability in height for such canopy geometries is relatively low (Nelson, 1997). The relatively high estimation accuracies obtained using the approach followed in this study strongly hints that field height measurement might be superfluous. This could have a significant impact on the industry that considers the current field protocol as expensive and time-consuming.

2 Volume Studies that attempted to estimate individual stem volume from lidar data are rare (eg, Persson et al., 2002; Heurich, 2008). These studies typically employed models developed from a combination of lidar and field data to derive

Tesfamichael et al.

533

Figure 5. Plot-level volume estimation: (a) estimated versus observed at 5 points/m2; (b) estimation error at 5 points/m2; (c) comparison of estimates at the three lidar point densities

certain tree attributes, with relatively high estimation accuracies reported. In contrast, this study relied on lidar data and auxiliary information (tree spacing and/or age), readily available for properly documented plantation stands. Our results showed that the relationship between observed and estimated volume of the mean tree was relatively strong at the three lidar point densities for both approaches of local maxima filtering (R2¼0.76–0.83; Table 5). Although the errors

of estimations (RMSE) were deemed large, the graphical illustrations did not show considerable systematic error (Figure 4). This has resulted in the non-significant difference between the observed and estimated tree volume (Table 5). It is also important to note that such accuracies were found in spite of underestimations in the number of trees detected from lidar CHMs. Plot-level volume estimation in this study was deemed satisfactory, based on the results

534

Figure 6. Overall difference in estimated tree volume due to changes in each variable for lidar point density of 5 points/m2 and tree spacing-based local maxima filtering

relating to the coefficient of determination and ANOVA for higher lidar point densities. The levels of error observed as underestimates are, on the other hand, high in comparison to the expectations of the commercial forest industry that usually requires accuracy within 10%. Underestimation was attributed to the loss of volume in small trees that are often missed during the detection process (eg, Persson et al., 2002; Maltamo et al., 2004; Heurich, 2008). This argument holds true in this study since our estimation of SPHA, which was used as a multiplication factor for plot-level aggregation, was underestimated by the lidar data. In order to substantiate this argument, a simple sensitivity analysis was performed to rank the importance of input variables using the data obtained from lidar point density of 5 points/m2 and tree spacing based local maxima filtering. Each of the inputs was varied based on the input variable RMSE values, and the resulting overall mean volume differences are depicted in Figure 6. Varying SPHA resulted in the largest mean difference in plot-level tree volume, while mean height and dominant height have relatively similar effect. The tendency of direct relationship between volume and its underestimation (Figure 5b) can also be explained by the underestimation of


SPHA. That is, assuming an inverse relationship between plot density and detection rate, it is highly likely that denser plots will result in a greater underestimation of volume. The approach to estimate plot-level volume, based on an average mean tree volume and SPHA, was different from that employed by Persson et al. (2002) and Heurich (2008), who calculated volume of each tree and then aggregated to the plot level. Comparability of the results was nevertheless notable, indicating the suitability of the method. It should be stressed that this method was developed for even-aged monoculture stands, typically characterized by homogenous physical structures. Although the approach by Persson et al. and Heurich has become common practice, it is important to consider its drawback. Deriving DBH from lidarderived height (and crown dimension) relies on model calibration using costly field data. Comparison among estimates indicated similarities among lidar point densities (Table 6), thereby indicating the suitability of low point density compared to the higher densities. This will have a significant contribution in reducing the cost of lidar surveys. The lidar data provider for this study, for example, indicated that there would approximately be a net 30% cost reduction for a survey of 3 points/m2 compared to 5 points/m2 for an area of 10,000 ha. Similarities were also observed between estimates of local maxima filtering methods, clearly showing the potential of the semi-variogram approach in the analysis of tree detection. The fact that this approach does not require any a priori knowledge of field conditions makes it more appealing, especially in cases where there is considerable difference in stocking between the time of planting and lidar data acquisition. Both the similarities in plot-level volume estimates across the density levels and between local maxima filtering approaches were realized in spite of the significant variations in SPHA. This can be explained by the fact that mean and dominant height may also be contributing to

Tesfamichael et al.

volume estimation to an extent of affecting the significance level. The contribution of these variables is manifested at the tree level. The height of trees identified using local maxima filtering often represents the tallest trees within a plot. Basal area derived from these trees is relatively high (equation 2) and is directly related to stem diameter (QDBH) (equation 3). Although SPHA is directly related to basal area, it is inversely related to stem diameter. It follows that finding the dominant trees contributed towards the overall similarities of plot-level volume estimates which might have negated the effect of SPHA to a certain degree. This corroborates the assertion that dominant trees explain most of the variability in volume (eg, Hyyppa et al., 2001; Maltamo et al., 2004; Heurich, 2008), although these sources derived volume for each tree. We concluded that the accuracy of volume estimation at the tree and plot level can be regarded as promising in light of the different methods followed in deriving the field and lidar volume. The field method computed the volume of each tree from known height and DBH, while the average tree volume was calculated using the lidar data. This may also hint at the ability of lidar data to accurately estimate plot-level mean DBH, a variable that has traditionally been difficult to assess from remotely sensed data.

V Conclusions This study outlines a methodology for quantifying timber resources in Eucalyptus grandis plantations using discrete return lidar height data and limited auxiliary input(s) from readily available information on plantations. It has demonstrated that two height variables (mean and dominant height) can be accurately estimated from trees identified from CHMs using local maxima filtering techniques. Combining estimates of height and stems per hectare, along with documented information on age of plantations, resulted in accurate estimates of mean tree volume. Furthermore, tree volume estimates were similar in all

535

cases of local maxima filtering approaches and lidar point densities, although significant variations were observed in SPHA estimates. Plot-level volume estimates were comparable to field observation particularly at higher lidar point densities, although underestimation was noted at all point densities. This underestimation was attributed to a smaller number of trees identified using lidar data than exist in the field. In conclusion, the overall approach of this study in minimizing inputs from field inventories should be considered indicative, rather than definitive, to future directions of lidar applications in the plantation industry. We suggest that future research should place more emphasis on improving SPHA estimation, given its importance as reported in this study. One approach could involve the development of advanced tree-counting procedures that are suitable for Eucalyptus plantation forests, which are typically characterized by overlapping crowns and near-rounded/flat canopy tops. Another approach would be to account for small trees using theoretical distribution functions, as illustrated by Maltamo et al. (2004), provided that reliable functions have already been developed. Acknowledgements This study was conducted under a Remote Sensing Cooperative project sponsored by the Council for Scientific and Industrial Research (CSIR), South Africa, and Mondi South Africa. We acknowledge the help of Yvonne Fletcher (Mondi South Africa) in the statistical analysis of the study. The authors would also like to thank Wesley Naidoo and Thamsanqa Mzinyane for their support throughout the field data collection. The authors are also grateful to Wesley Roberts (CSIR) for his constructive comments on the manuscript, as well as members of the cooperative for their inputs during the course of the study.

References Ackermann, F. 1999: Airborne laser scanning – present status and future expectations. ISPRS Journal of Photogrammetry and Remote Sensing 54, 64–67.

536 Andersen, H.-E., McGaughey, R.J. and Reutebuch, S.E. 2005: Estimating forest canopy fuel parameters using lidar data. Remote Sensing of Environment 94, 441–49. Anderson, E.S., Thompson, J.A. and Austin R.E. 2005: Lidar density and linear interpolator effects on elevation estimates. International Journal of Remote Sensing 26, 3889–900. Avery, T.E. and Burkhart, H.E. 2001: Forest Measurements (fifth edition). New York: McGraw-Hill, 456 pp. Baltsavias, E.P. 1999: A comparison between photogrammetry and laser scanning. ISPRS Journal of Photogrammetry and Remote Sensing 54, 83–94. Bortolot, Z.J. and Wynne, R.H. 2005: Estimating forest biomass using small footprint lidar data: an individual tree-based approach that incorporates training data. ISPRS Journal of Photogrammetry and Remote Sensing 59, 342–60. Boyd, D.S. and Danson, F.M. 2005: Satellite remote sensing of forest resources: three decades of research development. Progress in Physical Geography 29, 1–26. Brandtberg, T. 2007: Classifying individual tree species under leaf-off and leaf-on conditions using airborne lidar. ISPRS Journal of Photogrammetry and Remote Sensing 61, 325–40. Brandtberg, T. and Walter, F. 1998: Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis. Machine Vision and Applications 11, 64–73. Brandtberg, T., Warner, T.A., Landenberger, R.E. and McGraw, J.B. 2003: Detection and analysis of individual leaf-off tree crowns in small footprint, high sampling density lidar data from the eastern deciduous forest in North America. Remote Sensing of Environment 85, 290–303. Brown, A.G., Nambiar, E.K.S. and Cossalter, C. 1997: Plantations for the tropics – their role, extent, and nature. In Nambiar, E.K.S. and Brown, A.G., editors, Management of soil, nutrients, and water in tropical plantation, Canberra: ACIAR, 1–23. Brown, S. 2002: Measuring carbon in forests: current status and future challenges. Environmental Pollution 116, 363–72. Chen, J.M. and Cihlar, J. 1996: Retrieving leaf area index of boreal conifer forests using Landsat TM images. Remote Sensing of Environment 55, 153–62. Chen, Q., Baldocchi, D., Gong, P. and Kelly, M. 2006: Isolating individual trees in a savanna woodland using

Progress in Physical Geography 34(4) small-footprint lidar data. Photogrammetric Engineering and Remote Sensing 72, 923–32. Chen, Q., Gong, P., Baldocchi, D. and Tian, Y.Q. 2007: Estimating basal area and stem volume for individual trees from lidar data. Photogrammetric Engineering and Remote Sensing 73, 1355–65. Chubey, M.S., Franklin, S.E. and Wulder, M.A. 2006: Object-based analysis of IKONOS-2 imagery for extraction of forest inventory parameters. Photogrammetric Engineering and Remote Sensing 72, 383–94. Clark, M.L., Clark, D.B. and Roberts, D.A. 2004: Smallfootprint lidar estimation of sub-canopy elevation and tree height in a tropical rain forest landscape. Remote Sensing of Environment 91, 68–89. Clarke, C.R.E. and Wessels, A.M. 1995: Variation and measurement of pulp properties in eucalypts. In Potts, B.M., Borralho, N.M.G., Reid, J.B., Tibbits, W.N. and Raymond, C.A., editors, Eucalypt plantations: improving fibre yield and quality, Proceedings of the CRC-IUFRO Conference, Hobart: CRC for Temperate Hardwood Forestry, 93–100. Coetzee, J. 1992: A revised tree volume table for short rotation Eucalyptus grandis timber crops. ICFR Bulletin 9/92. Pietermaritzburg: Institute for Commercial Forestry Research, 5 pp. Coetzee, J. 1999: The use of tree volume equations stratified by height classes instead of a general single format for short rotation applications with Eucalyptus grandis. ICFR Bulletin 6. Pietermaritzburg: Institute for Commercial Forestry Research, 9 pp. Coops, N.C., Wulder, M.A., Culvenor, D.S. and St-Onge, B. 2004: Comparison of forest attributes extracted from fine spatial resolution multispectral and lidar data. Canadian Journal of Remote Sensing 30, 855–66. Drake, J.B., Dubayah, R.O., Clark, D.B., Knox, R.G., Blair, J.B., Hofton, M.A., Chazdon, R.L., Weishampel, J.F. and Prince, S.D. 2002: Estimation of tropical forest structural characteristics using large-footprint lidar. Remote Sensing of Environment 79, 305–19. Dralle, K. and Rudemo, M. 1996: Stem number estimation by kernel smoothing of aerial photos. Canadian Journal of Forest Research 26, 1228–36. Dye, P. and Versfeld, D. 2007: Managing the hydrological impacts of South African plantation forests: an overview. Forest Ecology and Management 251, 121–28. Esler, W.K. 2004: The basics of forest mensuration. Unpublished internal document. Pietermaritzburg: Mondi South Africa, 13 pp.

Tesfamichael et al. Falkowski, M.J., Evans, J.S., Martinuzzi, S., Gessler, P.E. and Hudak, A.T. 2009: Characterizing forest succession with lidar data: an evaluation for the Inland Northwest, USA. Remote Sensing of Environment 113, 946–56. Falkowski, M.J., Smith, A.M.S., Gessler, P.E., Hudak, A.T., Vierling, L.A. and Evans, J.S. 2008: The influence of conifer forest canopy cover on the accuracy of two individual tree measurement algorithms using lidar data. Canadian Journal of Remote Sensing 34 (Supplement 2), 338–50. Foody, G.M. and Curran, P.J. 1994: Estimation of tropical forest extent and regenerative stage using remotelysensed data. Journal of Biogeography 21, 223–44. Franco-Lopez, H., Ek, A.R. and Bauer, M.E. 2001: Estimation and mapping of forest stand density, volume, and cover type using the k-nearest neighbours method. Remote Sensing of Environment 77, 251–74. Garcıá, O. 1998: Estimating top height with variable plot sizes. Canadian Journal of Forest Research 28, 1509–17. Godsmark, R. 2009: The South African forestry and forest products industry 2008. Pretoria: Forestry South Africa. Goetz, S., Steinberg, D., Dubayah, R. and Blair, B. 2007: Laser remote sensing of canopy habitat heterogeneity as a predictor of bird species richness in an eastern temperate forest, USA. Remote Sensing of Environment 108, 254–63. Gonçalves, J.L.M., Barros, N.F., Nambiar, E.K.S. and Novais, R.F. 1997: Soil and stand management for short rotation plantations. In Nambiar, E.K.S. and Brown, A.G., editors, Management of soil, nutrients, and water in tropical plantation, Canberra: ACIAR, 379–417. Gougeon, F.A. 1995: A crown-following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images. Canadian Journal of Remote Sensing 21, 274–84. Graf, R.F., Mathys, L. and Bollmann, K. 2009: Habitat assessment for forest dwelling species using lidar remote sensing: Capercaillie in the Alps. Forest Ecology and Management 257, 160–67. Hall, R.J., Skakun, R.S., Arsenault, E.J. and Case, B.S. 2006: Modeling forest stand structure attributes using Landsat ETMþ data: application to mapping of above-ground biomass and stand volume. Forest Ecology and Management 225, 378–90.

537 Hall, S.A., Burke, I.C., Box, D.O., Kaufmann, M.R. and Stoker, J.M. 2005: Estimating stand structure using discrete return lidar: an example from low density, fire prone ponderosa pine forests. Forest Ecology and Management 208, 189–209. Heurich, M. 2008: Automatic recognition and measurement of single trees based on data from airborne laser scanning over the richly structured natural forests of the Bavarian Forest National Park. Forest Ecology and Management 255, 2416–33. Holmgren, J. 2004: Prediction of tree height, basal area, and stem volume in forest stands using airborne laser scanning. Scandinavian Journal of Forest Research 19, 543–53. Holmgren, J., Nilsson, M. and Olsson, H. 2003: Estimation of tree height and stem volume on plots using airborne laser scanning. Forest Science 49, 419–28. Hyyppa¨, J., Hyyppa¨, H., Leckie, D., Gougeon, F., Yu, X. and Maltamo, M. 2008: Review of methods of smallfootprint airborne laser scanning for extracting forest inventory data in boreal forests. International Journal of Remote Sensing 29, 1339–66. Hyyppa¨, J., Kelle, O., Lehikoinen, M. and Inkinen, M. 2001: A segmentation-based method to retrieve stem volume estimates from 3-D tree height models produced by laser scanners. IEEE Transactions on Geoscience and Remote Sensing 39, 969–75. Jensen, J.L.R., Humes, K.S., Vierling, L.A. and Hudak, A.T. 2008: Discrete return lidar-based prediction of leaf area index in two conifer forests. Remote Sensing of Environment 112, 3947–57. Kato, A., Moskal, L.M., Schiess, P., Swanson, M.E., Calhoun, D. and Stuetzle, W. 2009: Capturing tree crown formation through implicit surface reconstruction using airborne lidar data. Remote Sensing of Environment 113, 1148–62. Koch, B., Heyder, U. and Welnacker, H. 2006: Detection of individual tree crowns in airborne lidar data. Photogrammetric Engineering and Remote Sensing 72, 357–63. Koch, B., Straub, C., Dees, M., Wang, Y. and Weinacker, H. 2009: Airborne laser data for stand delineation and information extraction. International Journal of Remote Sensing 30, 935–63. Kotze, H. 2000: Growth and yield functions. Unpublished Mensuration and Modeling Consortium (MMRC) report. Pietermaritzburg: MMRC, 70 pp.

538 Leboeuf, A., Beaudoin, A., Fournier, R.A., Guindon, L., Luther, J.E. and Lambert, M.-C. 2007: A shadow fraction method for mapping biomass of northern boreal black spruce forests using Quickbird imagery. Remote Sensing of Environment 110, 488–500. Lefsky, M.A., Cohen, W.B., Acker, S.A., Parker, G.G., Spies, T.A. and Harding, D. 1999: Lidar remote sensing of the canopy structure and biophysical properties of Douglas-fir western hemlock forests. Remote Sensing of Environment 70, 339–61. Lefsky, M.A., Cohen, W.B., Parker, G.G. and Harding, D.J. 2002: Lidar remote sensing for ecosystem studies. Bioscience 52, 19–30. Lefsky, M.A., Cohen, W.B. and Spies, T.A. 2001: An evaluation of alternate remote sensing products for forest inventory, monitoring, and mapping of Douglas-fir forests in western Oregon. Canadian Journal of Forest Research 31, 78–87. Le Maitre, D.C., van Wilgen, B.W., Gelderblom, C.M., Bailey, C., Chapman, R.A. and Nel, J.A. 2002: Invasive alien trees and water resources in South Africa: case studies of the costs and benefits of management. Forest Ecology and Management 160, 143–59. Lim, K., Treitz, P., Wulder, M., St-Onge, B. and Flood, M. 2003: Lidar remote sensing of forest structure. Progress in Physical Geography 27, 88–106. Louw, J.H. and Scholes, M. 2002: Forest site classification and evaluation: a South African perspective. Forest Ecology and Management 171, 153–68. Lovell, J.L., Jupp, D.L.B., Newnham, G.J., Coops, N.C. and Culvenor, D.S. 2005: Simulation study for finding optimal lidar acquisition parameters for forest height retrieval. Forest Ecology and Management 214, 398–412. Magnusson, M., Fransson, J.E.S. and Holmgren, J. 2007: Effects on estimation accuracy of forest variables using different pulse density of laser data. Forest Science 53, 619–26. Ma¨kela¨, H. and Pekkarinen, A. 2001: Estimation of timber volume at the sample plot level by means of image segmentation and Landsat TM imagery. Remote Sensing of Environment 77, 66–75. Maltamo, M., Eerikaïnen, K., Pitka¨nen, J., Hyyppa¨, J. and Vehmas, M. 2004: Estimation of timber volume and stem density based on scanning laser altimetry and expected tree size distribution functions. Remote Sensing of Environment 90, 319–30.

Progress in Physical Geography 34(4) Maltamo, M., Hyyppa¨, J. and Malinen, J. 2006: A comparative study of the use of laser scanner data and field measurements in the prediction of crown height in boreal forests. Scandinavian Journal of Forest Research 21, 231–38. McAllister, D.M. and Valeo, C. 2007: A robust new method for the remote estimation of LAI in montane and boreal forests. International Journal of Remote Sensing 28, 1891–905. McCombs, J.W., Roberts, S.D. and Evans, D.L. 2003: Influence of fusing lidar and multispectral imagery on remotely sensed estimates of stand density and mean tree height in a managed loblolly pine plantation. Forest Science 49, 457–66. Means, J.E., Acker, S.A., Harding, D.J., Blair, J.B., Lefsky, M.A., Cohen, W.B., Harmon, M.E. and McKee, W.A. 1999: Use of large-footprint scanning airborne lidar to estimate forest stand characteristics in the western cascades of Oregon. Remote Sensing of Environment 67, 298–308. Morsdorf, F., Meier, E., Ko¨tz, B., Itten, K.I., Dobbertin, M. and Allgo¨wer, B. 2004: Lidar-based geometric reconstruction of boreal type forest stands at single tree level for forest and wildland fire management. Remote Sensing of Environment 92, 353–62. Mu¨ller, J. and Brandl, R. 2009: Assessing biodiversity by remote sensing in mountainous terrain: the potential of lidar to predict forest beetle assemblages. Journal of Applied Ecology 46, 897–905. Næsset, E. 2002: Predicting forest stand characteristics with airborne scanning laser using a practical twostage procedure and field data. Remote Sensing of Environment 80, 88–99. Næsset, E. 2004: Practical large-scale forest stand inventory using a small-footprint airborne scanning laser. Scandinavian Journal of Forest Research 19, 164–79. Næsset, E. 2007: Airborne laser scanning as a method in operational forest inventory: status of accuracy assessments accomplished in Scandinavia. Scandinavian Journal of Forest Research 22, 433–42. Nelson, R. 1997: Modeling forest canopy heights: the effects of canopy shape. Remote Sensing of Environment 60, 327–34. Nemani, R.R., Pierce, L., Running, S. and Band, L. 1993: Forest ecosystem processes at the watershed scale: sensitivity to remotely-sensed leaf area index estimates. International Journal of Remote Sensing 14, 2519–34.

Tesfamichael et al. Oliver, M.A. 1990: Kriging: a method of interpolation for Geographical Information Systems. International Journal of Geographic Information Systems 4, 313–32. Ørka, H.O., Næsset, E. and Bollandsa˚s, O.M. 2009: Classifying species of individual trees by intensity and structure features derived from airborne laser scanner data. Remote Sensing of Environment 113, 1163–74. Ozdemir, I. 2008: Estimating stem volume by tree crown area and tree shadow area extracted from pansharpened Quickbird imagery in open Crimean juniper forests. International Journal of Remote Sensing 29, 5643–55. Pallett, R.N. and Sale, G. 2004: The relative contributions of tree improvement and cultural practice toward productivity gains in Eucalyptus pulpwood stands. Forest Ecology and Management 193, 33–43. Patenaude, G., Hill, R.A. and Milne, R. 2004: Quantifying forest above-ground carbon content using lidar remote sensing. Remote Sensing of Environment 93, 368–80. ˚ ., Holmgren, J. and So¨dermann, U. 2002: Persson, A Detecting and measuring individual trees using an airborne laser scanner. Photogrammetric Engineering and Remote Sensing 68, 925–32. Peuhkurinen, J., Maltamo, M., Malinen, J., Pitka¨nen, J. and Packaleń, P. 2007: Preharvest measurement of marked stands using airborne laser scanning. Forest Science 53, 653–61. Pienaar, B. and Harrison, W.M. 1988: A stand table projection approach to yield prediction in unthinned evenaged stands. Forest Science 34, 804–808. Pienaar, B. and Kotze, H. 1998: Growth and yield functions within the growth and yield simulator. Unpublished Mensuration and Modeling Consortium (MMRC) report. Pietermaritzburg: MMRC, 58 pp. Pollock, R. 1998: Individual tree recognition based on a synthetic tree crown image model. In Proceedings of the International Forum on Automated Interpretation of High Spatial Resolution Digital Imagery for Forestry, 10–12 February, Pacific Forestry Centre, Canada Forest Service, Victoria, BC: Natural Resources Canada, 25–34. Popescu, S.C., Wynne, R.H. and Nelson, R.F. 2002: Estimating plot-level tree heights with lidar: local filtering with a canopy height based variable window size. Computers and Electronics in Agriculture 37, 71–95. Popescu, S.C., Wynne, R.H. and Nelson, R.F. 2003: Measuring individual tree crown diameter with lidar and assessing its influence on estimating forest volume and

539 biomass. Canadian Journal of Remote Sensing 29, 564–77. Pouliot, D.A., King, D.J., Bell, F.W. and Pitt, D.G. 2002: Automated tree crown detection and delineation in high resolution digital camera imagery of coniferous forest regeneration. Remote Sensing of Environment 82, 322–34. Reitberger, J. Schno¨rr, Cl., Krzystek, P. and Stilla, U. 2009: 3-D segmentation of single trees exploiting full waveform lidar data. ISPRS Journal of Photogrammetry and Remote Sensing 64, 561–74. Reutebuch, S.E., Andersen, H.-E. and McGaughey, R.J. 2005: Light detection and ranging (lidar): an emerging tool for multiple resource inventory. Journal of Forestry 103, 286–92. Rietz, D.N. and Smith, C.W. 2004: A preliminary study of above-ground allometric relationships of Eucalyptus nitens (Deane and Maiden) plantations in South Africa. ICFR Bulletin 14. Pietermaritzburg: Institute for Commercial Forestry Research, 34 pp. Roberts, S.D., Dean, T.J., Evans, D.L., McCombs, J.W., Harrington, R.L. and Glass, P.A. 2005: Estimating individual tree leaf area in loblolly pine plantations using lidar-derived measurements of height and crown dimensions. Forest Ecology and Management 213, 54–70. Sader, S.A., Waide, R.B., Lawrence, W.T. and Joyce, A.T. 1989: Tropical forest biomass and successional age class relationships to a vegetation index derived from Landsat TM data. Remote Sensing of Environment 28, 143–56. Scho¨nau, A.P.G. and Boden, D.I. 1982: Preliminary biomass studies in young eucalypts. South African Forestry Journal 120, 24–28. Schreuder, H.T., Gregoire, T.G. and Wood, G.B. 1993: Sampling methods for multiresource forest inventory. New York: Wiley, 464 pp. Schulze, R.E. 1997: South African atlas of agrohydrology and climatology. Report TT82/96. Pretoria: Water Research Commission, 286 pp. Schumacher, F.X. and Hall, F.D.S. 1933: Logarithmic expression of timber-tree volume. Journal of Agricultural Research 47, 719–34. Smith, S.L., Holland, D.A. and Longley, P.A. 2004: The importance of understanding error in lidar digital elevation models. International Archives of Photogrammetry, Remote Sensing, and Spatial Information Sciences 35, 996–1001.

540 Song, C. and Dickinson, M.B. 2008: Extracting forest canopy structure from spatial information of high resolution optical imagery: tree crown size versus leaf area index. International Journal of Remote Sensing 29, 5605–22. Steininger, M.K. 2000: Satellite estimation of tropical secondary forest above-ground biomass: data from Brazil and Bolivia. International Journal of Remote Sensing 21, 1139–57. Sua´rez, J.C., Ontiveros, C., Smith, S. and Snape, S. 2005: Use of airborne lidar and aerial photography in the estimation of individual tree heights in forestry. Computers and Geosciences 31, 253–62. Tesfamichael, S.G., Ahmed, F. and van Aardt, J.A.N. 2010: Investigating the impact of discrete return lidar point density on estimations of mean and dominant plot-level tree height in Eucalyptus grandis plantations. International Journal of Remote Sensing, in press. Tesfamichael, S.G., Ahmed, F., van Aardt, J.A.N. and Blakeway, F. 2009: A semi-variogram approach for estimating stems per hectare in Eucalyptus grandis plantations using discrete return lidar height data. Forest Ecology and Management 258, 1188–99. Thomas, V., Treitz, P., McCaughey, J.H. and Morrison, I. 2006: Mapping stand-level forest biophysical variables for a mixedwood boreal forest using lidar: an examination of scanning density. Canadian Journal of Forest Research 36, 34–47. Thomas, V., Treitz, P., McCaughey, J.H., Noland, T. and Rich, L. 2008: Canopy chlorophyll concentration estimation using hyperspectral and lidar data for a boreal mixedwood forest in northern Ontario, Canada. International Journal of Remote Sensing 29, 1029–52. Tickle, P.K., Lee, A., Lucas, R.M., Austin, J. and Witte, C. 2006: Quantifying Australian forest floristics and structure using small footprint lidar and large scale aerial photography. Forest Ecology and Management 223, 379–94. Turner, D.P., Cohen, W.B., Kennedy, R.E., Fassnacht, K.S. and Briggs, J.M. 1999: Relationship between leaf area index and Landsat TM spectral vegetation indices across three temperate zone sites. Remote Sensing of Environment 70, 52–68.

Progress in Physical Geography 34(4) van Aardt, J.A.N., Wynne, R.H. and Oderwald, R.G. 2006: Forest volume and biomass estimation using smallfootprint lidar distributional parameters on a persegment basis. Forest Science 52, 636–49. von Gadow, K. and Bredenkamp, B. 1992: Forest management. Pretoria: Academica, 164 pp. Wang, L., Gong, P. and Biging, G.S. 2004: Individual tree crown delineation and treetop detection in high spatial resolution aerial imagery. Photogrammetric Engineering and Remote Sensing 70, 351–58. Wehr, A. and Lohr, U. 1999: Airborne laser scanning – an introduction and overview. ISPRS Journal of Photogrammetry and Remote Sensing 54, 68–82. Woodcock, C.E., Strahler, A.H. and Jupp, D.L.B. 1988a: The use of variograms in remote sensing: I. Scene models and simulated images. Remote Sensing of Environment 25, 323–48. Woodcock, C.E., Strahler, A.H. and Jupp, D.L.B. 1988b: The use of variograms in remote sensing: II. Real digital images. Remote Sensing of Environment 25, 349–79. Wu, Y. and Strahler, A.H. 1994: Remote estimation of crown size, stand density, and biomass on the Oregon transect. Ecological Applications 4, 299–312. Wulder, M. 1998: Optical remote sensing techniques for the assessment of forest inventory and biophysical parameters. Progress in Physical Geography 22, 449–76. Wulder, M., Niemann, K.O. and Goodenough, D.G. 2000: Local maximum filtering for the extraction of tree locations and basal area from high spatial resolution imagery. Remote Sensing of Environment 73, 103–14. Wynne, R.H., Oderwald, R.G., Reams, G.A. and Scrivani, J.A. 2000: Optical remote sensing for forest area estimation. Journal of Forestry 98, 31–36. Zhao, K., Popescu, S. and Nelson, R. 2009: Lidar remote sensing of forest biomass: a scale-invariant estimation approach using airborne lasers. Remote Sensing of Environment 113, 182–96. Zimble, D.A., Evans, D.L., Carlson, G.C., Parker, R.C., Grado, S.C. and Gerard, P.D. 2003: Characterizing vertical forest structure using small-footprint airborne lidar. Remote Sensing of Environment 87, 171–82.

Estimating plot-level tree height and volume of

Estimating plot-level tree height and volume of

Suggest Documents

Estimating Tree Height and Diameter at Breast Height (DBH) - MDPI

Estimating Tree Height and Diameter at Breast Height ... - MDPIwww.researchgate.net › publication › fulltext › Estimating

Development of q-exponential models for tree height, volume and

Calculation of tree volume and surface area by the height ...

Estimating Stand Height and Tree Density in Pinus taeda plantations ...

Challenges to Estimating Tree Height via LiDAR in Closed-Canopy

Relationships of tree height and diameter at breast height revisited

Estimating Pinus palustris tree diameter and stem volume ... - UF SFRC

Estimating Individual Tree Diameter and Stem Volume Using Airborne

Tropical tree height allometry

Estimating Single Tree Stem Volume of Pinus sylvestris Using ... - MDPI

The limits to tree height

Tree height measurements and tree growth estimation in a mire ...

Hydraulic limits to tree height and tree growth - Natural Resource ...

ESTIMATING THE HEIGHT OF CMEs ASSOCIATED ... - IOPscience

Height Extraction and Stand Volume Estimation ...

practical methods of estimating volume and fresh

EVALUATION OF TREE HEIGHT AND NUMBER OF TREES ... - tf.llu.lv

Validity of self-reported height and weight for estimating ... - CiteSeerX

Estimation of Tree Height and Forest Biomass Using

Estimating Height from Arm Span Measurement in

Calculation of Tree Height and Canopy Crown from ...

Relationships of tree height and diameter at breast ... - Oxford Journals

ESTIMATION OF EUCALYPTUS TREE HEIGHT IN CLONAL AND