-1

Use of digital photography for analysis of canopy closure

A. Guevara-Escobar*, J. Tellez and E. Gonzalez-Sosa
Facultad de Ciencias Naturales, Centro Universitario s/n, Santiago de Querétaro, Universidad Autónoma de
Querétaro, Querétaro 76010, México; *Author for correspondence (e-mail: guevara@uaq.mx; phone:
+52-442-192-1200, ext: 5321; fax: +52-442-192-1200, ext: 5318)

Received 7 May 2003; accepted in revised form 9 January 2005

Key words: Canopy analyser, Canopy openness, Canopy structure, LAI-2000

Abstract

The relationships between trees and understory crops are very important in agroforestry systems. Also,
above ground interactions can be related to canopy structure. However, measurements of canopy structural
parameters, either destructive or indirect, are time-consuming or prohibitively expensive. The present work
explored the use of digital photography as a simple method to characterise the extent of canopy closure
(CC), defined as the area of tree canopies projected onto the horizontal ground surface beneath, and
expressed as a percentage of the ground covered. Measurements were made in two Eucalyptus (Eucalyptus
nitens, Deane and Maiden) plantations and a subtropical mixed legume woodland dominated by Albizia
(Albizia sp), Kidneywood (Eysenhardtia sp.) and Desert Fern (Lysiloma sp.). Images were captured at dawn
to minimise light scattering and the number of sunlit foliage elements. Mean CC estimates provided by
analysis of images obtained using digital cameras with contrasting performance, a Kodak DC-120 and a
Canon EOS D1, were similar in precision and accuracy both between the two cameras and to those
provided by a Li-Cor LAI-2000 canopy analyser. Bias between the estimates provided by the Kodak and
Canon cameras was �0.02, between the Kodak and LAI-2000 was �0.07 and between the Canon and LAI-
2000 was �0.05. Data from a pruning experiment using alder also demonstrated the repeatability of
estimates obtained with a photographic method using the Kodak camera. The number of ring sensors
within the LAI-2000 used to estimate CC affected agreement between the photographic method and the
LAI-2000.

Introduction

Tree canopy structure is a major factor defining
competitive interactions in agroforestry systems.
Due to growth, decay or management, the tree
canopy structure is constantly changing in time
and space. Canopy structure refers to the distri-
bution of positions, orientations, areas and shapes
of plant organs (Welles and Norman 1991). Tree
canopy structure must be characterised to inves-
tigate and explain its effect on biophysical and

ecological processes. Relationships between can-
opy dimensions, sapwood area or basal area are
often established from canopy attributes to avoid
labour-intensive sampling. Alternatively, canopy
attributes such as leaf area or light interception
may be measured indirectly using different optical
instruments, particularly photography.

Ansley et al. (1992) predicted leaf area using
photographs of tree profiles and a planimeter to
measure the area of a hand-drawn line delineating
the canopy profile. Grace and Fownes (1998)

Agroforestry Systems (2005) � Springer 2005
DOI 10.1007/s10457-005-0504-y



estimated leaf area using film slides and a dot grid
to manually count the number of dots intersecting
the canopy profile. Both authors found a good
relationship between estimates of leaf area
obtained by photography as compared to har-
vesting entire trees. Knowles et al. (1999) esti-
mated canopy closure (CC), defined as the area of
the tree canopy projected onto the horizontal
ground surface below, using digitised still images
taken from 8 mm video film; the canopy area was
determined using digital image analysis. They
found a good relationship between the extent of
CC and stand parameters such as basal area and
the ratio of green crown length to mean tree height.

Hemispherical photography combined with
digital image analysis is used routinely to estimate
canopy attributes such as leaf area or foliage
inclination angle (Chen et al. 1991). Photographs
are taken from below the canopy looking upwards
under conditions where there is high contrast
between leaves and the sky. Film-based hemi-
spherical photographs have disadvantages because
they cannot be reviewed in the field, and the film
needs to be processed before being digitised. Film
images lose resolution because they cannot pre-
serve a higher resolution than that provided by the
scanner used. Also, few digital cameras have a
fisheye lens, and digital reflex cameras that can
accommodate a fisheye lens are expensive. Frazer
et al. (2001) reported that estimates of canopy
closure were 1.4 times greater when obtained using
a digital camera than a film-based camera, both
using hemispherical lenses.

Comparisons of estimates of CC obtained using
film-based or digital hemispherical photographs
showed that image resolution, file compression,
shutter speed and heterogeneity in sky luminance
determine the accuracy of photographic estimates
(Chen et al. 1991; Macfarlane et al. 2000; Frazer
et al. 2001). Overexposure is the main problem
associated with errors in the measurement of CC
because this causes excessive light scattering and
diffraction along the edges of branches and leaves
(Chen et al. 1991; Frazer et al. 2001). The accuracy
of digital image analysis depends on the edge
definition between the structures, in this case the
canopy and sky areas. Edge definition or sharpness
is obtained by using the correct shooting speed,
adequate contrast and film or sensor sensitivity
(ISO speed). The accuracy of the imaging system
also depends on tree architecture. The quality of

digital images can also be improved by increasing
image resolution, and photographs are taken
under diffuse sky conditions because the range of
brightness of foliage is minimised.

The LAI-2000 canopy analyser (Li-Cor Inc.,
Lincoln, NE, USA) is an indirect method for
describing canopy attributes. This instrument
estimates leaf area index (LAI) and the fraction
of sky visible from beneath the canopy. The
LAI-2000 method is fast, but it is sensitive to
illumination conditions and stand boundary ef-
fects (Welles and Cohen 1996). These problems
are reduced using sensor view- caps, obtaining
measurements when the sun is low or protecting
the sensor from direct sunlight (Li-Cor 1992).
Although the effects of multiple light scattering
can be corrected (Leblanc and Chen 2001) actual
field LAI measurements are still routinely done
under a range of illumination conditions,
including direct sunlight.

The estimates of LAI obtained can be up to 40%
lower than those provided by direct measurements,
especially in canopies with large gaps (Welles and
Norman 1991). The protocol recommended by
Li-Cor (1992) results in substantial underestima-
tion of direct measurements of LAI in agroforestry
tree rows and vineyard canopies (Broadhead et al.
2003; Johnson and Pierce 2004). Estimates of LAI
provided by hemispherical photography and the
LAI-2000 differed only by 10% in forest canopies,
but this relationship is sensitive to sky conditions,
exposure and field of view (Welles and Cohen
1996). Larger differences in the estimates of LAI
obtained by hemispherical photography and the
LAI-2000 have also been found (Frazer et al.
1998). Similarly, Macfarlane et al. (2000) reported
that the photographic method underestimated LAI
by 16–30% when compared to the LAI-2000
method. However, good agreement between direct
methods and estimation of LAI with a LAI-2000
was found in a forest canopy (López-Serrano et al.
2000).

Estimates of CC obtained using digital cameras
are a potentially valuable alternative to the LAI-
2000 method because they are very economical,
and provide a permanent visual record of what
was measured. If image analysis of digital photo-
graphs can provide reliable estimates of CC with
the same precision as the LAI-2000, then canopy
management decisions could be made based on
this relatively straightforward and economical



approach. However, there are many kinds of
cameras providing a range of electronic capabili-
ties and lens quality. This would affect the accu-
racy of estimates of CC and calibration would be
necessary. The following study reported here
compared estimates of CC obtained with digital
photography using various camera settings and
those provided by the LAI-2000. Three digital
cameras were compared. It was hypothesised that
estimates of CC provided by the digital photog-
raphy approach would have a similar variance to
those provided by the LAI-2000. A second
hypothesis was that good agreement would be
obtained between estimates of CC provided by the
tested cameras.

Material and methods

Study locations

Measurements were made at three sites in Mexico.
The Queretaro site (1867 m.a.s.l. 20�36¢ N,
100�22¢ W) had 45-year old Eucalyptus trees
(Eucalyptus nitens (Deane and Maiden)), which
were spaced 10–20 m apart, 25–30 m in height and
with heavy branching. The Huimilpan site
(2318 m.a.s.l. 20�22¢ N, 100�16¢ W) consisted of a
12-year-old E. nitens plantation initially spaced at
4 · 6 m; the trees were 7–12 m tall. The Amazcala
site (1919 m.a.s.l. 20�41¢ N, 100�16¢ W) comprised
a mixed subtropical woodland dominated by Al-
bizia (Albizia sp.), Kidneywood (Eysenhardtia sp.)
and Desert fern (Lysiloma sp.); the trees were 6–
7 m in height. These sites were chosen to provide
low, medium and high values of CC.

Data from a fourth site at the Horticultural
Research Centre, Aokautere, New Zealand
(30 m a.s.l. 40�22¢ S, 175�40¢ E) are also presented
for comparison; this site was planted with 11-year
old alder trees (Alnus glutinosa (L.) Gaertn). Three
shade treatments were created by pruning trees to
heights of 2.5, 5.0 and 7.0 m above ground level.
These treatments provided three levels of trans-
mitted photosynthetic active radiation i.e. 17, 27
and 77% of full sunlight as determined from
measurements made at noon on three clear sky
days using a LI-191 SA quantum sensor (Li-Cor
Inc., Lincoln, NE, USA). Plot size was 5 by 7 m
for all pruning treatments; CC was determined at
the centre of each plot (Devkota et al. 2000).

Instrumentation and calculations

Three digital cameras of contrasting capability
were used. A Kodak DC-120 camera with a fixed
39–114 mm f/2.5–3.8 lens and 1280·960 pixel im-
age resolution (Eastman Kodak Corp., Rochester,
NY, USA). A Mavica FD-7 (Sony Corp., New
York, NY, USA) with a fixed 40–400 mm f/1.8–
2.9 lens and 640·480 pixel image resolution. A
Canon EOS D1 with an image resolution of
2464·1648 pixels (Canon Inc., Lake Success, NY,
USA). This camera was fitted with an EF 28–
70 mm f/3.5–5.6 lens.

The cameras were mounted on a tripod, levelled
and oriented to magnetic north. Automatic expo-
sure was used to reduce the possibility of overex-
posure (Macfarlane et al. 2000). A 10 s self-timer
was also used to avoid any movement of the
camera associated with the manual release of the
shutter button (Frazer et al. 2001). Photographs
were taken before sunrise to avoid large variation
in brightness across the picture and reflections of
direct sunlight from leaves or branches (Hale and
Edwards 2002). Digital files were stored as RAW
or KDC formats, which are the least compressed
file types for the Canon and Kodak cameras,
respectively. The Mavica camera stored images as
JPG files.

Each measurement with the LAI-2000 was made
by recording one reference reading ‘above the
canopy’ and four readings below the canopy. The
‘above canopy’ values were determined in an open
area whose radius was more than four times
greater than the adjacent stand height, and using
the same view-cap as was used for the below
canopy measurements (López-Serrano et al. 2000).
A 90� view-cap was used to reduce the bias in the
below canopy measurements caused by heteroge-
neity in gap distribution within the field of view of
the sensor (López-Serrano et al. 2000; Nackaerts
et al. 2000). All measurements were made using a
single LAI-2000 unit. When using this approach, it
is essential that sky conditions remain uniform for
the above and below canopy measurements
(Li-Cor 1992). The readings beneath the canopy
were made facing N, S, W and E at approximately
30 cm from the tripod.

The LAI-2000 measures diffuse sunlight by
five concentric detector rings associated with the
following zenith angles: 0–13�, 16–28�, 32–43�,
47–58� and 61–74�, where zero degrees represents



the direct upward-view. These five rings bands
are hereafter termed ring numbers 1, 2, 3, 4 and
5, respectively. Because the visual angle of the
cameras lens was smaller than that of the LAI-
2000, logged records were edited using the C2000
program and then recalculated using rings
number 1 and 2, 1–3, 1–4 or 1–5. The diffuse
non-interceptance value (DIFN) was used to
estimate CC as (1-DIFN).

Measurements

Test 1 Sets of photographs were obtained to
investigate the accuracy and precision of esti-
mates of CC obtained using the LAI-2000 and
digital photography. The first was collected be-
tween 26 and 28 June 2002 from each of the
three sites in Mexico. Sampling points were se-
lected using the following criteria: three parallel
line transects were laid at each site at least 10 m
apart to prevent sampling the same canopy gap.
All gaps that were approximately 10 m apart
within the transect were identified, at these
ground points, the extend of canopy closure was
measured using the LAI-2000. Four ground
points were randomly selected within the first
and fourth quartiles from the median estimates
of CC obtained at each site. Similarly, five
ground points were randomly selected within the
second and third quartiles. Selecting ground
points form each quartile ensured that a wider
range of canopy closure was included to evaluate
the methods of estimation of CC. The 18 se-
lected ground points were surrounded by trees to
avoid edge effects. A sample consisted of one
photograph obtained using the Kodak camera,
one with the Canon camera and corresponding
paired LAI-2000 readings. All observations were
made during periods of calm weather and clear
sky. Collection started at twilight 1100 h
Greenwich Mean Time (GMT), and approxi-
mately 50 min elapsed between the first and last
image taken. Focal length was set to 50 mm for
both cameras. For each site, a one-way ANOVA
test was used to compare the mean estimates of
CC obtained using digital photography and the
LAI-2000. Estimates of CC obtained using the
LAI-2000 were considered to provide the stan-
dard value. To verify if both estimates of CC
have the same precision, the Bartlett test was

used to compare the variance of the mean esti-
mates of CC. Regression analysis also was used
to compare estimates of CC obtained with the
different methods. Bias was calculated using the
procedures described by Bland and Altman
(1986). The difference between each pair of
estimates of CC obtained with the methods tes-
ted was calculated for each ground point (d).
The mean difference ( �d) and their standard
deviation were used to summarise the lack of
agreement between methods. Bias was estimated
by �d and 95% confidence intervals were also
calculated.

Test 2 A second set of photographs was col-
lected on 4 September 2002 at the Amazcala site.
Two ground points were selected within one
standard deviation from the mean estimate of CC
determined previously in Test 1 with the
LAI-2000. Photographs were taken using the
Canon camera at focal lengths of 28, 35 and
50 mm, respectively corresponding to 59�, 49� and
35� vertical angles of view. Three consecutive
photographs were taken together with one
LAI-2000 measurement. For each ground point,
these measurements were repeated four times every
10 min within a 40-min period. Sampling started
at twilight (1120 h GMT), weather was calm and
sky was cloudy. LAI-2000 measurements were
recomputed using the C-2000 software after elim-
inating data from one ring at a time as described
by Li-Cor (1992). This procedure was adopted to
examine which group of rings yielded the highest
correlation with the photographic estimates of CC.
Data were analysed as repeated measures, one
factor design, the factor being the vertical angle of
view. Replicates were the ground points and the
four repeated measures. One degree of freedom
contrasts were used to compare mean estimates of
CC. Pearson correlation of estimates of CC were
obtained excluding different sensor rings of the
LAI-2000 and the images obtained with different
focal lengths.

Test 3 A third set of 16 photographs was col-
lected on 16 September 2002 at the Amazcala site
to evaluate the relation between estimates of CC
obtained using the LAI-2000 and digital photog-
raphy at different sampling times starting at
twilight (1120 h GMT). A total of eight measure-
ments were made every 20 min for each ground
point during a 150-min period. Other times
of day were not considered because illumination



conditions may affect the photographic and
LAI-2000 estimates of CC (Chen et al. 1991;
Macfarlane et al. 2000; Frazer et al. 2001). Only
the Canon camera at a focal length of 50 mm was
used in this test. Measurements consisted of paired
photographs and LAI-2000 readings. The previ-
ously selected ground points in Test 2 were used
and considered as replicates. The tripod was
repositioned on each occasion at virtually the same
position and orientation. The LAI-2000 readings
were edited as described previously. Weather was
calm and sky conditions were cloudy. Linear
regression was used to explore the relationship
between time and the estimates of CC provided by
both methods.

Test 4 A fourth group of photographs was
obtained between 22 October 1998 and 7 May
1999 in a system containing alder at Aokautere,
New Zealand. Photographs were taken approxi-
mately every 10 days. Eight photographs per
pruning level were taken for each date. Photo-
graphs were taken at twilight, 30–45 min before
dusk, handholding the Kodak camera towards
the canopy and manually releasing the shutter.
Three sets of digital photographs were obtained
using the Mavica camera from 29 January to 21
February 1999.

All models were fitted using MEANS and GLM
procedures within SAS (SAS Institute, Cary, NY,
USA) to establish differences between means and
the significance of regressions. The minimum level
for significance was set at p £ 0.05. Residuals were
checked for normality and independence using
Kolmogorov–Smirnov and Durbin–Watson tests,
respectively.

Analysis of digital images

Images were analysed using Photopaint v 9.0
(Corel Corp., Ottawa, ON, Canada). The thresh-
old to classify pixels into ‘sky’ and ‘canopy’ was
determined for the first image of the set obtained
using each camera and date. The threshold was
saved in a file and then applied to the rest of the
images of the corresponding set. The threshold
was estimated again only when it was noticeable
that pixels pertaining to the sky were wrongly
identified as canopy and vice versa and then
applied to the remaining images. Classified images
were analysed using Sigmascan v 4.0 (SPSS Inc.,
Chicago IL, USA) to calculate canopy area. The
ratio of canopy area to frame area of the image
was expressed as a percentage and used as an
estimate of CC.

Results

Test 1: comparing cameras

Variance and the mean estimates of CC obtained
with the Canon and Kodak cameras set to 50 mm
focal length and those provided by the LAI-2000
were similar at all the sites examined in Mexico
(Table 1). The estimates of CC obtained using the
LAI-2000 presented in Table 1 were calculated
using rings 1–3; similar values were obtained when
the estimates were based on rings 1–2 or 1–4. Only
at the Amazcala site was a significant correlation
found between the estimates provided by digital
photography and the LAI-2000 (r

2
=0.5 and 0.61

Table 1. Canopy closure (CC) estimated by analysis of digital photographs taken with a zoom lens set to 50 mm focal length or the

LAI-2000 canopy analyser using sensor rings number 1–3.

Vertical angle of view

Basal area m
2
ha
�1

Species
b

CC

Kodak DC-120 SE Canon EOS D1 SE LAI-2000 SE

46� 35� 38�

Sites
a

Huimilpan 5.5 E 0.20 0.015 0.24 0.019 0.24 0.022

Queretaro 33.9 E 0.45 0.031 0.51 0.035 0.59 0.017

Amazcala 4.9 A, Ey, L 0.61 0.051 0.58 0.047 0.66 0.030

a
Sites located in Mexico.

b
E – Eucalyptus nitens; A – Albizia sp.; Ey – Eysenhardtia sp.; L – Lysiloma sp.



for the Kodak and Canon cameras, respectively;
p<0.05).

At each site, the relationship between estimates
of CC provided by the Kodak and Canon cam-
eras was linear (r

2
=0.90, 0.71 and 0.92 for

Amazcala, Queretaro and Humilpan, respectively;
p<0.054). The relationship between the Kodak
(y) and Canon (x) cameras using the data from all
sites was also linear (y=0.065+0.89x, r

2
=0.91,

p< 0.05; Figure 1).
Bias between estimates of CC obtained using

the Kodak and Canon cameras was �0.02 with
a 95% confidence interval of ±0.019. Bias be-
tween the Kodak camera and LAI-2000 esti-
mates of CC was �0.07±0.038. Bias between
the Canon camera and the LAI-2000 estimates
of CC was �0.05± 0.035. While the Kodak
camera had a fixed 160 ISO value, the Canon
camera automatically adjusted to the maximum
1600 ISO value and avoided extreme aperture
settings. The Canon camera used diaphragm
aperture settings higher than 8.4 while the
Kodak camera used settings as low as 5.0.
Images obtained using the Kodak camera were
darker, poorer in sharpness and some fine de-
tails were blurred. More effort was required to
define the colour threshold in the images from
the Kodak camera.

Test 2: focal length

Table 2 summarises the correlations between esti-
mates of CC obtained using the LAI-2000 and the
Canon camera set at different focal lengths. A focal
length of 50 mm was correlated with LAI-2000
estimates calculated using ring numbers 1–5, 1–4
and 1–3. The photographic method estimated a CC
of 0.77 when focal length was set at 50 mm, which
was higher than the value of 0.54 provided by the
LAI-2000 using ring numbers 1–3 (p< 0.01).
Similar estimates of CC were obtained for all three
focal lengths used with the photographic method
(Table 3). However, excluding different ring num-
bers from the calculations, and thus having differ-
ent vertical angles of view, yielded different
estimates of CC using the LAI-2000 (p<0.05).

Test 3: effect of time on estimates of CC

Over the 150-min time interval examined during
the morning, the estimates of CC obtained with
the LAI-2000 were not influenced by the time of
measurement. However, this was not the case for
estimates obtained using the Canon camera
because the regression over time was significant
(r
2
=0.58 and slope of 0.0003; Figure 2). By

Figure 1. Relationship between CC estimated by analysis of digital photographs taken at three sites in Mexico using two cameras with

zoom lenses set to 50 mm focal length.



contrast, no significant correlation was found
between estimates of CC obtained using the
photographic method and the LAI-2000. The
photographic estimate of CC of 0.73 was higher
than that of 0.56 obtained using the LAI-2000
(p<0.01). The photographic estimate of CC was
the same in Tests 2 and 3 when focal length was set
to 50 mm (0.77 vs. 0.73). The LAI-2000 also pro-
vided the same estimates of CC in Tests 2 and 3
(0.54 vs. 0.56).

Test 4: pruning treatments

The time course of CC differed between the
pruning treatments (Figure 3, p<0.05). The polled
standard error of the mean for the estimates of CC
(0.02) was identical for all pruning treatments (17,
27 and 77% of full sunlight) for the growing sea-
son.

Only in the 17% of full sunlight treatment was a
close relationship found for estimates of CC ob-
tained using the Mavica and Kodak cameras
(r
2
=0.73, p<0.05). When all three pruning treat-

ments were considered, a non-linear relationship
was obtained between estimates of CC obtained
using the Mavica (y) and Kodak (x) cameras (y=
0.86 � x0.03, r2=0.86, p<0.05). However, photo-
graphic estimates of CC differed between cameras
for the 27% of full sunlight treatment and were
more similar for the other pruning treatments
(Table 4, p<0.05). Also, the polled standard error
of the mean for the estimates of CC was similar for
both cameras (0.02). The bias between estimates of
CC obtained with the photographic method using
the Kodak and the Mavica cameras was �0.07
with a 95% confidence interval of ±0.018.

Discussion and conclusions

Camera comparison

Images obtained using the Kodak and Canon
cameras were equally useful in estimating CC at
the three sites examined. Considering the technical
differences between these cameras, it was expected
that the estimates of CC at Queretaro would be

Table 3. Estimates of CC within a subtropical mixed legume woodland obtained using different sensor rings of the Licor LAI-2000

canopy analyser and a Canon EOS D1 camera with the lens set at different focal lengths.

Vertical angle of view

LAI-2000 rings used
a

Focal length (mm)

1 1–2 1–3 1–4 1–5 28 35 50

7� 23� 38� 53� 68� 59� 49� 35�

CC 0.72 a
b

0.72 a 0.54 b 0.52 b 0.52 b 0.77 a 0.78 a 0.78 a

SE 0.007 0.007 0.017 0.020 0.018 0.011 0.010 0.007

a
Vertical angle of view correspond to the mean angle of the outermost ring.

b
Means with same letter were not significantly different (p<0.05).

Data are for to the Amazcala site, Mexico on 4 September 2002.

Table 2. Pearson correlation between estimates of CC within a subtropical mixed legume woodland obtained using different sensor

rings of the LAI-2000 canopy analyser and a Canon EOS D1 camera with the lens set at different focal lengths.

LAI-2000 rings used Vertical angle of view
b

Focal length (mm)
a

28 35 50

1 7� 0.58 0.45 0.54
1–2 23� 0.58 0.45 0.54
1–3 38� 0.65 0.69 0.82*
1–4 53� 0.77* 0.66 0.74*
1–5 68� 0.80* 0.76* 0.84*

a
Vertical angle of view was 59�, 49� and 35� for focal lengths of 28, 35 and 50 mm, respectively.

b
Vertical angle of view corresponds to the mid-angle of the outermost ring.

*p £ 0.05.
Data are for to the Amazcala site, Mexico on 4 September 2002.



statistically different. The data for the Queretaro
site shown in Figure 1 showed more scatter
because the canopy was the tallest at this site and
some small foliage objects were lost during image
processing of the Kodak images, as their chro-

matic values were very similar to those of the sky.
Images from the Canon camera registered a higher
canopy area because the definition of small bran-
ches and leaves at the top of the canopy was better
than in the images provided by the Kodak camera.

Figure 3. Development of CC for Alnus glutinosa at Aokautere, New Zealand, estimated by analysis of digital photographs captured

with a Kodak DC-120 camera. Vertical bars represent a 95% confidence intervals for the mean estimate of CC. Julian day 1 was on 1

January 1998.

Figure 2. Time trend of estimates of CC within a subtropical mixed legume woodland obtained during the morning (11:20 GMT) on

16 September 2002 at Amazcala, Mexico.



This also explained the lower regression coefficient
between these cameras at Queretaro (r

2
=0.71)

when compared to Humilpan and Amazcala
(r
2
=0.90 and 0.92, respectively), where the cano-

pies were much lower. Images from the Queretaro
site were processed again using the additive
threshold mode and the magic wand mask tool
(Corel Corp.) to manually select all foliage ele-
ments in the Kodak images and using the Canon
images as a reference. The resulting mean estimate
of canopy closure obtained using images from the
Kodak camera was 0.50. However, selecting foli-
age elements manually took 2 h per image,
whereas each image could be processed in under
3 min using the colour threshold file definition.

The Kodak camera consistently selected a
higher diaphragm aperture to compensate for the
low illumination environment, but this also
reduced the depth of field and increased the pos-
sibility of blurred foliage borders because some
objects were not in focus. Chen et al. (1991)
showed that variation in shutter speed for the same
image taken at the same time causes large varia-
tion in estimates of CC. Good definition of foliage
was also important at Amazcala because the can-
opy was denser and closer to the camera than at
the other sites; the shorter distance to the focused
object decreased the depth of field. The particu-
larly small and pinnate leaf form of the legumi-
nous tree species present at the Amazcala site
imposed a further requirement for clear definition
of the foliage border of the most distant canopy
stratum, which the Kodak camera could not define
adequately. The difference between estimates of
CC provided by the Kodak and Canon cameras
probably was smaller at Amazcala because the

problem of foliage definition was less pronounced
at this site (Table 1). However, the quality of
Canon images could be improved if the ISO speed
was set to its lower limit (200) rather than the 1600
ISO automatically selected. Although higher ISO
speeds are needed to take photographs under poor
light, they also increase the amount of noise in the
image.

Data from all tests indicate that the repeatability
of the photographic method was adequate for the
description of tree canopies. Agreement between
repeated measurements using the Kodak camera
was similar for all three pruning treatments in the
alder canopy in New Zealand (0.02 standard error
of the mean), and also similar to that observed at
the three sites in Mexico (0.015, 0.031 and 0.051;
Table 1). The consistent variability of the esti-
mates of CC obtained using the Kodak camera
indicates that the repeatability of the method was
unaffected by canopy development during the
growing season or differences in canopy structure;
this is clearly shown by the similarity of the 95%
confidence intervals for the estimates of CC for the
various pruning treatments (Figure 4). Repeat-
ability of the photographic method was also
unaffected by selection of camera or selected sites
because the standard error of the mean of esti-
mates of CC was similar in magnitude for the
Kodak, Canon and Mavica cameras in all tests
(0.01–0.03). Nonetheless, using the Canon would
be preferable to the Kodak camera because the
magnitude of the bias was smaller with respect to
the LAI-2000 (�0.05 vs. �0.07, respectively).
Similarly, using the Kodak camera would be
preferable to the Mavica because the bias between
the Kodak and Canon cameras was smaller than

Table 4. Canopy closure (CC) of pruned alder (Alnus glutinosa) estimated by analysis of digital images captured using Kodak DC-120

or Mavica FD-7 cameras. The pruning treatments correspond to levels of 17, 27 and 77% of full sunlight.

Julian day
a

CC SE
b

17% 27% 77%

Mavica Kodak Mavica Kodak Mavica Kodak

402 0.68 a
c

0.74 b 0.39 a 0.21 b 0.68 a 0.54 b 0.020

408 0.71 a 0.68 a 0.35 a 0.19 b 0.59 a 0.59 a 0.019

416 0.70 a 0.68 a 0.34 a 0.19 b 0.60 a 0.59 a 0.024

a
Julian day 1 was on 1 January 1998.

b
Pooled standard error.

c
Within pruning treatments, means in the same row were not significantly different (p<0.05).

Measurements were made during February 1999 at Aokautere, New Zealand.



between the Kodak and Mavica cameras (�0.02
vs. �0.07, respectively).

At Queretaro and Amazcala, where the foliage
was further from the camera or smaller, image
resolution was important. This problem was dis-
cussed by Yamamoto (2000), who examined can-
opy gap size, and concluded that higher resolution
images would estimate the gap size more precisely
than low resolution images. The use of higher
resolution also was advocated by Takenaka
(1987). Frazer et al. (2001) concluded that esti-
mates of CC derived from low image resolutions
(1024·768 or 640·480 pixels) combined with 1:4
file compression were lower than those obtained
from 1600·1200 uncompressed images provided
by a Nikon Coolpix 950 digital camera. Using the
same camera (Coolpix 950), Englund et al. (2000)
found that differences in digital image quality did
not affect estimates of CC. Nevertheless, the rela-
tionship between estimates of CC provided by the
Kodak and Canon cameras in the present study
(slope=0.89, r

2
=0.91) was close to that found by

Hale and Edwards (2002) for CC when comparing
hemispherical pictures from a film-based Nikon
FM2 and Nikon Coolpix 950 cameras over a range
of canopy densities (slope=0.905, r

2
=0.92). A

bias value of �0.003 to �0.04 could be added to
the CC estimates obtained using the Kodak cam-
era as correction factor with respect to the Canon
camera. This would be appropriate as the mea-
surement errors of both methods were compara-
ble, and also because the correlation between the
difference of the two estimates of CC (d) and their
average was small (r=0.2, Bland and Altman
(1986)).

Comparison between the LAI-2000 and cameras

Results from three field tests showed that the
digital photography and LAI-2000 approaches
provided estimates of CC with similar precision
because the variance within the results was similar.
Although the standard errors of the mean were
similar in Test 1 (Table 1), they were one order of
magnitude greater than in Tests 2 and 3 (Table 3
and Figure 2). This difference was attributed to the
sampling procedure used to obtain representative
ground points. Ground points used in Test 1 were
selected within each quartile of the median distri-
bution of canopy closure estimated with the

LAI-2000, whereas those used in Test 2 and 3 were
selected to be within one standard deviation of the
mean estimate of CC, with the result that variation
between the ground points was smaller.

The photographic and LAI-2000 methods pro-
vided similar estimates of CC except in Tests 2 and
3, when LAI-2000 rings 3, 4 and 5 were included in
the calculations. It is possible that the photo-
graphic and LAI-2000 methods spatially assess CC
differently. This possibility is supported by the
non-significant regressions between CC estimates
from either, the Canon or Kodak cameras, and the
LAI-2000. In fact, the high correlation at 68�
vertical angle of view between the LAI-2000 and
the camera was considered an artifact because the
camera was not able to capture light between 60�
vertical angle of view and the horizon, when focal
length was set to 28 mm (Table 2). Correlations of
estimates of CC also were low between hemi-
spherical photography and LAI-2000 methods in a
deeply shaded conifer-dominated forest or tropical
forest, even when the outermost zenithal rings
were disregarded (Machado and Reich 1999; Fer-
ment et al. 2001). Nonetheless, analyses based on
hemispherical photography and LAI-2000 mea-
surements revealed similar developmental trends
in CC and stand leaf area, despite significant
quantitative differences in their estimates (Frazer
et al. 1998; Paula and Lemos-Filho 2001). Also,
Peper and McPherson (2003) reported that esti-
mates based on digital analysis of photographs
obtained with a Kodak DC-50 camera and the
LAI-2000, showed good correlation with total leaf
area of isolated trees (r

2
>0.71).

The regression shown in Figure 2 suggests that
the sensor of the Canon EOS D1 was more
responsive to changes in illumination than the
sensors within the LAI-2000 and, hence, would
also require measurements to be made under
overcast conditions or low solar angles (Li-Cor
1992). Obtaining canopy images at other times of
day is not advisable, because foliage elements may
be sunlit and reflected light in images is difficult to
distinguish from the sky. Images containing lateral
chromatic aberration (halos) also have unpredict-
able effects on estimates of CC (Frazer et al. 2001).

It was concluded that digital photography is
suitable for characterising CC within agroforestry
systems, and that the estimates obtained exhibit
similar variability to those provided by the LAI-
2000 canopy analyser. Estimates of CC obtained



using the photographic method and a focal length
of 50 mm were comparable to those provided by
the LAI-2000 when considering rings 1 and 2 or 1–3
were used in the calculations. The reasons why the
estimates disagreed at lower angles of view require
further exploration. Estimates of canopy closure
obtained by analysis of images from different
cameras could be calibrated by including a cor-
rection factor within the calculation. However,
cameras with higher resolution should preferably
be used.

Acknowledgements

Financial support for this project was provided by
Consejo Nacional de Ciencia y Tecnologı́a under
grant SIHGO 19990206019. We also thank the
anonymous reviewers for suggestions to improve
this manuscript.

References

Ansley R.J., Price D.L., Dowhower S.L. and Carlson D.H.

1992. Seasonal trends in leaf area of honey mesquite trees:

determination using image analysis. J. Range Manage. 45:

339–344.

Bland J.M. and Altman D.G. 1986. Statistical methods for

assessing agreement between two methods of clinical mea-

surement. Lancet i: 307–310.

Broadhead J.S., Muxworthy A.R., Ong C.K. and Black C.R.

2003. Comparison of methods for determining leaf area in

tree rows. Agric. Forest Meteorol. 115: 151–161.

Chen J.M., Black T.A. and Adams R.S. 1991. Evaluation of

hemispherical photography for determining plant area index

and geometry of a forest stand. Agric. Forest Meteorol. 56:

129–43.

Devkota N.R., Kemp P.D., Valentine I. and Hodgson J. 2000.

Overview of shade tolerance of pasture species in relation to

deciduous tree, temperate silvopastoral systems. Agron. New

Zealand 30: 101–107.

Englund S.R., O’Brien J.J. and Clark D.B. 2000. Evaluation of

digital and film hemispherical photography and spherical

densitometry for measuring forest light environments. Can.

J. Forestry Res. 30: 1999–2005.

Ferment A., Picard N., Gourlet-Fleury S. and Baraloto C.

2001. A comparison of five indirect methods for character-

izing the light environment in a tropical forest. Ann. Forestry

Sci. 58: 877–891.

Frazer G., Lertzman K.P. and Trofymow J.A. 1998. Develop-

mental tends of canopy structure in coastal forests of British

Columbia. Northwest Sci. 72: 21–22.

Frazer G.W., Fournier R.A., Trofymow J.A. and Hall R.J.

2001. A comparison of digital and film fisheye photography

for analysis of forest canopy structure and gap light trans-

mission. Agric. Forest Meteorol. 109: 249–263.

Grace K.T. and Fownes H.J. 1998. Leaf area allometry and

evaluation of non-destructive estimates of total leaf area and

loss by browsing in a silvopastoral system. Agrofor. Syst. 40:

139–147.

Hale S.E. and Edwards C. 2002. Comparison of film and digital

hemispherical photography across a wide range of canopy

densities. Agric. Forest Meteorol. 112: 51–56.

Johnson L.F. and Pierce L.L. 2004. Indirect measurement of

leaf area index in California North Coast Vineyards. HortSci.

39: 236–238.

Knowles R.L., Horvath G.C., Carter M.A. and Hawke M.F.

1999. Developing a canopy closure model to predict over-

storey/understorey relationships in Pinus radiata silvopas-

toral systems. Agrofor. Syst. 43: 109–119.

Leblanc S.G. and Chen J.M. 2001. A practical scheme for

correcting multiple scattering effects on optical LAI mea-

surements. Agric. Forest Meteorol. 110: 125–139.

López-Serrano F.R., Landete-Castillejos T., Martı́nez-Millán J.

and Cerro-Barja A. 2000. LAI estimation of natural pine

forest using a non-standard sampling technique. Agric.

Forest Meteorol. 101: 95–111.

Li-Cor D. 1992. Plant Canopy Analyser Operating Manual.

Li-Cor Inc, Lincoln, NE, USA.

Macfarlane C., Coote M., White D.A. and Adams D.A. 2000.

Photographic exposure affects indirect estimation of leaf area

in plantations of Eucalyptus globulus Labill. Agric. Forest

Meteorol. 100: 155–168.

Machado J.L. and Reich P.B. 1999. Evaluation of several

measures of canopy openness as predictors of photosyn-

thetic photon flux density in deeply shaded conifer-domi-

nated forest understory. Can. J. Forestry Res. 29: 1438–

1444.

Nackaerts K., Coppin P., Muys B. and Hermy M. 2000.

Sampling methodology for LAI measurements with LAI-

2000 in small forest stands. Agric. Forest Meteorol. 101:

247–250.

Paula S.A. and Lemos-Filho J.P. 2001. Dynamics of canopy

in a semideciduous forest at the urban perimeter of Belo

Horizonte, MG. Rev. Brasileira de Bot. 24: 545–551.

Peper P.J. and McPherson E.G. 2003. Evaluation of four

methods for estimating leaf area of isolated trees. Urban

Forestry Urban Green. 2: 19–29.

Takenaka A. 1987. Analysis of light transmissivity of forest

canopies with a telephoto method. Agric. Forest Meteorol.

40: 359–369.

Welles J.M. and Cohen S. 1996. Canopy structure measurement

by gap fraction analysis using commercial instrumentation.

J. Exper. Bot. 47: 1335–1342.

Welles J.M. and Norman J.M. 1991. Instrument for indirect

measurement of canopy architecture. Agron. J. 83: 818–

825.

Yamamoto K. 2000. Estimation of the canopy gap size using

two photographs taken at different heights. Ecol. Res. 15:

203–208.