Ecological Society of America

A Practical Method for Mapping Trees Using Distance Measurements

Author(s): Emery R. Boose, Emery F. Boose and Ann L. Lezberg
Source: Ecology, Vol. 79, No. 3 (Apr., 1998), pp. 819-827
Published by: Ecological Society of America
Stable URL: http://www.jstor.org/stable/176581
Accessed: 29-10-2015 20:56 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].

Ecological Society of America is collaborating with JSTOR to digitize, preserve and extend access to Ecology.

http://www.jstor.org

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
Ecology, 79(3), 1998, pp. 819-827
(C 1998 by the Ecological Society of America

A PRACTICAL METHOD FOR MAPPING TREES

USING DISTANCE MEASUREMENTS
EMERY R. BooSE,1 EMERY E BOOSE,2 AND ANN L. LEZBERG'3
'HarvardForest, Harvard University, Petersham, Massachusetts 01366 USA
2340 North Street, Georgetown, Massachusetts 01833 USA

Abstract. Accurate maps of the locations of trees are useful for many ecological studies
but are often difficult to obtain with traditional surveying methods because the trees hinder
line of sight measurements. An alternative method, inspired by earlier work of E Rohlf
and J. Archie, is presented. This "Interpoint method" is based solely on tree diameter and
tree-to-tree distance measurements. A computer performs the necessary triangulation and
detects gross errors. The Interpoint method was used to map trees in seven long-term study
plots at the Harvard Forest, ranging from 0.25 ha (200 trees) to 0.80 ha (889 trees). The
question of accumulation of error was addressed though a computer simulation designed
to model field conditions as closely as possible. The simulation showed that the technique
is highly accurate and that errors accumulate quite slowly if measurements are made with
reasonable care (e.g., average predicted location errors after 1000 trees and after 10000
trees were 9 cm and 15 cm, respectively, for measurement errors comparable to field
conditions; similar values were obtained in an independent survey of one of the field plots).
The technique requires only measuring tapes, a computer, and two or three field personnel.
Previous field experience is not required. The Interpoint method is a good choice for
mapping trees where a high level of accuracy is desired, especially where expensive sur-
veying equipment and trained personnel are not available.
Key words: distance measurements; error accumulation; error correction; interpoint distances;
simulation model; spatial pattern; stand mapping; surveying; tree mapping; triangulation.

INTRODUCTION simpler and permit much higher accuracy to be ob-

tained in the field. The triangulations required to cal-
Accurate maps of the locations of trees are essential
culate tree coordinates from the field data are easily
for spatially explicit studies of tree populations and
performed by computer. The Rohlf-Archie method has
forest dynamics (e.g., Sterner et al. 1986, Condit et al.
been used in a number of published studies (e.g., Mitton
1992, Moeur 1993). However such maps are often dif-
and Grant 1980, Robertson 1984, Glitzenstein et al.
ficult to obtain with traditional surveying methods, be-
1986; with modifications in Kenkel 1988). However
cause the trees hinder line of sight measurements and
the technique as proposed is problematic for large num-
the transit must be moved frequently to map an entire
bers of trees, because it can be very difficult to isolate
stand. An alternative method is to establish a grid of
input errors (recording errors or gross measurement
points with a transit, lay out measuring tapes on the
errors). In addition the computer program provided by
ground at right angles between grid points, and then
Rohlf and Archie has been shown to generate large
estimate tree coordinates from the tapes by eye or with
errors with both field collected and computer simulated
a right-angled prism (e.g., Reed et al. 1989). Although
data, raising the question of whether unacceptable ac-
GPS (global positioning system) devices can provide
cumulation of error is inherent to the technique (Hall
accurate locations of bench marks, current hand-held
1991).
units are accurate only to -1-10 m and are adversely
In 1990 we developed our own computer program
affected by a heavy canopy. In addition there is the
(INTERPNT) for mapping trees using tree diameters
problem of measuring the vector between the GPS re-
and tree-to-tree distances, inspired by Rohlf and Ar-
ceiver and the center of the tree.
chie's method (but without least squares optimization),
In 1978 Rohlf and Archie proposed a method for
and incorporating practical improvements from exten-
mapping trees based entirely on tree diameter and tree-
sive field testing. These improvements included sim-
to-tree distance measurements. Distance measurements
plification of the information recorded in the field and
are preferable to angle measurements, because they are
the ability to isolate recording and gross measurement
errors. The resulting "Interpoint method" was used to
Manuscript received 18 November 1996; revised 2 May map trees in several long-term study plots at the Har-
1997; accepted 13 May 1997; final version received 9 June
1997.
vard Forest during the period 1990-1994. More re-
3 Present address: College of Forest Resources, University cently the critical question of accumulation of error
of Washington, Seattle, Washington 98195 USA. was resolved through extensive computer simulation of
819

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
820 EMERY R. BOOSE ET AL. Ecology, Vol. 79, No. 3

the technique. Our results suggest that the Interpoint each target tree is located it is flagged so that the sur-
method is both easy to use and highly accurate. In this veyors can recognize it as a potential reference tree.
paper we explain how the Interpoint method works, Target trees may be recorded in any order; we found
how the method was tested in the field, and how the that recording them in order of tag number helped us
method was tested by computer simulation. The IN- ensure that all trees were accounted for. For each target
TERPNT program (with documentation and field in- tree the three reference trees are recorded in clockwise
structions) is available on the Harvard Forest Web order as seen from above.
page.4 4) At the end of each field day, field measurements
are entered into a computer and analyzed. Possible er-
METHODS
rors are noted and investigated at the beginning of the
Interpoint method next field day.
Overview.-The Interpoint method requires that the Computer analysis.-Analysis of the field data by
plot to be surveyed have at least three reference points computer includes checking for errors and calculating
(bench marks) with known Cartesian coordinates. Trees Cartesian coordinates for all trees mapped to date, and
in the plot are numbered and the dbh (diameter at breast proceeds as follows:
height) measured. Each new tree (target tree) is located 1) The list of located target trees is sorted so that
by measuring the distances to three previously located each target tree is preceded in the list by the three
trees (reference trees) or bench marks. A computer cal- reference trees or bench marks that were used to locate
culates the coordinates of the center of the target tree it. This procedure simplifies field operations by making
by triangulation and averaging, using the three triangles it possible to record target trees in any order (if the
formed by the three distance measurements and the computer assumes that target trees are listed in the
measured diameters at breast height. Possible errors are order of actual measurement, then a single recording
identified in the data analysis and investigated in the mistake could render a large data set virtually unusa-
field. ble). The sort also detects two critical errors that are
Field measurements.-Field work for the Interpoint difficult or impossible to detect by hand: (a) missing
method includes establishing bench marks, labelling trees, i.e., references to trees that do not exist or have
trees, measuring tree dbh, and measuring tree-to-tree not yet been located; and (b) circular references, i.e.,
distances, and proceeds as follows: references (perhaps through a long series of trees) back
1) Three (or more) bench marks are located and to the target tree itself. These errors result from re-
permanently marked. Bench marks should be moder- cording the wrong tree number or selecting a tree not
ately spaced and their locations with respect to one yet located as a reference tree. Circular references are
another determined as accurately as possible. We found particularly difficult to isolate, since finding them may
it convenient to use one corner of the plot as a bench require examining all possible pathways from the target
mark, locating the other two bench marks at 15 m and tree to trees in the last data set that was sorted suc-
20 m along the two neighboring sides and carefully cessfully. Because the magnitude of such a search in-
adjusting the points to create a 3-4-5 right triangle (i.e., creases exponentially as the number of trees increases,
15-20-25 m). it is important that new field data be processed regularly
2) Each tree to be mapped is labelled and measured to avoid the possibility of an unreasonably long search.
for dbh to the nearest 0.1 cm. We used numbered alu- 2) Cartesian coordinates for the center of each target
minum tags at dbh height (1.37 m); the tags provided tree are calculated by triangulating from each of the
a convenient reference for setting the height of the three pairs of reference trees (or bench marks). Each
measuring tape for the tree-to-tree measurements. pair of reference trees yields two possible locations for
3) Beginning in the vicinity of the bench marks, and the target tree. The problem of mirror triangles is
moving gradually across the plot, each target tree is solved by recording the reference trees in clockwise
located by measuring the horizontal distance to three order (Fig. 1). This procedure is easier and less prone
reference trees (or bench marks). Reference trees close to error than recording all three left/right pairs as E J.
to the target tree (but not closer than 1 m) are selected. Rohlf and J. W. Archie suggest.
Pairs of reference trees that subtend an angle of <200 3) In rare cases triangulation fails because of an
or >1600 at the target tree are avoided, because such open triangle (where the length of one side is greater
pairs may magnify measurement errors (e.g., errors than the sum of the lengths of the other two sides).
may increase by a factor of 20 at angles of 50 or 1750). Here the computer adjusts the measured distances from
Distances are measured from bark to bark to the nearest the target tree to the two reference trees to make the
0.01 m, holding the measuring tape as level as possible triangle close and permit triangulation to continue. The
at dbh height on the target tree (a third person makes magnitude of the adjustment and the relevant tree num-
it easier to level the tape and a plumb bob makes it bers are noted and such cases are checked in the field
easier to measure distances to the bench marks). After for possible recording errors, measurement errors, or
pairs of reference trees subtending angles <200 or
4 URL = http://lternet.edu/hfr >160?.

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
April 1998 MAPPING TREES WITH DISTANCE MEASUREMENTS 821

(a) R2 sured distance (the measured bark to bark distance plus

,' ~- ^R2
one-half the dbh of each tree). Target trees are marked
RI in the computer output if the absolute value of any of
the three distance differences is >10 cm. Marked trees
near the top of the sorted list are checked first in the
,,0 field (often if a tree is marked then later trees measured
back to it are also marked, so the first marked trees are
investigated first).
5) Maps showing tree locations and tree numbers
(or other collected data such as tree species) may be
R3 generated by computer using the calculated coordinates
of the trees surveyed to date. Such maps are very useful
for locating particular trees and investigating problems
(b) Q in the field as the survey continues.
6) Possible errors identified in the computer anal-
I, N ysis (missing trees, circular references, open triangles,
I,, _ ' '
and marked trees) are all checked in the field. In our
RI R2 R 3 surveys, if a tree was marked but no recording or mea-
surement errors were discovered, one or more of the
reference trees were replaced with new reference trees
(referred to below as "reference tree replacement"). In
T rare cases this was done more than once for the same
target tree.
7) Overall accuracy may be improved by applying
a "distance correction factor" to the tree-to-tree mea-
(c) QC
surements in cases where there is independent and re-
liable evidence of measurement bias in a particular plot.
Field test
RI *R2 R3
Harvard Forest plots.-The Interpoint method was
used to map the locations of standing trees with dbh
- 5 cm in seven rectangular plots at the Harvard Forest
in central Massachusetts, USA (Table 1). The plots con-
tained mature stands and ranged in size from 0.25 ha
T (200 trees) to 0.80 ha (889 trees). Average slopes
FIG. 1. Elimination of mirror triangles in the analysis of ranged from level to 5?. One plot (Plot 1) was domi-
field data. Each pair of reference trees yields two possible nated by Tsuga canadensis; the others by Quercus ru-
locations for the target tree (T). Alternate triangulated posi- bra and Acer rubrum. The T. canadensis plot, which
tions are shown as open circles. Recording reference trees in has a small bog (10 m X 20 m) in the center of the
clockwise order from above (e.g., R1-R2-R3) leads to four
possible solutions for the location of the target tree: (a) The
plot, was mapped in the winter when it was possible
first solution arises if each of the three angles subtended at to measure tree-to-tree distances directly across the ice
the target tree is <180 and the recorded order yields the of the bog. The other plots were mapped in the summer.
correct left/right sequences for triangulation (R1-R2, R2-R3, Field measurements were performed by two or occa-
R3-R1). (b) Three other solutions arise if one of the three sionally three people. Species, crown position, and
angles subtended at the target tree is > 180 and the left/right
sequence for that pair must be reversed (here R1-R3 instead health were recorded for each tree. Tree-to-tree dis-
of R3-R1). The computer evaluates all four possible solutions tances were measured with fiberglass tape; measure-
and selects the one in which the three triangulated positions ments were not corrected for tape stretch.
are closest to one another. (c) If the order of reference trees Field check of accuracy.-Accumulation of error in
(clockwise or counterclockwise) is not known, then it is not
one plot (Plot 3) was checked by directly measuring
possible to choose between actual and alternate triangulated
positions in cases where the reference trees are nearly col- the distances between 18 pairs of widely spaced trees,
linear. including 10 pairs of trees -30 m apart and parallel
to the short axis of the plot and 8 pairs of trees -75
m apart and parallel to the long axis of the plot. Tree
4) For each target tree and associated reference tree pairs were selected so that no tree was closer than 5
(or bench mark) a "distance difference" is calculated m to the plot boundary, and the pairs were widely
as an indication of possible error. The distance differ- spaced in an effort to sample the entire plot. For each
ence is the calculated distance between trees (based on pair of trees, the dbh of each tree, the bark to bark
the calculated Cartesian coordinates) minus the mea- distance from breast height on one tree to breast height

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
822 EMERY R. BOOSE ET AL. Ecology, Vol. 79, No. 3

TABLE 1. Harvard Forest plots mapped with the Interpoint method.

Width Length Area No. Avg. Avg. Max. abs. Avg. abs. Avg. diff.
Plot (m) (m) (ha) trees dbh (cm) dist. (m) diff. (m) diff. (m) (m)
1 60 120 0.72 742 21.6 5.13 0.19 0.0175 -0.0008
2 50 160 0.80 889 15.2 4.35 0.17 0.0164 -0.0007
3 50 120 0.60 776 14.5 4.84 0.08 0.0108 0.0003
4 50 50 0.25 241 16.8 5.80 0.10 0.0193 0.0001
5 50 50 0.25 228 18.3 5.47 0.10 0.0224 0.0005
6 50 50 0.25 203 17.7 5.87 0.11 0.0179 -0.0002
7 50 50 0.25 200 21.0 4.99 0.10 0.0206 -0.0001
Total 3.12 3279
Weighted average 17.3 4.96 0.0163 -0.0003
Notes: Avg. dbh = average tree diameter at breast height. Avg. dist. = average measured distance between target tree and
reference tree, center to center. Max. abs. diff., avg. abs. diff., and avg. diff. = maximum absolute value, average absolute
value, and average distance difference (calculated distance between target tree and reference tree minus measured distance).

on the other tree, and the angle of the tape endpoints 6) What is the effect of a large measurement bias?
above or below horizonal were measured. Measured Simulation of errors.-Tree-to-tree measurements
distances between trees were adjusted using the man- include errors that always decrease measurements (neg-
ufacturer's suggested correction for the stretch of the ative bias; e.g., tape stretch), errors that always increase
fiberglass tape under tension. These values were used measurements (positive bias; e.g., catenary sag of
to calculate the horizontal distance between tree cen- tape), and errors that can be positive or negative (no
ters. This measured distance was then compared to the bias; e.g., departure of tree cross sections from circu-
calculated distance using the Cartesian coordinates lar). Of these, the two errors judged to be largest and
from the Interpoint method. most difficult to correct for were modeled in the sim-
ulation:
Simulation test 1) The elevation angle error resulting from failure
Objectives.-Though the tree maps generated from to hold the tape endpoints level. This error is always
our field data appeared to be quite accurate, our field positive and is calculated as s = L(l/cos A - 1) where
test was not sufficient to quantify the accumulation of s = error in length, L = actual length, and A = tilt
error or to rule out occasional catastrophic error. To angle of tape from horizontal. In the simulation the tilt
address these issues we developed a computer simu- angle was represented by a normal distribution function
lation of the technique in which we attempted to model (mean = 0) that provided both positive and negative
field conditions as closely as possible. Two measure- angles (but note that s is always positive, creating a
ment errors were simulated: (1) an elevation angle error small positive bias). A standard deviation of 10 was
associated with failure to hold the tape endpoints level used in all cases except one. Experimentation in the
(always positive), and (2) a linear measurement error field suggested that with care we could reliably hold
associated with human error in reading the tape (pos- the tape to within one-third degree of level by eye over
itive or negative). The simulation permitted us to ob- a distance of five meters, so our results probably did
serve the build up of location error (defined as the not underestimate the actual elevation angle error.
distance between actual and calculated tree location) 2) The linear measurement error associated with hu-
across a large plot for many hundreds of runs. The man error in reading the tape. This error may be pos-
following questions were investigated with the simu- itive or negative. In the simulation it was represented
lation: by a normal distibution function (mean = 0) whose
1) How much does location error accumulate across standard deviations were 1, 1.5, 3, 6, 12, or 24 cm.
the plot? How much does location error vary as a func- This wide range of standard deviations was chosen to
tion of the standard deviation (SD) of the linear mea- exceed likely conditions in the field. Field data from
surement error? For a given standard deviation of linear the Harvard Forest plots suggested that the actual stan-
measurement error, how much does location error vary dard deviation of the linear measurement error in the
between different runs? Harvard Forest surveys was on the order of 1.5 to 3
2) Does reference tree replacement improve overall cm. Note that in the simulation the average linear mea-
accuracy? surement error was at least an order of magnitude great-
3) Does correction of the positive bias caused by er than the average elevation angle error.
elevation angle errors improve overall accuracy? Simulation of tree mapping.-Target trees were lo-
4) Does the shape of the plot have an effect on cated at random in square or rectangular plots. Plots
overall accuracy? contained 1000 or 10000 trees, and tree density was
5) What is the effect of a single, large, uncorrected set to 1000 trees/ha based on the Harvard Forest field
error? data. The selection of reference trees was patterned as

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
April 1998 MAPPING TREES WITH DISTANCE MEASUREMENTS 823

closely as possible after the field practice. Tree-to-tree erence trees for all seven plots was 4.96 m (Table 1).
measurements were simulated by adding the elevation Measurement errors were detected by computer and
angle error and the linear measurement error to the corrected in the field for 2.1% of the target trees. An-
actual distance. other 4.5% of the target trees showed distance differ-
The practice of reference tree replacement was mod- ences >10 cm, but no errors were discovered in the
eled in selected runs by calculating the distance dif- field. Here one or more reference trees were replaced
ference for each target tree and its three reference trees. with new trees. At the end of the survey the maximum
If the distance difference exceeded 3.5 times the stan- absolute value of the distance differences in all plots
dard deviation of the linear measurement error, then a was 19 cm, and only 22 trees (0.67% of all trees) had
new set of measurement errors was chosen by the usual values >10 cm. The average absolute value of the dis-
process, simulating remeasurement in the field. If the tance difference across all plots was 1.63 cm, while the
new distance difference still exceeded 3.5 times the average signed difference was close to zero (-0.03
standard deviation of the linear measurement error,then cm).
all three reference trees (or as many as possible, but at The average time required across all seven plots to
least one) were replaced with new trees. tag each tree, measure dbh, identify species, and es-
The elevation angle error always introduces a small timate crown position and health was -6 d/ha. Of these
positive bias in the tree-to-tree measurements. The pos-
tasks, only the first two were necessary to map the trees,
sibility of correcting for this bias was modeled in se-
and estimating crown position required the most time.
lected runs by multiplying the simulated measured dis-
The average time required to locate target trees, mea-
tances between trees by the cosine of the estimated
sure distances to three reference trees, flag located
average elevation angle of the tape.
trees, analyze data, and make any necessary remea-
The effects of plot shape on overall accuracy were
tested by running the simulation for rectangular plots surements was -9 d/ha. These estimates were based
of different length-to-width ratios. The impacts of a on two person crews, 8-h days, and a tree density of
single, large, uncorrected error were tested by assigning -1000 trees/ha. In most cases the crews were under-
erroneous coordinates to one of the bench marks and graduate students with little or no field experience.
examining location errors across the plot. The effects Field check of accuracy.-A comparison of mea-
of a 0.25-1.0 cm measurement bias were tested by sured distances and calculated distances between wide-
setting the mean linear measurement error to a range ly spaced trees in Plot 3 showed close agreement (Table
of positive and negative values. 2). For the 10 pairs of trees parallel to the short axis
Statistical methods.-Trees for 1000 tree runs, or- (50 m) of the plot, the average measured distance was
dered on their distance from the origin, were divided 28.93 m, and the average absolute value of the distance
into 10 bands of 100 consecutive trees each; while trees differences was 4 cm. For the 8 pairs of trees parallel
for 10 000 tree runs were divided into 16 bands of 625 to the long axis (120 m) of the plot, the average mea-
trees each. For each band, location errors were sorted sured distance was 76.08 m, and the average absolute
into one of twenty equal classes to build a sample fre- value of the distance differences was 16 cm. Note that
quency distribution. In addition the sample mean, the for each shorter pair the two trees were roughly equi-
sample standard deviation, and the maximum location distant from the bench marks, while for each longer
error were calculated for each band. pair one tree was much farther from the bench marks
The sample frequency distribution of location error than the other tree.
for a single band typically has a bell shaped curve, In all cases but one, the calculated distance for the
somewhat skewed on the upper end. Since a large num- longer pairs was longer than the measured distance,
ber of samples is used to calculate the sample mean suggesting (as expected) a tendency toward positive
and the maximum for each band, the sample means and measurement bias and the accumulation of positive er-
maxima for each band over multiple runs should be ror with distance from the bench marks. Cartesian co-
nearly normally distributed. Each set of input condi- ordinates for the plot were recalculated using a distance
tions was tested with at least 100 runs, and the results correction factor of 0.9980, the ratio between the av-
were used to calculate sample frequency distributions erage measured distance between the longer pairs
of the mean and the maximum location error over all (76.08 m) and the average calculated distance between
runs for each band. the longer pairs (76.23 m). Calculated distances using
The effects of various corrective methods (e.g., ref- the new coordinates were then compared to the mea-
erence tree replacement) were evaluated by making at sured distances. For the 10 shorter pairs the average
least 100 runs for each method. The means of the mean
absolute value of the distance differences increased
location error in the final band of trees were then com-
slightly from 4 to 6 cm, while for the 8 longer pairs it
pared statistically using a t test.
decreased significantly from 16 to 6 cm.
RESULTS
Field test Simulation test
Harvard Forest plots.-The average measured dis- Accumulation of location error.-Results of 100
tance (center to center) between target trees and ref- runs for 1000 and 10000 trees suggest that location

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
824 EMERY R. BOOSE ET AL. Ecology, Vol. 79, No. 3

TABLE 2. Field check of accumulation of error in Plot 3 (50 m X 120 m) at Harvard Forest.

10 pairs of trees along width of plot 8 pairs of trees along length of plot
Meas. Calc. Abs. Meas. Calc. Abs.
dist. dist. Diff. diff. dist. dist. Diff. diff.
(in) (in) (in) (in) (in) (in) (in) (in)

No distance correction
Average 28.93 28.92 -0.01 0.04 76.08 76.23 0.15 0.16
Minimum 24.47 24.41 -0.06 0.02 65.08 65.20 -0.04 0.04
Maximum 32.57 32.55 0.07 0.07 87.61 87.82 0.24 0.24
Distance correction factor = 76.08/76.23 0.9980
Average 28.93 28.87 -0.06 0.06 76.08 76.09 0.00 0.06
Minimum 24.47 24.37 -0.10 0.01 65.08 65.08 -0.19 0.00
Maximum 32.57 32.50 0.01 0.10 87.61 87.66 0.08 0.19
Notes: Measured distances between widely separated trees were compared to calculated distances using two sets of calculated
coordinates (no distance correction and distance correction factor = 0.9980). Meas. dist. = measured distance between trees,
center to center. Calc. dist. = calculated distance between trees, center to center. Diff. = calculated distance minus measured
distance. Abs. diff. = absolute value of calculated distance minus measured distance.

errors increase slowly as one moves across the plot, first two or three bands and nearly constant thereafter,
away from the initial bench marks (Fig. 2). As ex- with a slight increase in the final band which may have
pected, location errors (measured here as the means of resulted from the more limited choice of reference trees
the mean location error) were roughly proportional to in the far corner of the plot. The slope in the middle
the standard deviation of the linear measurement error. of the curve could probably be extrapolated to furnish
In each case the slope of the curve was greatest in the a good estimate of the location errors for plots with >
10000 trees.
60 - 1000 Trees The simulation showed moderate variation between
runs but no evidence of catastrophic error (Table 3).
50 - Even for exceptionally poor measurements (SD of linear
_
measurement error = 24 cm, SD of elevation angle error
-
40
40 = 20), the mean of the mean location error in the final
0 30 1 2.0 band of 10000 trees was only 1.23 m. The maxima of
the maximum location error for a SD of linear mea-
30 -
cz - surement error of 3 cm were 32 cm and 51 cm for 1000
U 20 6.0 and 10 000 trees, respectively; note that these were the
0

10 =- 3.0 worst locations errors in 100 000 and 1 000 000 trees.
G Cow to Wv 1.5 Reference tree replacement.-The value of reference
; ~~1.0
tree replacement, which was used in mapping all of the
1 2 3 4 5 6 7 8 9 10
Harvard Forest plots, was tested by comparing 100-
60 10 000 Trees 300 runs for each of the initial conditions described in
12.0 Table 3, both with and without reference tree replace-
50 , ment. Results for the final band for both 1000 and
E 10000 trees showed that reference tree replacement
o 40
improved the means of the mean location error in four
0
of six cases, but worsened it in the other two cases. In
a 30 -6.
.o- 6.0 all six cases the standard deviation of the mean location
0
C 20 error was improved. The results suggested a general
0 ~~~~~~~~~~~~~~~3.0
trend toward improvement; however no cases were sta-
-J ~~~~~~~~~~~~~~~~1.5
tistically significant at the 0.05 confidence level.
10 1.0
Elevation angle bias correction.-The value of cor-
0
recting for elevation angle bias was tested by compar-
1 2 3 4 5 6 7 8 9 1011 1213141516
ing location errors in sets of 100 runs with an angle
Band error of 1? and angle bias corrections of 0.0 (no cor-
FIG. 2. Variation in simulated location errors (i.e., dis- rection) to 2.5? in 0.5? increments (Fig. 3). Results
tances between actual and calculated tree locations) in square showed that the location error in the final band de-
plots of 1000 and 10000 trees as a function of the standard creased as the angle correction increased from no cor-
deviation of the linear measurement error (cm). Points show
the means of the mean location error for a given band for
rection to 1.00, but then increased beyond the uncor-
100 runs of the simulation model. Standard deviation of el- rected value as the angle correction increased from 1.0?
evation angle error =10. to 2.5?. A t test showed that the reduced mean of the

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
April 1998 MAPPING TREES WITH DISTANCE MEASUREMENTS 825

TABLE 3. Computer simulation of the Interpoint method for square plots of 1000 trees and 10000 trees.

Linear
ineas. Elev. angle Mean of the SD of the Max. of the Mean of the SD of the Max. of the
SD (cm) SD (degree) means (m) means (m) means (m) maxima (m) maxima (m) maxima (m)
1000 Trees Band 10 (trees 901-1000)
1.0 1.0 0.039 0.013 0.090 0.069 0.015 0.119
1.5 1.0 0.050 0.019 0.125 0.093 0.023 0.168
3.0 1.0 0.088 0.037 0.231 0.172 0.048 0.320
6.0 1.0 0.170 0.074 0.446 0.338 0.100 0.659
12.0 1.0 0.332 0.149 0.906 0.669 0.196 1.260
24.0 2.0 0.691 0.302 1.880 1.372 0.376 2.604
10000 Trees Band 16 (trees 9376-10000)
1.0 1.0 0.098 0.031 0.200 0.144 0.031 0.236
1.5 1.0 0.105 0.041 0.249 0.172 0.043 0.300
3.0 1.0 0.148 0.070 0.402 0.277 0.074 0.511
6.0 1.0 0.274 0.136 0.707 0.529 0.145 0.929
12.0 1.0 0.548 0.276 1.475 1.063 0.304 1.964
24.0 2.0 1.232 0.480 2.536 2.252 0.522 3.584
Notes: Each row shows the results in the final band of trees after 100 runs of the simulation model at the specified linear
measurement error and elevation angle error. For each run there is a mean and a maximum location error in the final band.
The values in the table describe the distribution of these 100 means and 100 maxima.

mean location error for the 1.00 correction group was is a greater distance (and thus a greater number of trees)
better than no correction at a significance level of 0.02; between the initial bench marks and the far end of the
while the 2.5? correction was worse than no correction plot; also there may be some loss of accuracy caused
at a significance level of 0.00. Elevation angle bias by a greater number of trees next to the plot boundary,
correction thus appears to be effective if there is a where the choice of reference trees is more limited.
reliable estimate of the actual angle error for a given Effect of a large error.-The effect of a single, large,
plot; otherwise there is a significant risk of overcom- uncorrected error in a single run was tested by intro-
pensating and decreasing overall accuracy. ducing an initial bench mark error in a plot of 160 trees
Shape of the plot.-The effect of the shape of the and examining the results for each band of 10 trees
plot on location error was tested by comparing location (Fig. 4). The distance differences and location errors,
errors for 100 runs for rectangular plots of the same though initially large, decreased rapidly and by the final
area but different length-to-width ratios (Table 4). As band were comparable to a similar run made with all
expected, the mean and standard deviation of the mean bench marks correctly located. These results suggest
location error in the final band increased as the length- that the averaging process of the Interpoint method
to-width ratio increased. In a long, narrow plot there (using three triangles to locate each target tree) is quite
effective in reducing the propagation of a large error.
30 Measurement bias.-The effects of a systematic
measurement bias were tested by comparing location
25 errors for different values of the mean linear measure-
ment error (Fig. 5). Results of 100 runs for 1000 and
? 20 10 000 trees showed that the location error was directly
0
15
0
TABLE 4. Computer simulation of the Interpoint method for
10 1000 trees in rectangular plots of the same area but different
0
length-to-width ratios.
5
Plot Mean SD Max. Mean SD Max.
0 length/ of the of the of the of the of the of the
0.0 0.5 1.0 1.5 2.0 2.5 width means means means maxima maxima maxima
ratio (i) (i) (i) (i) (i) (i)
Elevationangle adjustment(degrees)
1 0.088 0.037 0.231 0.172 0.048 0.320
FIG. 3. Simulated location errors as a function of eleva- 2 0.121 0.054 0.281 0.213 0.062 0.369
tion angle bias correction. Bars show the means of the mean 3 0.122 0.057 0.315 0.211 0.064 0.387
location error (filled bars), and the means of the mean location 4 0.130 0.063 0.343 0.222 0.070 0.434
error plus one standard deviation (open bars) and plus two Notes: Each row shows the distribution of the mean and
standard deviations (hatched bars), for the last band in square the maximum location error in the final band (trees 901-1000)
plots of 1000 trees for 100 runs of the simulation model. for 100 runs of the simulation model. Standard deviation of
Standard deviation of linear measurement error = 3 cm, and linear measurement error = 3 cm, and standard deviation of
standard deviation of elevation angle error = 1?. elevation angle error = 1?.

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
826 EMERY R. BOOSE ET AL. Ecology, Vol. 79, No. 3

60 100 - 1000 Trees

40 . 10 000 Trees
20_ 80
0
E
02 -20 o 60
a)
0 -40
(i -60 0 Maximum difference .? 40
C -80 * Minimum difference
0
v Avg. absolute difference
-1 00 20
20
v error
Avg. location
-120
1
-140 I I I I
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1516 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Band
Mean linear measurement error (cm)
FIG. 4. Simulated errors by band in a square plot of 160
trees after the introduction of a large bench mark error. One FIG. 5. Simulated location errors in square plots of 1000
of seven bench marks, actually located at (5, 5), was assigned and 10 000 trees -s a function of the mean linear measurement
erroneous coordinates (4, 6), a distance of 1.41 m from the error. Points show the means of the mean location error in
true location. Values show the maximum, minimum, and av- the final band for 100 runs of the simulation model. Standard
deviation of linear measurement error = 3 cm, and standard
erage absolute value of the distance differences (calculated
distance between target tree and reference tree minus mea- deviation of elevation angle error = 1?. Results for corre-
sured distance), and the average location error, for each band sponding negative values of the mean linear measurement
of 10 trees. Standard deviation of linear measurement error error were nearly identical.
= 1.5 cm, and standard deviation of elevation angle error
10.
mutation of error reported by Hall (1991), even when
measurement accuracy is quite poor. On the contrary,
proportional to the mean linear measurement error for the averaging process appears to reduce the propaga-
larger values of the mean. However local error, mea- tion of large, individual errors.
sured as the difference between the calculated distance Drawbacks of the Interpoint method include:
between target tree and reference tree minus the (sim- 1) Field data must be analyzed and problems
ulated) measured distance, remained unchanged. Thus checked on a regular (preferably daily) basis.
a large measurement bias has a significant impact on 2) When a target tree is marked and no immediate
the accumulation of location error across the plot, but errors are discovered, backtracking to search for errors
little impact on the accuracy of a tree's calculated po- among previous trees is time consuming and often im-
sition relative to its immediate neighbors. Note that practical. Our field experience suggests that reference
actual measurement bias should be well under 1 cm in tree replacement solves this problem in nearly all cases,
most cases. For example, over a distance of 5 m, a without the need for backtracking. We suspect that ref-
catenary sag of the tape of 2.5 cm causes an error of erence tree replacement helps to contain the effects of
+0.03 cm, an elevation angle error of 1?causes an error previous minor errors, though the results of the sim-
of +0.08 cm, stretch of a steel tape at 2 kg tension ulation on this point were not conclusive.
causes an error of -0.01 cm, stretch of a fiberglass 3) Measurement bias must be kept to a minimum
tape at 2 kg tension causes an error of -0.10 cm, and to control accumulation of error, particularly in large
a temperature change of 10? C causes a steel tape to plots. An independent check of accuracy is recom-
expand or contract by 0.06 cm. mended for plots >100 trees, e.g., through a series of
long distance tree-to-tree or tree-to-bench mark mea-
DISCUSSION
surements. Measurement bias, if discovered, can be
Any evaluation of surveying methods must consider corrected for to improve overall accuracy. For plots
accuracy, speed, equipment, and personnel. After ex- >1000 trees, accumulation of error can be controlled
tensive field testing we found that the Interpoint method by combining the Interpoint method with traditional
is easy to use, requires no special equipment or training, methods; e.g., by using a transit to divide the plot into
results in a computerized data set, and virtually elim- subplots and using the Interpoint technique to map trees
inates gross errors. Mapping can be done while the trees in each subplot.
are in leaf, because there is no need for extended line 4) Tree-to-tree distance measurements may be dif-
of sight measurements. The simulations showed that ficult or impractical in steep or rough terrain.
the technique is highly accurate for plots of moderate Different surveying methods provide different com-
size if reasonable care is taken in setting out the bench promises between time and accuracy. For example, the
marks and measuring tree-to-tree distances. The tech- times reported by Reed et al. (1989) for mapping trees
nique is best at estimating the relative locations of with the prism technique (1 d/ha for three people) are
neighboring trees, an advantage for neighborhood stud- significantly shorter than the times reported here (9
ies. We found no evidence for the catastrophic accu- d/ha for two people), though Reed's estimate assumed

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions
April 1998 MAPPINGTREES WITH DISTANCEMEASUREMENTS 827

an experienced crew and did not include the time re- R. Lent, P. Micks, K. Newkirk, E. Nilson, T. Peterson, J.
Quisel, A. Smyth, and T. Zebryk for assistance with the field
quired for a surveyor to set out the original 50-m grid.
work. The research was supported by the National Science
On the other hand the Interpoint method appears to be Foundation and is a contribution from the Harvard Forest
significantly more accurate than the prism technique; Long-Term Ecological Research Program.
the average difference in coordinates for trees remea- LITERATURE CITED
sured by Reed et al. (1989) in 18 50 X 50 m plots was
Condit, R., S. P. Hubbell, and R. B. Foster. 1992. Short-term
16 cm (corresponding to a Euclidean distance of -23 dynamics of a neotropical forest: change within limits. Bio-
cm), while our simulation of the Interpoint method pre- science 42:822-828.
dicts an average location error of -5.5 cm (for a 50 Glitzenstein, J. S., P. A. Harcombe, and D. R. Streng. 1986.
X 50 m plot, SD of linear measurement error = 3 cm, Disturbance, succession, and maintenance of species di-
versity in an east Texas forest. Ecological Monographs 56:
SD of elevation angle error = 10) with even better ac-
243-258.
curacy on a local scale. Hall, R. B. W. 1991. A re-examination of the use of interpoint
The Interpoint method can be extended in a variety distances and least squares in mapping forest trees. Ecology
of ways. New technology for measuring distances (e.g., 72:2286-2289.
Kenkel, N. C. 1988. Pattern of self-thinning in jack pine:
laser or ultrasonic range finders) may be faster than
testing the random mortality hypothesis. Ecology 69:1017-
measuring tapes, though such devices should be tested 1024.
for accuracy and especially for measurement bias be- Mitton, J. B., and M. C. Grant. 1980. Observations on the
fore being used with this technique. Stationary objects ecology and evolution of quaking aspen, Populus tremu-
other than trees (e.g., herbs or shrubs) can be mapped loides, in the Colorado front range. American Journal of
Botany 67:202-209.
by measuring distances between plants or between Moeur, M. 1993. Characterizing spatial patterns of trees us-
plants and grid points. And the method can be extended ing stem-mapped data. Forest Science 39:756-775.
to map objects in three dimensions, as Rohlf and Archie Reed, D. D., H. 0. Liechty, and A. J. Burton. 1989. A simple
(1978) suggest, by selecting four bench marks with procedure for mapping tree locations in forest stands. For-
est Science 35:657-662.
known three-dimensional coordinates, and for each tar- Robertson, J. G. M. 1984. Acoustic spacing by breeding
get object measuring distances to four reference ob- males of Uperoleia rugosa (Anura: Leptodactylidae). Zeit-
jects. schrift fur Teirpsychologie 64:283-297.
Rohlf, F. J., and J. W. Archie. 1978. Least-squares mapping
ACKNOWLEDGMENTS using interpoint distances. Ecology 59:126-132.
The authors thank T. Allison, C. Canham, D. Foster, J. Sterner, R. W., C. A. Ribic, and G. E. Schatz. 1986. Testing
Glitzenstein, D. Katz, and A. Lewis for many helpful sug- for life historical changes in spatial patterns of four tropical
gestions, and M. Fluet, J. Gerwin, M. Kennon, K. LeClaire, tree species. Journal of Ecology 74:621-633.

This content downloaded from 128.103.149.52 on Thu, 29 Oct 2015 20:56:37 UTC
All use subject to JSTOR Terms and Conditions