remote sensing

Article Comparison of Canopy Volume Measurements of Scattered Eucalypt Farm Derived from High Spatial Resolution Imagery and

Niva Kiran Verma 1,2, David W. Lamb 1,2,*, Nick Reid 1,3 and Brian Wilson 1,3 1 Cooperative Research Centre for Spatial Information (CRCSI), University of New England, Armidale, NSW 2351, Australia; [email protected] (N.K.V.); [email protected] (N.R.); [email protected] (B.W.) 2 Precision Agriculture Research Group and School of Science and Technology, University of New England, Armidale, NSW 2351, Australia 3 Ecosystem Management, School of Environmental and Rural Science, University of New England, Armidale, NSW 2351, Australia * Correspondence: [email protected]; Tel.: +61-2-6773-3565; Fax: +61-2-6773-2844

Academic Editors: Angela Lausch, Marco Heurich, Nicolas Baghdadi and Prasad S. Thenkabail Received: 16 November 2015; Accepted: 25 April 2016; Published: 5 May 2016

Abstract: Studies estimating canopy volume are mostly based on laborious and time-consuming field measurements; hence, there is a need for easier and convenient means of estimation. Accordingly, this study investigated the use of remotely sensed data (WorldView-2 and LiDAR) for estimating height, canopy height and crown diameter, which were then used to infer the canopy volume of remnant eucalypt trees at the Newholme/Kirby ‘SMART’ farm in north-east New South Wales. A regression model was developed with field measurements, which was then applied to remote-sensing-based measurements. LiDAR estimates of tree dimensions were generally lower than the field measurements (e.g., 6.5% for tree height) although some of the parameters (such as tree height) may also be overestimated by the clinometer/rangefinder protocols used. The WorldView-2 results showed both crown projected area and crown diameter to be strongly correlated to canopy volume, and that crown diameter yielded better results (Root Mean Square Error RMSE 31%) than crown projected area (RMSE 42%). Although the better performance of LiDAR in the vertical dimension cannot be dismissed, as suggested by results obtained from this study and also similar studies conducted with LiDAR data for tree parameter measurements, the high price and complexity associated with the acquisition and processing of LiDAR datasets mean that the technology is beyond the reach of many applications. Therefore, given the need for easier and convenient means of tree parameters estimation, this study filled a gap and successfully used 2D multispectral WorldView-2 data for 3D canopy volume estimation with satisfactory results compared to LiDAR-based estimation. The result obtained from this study highlights the usefulness of high resolution data for canopy volume estimations at different locations as a possible alternative to existing methods.

Keywords: LiDAR; canopy volume; farmscape

1. Introduction The measurement of volume is important in assessing the economic value of a tree [1]. Tree volume measurements may include stem volume (volume of trunk from ground to tip), canopy volume or total tree volume (the sum of the volume of the trunk and canopy). Canopy volume includes the entire live canopy of a tree from the base of the crown to the highest point and from the centre of the crown out to the furthest tips, but does not include dead branches. Canopy volume is an important parameter in the study of yield estimates in [2]. Tree canopy geometric

Remote Sens. 2016, 8, 388; doi:10.3390/rs8050388 www.mdpi.com/journal/remotesensing Remote Sens. 2016, 8, 388 2 of 16 characteristics are directly related to tree growth and productivity, and hence can be used to estimate tree and growth, yield and water consumption, as well as assessing tree health and long-term productivity. Measurements of canopy volume are important for managing trees intermixed with crops or pastures in a farmscape. Canopy volume can be measured manually, but measuring canopy size can be a challenging and time consuming task due to the sometimes complicated growth structures and irregular shapes of trees. The conventional estimation of canopy volume by manual measurement of crown diameter and canopy height assumes an appropriate 3D crown shape. Because of varying crown shape, and extent and positioning of branches, it is difficult to calculate the actual volume of a specific canopy outline [3], so most published models assume that the canopy is a solid geometric object [3–5]. Canopy volume is generally calculated using the following predefined volume formula [4].

Canopy Volume “ Canopy Height ˆ (Crown Diameter)2 ˆ Multiplier (1)

The choice of multiplier varies with crown shape which is typically approximated as a spheroid, ellipsoid, right circular cone or right circular cylinder [4,5]. The numerical value of the multiplier is the fractional volume (e.g., to 4 significant figures [4]) of the shape relative to that of a right cylinder of the same diameter and height. Hence crown shape is an important parameter to be assessed in canopy volume estimations. Troxel et al. [5] examined interactions between age (years from transplanting), bole size (diameter) and crown dimensions (tree height, crown diameter, crown volume) in young urban trees in New Haven, Connecticut, to identify allometric relationships, and to compare growth rates of the 10 most commonly planted species. In their study they estimated canopy volume using the above mentioned volume formula. Rautiainen et al. [6] addressed the importance of crown shape in understanding various stand features including crown shape and crown volume and modelled crown shape from remote sensing data in Finland. Results indicated that the basic dimensions of tree crowns were better predicted for pines. Tumbo et al. [7] investigated methods for rapid mapping of canopy volumes in citrus groves and correlated canopy volume measurements using manual methods, laser, and ultrasonic sensors. They concluded laser and ultrasonic systems to be valuable tools for automatic mapping of canopy volumes in citrus groves. Nelson [8] characterized the effects of the canopy shape assumption on laser measurements such as average canopy height and height variability, and on basal area, volume, and biomass estimates. The results for Costa Rican tropical indicated simulated laser measurements of average height, canopy profile area, and canopy volume increased 8%–10% when a parabolic rather than a conic shape was assumed. An elliptic canopy was 16%–18% taller, on average, than a conic canopy, and a spheric canopy was 23%–25% taller. However, the degree to which these estimates were affected was found to be dependent on study area. The relationships between canopy volume and several tree dimensions such as diameter at breast height, tree height, and crown shape are often used for estimating canopy volume. Allometric equations are commonly developed that predict tree and canopy volume as well as the biomass of several tree components from either diameter at breast height (DBH), tree height or both [9,10]. Figure1 represents the work flow for the calculation of canopy volume based on manual measurements of several tree dimensions (Figure2) in the field. As on-ground field measurements can be time-consuming and costly, numerous remote sensing techniques have been examined for estimating canopy volume. Like on-ground measurements, these methods are indirect and involve the remote measurement of various tree dimensions from which canopy volume may be inferred through predictive modelling (Figure1, Equation (1)). LiDAR (light detection and ranging) is an example of distance (ranging) measuring technology, which relies on the principle of ‘time of flight’. Laser pulses are directed from a source (e.g., mounted on an aircraft) and on striking the target, a portion of the incident beam is scattered back towards the source. High-speed detectors and electronics calculate the time of flight between the emission of the pulse and the return of the back-scattered component, from which the distance (range) between source and target is calculated. LiDAR captures elevation information from a canopy as well as the Remote Sens. 2016, 8, 388 Remote Sens. 2016, 8, 388 Remote Sens. 2016, 8, 388 3 of 16 canopy as well as the ground beneath and can be used to assess complex 3D patterns of canopy and canopyforest stand as well structure as the ground(e.g., [11–13]). beneath LiDAR-derive and can be usedd measurements to assess complex such as3D tree patterns height, of trunkcanopy height and groundforest(perpendicular stand beneath structure distance and can (e.g., bebetween [11–13]). used to ground assess LiDAR-derive complex level andd 3D measurements the patterns lowest of canopypoint such ofas and treethe forest canopy)height, stand trunk and structure heightcrown (e.g.,(perpendiculardiameter [11–13 can]). LiDAR-derivedthen distance be used between to measurements estimate ground canopy level such volumeand as tree the height, usinglowest trunkformulae point height of based the (perpendicular canopy) on Equation and distance crown(1). In betweendiameteraddition ground tocan estimating then level be andusedindividual the to lowestestimate tree pointcanopy canopy of thevo volumelume canopy) and using andbiomass, crownformulae LiDAR diameter based can canonbe usedEquation then to be measure used (1). toIn estimateadditionother forest canopyto estimating structural volume individual usingcharacteristics formulae tree canopy basedsuch voas on lume Equationcrown and diameter biomass, (1). In addition [14,15],LiDARto canpercent estimating be used canopy to individual measure cover treeother[12,16,17], canopy forest stand volume structural volume and biomass, characteristics[18]. LiDAR such can be as used crown to measure diameter other [14,15], forest structuralpercent canopy characteristics cover such[12,16,17], as crown stand diameter volume [ 14[18].,15 ], percent canopy cover [12,16,17], stand volume [18]. Tree Characteristics Measurement

Treerequired Characteristics for volume Measurement extraction required for volume extraction (c) (c)

Tree Height Tree Trunk Height (a) (b) Tree Height Tree Trunk Height (a) (b) (Using Range finder and Clinometer) (Using Range finder and Clinometer)

Crown Shape Crown Diameter/Area Canopy Height Crown Shape Crown Diameter/Area Canopy Height Visual Using Range finder Visual Using Range finder Combined Solid Geometric Combined VolumeSolid Geometric Formula Volume Formula

Canopy Volume Canopy Volume

FigureFigure 1.1. Flow diagram for canopy volume estimations. Figure 1. Flow diagram for canopy volumevolume estimations.estimations.

Figure 2. Schematic diagram indicating canopy dimensions required to estimate canopy volume. Figure 2. Schematic diagram indicating canopy dimensions required to estimate canopy volume. Figure 2. Schematic diagram indicating canopy dimensions required to estimate canopy volume. ManyMany studiesstudies have have yielded yielded promising promising results results with with LiDAR LiDAR to assessto assess single single tree tree heights heights [19– 25[19–25]] and forestand Manyforest plot or studiesplot stand or heighthave stand yielded and height have promising achievedand have results good achieved agreement with goodLiDAR between agreement to assess LiDAR singlebetween derived tree LiDAR parametersheights derived[19–25] and et al. et al. aandparameters variety forest of heightplotand aor measuresvariety stand ofheight [13 height,26 ,27and ].measur However,havees achieved [13,26,27]. studies good by However, Rönnholm agreement studies between[ 28by] andRönnholm HuangLiDAR etderived al. [[28]29] haveparametersand Huang underestimated andet al. a[29] variety tree have height of underestimated height using measur systems treees with[13,26,27]. height a small using However, footprint, systems studies due with to differencesaby small Rönnholm footprint, in the et density al.due [28] to andanddifferences coverageHuang etin of al.thelaser [29] density pulses,have underestimatedand the canopycoverage height of tree lase model,heightr pulses, bare-groundusing the systems canopy elevation-extraction with height a small model, footprint, bare-ground algorithms due to anddifferenceselevation-extraction the shape in andthe speciesdensity algorithms of and trees. andcoverage Hunter the shape etof al. lase and[30r] species alsopulses, showed of the trees. canopy that Hunter ground height et al. based [30]model, measurementsalso bare-groundshowed that of heightelevation-extractionground exceeded based measurements LiDAR algorithms measurements of and height the of shapeexceeded height and by Lispecies anDAR average measurementsof trees. of 1.4 Hunter m. However of et heightal. [30] they byalso concluded an showed average that of itground1.4 was m. unclearHowever based whethermeasurements they concluded this was of due heightthat to it overestimationwasexceeded unclear Li DARwhether of fieldmeasurements this height was or due underestimation of to height overestimation by an byaverage LiDAR,of field of or1.4height a m. combination Howeveror underestimation they of the concluded two. by TheLiDAR, that other it or was tree a combination unclear attribute whether that of canthe this betwo. was directly The due other measuredto overestimation tree attribute with LiDAR thatof field can is crownheight diameter.or underestimation Popescu et by al. [LiDAR,14] estimated or a combination tree crown diameter of the two. to estimate The other forest tree biomass attribute and that stand can 3 3 Remote Sens. 2016, 8, 388 4 of 16 volume using multiple return, small-footprint LiDAR data. The study demonstrated that airborne laser provides the tool to reliably measure not only tree height but also crown dimensions, thus improving estimates of forest volume and biomass. Hyyppä et al. [31] estimated stem diameters using a correlation with crown diameters. Persson et al. [32] estimated tree height and crown diameter of detected trees with a RMSE of 0.63 m and 0.61 m, respectively. Lefsky et al. [33] measured the canopy structure of closed- Douglas-fir/western hemlock forests using SLICER surface LiDAR system and developed a novel technique to characterize the three-dimensional aspects of canopy structure. In Australia, numerous studies have examined LiDAR applications in , riparian zones, open woodlands and tall eucalypt forests [34–39], mostly measuring tree characteristics (both individual tree dimensions and forest stand characteristics) [40,41]. Tickle et al. [42] investigated the use of LiDAR data and large scale (1:4000) photography (LSP) to quantify the floristics and structure of mixed-species forests near Injune, central-eastern Queensland. They observed a robust relationship between LiDAR and field measurements of individual tree and stand height (r2 = 0.84–0.91). Using these relationships, they estimated floristics, height and foliage projective cover distributions for forests contained within the PSU (primary sampling unit) grid, which were then extrapolated to the entire study area. The research demonstrated that sampling using LSP and/ or LiDAR can provide quantitative assessments of floristics and key structural attributes (height, cover) at landscape scale. Lee and Lucas [43] employed an alternative approach to develop canopy height models, which was only partially useful for mapping and measuring stem dimensions in complex, multi-layered forests. They developed a Height-Scaled Crown Openness Index (HSCOI) to maximize the amount of information retrieved from scanning LiDAR for mixed-species woodlands and open-forests near Injune, Queensland. They concluded that HSCOI should not replace the CHM (canopy height model) surface but could be useful as a complementary tool for assessing forests with a highly variable structure. Although several Australian studies have used LiDAR to measure tree dimensions, no substantial work has yet been done to measure canopy volume, which is important for estimating biomass. Image data from both airborne and spaceborne remote sensing platforms have been used to determine the relationships between tree height, crown diameter and crown cover for open forests and scattered trees [44–46]. Verma et al. [46] used crown projected area and heights of individual and clustered eucalypt trees to infer DBH. Few studies have explored the feasibility of using 2D, remotely sensed, multispectral data for canopy volume estimation. Ozdemir [44] estimated tree volume from pan sharpened QuickBird imagery in Crimean juniper open-forests. Greenberg et al. [45] generated regional-scale, above-ground, biomass estimates using a novel approach with hyperspectral remotely sensed imagery. They related the area of vegetation shadow to individual tree dimensions by assuming that tree crowns were symmetrical, and concluded that the measurement of shadow was a promising technique for estimating DBH and crown area. Numerous methods for estimating tree parameters from multispectral, remotely sensed data have been trialled with varying success. Verma et al. [46,47] demonstrated that it was possible to delineate tree canopies and infer projected crown area from multispectral, remotely sensed imagery. These studies involved 2D measurements of tree parameters and could not be used directly to estimate trunk height or canopy volume. The key question, then, is whether in the absence of remotely derived measures of trunk height, we can infer canopy volume based on crown projected area and crown diameter? Few studies have explored this possibility. Popescu and Wynne [23] used pan-sharpened QuickBird imagery in Crimean juniper open-forests to predict tree volume from projected crown area with a root-mean-square error of 15.2%. More studies are required to determine whether this approach is more generally applicable. The aim of this current study was to determine the usefulness of optical remote sensing data as an alternative to LiDAR for estimation of 3D derivatives like canopy volume, by minimizing field based measurements and complex and expensive LiDAR data processing. This research aimed to directly compare the performance of two sensor systems (airborne LiDAR and spaceborne multispectral systems) and two approaches for estimating tree canopy volume of scattered Eucalyptus trees; one using regression relationships between crown diameter and crown Remote Sens. 2016, 8, 388

This research aimed to directly compare the performance of two sensor systems (airborne LiDAR and spaceborne multispectral systems) and two approaches for estimating tree canopy Remotevolume Sens. of2016 scattered, 8, 388 Eucalyptus trees; one using regression relationships between crown diameter5 of 16 and crown projected area as derived from the image-based World View-2 remote sensing system, and the other from measurements of canopy height and diameter acquired using a LiDAR system. projectedThe primary area asobjective derived was from to the investigate image-based how World well View-2canopy remotevolume sensing of scattered system, eucalypt and the trees other in frompasture measurements can be estimated of canopy using height LiDAR and and diameter satellit acquirede imagery, using using a LiDAR field-based system. measurements The primary of objectivecanopy volume was to investigate as the benchmark. how well Also, canopy owing volume to the of complexity scatteredeucalypt associated trees with in LiDAR pasture data, can bethe estimatedstudy explored using LiDARthe possibility and satellite of using imagery, multispectral using field-based imagery alone measurements to estimate of canopy canopy volume volume given as thethat benchmark. canopy volume Also, is owing a 3D toparameter. the complexity associated with LiDAR data, the study explored the possibility of using multispectral imagery alone to estimate canopy volume given that canopy volume is2. a Materials 3D parameter. and Methods

2. Materials and Methods 2.1. Study Area 2.1. StudyThe Areastudy area was the ‘SMART’ farm [48] owned and operated by the University of New EnglandThe study (UNE), area Armidale, was the ‘SMART’ New South farm Wales, [48] owned Australia and operated (longitude by the151°35 University′40′′E to of 151°37 New England′12′′E and (UNE),latitude Armidale, 30°26′09′′ NewS to 30°25 South′12 Wales,′′S; Figure Australia 3). With (longitude a total area 151 of˝35 4837140 11ha,E tothe 151 farm˝37 1includes1211E and large latitude tracts 30of˝26 natural10911S to forest 30˝25 cover11211S; with Figure several3). With Eucalyptus a total area species, of 4837 namely ha, the farm Apple includes Box (Eucalyptus large tracts bridgesiana of natural ), forestStringybark cover with (Eucalyptus several Eucalyptuscaliginosa), species,Red Gum namely (Eucalyptus Apple blakelyi Box (Eucalyptus), White Gum bridgesiana (Eucalyptus), Stringybark viminalis) (Eucalyptusand Yellow caliginosa Box (Eucalyptus), Red Gum melliodora (Eucalyptus), and blakelyi is partitioned), White Gum into (Eucalyptus grazed and viminalis ungrazed) and Yellowareas of Boxwoodland (Eucalyptus and melliodora open pasture.), and Approximately is partitioned into one grazed third andof the ungrazed area is forested; areas of woodland a third is andwoodland, open pasture.and the Approximately remainder native one thirdpasture. of the Most area isof forested;the property a third is is woodland,managed in and an the agriculturally remainder nativeunmanipulated pasture. Most manner of the other property than grazing. is managed Most in of an th agriculturallye study area is unmanipulated unfertilized or manner little fertilized other thannatural grazing. pasture Most grazed of the by study sheep. area The is unfertilizedportion of the or littlestudy fertilized area chosen natural for pasturethe collection grazed of by laser sheep. and Theground portion datasets of the studywas approximately area chosen for 200 the ha collection (study site, of laser Figure and 3), ground limited datasets in size was by the approximately LiDAR data 200acquisition ha (study footprint. site, Figure 3), limited in size by the LiDAR data acquisition footprint.

FigureFigure 3. 3.Location Locationmap map of of the the study study site site in in north-eastern north-eastern NSW, NSW, Australia. Australia. Source: Source: open open access access data. data.

2.2. Field Measurements of Canopy Volume High spatial resolution (15 cm), colour infrared (CIR) airborne imagery of the study area was first used to identify candidate trees in the study site, the coordinates of which were later used to identify trees in the field using GPS based positioning.5 Single eucalypt trees were selected randomly using the orthorectified imagery, and care was taken that the trees were well distributed across Remote Sens. 2016, 8, 388 6 of 16 the study site. Field measurements were made between September and December 2012 (spring to summer, i.e., ‘leaf on’, as eucalypts are evergreen). In order to establish the relationship between crown variables, other tree dimensions and canopy volume, 79 trees of five Eucalyptus species were sampled. The number of each species varied depending on occurrence across the study area. Tree height (TH), crown diameter (CD) and canopy height (CH) were manually measured using a laser rangefinder (MDL LaserAce 300, Measurement Devices Ltd., Scotland, UK) and clinometer. Tree height was measured directly with the rangefinder by taking two measurements. Tree height measured using the clinometer was used to compare the performance of both instruments, which were very similar. Canopy height was measured by measuring trunk height, and subtracting it from total tree height (Figure2). In order to measure crown projected area (CA), the crown diameter of individual trees was measured following the protocol described in Verma et al. [46]: firstly a pair of vertical range poles were placed in the ground to delineate the edges of the tree/cluster canopy along a cardinal axis, for example N-S, passing through the tree stem or the estimated centre of the cluster. The crown periphery for locating the pole positions was located using a clinometer set to ‘vertical view, in effect acting as a crown mirror. The crown diameter (d) along this direction was then measured using a laser range finder positioned at right-angles to, and a known distance well back from, the line between the poles. This measurement avoided errors that would otherwise be incurred by using tapes to measure the straight-line distance between the poles with tree stems in between. This measurement was undertaken for six cardinal directions namely N, ENE, ESE, S, WSW and WNW, respectively [46], and the average diameter, which is effectively the average crown spread, d, calculated [49]. The crown projected area was then calculated using:

CA “ πd2{4 (2)

Field-measured canopy volume (CVfield) of individual trees was calculated using Equation (1). A visual assessment of the canopies determined that the trees in the study area were parabolic in shape. According to the work of Coder [4] the volume can then be approximated as a paraboloid with a multiplier value (Equation (1)) of 0.3927. Regression equations were developed between canopy volume (CVfield), and crown diameter (CDfield) and projected crown area (CAfield), to determine canopy volume from the satellite data, which was able to provide estimates of CD and CA (Section 2.4) but unable to provide measurements of canopy height (discussed later).

2.3. LiDAR Data Acquisition and Post-Processing The Airborne Laser Scanning system used for the project was the Trimble Harrier 68i/G1 system flown on 1 June 2013. It consisted of a Riegl LiDAR scanning instrument, Applanix POS/AV 410 Inertial Motion System, with 12 channels and a dual frequency GPS receiver. The full waveform LiDAR dataset had the parameters in Table1.

Table 1. LiDAR data acquisition parameters.

Parameter Value Unit Scanning angle 60 degrees Flight speed 216 km¨h´1 Flight height 375 m Scan rate 192 Hz Pulse rate 400 kHz Swath width 433 m Swath overlap 37 % Along track point spacing 0.31 m (along track) Across track point spacing 0.31 m (across track) Outgoing pulse density 10.26 m´2 Calculated spot footprint 0.19 metres Remote Sens. 2016, 8, 388

Table 1. LiDAR data acquisition parameters.

Parameter Value Unit

Remote Sens. 2016, 8, 388 Scanning angle 60 degrees 7 of 16 Flight speed 216 km∙h−1 Flight height 375 m The LiDAR data were providedScan rate in LiDAR Exchange 192 Format (LAS), Hz having first been classified into ground and non-groundPulse points rate by the data provider 400 using proprietary kHz software (Terrascan). An intensity image was createdSwath from width the point clouds. 433 The selected individual m trees measured in the field were then identified inSwath the point overlap cloud data. The 37 tree height and % trunk height were measured for each tree using theAlong software track FUSION/LDV point spacing (Robert 0.31 J. McGaughey, m (along Pacifictrack) Northwest Research Station, Version 3.10, BuildAcross date track 16 Maypoint 2012; spacing Figure 4). Canopy 0.31 height m (across was track) calculated as the difference between tree height andOutgoing trunk height. pulse density Tree crown parameters 10.26 were extracted m−2 automatically by the method of image segmentation.Calculated spot footprint 0.19 metres

FigureFigure 4. Tree 4. Tree dimensions dimensions (total (total tree tree height height andand trunktrunk height) height) from from FUSION/LDV FUSION/LDV software. software. The The colourscolours green, green, yellow yellow and and red red represents represents the the canopy canopy at at different heights.heights. Blue Blue represents represents ground ground level. level.

The Theautomatic automatic extraction extraction of ofcrown crown parameters parameters fr fromom LiDARLiDAR was was a a two-step two-step process process involving involving (i) (i) CanopyCanopy Height Height Model Model generationgeneration (CHM); (CHM); and (ii)and segmentation (ii) segmentation of the CHM of image the into CHM homogeneous image into homogeneousobjects representing objects representing individual tree individual canopies. The tree CHM canopies. generation The was CHM carried generation out using was interpolated carried out usingtwo interpolated grid inputs, two namely grid a Digitalinputs, Terrainnamely Model a Digital (DTM) Terrain and a DigitalModel Surface (DTM) Model and a (DSM) Digital of 1Surface m Modelspatial (DSM) resolution. of 1 m Thespatial DTM resolution. was a digital The representation DTM was ofa thedigital topographic representation surface whereof the the topographic height surfacevalues where in the the terrain height were values known in [50the]. DSMterrain was we are representation known [50]. of DSM the features was a aboverepresentation ground level. of the featuresA canopy above height ground model, level. also A referred canopy to height as a normalized model, also digital referred surface to model as a normalized (nDSM), was digital created surface by modelsubtracting (nDSM), thewas DTM created from theby DSM.subtracting The vertical the DTM resolution from of the the DSM. CHM wasThe importantvertical resolution as extraction of the of the tree height information largely depended on the accuracy of the CHM. The LiDAR data set CHM was important as extraction of the tree height information largely depended on the accuracy of for this study had a vertical accuracy better than ˘30 cm (at 95% confidence level) and horizontal the CHM. The LiDAR data set for this study had a vertical accuracy better than ±30 cm (at 95% accuracy better than ˘80 cm (at 95% confidence level). Since both the DTM and the DSM were created confidenceusing interpolation, level) and horizontal and given the accuracy additive better nature th ofan the ±80 vertical cm errors(at 95% from confidence a number oflevel). contributing Since both the DTMprocesses and the in calculating DSM were the created CHM, using this yields interpolation, a vertical error and ofgiven approximately the additive 1 m. nature Theprocessing of the vertical errorswas from done a number in ArcGIS of version contributing 10. The processes image segmentation in calculating (Object the based CHM, classification) this yields softwarea vertical used error of approximatelyto extract trees 1 m. from The the processing CHM was was eCognition done in (eCognition ArcGIS version Developer 10. The 8, Munich, image segmentation Germany, GmbH), (Object basedwhich classification) offers a wide software range of segmentationused to extract algorithms. trees from Segmentation the CHM isthe was critical eCognition first step in(eCognition object Developerbased classification,8, Munich, andGermany, reduces GmbH), the image which into discrete offers regions a wide or objectsrange thatof segmentation are homogeneous algorithms. with Segmentationregard to spatialis the or critical spectral first characteristics step in object [51]. Thebased image classification, segmentation and was reduces carried out the using image the into discrete‘fractal regions net evolution or objects approach’ that are (FNEA), homogeneous which relies with on using regard a set to of sp user-definedatial or spectral parameter characteristics settings, [51]. The image segmentation was carried out using the ‘fractal net evolution approach’ (FNEA),

7 Remote Sens. 2016, 8, 388 which relies on using a set of user-defined parameter settings, which affects the segmentation results Remote Sens. 2016, 8, 388 8 of 16 [52,53]. The important outcome of the segmentation process is to identify objects that are representative of the features. The derived CHM was a grid in a single band image with the objects partitionedwhich affects into thelighter segmentation and darker results areas. [52, 53Therefor]. The importante, a ‘contrast outcome split’ of thesegmentation segmentation algorithm process in eCognitionis to identify was employed objects that for are tree representative extraction. of The the features.algorithm The splits derived bright CHM and was dark a grid objects in a single in image basedband on imagea threshold with the that objects maximizes partitioned the contrast into lighter between and darker the resulting areas. Therefore, bright objects a ‘contrast (pixel split’ values segmentation algorithm in eCognition was employed for tree extraction. The algorithm splits bright above threshold) and dark objects (pixel values below the threshold). The algorithm evaluates the and dark objects in image based on a threshold that maximizes the contrast between the resulting optimal threshold separately for each image object in the image object domain (set of image objects). bright objects (pixel values above threshold) and dark objects (pixel values below the threshold). Firstly,The the algorithm algorithm evaluates executes the optimal a chessboard threshold segmen separatelytation for (splits each image the pixels object indomain the image into object a square imagedomain object), (set and ofimage then performs objects). Firstly, the split the on algorithm each square. executes Secondly, a chessboard to optimize segmentation this separation, (splits the the algorithmpixels domainconsiders into different a square pixel image values object), as anda pote thenntial performs threshold, the splitwithin on the each range square. provided Secondly, by the user toparameters optimize this and separation, with values the selected algorithm based considers on the different input pixelstep size values and as stepping a potential parameter. threshold, The test thresholdwithin the rangecausing provided the largest by the contrast user parameters is chosen and as withthe best values threshold selected and based used on thefor inputsplitting. step The test thresholdssize and stepping range parameter.from the minimum The test threshold threshol causingd to the the maximum largest contrast threshol is chosend, with as intermediate the best valuesthreshold chosen andaccording used for to splitting. the step Thesize test(the thresholdssizes of steps range used from by the the minimum algorithm threshold to move to from the the minimummaximum threshold threshold, to withthe intermediatemaximum threshold), values chosen and according step parameters to the step sizecalculate (the sizes each of steps step by used by the algorithm to move from the minimum threshold to the maximum threshold), and step addition or multiplication (the value is either added to the threshold or multiplied by the threshold). parameters calculate each step by addition or multiplication (the value is either added to the threshold Finally, the mean of possible bright and dark border pixels is used to calculate either the edge ratio or multiplied by the threshold). Finally, the mean of possible bright and dark border pixels is used to or edgecalculate difference, either thewhich edge is ratio then or used edge difference,to define the which contrast is then between used to define two objects. the contrast The between selection of segmentationtwo objects. parameters The selection was of segmentationhighly subjective parameters and determined was highly subjective through anda combination determined throughof trial and error,a combinationand ultimately of trial user and experience error, and [47]. ultimately For th useris study, experience a minimum [47]. For and this maximum study, a minimum threshold and of 20 and maximum110, respectively, threshold a of step 20 and size 110, of respectively, 5, and an aadd step function size of 5, andfor the an add stepping function type for theparameter stepping was foundtype suitable parameter and was effective found suitablefor tree and extraction. effective forFigure tree extraction.5 shows the Figure different5 shows steps the different of contrast steps split algorithmof contrast used split in the algorithm study and used Figu in there study 6 shows and the Figure segmented6 shows the objects. segmented objects.

FigureFigure 5. Contrast 5. Contrast split split segmentation segmentation steps steps applied applied on singleon single band canopyband canopy height model height (CHM) model image (CHM) imageto splitsto splits bright bright and darkand dark objects objects based onbased a threshold on a thre thatshold maximized that maximized the contrast the between contrast the resultingbetween the resultingbright bright and dark and objects. dark objects.

8 Remote Sens. 2016, 8, 388 9 of 16 Remote Sens. 2016, 8, 388

(a) (b)

FigureFigure 6.6.An An example example of of the the (a) ( LiDAR-deriveda) LiDAR-derived canopy canopy height height model; model; and (b and) the ( segmentationb) the segmentation results. results.

The tree polygons were exported to ArcGIS 10 and the crown diameter (CDLiDAR) of each tree The tree polygons were exported to ArcGIS 10 and the crown diameter (CDLiDAR) of each tree was estimated by measuring the lengths of the major and minor axes of the canopy polygon and was estimated by measuring the lengths of the major and minor axes of the canopy polygon and averaging the two. Canopy heights were estimated by subtracting trunk height from tree height. averaging the two. Canopy heights were estimated by subtracting trunk height from tree height. The The canopy volume (CVLiDAR) was then calculated using Equation (1) with the same multiplier used canopy volume (CVLiDAR) was then calculated using Equation (1) with the same multiplier used for for the field-based measurements. the field-based measurements. 2.4. Delineation of Tree Attributes from WorldView2 Data 2.4. Delineation of Tree Attributes from WorldView2 Data A multispectral, PAN-sharpened, WorldView2 image (8-bit, 1 January 2012) was acquired with A multispectral, PAN-sharpened, WorldView2 image (8-bit, 1 January 2012) was acquired with a spatial resolution of approximately 50 cm in four spectral bands, Band 1 (NIR 0.7–1 µm), Band 2 a spatial resolution of approximately 50 cm in four spectral bands, Band 1 (NIR 0.7–1 μm), Band 2 (Red 0.6–0.7 µm), Band 3 (Green 0.5–0.6 µm) and Band 4 (0.4–0.5 µm). The image coordinate reference (Red 0.6–0.7 μm), Band 3 (Green 0.5–0.6 μm) and Band 4 (0.4–0.5 μm). The image coordinate system was WGS 84 UTM Zone 56 S. The crown projected area of each tree (CAWV2) was determined reference system was WGS 84 UTM Zone 56 S. The crown projected area of each tree (CAWV2) was manually by onscreen digitization/segmentation of the canopies (ArcGIS version 10) and counting the determined manually by onscreen digitization/segmentation of the canopies (ArcGIS version 10) canopy pixels, assuming a pixel dimension of 50 cm ˆ 50 cm. The crown diameter (CDWV2) of each and counting the canopy pixels, assuming a pixel dimension of 50 cm × 50 cm. The crown diameter tree was determined by measuring the lengths of the major and minor axes of the canopy polygon (CDWV2) of each tree was determined by measuring the lengths of the major and minor axes of the and averaging the two. In order to estimate canopy volume from the WorldView-2 imagery (CVWV2), canopy polygon and averaging the two. In order to estimate canopy volume from the WorldView-2 the derived values of CDWV2 and CAWV2 were substituted into the regression equation developed imagery (CVWV2), the derived values of CDWV2 and CAWV2 were substituted into the regression between on-ground measurements of canopy volume (CVfield), and crown diameter (CDfield) and equation developed between on-ground measurements of canopy volume (CVfield), and crown crown projected area (CAfield). diameter (CDfield) and crown projected area (CAfield). 2.5. Evaluating the Performance of the Two Techniques 2.5. Evaluating the Performance of the Two Techniques The derived canopy volumes for each remote sensing method (CVLiDAR, CVWV2) were The derived canopy volumes for each remote sensing method (CVLiDAR, CVWV2) were compared compared to the field-measured values (CVfield) and a root mean squared prediction error, to the field-measured values 2(CVfield) and a root mean squared prediction error, RMSE “ CVpredicted ´ CVactual , calculated. Analysis and model evaluation was done using RMSE =cCV – CV , calculated. Analysis and model evaluation was done using the the statistical´ softwarepredicted R (Studioactual Version¯ 0.97.318). statistical software R (Studio Version 0.97.318). 3. Results 3. Results The analysis results of field and LiDAR measurements and canopy volume estimations are presentedThe analysis in this section. results of field and LiDAR measurements and canopy volume estimations are presented in this section. 3.1. Field Measurements of Tree Parameters 3.1. FieldSummary Measurements statistics of ofTree the Parameters measured trees are given in Table2. Due to the variability in tree species,Summary and the statistics irregular of dimensionsthe measured of trees the treesare give characteristics,n in Table 2. treeDue heightto the andvariability trunk heightin tree (species,i.e., tree heightand the´ irregularcanopy height) dimensions exhibited of the significant trees characteristics, variability (approx. tree height 25% andand 15%, trunk respectively). height (i.e. Fieldtree height measurements − canopy of height) tree height exhibited ranged significant from 12.7 mvariability to 42.8 m (approx. with a SD 25% of 5.3and m 15%, while respectively). trunk height measurementsField measurements ranged of from tree 3.6height m to ranged 16.2 m withfrom a 12.7 SD ofm 2.6to m.42.8 m with a SD of 5.3 m while trunk height measurements ranged from 3.6 m to 16.2 m with a SD of 2.6 m.

9 Remote Sens. 2016, 8, 388 RemoteRemote Sens. Sens.2016 2016, 8,, 3888, 388 10 of 16

Noting that the diameter measurements of each canopy (CDfield) were acquired for 6 cardinal directionsNoting and that the the average diameter calculated, measurements the standard of each deviation canopy of(CD thisfield average) were acquired diameter for value 6 cardinal varied Table 2. Summary statistics for single trees from field measurements; n = 79 was the number of trees directionsbetween 0.4% and and the 34%.average calculated, the standard deviation of this average diameter value varied betweenused in 0.4% model and development. 34%. Table 2. Summary statistics for single trees from field measurements; n = 79 was the number of trees Tree Characteristics Min Max Mean Std.Dev. Tableused in 2. model Summary development. statistics for single trees from field measurements; n = 79 was the number of trees used in modelCrown development. diameter (CDfield, m) 6.8 30.5 15.2 4.9 2 CrownTree projected Characteristics area (CAfield, m ) 36.3 Min 731.8 Max 210.4 Mean 129.1 Std.Dev. CrownTreeTree diameter Characteristics height (CD (m)field, m) 12.7 Min 6.8 42.8 Max 30.5 21.3 Mean 15.2 5.3 Std.Dev. 4.9 CrownCrown Canopyprojected diameter height area (CD (CA (m)fieldfield, m), m 2) 8.836.3 6.8 30.6731.8 30.5 16.1210.4 15.2 4.1 129.1 4.9 Crown projectedTrunkTree height Height area (m) (m)(CA field, m2) 3.636.3 12.7 16.2731.8 42.8 5.2210.4 21.3 2.6 129.1 5.3 3 217.9 9040.8 1840.2 1533.1 Canopy vol.CanopyTree (CV fieldheight height, m (m)) (Equation(m) (1)) 12.7 8.8 42.830.6 21.316.1 4.15.3 CanopyTrunk Height height (m)(m) 3.68.8 16.230.6 16.1 5.2 2.64.1 Canopy vol.Trunk (CV fieldHeight, m3) (m)(Equation (1)) 217.9 3.6 9040.8 16.2 1840.2 5.2 1533.1 2.6 Noting that the diameter measurements of each canopy (CDfield) were acquired for 6 cardinal Canopy vol. (CVfield, m3) (Equation (1)) 217.9 9040.8 1840.2 1533.1 directions and the average calculated, the standard deviation ofthis average diameter value varied between 0.4% and 34%. 3.2. Canopy Volume Estimations 3.2.3.2. Canopy Canopy Volume Volume Estimations Estimations Scatter plots of canopy volume (CVfield) versus crown diameter (CDfield) and crown projected area Scatter plots of canopy volume (CVfield) versus crown diameter (CDfield) and crown projected area (CAScatterfield) are plots given of canopy in Figures volume 7 and (CV 8,field respectively,) versus crown along diameter with (CDthe fieldbest-fit,) and polynomial crown projected regression area (CAfield) are given in Figures 7 and 8, respectively, along with the best-fit, polynomial regression (CAcurves.field) are The given derived in Figures regression7 and 8equations respectively, correspon along withding theto these best-fit, curves polynomial are given regression in Table curves.3. Thecurves. derived The regression derived regression equations equations corresponding correspon to theseding curves to these are curves given inare Table given3. in Table 3.

Figure 7. Scatterplot between canopy volume (CVfield), calculated using Equation (1), and measured Figure 7. Scatterplot between canopy volume (CVfield), calculated using Equation (1), and measured crownFigure diameter7. Scatterplot (CD fieldbetween). The solid canopy curve volume is the (CVbest-fit,field), polynomialcalculated using regression Equation equation. (1), and measured crown diameter (CD ). The solid curve is the best-fit, polynomial regression equation. crown diameter (CDfieldfield). The solid curve is the best-fit, polynomial regression equation.

Figure 8. Scatterplot between canopy volume (CVfield), as calculated using Equation (1), and measured Figure 8. Scatterplot between canopy volume (CVfield), as calculated using Equation (1), and crown projected area (CAfield) derived from field measurements. The solid curve is the best-fit, measuredFigure 8. crownScatterplot projected between area canopy(CAfield ) volumederived (CVfromfield field), as measurem calculatedents. using The Equationsolid curve (1), is andthe polynomial regression equation. measuredbest-fit, polynomial crown projected regression area equation. (CAfield) derived from field measurements. The solid curve is the best-fit, polynomial regression equation. 10 10 Remote Sens. 2016, 8, 388 Remote Sens. 2016, 8, 388 11 of 16

Table 3. Derived best-fit regression parameters for calculating canopy volume (CVfield) from

EquationTable 3. (1)Derived using best-fit field measuremen regression parametersts of crown for calculatingprojected canopyarea (CA volumefield) and (CV crownfield) from diameter Equation (CDfield). Multiplier(1) using fieldvalue measurements = 0.3927 (n = of 79). crown projected area (CAfield) and crown diameter (CDfield). Multiplier value = 0.3927 (n = 79). Equation R2 F-stat p 2 2 CVfield = 0.008 × (CA fieldEquation) + 6.5673 × – 25.199 R 0.93 F-stat993.0 p<0.0001 2 CVfield = 15.11 × (field)2 – 218.58 × CD field + 1222.6 0.94 415.6 <0.0001 CVfield = 0.008 ˆ (CAfield) + 6.5673 ˆ CAfield – 25.199 0.93 993.0 <0.0001 2 CVfield = 15.11 ˆ (CDfield) – 218.58 ˆ CDfield + 1222.6 0.94 415.6 <0.0001 The two models represented in Table 3 indicated that CV can be estimated from optical remote sensingThe datasets two models using represented CA and CD in as Table these3 indicated are the thatvariables CV can which be estimated can directly from optical be measured remote from opticalsensing remote datasets sensing using data. CA and However, CD as these the CV are estima the variablestes using which CD can (F = directly 415.6) beis better measured than from using CA (F =optical 993.0). remote The application sensing data. of However, the two theregression CV estimates equations using CDin Table (F = 415.6) 3 to isestimate better than canopy using volume CA is depicted(F = 993.0). in Figure The application 9, which of compares the two regression the predicted equations canopy in Table volumes3 to estimate (CVWV2 canopy) against volume the field measuredis depicted values, in Figure the9 ,latter which including compares thethe predicted canopy canopyheight volumes parameter (CV WV2from) against Equation the field(1). Both scatterplotsmeasured values,include the a 1:1 latter equivalence including the line canopy (dashed height line) parameter and a best-fit from Equation regression (1). Both curve. scatterplots The accuracy of includeusing WorldView-2 a 1:1 equivalence imagery line (dashed to infer line) canopy and a best-fitvolume regression from CA curve. and TheCD accuracydiffered, of as using shown in WorldView-2 imagery to infer canopy volume from CA and CD differed, as shown in Figure9a,b, Figure 9a,b, respectively. An RMSE of 781.03 m3 (42%) was observed with CA as the predictor respectively. An RMSE of 781.03 m3 (42%) was observed with CA as the predictor variable, compared 3 variable, compared to 3an RMSE of 575.5 m (31%) with CD as the predictor. The best-fit regression of to an RMSE of 575.5 m (31%) with CD as the predictor. The best-fit regression of CA-derived CVWV2 on CA-derived CVWV2 on CVfield was a power function and explained 65% of the variance, whereas a CVfield was a power function and explained 65% of the variance, whereas a second-order polynomial second-order polynomial regression of CD-derived CVWV2 on CVfield explained 76% of the variance. regression of CD-derived CVWV2 on CVfield explained 76% of the variance.

(a) (b)

FigureFigure 9.9. ScatterplotsScatterplots ofof canopy canopy volume volume from from WV2 WV2 using (usinga) crown (a) projectedcrown projected area (CA); andarea ( b(CA);) crown and (b) crowndiameter diameter (CD) with (CD) crown with volume crown (CV) volume as the (CV) predictor as the variable predictor from fieldvariable measurements. from field The measurements. dashed Thelines dashed are the lines 1:1 equivalenceare the 1:1 equivalence between measured between and me predictedasured valuesand predicted and the solid values lines and the the best-fit solid lines theregression best-fit regression curves (power curves and (power polynomial, and respectively).polynomial, respectively).

FigureFigure 10 10 shows shows the the scatterplot scatterplot ofof LiDAR-derivedLiDAR-derived canopy canopy volume volumeversus versusfield-measured field-measured CV. CV. CVCVLiDARLiDAR waswas calculated calculated from from EquationEquation (1) (1) using using both both canopy canopy height height and crown and diameter crown deriveddiameter from derived 3 fromthe the LiDAR LiDAR data. data. Despite Despite the lower the RMSE lower for RMSE the LiDAR-derived for the LiDAR-derived canopy volume canopy estimates volume (490.8 estimates m , 26% error), CV systematically underestimated CV at larger canopy volumes (>4000 m3). (490.8 m3, 26% error),LiDAR CVLiDAR systematically underestimatedfield CVfield at larger canopy volumes (>4000 The individual parameters, CD and CH, extracted from the LiDAR data are plotted against the m3). corresponding field measurements in Figure 11a,b. The LiDAR method estimated tree heights with an The individual parameters, CD and CH, extracted from the LiDAR data are plotted against the RMSE of 1.44 m (error of 6.5%) when compared to the field measurements. corresponding field measurements in Figure 11a,b. The LiDAR method estimated tree heights with an RMSE of 1.44 m (error of 6.5%) when compared to the field measurements.

11 RemoteRemote Sens. Sens. 20162016, 8,,8 388, 388 12 of 16 Remote Sens. 2016, 8, 388

RMSE = 491 m3 RMSE = 491 m3

FigureFigure 10. 10. ScatterplotScatterplot between between canopy canopy volume volume predicte predictedd using using the the LiDAR-derived LiDAR-derived values values of of canopy canopy Figure 10. Scatterplot between canopy volume predicted using the LiDAR-derived values of canopy heightheight and and crown crown diameter diameter (Equation (Equation (1)) (1)) and and the the field field measured measured values. values. The Thedashed dashed line is line the is 1:1 the height and crown diameter (Equation (1)) and the field measured values. The dashed line is the 1:1 equivalence1:1 equivalence between between measured measured and predicted and predicted values values and the and solid the line solid is the line best-fit is the best-fitregression regression curve equivalence between measured and predicted values and the solid line is the best-fit regression curve (polynomial).curve (polynomial). (polynomial).

(a) (b) (a) (b) FigureFigure 11. 11. ScatterplotsScatterplots of (a)of LiDAR-derived (a) LiDAR-derived crown diameter crown diameter and field and measurements; field measurements; and (b) Figure 11. Scatterplots of (a) LiDAR-derived crown diameter and field measurements; and (b) LiDAR-derivedand (b) LiDAR-derived canopy canopyheight and height field and measurem field measurements.ents. The dashed The dashed lines lines are arethe the 1:1 1:1 equivalence equivalence LiDAR-derived canopy height and field measurements. The dashed lines are the 1:1 equivalence betweenbetween measured measured and and derived derived values values and and the the solid solid lines lines the the best-fit best-fit regression regression curves curves (liner (liner and and between measured and derived values and the solid lines the best-fit regression curves (liner and polynomial,polynomial, respectively). respectively). polynomial, respectively). 4. Discussion 4.4. Discussion Discussion As the field measurements of canopy height were acquired using a digital clinometer and AsAs the the field field measurements measurements of of canopy canopy height height were were acquired acquired using using a a digital digital clinometer clinometer and and rangefinder, there was a potential error in overestimating the height associated with locating the rangefinder,rangefinder, there there was was a a potential potential error error in in overestimating overestimating the the height height associated associated with with locating locating the the very top of the tree canopy from ground level. It can be shown that for spherical canopies the veryvery top top of theof the tree tree canopy canopy from from ground ground level. level. It can It be can shown be thatshown for that spherical for spherical canopies canopies the relative the relative error is proportional to the product of the height and radius of the (assumed spherical) errorrelative is proportional error is proportional to the product to the of theproduct height of and the radius height of and the radius (assumed of the spherical) (assumed canopy spherical) and canopy and inversely proportional to the baseline distance from the clinometer/rangefinder and the inverselycanopy and proportional inversely toproportional the baseline to distance the baseline from distance the clinometer/rangefinder from the clinometer/rangefinder and the base and of the the base of the tree. Given the dimensions of the trees in question and the baseline distances used, this tree.base Given of the the tree. dimensions Given the of dimensions the trees in questionof the trees and in the question baseline and distances the baseline used, thisdistances can be used, as large this can be as large as 40%. For parabolic canopies this overestimation can be significantly less. In this ascan 40%. be For as paraboliclarge as 40%. canopies For parabo this overestimationlic canopies canthis beoverestimation significantlyless. can Inbe thissignificantly work, and less. given In the this work, and given the parabolic nature of the canopies observed, we estimate the propensity for parabolicwork, and nature given of thethe canopies parabolic observed, nature weof the estimate canopies the propensityobserved, forwe overestimatingestimate the propensity the height offor overestimating the height of the trees to be ~10%. The irregular shape of the crowns of the individual theoverestimating trees to be ~10%. the Theheight irregular of the trees shape to of be the ~10%. crowns The of irregular the individual shape treesof the was crowns also of apparent the individual in the trees was also apparent in the calculation of the individual crown diameters. calculationtrees was ofalso the apparent individual in the crown calculat diameters.ion of the individual crown diameters. Both Figures 7 and 8 and the regression statistics in Table 3 indicated that the canopy volume of Both Figures 7 and 8 and the regression statistics in Table 3 indicated that the canopy volume of Eucalyptus trees can be inferred using crown diameter or crown projected area, without the need for Eucalyptus trees can be inferred using crown diameter or crown projected area, without the need for 12 12 Remote Sens. 2016, 8, 388 13 of 16

Both Figures7 and8 and the regression statistics in Table3 indicated that the canopy volume of Eucalyptus trees can be inferred using crown diameter or crown projected area, without the need for measuring canopy height, which bodes well for using remotely sensed imaging systems. Both equations in Table3 tended to overestimate canopy volume measured in the field, most likely due to the visual classification of mixed boundary pixels as canopy. As the imagery was orthorectified, it was often difficult to determine whether the canopy fringes observed in the imagery were parts of the crown or shadow cast upon the underlying ground surface. If shadow fringes were generally classified as tree canopy, both CA and CD would have been overestimated, yielding higher CV values. The CA-derived CVWV2 estimates were higher than the corresponding CD-derived CVWV2 values because of the additive effect of misclassified pixels in calculating CA. The crown diameter measure, on the other hand, was only an average of two canopy (major and minor) axes. Ideally, an objective classification procedure would apportion intermediate pixels according to the level of mixing, and this is recommended for further work. At higher canopy volumes (>4000 m3, i.e., trees with huge crowns), both techniques tended to underestimate CVfield (probably due to additional textural effect), although there was considerable spread in the predicted versus actual values for larger canopies. Given the potential errors associated with the clinometer/rangefinder measurements, it is not surprising that the field values exceeded the LiDAR values for taller trees. Of course the LiDAR method may also be contributing to the progressively larger difference between the remote sensing and field based measurements owing to the fact that the larger trees tend to exhibit greater porosity (visual assessment only) and this may result in ‘pits’ in the LiDAR surface profiles. Studies by Rönnholm et al. and Huang et al. [28,29] have also showed that tree heights are typically underestimated by small footprint laser scanning systems for several reasons: (a) variability in the density and coverage of laser pulses; (b) differences in the algorithms used to obtain the canopy height model; (c) the amount and height of understory vegetation obscuring the ground surface; (d) differences in algorithms used to calculate the bare ground elevation; (e) the sensitivity of the laser system and the algorithms used for signal processing; and (f) lastly the tree shape and tree species. The present study used full waveform LiDAR with a point density of 10 m´2, which should have been sufficient to generate accurate estimates of tree height and other tree dimensions [14]. The LiDAR contribution to the discrepancy (i.e., underestimation) was most likely due to the flying height. Other factors that influence the crown diameter and canopy height measurements may be the number of pulses hitting the canopy and understorey, which would fail to provide an accurate estimate of the trunk height, and due to confusion in estimating the lower limit of the canopy. Given the diameter value is squared in Equation (1), even a minor underestimation is exaggerated.

5. Conclusions Field-based measurements provide the best estimate of the structural dimensions of forests but are often expensive, labour-intensive and time consuming to collect. Several studies have used LiDAR to measure tree dimensions with high accuracy, but the price and complexity associated with LiDAR acquisition and processing restrict its application. Therefore, in the quest for cheaper and more convenient means of tree parameter estimation, this study estimated the canopy volume of individual Eucalyptus trees using crown diameter and crown projected area derived from WorldView-2 (WV2) data. The results showed that WV2-derived crown projected area and crown diameter were strongly correlated with canopy volume, and that crown diameter yielded better results (RMSE 31%) than crown projected area (RMSE 42%). The results indicated the possible use of high-resolution optical data as an alternative to existing methods for measuring 3D attributes of scattered trees and spatially separated tree copses. The remote sensing advantages associated with the use of optical data include avoiding the high cost of LiDAR and associated complex data processing, as well as the ready availability of optical data compared to LiDAR. We note that our results may not be applicable to forests with closed canopies, as understorey is also an important component of forest ecosystems. Our approach could also perhaps be enhanced by incorporating terrain characteristics like slope, aspect, Remote Sens. 2016, 8, 388 14 of 16 rainfall, etc., which affect volume estimates. In addition, this study used contrast split segmentation for LiDAR CHM data segmentation, which is relatively unexplored but could be more frequently used, based on our results, compared to the widely used multi-resolution rule-based segmentation. Future research should extend our study by testing remotely sensed datasets of varying resolution against volume estimates in other forest and savanna types.

Acknowledgments: This work was partially funded by the CRC for Spatial Information (CRCSI), established and supported under the Australian Government Cooperative Research Centres Programme. One of the authors (NKV) wishes to acknowledge the receipt of a Postgraduate ‘Top-up’ Scholarship from the CRCSI. Author Contributions: Conceived and designed the experiments: Niva Kiran Verma and David Lamb. Performed the experiments: Niva Kiran Verma. Analysed the data: Niva Kiran Verma. Wrote the paper: Niva Kiran Verma, David Lamb, Nick Reid. Edited: Nick Reid, Brian Wilson. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Zaman, Q.; Schuman, A.W. Performance of an ultrasonic tree volume measurement system in commercial citrus groves. Precis. Agric. 2005, 6, 467–480. [CrossRef] 2. Usha, K.; Singh, B. Potential applications of remote sensing in horticulture—A review. Sci. Hortic. 2013, 153, 71–83. [CrossRef] 3. Dubrasich, M.E.; Hann, D.W.; Tappeineer, J.C., II. Methods for evaluating crown area profiles of forest stands. Can. J. For. Res. 1997, 27, 385–392. [CrossRef] 4. Coder, K.D. Tree Biomechanics Series: Crown Shape Factors & Volumes. Available online: http://www. .uga.edu/warnell/service/library/for00--032/2000 (accessed on 27 January 2014). 5. Troxel, B.; Piana, M.; Ashton, M.S.; Dunning, C.M. Relationships between bole and crown size for young urban trees in the northeastern USA. Urban For. Urban Green. 2013, 12, 144–153. [CrossRef] 6. Rautiainen, M.; Mõttus, M.; Stenberg, P.; Ervasti, S. Crown envelope shape measurements and models. Silv. Fenn. 2007, 42, 19–33. [CrossRef] 7. Tumbo, S.D.; Salyani, M.; Whitney, J.D.; Wheaton, T.A.; Miller, W.M. Investigation of laser and ultrasonic ranging sensors for measurements of citrus canopy volume. Appl. Eng. Agric. 2002, 18, 367–372. [CrossRef] 8. Ross, N. Modeling forest canopy heights: The effects of canopy shape. Remote Sens. Environ. 1997, 60, 327–334. 9. Eamus, D.; McGuinness, K.; Burrows, W. Review of Allometric Relationships for Estimating Woody Biomass for Queensland, the Northern Territory and Western Australia; National Carbon Accounting System Technical Report No. 5A. Australian Greenhouse Office: Canberra, Australia, 2000. 10. Keith, H.; Barrett, D.; Keenan, R. Review of Allometric Relationships for Estimating Woody Biomass for New South Wales, The Australian Capital Territory, Victoria, Tasmania and South Australia; National Carbon Accounting System Technical Report No. 5B. Australian Greenhouse Office: Canberra, Australia, 2000. 11. Kini, A.U.; Popescu, S.C. TreeVaW: A versatile tool for Analysing Forest Canopy LiDAR Data—A preview with an eye towards the future. In Proceedings of the ASPRS Fall Conference, Kansas City, MO, USA, 12–16 September 2004; pp. 1–10. 12. Lefsky, M.A.; Cohen, W.B.; Parker, G.G.; Harding, D.J. LiDAR remote sensing for ecosystem studies. Bioscience 2002, 52, 19–30. [CrossRef] 13. Naesset, E.; Okland, T. Estimating tree height and tree crown properties using airborne scanning laser in a Boreat nature reserve. Remote Sens. Environ. 2002, 79, 105–115. [CrossRef] 14. Popescu, S.C.; Wynne, R.H.; Nelson, R.F. Measuring individual tree crown diameter with LiDAR and assessing its influence on estimating forest volume and biomass. Can. J. Remote Sens. 2003, 29, 564–577. [CrossRef] 15. Hemery, G.E.; Savill, P.S.; Pryor, S.N. Applications of the crown diameter–stem diameter relationship for different species of broadleaved trees. For. Ecol. Manag. 2005, 215, 285–294. [CrossRef] 16. Hyyppa, J.; Hyyppa, H.; Leckie, D.; Gougeon, F.; Yu, X.; Maltamo, M. Review of methods of small footprint airborne laser scanning for extracting data in boreal forests. Int. J. Remote Sens. 2008, 29, 1339–1366. [CrossRef] 17. Lim, K.; Hopkinson, C.; Treitz, P. Examining the effects of sampling point densities on laser canopy height and density metrics. For. Cron. 2008, 84, 876–885. [CrossRef] Remote Sens. 2016, 8, 388 15 of 16

18. Segura, M.; Kanninen, M. Allometric models for tree volume and total aboveground biomass in a tropical humid forest in Costa Rica. Biotropica 2005, 37, 2–8. [CrossRef] 19. Brandtberg, T.; Warner, T.; Landenberg, R.; McGraw, J. 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 Sens. Environ. 2003, 85, 290–303. [CrossRef] 20. Holmgren, J.; Nilsson, M.; Olsson, H. Estimation of tree height and stem volume on plots using airborne laser scanning. For. Sci. 2003, 49, 419–428. 21. McComb, J.W.; Roberts, S.D.; Evans, D.L. Influence of fusing LiDAR and multispectral imagery on remotely sensed estimates of stand density and mean tree height in a managed loblolly pine . For. Sci. 2003, 49, 457–466. 22. Holmgren, J.; Persson, A. Identifying species of individual trees using airborne laser scanner. Remote Sens. Environ. 2004, 90, 415–423. [CrossRef] 23. Popescu, S.C.; Wynne, R.H. Seeing the trees in the forest: Using LiDAR and multispectral data fusion with local filtering and variable window size for estimating tree height. Photogramm. Eng. Remote Sens. 2004, 70, 589–604. [CrossRef] 24. Yu, X.; Hyyppä, J.; Kaartinen, H.; Maltamo, M. Automatic detection of harvested trees and determination of forest growth using airborne laser scanning. Remote Sens. Environ. 2004, 90, 451–462. [CrossRef] 25. Roberts, S.D.; Dean, T.J.; Evans, D.L.; McCombs, J.W.; Harrington, R.L.; Glass, P.A. Estimating individual tree leaf area in loblolly pine plantations using LiDAR-derived measurements of height and crown dimensions. For. Ecol. Manag. 2005, 213, 54–70. [CrossRef] 26. Means, J.E.; Acker, S.A.; Fitt, B.J.; Renslow, M.; Emerson, L.; Hendrix, C.J. Predicting forest stand characteristics with airborne scanning LiDAR. Photogramm. Eng. Remote Sens. 2000, 66, 1367–1371. 27. Næsset, E.; Bjerknes, K.-O. Estimating tree heights and number of stems in young forest stands using airborne laser scanner data. Remote Sens. Environ. 2001, 78, 328–340. [CrossRef] 28. Rönnholm, P.; Hyyppä, J.; Hyyppä, H.; Haggrén, H.; Yu, X.; Kaartinen, H. Calibration of laser-derived tree height estimates by means of photogrammetric techniques. Scand. J. For. Res. 2004, 19, 524–528. [CrossRef] 29. Huang, S.; Price, D.; Titus, S.J. Development of ecoregion-based height–diameter models for white spruce in boreal forests. For. Ecol. Manag. 2000, 129, 125–141. [CrossRef] 30. Hunter, M.O.; Keller, M.; Victoria, D.; Morton, D.C. Tree height and tropical forest biomass estimation. Biogeosciences 2013, 10, 8385–8399. [CrossRef] 31. Hyyppa, J.; Kelle, O.; Lehikoinen, M.; Inkinen, M. A segmentation-based method to retrieve stem volume estimates from 3-D tree height models produced by laser scanners. IEEE Trans. Geosci. Remote Sens. 2001, 39, 969–975. [CrossRef] 32. Persson, A.; Holmgren, J.; Söderman, U. Detecting and measuring individual trees using an airborne laser scanner. Photogramm. Eng. Remote Sens. 2002, 68, 925–932. 33. Lefsky, M.A.; Cohen, W.B.; Acker, S.A.; Parker, G.G.; Spies, T.A.; Harding, D. LiDAR remote sensing of the canopy structure and biophysical properties of Douglas-fir western hemlock forests. Remote Sens. Environ. 1999, 70, 339–361. [CrossRef] 34. Dowling, R.; Accad, A. Vegetation classification of the riparian zone along the Brisbane River, Queensland, Australia using Light Detection and Ranging (LiDAR) data and forward looking digital video. Can. J. Remote Sens. 2003, 29, 556–563. [CrossRef] 35. Lovell, J.L.; Jupp, D.L.B.; Culvenor, D.S.; Coops, N.C. Using airborne and ground-based ranging LIDAR to measure canopy structure in Australian forests. Can. J. Remote Sens. 2003, 29, 607–622. [CrossRef] 36. Lucas, R.M.; Cronin, N.; Lee, A.; Moghaddam, M.; Witte, C.; Tickle, P. Empirical relationships between AIRSAR backscatter and LiDAR-derived forest biomass, Queensland, Australia. Remote Sens. Environ. 2006, 100, 407–425. [CrossRef] 37. Moffiet, T.; Mengersen, K.; Witte, C.; Kinga, R.; Denham, R. Airborne laser scanning: Exploratory data analysis indicates potential variables for classification of individual trees or forest stands according to species. ISPRS J. Photogramm. Remote Sens. 2005, 59, 289–309. 38. Witte, C.; Denham, R.; Turton, D.; Jonas, D.; Tickle, P.; Norman, P. Airborne laser scanning: A tool for monitoring and assessing the forests and woodlands of Australia. In Proceedings of the 10th Australasian Remote Sensing Conference, Adelaide, Australia, 21–25 August 2000; pp. 348–362. Remote Sens. 2016, 8, 388 16 of 16

39. Weller, D.; Denham, R.; Witte, C.; Mackie, C.; Smith, D. Assessment and monitoring of foliage projected cover and canopy height across native vegetation in Queensland, Australia, using laser profiler data. Can. J. Remote Sens. 2003, 29, 578–591. 40. Bunting, P.; Lucas, R.M. The delineation of tree crowns within CASI data of Australian mixed species woodlands. Remote Sens. Environ. 2006, 101, 230–248. 41. Held, A.; Phinn, S.; Scarth, P.; Stanford, M.; Ticehurst, C.; Hartini, S.; Lymburnber, L. Hyperspectral mapping of and mangroves. In Proceedings of the International Geosciences and Remote Sensing Symposium, Sydney, Australia, 9–13 July 2001. 42. Tickle, P.K.; Lee, A.; Lucas, R.M.; Austin, J.; Witte, C. Quantifying Australian forest floristics and structure using small footprint LiDAR and large scale aerial photography. For. Ecol. Manag. 2006, 223, 379–394. 43. Lee, A.C.; Lucas, R.M. A LiDAR-derived canopy density model for tree stem and crown mapping in Australian forests. Remote Sens. Environ. 2007, 111, 493–518. [CrossRef] 44. Ozdemir, I. Estimating stem volume by tree crown area and tree shadow area extracted from pan-sharpened Quickbird imagery in open Crimean juniper forests. Int. J Remote Sens. 2008, 29, 5643–5655. [CrossRef] 45. Greenberg, J.A.; Dobrowski, S.Z.; Ustin, S.L. Shadow allometry: Estimating tree structural parameters using hyperspatial image analysis. Remote Sens. Environ. 2005, 97, 15–25. [CrossRef] 46. Verma, N.K.; Lamb, D.W.; Reid, N.; Wilson, B. An allometric model for estimating DBH of isolated and clustered Eucalyptus trees from measurements of crown projection area. For. Manag. 2014, 326, 125–132. [CrossRef] 47. Verma, N.; Lamb, D.W.; Reid, N.; Wilson, B. A comparative study of land cover classification techniques for ‘farmscapes’ using very high-resolution remotely sensed data. Photogramm. Eng. Remote Sens. 2014, 80, 461–470. 48. Lamb, D.W.; Trotter, M.; Murphy, A.; Young, I. UNE SMART Farm: Showcasing the value of broadband connectivity in the business and lifestyle of farming. In Proceedings of the Digital Rural Futures Conference, University of New England, Australia, 26–28 June 2013. 49. Sumida, A.; Komiyama, A. Crown spread patterns for five deciduous broad-leaved woody species: Ecological significance of the retention patterns of larger branches. Ann. Biol. 1997, 80, 759–766. [CrossRef] 50. Dash, J.; Steinle, E.; Singh, R.P.; Bahr, H.P. Automatic building extraction from laser scanning data: An input tool for disaster management. Adv. Space Res. 2004, 33, 317–322. [CrossRef] 51. Ryherd, S.; Woodcock, C. Combining spectral and texture data in the segmentation of remotely sensed images. Photogramm. Eng. Remote Sens. 1996, 62, 181–194. 52. Baatz, M.; Schape, A. Multiresolution segmentation—An optimization approach for high quality multi-scale image segmentation. In Angewandte Geographische Informationsverarbeitung XII, Beitra¨ge zum AGIT-Symposium Salzburg; Wichmann: Heidelberg, Germany, 2000; pp. 12–23. 53. Benz, U.C.; Hoffmann, P.; Willhauck, G.; Lingenfelder, I.; Heynen, M. Multi-resolution object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS J. Photogramm. Remote Sens. 2004, 58, 239–258.

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).