Estimation of leaf area index in eucalypt forest using ...

Agricultural and Forest Meteorology 143 (2007) 176–188 www.elsevier.com/locate/agrformet

Estimation of leaf area index in eucalypt forest using digital photography Craig Macfarlane a,b,*, Megan Hoffman a, Derek Eamus c, Naomi Kerp d, Simon Higginson e, Ross McMurtrie f, Mark Adams a a

The Centre of Excellence in Natural Resource Management, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia b Ecosystems Research Group, School of Plant Biology (M090), Faculty of Natural and Agricultural Sciences, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia c Institute for Water and Environmental Resource Management, University of Technology, Sydney, NSW 2000, Australia d Alcoa World Alumina Australia, PO Box 172, Pinjarra, WA 6208, Australia e CSIRO Land and Water, Centre for Environment and Life Sciences, PO Box 5 Wembley, WA 6913, Australia f School of Biological, Earth & Environmental Sciences, University of New South Wales, Sydney, NSW 2000, Australia Received 29 August 2005; received in revised form 13 October 2006; accepted 22 October 2006

Abstract We tested whether leaf area index (L) in eucalypt vegetation could be accurately estimated from gap fraction measurements made using both fisheye and non-fisheye digital photography. We compared methods that measure the gap fraction at a single zenith angle (08 or 578), with fisheye photography that measures the gap fraction at multiple zenith angles. We applied these methods in an unthinned stand of the broadleaf tree species Eucalyptus marginata that had an initial L of 3. We removed one-third of the trees and reapplied the methods, and then removed another one-third of the trees and applied the methods a third time. L from the photographic methods was compared to L obtained from destructive sampling and allometry. We found that L was accurately estimated from non-fisheye images taken at the zenith, providing that the total gap fraction was divided into large, between-crown gaps and smaller, within-crown gaps, prior to using the Beer–Lambert law to estimate L. This rapid and simple method corrected for foliage clumping and provided estimates of crown porosity, crown cover, foliage cover and the foliage clumping index at the zenith, but required an assumption about the light extinction coefficient at the zenith. Fisheye photography also provided good estimates of L but only if the images were corrected for the gamma function of the digital camera, and the combined Chen–Cihlar and Lang– Xiang method of correcting for foliage clumping was used. The clumping index derived from fisheye images was insensitive to thinning but the calculated foliage projection coefficient was. Methods of obtaining and analysing gap fraction and gap size distributions from fisheye photography need further improvement to separate the effects of foliage clumping and leaf angle distribution. # 2006 Elsevier B.V. All rights reserved. Keywords: Eucalypt forest; Leaf area index; Digital photography; Gap fraction; Canopy cover; Clumping index

* Corresponding author at: Ecosystems Research Group, School of Plant Biology (M090), Faculty of Natural and Agricultural Sciences, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia. Tel.: +61 8 6488 7924; fax: +61 8 6488 7925. E-mail addresses: [email protected] (C. Macfarlane), [email protected] (M. Hoffman), [email protected] (D. Eamus), [email protected] (N. Kerp), [email protected] (S. Higginson), [email protected] (R. McMurtrie), [email protected] (M. Adams). 0168-1923/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2006.10.013

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

1. Introduction Eucalypts are increasingly important worldwide— both to industry and to local and regional carbon and water budgets (see, for example, Hubbard et al., 2004), and there is an obvious need to obtain accurate estimates of their leaf area index (L). L is an essential input into many models of eucalypt growth and water use (Linder, 1985; Comins and McMurtrie, 1993; Beadle et al., 1995; Landsberg and Hingston, 1996; Hingston et al., 1998; Snow et al., 1999; Croton and Barry, 2001) as well as being an essential component of comparative studies of many leaf-level attributes such as transpiration and water use efficiency (e.g. Hubbard et al., 2004). Estimates of L are difficult to obtain in forest owing to the labour required to destructively sample many trees, and to recognised problems with indirect methods of estimating L. This has prompted the development of visual guides for estimating L (e.g. Sampson et al., 1997; e.g. Cherry et al., 2002) and, mainly within Australia, the development and use of the ‘‘Adelaide method’’ (Andrew et al., 1979), which is based on counting the number of similar clumps of foliage in the canopy. These visual methods require a limited amount of destructive sampling for calibration (O’Grady et al., 1999) and can still be time consuming. The Licor LAI-2000 plant canopy analyser (PCA, Licor Inc., Lincoln, Nebraska) has gained wide acceptance for estimating L but the cost of this instrument can be prohibitive. Furthermore, the PCA is known to underestimate L by 10–40% (Macfarlane et al., 2000), partly because of scattering of blue light (Chen, 1996). Hemispherical or fisheye photography is a cheaper alternative to the PCA and has been applied successfully in stands of Eucalyptus globulus (Macfarlane et al., 2000). Both fisheye photography and the PCA measure the gap fraction at multiple viewing angles in order to analytically separate and quantify both foliage area and foliage angle. Technical and theoretical obstacles have until recently prevented wide spread adoption of fisheye photography. Fisheye photography required metering exposure, adjusting lens settings, film development, negative scanning, conversion of colour images to black and white, and image analysis with specialised software. The accurate reproduction of pixel brightness theoretically requires control using an optical density wedge throughout the process (Wagner, 1998), but this has rarely been undertaken. Other sources of error in fisheye photography include photographic exposure and choice of threshold. Darker images give larger L and automatic exposure metered beneath the canopy can result in

177

inaccurate estimates of L (Macfarlane et al., 2000; Zhang et al., 2005). Digital photography eliminates several steps from fisheye photography and attempts have been made to resolve other issues. Correct exposure can be determined beneath tall canopies with a handheld ‘spot’ light meter (Olsson et al., 1982; Clearwater et al., 1999; Walter et al., 2003). To some extent varying the threshold can correct for incorrect exposure, but thresholding is often subjective and can introduce errors as well as remove them. Wagner (2001) and Wagner and Hagemeier (2006) demonstrated that, when exposure is inconsistent, the two-value threshold method provides better estimates of L and of the leaf angle distribution than analysis of binary thresholded images (images converted from greyscale to black-andwhite based on a single threshold). Leblanc (2004) recently released DHP (Digital Hemispherical Photography), the first freely available software product that incorporates a two-value threshold method. Notwithstanding these improvements in image capture and analysis, the interpretation of results from fisheye photography can be problematic. Most indirect methods estimate the effective plant area index (Lt) rather than actual L as a result of the contribution of woody elements to the total plant cover, which results in overestimation of L, and clumping of foliage, which results in underestimation of L. Generally, corrections for woody area are between 5 and 30% (Gower et al., 1999) while corrections for clumping may be larger (Fassnacht et al., 1994; Kucharik et al., 1997). The clumping index, V(u), is the ratio of effective plant or leaf area index to the actual plant or leaf area index. V(u) equals 1 when foliage is randomly distributed in a canopy but is less than 1 as foliage becomes more clumped. Lt, V(u) and the foliage projection coefficient, G(u), together determine the canopy gap fraction, P(u) according to the modified Beer–Lambert law (Eq. (1) based on Nilson, 1971) where u is the zenith angle. Hence, to accurately estimate L from gap fraction distributions it is necessary to also know V(u) and the ratio of woody area to total plant area.   GðuÞVðuÞLt PðuÞ ¼ exp  cosðuÞ

(1)

Most of the clumping in broadleaf canopies is at the tree level; large, non-random gaps exist between broadleaf trees while the foliage within crowns tends to be more randomly distributed (Kucharik et al., 1997). Clumping indices can be calculated from an analysis of the gap size distribution (Chen and Cihlar, 1995) in addition to

178

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

the gap fraction distribution. Chen and Cihlar’s (1995) derivation of V(u) tended to underestimate V(u) and was later corrected by Leblanc (2002). The process is an iterative one of removal of non-random gaps until the gap size distribution resembles that of a canopy with the same gap fraction but with randomly distributed foliage. Instruments such as the TRAC (Chen and Cihlar, 1995) estimate the gap size distribution from light transmittance along a transect; Leblanc’s (2004) software DHP integrates with TRACWin (the software for the TRAC, Leblanc and Chen, 2002) to obtain the gap size distribution from fisheye photographs. Kucharik et al. (1999) used Chen and Cihlar’s (1995) uncorrected method to obtain V(u) from the crown ratio (ratio of crown depth to crown diameter), crown porosity (proportion of sky within crown envelopes) and crown cover (the proportion of ground area covered by the vertical projection of crowns, assuming the crowns to be solid, Walker and Tunstall, 1981). Crown porosity and cover have been estimated using both a custom developed digital camera (the multiband vegetation imager, Kucharik et al., 1997) and the PCA (Kucharik et al., 1999). From simulated canopy images, Kucharik et al. (1999) developed empirical equations for both the minimum clumping index at the zenith, V(0), and V(u). In this study we estimated crown cover and porosity from canopy photographs obtained with a consumer grade digital camera (Nikon Coolpix 4500). We applied the corrected Chen and Cihlar method (Leblanc, 2002) of quantifying V(0) from such photographs by manually removing large gaps between trees from the images. We compared this simple method of obtaining Lt and V(0) with Lt and V(u) obtained from fisheye photography using DHP-TRACWin. The methods were tested on a single stand of Eucalyptus marginata with L of 3 before thinning, and again after two thinnings of the stand that each reduced L by one-third and increased the degree of tree-level foliage clumping to simulate sparse, clumped stands. Further comparisons were made on the unthinned stand with the PCA (Welles and Norman, 1991), the Adelaide method (Andrew et al., 1979) and digital photographs taken at a zenith angle of 578 (Bonhomme and Chartier, 1972). We expected to find that no correction for foliage clumping would result in underestimates of L, and that V(0) estimated from crown cover and porosity using the corrected Chen and Cihlar method (Leblanc, 2002) would agree with that estimated from fisheye photography using the same method implemented in DHP-TRACWin. We also expected that correcting for foliage clumping would result in accurate estimates of the foliage projection coefficient and L from fisheye photography.

2. Materials and methods 2.1. Site description and experimental design A square 0.25 ha plot was established on near-level ground in 12 year old, 12 m tall jarrah (E. marginata) forest established on a rehabilitated minesite operated by Alcoa World Alumina Australia near Dwellingup, Western Australia (328360 4100 S, 1168010 3900 E). The plot contained 1066 trees of mainly E. marginata with a few smaller E. calophylla trees. The diameter at breast height (1.3 m) over bark of all stems was measured; initial basal area was 13.6 m2. Trees were then randomly allocated to one of three groups that were either to be unharvested, harvested in the first thinning or harvested in the second thinning. The photographic methods described below were applied prior to any harvesting, and after each of the two thinnings. The Adelaide method and the PCA were only applied to the unthinned stand, and 578 photographic images were collected on the unthinned and once-thinned stands only. No corrections were made for woody area for any of the indirect methods. Hence, all results from photographic methods and the PCA are referred to as plant area index, Lt. Results from allometry and the Adelaide method are estimates of actual L. 2.2. Estimation of L from allometry and the Adelaide method L was estimated on the thinned and unthinned stands using allometric regressions of leaf area versus diameter at breast height over bark, and the Adelaide method (Andrew et al., 1979). Twelve trees that represented the range of stem diameters at the site were selected for destructive sampling from the trees to be harvested, and felled during the 3 days of measurements. For each of the 12 trees, all live branches were removed and stratified into three or four groups based on their diameters. The total mass of branches in each group was measured to the nearest gram, and three to five sample branches were selected from each group. These branches were weighed then stripped of leaves and their total leaf area measured with a calibrated leaf area meter. The mean ratio of leaf area to fresh branch mass of the sample branches was used to calculate the leaf area of all the branches in that size class. These were summed for each sample tree to estimate its total leaf area. A logarithmic regression was developed to predict total tree leaf area from tree diameter and used to calculate the leaf area of all trees in the stand, both before thinning and after each thinning. These were

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

summed to obtain the total leaf area of the stand, which was divided by 0.25 ha to obtain L. The Adelaide method was applied as described above except that each tree was scored, prior to any harvesting, for the number of similar sized ‘branch modules’ that it contained. A regression was constructed between number of modules of the 12 sample trees and their leaf area. This regression was applied to all the 1066 trees in the plot and L calculated as above. 2.3. Estimation of plant area index from digital photography 2.3.1. Cover photography All images were collected as FINE JPEG images with maximum resolution (3,871,488 pixels total). To measure cover, the Nikon Coolpix 4500 was set to automatic exposure, Aperture-Priority mode and F2 lens. This provided a consistent zoom with angle of view of approximately 358 across the diagonal, or about 0–158 zenith angle range, which is similar to the first ring of the Licor LAI-2000 PCA, as used by Kucharik et al. (1999). The aperture was set to minimum (i.e., the maximum possible F-stop of 9.6). During late afternoon, 25 images were collected on a grid with sample locations spaced 10 m apart (Fig. 1). The lens was pointed directly upwards and the camera lens was levelled using a bubble level fixed to an aluminium plate fitted between the camera tripod mount and tripod. Cover images were analysed using Adobe PhotoShop1 7.0 as follows: on each image the large gaps

179

between tree crowns were selected using the ‘wand’ tool with the SHIFT key held down, and the total number of pixels contained in large gaps (gL) recorded from the Image Histogram; all gaps were then selected using the Select Similar menu item and the total number of pixels in gaps (gT) recorded from the Image Histogram. The fractions of foliage cover ( f f, defined as ‘‘the proportion of ground area covered by the vertical projection of foliage and branches’’ (Walker and Tunstall, 1981)) and crown cover ( f c) were calculated as: fc ¼ 1 

gL 3; 871; 488

ff ¼ 1 

gT 3; 871; 488

Crown porosity (F) is the inverse of canopy cover, defined by Walker and Tunstall (1981) as ‘‘the proportion of the ground area covered by the vertical projection of foliage and branches within the perimeter of the crowns of individual plants’’. F was calculated from Eq. (2). F¼1

ff fc

(2)

Lt was estimated using a modified version of the Beer– Lambert law as follows, where k (k = G(u)/cos(u)) is the extinction coefficient: Lt ¼  f c

lnðFÞ : k

(3)

Eq. (3) is equivalent to Eq. (5) from Leblanc (2002) except that Leblanc (2002) expressed f c on a linear basis rather than an area basis. Lt was also calculated from f f and the Beer–Lambert law with no clumping correction as: Lt ¼ 

lnð1  f f Þ k

(4)

From Eqs. (1)–(3) we calculated the clumping index at the zenith as follows: Vð0Þ ¼

Fig. 1. Indicative sampling positions for fisheye photography and the Licor LAI-2000 (crosses), cover images (circles) and 57.58 images (arrows) in the 0.25 ha plot of E. marginata.

ð1  FÞ lnð1  f f Þ : lnðFÞ f f

(5)

Eq. (5) is equivalent to the corrected Chen and Cihlar (1995) method (Eq. (11) in Leblanc, 2002) and does not require knowledge of either L or k. V(0) and V(u) were also calculated from f c, F and crown ratio after Kucharik et al. (1999), except that V(0) from our Eq. (5) was used instead of that from Kucharik et al.’s (1999) Eq. (3), because their derivation of V(0) was based on the uncorrected Chen and Cihlar

180

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

(1995) method. Crown ratio (ratio of crown depth to crown diameter) was estimated visually in the field as 1. In Eqs. (3) and (4) we assumed k = 0.5. k for forest is usually between 0.3 and 0.6 (Larcher, 1975; Linder, 1985; Waring and Running, 1998). Bre´da (2003) summarised a range of k values for broadleaved stands, which ranged from 0.42 to 0.58. An extinction coefficient of 0.5 is commonly assumed for eucalypt canopies (Linder, 1985; Beadle et al., 1995; Landsberg and Hingston, 1996; Battaglia and Sands, 1997; Hingston et al., 1998; Snow et al., 1999; Battaglia et al., 2004). A recent study in E. nitens found that k measured at the base of crowns was 0.53 and was unaffected by pruning (Pinkard et al., 1999) and Bre´da (2003) cited a value for k of 0.50 for forest of E. globulus. However, k is usually assumed or measured for the purposes of modelling light transmission at multiple zenith angles, not only at the zenith (the sun is rarely directly overhead). Mean k = 0.5 across a wide range of zenith angles does not imply a spherical leaf distribution. In fact, for a spherical leaf distribution, k = 0.5 only at the zenith. Assuming typical mean leaf angles of 60–808 (King, 1997), k at the zenith is unlikely to be larger than 0.5 for eucalypt canopies but could be smaller. Bolstad and Gower (1990) found that errors of L estimation were three times greater if k was assumed rather than measured locally. Line quantum sensors and other portable light meters have been used to measure k (e.g. Baldocchi et al., 1985; e.g. Bolstad and Gower, 1990; Vose and Swank, 1990). However, these methods require foreknowledge of L, and instruments such as the PCA and Decagon AccuPar (Hyer and Goetz, 2004) are susceptible to the effects of light scattering, hence, estimates of k from these methods may be hardware specific. Furthermore, accurate estimates of k within crowns are difficult to obtain; k is as much a function of the distribution of foliage clumps as it is the angular distribution of foliage elements (Chen and Black, 1991), hence, the best estimates of k are derived from stands with minimal clumping. 2.3.2. Fifty-seven degree photography Just prior to sunset on Days 1 and 2, 32 images were collected with a Sony DSC-717 at a zenith angle of approximately 578 at the locations and in the directions indicated by the arrows in Fig. 1. If necessary, the camera was adjusted to one side of the predetermined angles to avoid large stems. The camera was operated in Aperture-Priority mode with the lens set to maximum zoom, minimum aperture and automatic exposure. All images were collected as FINE JPEG images with maximum resolution (4,915,200 pixels total). At

maximum zoom the vertical angle of view was approximately 88, hence, the gap fraction was measured from 538 to 618. The gap fraction at u = 578 (P(57)) was measured using Adobe Photoshop 7.01 with the same method as for f f. From P(57), the effective Lt was estimated using the Beer–Lambert law (Eq. (6)). The extinction coefficient is largely independent of the foliage angle distribution at this zenith angle and was taken as k = 0.91 (Bonhomme and Chartier, 1972). Lt ¼ 

lnðPð57ÞÞ 0:91

(6)

2.3.3. Fisheye photography At sunset each day, nine fisheye images were collected within each plot as indicated in Fig. 1 with the FC-E8 lens converter on the Coolpix 4500 and the lens set to F1. The camera was pointed directly upwards and levelled. Aperture was set to minimum and, with the camera set to Aperture-Priority mode, exposure was metered in an adjacent clearing, the shutter speed noted, then the mode changed to Manual and the shutter speed lowered by two stops relative to that metered below open-sky. Exposure was checked regularly beneath the canopy using a spot light meter (Sekonic L-508, Sekonic Corporation, Tokyo, Japan) that had been cross-calibrated with the camera earlier, but light conditions changed little during the time required to collect nine images. All images were collected as FINE JPEG images with maximum resolution (3,871,488 pixels total). Foliage clumping indices, k and Lt were estimated from fisheye images using DHP and TRACWin as outlined in the manual (Leblanc, 2004). To assess the effect of the camera’s gamma function on gap fraction and LAI retrieval we analysed all images both with an early version of DHP (1.9d) which assumed a gamma of 1 for images, and DHP version 4.2.2, which allows the user to specify the gamma function. We assumed a gamma of 2.2 for images from the Nikon Coolpix 4500 (Leblanc, 2004). The gamma function of a film or digital image analysis system (e.g. camera) describes the relation between the actual light intensity during photography and the resulting density of the negative or brightness value of the pixel (Wagner, 1998). Images were divided into six rings for thresholding, and the gap fraction and foliage projection coefficient at these six angles were estimated by DHP. Input files for TRACWin (version 3.7.3) were generated from zenith angles of 10–808 with a step of 18 and markers at 458. The element clumping index at all zenith angles was estimated using the CC and CLX methods (Leblanc,

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

2004; Leblanc et al., 2005) and Lt estimated at a zenith angle (u) of 57.38, as well as at zenith angles ranging from 168 to 458, from the gap fraction and foliage clumping index at those angles. TRACWin assumes a spherical leaf distribution to calculate Lt at u = 57.38 but calculates G(u) and Lt simultaneously for other zenith angles. G(u) was obtained from DHP version 4.2.2 prior to correction for foliage clumping and from TRACWin 3.7.3 after correction for foliage clumping. TRACWin provides estimates of G(u) from zenith angles of 10–808 with a step of 18: these were averaged in such a way as to provide mean estimates of G(u) within the same six rings used by DHP. Lt and the mean leaf angle before and after each thinning were also obtained from Hemiview (Delta-T Devices, Ltd., Burwell, Cambridge, UK). 2.4. Estimation of plant area index with the Licor LAI-2000 PCA Canopy gap fraction and effective Lt readings were made with the Licor LAI-2000 plant canopy analyser (Licor Inc., Lincoln, Nebraska) at the same time and location as the fisheye images prior to any thinning. The PCA was operated in remote mode. The ‘‘above’’ sensor was located in a clearing 200 m from the sample plot, levelled and set to record every 15 s. At each sample location within the plot, three readings were made and their average recorded. The light measurements from the fifth ring were not used in the calculation of effective Lt to eliminate possible interference by surrounding vegetation and to minimise the effects of blue light scattering on the measured gap fraction distribution. Gap fractions and effective Lt were calculated by the PCA software. 2.5. Statistical analyses All statistical analyses were performed using Minitab Release 13 (# 1999 Minitab Inc. State College, Philadelphia, PA, USA), except reduced major axis regression, which was performed using PAST version 1.40 (Hammer et al., 2001).

181

3. Results There was a good correlation between tree leaf area (LA, m2) and tree diameter at breast height over bark (D, mm; ln LA = 2.29 ln D  8.92, R2 = 0.91). Initial leaf area index (L), estimated using allometry, was 3 and initial stand basal area was 13.6 m2. The Adelaide method yielded L = 3.3. Each thinning reduced basal area, L, foliage cover ( f f) and crown cover ( f c) by approximately one-third (Table 1). Crown porosity (F) was largely unaffected by thinning although the first thinning may have slightly increased crown porosity, perhaps by removing some overlapping crowns. The clumping index at the zenith calculated from cover images, V(0), decreased as the plot was thinned (Table 1). Calculating V(0) after two thinnings was problematic because 5 of the 25 cover images contained no foliage, hence, crown porosity and V(0) could not be calculated for these images. These images were omitted to calculate V(0) in Table 1. As an alternative we calculated V(0) for the whole plot from the cumulative total number of random and non-random gap pixels of all 25 images. This yielded a similar result for V(0) because a slightly larger crown porosity was also calculated (0.17). Lt from cover photographs (Eq. (3)) agreed very closely with L from allometry (Table 1). As expected, Lt from the Beer–Lambert law with no clumping correction (Eq. (4)) underestimated L by 30–50% and the error increased after each thinning (Table 1), reflecting the increased clumping of the canopy. Effective Lt from 578 images underestimated L by 17–27%; Lt was 2.30  0.11 for the unthinned stand and 1.69  0.15 for the once-thinned stand. Applying the same methods for cover images to 578 images and using Eq. (5), we obtained an estimate of V(57) of 0.93 in the unthinned stand, indicating that not all of the underestimation by 578 images resulted from foliage clumping. If the estimates from the once-thinned stand were corrected for clumping then the estimate of L from these images would have agreed well with that from allometry. Good estimates of L were obtained from fisheye photography only if the images were corrected for both

Table 1 Basal area (m2), leaf area index from allometry (L) and measures of stand architecture, including plant area index (Lt), derived from canopy cover images of the 0.25 ha plot of E. marginata, before and after thinning

Unthinned Once thinned Twice thinned

Basal area

L allometry

Lt, Eq. (3)

Lt, Eq. (4)

V(0), Eq. (5)

Foliage cover

Crown cover

Crown porosity

13.6 9.2 4.7

3.00 2.03 1.05

3.02 (0.21) 2.00 (0.19) 0.92 (0.16)

2.18 (0.20) 1.32 (0.15) 0.51 (0.10)

0.72 (0.034) 0.65 (0.026) 0.53 (0.027)

0.63 (0.042) 0.46 (0.042) 0.21 (0.037)

0.73 (0.049) 0.56 (0.005) 0.25 (0.005)

0.13 (0.009) 0.17 (0.001) 0.16 (0.002)

Lt, cover and porosity measures, and clumping indices are means (S.E.) of 25 images, except for V(0) after the second thinning (see text).

182

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

Table 2 Indirect estimates of plant area index (Lt; mean  S.E.) of the 0.25 ha plot of E. marginata derived from nine fisheye images before and after thinning Gamma function [clumping correction; zenith angle]

Unthinned Once thinned Twice thinned

2.2 [CC; 578]

2.2 [CLX; 578]

2.2 [CC; 16–458]

2.2 [CLX; 16–458]

1.0 [CC; 578]

1.0 [CLX; 578]

1.0 [CC; 16–458]

1.0 [CLX; 16–458]

2.2 [hemiview; 0–908]

1.56 (0.09) 1.27 (0.07) 0.95 (0.04)

1.96 (0.11) 1.52 (0.09) 1.12 (0.06)

1.54 (0.02) 1.06 (0.01) 0.56 (0.01)

1.74 (0.02) 1.26 (0.01) 0.74 (0.02)

1.99 (0.06) 1.64 (0.08) 0.93 (0.06)

2.58 (0.10) 2.00 (0.14) 1.12 (0.07)

2.31 (0.09) 1.69 (0.11) 0.79 (0.07)

3.09 (0.15) 2.24 (0.16) 0.98 (0.09)

1.77 (0.04) 1.16 (0.04) 0.84 (0.04)

Lt was estimated using Hemiview with no correction for foliage clumping, and using DHP-TRACWin with corrections for foliage clumping estimated from either the corrected Chen and Cihlar method (CC, Leblanc, 2002) or the combined Chen and Cihlar, and Lang and Xiang method (CLX, Leblanc et al., 2005). Images were analysed with either no gamma correction (2.2) or with their gamma corrected to one.

the camera’s gamma function and for foliage clumping using the CLX method (Table 2). Correcting for the camera’s gamma usually increased estimated L by 30– 80%, with some exceptions. The CLX method of correcting for clumping increased estimates of L by 13– 34% compared to the Chen–Cihlar method alone. L estimated in the zenith range 16–458 was generally less than that estimated at 578, but the differences were not very large except for uncorrected images (gamma = 2.2) from the twice-thinned stand. L from Hemiphot resembled that from DHP-TRACWin using uncorrected images, and was similar to that calculated by the PCA (1.64  0.11) for the unthinned stand. The gap fractions obtained from gamma-corrected, fisheye images were similar to those obtained from cover images, 578 images and the PCA, although there was some evidence that the fisheye images underestimated cover near the zenith (Fig. 2). In contrast, gap fractions obtained from uncorrected fisheye images (gamma = 2.2) were clearly larger than those from the cover and 578 images (Fig. 2). Prior to correcting for foliage clumping, G(u) was close to 0.5 at all zenith angles for both the unthinned and once-thinned stands (Fig. 3), but deviated significantly from the assumption of a spherical leaf angle distribution (G(u) = 0.5) for the twice-thinned stand. After correcting for foliage clumping using the CLX method, a more spherical leaf angle distribution was calculated for the twicethinned stand, at least for zenith angles less than 608. However, a less spherical, more planophile leaf angle distribution was calculated for the unthinned and oncethinned stands, and the discrepancy between the twicethinned and other stands remained. The mean leaf angle obtained from both Hemiview (63  48) and the PCA (61  68) for the unthinned stand were similar to that for an ideal spherical leaf distribution (608, Jarvis and Leverenz, 1983). The mean leaf angle from Hemiview increased after the first (70  38) and second (86  28) thinning, illustrating the effect of foliage clumping on

calculated G(u). G near the zenith (G(7.5)), and therefore also k near the zenith, obtained from the fisheye images ranged between 0.4 and 0.65 (Fig. 3), but was sensitive to thinning and to the method used to correct for foliage clumping. Unlike G(u), V(u) from fisheye images was insensitive to thinning (Fig. 4). In contrast, the cover images showed an obvious increase in foliage clumping as the stand was thinned (Fig. 4 and Table 1). V(u) from the CLX method was always less than that from the CC

Fig. 2. Gap fraction vs. zenith angle calculated from fisheye images (n = 9) using DHP 1.9d (gamma = 2.2) and DHP 4.2.2 (gamma = 1.0) for the unthinned stand (open circles), once thinned stand (open squares) and twice thinned stand (open triangles). Gap fractions are also shown for the Licor LAI-2000 (unthinned stand only, open diamonds, n = 9), and cover (n = 25) and 578 (n = 32) images (closed symbols).

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

183

Fig. 3. Foliage projection coefficient vs. zenith angle before and after thinning, calculated from: (a) DHP 4.2.2 (gamma = 1.0) at six zenith angle ranges (n = 9) and (b) DHP 4.2.2 and TRACWin 3.7.3 after correction for foliage clumping using the combined Chen–Cihlar and Lang–Xiang (CLX, Leblanc et al., 2005) method. The dashed line represents the spherical leaf distribution. For the TRACWin data n = 135 for zenith angles between 168 and 758, but n = 54 for the zenith range 10–158 and n = 45 for the zenith range 76–808. Each data point is the mean and its standard error.

method, but both showed little evidence of increased foliage clumping after the first or second thinning especially at zenith angles less than 458. There was some evidence that, as the stand was thinned, V(u) from the CC method was decreasing at u = 57.58, where G  0.5 (Table 3) suggesting that the CC method was more sensitive to thinning than the CLX method. V(u) from the Kucharik model agreed well with that from TRACWin only for the unthinned stand and only for the Table 3 Clumping indices from either the corrected Chen and Cihlar method (CC, Leblanc, 2002) or the combined Chen and Cihlar, and Lang and Xiang method (CLX, Leblanc et al., 2005) calculated from gap fractions measured within the zenith range 54–608

Unthinned Once thinned Twice thinned

CC

CLX

0.89 (0.006) 0.82 (0.008) 0.79 (0.008)

0.69 (0.009) 0.68 (0.008) 0.66 (0.011)

Each result is the mean and its standard error of 63 measurements (7 zenith angles from 9 gamma-corrected fisheye images).

Fig. 4. Element clumping index vs. zenith angle before and after thinning. Clumping indices were derived from cover images using Eq. (5) (black circles, n = 25) and the Kucharick model (Kucharik et al., 1999, solid line), and from DHP-TRACWin (gamma = 1.0, n = 9) using the Chen–Cihlar method (Chen and Cihlar, 1995, short-dashed line) and the combined Chen–Cihlar and Lang–Xiang (CLX) method (Leblanc et al., 2005, long-dashed line). Standard errors of the means are not presented for the sake of clarity but were typically 0.03.

CC method (Fig. 4). V(57) from the 578 images (0.93) was also more similar to V(57) from fisheye images calculated using the CC method (0.79–0.89) than the CLX method (0.66–0.69), despite the CLX method giving the best estimates of L. 4. Discussion Error in the allometric method of estimating L may be as large as 25% owing to the cumulative errors in

184

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

each step of the process, which include measurement of tree diameters, separation of foliage from branches, estimation of foliage area to branch weight ratios, modelling of tree diameter to tree leaf area relationships, and extrapolation of these relationships to the whole stand (Chen et al., 1997). Estimates of L or Lt that agreed with allometry to within 25% were obtained from three methods in this study: the Adelaide method, cover photography (Eq. (3)) and fisheye photography. The Adelaide method, as applied in this study, was essentially the same as the allometric method except that a visual estimate of ‘branch modules’ was substituted for stem diameter. It took two people more than 1 day’s work to obtain an estimate of L using the Adelaide method, ignoring the time required for destructive sampling of the twelve trees. It is thus more appropriate to stands with far fewer trees than the 4000 ha1 of the present study site, and that have lower canopies from which individual branches can be easily sampled, rather than requiring felling of whole trees, as in this study. The Adelaide method was originally intended to be rapid, simple and minimally destructive (Andrew et al., 1979) and remains a useful tool in low, sparse vegetation with pronounced clumping of foliage at the branch level. In addition to providing estimates of Lt that agreed with L from allometry, collection and analysis of digital cover images was quicker than any other method. Although more images are required, they are easier to collect than fisheye images, as they do not require careful control of photographic exposure, and are quicker and simpler to analyse. Cover photography yields additional information to Lt such as crown cover and porosity, foliage cover and V(0). Like the multiband vegetation analyser (Kucharik et al., 1997), corrections for woody area are likely to be smaller for cover images than for either hemispherical-sensor based methods or the 578 method because stems contribute little to the Lt ‘seen’ by an upward looking, narrow angle lens (Kucharik et al., 1998). Gap fractions estimated from cover images are also likely to be more accurate than those from either the PCA or fisheye photography. Initial tests showed that gap fractions computed from cover images were largely insensitive to photographic exposure (data not presented), which we attribute to the much higher resolution at which the canopy is photographed compared to fisheye photography. Although not tested, the good performance of the cover photography method also suggests that the method is probably insensitive to the gamma function of the digital camera, owing to the high resolution of the images. A small area of canopy was represented by nearly four million pixels (for the Nikon Coolpix 4500) in the

cover images, whereas the fisheye images represent the whole canopy in less than two million pixels. Blennow (1995) found that the choice of threshold was less critical to high resolution images than to low resolution images because the frequency of mixed pixels is reduced in the high resolution images. The lower resolution of fisheye photography is also likely to result in small gaps being undetected. Owing to their likely greater accuracy, measurements of gap fractions and estimates of foliage clumping obtained from cover images can provide a valuable ‘reality-check’ for other methods. The total gap fraction, or foliage cover, is an objective measure that is very reproducible between operators and could be related directly to L by empirical relationships derived from allometry and destructive sampling. Crown cover and porosity require the visual selection of large, between-crown gaps. Typically there are a few large gaps that obviously fit this category, many smaller gaps that obviously do not, and a small number of intermediate gaps that may or may not be classified as large depending on the operator. A comparison of porosity (%) values estimated by two independent operators from the images collected in this study gave very similar results: F1 = 1.04 F2  1.1, R2 = 0.88, n = 24. The slope did not differ from one (reduced major axis regression, P = 0.56) and the 95% confidence interval for the intercept included zero. Estimation of Lt is fairly robust to variations in estimation of crown cover and porosity. As more gaps are assigned to the ‘large gap basket’ both crown cover and crown porosity decrease. Simulations using Eq. (3) show that a 30% change in porosity, say 20% versus 15%, results in only a 10% change in estimated Lt. Estimation of Lt from cover is most error-prone when cover is very high and porosity is very low. While the method described here was developed to deal with sparse, clumped canopies, it could be extended to denser canopies, especially if higher resolution SLR digital cameras, with better quality optics, are used to improve classification of small gaps. The impact of the camera’s gamma function on cover photography remains to be tested. Despite potential errors that result from the choice of k value, operator variation and the contribution of branches to plant cover, the accuracy of the method described here appears comparable to that of destructive sampling for this forest type. The good agreement of Lt from the cover images and Eq. (3), with L from allometry supports the assumption of k = 0.5 near the zenith for E. marginata. This assumption was also somewhat supported by the results from fisheye photography. We have also obtained k  0.5 from fisheye photographs of forest of E. wandoo

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

and E. diversicolor (Macfarlane et al. unpublished data). However, poor sampling near the zenith by hemispherical sensors means that estimates of k near the zenith from these methods are suspect. The correction for wood from images taken at the zenith is unlikely to be large (Kucharik et al., 1998) and it is likely that the true value of k for foliage only, within crowns of E. marginata, is between 0.45 and 0.5. The assumption of k  0.5 within eucalypt crowns seems a reasonable initial assumption, but may be significantly in error for eucalypts with very vertical foliage. Such species include many important plantation and forest species such as E. globulus, E. regnans, E. nitens, E. viminalis and E. obliqua, amongst others (King, 1997). Inaccurate measurement of L by fisheye photography has been attributed to incorrect exposure or threshold, and foliage clumping. In this study we found that it was possible to obtain good estimates of L for broadleaved eucalypt forest from fisheye photography if: (1) exposure was controlled relative to above-canopy sky brightness, (2) images were corrected for the gamma function of the digital camera, (3) a regional, two-value threshold was used instead of a global, binary threshold, and (4) the gap fraction distribution was corrected for foliage clumping using the combined Chen and Cihlar (1995) and Lang and Xiang (1986) methods. The importance of exposure control is now well documented (Chen et al., 1991; Macfarlane et al., 2000; Zhang et al., 2005). Wagner (1998) illustrated the importance of correcting images for the gamma function of chemical films and this study has shown that large errors can result from failing to account for the gamma function of digital cameras. Leblanc et al. (2005) also found that L was underestimated using an earlier version of DHPTRACWin that lacked a correction for the camera’s gamma. Methods for measuring the gamma function (Wagner, 1998) can be applied to digital cameras as well as film scanners and film, but this study has shown that gamma = 2.2 for the Nikon Coolpix 4500 provides accurate gap fractions and estimates of L. Wagner (2001) found that the two-threshold method greatly reduced the effects of exposure on Lt derived from fisheye photographs. We have found that the twovalue threshold method could not correct for an incorrect gamma function. Estimates of L from DHPTRACWin using uncorrected images (gamma = 2.2) were similar to those from Hemiview, which uses a global, binary threshold, despite corrections for foliage clumping. L was underestimated by up to 50% even after corrections were applied for foliage clumping, similar to the findings of Leblanc et al. (2005) in boreal forest. Hence, we conclude that correcting images for

185

the camera’s gamma function and correcting the gap fraction distribution for foliage clumping are more important than the thresholding method used. There is considerable uncertainty as to the value of the clumping indices that should be used to correct indirect estimates of Lt because it is not possible to directly measure a clumping index; it can only be inferred from other measurements. Some early studies (e.g. Smolander and Stenberg, 1996) estimated a mean clumping index as the ratio of effective Lt from the PCA to L from destructive sampling and, as a result, probably underestimated V because of blue light scattering (Chen, 1996). It has since become apparent that V is a function of zenith angle, and that it should not be derived using Lt from the PCA. This understanding has been advanced by the development of instruments such as the TRAC and MVI, and associated analytical methods, to quantify V. We found that the fairly extreme CLX method for calculating V(u) was necessary to obtain good estimates of L, and that both the CC and CLX methods were fairly insensitive to the increased clumping that must have resulted from thinning of the stand. In contrast, estimates of V(0) from cover images were quite sensitive to thinning; this is despite the fact that both used the same model (Eq. (5) for cover images and Eq. (11) from Leblanc (2002) for fisheye photography). We are unable to explain why the clumping index methods employed in TRACWin were insensitive to thinning, but believe that estimates from the CLX method, in particular, were unreliable, based on examination of the corresponding foliage projection coefficients and comparison with estimates from cover and 578 photography. The foliage projection coefficient should be unaffected by thinning: the foliage angle distribution should not become markedly more erectophile overnight because one third of the trees in the stand have been removed. The different foliage angle distributions calculated by TRACWin (Fig. 3) are evidence that TRACWin did not retrieve an accurate G(u) for all three thinnings. Instead of calculating an unchanged G(u) but a decreasing V(u) as the stand was thinned, TRACWin calculated a changing G(u) but largely unchanged V(u). In the absence of corrections for foliage clumping, G(u) is as much a function of the distribution of foliage clumps as it is the angular distribution of foliage elements (Chen and Black, 1991). Despite the good estimates of L obtained from fisheye images using the CLX method, the estimates of V(57) were much smaller than those from 578 images or the Kucharik model. We can not explain how DHP-TRACWin derived good estimates of L using the CLX clumping indices because

186

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

TRACWin assumes G(57) = 0.5, an assumption that should be valid, and the gap fractions also appeared fairly accurate. This study does show that further work is needed to remove the influence of foliage clumping on estimates of G(u) from fisheye photography. Thinning experiments such as this study provide an excellent means of testing the sensitivity of estimated G(u) and V(u) to photographic and analytical methods. The 578 method underestimated L by 15–30%. This method has the advantage that k is reasonably certain. Underestimation of L by the 578 method in the unthinned stand did not appear to result entirely from foliage clumping, and was instead more likely the result of a combination of under-sampling and biased sampling of the canopy. Unlike hemispherical sensors, or even the cover images, the 578 telephoto images sampled a very small zenith and azimuth angle. We collected thirty-two 578 images but this may have been insufficient. The overall gap fraction is smaller at a zenith angle of 578, and more samples are needed to prevent underestimation of L at small gap fractions (Radtke and Bolstad, 2001; Coops et al., 2004). After the first thinning, the gap fraction would have increased and under-sampling may have been less important. Sampling for 578 images was also problematic because of the large number of tree stems; if aimed randomly the image frequently contained mainly stems. We moved the camera to one side to avoid large stems, which introduced some subjectivity into the method, and may have biased our sampling towards gaps. If the sampling for 578 images were entirely random then very large woody area corrections (>30%) could be expected (Kucharik et al., 1998). The large number of images required and potentially large woody area corrections are the two major obstacles to adoption of the 578 method.

images. Cover images have a well-defined footprint that should suit typical smaller, rectangular research or inventory plots better than the large circular footprint of hemispherical sensors. Other advantages of the cover method are that images can be collected under clear or cloudy sky conditions during normal working hours, provided midday sun is avoided. The apparent advantage of the fisheye methods, that the extinction coefficient is simultaneously estimated, is not realised in practice. We conclude that future investigations of fisheye methods for estimating leaf area should concentrate on improving methods to obtain accurate gap fraction distributions, and to separate the effects of foliage angle distribution from those of foliage clumping. To this end, it appears important that fisheye images be adjusted for the gamma function of the camera used. Cover images, and high resolution images at other angles, could assist these investigations by providing accurate gap fraction estimates for comparison and greater use should be made of such methods for studying canopy architecture and leaf area.

5. Conclusions

Andrew, M.H., Noble, I.R., Lange, R.T., 1979. A non-destructive method for estimating the weight of forage on shrubs. Austr. J. Range. 1, 225–231. Baldocchi, D.D., Hutchinson, B.A., Matt, D.R., McMillen, R.T., 1985. Canopy radiative transfer models for spherical and known leaf inclination angle distributions: a test in an oak-hickory forest. J. Appl. Ecol. 22, 539–555. Battaglia, M., Sands, P., 1997. Modelling site productivity of Eucalyptus globulus in response to climatic and site factors. Austr. J. Plant Physiol. 24, 831–850. Battaglia, M., Sands, P., White, D., Mummery, D., 2004. CABALA: a linked carbon, water and nitrogen model of forest growth for silvicultural decision support. For. Ecol. Manage. 193, 251–282. Beadle, C.L., Honeysett, J.L., Turnbull, C.R., White, D.A., 1995. Site limits to achieving genetic potential. In: Potts, B.M., et al. (Eds.), Eucalypt Plantations: Improving Fibre Yield and Quality. Proc. CRC for Temperate Hardwood Forestry, Hobart, pp. 325–330.

Better leaf area measures are central to a wide range of studies including hydrology, carbon and nutrient cycling, and global change. Kucharik et al. (1998) have already shown that the MVI can be used to improve estimates of Lt from other methods by correcting for foliage clumping. We have extended that approach to estimation of cover, crown porosity, leaf area index and clumping index, all from easily obtainable digital photographic images. The image resolution and image quality attainable from consumer-grade digital cameras makes this approach inexpensive, rapid and simple. Cover images are less sensitive to exposure and choice of threshold, and can detect smaller gaps than fisheye

Acknowledgements The financial support of the Australian Research Council (Discovery grant DP0344927) to the senior author is acknowledged. We wish to thank Alcoa World Alumina Australia for access to the experimental site and tree felling services, Andrew Grigg and Nicole Adams for assistance with harvesting, Don White and CSIRO Forestry and Forest Products for access to the LICor LAI-2000 PCA, and Jim Croton, Sven Wagner and Sylvain Leblanc for useful discussions and comments on earlier versions of the manuscript. References

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188 Blennow, K., 1995. Sky view factors from high-resolution scanned fish-eye lens photographic negatives. J. Atmos. Ocean Tech. 12, 1357–1362. Bolstad, P.V., Gower, S.T., 1990. Estimation of leaf area index in fourteen southern Wisconsin forest stands using a portable radiometer. Tree Physiol. 7, 115–124. Bonhomme, R., Chartier, P., 1972. The interpretation and automatic measurement of hemispherical photographs to obtain sunlit foliage area and gap frequency. Israel J. Agric. Res. 22, 53–61. Bre´da, N.J.J., 2003. Ground-based measurements of leaf area index: a review of methods, instruments and current controversies. J. Exp. Bot. 54, 2403–2417. Chen, J.M., 1996. Canopy architecture and remote sensing of the fraction of photosynthetically active radiation absorbed by boreal conifer forests. IEEE Trans. Geosci. Rem. Sens. 34, 1353–1368. Chen, J.M., Black, T.A., 1991. Measuring leaf area index of plant canopies with branch architecture. Agric. For. Meteorol. 57, 1–12. Chen, J.M., Black, T.A., Adams, R.S., 1991. Evaluation of hemispherical photography for determining plant area index and geometry of a forest stand. Agric. For. Meteorol. 56, 129–143. Chen, J.M., Cihlar, J., 1995. Quantifying the effect of canopy architecture on optical measurements of leaf area index using two gap size analysis methods. IEEE Trans. Geosci. Rem. Sens. 33, 777–787. Chen, J.M., Rich, P.M., Gower, S.T., Norman, J.M., Plummer, S., 1997. Leaf area index of boreal forests—theory, techniques and measurements. J. Geophys. Res. 102 (D24), 29429–29443. Cherry, M., Macfarlane, C., Smethurst, P., Beadle, C., 2002. Visual Guide to Leaf Area Index of Eucalypt Plantations. Cooperative Research Centre for Sustainable Production Forestry, Hobart, 24 pp. Clearwater, M.J., Nifinluri, T., van Gardingen, P.R., 1999. Forest fire smoke and a test of hemispherical photography for predicting understorey light in Bornean tropical rain forest. Agric. For. Meteorol. 97, 129–139. Comins, H.N., McMurtrie, R.E., 1993. Long-term biotic response of nutrient-limited forest ecosystems to CO2-enrichment: equilibrium behaviour of integrated plant-soil models. Ecol. Appl. 3, 666–681. Coops, N.C., Smith, M.L., Jacobsen, K.L., Martin, M., Ollinger, S., 2004. Estimation of plant and leaf area index using three techniques in a mature native eucalypt canopy. Austr. Ecol. 29, 332–341. Croton, J.T., Barry, D.A., 2001. WEC-C: a distributed, deterministic catchment model—theory, formulation and testing. Environ. Model. Soft. 16, 583–599. Fassnacht, K., Gower, S.T., Norman, J.M., McMurtrie, R.E., 1994. A comparison of optical and direct methods for estimating foliage surface area index on forests. Agric. For. Meteorol. 71, 183–207. Gower, S.T., Kucharik, C.J., Norman, J.M., 1999. Direct and indirect estimation of leaf area index, f PAR, and net primary production of terretrial ecosystems. Rem. Sens. Environ. 70, 29–51. Hammer, Ø., Harper, D.A.T., Ryan, P.D., 2001. PAST: paleontological statistics software package for education and data analysis. Palaeontol. Electron. 4, 9. Hingston, F.J., Galbraith, J.H., Dimmock, G.M., 1998. Application of the process-based model BIOMASS to Eucalyptus globulus subsp. globulus plantations on ex-farmland in south western Australia I. Water use by trees and assessing risk of losses due to drought. For. Ecol. Manage. 106, 141–156. Hubbard, R.M., Ryan, M.G., Giardina, C.P., Barnard, H., 2004. The effect of fertilization on sap flux and canopy conductance in a Eucalyptus saligna experimental forest. Glob. Change Biol. 10, 427–436.

187

Hyer, E.J., Goetz, S.J., 2004. Comparison and sensitivity analysis of instruments and radiometric methods for LAI estimation: assessments from a boreal forest site. Agric. For. Meteorol. 122, 157–174. Jarvis, P.G., Leverenz, J.W., 1983. Productivity of temperate, deciduous and evergreen forests. In: Lange, O.L., Nobel, P.S., Osmond, C.B., Ziegler, H. (Eds.), Physiological Plant Ecology. IV. Encyclopedia of Plant Physiology, vol. 12D. Springer-Verlag, New York, pp. 233–280. King, D.A., 1997. The functional significance of leaf angle in Eucalyptus. Austr. J. Bot. 45, 619–639. Kucharik, C.J., Norman, J.M., Gower, S.T., 1998. Measurements of branch area and adjusting leaf area index indirect measurements. Agric. For. Meteorol. 91, 69–88. Kucharik, C.J., Norman, J.M., Gower, S.T., 1999. Characterization of radiation regimes in nonrandom forest canopies: theory, measurements, and a simplified modeling approach. Tree Physiol. 19, 695–706. Kucharik, C.J., Norman, J.M., Murdoch, L.M., 1997. Characterizing canopy nonrandomness with a multiband vegetation imager (MVI). J. Geophys. Res. 102 (D24), 29,455–29,473. Landsberg, J.J., Hingston, F.J., 1996. Evaluating a simple radiation/ dry matter conversion model using data from Eucalyptus globulus in Western Australia. Tree Physiol. 16, 801–808. Lang, A.R.G., Xiang, Y., 1986. Estimation of leaf area index from transmission of direct sunlight in discontinuous canopies. Agric. For. Meteorol. 35, 229–243. Larcher, W., 1975. Physiological Plant Ecology. Springer-Verlag, Berlin. Leblanc, S.G., 2002. Correction to the plant canopy gap-size analysis theory used by the tracing radiation and architecture of canopies instrument. Appl. Opt. 41, 7667–7670. Leblanc, S.G., 2004. Digital Hemispherical Photography Manual, Draft Version 1. 0 (17 September 2004) Canada Centre for Remote Sensing, Natural Resources Canada, Ottawa. Leblanc, S.G., Chen, J.M., 2002. TRAC Manual, version 2.1. Canada Centre for Remote Sensing, Natural Resources Canada, Ottawa. Leblanc, S.G., Chen, J.M., Fernandes, R., Deering, D.W., Conley, A., 2005. Methodology comparison for canopy structure parameters extraction from digital hemispherical photography in boreal forests. Agric. For. Meteorol. 129, 187–207. Linder, S., 1985. Potential and actual production in Australian forest stands. In: Landsberg, J.J., Parsons, W. (Eds.), Research for Forest Management. CSIRO, Melbourne, pp. 11–35. Macfarlane, C., Coote, M., White, D.A., Adams, M.A., 2000. Photographic exposure affects indirect estimation of leaf area in plantations of Eucalyptus globulus Labill. Agric. For. Meteorol. 100, 155–168. Nilson, T., 1971. A theoretical analysis of the frequency of gaps in plant stands. Agric. For. Meteorol. 8, 25–38. O’Grady, A.P., Eamus, D., Hutley, L.B., 1999. Transpiration increases during the dry season: patterns of tree water use in eucalypt openforests of northern Australia. Tree Physiol. 19, 591–597. Olsson, L., Carlsson, K., Grip, H., Perttu, K., 1982. Evaluation of forest-canopy photographs with diode-array scanner OSIRIS. Can. J. For. Res. 12, 822–828. Pinkard, E.A., Battaglia, M., Beadle, C.L., Sands, P.J., 1999. Modeling the effect of physiological responses to green pruning on net biomass production of Eucalyptus nitens. Tree Physiol. 19, 1–12. Radtke, P.J., Bolstad, P.V., 2001. Laser point-quadrat sampling for estimating foliage-height profiles in broad-leaved forests. Can. J. For. Res. 31, 410–418.

188

C. Macfarlane et al. / Agricultural and Forest Meteorology 143 (2007) 176–188

Sampson, D.A., Allen, H.L., Lunk, E.M. and Blevins, D.P., 1997. The Ocular LAI Comparator for Loblolly Pine Version 4.1. Smolander, H., Stenberg, P., 1996. Response of LAI-2000 estimates to changes in plant surface area index in a Scots pine stand. Tree Physiol. 16, 345–349. Snow, V.O., Bond, W.J., Myers, B.J., Theiveyanathan, S., Smith, C.J., Benyon, R.G., 1999. Modelling the water balance of effluent irrigated trees. Agric. Water Manage. 39, 47–67. Vose, J.M., Swank, W.T., 1990. Assessing seasonal leaf dynamics and vertical leaf area distribution in eastern white pine (Pinus strobus L.) with a portable light meter. Tree Physiol. 7, 125–134. Wagner, S., 1998. Calibration of grey values of hemispherical photographs for image analysis. Agric. For. Meteorol. 90, 103–117. Wagner, S., 2001. Relative radiance measurements and zenith angle dependent segmentation in hemispherical photography. Agric. For. Meteorol. 107, 103–115.

Wagner, S., Hagemeier, M., 2006. Method of segmentation affects leaf inclination angle estimation in hemispherical photography. Agric. For. Meteorol. 139, 12–24. Walker, J., Tunstall, B.R., 1981. Field Estimation of Foliage Cover in Australian Woody Vegetation. Technical Memorandum 81/19. CSIRO, Canberra, 18 pp. Walter, J.-M.N., Fournier, R.A., Soudani, K., Meyer, E., 2003. Integrating clumping effects in forest canopy structure: an assessment through hemispherical photographs. Can. J. Rem. Sens. 29, 388–410. Waring, R.H., Running, S.W., 1998. Forest Ecosystems. Analysis at Multiple Scales. Academic Press, California, 370 pp. Welles, J.M., Norman, J.M., 1991. Instrument for indirect measurement of canopy architecture. Agron. J. 83, 818–825. Zhang, Y., Chen, J.M., Miller, J.R., 2005. Determining digital hemispherical photograph exposure for leaf area index estimation. Agric. For. Meteorol. 133, 166–181.