Draft sampling marbled murrelets at sea: tests of survey methods and designs

              DRAFT
5/2499
SAMPLING MARBLED MURRELETS AT SEA:
TESTS OF SURVEY METHODS AND DESIGNS
Martin G. Raphael, Diane M. Evans, and Randall J. Wilk
USFS Pacific Northwest Research Station
Olympia, Washington
Introduction
The marbled murrelet (Brachyramphus marmoratus) is a small, nondescript seabird that
was federally listed as threatened under the Endangered Species Act in 1992. This species has a
unique natural history among seabirds in that it travels up to 50 mi inland away from the ocean to
establish nest sites in the canopies of large trees in older-successional forests (Hamer and Nelson
1995). Because of its association with late-successional forest habitats, the marbled murrelet
became a focal species of the Northwest Forest Plan (NWFP; USDA/USDI 1994), a multi-agency
initiative that provided management direction to maintain and restore populations and
habitat of late-successional and old-growth species on federal lands in western Washington,
Oregon, and northern California One of the critical components of the NWFP and of the
USFWS’s Recovery Plan for the marbled murrelet (U.S. Fish and Wildlife Service 1997) is the
ability to accurately assess the status of the population and to monitor population trends over
time. Most of the typical demographic indices used for wildlife species, such as reproductive
success, juvenile and adult survival, and longevity, are extremely difficult to obtain for marbled
murrelets. Active nests are widely dispersed and difficult to locate due to the murrelet’s secretive
nesting behavior (Hamer and Nelson 1995, Nelson and Hamer 1995). Mark-recapture
techniques refined in the last 3-4 years may lead to refined survival estimates (Cook et al. 1998)
but still have limited utility in locating nests. Thus, researchers and managers have concurred
that the most appropriate approach to population monitoring is to track changes at sea (U.S. Fish
and Wildlife Service 1997, Madsen et al. 1997).
Within the NWFP area, marine surveys for murrelets were conducted by at least five
different groups during 1990-97. The size of vessels, number of observers, type and
configuration of transect used, and the data collected differed among these groups, yet the extent
to which these differences affected the population parameters ultimately reported have not been
tested. The Effectiveness Monitoring Plan for the marbled murrelet (Madsen et al. 1997)
specified that monitoring under this program would be conducted using consistent methodology
(a protocol) throughout the NWFP area. Before this protocol could be defined, a thorough
examination of the different methods and an assessment of which techniques best achieved
population estimates and indices of productivity was needed.
Marbled murrelets are counted on the water from boats by scanning the area ahead and to
the side of the moving vessel. The biggest differences in the methods currently employed are (1)
whether data are collected as a strip or line transect (see Buckland et al. 1993); (2) if collected as
a line transect, whether the distance of the target from the line is determined by a direct estimate
of perpendicular distance or a calculation of perpendicular distance from a direct estimate of
radial distance and a measure of the angle of the target from the line; (3) whether a single or two
observers are used; and (4) how the productivity index (the ratio of juveniles to adults) is
calculated (Raphael and Evans 1997). We tested several aspects of marine survey methods
during the breeding seasons of 1997 and 98. Our objectives were to:
(1) determine if density estimates differed when using one vs. two observers,
(2) assess observer variability and accuracy in estimating perpendicular and radial distances,
(3) determine if density estimates differed between perpendicular- or radial-distance based
calculations,
(4) assess observer variability in identifying adult and juvenile murrelets,
(5) determine if the spatial distribution of adult murrelets differed from juveniles and if the
ability to detect adults differed from that of juveniles, such that productivity indices were
affected,
(6) compare age ratios calculated with three different approaches, and
(7) investigate the effect of transect configuration (straight line vs. other configurations) on
density estimates.
Results from these investigations will be incorporated into development of the survey protocol
for the NWFP Effectiveness Monitoring Program.
General Methods
Our study was conducted in the inland waters of Puget Sound, WA, including Hood
Canal (HC) along the eastern side of the Olympic Peninsula and the inter-island passages of the
San Juan Island (SJI) archipelago. Observers were positioned in the bow of a 17-ft Boston
Whaler equipped with 90-hp outboard motor. This placed observer height at approximately 2 m
above the surface of the water. Observers were experienced in estimating perpendicular
distances and identifying seabirds. Distance estimates were visually recalibrated each day by
deploying a line behind the boat at various distances from 0-100 m. The line was marked with
knots every 5 m and with buoys every 25 m. With few exceptions, experiments were conducted
at Beaufort sea state < 2. Given that the specific methods for each experiment differed, the
remainder of the methods and results are presented by experiment.
One versus Two Observers
Methods.---Data were collected while conducting a ‘typical’ extensive survey in the SJI. Line
transects were run parallel to shore at a 300-m distance. Observers estimated the perpendicular
distance from the transect line of all seabirds encountered within 200 m either side of the boat.
Observer behavior was standard for line transects: observers focused more effort within shallow
angles in front of the boat, but continued to scan out to the sides. Data were recorded into a
handheld micro cassette recorder. A transect or group of transects was run first with one
observer, then repeated shortly after or the next day with two observers. One observer covered
the entire 1800 scanning window; two observers split the scanning window to each cover from
the bow to 900 to one side. A total of 9 transects ranging 4-22 km in length were repeated in each
of three time intervals beginning in mid-July 1998.
Because it was logistically difficult to design an experiment that provided simultaneous
observations of one and two observers, the basis of comparison was the density estimate derived
from the program DISTANCE (Buckland et al. 1993) for the transects as a group for each time
interval. We categorized perpendicular distances from the line into 6 classes: 0-25 m, 25-50 m,
50-75 m, 75-100 m, 100-150 m, 150-200 m. We expected that birds moved around from the
time a transect was run with one observer to when the same ground was covered with two
observers (maximum of 24 hours later), but we assumed that the overall population sampled
remained constant across the transects within the time interval. However, in an attempt to reduce
the added variability from nonsynchronous observations, we analyzed the data two ways:
including just the transects that were covered on the same day with one and two observers, and
including all transects. Seven transects were run on the same day in interval 1 (mid July), 3 in
interval 2 (early August), and 9 in interval 3 (mid August). Note that ‘same day’ does not mean
that the 7 transects in interval 1 were run on the same calendar day, but rather that a given
transect was covered by one observer and two observers on the same day. Covering all 9
transects within a time interval spanned several calendar days.
We also compared the number of murrelets observed per transect by one observer
compared to two observers with a paired t-test, using transects as the samples.
Results.---Density across transects was higher or lower with one observer depending on the time
interval and the subset of transects considered (same day or all). When transects run on the same
day were considered, there was no difference in density estimates in mid July, one observer had a
higher density than two observers in early August, and two observers had a higher density than
one observer in late August (Figure 1). When all transects were considered, there was again no
difference in mid July, and two observers had a higher density estimate than one observer for
early and mid August (Figure 1). Density estimates from the DISTANCE program are most
influenced by the number of birds recorded closer to the transect line. The patterns in density
estimates reported above were driven by the number of murrelets recorded 0-25 m and 25-50 m
from the line. When all transects were considered, two observers recorded significantly more
murrelets (X2 > 3.8, p < 0.05 in all cases) 0-50 m from the line than one observer in late August,
although not in the other intervals (Figure 2). When only transects run on the same day were
considered, one observer recorded significantly more murrelets close to the line than two
observers in interval 2, and the reverse was true in interval 3 (Figure 3). Combining all 3
intervals, two observers recorded significantly more murrelets than one observer only when all
transects were considered (Figure 4). One observer generally recorded significantly more
murrelets than two observers at distances > 100 m from the line (Figures 2, 3, and 4).
Results of paired t-tests showed no significant differences in the number of murrelets
seen per transect per number of observers. One observer recorded, on average, 3 fewer murrelets
per transect than two observers when transects run on the same day were considered (t = -0.378,
p = 0.71), and 6 fewer murrelets on average when all transects were considered (t =
-0.84, p = 0.41; Table 1).
Discussion.--Due to the lack of simultaneous observations, this experiment was limited in its
ability to assess the consequences of using one vs. two observers to conduct murrelet surveys.
Because the absolute number and spatial distribution of murrelets at any one site is in flux,
density estimates will vary at an unknown magnitude within relatively short periods of time. Our
assumption that murrelets might move within and between transects but that the overall
population across transects would remain constant within a time interval is questionable. Our
sample transects did not cover all probable areas to which murrelets could move to or from.
While we had circumstances under which two observers outperformed one observer (as might be
expected), the reverse also was true. Thus, these results do not provide a definitive answer to this
aspect of survey methodology.
Our ‘typical’ extensive survey enlists a single observer but the driver serves as a backup
and points out birds that might be missed by the observer. Thus, our ‘one observer’ sample may
more accurately reflect 1.25 observers.
Mid July Early Aug Late Aug
0
10
20
30
40
50
60
70
80
90
Density (#/ Sq. Km)
1 Obs
2 Obs
Transects on Same Day
Mid July Early Aug Late Aug
0
10
20
30
40
50
60
70
80
90
Density (# / Sq. Km)
1 Obs
2 Obs
All Transects
Intervalsrunasseparatesamples
Figure 1. Comparison of density estimates (from program DISTANCE) obtained by one and
two observers. ‘Transects on same day����� refers to the sample of transects covered by one and two
observers on the same day (n = 7 in mid July, n = 3 in early Aug, and n = 9 in mid August). ‘All
transects’ includes those run on the same or one day apart (n = 9 in each interval).
25 50 75 100 150 200
Distance Band
0
10
20
30
40
50
60
70
80
1ob
2ob
Interval 1 (Mid July)
* *
25 50 75 100 150 200
Distance Band
0
50
100
150
200
250
# Murrelets
1ob
2ob
Interval 2 (Early Aug)
*
*
25 50 75 100 150 200
Distance Band
0
50
100
150
200
250
300
# Murrelets
1ob
2ob
Interval 3 (Mid Aug)
* *
*
*
Figure 2. Number of murrelets recorded by distance from the transect line from all transects (run on same or next day). Significant
differences noted with ‘*’.
25 50 75 100 150 200
Distance Band
0
10
20
30
40
50
60
1ob
2ob
Interval 1 (Mid July)
*
25 50 75 100 150 200
Distance Band
0
10
20
30
40
50
60
# Murrelets
1ob
2ob
Interval 2 (Early Aug)
*
*
*
*
25 50 75 100 150 200
Distance Band
0
50
100
150
200
250
300
# Murrelets
1ob
2ob
Interval 3 (Mid Aug)
* *
*
*
Figure 3. Number of murrelets recorded by distance from the transect line from transects run by one observer on the same day as with
two observers. Significant differences noted with ‘*’.
25 50 75 100 150 200
Distance Bands
0
100
200
300
400
500
600
# Murrelets
1ob
2ob
Transects Run On Same Day
*
25 50 75 100 150 200
Distance Bands
0
100
200
300
400
500
600
# Murrelets
1ob
2ob
Transects Run on Same Day
or 1 Day Apart
*
* *
*
Figure 4. Distribution of murrelets from the transect line pooled across three time intervals (mid
July, early August, mid August). Significant differences noted with ‘*’.
Table 1. To be included.
Observer Variability and Accuracy in Estimating Perpendicular and Radial Distances
Methods.---We compared the direct perpendicular distance estimates of three simultaneous
observers on n = 113 targets under typical survey conditions (see above) over several weeks to
assess variability. Most of the targets were marbled murrelets; remaining targets were other
seabirds. Observers looked at the same target at the same time and recorded their distance
estimates independently. Data were analyzed with two ANOVAs, one using observer as the
factor and one using observer, sea condition, and the interaction of observer and sea condition as
factors. Sea condition was defined as wave height less than or greater than 12 cm. Data also
were compared with paired t-tests.
We tested the accuracy of distance estimates using three other observers and running
random line transects through an array of stationary targets (small, anchored buoys). We
compared, with paired t-tests, each observer’s estimate of radial distance with the actual radial
distance as measured with a laser rangefinder at the same time and from the same observation
point. We also compared, with paired t-tests, each observer’s direct estimate of perpendicular
distance to the perpendicular distance calculated from the laser-measured radial distance and the
angle obtained from the bearings recorded by the radial distance observer. Two bearings were
read from a digital compass for each target: the bearing of the transect line, and the bearing of the
target. Difference between these measures was the angle of the target from the transect line.
Perpendicular distance was calculated as:
perpendicular distance = absolute value(radial distance * sin(angle of target from line))
Each observer was not estimating both radial and perpendicular distances at each target. For a
given number of targets, each observer was assigned one method and continued to use that
method until a sample of ~40 had been recorded. Observers then changed to a different method
for the next 40+ targets.
Results.---Of the three observers simultaneously estimating perpendicular distance to the same
target, one observer consistently had a higher estimate than the other two. This was not a
significant difference when just observer was considered in ANOVA models (F = 0.648, p =
0.52; Figure 5). In paired t-tests, distances of this observer were significantly greater compared
to one of the other observers, by 5 meters on average (Table 2). Differences between other pairs
of observers were 1.5 and 2.75 m, on average, per target. Sea condition affected distance
estimates for two of the three observers (F = 5.18, p = 0.02; Figure 5), with lower average
distances at wave heights > 12 cm compared with < 12 cm.
When directly compared with a known distance, observers under- and overestimated
radial distances, with average mean differences ranging 0.4 - 9.6 m (Figure 6). All three
observers overestimated perpendicular distance when compared to a perpendicular distance
calculated from the laser-measured radial distance. Direct estimates were on average 5.4 - 22 m
greater than laser-based calculated distances (Figure 7).
OB1 OB2 OB3
0
10
20
30
40
50
60
Mean Distance (m)
ANOVA F = 0.648, p > 0.5
observer as factor
OB1 OB2 OB3
SS0 SS1 SS0 SS1 SS0 SS1
0
10
20
30
40
50
60
observer, sea condition,
and interaction as factors
F Sign.
Observer 1.69 p > 0.1
Sea Cond. 5.18 p < 0.02
Obs * Sea 1.39 P > 0.25
ANOVA F = 1.85, p > 0.1
Figure 5. Comparison of mean estimated perpendicular distances among three simultaneous
observers and under two sea conditions.
Table 2. Differences in direct perpendicular estimates between pairs of simultaneous observers.
Mean Observer Mean
Observer Distance (n) 95% CI Pair Difference (n) t Sig.
Obs1 38.24 (130) 33.5-43.0
Obs2 39.73 (130) 34.8-44.7
Obs3 43.20 (113) 37.1-49.3
1-2 -1.49 (130) -1.32 0.19
1-3 -4.60 (113) -3.79 0.00
3-2 2.75 (113) 1.67 0.98
Radial Laser Radial Laser Radial Laser
0
10
20
30
40
50
60
70
80
90
Distance (m)
OB1 OB2 OB3
Radial overstimated
on avg. 0.4 m
Radial underestimated
on avg. 5.7 m
Radial overestimated
on avg. 9.6 m
t = 0.37
p > 0.5
n = 60 t = -4.2
p < 0.01
n = 36
t = 3.66
p < 0.01
n = 58
OB1 OB2 OB3
Direct Laser Direct Laser Direct Laser
0
10
20
30
40
50
60
70
80
Distance (m)
t = 3.9
p < 0.01
n = 35
Direct
overestimated
on avg 5.4 m
Direct
overestimated
on avg 10 m
Direct
overestimated
on avg 22 m
t = 4.0
p < 0.01
n = 43
t = 7.3
p < 0.01
n = 60
Figure 6. Performance of individual observers in estimating radial distances, as tested against a
laser rangefinder.
Figure 7. Comparison of observers’ direct estimates of perpendicular distance with a
perpendicular distance calculated from laser rangefinder-measured radial distances and angles.
0 10 20 30 40 50 60 70 80 90
Radial Angle (degrees) from Boat
0
1
2
3
4
5
6
Perpendicular Distance (m)
1 degree
2 degree
3 degree
Figure 8. Effect of angle errors on calculated perpendicular distance at a radial distance of 100
m.
Discussion.---Most survey teams deploy a line from the stern of the boat and move the line in
and out to calibrate observers to different straight-line distances. Observers receive instant
feedback on the accuracy of their estimate. However, objects on the water in front of the boat
appear in a different perspective, and distances may be more difficult to gauge. This is
particularly true if trying to translate to perpendicular distance. By using a rangefinder, it is
possible to apply the same training to direct perpendicular estimates that are used for straight-line
estimates. Evaluating the variability among observers’ distance estimates early on allows for
sufficient training to reduce that variability. In addition, consistent over- or underestimates can
be accounted for in data analysis by applying a correction factor.
We used laser-measured distance to calculate the ‘true’ perpendicular distance in one
experiment. However, this calculation is equally dependent on the angle of the target from the
line. Our comparison of perpendicular to radial estimates was not completely controlled in that
the boat may have drifted off course between the time that the observer estimating perpendicular
distance recorded her/his estimate and the observer with the compass obtained two bearings.
This did occur, especially early on in the experiment, but the extent to which this may have
affected results is unknown. In addition, the compasses on occasion gave inconsistent readings
from the same point. This was not quantified during the experiment, but subsequent tests of the
compass showed a consistent 2-30 difference in repeated readings. Differences (‘error’) of 1-20
would not have a great influence on the perpendicular distance, but larger errors, particularly if
the boat moved between one method and the other, could explain differences in direct vs.
calculated perpendicular estimates (Figure 8). An improved design would be to obtain a laser-measured
perpendicular distance to use as the reference distance from the line, eliminating the
angle calculation in establishing the ‘true’ distance.
Direct Perpendicular vs. Radial Distance-based Density Estimates – Experiment I
Methods.--We established random line transects through an array of stationary targets (small,
anchored buoys). When a target was selected, the boat was slowed or stopped. One observer
directly estimated perpendicular distance from the transect line, a second observer estimated the
radial distance and recorded compass bearings of the transect line and the target, and the third
observer measured the radial distance with a laser rangefinder at the same time and from the
same observation point as the observer estimating radial distance. We then computed three
perpendicular distances for each target: the direct estimate, the distance calculated from the
estimated radial distance and angle, and the distance calculated from the laser-measured radial
distance and angle (see above). Pooling across three observers, we compared density estimates
derived from the program DISTANCE (Buckland et al. 1993) for the three methods. For the
DISTANCE program, we established an arbitrary transect length of 10 km, used all targets as one
sample per method, and proportioned observations into 1 of 6 distance bands as described above.
We also compared methods under real survey conditions using two simultaneous
observers. One observer conducted a typical extensive survey, estimating perpendicular distance
to all murrelet and nonmurrelet targets. The second observer worked independently, estimating
radial distances and obtaining compass bearings for murrelet targets only. Unlike the experiment
with the stationary array, it was not intended that observers recorded the same target at the same
time, or even observed the same targets. The second observer was positioned behind the first
observer to minimize the chances of cuing each other to birds. We conducted this experiment on
14 transects ranging from 2-36 km in length. We compared numbers of murrelets detected per
transect with paired t-tests, and compared density estimates (from program DISTANCE) derived
from pooling across transects.
Results.--- Density estimates from the perpendicular distances derived from the three methods
used in the stationary array were based on 143 targets pooled across observers. The density
based on direct estimates of radial distance was highest, the density based on laser-measured
radial distance was slightly lower, and the density from the direct estimate of perpendicular
distance was lowest (Figure 9).
Under real survey situations, the observer estimating radial distances and taking compass
bearings recorded fewer murrelets than the observer estimating perpendicular distance. The
difference was not significant across all transects, averaging 12 fewer murrelets per transect (t =
1.467, p = 0.163). However, looking at transects individually suggested that the differences
occurred when the encounter rate of murrelets was higher and sustained, exceeding 20 birds/km
for a number of km (Table 3). Despite differences in the number of murrelets detected, density
estimates were similar between the two methods (Figure 10). The numbers of murrelets and
murrelet groups (not weighted by number in the group) recorded within 50 m of the transect line
were similar between methods (Figures 11 and 12). Observers estimating radial distances
recorded significantly fewer murrelets than expected at distances > 75 m from the transect line
(X2 = 6.0, p < 0.02 in each case; Figure 11), but these had little influence on density estimates.
Direct Radial Laser
0
50
100
150
200
Density (# / Sq. Km)
n = 143
Direct Perpendicular
Distance
Radial
Distance
0
10
20
30
40
Density (# / Sq. Km)
n = 1304 n = 1159
6 distance bands (0-200 m)
0
10
20
30
40
Direct Perpendicular
Distance
Radial
Distance
n = 1146 n = 1089
5 distance bands (0-150 m)
Figure 9. Density estimates (from program DISTANCE) and 95% confidence intervals from
three methods of obtaining perpendicular distances: direct estimates, calculated from estimates
of radial distance and measured angles, and calculated from laser-measured radial distance and
measured angles.
Figure 10. Comparison of density estimates based on perpendicular and radial distance
estimates obtained under real survey conditions.
Table 3. Comparison by transect of the number of murrelets detected using perpendicular
distance method and the radial distance and angle method. Table is organized by increasing
encounter rate (no. murrelets/km).
Perpendicular Method Radial Method
Date Transect Km Total MM No./km Total MM No./km
8/27/98 ORES 36 25 0.7 24 0.7
8/20/98 LOSE 8 7 0.9 5 0.6
8/20/98 DECA 22 49 2.2 46 2.1
8/24/98 WALD 16 48 3.0 46 2.9
8/21/98 LOSO 10 33 3.3 34 3.4
8/24/98 ORWE 14 55 3.9 51 3.6
7/28/98 LOSO 10 54 5.4 52 5.2
8/27/98 CYPR 22 118 5.4 110 5.0
8/25/98 ORSW 4 22 5.5 24 6.0
8/27/98 CONE 2 11 5.5 12 6.0
8/25/98 WASP 6 47 7.8 45 7.5
8/25/98 JONE 6 49 8.2 51 8.5
8/21/98 LOSW 14 214 15.3 218 15.6
8/10/98 SJNO 10 239a 23.9 160a 16.0
8/24/98 SJNO 12 339a 28.2 230a 19.2
8/25/98 CRAN 4 147b 36.7 157b 39.2
a Greatest differences in number of murrelets between methods. Encounter rate > 20 birds/km.
b High encounter rate, but numbers of birds similar. Length of transect short.
25 50 75 100 150 200
Distance Bands (m)
0
100
200
300
400
Number of Murrelets
Direct
Radial
16 transects
X 2 = 49.5, p < 0.001
*
*
25 50 75 100 150 200
Distance Bands (m)
0
50
100
150
200
Number of Murrelet Groups
Direct
Radial
16 transects
X 2 = 7.9, p > 0.10
Figure 11. Distribution of murrelets from the transect line as detected by observers conducting
simultaneous but independent surveys using two methods of obtaining perpendicular distance.
Significant differences noted with ‘*’.
Figure 12. Distribution of murrelet groups (not weighted by the number in a group) from the
transect line as detected by observers conducting simultaneous but independent surveys using
two methods of obtaining perpendicular distance.
Discussion.— Based on density estimates alone, the results from the experiment conducted under
real survey conditions suggest that it makes little difference if observers estimate perpendicular
distances or estimate radial distances and take compass bearings. However, if the actual number
of murrelets recorded is critical, the results suggest that using the compass to obtain two bearings
per target may cause the observer to miss birds in high-density areas. (The differences can’t
really be explained by observer variability.) The observer with the compass should have had an
advantage in that she/he was only recording murrelets, while the other observer was recording all
species. This should be considered in terms of the additional information (other than density
estimates) that is obtained from survey data, such as the number of adults and juveniles to
compute age ratios. It may be proposed that ‘interference’ of the compass could be minimized
by estimating angles as well as distances. This has two weaknesses. First, the error associated
with estimates of any kind would be compounded. One reason that the radial method has been
supported is that it may be easier to estimate a radial distance than a perpendicular distance
because a radial distance is equivalent to a direct line of sight. Combined with an angle that is
accurately measured, the calculated perpendicular distance could be more accurate. However, if
the accuracy of the angle is reduced, the benefit of using this method becomes questionable.
Secondly, the magnitude of errors in angle measures would likely increase if observers were
estimating rather than measuring them. What was a 1-20 error in the compass itself could
become much greater, with a greater error associated with the calculated perpendicular distance.
This would most affect targets closer to the transect line (extrapolate from Figure 8), which are
the more critical for density estimates with the program DISTANCE. The potential impact of
angle estimation error should be tested before estimates of angles are considered as part of the
survey methodology.
The density derived from direct perpendicular distance estimates was lower than the
density derived from a known radial distance (the experiment with a fixed array). As discussed
on pg. 10, this comparison was weakened by the use of a calculated perpendicular distance as a
reference point, rather than a measured perpendicular distance. This result also may reflect
observers’ tendency to overestimate perpendicular distances. This could have a subtle effect on
how numbers of murrelets are assigned to distance classes within the DISTANCE program,
affecting the density estimate. The density derived from radial distance estimates was higher
than that from the known radial distance, although not significantly so. From a conservation
standpoint, one might consider whether an underestimate or overestimate of density is more
prudent. Again, the degree to which these methods actually differ can’t be conclusively
determined until tests with improved controls are conducted.
Direct Perpendicular vs. Radial Distance-based Density Estimates – Experiment II
Methods.��We deployed a sample of 36 buoys (3.8 l plastic milk jugs) tied to 0.23-kg (8-oz)
weights in waters of southern Puget Sound where water depth averaged about 6 m. Buoys were
spaced at random intervals throughout a rectangular area approximately 0.5 km wide and 10 km
long. We used two boats to run transect lines through the array of buoys. Multiple passes were
made from various angles to obtain a larger sample of distance estimates. Each boat was staffed
by 3 people; two experienced observers made distance estimates and the other person piloted the
boat. Prior to collecting observations, a 100-m line with visible markers at 25-m intervals was
trailed behind each boat to help calibrate distance estimates. For a set of observations, usually
about 40, one observer directly estimated the perpendicular distance between a buoy and the
transect line. The other observer made an estimate of the radial distance from the boat to the
target and used a digital compass to record the azimuth of the transect and the azimuth to the
target. When the boat progressed to a point where the target was abeam (along a perpendicular
line from the transect) one observer (the one doing the direct perpendicular estimate) measured
the actual perpendicular distance to the target using a laser rangefinder.
After completing a set of
observations, the observers switched roles. This cycle was repeated over 1.5 days, 14-15 January
1999. Weather conditions varied from calm to breezy with overcast skies or light rain. Waves
were flat to moderate (sea state 0 to 1). For the radial estimates, perpendicular distance (x) was
calculated as x = sin (ÿ)*r where ÿ was the angular deviation from the transect line to the target
and r was the radial distance from the observer to the target. Angular deviation was calculated as
the difference in azimuths of the transect line and the target.
To compare direct and radial estimates of perpendicular distance with actual distance, we
computed a regression with actual distance as the independent variable and the estimate as
dependent variable. We also computed the difference between each estimate and the actual
distance and compared mean differences using matched-pair t-test. We also grouped the data
into distance intervals of 0-25, 25-50, 50-75, 75-100, and 100-125 m and used program
DISTANCE to compute detectability functions and density of the buoys for each method. We
assumed a total transect length of 100 km for density calculations. The density estimates were
pooled across observers and all observations for one method were entered into the program as a
single sample.
Results.–We obtained a total of 321observations; because of missing data in each method, the
total sample with complete data was 305. For the radial estimate, a regression of the estimate (y)
on the actual measured distance (x) was y = 0.80x + 4.71; R2 = 0.68; SE y|x = 10.27 (Figure 13).
For the direct estimate, the regression was y = 0.89x + 5.52; R2 = 0.78; SE y|x = 8.67 (Figure 14).
These regression results indicate that the direct estimate was slightly less biased at greater
distances than the radial estimate (slope was closer to 1.0) and that the direct estimate was more
precise (lower SE y|x, a measure of the variability of the estimates). However, the intercept was
greater for the direct estimate, indicating a tendency to overestimate distances to closer targets.
Results of matched-pair t-tests differed from the regression in that the mean difference between
radial and measured perpendicular distance did not differ from 0 (mean difference = 0.87 m, t =
1.40, df = 304, P = 0.164), whereas the mean difference did differ from 0 between the direct and
measured estimates (mean difference = 2.44 m, t = 4.80, df = 305, P = 0.000). Thus, this test
indicated that the direct estimate was more biased than the radial estimate but the amount of bias
was small relative to the precision of the distance estimates.
We computed the absolute value of differences between the measured distance and each of the
two estimates and computed the mean of these absolute differences. For the direct estimate, the
mean of the absolute difference was 6.7 m. For the radial estimate, the mean difference was 8.0
m, indicating, as in the regression results, that the direct estimates were more precise than the
radial estimates (matched pair t-test, t = 2.79, df = 304, P = 0.006). For the direct estimate, 28%
of our estimates were within 10% of the actual measured perpendicular distance compared to
21% for the radial estimates. We also computed the absolute difference between each estimate
and the actual distance expressed as a percent of the actual distance. The mean difference for the
direct estimate was 37% (median = 21%); mean difference for the radial estimate was 51%
(median = 25%). These results indicate that the direct estimate performed somewhat better than
the radial estimate, but both are relatively imprecise.
Results varied among the four observers (Figure 15). Numbers of observations for each of the 4
observers varied from 69 to 84. Using the radial method, 2 observers had mean differences
between their estimates and the actual distance that did not differ from 0; one observer
overestimated and one observer underestimated the true distance. Using the direct method, the
estimates of two observers also included 0, and two observers significantly underestimated the
true distance.
We grouped the perpendicular distances into four distance bands under each method (Figure 16)
and submitted these grouped data to program DISTANCE to compute detectability functions and
density estimates based on each of the three datasets (actual, direct, radial), we found little
difference between methods (Figure 17). The detectability function based on actual
perpendicular distances was intermediate between the higher radial-based function and the lower
direct-based function but confidence intervals overlapped among all estimates. Similarly, the
computed density estimate based on the actual distances was intermediate between the higher
density from the radial estimate and the lower density from the direct estimate and all confidence
intervals overlapped. These results parallel those from our other experiments (see above).
Discussion.–Although the direct estimates performed better in most of the tests we conducted (8
of 10 comparisons), the differences were not strong. We found that the direct estimate was
somewhat more precise than the radial method, but that both estimates were biased. Because the
differences were slight, we conclude that either method could be used for surveys using the
protocols we employed. However, if observers estimate azimuths, radial estimates will have
greater error. We used a digital compass to make relatively precise measurements; visual
estimates of azimuths, or estimates using angle boards marked at 5-degree intervals will lead to
greater error in the radial method.
Neither method is as accurate as we would prefer. As noted above, the average difference
between the actual perpendicular distance and the estimated distance was 8.0 m for the radial
estimate (51% error) and 6.7 m for the direct estimate (37% error). Only about a quarter of all
observations fell within plus or minus 10% of the actual distance under either method.
We are concerned about other problems using the radial method, as discussed in Experiment I.
Because more time is needed to collect radial distances and azimuths, there is a risk of missing
birds. Our results, for example, suggest that observers missed birds significant numbers of
murrelets using the radial method compared with the direct method (Table 3) on one transect
with high numbers of encounters. On transects with lower numbers of encounters, numbers of
birds were very similar for the two methods.
0 10 20 30 40 50 60 70 80 90 100
Actual Perpendicular Distance (m)
0
20
40
60
80
100
Radial Perpendicular Estimate (m)
0 10 20 30 40 50 60 70 80 90 100
Actual Perpendicular Distance (m)
0
20
40
60
80
100
Direct Perpendicular Estimate (m)
Figure 13. Relationship between actual perpendicular distance to fixed buoys (measured with
laser rangefinder) and distance estimated using radial distance and angle. Heavy line indicates
equal estimates (slope = 1.0); thinner line is regression between actual and radial estimates.
Figure 14. Relationship between actual perpendicular distance to fixed buoys (measured with
laser rangefinder) and direct estimate of perpendicular distance. Heavy line indicates equal
estimates (slope = 1.0); thinner line is regression between actual and direct estimates.
N = 78 69 84 75
Observer
BM DE JH RW
Deviation from Actual (m)
2
0
-2
-4
-6
-8
Direct Estimate
Radial Estimate
Figure 15. Variation among observers in difference between estimated and actual perpendicular
distances to buoys using direct (top) and radial (bottom) estimation technique. Values are mean
and 95% confidence intervals.
0 - 25 25 - 50 50 - 75 75 - 100
Distance Band (m)
0
50
100
150
200
Number of Targets
Actual
Direct
Radial
Direct Radial Actual
Method
0.014
0.016
0.018
0.020
0.022
0.024
0.026
0.028
f(0)
Detectability Function
Direct Radial Actual
Method
20
25
30
35
40
45
Density (n/sq. km)
Density
Figure 16.
Number of buoys observed in four perpendicular distance classes using three techniques. Actual
= perpendicular distance from the boat to a buoy measured using a laser rangefinder; Direct =
perpendicular distance between the projected transect line and a bouy as visually estimated by an
observer; Radial = perpendicular distance calculated from an observer’s estimate of radial
distance to the buoy and the angle between the transect line and the buoy. These grouped
distance data were used as input to program DISTANCE to compute detectability functions and
density estimates for each method.
Figure 17. Detectability and density of buoys from 3 measures of perpendicular distance. Direct
= direct estimate of perpendicular distance between target and transect line; radial =
perpendicular distance calculated from estimate of radial distance between observer and target
and angle between transect line and target; Actual = perpendicular distance between observer and
target as measured with a laser rangefinder. Detectability and density were calculated using
program distance assuming 100 km of transect. Sample size was 305 targets in each case.
Observer Variability in Identifying Adult and Juvenile Marbled Murrelets
Methods.----We compared the classification of marbled murrelets by three simultaneous
observers on n = 102 murrelet groups under typical survey conditions (see above) over several
weeks during July and August 1998 to assess variability. Observers looked at the same groups
at the same time and recorded each murrelet as 1 of 11 possible age classes adapted from Strong
1995. Each of 5 classes could be recorded as definite or probable, with the 11th class being
‘unknown’. Comparisons were made of the total numbers per class and of the juvenile:adult
ratios calculated from the data. Adults were defined as definite class 1+2+ 3 + probable class
1+2+3. Juveniles were defined as definite + probable class 5. Unknowns were class 11. Class 4
birds were not included in calculations, as by definition these are birds in basic plumage that
could be fully-molted adults or juveniles. In this experiment, no birds were classified as class 4
because questionable birds were examined until a determination of adult or juvenile could be
made, or they were classed as unknown.
Results.—There were differences in how observers classed murrelets, but none of these affected
the total number of adults recorded (Table 4). Total number of adults differed by < 3%. The
sample of juveniles was much smaller, where differences of 1-2 birds could be important.
Observer 1 recorded the fewest juveniles and the most adults, yielding the lowest juvenile:adult
ratio. However, all of the ratios were well below the level needed for population stability if
survivorship is assumed to be 0.85 (Beissinger 1995). Thus, all three ratios would have led to
the same conclusion about population status.
Discussion.—The variability we observed was generally minimal, although one observer may
have misidentified 1-2 juveniles as adults. One would expect that if confusion was to occur, it
would be with advanced-molt birds. In this sample, all three observers had very similar numbers
of definite class 3 birds and no probable class 3's, so it appears that this isn’t the source of the
difference in numbers of juveniles. Variability in classification among observers likely could be
reduced with more orientation to plumage classes, more training by actual observations, and with
better understanding of when to classify a bird as definite, probable, or unknown.
Table 4. Classification of marbled murrelets by three simultaneous observers.
OB1 OB2 OB3
Class 1 - Definite 152 133 141
Class 1 - Probable 4 10 0
Class 2 - Definite 38 45 46
Class 2 - Probable 1 0 9
Class 3 - Definite 21 22 19
Class 3 - Probable 0 0 0
Total Adult 216 210 215
Class 4 - Definite 0 0 0
Class 4 - Probable 0 0 0
Class 5 - Definite 7 10 9
Class 5 - Probable 0 0 0
Total Juvenile 7 10 9
Unknown 6 13 6
Juv:Ad Ratio 0.032 0.048 0.042
Differences in Spatial Distribution and Detectability of Adult and Juvenile Murrelets
Methods.—To assess detectability, we compiled data from 16 line transects totaling 170 km that
were repeated every 10 days from early June - late August in 1997 and 1998. Transects were 300
m from shore. We compared the probability density function from the program DISTANCE of
adults to that of juveniles for the 10-day intervals in which juveniles were present. Density
functions were based on direct estimates of perpendicular distance from the transect line.
Distances were assigned to one of six bands: 0-25 m from the line, 0-25 m, 25-50 m, 50-75 m,
75-100 m, 100-150 m, and 150-200 m.
To compare distributions, we combined extensive and intensive survey data from the SJI
in 1997 and 1998. We computed the actual distance from shore of all known-age murrelets (n =
6724) based on their distance from the transect line, the transects’s distance from shore, and the
direction of the boat for each transect. All murrelets had been identified as adult, juvenile, or
unknown in 1997. For the 1998 data, where murrelets were classified as 1 of 11 classes, we
included definite and probable class 1, 2 and 3 murrelets as adults and definite and probable class
5 birds as juveniles. We compared seasonal distribution from shore between years across 6 time
intervals from early June - late August. We also investigated the distribution within the SJI by
comparing the patterns of peak and low numbers of adults and juveniles at 8 locations across 6
seasons in both years.
Results.—Adults and juveniles were similarly distributed across distance bands from the transect
line (Figure 18). Both distributions followed the expected pattern of more birds detected closer
to line and fewer as distance from the line increased. As a result, probability density functions
were similar (Figure 19), suggesting no difference in the ability to detect adults compared with
juveniles based on their distribution around line transects run at 300 m from shore.
Based on actual distance from shore, the distribution of juveniles was skewed slightly
closer to shore compared with adults, although the mean difference was not statistically
significant. In 1997, juveniles moved closer to shore as the season progressed from late July to
late August, whereas adult distribution remained stable (Figure 20). Mean distance from shore
for juveniles exceeded 300 m in late July, and decreased to < 250 m by late August. This pattern
was repeated somewhat in 1998, but the decrease in mean distance from shore occurred earlier in
the summer and did not continue to decrease through the last season. Mean distance from shore
of adults generally exceeded 300 m, and exceeded 350 m during late June 1997 (Figure 20).
There were no obvious differences in the patterns of adult and juveniles at 8 locations
around the SJI during 1997 (Figure 21). In general, places that showed increases in adults as the
summer progressed also had higher numbers of juveniles, and vice versa. In 1998, 4 of the 8
locations showed contrasting patterns. At 2 sites (SJSW and LOSO), juveniles increased as the
number of adults decreased. At two other sites (LOSW and WASP), juveniles decreased as the
number of adults increased (Figure 21).
25 50 75 100 150 200
0
100
200
300
400
500
600
700
NUMBER (N)
AHY
25 50 75 100 150 200
0
10
20
30
40
50
60
70
HY
DISTANCE CUT POINT (m)
0
0.01
0.02
0.03
0.04
HY
AHY
Survey Intervals 5-9 only in 1997
0
0.005
0.01
0.015
0.02
Survey Intervals 3-8 only in 1998
AHY
HY
Figure 18. Distribution of adult (AHY) and juvenile (HY) marbled murrelets from the transect
line across 16 transects surveyed in each of five time intervals (intervals 5-9) during mid July -
late August 1997. Distributions similar for 1998.
Figure 19. Probability density functions [f(0)] for adult (AHY) and juvenile (HY) marbled
murrelets during July-August 1997 and 1998.
EarlyJune
Late June
Early July
Late July
Early Aug
Late Aug
0
100
200
300
400
Mean Distance From Shore (m)
1997
189
310
119
350 855
1110
24
72
52
Early June
Late June
Early July
Late July
Early Aug
Late Aug
0
100
200
300
400
Mean Distance From Shore (m)
1998
114
453
652
1003
1357 868
22
42
32
Adult
Juv
Figure 20. Distribution from shore of adult and juvenile marbled murrelets by season and year. Numbers above bars are numbers of
murrelets.
WALD
SJSW
SJNO
ORSE
LOSW
LOSO
DECA
WASP
COMPLEX
Seasonal Distribution of
Adult and Juvenile Murrelets
Adult
1997
Adult
1998
Juvenile
1997
Juvenile
1998
Figure 21. Distribution of numbers of adult and juvenile marbled murrelets within and between
seasons at eight locations within the San Juan Island archipelago. Heights of bars are not
comparable between locations and between adults and juveniles. Only the patterns (peaks and
lows) are comparable for each location. Seasons = 1-15 June, 16-30 June, 1-15 July, 16-31 July,
1-15 Aug, and 16-31 Aug.
Effect of Transect Configuration on Density Estimates
Methods.--- We tested two configurations, a zig zag and rectangular transect, against a ‘straight
line’ configuration (uniform distance from shore) in the SJI in 1997 (Figure 22 -need). The
straight-line transects were a subset of our established core transects. They followed the contour
of the shoreline at a distance of 300 m from shore and were comprised of 2-km segments. We
used marine radar to maintain course at this uniform distance. The rectangular configuration
began as a core straight-line transect (i.e., on an established line 300 m from shore). The boat
traveled parallel to shore along the straight-line transect for 750 m, then turned 900 and headed
perpendicular from shore to a distance of 700 m (positioning the boat at 1000 m from shore). The
boat then turned to continue parallel to shore for a distance of 500 m, then headed straight back to
shore for a distance of 700 m back to the 300-m line. The boat then finished the 2-km segment by
traveling along the 300-m line for 750 m. Each subsequent 2-km segment was repeated with this
same design until the end of the transect had been reached. Distances for each jog were measured
with a GPS unit and the distance from shore (300 or 1000 m) confirmed with the radar. The zig
zag transect began at the same origin (landmark) as a straight line transect, but at a point 100 m
from shore. The line then proceeded at an arbitrary angle away from the shore until a distance of
1000 m from shore had been reached. The line then headed at another arbitrary angle back to
shore to a distance of 100 m from shore. This pattern was repeated until the ending point of the
core transect was reached. Each leg of the zig zag transect was recorded and stored with a GPS
unit so that the ‘route’ could be recalled from the GPS and repeated. Legs did not equate to 2-km
segments. The 100-m and 1000-m distances from shore were determined with marine radar.
For all three methods, observers estimated the perpendicular distance of all murrelet groups
up to 200 m either side of the transect line. In addition, the actual distance from shore of each
murrelet group was measured by radar for groups encountered when the boat was not traveling
parallel to shore at a known distance. A total of 10 transects were run with the three
configurations. Straight-line configurations were run first; the transect area was repeated with the
zig zag and rectangular pattern either immediately after (n = 4) or the next day (n = 6). The 10
transects were conducted during 3 time intervals: last week of July (n = 4), early-mid August (n =
2), and late August (n = 4). We compared the number of murrelets detected by transect by
method, the number detected per km by method and time interval, and the density estimates from
program DISTANCE by method for each interval across transects within the interval. For
DISTANCE, murrelet groups were assigned to 1 of 6 perpendicular distance classes (0-25 m, 25-
50 m, 50-75 m, 75-100 m, 100-150 m, and 150-200 m).
Results.—The zig zag configuration had the lowest number of total murrelets detected, the lowest
number per linear km, and the lowest density estimate of the three methods (Figures 23 and 24).
Rectangular configurations yielded the highest numbers of birds in 2 of the 3 intervals, but lower
numbers/km compared with the straight-line method. Differences in density estimates between
the straight line and rectangle may be a sample size problem. The straight-line density estimate
was much higher in the interval when only two transects were covered, and the rectangular
estimate was higher in the interval when more transects were covered. The zig zag method was
similar to the other two methods only in the first interval, when the number of murrelets detected
was very low.
Discussion.---The zig zag method spent less time sampling the area where murrelets are more
likely to be in the SJI, thus we detected fewer murrelet groups on transects of this configuration.
However, this may not be a problem from a monitoring standpoint, when the most appropriate
sampling approach may not be to target the highest-density areas (Tim Max and others, pers.
comm.). The zig zag method has the benefit of covering a range of distances from shore, and
may be a good way to cover large expanses of relatively shallow water where murrelets are more
evenly distributed away from shore (i.e., bays or coves). The rectangular configuration is more
similar to a typical extensive survey in that it spends most of its time running parallel to shore at
an ‘optimum’ distance, but also forays out to a greater distance as a spot check of the distribution
of murrelets further from shore. Our design did not incorporate a line heading toward shore,
because at 300 m from shore and with a transect width of 200 m either side of the line, we were
essentially sampling from 100 m from shore. The rectangular configuration may be prone to
double counting near the junctions of the perpendicular lines (leaving and returning to the line
parallel to shore), especially if the transect width is large (i.e., 200 m either side of the boat, as in
this case).
On more than half of the samples, the straight-line method was run a day before the
rectangular and zigzag methods. Even when all methods were run in succession on the same day,
the time lapsed from one method to the next is certainly enough for birds to move into or out of
the sampling area, especially if the transect is 8-10 km long. This lack of simultaneous
observations limits these comparisons to that extent. The only way to remove this confounding
factor under field conditions would be to run three boats at the same time.
Late July Early Aug Late Aug
0
50
100
150
200
250
300
Total Murrelets
Straight
Rectangle
ZigZag
Late July Early Aug Late Aug
0
5
10
15
Number of Murrelets / Km
Straight
Rectangle
ZigZag
Figure 23. Comparison of the total number of murrelets and the number per linear km detected on three transect configurations during
three time intervals in the San Juan Islands in 1997.
Straight
Rectangle
ZigZag Straight
Rectangle
ZigZag Straight
Rectangle
ZigZag
0
50
100
150
200
Density (#/ Sq. Km)
Late July Early Aug Late Aug
n = 4 transects n = 2 transects n = 4 transects
Figure 24. Marbled murrelet density estimates (program DISTANCE) for each of three transect configurations for three time intervals
in the San Juan Islands, 1997. Densities estimated using transects within an interval as the sample.
Literature Cited
Beissinger, S.R. 1995. Population trends of the marbled murrelet projected from demographic
analyses. P. 385-393 In Ralph, C.J., G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, eds. Ecology
and conservation of the marbled murrelet. Gen. Tech. Rep. PSW-152. Albany, CA: U.S.
Department of Agriculture, Forest Service, Pacific Southwest Research Station.
Buckland, S.T., D.R. Anderson, K.P. Burnham, and J.L. Laake. 1993. Distance sampling:
estimating abundance of biological populations. New York: Chapman and Hall.
Cooke, F. 1998. Marbled murrelet studies at Desolation Sound, British Columbia. Abstract. P.
33 In Pacific Seabird Group 25th Annual Meeting, 21-25 January 1998, Monterey, CA.
Hamer, T.E., and S.K. Nelson. 1995. Characteristics of marbled murrelet nest trees and nesting
stands. P. 69-82 In Ralph, C.J., G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, eds. Ecology and
conservation of the marbled murrelet. Gen. Tech. Rep. PSW-152. Albany, CA: U.S. Department
of Agriculture, Forest Service, Pacific Southwest Research Station.
Madsen, S., D. Evans, T. Hamer, P. Hensen, S. Miller, S. K. Nelson, D. Roby, and M. Stapanian.
1997. Marbled murrelet effectiveness monitoring plan for the Northwest forest plan. Interim
report to the USDA/USDI IAC and REIC. 55 pp.
Nelson, S. K., and T.E. Hamer. 1995. Nesting biology and behavior of the marbled murrelet. P.
57-67 In Ralph, C.J., G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, eds. Ecology and conservation
of the marbled murrelet. Gen. Tech. Rep. PSW-152. Albany, CA: U.S. Department of
Agriculture, Forest Service, Pacific Southwest Research Station.
Raphael, M.G. and D.M. Evans. 1997. An evaluation of alternative techniques for estimating
productivity of the marbled murrelet, Puget Sound, Washington. Report to the U.S. Fish and
Wildlife Service, December 15, 1997.
U.S. Department of Agriculture and Department of the Interior (USDA/USDI). 1994. Record of
decision for amendments to Forest Service and Bureau of Land Management planning documents
within the range of the Northern Spotted Owl. Portland, OR: Forest Service, Bureau of Land
Management.
U.S. Department of the Interior, Fish and Wildlife Service. 1997. Recovery plan for the marbled
murrelet (Brachyrhamphus marmoratus) in Washington, Oregon, and California. Portland, OR.