(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies areencouraged to visit:

http://www.elsevier.com/copyright

Author's personal copy

Agricultural and Forest Meteorology 168 (2013) 130– 139

Contents lists available at SciVerse ScienceDirect

Agricultural and Forest Meteorology

jou rn al h om epa ge: www.elsev ier .com/ locate /agr formet

Estimation of canopy properties in deciduous forests with digital hemisphericaland cover photography

Francesco Chianuccia,b,∗, Andrea Cutinia

a Research Centre for Silviculture, Agriculture Research Council, Viale S. Margherita 80, 52100 Arezzo, Italyb Department for Innovation in Biological, Agro-Food, and Forestry Systems, University of Tuscia, Via San Camillo de Lellis, 01100 Viterbo, Italy

a r t i c l e i n f o

Article history:Received 5 March 2012Received in revised form 30 August 2012Accepted 2 September 2012

Keywords:Leaf area indexCover photographyDigital hemispherical photographyLight extinction coefficientClumping

a b s t r a c t

Rapid, reliable and meaningful estimates of forest canopy are essential to the characterization of forestecosystems. In this paper the accuracy of digital hemispherical (DHP) and cover (DCP) photography for theestimation of canopy properties in deciduous forests was evaluated. Leaf area index (LAI) estimated fromboth these photographic methods and from light transmittance data derived from DHP were comparedwith direct measurements obtained by litter traps (LAILT) and an AccuPAR ceptometer. Also, comparisonwith different gap fraction methods used to calculate LAI in DHP and LAI-2000 PCA were performed.

We applied these methods in four forest stands of Quercus cerris, two stands of Castanea sativa andfour stands of Fagus sylvatica, the most common deciduous species in Italy, where LAILT ranged from 3.9to 7.3. Both photographic methods provided good indirect estimates of LAILT. The DCP method providedestimates of crown porosity, crown cover, foliage cover and the clumping index at the zenith, but requiredassumptions about the light extinction coefficient at the zenith (k), to accurately estimate LAI. Coverphotography provided good indirect estimates of LAI assuming a spherical leaf angle distribution, eventhough k appeared to decrease as LAI increased, thus affecting the accuracy of LAI estimates in DCP. Incontrast, the accuracy of LAI estimates in DHP appeared insensitive to LAILT values, but the method wassensitive to photographic exposure and more time-consuming than DCP.

The studied stands were characterized by higher within-crown clumping than between-crowns clump-ing; only the segmented analysis of gap fraction for each ring of the fisheye images was found to providereliable and useful clumping index in DHP. The 1-azimuth segment method employed in PCA poorlydetected clumping in dense canopies.

The correlation between transmittance estimates by DHP with values measured at noon with theAccuPAR ceptometer was linear and significant, although the variability observed in reference measuressuggested that results obtained with the ceptometer should be treated with caution.

We conclude both photographic methods are suitable for dense deciduous forests. Cover photographyholds great promise as a means to quickly obtain inexpensive estimates of LAI over large areas. However,in situations where no direct reference measurements of k are available, we recommend using both DHPand DCP, in order to cross-calibrate the two methods; DCP could then be used for more routinely indirectmeasurement and monitoring of LAI.

© 2012 Elsevier B.V. All rights reserved.

Abbreviations: DCP, digital cover photography; DHP, digital hemispherical pho-tography; gL, large gaps; gT, total gaps; fc , crown cover; ff , foliage cover; ϕ, crownporosity; GLA, Gap Light Analyzer 2.0 software; k, light extinction coefficient at thezenith; LAI, indirect leaf area index; LAILT, leaf area index measured by litter traps;LAI C, LAI with correction for clumping; LAI NC, LAI without correction for clump-ing; SLA, specific leaf area; ˝0, zenithal clumping index calculated from DCP; ˝APP,Apparent clumping index; ˝LX, Clumping index calculated from DHP.

∗ Corresponding authors at: Research Centre for Silviculture, Agriculture ResearchCouncil, Viale S. Margherita 80, 52100 Arezzo, Italy. Tel.: +39 0575 353021;fax: +39 0575 353490.

E-mail address: Francesco.chianucci@entecra.it (F. Chianucci).

1. Introduction

A key aspect of characterizing forest stands is obtaining accu-rate and meaningful measures of the forest canopy (Chen et al.,1997). Crowns and canopy are more sensitive and react morepromptly than other stand structural components to abiotic aswell as biotic disturbances, and thus show effective potential foruse in monitoring the status of forest ecosystems and establish-ing long-term research programs (Cutini, 2003; Macfarlane, 2011).Measurements of leaf area index (LAI), defined as half the totalgreen leaf area per unit ground surface area (Chen and Black, 1992),and the forest light environment are some of the primary vari-ables used to characterize forest canopies. These variables affect

0168-1923/$ – see front matter © 2012 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.agrformet.2012.09.002

Author's personal copy

F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139 131

both ecological and processed-based canopy photosynthesis mod-els (Amiro et al., 2000; Chen et al., 1997). Despite the well-knownimportance of these variables, it is very difficult to gain accurateand reliable estimates of them.

Over the last two decades, much attention has been given toindirect measurements of canopy properties in forested ecosys-tems using ground-based instruments (Bréda, 2003; Cutini et al.,1998; Dufrêne and Bréda, 1995; Welles, 1990). This approach isdesirable mainly because harvesting of trees for direct measure-ment is labor intensive, time-consuming, destructive, and practicalonly in small areas (Bréda, 2003; Macfarlane et al., 2007c). Oneless destructive method using allometric relationships, namelythe allometric method, has been applied frequently in forestedecosystems. The method enables estimation of LAI by measurementof more readily acquirable variables such as diameter at breastheight or total height. However, this method is destructive, tediousand time-consuming, and not immune to error because it is site-and species-dependent (Bréda, 2003). Remotely sensed vegetationindices have novel potential, but still need cross-calibration usingground-based estimates (Bréda, 2003).

Indirect methods infer LAI and other canopy properties, suchas the forest light environment, using measurements of radiationtransmission through the canopy and using radiative transfer the-ory (Ross, 1981). Of these techniques, use of a Licor LAI-2000 PlantCanopy Analyzer (PCA; Licor Inc., Lincoln, NE, USA) has gainedwide acceptance for estimating LAI. Also, the AccuPAR ceptometer(Decagon Devices, Pullman, WA, USA) is a widely used instrumentfor measuring incident photosynthetically active radiation (PAR)and transmittance in forested stands (Bréda, 2003). Despite spe-cific constraints related these methods (for an overview, see Bréda,2003; Jonckheere et al., 2004), high cost of the instruments limitstheir broad scale application (Macfarlane et al., 2007c).

Digital photography is a less expensive and readily availablealternative. Since the 1970s, photography has been used to quan-titatively describe the architecture and radiative regime of forestcanopies. However, because it involves many time-consumingsteps, it has been progressively abandoned (Bréda, 2003). Recentadvances in digital photography have led to a resurgence of inter-est in photography for indirect quantification of canopy properties(i.e., Chen et al., 1991; Leblanc et al., 2005; Macfarlane et al., 2007c;Van Gardingen et al., 1999). Digital cameras greatly simplifiedimage capture and processing when compared to film cameras(Macfarlane, 2011). Also, numerous commercial software packagesand freeware programs for canopy analysis have been developedrecently (Frazer et al., 1999; Jarcuska, 2008; Jonckheere et al., 2005).

Digital hemispherical photography (DHP), also known as fish-eye photography, is the most widely used of several photographictechniques for describing forest canopies. The method providesestimates of canopy properties by measuring the gap fraction atmultiple zenith angles. It is inexpensive because it uses a standardcamera with a fish-eye lens. However, the sensitivity of outputsto exposure and image processing is a major drawback (Cescatti,2007; Jonckheere et al., 2004; Zhang et al., 2005). Macfarlaneet al. (2007c) recently proposed digital cover photography (DCP), amethod based on a single zenithal gap fraction measurement. Thetechnique was successfully applied in Eucalyptus forest stands inAustralia (Macfarlane et al., 2007c), in Populus euphratica standsin China (Zhongguo et al., 2009) and in an open savanna ecosys-tem in California, USA (Ryu et al., 2010b). Also, Boer et al. (2008)and Fuentes et al. (2008) used this technique to calibrate remotelysensed vegetation indices. The method provides indirect estimatesof LAI and other canopy properties, and is rapid and insensitive toexposure, but requires an assumed zenithal light extinction coef-ficient (k). Previously, LAI estimated from DCP has been testeddirectly only in forest stands with low LAI (<3.0 m2 m−2) values,so tests in denser forest stands with higher LAI values are needed.

Our goal was to evaluate the accuracy of digital photogra-phy for estimating canopy properties in dense deciduous forests,using both older (DHP) and recently developed (DCP) techniques.Forested stands of the most common deciduous species in Italy,which differed in age, stand structure and silvicultural treatment,were analyzed. We first compared LAI estimated from digital pho-tography with LAI estimated from litterfall, a direct approach thatwas considered the most precise method for estimating LAI indeciduous stands (Chen et al., 1997; Dufrêne and Bréda, 1995). Wealso compared light estimates from DHP and canopy transmittancevalues measured with an AccuPAR ceptometer; the latter method isconsidered the direct measurement point of reference for transmit-tance. Other indirect methods commonly used to estimate LAI, suchas LAI-2000 PCA and the canopy transmittance method (Cutini,1996), were also tested, to compare the performance of digital pho-tography with other indirect methods. We evaluated different gapfraction methods applicable in DHP and LAI-2000 PCA, in order toassess strength and weakness of the different approaches.

2. Materials and methods

2.1. Study sites

The study involved ten 0.5–1 ha permanent research plots indeciduous stands in three different forested areas of central Italy,which were established in previous studies (Cutini et al., 1998).All ten permanent plots were represented by coppices in conver-sion into high forest either by thinning or by natural evolution.The selected stands showed differences in species composition,age (29–66 years) and basal area (14.7–45.3 m2 ha−1), a resultof the different silvicultural treatments applied. Specifically, tenpermanent plots, four of Turkey oak (Quercus cerris L.), two ofchestnut (Castanea sativa Mill.), and four of beech (Fagus sylvaticaL.), were sampled in Massa Marittima (43◦08′N, 10◦54′E), AbbadiaSan Salvatore (42◦51′N, 11◦40′E), and Catenaia (43◦59′N, 11◦55′E),respectively. Species composition in each stand was pure. Table 1lists the main stand characteristics of each plot.

2.2. Direct measurements of LAI and of canopy transmittance

Reference LAI values were directly estimated using litter traps(LAILT). Cutini et al. (1998) described the procedure and accuracyof the method. In each plot, 9–15 litter traps were set out on a grid,at 1 m above ground level, with traps spaced 7–20 m apart, basedon the stand homogeneity and structure. Litterfall was collectedevery 2 weeks in fall and winter, with the last collection timedsoon after last leaf fall. Litter was sorted into its main components,and then dried for at least 24 h to a constant weight in a forced-airstove at 85 ± 2 ◦C (Cutini et al., 1998). This method provides leafdry mass per unit of ground area (g m−2). Specific leaf area (SLA;cm2 g−1) was estimated from a sub-sample of about 200 leaves foreach stand with samples collected at every litterfall. The area ofunwrinkled and undamaged leaves was measured with the Li-Cor3000 area meter (Licor Inc., Lincoln, NE, USA), and the dry weightwas measured. The resulting SLA was corrected using a shrinkagecoefficient (Vansereven, 1969), which was estimated from a sub-sample of green leaves collected in the vicinity of the research plots.Finally, the total dry mass of leaves collected was converted intoLAILT by multiplication of the dry weight by the corrected SLA.

Canopy transmitted light in the PAR waveband was measuredduring July 2011 using AccuPAR ceptometers (Decagon Devices,Pullman, WA, USA) at midday on sunny days. Incident radiationwas measured in open areas, which were located in the vicinityof the experimental plots. Below-canopy radiation was measuredon a grid of 9–15 sample points, which were located within 2 m of

Author's personal copy

132 F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139

Table 1Main stand characteristics of the studied stands.

Species Elevation (m.a.s.l.) Silvicultural treatment Age (years) Stems (n ha−1) Basal area (m2 ha−1)

Q. cerris 566 Natural evolution 51 4509 30.2568 1 medium thinning 51 3181 33.2589 1 light thinning 51 4198 31.1594 2 medium-heavy thinnings 54 402 21.2

F. sylvatica 1080 Seed cutting 66 108 14.71100 2 medium-heavy thinnings 66 419 34.61050 Natural evolution 66 3324 45.31020 3 light-medium thinnings 66 701 33.5

C. sativa 870 2 medium thinnings 29 580 28.4850 3 medium thinnings 29 447 23.9

the litter traps (Cutini, 1996). For each sample point, four measure-ments at cardinal directions were recorded, averaged and stored inthe instruments, for a total of 36–60 readings per plot. Transmit-tance was calculated as the percent fraction of below-canopy lightdivided by the incident radiation.

2.3. Estimation of canopy properties from digital photography

All images were collected as “fine” quality at maximum resolu-tion (2272 pixels × 1704 pixels) and as JPEG images using a NikonCoolpix 4500, one of the most widely used cameras for estimat-ing forest canopy properties. Photographs were taken at a heightof 1.5 m, on a grid of sample points, which were located nearthe litter traps. Images were collected monthly between Apriland September 2011, to evaluate leaf expansion from emergenceto leaf-off; accordingly, the maximum leaf area expansion wasachieved in July and the images collected in July were used to calcu-late LAI and for comparison with LAI from litter traps. In each plot,15–25 cover images (DCP) and 9–15 fisheye images (DHP) were col-lected. Acquisition parameters for both methods were set followingMacfarlane et al. (2007c) as summarized in Table 2 and describedbelow.

Fisheye images were collected close to sunrise (or sunset) underuniform sky conditions. The camera was equipped with a FC-E8fisheye lens converter and was set to F1; the lens was aligned tomagnetic north and pointed upward using a self-leveling tripod.The aperture was set to the minimum (F 5.3) and with the cam-era in aperture-priority (A) mode; the exposure was metered in anadjacent clearing. Subsequently, the mode was changed to manual(M) and the shutter speed was lowered by two stops in comparisonto the exposure metered in the clearing.

The gamma function of all fisheye images was corrected to 1.0using Irfanview 3.95 before conducting hemispherical softwareimage analysis. The blue channel of the images was sharp-ened (medium) and then analyzed in Winscanopy 2004a (RegentInstruments, Ste-Foy, Quebec, Canada) for estimation of LAI and

Table 2Camera setup for fisheye (DHP) and cover (DCP) photography.

Parameters DHP DCP

Camera Nikon Coolpix 4500Orientation NorthLens FC-E8 fisheye FixedLens set F1 F2Field of view ∼180◦ ∼30◦

Mode Manual Aperture-priorityF-number 5.3 9.6Exposure Manual (see text) AutomaticFormat JpegResolution 2272 × 1704Quality FINETime of day Close to sunrise (or sunset) During day

openness, as described by Macfarlane et al. (2007b). The hemi-sphere of each image was divided into 7 zenith angle rings and 8azimuth segments. The zenithal angle range used for the analyseswas 0–70◦.

Among the outputs, we obtained LAI without correction forclumping (LAI NC) by averaging the generalized LAI-2000 method(Welles and Norman, 1991) and the ellipsoidal LAI method(Campbell, 1986), for unmodified gap fraction data (linear method).We obtained LAI corrected for clumping (LAI C) by averaging thegeneralized LAI-2000 method and the ellipsoidal LAI method, fortransformed gap fraction data (log method; Lang and Xiang, 1986).Also, we estimated the percent fraction of direct, relative and totaltransmittance (respectively the direct, indirect, and total site fac-tor outputs, multiplied by 100) under the assumption of a standardovercast sky model (Anderson, 1966). A mean clumping index(˝LX) was calculated as the ratio of LAI NC to LAI C.

For comparison with the LAI-2000 PCA, fisheye images were fur-ther analyzed in Winscanopy, by setting 5 zenith angle rings, similarto those used in LAI-2000 PCA, and either 1 azimuth segment or 8azimuth segments. One azimuth segment replicates the LAI-2000PCA and was used to directly compare the two devices. For eachimage, gap fraction was computed as the average for each ring.For multiple images (namely the images across the stand), LAI wascalculated by applying two gap fraction methods:

LAI = 2

∫ �/2

0

−[ln P0(�)] cos � sin �d� (1)

LAI = 2

∫ �/2

0

−[ln P0(�)] cos � sin �d� (2)

An apparent clumping index (˝APP) was calculated as the ratioof Eq. (1) to Eq. (2).

This approach (henceforth defined as ‘1-azimuth segmentmethod’) assumes foliage elements were randomly distributed inazimuth (within each ring), and was originally proposed for LAI-2000 PCA by Ryu et al. (2010a). In fact, design limitation in PCA doesnot allow dividing each ring of each individual measurement (orfisheye image) into azimuth segments. As a consequence, the ˝APPis unable to consider clumping effect at the shoot scale (within-crown), but potentially accounts for clumping at the stand level(between-crowns) (Ryu et al., 2010a).

By contrast, DHP enables the division of each ring of each imageinto azimuth segments (log method; Lang and Xiang, 1986), thusconsidering clumping effect at the shoot scale. The 8-azimuth seg-ments method was compared with the 1-azimuth segment method,in order to investigate the difference between the two approaches,and to test the degree of clumping at different scales from within-crown level (˝LX) to between-crowns level (˝APP).

The blue channel of the images was also analyzed with the free-ware Gap Light Analyzer 2.0 (GLA; Frazer et al., 1999). Since pixelclassification between sky and canopy (thresholding) in GLA was

Author's personal copy

F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139 133

performed manually, while it is automatically determined in Win-scanopy, we used the latter automatic threshold for both softwareanalyses, to avoid subjectivity (Jarcuska et al., 2010). We consid-ered LAI NC to be the LAI 5 ring output result. Also, we estimatedthe percent transmittance (% transmittance direct, diffuse and totaloutputs) under the assumption of a standard overcast sky model.

Unlike fisheye imagery, DCP uses a narrow field of view (about30◦); the method does not require a fisheye converter and expo-sure is automatically set. Cover images were collected during themorning under uniform sky conditions. The camera was set to F2,aperture-priority mode (A) and minimum aperture (F 9.6). The fixedlens was pointed upward using a self-leveling tripod.

Images were analyzed in color using the freeware GIMP 2.6(GNU image manipulation program; www.gimp.org). Large gaps(gL) between tree crowns were selected using the ‘fuzzy’ tool, andthe total number of relative pixels was recorded from the his-togram. All gaps were then selected using the ‘select by color’ tool,and the relative number of pixels within each gap (gT) was recordedfrom the histogram. The number of pixels contained in large gapsas well as in total gaps was used to estimate crown cover (fc, Eq.(3)), foliage cover (ff, Eq. (4)), porosity (ϕ, Eq. (5)) as described byMacfarlane et al. (2007c):

fc = 1 − gL

2272 × 1704(3)

ff = 1 − gT

2272 × 1704(4)

� = 1 − fffc

(5)

Among the outputs, the method provided LAI, with either cor-rection for clumping (LAI C, Eq. (6)) or without correction (LAI NC,Eq. (7)) and making use of a modified Beer–Lambert law:

LAI · C = −fcln (�)

k(6)

LAI · NC = − ln (1 − ff )k

(7)

We also calculated the clumping index at the zenith (Eq. (8)):

˝0 = (1 − �) ln(1 − ff )ln (�)ff

(8)

However, DCP needs a zenithal light extinction coefficient (k)to calculate LAI. Measures of zenithal extinction coefficients arechallenging to obtain (Ryu et al., 2010b); also, k for DCP has beenaccurately evaluated only for Eucalyptus spp. (Macfarlane et al.,2007a, 2007b). Because measuring k is difficult, we first assumed aspherical leaf angle distribution (zenithal extinction coefficient of0.50). However, we also calculated k by combining direct measure-ments of LAI by the litterfall method with measurements of crowncover and porosity from DCP (Eq. (9)), to verify our assumption:

k = − fc ln (�)LAILT

(9)

2.4. Comparison of LAI-2000 PCA and AccuPAR ceptometer

The LAI-2000 PCA measurements were performed in each plot inJuly 2011, just after dawn or close to sunset, under uniform sky con-ditions. One above-canopy reference measurement for each plotwas recorded in clearings near each study area. The fisheye lens ofthe instrument was covered by a 90◦ view cap to avoid the influenceof surrounding trees on the reference measurements (Cutini et al.,1998). Nine to fifteen below-canopy measurements were recordedwithin each plot at the same grid points used for photography.Gap fraction was calculated for each ring from the raw data col-lected by the instrument. LAI was then estimated for each plot by

applying two different gap fraction methods (Eqs. (1) and (2); Ryuet al., 2010a). We assumed LAI NC and LAI C to be LAI calculated byEqs. (1) and (2), respectively, even though clumping effects are notconsidered at the shoot scale (within-crown) herein. An Apparentclumping index (˝APP) was calculated as the ratio of Eq. (1) to Eq.(2). The output results were compared with those calculated fromDHP using 1-azimuth segment method (see Section 2.3).

LAI was also estimated from transmitted light values measuredwith AccuPAR ceptometers, making use of the Beer–Lambert law(Bolstad and Gower, 1990; Cutini, 1996; Pierce and Running, 1988).This procedure needs a light extinction coefficient (k), which in thisstudy was assumed to be k = 0.50, because of the difficulty of directlymeasuring k, as previously outlined (see Section 2.3).

2.5. Statistical analyses

The digital photography output results were statistically ana-lyzed by comparing methods (DHP vs DCP) and software packages(Winscanopy vs GLA in DHP). The output results were first com-pared with the reference methods, to evaluate the accuracy of thedigital photography methods. We first compared LAILT with digitalphotography output results by analyzing the root mean-squarederror (RMSE). Methods with a small RMSE were then comparedusing reduced major axis (RMA) regression because we judged themeasurement errors were similar in both the direct and indirectmethods and we were also interested in testing to see if the trueslope of the relationship was equal to one (Warton et al., 2006). Wealso analyzed the RMSE of the other indirect methods of estimatingLAI (LAI-2000 PCA and canopy transmittance methods), to evalu-ate the performance of the different methods in relation to digitalphotography. Also, we compared the transmittance estimated withthe AccuPAR ceptometer with DHP output results by fitting RMAregression lines.

All statistical analyses were performed using R version 2.13.1(R Development Core Team, 2011), with the lmodel2 packageuploaded.

3. Results

The studied stands were characterized by dense canopy coverwith different attributes, as a consequence of the different sil-vicultural treatments applied. Leaf litter ranged from 3.312 to4.030 Mg ha−1 in Turkey oak, from 2.910 to 3.941 Mg ha−1 in beech,and from 2.856 to 3.236 Mg ha−1 in chestnut. LAILT ranged from3.9 to 7.3 m2 m−2 with a coefficient of variation of 23.5%; morespecifically, LAILT averaged 4.4 (±0.2) m2 m−2 in Turkey oak, aver-aged 6.6 (±0.5) m2 m−2 in beech, and averaged 4.6 (±0.2) m2 m−2

in chestnut. Transmittance values ranged from 1.6% to 11.6% witha coefficient of variation of 44.4% (Table 3).

The gap fraction obtained from fisheye images, cover images andLAI-2000 PCA were similar, although DCP provided slightly largergap faction at the zenith (Fig. 1). In contrast, gap fraction at thezenith obtained from AccuPAR was noticeably lower than thoseobtained from the other methods (Fig. 1).

From DCP, crown cover, foliage cover, and crown porosityranged from 0.86 to 0.99 (average 0.92 ± 0.01), 0.82 to 0.95 (aver-age 0.88 ± 0.02), and 0.03 to 0.10 (average 0.05 ± 0.01), respectively.Cover increased as LAILT increased, while crown porosity decreased(Fig. 2).

The zenithal clumping index from DCP (˝0) averaged0.81 ± 0.03, while the clumping index calculated from DHP(˝LX) averaged 0.87 ± 0.01. Both clumping indices increased withincreasing LAILT (Figs. 2 and 3), but they were not correlated. Theapparent clumping indices calculated from both DHP and LAI-2000PCA averaged 0.97 ± 0.01 and 0.94 ± 0.01, respectively; both the

Author's personal copy

134 F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139

Table 3Canopy properties of the studied stands.

Species Leaf litter (Mg ha−1) SLA (cm2 g−1) LAILT (m2 m−2) Transmittance (%)

Q. cerris 3.604 116.05 (6.40) 4.44 6.20 (1.52)Q. cerris 3.312 115.73 (3.21) 3.91 8.21 (2.02)Q. cerris 3.968 115.54 (13.61) 4.78 8.05 (3.69)Q. cerris 4.030 103.80 (9.13) 4.44 7.34 (2.20)

F. sylvatica 2.910 160.73 (17.75) 5.16 11.56 (4.88)F. sylvatica 3.941 167.32 (11.67) 7.34 1.63 (0.29)F. sylvatica 3.854 169.81 (19.78) 7.14 3.14 (2.01)F. sylvatica 3.259 190.76 (18.85) 6.66 4.26 (1.36)

C. sativa 3.236 140.40 (22.15) 4.83 6.72 (3.09)C. sativa 2.856 146.81 (19.62) 4.46 8.92 (3.29)

apparent clumping indices significantly differed from the otherclumping indices calculated from digital photography (p > 0.01,paired t-test).

The zenithal extinction coefficient calculated from DCP andLAILT (Eq. (9)) averaged 0.51 ± 0.02, and agreed closely to the spher-ical leaf angle distribution assumed (k = 0.50).

However, the k values appeared sensitive to canopy densityand decreased as LAILT increased (Fig. 2). Moreover, there was sig-nificant effect of species on calculated k (ANCOVA, p > 0.01); the

Fig. 1. Gap fraction vs zenith angle calculated from DHP (circles) and PCA (dia-monds). Gap fractions near the zenith are also presented for DCP (square) andAccuPAR (triangle). Standard errors are reported.

Fig. 2. Foliage cover (filled circles), crown cover (empty circles), crown porosity(squares), zenithal extinction coefficient (triangles), zenithal clumping index (dia-monds) from DCP vs LAILT from the litterfall method.

zenithal extinction coefficient for Turkey oak, beech and chestnutaveraged 0.57 ± 0.03, 0.46 ± 0.02 and 0.47 ± 0.06, respectively.

The foliage projection coefficient (G(�)) calculated from DHPeven showed an approximately spherical leaf angle distribution(∼0.50), at least for zenith angle up to 30◦, in particular when cor-rection for clumping was applied (Fig. 4). Both G(�) (averaged for allzenith angles) and mean leaf angle decreased as LAILT increased, butk near the zenith increased (Fig. 3). There was no effect of specieson both G(�) and k (ANCOVA).

With regard to LAI, similar RMSE values were obtained from DHPusing Winscanopy software and DCP (Table 4). The smallest RMSE

Fig. 3. Foliage projection coefficient, averaged for all zenith angles (filled circles),zenithal extinction coefficient (empty circles), clumping index (filled triangles),mean leaf angle (open triangles) from DHP vs LAILT from the litterfall method. Toimprove readability, only the outputs calculated after correction for clumping weredrawn; also, the dashed lines indicated the mean leaf angle trend.

Fig. 4. Foliage projection coefficient vs zenith angle before (circles) and after(squares) correction for clumping from DHP. The dashed line represents the spher-ical leaf angle distribution. Foliage projection coefficients near the zenith are alsopresented from DCP (diamond). Standard errors are reported.

Author's personal copy

F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139 135

Table 4Root mean-squared error of leaf area index (LAI) values from indirect methods andfrom littertraps.

Method NC C

Digital cover photography 0.41 0.48Digital hemispherical

photography – GLA0.72 –

Digital hemisphericalphotography – Winscanopy

0.48 0.48

PCA-LAI 2000 0.58 0.51Canopy transmittance method 0.68 –

value was obtained from DCP, without correcting for clumping.When correction for clumping was applied, the RMSE values fromLAI-2000 PCA were similar to those obtained from digital photogra-phy. In contrast, poor results were obtained from the LAI-2000 PCAwithout correction for clumping and canopy transmittance meth-ods. Using GLA software, DHP produced poorer results (Table 4).

Methods in which the RMSE was below or equal to 0.50 werefurther analyzed using RMA regression (Table 5). On the basis ofthe RMA regression results, the DHP method outperformed the DCPmethod, because the slopes were closer to unity and the interceptswere closer to zero than for the DCP method (Table 5), even thoughboth methods showed high correlation with LAILT values. Also, abias from a 1:1 relationship was observed when LAI values fromDCP were corrected for clumping, because the slope significantlydiffered from unity, in spite of the strong correlation with LAILTvalues (Table 5).

On average, DHP underestimated LAI (Fig. 5) when com-pared with LAILT values, with an average underestimation of−23.3% ± 2.9%. Underestimation was lower when correction forclumping was applied in DHP, arriving at −11.7% ± 3.0%. The accu-racy of LAI estimates from DHP appeared insensitive to LAILT values,unlike DCP (Fig. 5).

Prior to correcting for clumping, DCP generally underestimatedLAI, with an average underestimation of −11.7% ± 3.3%, which wassimilar to results obtained from the corrected LAI values from DHP(Fig. 5). After correcting for clumping, DCP overestimated LAI in theless dense canopies (LAILT < 5.5), while underestimating LAI in themore dense canopies (LAILT > 5.5, Fig. 5).

LAI calculated using 1-azimuth segment method from PCA andDHP were highly correlated (Table 6); DHP yielded slightly larger

Table 5RMA regression coefficients for leaf area index (LAI) from indirect methods and fromlitter traps. The expression used for regressions was: LAIindirect = a LAILT + b.

Method Correction a b R2 p

DCP NC 0.72 0.79 0.85 0.01C 0.50 3.02 0.88 0.01

DHP – Winscanopy NC 0.83 −0.36 0.82 0.01C 0.91 −0.18 0.87 0.01

Highlighted in bold are the intercepts for which the 95% confidence interval includeszero, and slopes that do not significantly differ from unity (p < 0.05).

Table 6RMA regression coefficients obtained from LAI-2000 PCA (y-axis) and DHP (x-axis).The expression used for regression was: yPCA = a xDHP + b.

y a b R2 p

LAI NC 1.36 −1.51 0.91 0.01LAI C 1.34 −1.35 0.83 0.01˝APP 4.65 −3.59 0.20 0.14

Highlighted in bold are the intercepts for which the 95% confidence interval includeszero, and slopes that do not significantly differ from unity (p < 0.05). LAI NC refersto the LAI calculated by taking the log of the gap fractions averaged for multiplesamples; LAI C refers to the LAI computed by averaging the logs of the individualgap fraction of each sample. ˝APP is the apparent clumping index calculate as theratio of LAI NC to LAI C.

Fig. 5. Leaf area index from digital photography vs LAILT from the litterfall method.For DHP, data are presented, either with Lang and Xiang (1986) correction for clump-ing (filled circles) or without correction for clumping (empty circles). For DCP, dataare presented, either with correction for clumping (filled squares) or without cor-rection for clumping (empty squares). The dashed line indicates the 1:1 relationship.

estimates of LAI. By contrast, the apparent clumping indices werenot correlated and differed significantly (p > 0.05, paired t-test,Table 6). Leaf area indices calculated from 1-azimuth segmentmethod were statistically different from those obtained from 8-azimuth segments method (p > 0.01, paired t-test).

The differences between DCP and DHP were analyzed by com-paring foliage and crown cover from DCP and canopy closure(1-openness) from DHP, despite the fact the two measures werenot directly comparable (Fig. 6). In plots with fc > 0.9 there was goodagreement between crown cover from DCP and 1-openness fromDHP; LAI from the two methods also agreed well in these plots.In plots with fc < 0.9 there was poorer agreement between the twomethods.

Foliage cover and crown cover were strongly correlated withLAI, calculated from 8-azimuth segments method in DHP; 95% con-fidence intervals for all the intercepts did not include zero:

ff = 0.72 + 0.040 LAI NC, p = 0.01, R2 = 0.76 (10)

fc = 0.78 + 0.036 LAI NC, p = 0.01, R2 = 0.69 (11)

ff = 0.72 + 0.035 LAI C, p = 0.01, R2 = 0.67 (12)

fc = 0.78 + 0.031 LAI C, p = 0.01, R2 = 0.60 (13)

With regard to transmittance, a strong correlation between thevalues estimated from DHP and with the AccuPAR ceptometer wasobserved (Table 7), regardless of which radiation type (direct, dif-fuse, or total) was considered, and regardless of which software

Author's personal copy

136 F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139

Fig. 6. Comparison of canopy closure (1-openness), calculated from DHP, and eitherfoliage cover (filled triangles) or crown cover (empty triangles) from DCP. The 1:1relationship between canopy closure and crown cover is indicated (dashed line).

package was used for DHP. None of the regression lines differed sig-nificantly from the 1:1 relationship. On average, the DHP methodoverestimated transmittance values using Winscanopy software,but underestimated transmittance using GLA software, althoughthe average differences were rather small for both software pack-ages (approximately ±5–10%), regardless of the radiation typeconsidered. However, the high RMSE obtained (∼1.50) suggestedthe results should be interpreted with caution. The high variabil-ity observed in transmittance values measured with a ceptometer(CV = 44.4%) probably affected the accuracy of the reference mea-surements; as such, measurements taken at midday could not bereliable or satisfactory for comparing the two methods.

There was good agreement between the two software packages(GLA and Winscanopy) used to process the fisheye photographyimages (Table 8). A strong correlation was identified between the

Table 7RMA regression coefficients obtained from fisheye photography and AccuPAR cep-tometer. The expression used for regressions was: yDHP = a xAccuPAR + b.

Software y a b R2 p

GLA DIF 1.09 −1.09 0.76 0.01DIR 1.52 −3.73 0.69 0.02TOT 1.23 −1.83 0.72 0.01

WINSCANOPY DIF 1.35 −1.95 0.86 0.01DIR 1.57 −3.48 0.74 0.01TOT 1.30 −1.27 0.70 0.02

Highlighted in bold are the intercepts for which the 95% confidence interval includeszero, and slopes that do not significantly differ from unity (p < 0.05). DIF, DIR andTOT denote diffuse, direct and total percent transmitted radiation, respectively.

Table 8RMA regression coefficients obtained from fisheye photography using GLAand Winscanopy software packages. The expression for regressions was:yGLA = a xWinscanopy + b.

y a b R2 p

LAI NC 0.99 −0.30 0.89 0.01Openness 0.85 −0.79 0.80 0.01DIF 0.80 0.53 0.79 0.01DIR 0.98 −0.43 0.87 0.01TOT 0.96 −0.70 0.80 0.01

Highlighted in bold are the intercepts for which the 95% confidence interval includeszero, and slopes that do not significantly differ from unity (p < 0.05). Openness is thefraction of open sky measured under the canopy. DIF, DIR and TOT denote diffuse,direct and total percent transmitted radiation, respectively. LAI NC refers to LAI 5ring outputs in GLA and to the averaged linear General and Ellipsoidal LAI outputsin Winscanopy, respectively.

two software outputs in all variables tested, and all regressionswere significant (Table 8). Overall, the RMA regressions confirmedthe tendency of GLA to underestimate the variables compared withthe performance of Winscanopy.

4. Discussion

Although data were collected from stands with different speciescompositions, ages, structures and silvicultural treatments, LAIestimated by litter traps was in the range of values reportedfor deciduous forests (Jarvis and Leverenz, 1983); in particular,previous studies reported LAI values estimated by litter trapsfor beech forests ranging from 5.0 to 10.2 m2 m−2 (Bréda, 2003;Lebourgeois et al., 2005; Meier and Leuschner, 2008), and LAI val-ues for deciduous forests of chestnut and oak spp. ranging from 1.71to 7.45 m2 m−2 (Dufrêne and Bréda, 1995; Thimonier et al., 2010).

Despite the great differences in the plot area sampled betweenlitterfall and indirect methods, estimates of LAI from digital pho-tographic methods that agreed with LAI from litter traps to within±25% were obtained; thus, LAILT can represent the DHP plot area.

Dense and overstocked forest stands usually represent chal-lenges to the accurate estimation of canopy properties. In thepresent study, both digital photographic methods provided esti-mates of canopy properties that satisfactorily agreed with thereference methods, in spite of the differences in theoretical back-ground, field procedures and processing steps.

With respect to LAI, fisheye photography showed a tendencyto underestimate LAI compared with the litter trap method. Aver-age underestimation decreased when correction for clumping wasapplied, which agrees with results from other studies (Chen andCihlar, 1995; Lang and Xiang, 1986; Leblanc, 2002; Macfarlaneet al., 2007b; Van Gardingen et al., 1999). This result is probablycaused by the tendency of foliage to concentrate in the upper partof the crown, which is frequent in stands with dense canopy cover,thus resulting in a non-random distribution of foliage within thecanopy (Sampson and Allen, 1995). This was also supported bythe clumping indices calculated from both DCP (˝0 ∼ 0.81) andDHP (˝LX ∼ 0.87), using 8-azimuth segments method, which inval-idates the random azimuth distribution of foliage assumed in ˝APP.Moreover, within-crown clumping appeared to decrease as LAIincreased, when presumably the degree of clumping than can bedetected was less (Van Gardingen et al., 1999). On the other hand,both DHP and PCA showed ˝APP > 0.9, which would indicate littlebetween-crowns clumping. As such, dense canopies were charac-terized by higher within-crown clumping than between-crownsclumping. It is reasonable to infer that the more azimuth segmentswere used, the more heterogeneity is detected (Van Gardingenet al., 1999). This also implies that small segments can detectclumping better than large segments. In DHP, the whole circularazimuth range was used, while in PCA it was restricted to 90◦ of theview cap, namely one quarter of the azimuth segment size used inDHP; the ˝APP from DHP detected 3% of clumping effect, while ˝APPfrom PCA detected 6% of clumping effect. Hence, quadrupling theazimuth size reduced the level of clumping that can be detected by50%. We considered the significant differences observed betweenthe two apparent clumping indices as a consequence of the differ-ent azimuth size used in DHP and PCA. To sum up, it is likely thatuse of a singular azimuth segment may be unable to yield a reli-able and useful clumping index in dense canopies. However, the1-azimuth method might be more applicable in sparse and hetero-geneous canopies, in which clumping highly occurs at stand scale,such as savanna ecosystems, for which the method was originallydeveloped (Ryu et al., 2010a).

Gap fraction estimated at the zenith from DCP was quite largerthan those from DHP, probably because DCP had higher vertical

Author's personal copy

F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139 137

resolution, which may result in better accuracy in detecting smallgaps in dense canopies, as compared to DHP. Previous studiesreported the accuracy of DCP in assessing gap fraction in sparseto moderately dense canopies (Macfarlane et al., 2007b; Pekin andMacfarlane, 2009; Ryu et al., 2010b); the current study even sup-ported the accuracy of DCP in assessing gap fraction in more densecanopies.

Since gap fraction estimated from DCP was correctly estimated,and since the clumping correction used in DCP can be consid-ered accurate, we were able to derive reliable k values to verifythe spherical leaf angle assumption. On average, the calculated ksuggested a nearly spherical leaf angle distribution; nonetheless,an apparent sensitivity of k to LAI was observed in the currentstudy, which was in accordance with previous reports (Cannellet al., 1987; Johansson, 1989; Macfarlane et al., 2007a; Smith et al.,1991). However, different k values would probably reflect speciesdifferences, rather than LAI differences, as observed in the currentanalysis. As such, the different k obtained for the three species canbe regarded as reliable, and can be further applied for routine LAIestimation in deciduous stands. Nonetheless, Bolstad and Gower(1990) found error in LAI estimation was 6% when k was measured,but increased to 20% when k was assumed. So, we considered thesystematic bias from 1:1 observed in corrected LAI outputs resultsfrom DCP as a consequence of discrepancies from the spherical leafangle distribution assumed, for which k = 0.50 at the zenith. Thisimplies that direct measurement of k is preferable, whereas it waspossible.

In contrast to the DCP method, the frequency of LAI under-estimation was relatively constant with corrected DHP (∼12%),because the extinction coefficient is not assumed but automaticallyestimated. DHP also suggested a nearly spherical leaf angle distri-bution, although foliage angle distribution function calculated fromfisheye images should be treated with caution (Macfarlane et al.,2007a). The zenithal extinction coefficient calculated from fish-eye image appeared to increase as LAI increased, while the meanleaf angle decreased, thus showing an opposite trend comparingwith DCP. Similar results were also observed by Macfarlane et al.(2007a). Accordingly, poor sampling near the zenith from hemi-spherical sensor limits the accuracy of k near the zenith from DHP(Macfarlane et al., 2007c).

An inherent limitation of hemispherical photography is the needto set an adequate exposure manually, which dramatically affectsLAI estimation, especially in dense canopies (Chen et al., 1991;Zhang et al., 2005). In the present study, adjustment by two stopsof overexposure relative to the open sky reference was used, basedon previous reports (Macfarlane et al., 2007c; Zhang et al., 2005),and was supported by the accurate gap fraction obtained from DHP.We did not test the effect of correcting or not correcting the gammafunction on the results, but clearly if the correction is applied, asin the current study, it represents an additional step in the DHPimage-processing procedure, which was already more tedious andtime-consuming (Jonckheere et al., 2004; Macfarlane et al., 2007c)compared with DCP.

The best agreement between crown cover from DCP and canopyclosure from DHP was achieved when crown cover was above 0.9.Although the two measures were not directly comparable, we couldinfer the number of images collected may have been insufficientand biased at the sites where canopies were less homogeneous(crown cover was <0.9), and the resulting differences in LAI fromthe two photographic methods were higher.

One advantage of DHP is the ability to estimate parametersrelated to the forest light environment. In the present study weobtained good estimates of transmittance, which agreed with thoserecorded with an AccuPAR ceptometer in sunny conditions. Allof the radiation parameters were correlated with the referencemethod. However, the high RMSE suggested these results must be

treated with caution. Some authors note instantaneous diffuse lighttransmission measured on overcast days is more stable and showedlower spatial variability than those measured at midday, and thusfewer sample points are required to achieve good accuracy (Messierand Puttonen, 1995; Parent and Messier, 1996). It is likely the num-ber of sample points collected should be larger than that used in thecurrent study. Also, it is reasonable to infer measurement of diffusetransmitted light with a ceptometer under completely overcast skyconditions could improve the accuracy of the reference methods;thus, diffuse transmitted light measured with DHP and ceptometerscould provide more reliable parameters for comparison of the twomethods (Gendron et al., 1998), particularly for dense canopies.

The two software packages used in DHP analysis showedmarked differences in performance. Overall, Winscanopy per-formed better than GLA. One advantage of the former is that pixelclassification in the canopy and sky (thresholding) is performedautomatically, whereas in GLA it is performed manually; thus,greater human input is required with GLA, which may lead to inad-equate image registration (Jarcuska et al., 2010), thereby reducingthe accuracy of the estimates. In the current study we used theautomatic Winscanopy threshold for both software analyses. Thismight have resulted in poorer estimates of LAI using GLA, sincethe two programs differ in the models used. Another advantage ofWinscanopy is that the most recent releases of the software are ableto provide DCP estimates (since the 2006a version), thus avoidingsubjectivity in separating large gaps from small gaps. However, inthe current study large gaps from DCP were estimated by two dif-ferent operators and very little differences were observed betweenthem. One advantage of GLA is the software is available for free.

In summary, the agreement between the two photographic anddirect methods for estimating LAI and transmittance was satis-factory. Both the DHP and DCP methods showed good potentialto replace other indirect methods, such as the LAI-2000 PCA andAccuPAR ceptometer, in broadleaf forests, which agrees with thefindings of other studies (Leblanc et al., 2005; Macfarlane et al.,2007b). Nevertheless, users should consider the following advan-tages and disadvantages of the two photographic methods.

• DHP is sensitive to photographic exposure, which is a major causeof error (Chen et al., 1991; Zhang et al., 2005). Some previousstudies also found the gamma function and thresholding stronglyaffect the accuracy of the results (Cescatti, 2007; Jonckheere et al.,2005; Macfarlane et al., 2007b). Consequently, this method istedious and time-consuming because several image-processingsteps are required. Photographs must be taken close to sunrise(or sunset), not during common working hours. The main advan-tages of the method are DHP does not require assumptions to bemade concerning the extinction coefficient and DHP also allowscharacterization of radiation regimes in forest canopies.

• DCP does not require measurement of photographic exposure;the method is quick, simple, and can be applied during nor-mal working hours (Pekin and Macfarlane, 2009). DCP is ableto provide additional information aside from LAI, such as crowncover, porosity, foliage cover and clumping index at the zenith(Macfarlane et al., 2007c). The major drawback of this method,the need to know the value of k, was confirmed in our study.

5. Conclusions

Both digital photographic methods are useful and inexpen-sive tools to quantify forest canopy properties. Since good resultswere obtained using a compact camera, with about 4 megapixelresolution, it is reasonable to infer higher quality cameras couldimprove the performance of both photographic methods (Pekinand Macfarlane, 2009). The more recently developed DCP method

Author's personal copy

138 F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139

offers strong potential for indirect measurement and monitor-ing of LAI and other canopy properties in deciduous forests. Evenwithout measurement of the zenithal light extinction coefficient,the method provided accurate estimates of LAI in dense decidu-ous forests, assuming a spherical leaf angle distribution. Also, DCPis able to measure several parameters that characterize a forestcanopy, such as crown cover, which do not require either conver-sion to LAI nor the foliage inclination angle distribution. So, coverphotography represents a useful new procedure for monitoringdeciduous forests because it is fast, inexpensive, and simple to use,permitting widespread use of the method. However, in situationswhere no direct reference measurements of k are available, andestimates of LAI are required, use of DHP in conjunction with DCPcan provide a valuable ‘reality-check’ to locally calibrate k, by meansof correlations between LAI from DHP and fc, ff from DCP; cover pho-tography could then be used more routinely to large areas, basedon the cross-calibrated k values.

Acknowledgments

This research was supported by RI.SELV.ITALIA Research Pro-gram 3.1 – “Silviculture, productivity and conservation of forestecosystems” research project and by the Research Program D.M.19477/7301/08 – “Maintenance of collections, databases, and otheractivities of public interest” funded by the Italian Ministry ofAgriculture and Forest Policies. Field assistance by the technicalstaff of the Research Centre for Silviculture (CRA-SEL) is grate-fully acknowledged, with particular thanks to Tessa Giannini andUmberto Cerofolini. We would like to thank Giovanni Vagnoli andGiovanni Venturi for assistance with field data collection. We alsothank an anonymous reviewer for the helpful and insightful com-ments, which help us to greatly improve the manuscript.

References

Amiro, B.D., Chen, J.M., Liu, J., 2000. Net primary productivity following forest firefor Canadian ecoregions. Can. J. Forest Res. 30, 939–947.

Anderson, M.G., 1966. Stand structure and light penetration. II. A theoretical analysis.J. Appl. Ecol. 3, 41–54.

Boer, M.M., Macfarlane, C., Norris, J., Sadler, R.J., Wallace, J., Grierson, P.F., 2008.Mapping burned areas and burn severity patterns in SW Australian eucalyptforest using remotely-sensed changes in leaf area index. Remote Sens. Environ.112, 4358–4369.

Bolstad, P.V., Gower, S.T., 1990. Estimation of leaf area index in fourteen southernWisconsin forest stands using a portable radiometer. Tree Physiol. 7, 115–124.

Bréda, N., 2003. Ground-based measurements of leaf area index: a review of meth-ods, instruments and current controversies. J. Exp. Bot. 54, 2403–2417.

Campbell, G.S., 1986. Extinction coefficients for radiation in plant canopies calcu-lated using an ellipsoidal inclination angle distribution. Agric. For. Meteorol. 36,317–321.

Cannell, M.G.R., Milne, R., Sheppard, L.J., Unsworth, M.H., 1987. Radiation intercep-tion and productivity of willow. J. Appl. Ecol. 24, 261–278.

Cescatti, A., 2007. Indirect estimates of canopy gap fraction based on the linearconversion of hemispherical photographs. Methodology and comparison withstandard thresholding techniques. Agric. For. Meteorol. 143, 1–12.

Chen, J.M., Black, T.A., 1992. Defining leaf area index for non-flat leaves. Plant CellEnviron. 15, 421–529.

Chen, J.M., Black, T.A., Adams, R.S., 1991. Evaluation of hemispherical photographyfor determining plant area index and geometry of a stand. Agric. For. Meteorol.56, 129–132.

Chen, J.M., Cihlar, J., 1995. Quantifying the effect of canopy architecture on opti-cal measurements of leaf area index using two gap size analysis methods. IEEETrans. Geosci. Remote Sens. 33, 777–787.

Chen, J.M., Rich, P.M., Gower, S.T., Norman, J.M., Plummer, S., 1997. Leaf area indexof boreal forests: theory, techniques, and measurements. J. Geophys. Res. 102,29429–29443.

Cutini, A., 1996. The influence of drought and thinning on leaf area index estimatesfrom canopy transmittance method. Ann. Sci. For. 53, 595–603.

Cutini, A., 2003. Litterfall and leaf area index in the CONECOFOR permanent moni-toring plots. J. Limnol. 61, 62–68.

Cutini, A., Matteucci, G., Mugnozza, G.S., 1998. Estimation of leaf area index with theLi-Cor LAI 2000 in deciduous forests. Forest Ecol. Manage. 105, 55–65.

Dufrêne, E., Bréda, N., 1995. Estimation of deciduous forest leaf area index usingdirect and indirect methods. Oecologia 104, 156–162.

Frazer, G.W., Canham, C.D., Lertzman, K.P., 1999. Gap Light Analyzer (GLA), Version2.0: Imaging Software to Extract Canopy Structure and Gap Light Transmis-sion Indices from True-color Fisheye Photographs, User’s Manual and ProgramDocumentation. Simon Fraser University, Burnaby, BC, Canada, pp. 36.

Fuentes, S., Palmer, A.R., Taylor, D., Zeppel, M., Whitley, R., Eamus, D., 2008. Anautomated procedure for estimating the leaf area index (LAI) of woodlandecosystems using digital imagery, MATLAB programming and its applicationto an examination of the relationship between remotely sensed and field mea-surements of LAI. Funct. Plant. Biol. 35, 1070–1079.

Gendron, F., Messier, C., Comeau, P., 1998. Comparison of various methods for esti-mating the mean growing season percent photosynthetic photon flux densityin forests. Agric. For. Meteorol. 92, 55–70.

Jarcuska, B., 2008. Methodological overview to hemispherical photography, demon-strated on an example of the software GLA. Folia Oecol. 35, 66–69.

Jarcuska, B., Kucbel, S., Jaloviar, P., 2010. Comparison of output results from two pro-grammes for hemispherical image analysis: gap light analyser and Winscanopy.J. For. Sci. 56 (4), 147–153.

Jarvis, P.G., Leverenz, J.W., 1983. Productivity of temperate, deciduous and evergreenforests. In: Lange, O.L., Nobel, P.S., Osmond, C.B., Ziegler, H. (Eds.), PhysiologicalPlant Ecology, IV. Encyclopedia of Plant Physiology, vol. 12D. Springer-Verlag,New York, pp. 233–280.

Johansson, T., 1989. Irradiance within the canopies of young trees of European aspen(Populus tremula) and European birch in stands of different spacing. Forest Ecol.Manage. 28, 217–236.

Jonckheere, I., Fleck, S., Nackaerts, K., Muys, B., Coppin, P., Weiss, M., Baret, F.,2004. Review of methods for in situ leaf area index determination: Part I.Theories, sensors and hemispherical photography. Agric. For. Meteorol. 121,19–35.

Jonckheere, I., Nackaerts, K., Muys, B., Coppin, P., 2005. Assessment of automatic gapfraction estimation of forests from digital hemispherical photography. Agric. For.Meteorol. 132, 96–114.

Lang, A.R.G., Xiang, Y., 1986. Estimation of leaf area index from transmission of directsunlight in discontinuous canopies. Agric. For. Meteorol. 35, 229–243.

Leblanc, S.G., 2002. Correction to the plant canopy gap-size analysis theory usedby the tracing radiation and architecture of canopies instrument. Appl. Opt. 41,7667–7670.

Leblanc, S.G., Chen, J.M., Fernandes, R., Deering, D.W., Conley, A., 2005. Methodologycomparison for canopy structure parameters extraction from digital hemispher-ical photography in boreal forests. Agric. For. Meteorol. 129, 187–207.

Lebourgeois, F., Bréda, N., Ulrich, E., Granier, A., 2005. Climate-growth relationshipsof European beech (Fagus sylvatica L.) in the French permanent plot network(RENECOFOR). Trees 19, 385–401.

Macfarlane, C., 2011. Classification method of mixed pixels does not affect canopymetrics from digital images of forest overstorey. Agric. For. Meteorol. 151,833–840.

Macfarlane, C., Arndt, S., Livesley, S.J., Edgar, A.C., White, D.A., Adams, M.A., Eamus,D., 2007a. Estimating of leaf area index in eucalypt forest with vertical foliage,using cover and fullframe fisheye photography. Forest Ecol. Manage. 242,756–763.

Macfarlane, C., Grigg, A., Evangelista, C., 2007b. Estimating forest leaf area usingcover and fullframe fisheye photography: Thinking inside the circle. Agric. For.Meteorol. 146, 1–12.

Macfarlane, C., Hoffman, M., Eamus, D., Kerp, N., Higginson, S., McMurtrie, R., Adams,M.A., 2007c. Estimation of leaf area index in eucalypt forest using digital pho-tography. Agric. For. Meteorol. 143, 176–188.

Meier, I.C., Leuschner, C., 2008. Leaf size and leaf area index in Fagus sylvaticaforests: competing effects of precipitation, temperature and nitrogen availabil-ity. Ecosystems 11, 655–669.

Messier, C., Puttonen, P., 1995. Spatial and temporal variation in the light environ-ment of developing Scots pine stands: the basis for a quick and efficient methodfor characterizing light. Can. J. For. Res. 25, 343–354.

Parent, S., Messier, C., 1996. A simple and efficient method to estimate micrositelight availability under a forest canopy. Can. J. For. Res. 26, 151–154.

Pekin, B., Macfarlane, C., 2009. Measurement of crown cover and leaf area indexusing digital cover photography and its application to remote sensing. RemoteSens. 1, 1298–1320.

Pierce, L.L., Running, S.W., 1988. Rapid estimation of coniferous forest leaf area indexusing a portable integrating radiometer. Ecology 69, 1762–1767.

R Development Core Team, 2011. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria.

Ross, J., 1981. The Radiation Regime and Architecture of Plant Stands. Junk, TheHague, pp. 391.

Ryu, Y., Nilson, T., Kobayashi, H., Sonnentag, O., Law, B.E., Baldocchi, D.D., 2010a. Onthe correct estimation of effective leaf area index: does it reveal information onclumping effects? Agric. For. Meteorol. 150, 463–472.

Ryu, Y., Sonnentag, O., Nilson, T., Vargas, R., Kobayashi, H., Wenk, R., Baldocchi, D.D.,2010b. How to quantify tree leaf area index in an open savanna ecosystem: amulti-instrument and multi-model approach. Agric. For. Meteorol. 150, 63–76.

Sampson, D.A., Allen, H.L., 1995. Direct and indirect estimates of leaf area index (LAI)for lodgepole and loblolly pine stands. Trees 9, 119–122.

Smith, F.W., Sampson, D.A., Long, J.N., 1991. Comparison of leaf area index esti-mates from tree allometrics and measured light interception. For. Sci. 37, 1682–1688.

Thimonier, A., Sedivy, I., Schleppi, P., 2010. Estimating leaf area index in differenttypes of mature forest stands in Switzerland: a comparison of methods. Eur. J.Forest Res. 129, 543–562.

Author's personal copy

F. Chianucci, A. Cutini / Agricultural and Forest Meteorology 168 (2013) 130– 139 139

Van Gardingen, P.R., Jackson, G.E., Hernandez-Daumas, S., Russell, G., Sharp,L., 1999. Leaf area index estimates obtained for clumped canopiesusing hemispherical photography. Agric. For. Meteorol. 94, 243–257.

Vansereven, J.P., 1969. Recherches sur l’écosysteème foret. Série B: La chenaiemélangée calcicole de Virelles-Blaimont. Contribution n.30: L’ index foli-aire et sa measure par photoplanimétrie. Bull. Soc. R. Bot. Belg. 102, 373–385.

Warton, D.I., Wright, I.J., Falster, D.S., Westoby, M., 2006. Bivariate line-fitting modelsfor allometry. Biol. Rev. 81, 259–291.

Welles, J.M., 1990. Some indirect methods of estimating canopy structure. RemoteSens. Rev. 5, 31–43.

Welles, J.M., Norman, J.M., 1991. Instrument for indirect measurement of canopyarchitecture. Agron. J. 83, 818–825.

Zhang, Y., Chen, J.M., Miller, J.R., 2005. Determining digital hemispherical pho-tograph exposure for leaf area index estimation. Agric. For. Meteorol. 133,166–181.

Zhongguo, M.A., Chen, X., Guli, J., Chang, C., 2009. Study on LAI estimation ofbroadleaf forests in arid areas using digital hemispherical photography: theTarim River Basin, China. J. Environ. Technol. Eng. 2 (3), 106–111.