Estimating tree crown dimensions using digital analysis of vertical photographs

johnnepaleseΗλεκτρονική - Συσκευές

10 Οκτ 2013 (πριν από 4 χρόνια και 29 μέρες)

240 εμφανίσεις

Agricultural and Forest Meteorology 100 (2000) 199212
Estimating tree crown dimensions using digital analysis
of vertical photographs
P.L.Brown
a;b;
,D.Doley
a;b
,R.J.Keenan
a;c
a
Cooperative Research Centre for Tropical Rainforest Ecology and Management,Cairns,Queensland 4870,Australia
b
Department of Botany,The University of Queensland,St Lucia,Queensland 4072,Australia
c
Bureau of Rural Sciences,P.O.Box E11,Kingston,ACT 2604,Australia
Received 10 May 1999;received in revised form30 August 1999;accepted 17 September 1999
Abstract
The aimof this study was to develop a relatively rapid,simple and repeatable method to characterize the crown architecture
of Queensland maple (Flindersia brayleyana F.Muell.) by vertical hemispherical photography.Calibration scales were
developed to provide horizontal planes of measurement at heights from15 to 40 mabove the camera.Hemispherical images
of tree crowns were merged with the specic calibration scale appropriate for the mid-height of the crown in order to adjust
for image distortion during measurement.Merged images were analyzed by three procedures to yield measurements of crown
diameter and projected area.The most precise measurement of projected crown dimensions was assumed to be provided by
a complete analysis of a digital image of the crown,after removal of the trunk image.Crown diameters estimated by tape
measurements on the ground were relatively imprecise ( R
2
D0.54).Crowns classied as symmetrical were described much
more precisely (R
2
D0.72) than those classied as asymmetrical ( R
2
D0.37).The precision of estimates of crown dimensions
by tape did not alter signicantly over the range of crown sizes sampled (210 mdiameter in trees 1540 mtall).This novel
image analysis method can yield relatively precise and accurate measurements on the relatively shallow crowns of tall trees,
often found in rainforests,regardless of the degree of asymmetry in crown shapes.However,the method is slower to apply than
conventional tape techniques and additional equipment is required.The principal advantage of the photographic technique is
the opportunity for precise measurements of changes in crown size and foliage projected area over time.©2000 Published by
Elsevier Science B.V.All rights reserved.
Keywords:Tree crowns;Crown asymmetry;Crown diameter;Crown projected area;Hemispherical photography;Radial distortion
1.Introduction
Vertical photographs of forest canopies provide
detailed information about radiation regimes within
forests,which can be used to predict individual tree
and stand growth (Koop,1989).In tropical rainforests,

Corresponding author.Tel.:C61-7-3365-2768;
fax:C61-7-3365-1699.
E-mail address:p.brown@botany.uq.edu.au (P.L.Brown).
where the availability of light is most limiting to the
growth of individual trees below the main canopy,
applications of radiation measurements may prove
crucial for improving existing tree growth prediction
models (Landsberg and Gower,1997).Vertical pho-
tographs currently contain some of the most precise
and detailed information on forest radiation regimes
but their utility has been restricted largely by the
difculty of making accurate repeatable estimates.
Application of radiation-based tree growth models
0168-1923/00/$  see front matter ©2000 Published by Elsevier Science B.V.All rights reserved.
PII:S0168- 1923( 99) 00138- 0
200 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
to individual trees requires improvements in the de-
scription of radiation interactions by their crowns in
structurally complex forests.These interactions need
to be estimated with considerable precision,and im-
portant small-scale changes in crown dimensions are
difcult to estimate in any tall or complex forests,
such as those of the wet tropics.
There is increasing global interest in expanding
the range of species used in timber plantations,as
a substitute for native forest utilization.This inter-
est is particularly strong in the wet tropics,where
many broad-leaved species are being proposed for
plantation management (Evans,1982;Rowe et al.,
1991;Ferris-Kaan,1992).The successful adoption
of broad-leaved species,many with irregular crown
shapes at a relatively early age,depends largely on the
denition of their optimumsilvicultural requirements.
An understanding of a species crown architecture is
fundamental to designing silvicultural prescriptions
(Dawkins,1963;Suri,1975;Samarasinghe,1995;
Dhanesh Kumar et al.,1997),and this understanding
is usually built upon accurate measurements of plant
dimensions which reect the photosynthetic potential
of individual trees.
Plant dimensions such as crown projected area
and diameter are often used as surrogates for crown
photosynthetic potential and foliage area (Jahnke and
Lawrence,1965),but they are usually derived from
relatively imprecise measurements taken from the
ground (Philip,1994).Consequently,there is a need
for better techniques for measuring these attributes.
The measurement of crown dimensions can be par-
ticularly difcult in rainforest broad-leaved species,
because of irregular crown shapes.Light environments
of tropical rainforests are often more heterogeneous
than those of temperate forests (Whitmore and Burn-
ham,1984) and the development of asymmetric crown
shapes has been postulated as a plastic response to
this heterogeneity (Umeki,1995).Most conventional
methods for measuring crown diameter and projected
area assume that crown projections are either a solid
circle or solid ellipsoid shape (Wilkinson,1995;Big-
ing and Gill,1997).Therefore,such methods vary
in precision according to the degree crowns depart
fromsimple geometric shapes (Biging and Gill,1997;
Zeide,1998).
Airborne and satellite remote sensing techniques,
that measure specic forest attributes,are rapidly im-
proving in their discriminative capacity.For example,
surface Lidar remote sensing systems now have the
capacity to rapidly obtain information at the stand
level on canopy height,total biomass,leaf biomass,
basal area and percentage cover (Means et al.,1999).
Lidar systems can operate at a nominal footprint
diameter of 10 m yet currently lack the capacity to
allow measurements of the dimensions of individ-
ual tree crowns (Lefsky et al.,1999).Stereo pairs
of co-registered aerial photographs have been used
effectively to monitor changes in co-dominant and
emergent tropical rainforest crown dimensions over
the longer term (18 years) (Herwitz et al.,1998a).
However,even when using photographs taken at
large-scales from low ying aircraft,such techniques
currently lack the discrimination needed to accurately
monitor smaller changes in tree crown dimensions in
complex forest systems,particularly on individuals
with irregular crown shapes or those growing in sub-
dominant positions (Herwitz et al.,1998b).Vertical
photography has also been used to determine tree
crown dimensions.Koop (1989) used stereo-pairs
of vertical photographs taken a short distance apart
to convert small image displacements between the
photographs into distance and crown area estimates.
Acceptable precision was achieved even on smaller
crown sizes,but at considerable cost per tree.Koike
(1985) reconstructed two-dimensional canopy pro-
les from clusters of vertical photographs processed
by computed tomography.This procedure is very de-
manding of resources particularly when the distance
between the camera and subject varies substantially.
Chen et al.(1991) measured plant area index using
both the analysis of hemispherical photographs and
a Plant Canopy Analyzer (Li-Cor LAI-200) on the
same crowns.They found that both methods were in
agreement when the hemispherical photographs were
underexposed four to ve stops below that recom-
mended by an upward facing light meter.Although,
the study by Chen et al.(1991) renes and validates
the use of hemispherical photos to measure plant area
index the need remains for a relatively quick,sim-
ple and accurate method of estimating small crown
diameters on tall trees.
An example of a tropical forest species which fre-
quently has small,irregular shaped crowns on tall trees
and for which crown architectural descriptions are
needed is Queensland maple (Flindersia brayleyana F.
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 201
Muell.).This species is highly prized as a cabinet tim-
ber and is endemic to the wet tropics region of eastern
Australia,where its supply from native forests has all
but ceased (Cameron and Jermyn,1991).Bazzaz and
Pickett (1980) suggest maple is a middle-order suc-
cession species and evidence is increasing that it can
survive in a wide range of light conditions (Thompson
Fig.1.Images of crown projected areas typifying the asymmetrical group (six above) and symmetrical group (six below).A circle of
known dimensions at a known height encompasses each crown however the relative scale of these circles has been altered for individual
crowns to a standard diameter for the diagram.
et al.,1988;Swanborough et al.,1998).With weak
apical control and highly plastic growth,maple tends
to develop asymmetric crowns that can be difcult to
describe adequately by simple geometric shapes (see
Fig.1).This irregularity may be due to crown shyness
(Jacobs,1955;Ng,1976),with crown development
being strongly inuenced by surrounding crowns.
202 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
The aim of this study was to develop a simple,re-
peatable method for measuring crown diameter and
projected area using vertical photographs.This aim
evolved out of the need to characterize the crown ar-
chitecture of Queensland maple,for we were partic-
ularly interested in obtaining accurate measurements
on crowns with irregular or asymmetrical shapes.
2.Methodology
The photographic method of crown measurement
was developed by calibrating images taken of objects
with known dimensions at known distances from the
camera.The method was then applied in a 66-year-old
plantation of Queensland maple.This plantation is
at Gadgarra State Forest,17

17
0
S 145

43
0
E,and
640 m above sea level on the Atherton Tablelands in
north Queensland.Measurements using the photo-
graphic method were compared with a conventional
ground-based method using a tape (Philip,1994).
Both methods were applied to the same maple crowns.
Crown diameters were measured on 147 maple trees
in an unthinned,66-year-old plantation.The measured
stand was then thinned in order to release the best
potential crop trees.This thinning operation isolated
and facilitated the denition of the crowns used in this
study.Also,all the understorey plants were removed
in order to obtain a clear view of the canopy structure.
Estimation of projected crown dimensions by tape
involved two measurements of crown diameter,on
NorthSouth and EastWest axes as located by com-
pass (Philip,1994).Most crowns were asymmetrical
to varying degrees,so measurements were made with-
out reference to the position of the stem.Crown ex-
tremities were located by looking vertically from the
ground and crown diameter was calculated as the mean
of the two axis lengths.
The photographic method involved:
1.Calibration of image estimates using standards;
2.Rening methods to capture crown images;
3.Processing images using image analysis software;
4.Calculating crown dimensions fromprocessed im-
ages in`Hemiphot'(ter Steege,1994),a software
package designed for hemispherical image analy-
sis.
A Canon hemispherical lens (7.5 mm) was used
throughout this study.Its use was not driven by an
inherent superiority in its ability to record the visible
extent of crowns for all spherical lenses create radial
distortions,particularly in the periphery of the eld
and these limit the accurate estimation of area in im-
age components.The best lens for a given task is the
longest focal length that can capture the entire dimen-
sions of the largest size crown(s) under investigation,
as this minimizes radial distortions.Seldom is a sin-
gle crown so large that it requires a hemispherical lens
for complete capture.However,in studies that require
additional information on both a target crown and its
spatial relationships with its nearest neighbors,a hemi-
spherical lens with its large eld of view may have
advantages.A hemispherical lens was chosen for this
study because hemispherical images were required for
stand canopy analysis in the`Hemiphot',image anal-
ysis package (ter Steege,1994).
2.1.Calibration of image estimates using standards
To enable horizontal projected area calculations to
be derived from a spherical image,allowance had to
be made for the effects of both perspective and radial
distortion.Perspective affects all images regardless of
the lens focal length,but radial distortion occurs only
with a spherical lens.The effects of radial distortion
are largest near the lens boundary and least near the
lens axis and they increase as the focal length of the
lens decreases.Calibrations for radial distortion have
been noted for a variety of hemispherical lenses (Her-
bert,1987;Clark and Follin,1988),and the process
should be completed for each lens.
Templates for the measurement of distance on pho-
tographic images were made from a set of six calibra-
tion images specic for a Canon 7.5 mm lens.To pre-
pare these calibration images nine pegs were placed
2 m apart along a straight line in a relatively at and
open eld.The centre peg was used as the beginning
of a line perpendicular to the rst line,along which
six positions were marked off at 5 mintervals between
15 and 40 m.
Six exposures were taken of the row of pegs,one at
each 5 minterval between the 15 and 40 mpoint along
the perpendicular line.The lens was aligned so that the
top of the centre peg was at the image centre.Resulting
negatives were scanned at 50 50 screen pixel scale
(resolution 2000 dpi (dots per inch)) then imported
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 203
Fig.2.Image manipulation and the effect of thinning on crown
visibility.Image (a):Full hemispherical view taken prior to thin-
ning.Image (b):50 50 pixel subset of the centre of image (a)
(note the difculty in dening the full extent of individual crowns).
Image (c):Same location as image (a) taken after thinning.Image
(d):50 50 pixel subset of the centre of image (c).Image (e):
Merging the area-dening template on image (d).Image (f):Final
merged image,ready for cover calculation in Hemiphot.
into the Adobe Illustrator drawing package (Adobe,
1997).Four concentric rings were drawn around the
centre peg giving the appearance of a shooting target
or`bulls-eye'(see Fig.2e).
The limits to calibration dimensions were estab-
lished from those encountered in the measured plan-
tation,where tree heights ranged from 15 to 40 m and
crown diameters up to 10 m.The rowof pegs in images
represented a horizontal plane in the forest canopy and
the photo point the actual lens position near the forest
oor.Due to the effects of radial distortion,the xed
distances between pegs appeared in images to be sym-
metrically and progressively reduced away from the
centre peg.Due to perspective,the further the line of
pegs was from the camera,the smaller the resulting
sets of rings in templates.
2.2.Capturing crown images
Throughout this study 400 ASA,Ilford,HP5 lm
was used in a Canon T60 body with the Canon
7.5 mm sh-eye lens (1:5.6).Procedures for tak-
ing vertical photographs in forest environments are
well documented (Becker et al.,1989;Barrie et
al.,1990;Chen et al.,1991;Schaffer et al.,1991)
(www.gemlab.ukans.edu/hp/).Photographs of the
crowns were taken at a time when the contrast be-
tween sky and vegetation components was maximal.
This required even illumination in the sky background
and no direct sunlight being reected by target vegeta-
tion.These conditions usually occurred at sunrise and
sunset or on days with a dense and even cloud cover.
Windy periods were avoided because of crown sway.
Consistency in the speed of exposure relative to the
prevailing light conditions is important.Chen et al.
(1991) showed that variation in shutter speed for the
same image taken at the same time causes large vari-
ation in canopy cover estimates.In this study,prior
to each exposure,a light reading was taken with a
Minolta Auto Meter IV F.After some testing,it was
found the best contrast was achieved by increasing
the shutter speed (underexposing) three conventional
levels above that recommended by the meter.Shut-
ter speeds calculated this way are not independent of
the site cover (Chen et al.,1991).However,indepen-
dent readings were not available due to an absence of
nearby canopy openings of sufcient size.
2.3.Selection of photo-points in the forest
Ground points for photographs were selected using
the following three criteria:
1.Photo-points must be vertically below the target
crown;
2.Photo-points allowed the isolation of the target
crowns area from its neighbors;
204 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
3.The number of photo-points for each crown was
a function of its size (horizontal extent).Some
large crowns were photographed fromfour points,
while the smallest crowns were photographed
from one point.
The lens was orientated as close as possible to ver-
tical using a tripod with a geared head and nely cal-
ibrated bubble level on the camera body.The recent
innovation of a self-leveling mount (Oberbauer et al.,
1993;Rich et al.,1993) (marketed by Delta T Devices,
Cambridge UK) would make this process faster.
2.4.Image processing and analysis
Exposed negatives were scanned twice using a
Kodak Professional RFS 2035 Film Scanner.Critical
factors in scanning include resolution,contrast and
brightness.The rst image was scaled to take in the
entire hemisphere at a resolution of 1000 dpi and
used to calculate the image centre.The second image,
which took in a 50 50 pixel (screen resolution) sub-
set of the image centre,focused on the target crown
and its nearest neighbors.It contained only those
components in the inner 30

solid angle surround-
ing the lens axis and was digitized at a resolution of
2000 dpi (Fig.2).The same screen scale and resolu-
tion was used throughout all scanning procedures and
a contrast value of 20,the maximum for the Kodak
Film Scanner,was used throughout the study.
Brightness settings varied between 5 and 10,de-
pending on the visibility of texture within vegeta-
tion components.This variation in brightness settings
caused less than 2% variation in cover readings.The
texture of vegetation was an unnecessary detail in
this study that could confound results.Therefore,gray
tones representing vegetation were kept as close to
black as possible.
All images (including calibration templates) were
saved in a PCX format and converted to a 400 400
screen pixel scale in`Adobe Photoshop'(Adobe Sys-
tems.,1996).This was the only common format and
maximum size acceptable in`Hemiphot'(ter Steege,
1994).Using the above settings in a more recent model
of scanner,the Nikon Coolscan II LS-20E,images
were obtained with a crisper appearance,but with sim-
ilar cover readings (<1% variation).
When using a spherical lens,knowledge of the
proximity of crown components to the centre of the
image or lens axis is necessary to determine the accu-
racy of area measurements.Heavy shade from dense
vegetation on the image periphery made it impossible
to clearly dene the edge of the full hemispherical
images of crowns.Inverting the image (sky-black,
vegetation-white) improved edge resolution,but it
was still not always possible to clearly dene the min-
imum of three boundary points,necessary to locate
the image centre.
The image boundary is constant and set by the lens
focal length.Acircle at this specic diameter was cre-
ated and used as a template to t as closely as possi-
ble on the diffuse boundary of the full hemispherical
images of the crown.The inverted crown image was
overlain with a circle template in Adobe Photoshop
(Adobe Systems.,1996).
2.5.Manipulation of crown images
Digitized crown images were recorded in 256
gray-tones.Converting all sky components to white
and all vegetation components to black increased the
accuracy of nal cover estimates in Hemiphot by
sharpening sky-vegetation boundaries.The Adobe
Photoshop program(Adobe Systems.,1996) was used
to perform black and white conversions at a variable
threshold level.
A histogram was created within Adobe Photoshop
(Adobe Systems.,1996),representing the pixel dis-
tribution of all 256 gray-tones,from light to dark,
within the image.Within this histogram the distribu-
tion of pixels representing vegetation components and
sky could be separated and the accuracy of the photo-
graphic methodology in maximizing contrast between
sky and vegetation could be determined.
2.6.Isolation of a target crown
Each image targeted a specic crown and this crown
lay at or close to the image centre.So that cover val-
ues represented only the crown of interest in each im-
age target crowns were isolated from their neighbors,
by erasing all neighboring crowns and trunks from
images in Adobe Photoshop (Adobe Systems.,1996).
The trunk connected to the target crown was erased
up to the point where it rst overlapped the foliage
of the target crown.Using this point of rst foliage
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 205
overlap,the trunk was erased perpendicular to its axial
direction.
2.7.Merging crown images with an calibration
templates
The height of the calibration template (distance
from photo point to the perpendicular line of pegs)
corresponded to the height of the crown mid-point,or
the mean of live crown base and top height (Philip,
1994).The accuracy of crown projected area estimates
is directly proportional to the total tree height/crown
depth ratio.The larger the distance between the mea-
surement plane and the actual locations of maximum
horizontal extension the greater the error resulting
frominadequate calibration for perspective in images.
For example,conifer stands with deep narrow crowns
are unsuited to its application,due to the high proba-
bility that the applied plane of horizontal measurement
would be a relatively large vertical distance from the
actual locations of maximum horizontal extension.
Acceptable values of the tree height/crown depth
ratio exceed 2.5 and all trees in the present study sat-
ised this criterion.
The modied image of a single,isolated and
trunk-less crown was merged with the appropriate
calibration template in Adobe Photoshop (Adobe
Systems.,1996).In these merged images,the crown
image formed the bottom layer and the appropriate
calibration template the upper layer.Before merg-
ing,the calibration template was reduced to a 50%
transparency (using the opaque tool) so that both
the target crown and the templates concentric circles
were visible in the overlaid image (Fig.2).Circles
and sections of the upper layer overlapping the crown
were removed using the eraser tool resulting in im-
ages where crowns were enclosed by a single circle.
These layers were merged and saved in PCX format
for analysis in Hemiphot (Fig.2).
Two methods were tested for positioning the cali-
bration template on the modied crown image:
1.Centering the calibration template on the centre
of the photographic image (theoretically the most
accurate placement);
2.Centering the area template on the approximate
centre of the target crown (without reference to
the image centre).
Photographic points on the ground were selected
to minimize the linear difference between the cen-
tre of the target crown and centre of the image.
Centering the calibration template on the crown
saved considerable processing time and caused no
more than 3% difference from the estimates de-
rived from photographic-centered templates.This
acceptable reduction in accuracy was attributed to
the fact that all target crowns were located within
20

of the lens axis,a zone of limited radial
distortion.
2.8.Hemiphot analysis
Each concentric ring of the calibration template rep-
resented an increment in radius of 2 m from its cen-
tre.Given the limited depth of the tree crowns (usu-
ally less than 5 m) and their height above the lens (be-
tween 15 and 25 m),the height of each crown mid-
point generally corresponded with the height of its
lateral extremities.As a result the circular calibration
template tted to each target crown depicted an imag-
inary yet plausible horizontal surface corresponding
to the height of the target crown's maximum lateral
extension.
Percentage crown cover was calculated with a cir-
cular calculation template (annex) in the Hemiphot
program (ter Steege,1994).The user sets both the
size and position on the image of this circular calcula-
tion template.The merged crown image was imported
into Hemiphot and the circular calculation template
was overlaid onto the merged image circle (of known
diameter),which encompassed the target crown.The
percentage cover of the target crown within this
merged image circle could then be derived and con-
verted into an estimate of crown projected area.Final
estimates of crown projected area were made from
one to four,but mostly two different images of the
target crown.Crown diameter was calculated by the
diameter of a circle equivalent of the crown-projected
area.
Given that an absolute measurement of crown diam-
eter was not possible for the trees under investigation,
the above Hemiphot-based procedure was considered
to be the most precise estimate and was used as a stan-
dard against which a number of simpler modications
for this photographic technique could be compared.
206 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
2.9.Linear image estimates
A rapid estimate of crown size on photographic
images can be obtained by measuring linear dimen-
sions at xed orientations.All images were orientated
with the north point at the top and the crown di-
ameters on the cardinal axes were measured using
the pixel-locating tool in Adobe Illustrator (Adobe,
1997).Linear pixel measurements were converted
into distance by calibration with the pixel diameter of
the encompassing circle,which was known for each
tree.This technique was applied in the following two
ways:
1.A rigid rule where the cut-off point depicting the
crown's extremities in each cardinal direction was
at the point of greatest lateral extension,no matter
how small or isolated the associated cluster of
leaves may be;
2.A exible rule whereby account was taken of
small and isolated lateral extremities and a mean
crown extremity was estimated by eye.
It was observed that most crowns were made up
of clusters of discrete units (Fig.1) and as a re-
sult it was expected that the rigid approach,that
took no account of gaps in cover within the crowns
area,would result in overestimates of crown pro-
jected area.The second approach was developed as a
rapid,subjective method for increasing the accuracy
of these linear image estimates.The outcomes of
both of these approaches have been compared in the
results.
To further analyze the differences between results
from the tape and photographic techniques,crowns
were divided into two groups according to their degree
of asymmetry,viz.:
1.Symmetric crowns were made up of one-two dis-
crete units and were often lightly skewed to the
cardinal axes (69% of the total sample).A single
or two unit crown was determined as symmetric if
a superimposed circle that encompassed the entire
crown with the minimumdiameter,had more than
70% vegetation cover within it.In encompassing
circles with less than 70% cover,the cover distri-
bution must be even throughout all quadrants.The
total amount of cover contained in each quadrant
was estimated and if any two of these four values
varied more than 50% the crown was deemed as
asymmetric;
2.Asymmetric crowns with either three or more,
discrete units or one to two heavily skewed or
highly irregular crown units according to the
above rule.A crown unit was considered discrete
when the distance between adjacent clusters was
greater than the diameter of clusters.Crowns
deemed as asymmetric made up 31% of the total
sample.
This description of crown symmetry differs some-
what fromUmeki (1995) who places more importance
on the spatial relationship between crown and trunk.
Tape and linear image estimates were compared
with Hemiphot estimates through t-tests that utilized
regression coefcients (two sample t-tests) and a
relative variable derived from tape or rule method es-
timate/Hemiphot estimate (one sample t-tests).T-tests
on regression coefcients were calculated with sig-
nicance levels adjusted according to Bonferoni's
rule.
3.Results and discussion
There was a signicant but not close relationship
between the estimates of crown diameter obtained
by tape in the eld and image estimates obtained by
Hemiphot analysis (R
2
D0.54) (Fig.3).Both the xed
and exible image rule methods had signicant and
close relationships with Hemiphot,with R
2
values of
0.9 and 0.88,respectively (see Fig.4,and Table 1).
The tape method gave signicantly lower estimates of
crown diameter than the Hemiphot method,with 78%
of the tape estimates being lower than Hemiphot esti-
mates.The xed rule method gave signicantly larger
crown diameters than the Hemiphot method (99.3%of
xed rule estimates >Hemiphot estimates).Whereas
the exible rule method resulted in estimates with no
signicant difference in diameter from the Hemiphot
estimates.However,the estimates derived from the
exible rule method may not be consistently repeat-
able due to the high degree of subjectivity implicit in
this rules application.
A comparison of tape with Hemiphot estimates re-
vealed a stronger relationship for symmetric crowns
(R
2
D0.722) than asymmetric crowns (R
2
D0.37).
Therefore,the largest differences between crown di-
ameters derived from Hemiphot analysis and tape
occurred in the most asymmetric or multiple-clumped
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 207
Fig.3.The relationship between tape,xed rule and Hemiphot methods using absolute estimates on both axes.Solid circles and the
broken lines represent the xed rule crown estimates and regression.Open triangles and the solid line represent the tape crown estimates
and regression.There is a signicant difference between these two regression slopes.
crowns and the smallest differences between the two
techniques occurred in the most symmetric crowns
(Fig.5 and Table 1).The accentuation of differences
between the image and tape techniques,as crowns
become more asymmetric,is probably a result of
Fig.4.The relationship between exible rule,xed rule and Hemiphot methods using absolute estimate values on both axes.Solid circles
and the broken lines represent the xed rule crown estimates and regression.Open triangles and the solid line represent the exible rule
crown estimates and regression.There is a signicant difference between these two regression slopes.
the greater precision achievable with the Hemiphot
technique.This increased precision is attributed to
the inherent ability of the Hemiphot based method to
account for the nuances of irregular crown shapes,
unlike xed rule and tape techniques.
208 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
Table 1
Correlation coefcients with slope and error parameters for linear regressions on relationship between all methods investigated
a
Independent Dependant A B R
2
Similarity
Hemiphot tape −0.386 0.367 0.927 0.071 0.54 a
Hemiphot xed rule −0.295 0.187 1.270 0.036 0.90 b
Hemiphot exible rule 0.037 0.146 0.942  0.028 0.88 a
Fixed rule tape −0.021 0.362 0.705 0.052 0.56 c
Flexible rule tape −0.145 0.350 0.925 0.071 0.54 d
Hemiphot (symmetric)
b
tape −0.776 0.320 1.054 0.064 0.72 a
Hemiphot (asymmetric)
b
tape −0.404 0.964 0.8828 0.171 0.37 a
a
The linear relationships between methods which have the same letter in the similarity column are not signicantly different ( PD0.05).
b
Symmetric and asymmetric categories were developed by dividing all crowns according to their degree of asymmetry (see text in
Section 2).
The differences between tape and Hemiphot es-
timates for the same crown did not vary signi-
cantly over the range of crown sizes measured (Fig.
6).This consistent pattern of error between the
tape and Hemiphot techniques occurred within both
the asymmetrical and symmetrical crown groups
(Fig.7).Differences between estimates from xed
rule,exible rule and Hemiphot techniques did not
vary signicantly over the range of crown sizes
(Fig.8).
Using the Hemiphot based method as the standard
against which to compare other methods,the most
accurate and precise technique for measuring crown
Fig.5.The relationship between tape and Hemiphot methods for crowns in the asymmetrical and symmetrical categories using absolute
estimate values on both axes.Open circles and the solid line represent the symmetrical group and regression.Closed diamonds and the
broken line represent the asymmetrical category and regression.There is no signicant difference between these two regression slopes.
diameters in this trial was found to be the exible
rule method followed by the xed rule method.The
tape method,though more precise when applied to
symmetric than asymmetric crowns was overall the
least precise and accurate method for measuring crown
diameter in this study (Table 1).
In the present study,estimating crown dimen-
sions by the Hemiphot method took approximately
three times as long as conventional tape methods.
The estimation of crown dimensions by the xed
and exible rule methods took approximately twice
as long as the tape method.It would be difcult to
justify the use of these new image techniques over
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 209
Fig.6.The pattern of differences between the tape,xed rule and Hemiphot methods.The solid circles and broken line represents a
relative measure and regression for the difference between the xed rule and Hemiphot techniques where YDxed rule/Hemiphot estimate.
Therefore,Y values of 1 indicate identical measures by both techniques.The open triangles and solid line represents a relative measure
and regression on the difference between the tape and Hemiphot techniques where YDtape/Hemiphot estimate.Both regressions are not
signicantly different from zero.
Fig.7.The pattern of difference between the tape and Hemiphot methods in the asymmetrical and symmetrical crown shape categories.
Open circles and the solid line represents a relative measure and regression on the difference between the tape and Hemiphot techniques
in the symmetrical category,where YDtape/Hemiphot estimate.The solid diamond and broken line represents the same relative measure
and regression for the asymmetrical category.Both regressions are not signicantly different from zero.
210 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
Fig.8.The pattern of differences between the xed rule,exible rule and Hemiphot methods.Open triangles and solid line represents
a relative measure and regression on the difference between the exible rule and Hemiphot techniques where YDexible rule/Hemiphot
estimate.Therefore,Y values of 1 indicate identical measures by both techniques.The solid circles and broken line represents a relative
measure and regression for the difference between the xed rule and Hemiphot techniques,where YDxed rule/Hemiphot estimate.Both
regressions are not signicantly different from zero.
conventional tape methods,when measurements of
symmetrical crowns are required once only,partic-
ularly where crown diameter exceeds one third of
tree height.However,both the exible rule and in
particular the Hemiphot method give the possibility
of accurate,repeated measurements that reveal small
changes in crown and canopy gap dimensions over
time.For example,when using the Hemiphot method,
the presence or absence of vegetation within one
pixel can be recorded (which may represent a 16 cm
2
area at a subject height of 15 m).Most conventional
techniques for calculating crown dimensions are less
sensitive for quantifying small changes in these di-
mensions,including the accurate,non-meteorological
approach involving the calculation of species spe-
cic allometric relationships to sapwood basal area
at breast height (Waring,1983).The Hemiphot based
method could have distinct advantages over conven-
tional techniques in a wide range of studies including
monitoring:
 Crown response to thinning;
 Crown interactions with competing or coexisting
neighbors;
 Crown condition and forest health;
 The effects of defoliation by insects or storms and
recovery following these events;
 Crown thickening or inll as well as crown expan-
sion,in response to a range of silvicultural treat-
ments.
The capture of forest radiation regimes with pho-
tographs is completely non-destructive to the target
canopy.If the processing of crown and canopy images
can be simplied to the point where it is carried out
with conventional photographic equipment,digitizers
and personal computers it may become a common-
place forest measurement tool.Such a tool could be
most relevant to silvicultural research in developing
tropical nations where research resources are scarce
and the need for detailed information on individual
trees and light interactions in species rich forests and
plantations of broad-leaved species is in immediate
demand.
P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212 211
Acknowledgements
Financial support from the Cooperative Research
Centre for Tropical Rainforest Ecology and Manage-
ment in Cairns,North Queensland.Support from the
Forestry Ofce in Atherton,part of the Queensland
Department of Primary Industries (DPI),in particu-
lar help was given from DPI staff:Mark Annandale,
Arthur Wright,Mila Bristow and Kerry Hanrahan.
References
Adobe Systems.,1996.Adobe photoshop:photo design and
production tool.Adobe Systems Incorporated,San Jose.
Adobe,Systems.,1997.Adobe illustrator:illustration tool.Adobe
Systems Incorporated,San Jose.
Barrie,J.,Greatorex-Davies,J.N.,Parsell,R.J.,Marrs,R.H.,
1990.A semi-automated method for analysing hemispherical
photographs for the assessment of woodland shade.Biol.
Conserv.54,327334.
Bazzaz,F.A.,Pickett,S.T.A.,1980.Physiological ecology of
tropical succession.Ann.Rev.Ecol.Systems 11,287310.
Becker,P.,Erhart,D.W.,Smith,A.P.,1989.Analysis of forest
light environments part I.Computerized estimation of solar
radiation from hemispherical canopy photographs.J.Agric.For.
Meteorol.44,217232.
Biging,G.S.,Gill,S.J.,1997.Stochastic models for conifer tree
crown proles.For.Sci.43 (1),2534.
Cameron,D.M.,Jermyn,D.,1991.Review of plantation
performance of high value rainforest species.CSIRO and
Queensland Forestry Service.
Chen,J.M.,Black,T.A.,Adams,R.S.,1991.Evaluation of
hemispherical photography for determining plant area index and
geometry of a forest stand.Agric.For.Meteorol.56,129143.
Clark,J.A.,Follin,G.M.,1988.A simple equal area calibration
for sheye photography.Agric.For.Meteorol.44,1925.
Dawkins,H.C.,1963.Crown diameters:their relation to bole
diameter in tropical forest trees.Commonwealth For.Rev.42,
318333.
Dhanesh Kumar,P.,Rajesh,N.,Santhosh Kumar,K.,Vidyasagar,
K.,Anaz,M.A.,1997.Crown diameter/bole diameter
relationship as an aid to thinning in teak ( Tectona grandis L.).
Ind.J.For.20 (4),355361.
Evans,J.,1982.Plantation Forestry in the Tropics.Claredon Press,
Oxford,New York,p.472.
Ferris-Kaan,R.,1992.Forests and biodiversit y  a brief review
of the issues.In:Ferris-Kaan,R.(Ed.),Managing Forests for
Biodiversity.Forestry Commission,Edinburgh,pp.12.
Herbert,T.J.,1987.Area projections of sheye photographic
lenses.Agric.For.Meteorol.39,215223.
Herwitz,S.R.,Slye,R.E.,Turton,S.M.,1998a.Co-registered aerial
stereopairs from low-ying aircraft for the analysis of long-term
tropical rainforest canopy dynamics.Photogram.Eng.Remote
Sens.64 (5),397405.
Herwitz,S.R.,Slye,R.E.,Turton,S.M.,1998b.Redening the
ecological niche of a tropical rainforest canopy tree species
using airborne imagery:long-term crown dynamics of Toona
ciliata.J.Trop.Ecol.14 (5),683703.
Jacobs,M.R.,1955.The Growth Habits of Eucalypts.Australian
Government Printer,Canberra,p.198.
Jahnke,L.S.,Lawrence,D.B.,1965.Inuence of photosynthetic
crown structure on potential productivity of vegetation,based
primarily on mathematical models.Ecology 46 (3),319326.
Koike,F.,1985.Reconstruction of two-dimensional tree and forest
canopy proles using photographs.J.Appl.Ecol.22,921929.
Koop,H.,1989.Forest Dynamics.SILVI-STAR:A Comprehensive
Monitoring System.Springer,Berlin,p.229.
Landsberg,J.J.,Gower,S.T.,1997.Applications of Physiological
Ecology to Forest Management.Academic Press,San Diego,
p.354.
Lefsky,M.A.,Harding,D.,Cohen,W.B.,Parker,G.,Shugart,H.H.,
1999.Surface Lidar remote sensing of basal area and biomass
in deciduous forests of Eastern Maryland.USA Remote Sens.
Environ.67,8398.
Means,J.E.,Acker,S.A.,Harding,D.J.,Blair,J.B.,Lefsky,M.A.,
Cohen,W.B.,Harmon,M.E.,M
c
Kee,W.A.,1999.Use of
large-footprint scanning airborne Lidar to estimate forest stand
characteristics in the Western Cascades of Oregon.Remote
Sens.Environ.67,298308.
Ng,F.S.P.,1976.Shyness in trees.Nat.Malaysiana 2,153159.
Oberbauer,S.F.,Clark,D.B.,Clark,D.A.,Rich,P.M.,Vega,G.,
1993.The light environment,gas exchange and annual growth
of saplings of three species of rainforest trees in Costa Rica.
J.Trop.Ecol.9 (4),511523.
Philip,M.S.,1994.Measuring Trees and Forests.CAB
International,Wallingford,UK,p.310.
Rich,P.M.,Clark,D.B.,Clark,D.A.,Oberbauer,S.F.,1993.
Long-term study of solar radiation regimes in a tropical wet
forest using quantum sensors and hemispherical photography.
Agric.For.Meteorol.65 (12),1993.
Rowe,R.,Sharma,N.,Browder,J.,1991.Deforestation:problems,
causes and concerns.In:Sharma,N.(Ed.),Managing the
World's Forests.Kendall/Hunt Publishing Co.,Dubuque,pp.
3345.
Samarasinghe,S.J.,1995.Thinning guidelines for tree species of
different successional status.J.Trop.For.Sci.8 (1),4452.
Schaffer,B.,Ploetz,R.C.,Mason,L.J.,1991.A simple method
for quantifying light transmission through hemispherical
photographs of plant canopies.Hortscience 26,208.
Suri,S.K.,1975.Quantitative thinning model with particular
reference to Sal forests of Changbhakahar Forest Division
(Madhya Pradesh).Ind.For.101 (1),8089.
Swanborough,P.W.,Doley,D.,Keenan,R.J.,Yates,D.J.,1998.
Photosynthetic characteristics of Flindersia brayleyana and
Castanospermum australe from tropical lowland and upland
sites.Tree Physiol.18,341347.
ter Steege,H.,1994.HEMIPHOT:a programme to analyze
vegetation indices,light and light quality from hemispherical
photographs.The Tropenbos Foundation,Wagningen,The
Netherlands.
Thompson,W.A.,Stocker,G.C.,Kreidemann,P.E.,1988.Growth
and photosynthetic response to light and nutrients of Flindersia
212 P.L.Brown et al./Agricultural and Forest Meteorology 100 (2000) 199212
brayleyana F.Muell.,a rainforest tree with broad tolerance to
sun and shade.Aust.J.Plant Physiol.15,299315.
Umeki,K.,1995.A comparison of crown asymmetry between
Picea abies and Betula maximowicziana.Can.J.For.Resour.
25,18761880.
Waring,R.H.,1983.Estimating forest growth and efciency in
relation to canopy leaf area.In:Macfadyen,A.,Ford,E.D.
(Eds.),Advances in Ecological Research.Academic Press,pp.
327354.
Whitmore,T.C.,Burnham,C.P.,1984.Tropical Rainforests of the
Far East:With a Chapter on Soils by C.P.Burnham.Clarendon
Press,Oxford,Oxfordshire,p.352.
Wilkinson,D.M.,1995.Modeling tree crowns as geometric solids.
Arboric.J.19,387393.
Zeide,B.,1998.Fractal analysis of foliage distribution in loblolly
pine crowns.Can.J.For.Resour.28,106114.