ArticlePDF Available

The Effect of Animal Movement on Line Transect Estimates of Abundance


Abstract and Figures

Line transect sampling is a distance sampling method for estimating the abundance of wild animal populations. One key assumption of this method is that all animals are detected at their initial location. Animal movement independent of the transect and observer can thus cause substantial bias. We present an analytic expression for this bias when detection within the transect is certain (strip transect sampling) and use simulation to quantify bias when detection falls off with distance from the line (line transect sampling). We also explore the non-linear relationship between bias, detection, and animal movement by varying detectability and movement type. We consider animals that move in randomly orientated straight lines, which provides an upper bound on bias, and animals that are constrained to a home range of random radius. We find that bias is reduced when animal movement is constrained, and bias is considerably smaller in line transect sampling than strip transect sampling provided that mean animal speed is less than observer speed. By contrast, when mean animal speed exceeds observer speed the bias in line transect sampling becomes comparable with, and may exceed, that of strip transect sampling. Bias from independent animal movement is reduced by the observer searching further perpendicular to the transect, searching a shorter distance ahead and by ignoring animals that may overtake the observer from behind. However, when animals move in response to the observer, the standard practice of searching further ahead should continue as the bias from responsive movement is often greater than that from independent movement.
Content may be subject to copyright.
The Effect of Animal Movement on Line
Transect Estimates of Abundance
Richard Glennie*, Stephen T. Buckland, Len Thomas
Centre for Research into Ecological and Environmental Modelling, University of St Andrews, St Andrews, UK
Line transect sampling is a distance sampling method for estimating the abundance of wild
animal populations. One key assumption of this method is that all animals are detected at
their initial location. Animal movement independent of the transect and observer can thus
cause substantial bias. We present an analytic expression for this bias when detection with-
in the transect is certain (strip transect sampling) and use simulation to quantify bias when
detection falls off with distance from the line (line transect sampling). We also explore the
non-linear relationship between bias, detection, and animal movement by varying detect-
ability and movement type. We consider animals that move in randomly orientated straight
lines, which provides an upper bound on bias, and animals that are constrained to a home
range of random radius. We find that bias is reduced when animal movement is constrained,
and bias is considerably smaller in line transect sampling than strip transect sampling pro-
vided that mean animal speed is less than observer speed. By contrast, when mean animal
speed exceeds observer speed the bias in line transect sampling becomes comparable
with, and may exceed, that of strip transect sampling. Bias from independent animal move-
ment is reduced by the observer searching further perpendicular to the transect, searching
a shorter distance ahead and by ignoring animals that may overtake the observer from be-
hind. However, when animals move in response to the observer, the standard practice of
searching further ahead should continue as the bias from responsive movement is often
greater than that from independent movement.
Line transect sampling is a distance sampling method [1,2], widely used for estimating the
abundance of wild animal populations. Lines are placed randomly across a study region, or
more usually a systematic grid of lines is randomly positioned, and the observer traverses each
line, recording the perpendicular distance from the line to each detected animal. Typically, the
observer fails to detect all animals in the surveyed strip, and these distances are used to model
the probability of detection, and hence to estimate abundance.
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 1/15
Citation: Glennie R, Buckland ST, Thomas L (2015)
The Effect of Animal Movement on Line Transect
Estimates of Abundance. PLoS ONE 10(3):
e0121333. doi:10.1371/journal.pone.0121333
Academic Editor: Marco Festa-Bianchet, Université
de Sherbrooke, CANADA
Received: November 18, 2014
Accepted: February 10, 2015
Published: March 23, 2015
Copyright: © 2015 Glennie et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any
medium, provided the original author and source are
Data Availability Statement: All relevant data are
within the paper.
Funding: This work was supported by the University
of St Andrews (; RG,
STB, LT) and by a summer scholarship and PhD
grant from The Carnegie Trust for the Universities of
Scotland ( to RG. The
funders had no role in study design, data collection
and analysis, decision to publish, or preparation of
the manuscript.
Competing Interests: The authors have declared
that no competing interests exist.
The method relies on three key assumptions, and if these are not met then estimates of
abundance can have substantial bias [1] (pp. 2937). These are:
1. animals on the line are detected with certainty;
2. measurements are exact;
3. animals are detected at their initial location.
When the first assumption is not met, double-observer (or multi-observer) methods, which
combine distance sampling with mark-recapture, may be used [3,4]. Similarly, if the second as-
sumption is violated, measurement error models [5,6] can be adopted. We concentrate on vio-
lations of the third assumption.
Conceptually, line transect sampling is a snapshot method [1](pp.31),thatis,weconsiderthat
counters between a moving observer and immobile animals [7]. There are models [8,9]foren-
counters between mobile animal populations and observers, but these are limited in use: they rely
upon quantities that are difficult to determine (mean animal speed, encounter radius) and they as-
sume an ideal free gas movement model which can be unrealistic [10]. Making the third assump-
tion avoids such problems. This assumption can be violated in two ways: animals can move in
response to the observer, or move independently of the observer. Responsive movement is a com-
mon problem in distance sampling surveys and the bias it causes can be reduced by modelling the
movement [11,12] or using double-observer methods [13]. In a few specific cases, independent
movement has been modelled (for fish: [14]; for seabirds: [15]) to reduce bias; however, these
methods are ad hoc and specific to their application. We do not explore responsive movement
here; instead, we consider how animal movement independent of the observer affects bias. For
whale surveys, it was concluded that such movement may seriously bias the estimate of mean den-
sity if the speed of the whales approaches that of the observer, if the encounter region is long rather
than wide, or if the probability of sighting a whale is not strongly dependent on the time it spends
within that region [16]. Subsequently, simulation revealed that bias was negligible if mean animal
speed was one quarter of that of the observer, but not if animal speed was one half that of the ob-
server [17]. This conclusion has now been adopted within distance sampling [1] (pp. 131, 173)
and is used to determine if independent animal movement is a problem in particular studies [18].
This is despite these results being pertinent only to the single simulation scenario considered.
In this paper, we quantify the bias caused by independent animal movement in some exam-
ple cases and discuss the non-linear relationship between movement, detection and bias. We
first consider strip transect sampling which is a special case of line transect sampling in which
all animals within a strip are assumed to be detected [1] (pp. 23). We derive an analytic ex-
pression for the bias caused when animals move in straight lines, each with the same constant
speed but an individual random direction, within a rectangular study area. We then generalize
to surveys in which detectability falls off with distance from the line, and for which animal
movement is more complex, using simulation to quantify bias. Finally, we discuss what these
results reveal about bias caused by animal movement independent of the observer.
Strip Transect Sampling
Consider estimating the abundance of an animal population using strip transect sampling
where the study area is rectangular of length aand breadth Lwith corners, in (x,y)-coordi-
nates, at (a/2, 0), (a/2, 0), (a/2, L), (a/2, L). We assume the number of animals within the
study area is fixed, and so adopt a wrap-around model: for each animal that exits across one
boundary of the study area, a new animal enters immediately at the same distance along the
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 2/15
opposite boundary. Intuitively, we think of the study area as part of a wider region. We assume
the (strip) transect is placed randomly within this region and so may take the transect to be at
x= 0. Let the transect have half-width w, where 2w<a, extending out to either side of x=0
and length L, so it transverses the entire study region in the y-direction.
We assume animal itravels in a straight line at speed uand in a direction θ
where θ
is a
random deviate from a uniform distribution U(0, 2π](Fig. 1). Thus, the equations of motion of
the ith animal are given by:
yi;t¼yi;0þucos ðyiÞt
xi;t¼xi;0þusin ðyiÞt
where tis time and (x
) is the initial (t= 0) location of animal i.
The observer travels along the centerline, x= 0, at speed v>0, starting at y= 0. Thus, the
position of the observer at time tis given by (0, z
) where z
=vt. We define the encounter region
to be the area around the observer within which every animal is recorded. No animals are re-
corded outside the encounter region. Here, we assume the encounter region is rectangular, ex-
tending a distance weither side of the centerline and a distance l, which we call the lookahead
distance, in front of the observer (Fig. 1). The encounter region is wrapped using the wrap-
around model to fix its size. (There are practical methods of implementing a wrap-around de-
sign with strip transects, using buffer zones [19].)
When estimating abundance using strip transect sampling, it is assumed that the animals
are immobilethis is termed a snapshot method as the survey is assumed to take place instan-
taneously [1] (pp. 31). When true, the expected proportion of animals within the strip is
a. However, if animals are mobile then they can move into the encounter region and so
be recorded; hence, we expect to record more animals, which causes positive bias in
abundance estimates.
Animals not initially inside the encounter region can enter in three ways: from in front,
from the side and from behind. We shall consider each in turn. First, bias is not caused by ani-
mals travelling toward the observer and entering the encounter region from in front since on
average in any interval of time an equal number of animals, that if immobile would be re-
corded, will move away from the observer and not be recorded. Second, bias is caused by ani-
mals entering the encounter region from the side. Any point along the strip is within the
encounter region of length lahead of the observer for l
vunits of time. In that time, all animals
with direction θand speed utravel lusinðyÞ
vunits of distance perpendicular to the line. Hence the
proportion of all such animals not initially inside the encounter region entering from the side
and so being recorded is given by lujsinðyÞj
av . Given that θis uniform on (0, 2π], the expectation of
this proportion is
P2¼luE½j sin ðyÞj
av ¼2lu
as E½jsinðyÞj ¼ 2
Lastly, animals can only enter the encounter region from behind by overtaking the observer;
hence, no bias is caused when animals travel slower than the observer. Under our movement
model, only animals that are initially behind the transect, y<0, can overtake the observer and
enter the encounter region. By the equations of motion given above, an animal with direction θ
and initial location (x
) overtakes the observer at time
vucos ðyÞ
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 3/15
Thus, to have 0<T<L
vwhen y
<0 and have the animal recorded, we require that the follow-
ing inequalities are satised:
u<cos ðyÞ
vðucos ðyÞvÞy00
for some k2[0, T
] where Td¼l
ucosðyÞvis the time the animal is within the encounter region.
From these inequalities, we see that within the subset of those animals with direction θsuch
Fig 1. Strip transect with encounter region. Transect line (dashed grey line) with the strip extending a
distance wout to either side (solid grey lines). The observer and encounter region at time t1 (dashed black)
and t(solid black) are shown, together with an animal moving with speed uin direction θ
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 4/15
that cosðyÞ>v
u, the proportion that are recorded is given by
ð2wþujsin ðyÞjTdÞL
vðucos ðyÞvÞ
The expected value of this proportion with respect to θis
cos 1v
av 2w
ul 1v
av cos 1v
since EjsinðyÞj jcosðyÞ>v
uÞand EcosðyÞjcosðyÞ>v
. Finally, since the
proportion of animals with required direction θis given by PcosðyÞ>v
, the
total proportion of animals causing this bias is
pacos 1v
We are now able to give an expression for the total bias in the abundance estimate. If Pis
the total expected proportion of animals detected in the strip transect, then when uv,P=P
and when u>v,P=P
. Thus, strip transect sampling produces an abundance
estimate with percentage bias given by
pwv uv
pwv þffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pcos 1v
In particular, if movement is absent (u= 0) or animal speed is less than observer speed (uv)
and the observer only searches abeam, not ahead (l= 0), then there is no bias. Nevertheless, the
bias caused can be substantial. For example, if we assume that the observer searches as far
ahead as to the side, l=w, and the animals are slower than the observer, u<v, then bias is the
ratio of animal speed to observer speed, divided by π. Thus, if animal speed is one third observ-
er speed, bias is 1/(3π) = 10.6%.
Line Transect Sampling
In line transect sampling, some animals in the covered strip may be undetected. Movement in-
dependent of the observer causes bias in the estimated probability of detection, and this bias
combines with the bias arising from movement of animals into the encounter region, so that
the formulae of the previous section no longer apply.
The detection process can be modelled as either a continuous [20,21] or a discrete [20,22]
hazard-rate process. The discrete model is more appropriate for surveys of whales, when detec-
tion cue is primarily a whale blow, or for songbirds, if the cue is a songburst. Hiby adopted a
discrete hazard-rate model [17]. To avoid introducing another parameter (cue rate), we adopt
a continuous-time model.
By combining animal movement with a hazard-rate detection process, the problem of deriv-
ing an analytic expression for bias appears intractable as it is a non-linear function of the mean
probability of detection. Therefore, a simulation study was performed in order to quantify bias
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 5/15
and investigate the relationship between bias, movement and detection. We first describe the
simulation setup (Section 1) and then present and discuss the results as they relate to bias (Sec-
tion 2), the detection function (Section 2.1), lookahead distance (Section 2.2) and truncation
distance (Section 2.3).
1 Simulation Setup
We consider two animal movement types at a range of speeds relative to the observer. We also
vary the maximal distance the observer searches ahead (the lookahead distance) and the shape
of the detection function: both control the size of the encounter region around the observer.
The study area and transect are defined similarly to the strip transect case given above. The
study area was one square kilometre and a wrap-around model was used. The transect was one
kilometre long and bisected the study area. There were 100 animals distributed, for each simu-
lation, randomly across the study area.
Two types of animal movement were considered. First, the linear movement model, the
same as in the strip transect case, was used, where all animals moved in straight lines at the
same speed but each in a randomly assigned direction. It is particularly useful to quantify the
bias arising from this model: for any other random animal movement, each animal will travel
through a smaller range of perpendicular distances from the line compared with linear move-
ment, and so bias will be less as bias depends critically upon how far an animal travels from its
initial location. Thus, the bias arising from linear movement is an upper bound for the bias
caused by any other type of random animal movement. The second model considered is a
home range model: each animal is contained within a circular home range with a randomly as-
signed radius from a Gamma distribution with mean 20 and variance 11. Inside its home
range, the animal performed a correlated random walk, where the direction of travel was per-
turbed at each time step of the simulation by adding a random deviate from a wrapped normal
with zero mean and 0.1 variance. When an animal reached the boundary of its home range, it
was reflected back by changing its direction of travel by πradians. By restricting the range of
perpendicular distances from the line the animal may travel through, we expect the home
range model to produce smaller bias than the linear movement model; clearly, however, this
will depend critically on the radius of the home ranges. If home range radius greatly exceeds
the strip half-width w, then the constraints to movement would have little effect on the bias. If,
however, the radius is smaller than w, as might occur with many songbird territories, then the
constraints on movement should result in appreciably lower bias. For each movement model
we consider animal speeds from 0 m s
to 20 m s
where all animals have the same speed. If
each animals speed were to vary around a population mean, we expect bias would be similar
(as indicated by some preliminary simulation) since bias depends strongly on the mean speed
of the population rather than on an individuals instantaneous speed.
The observer travels along the line at 10 m s
and the probability of detecting an animal is
modelled as a continuous hazard-rate process. However, as simulation studies are, by nature,
discrete we integrate this hazard over the time step used in the simulation to give the probabili-
ty of being detected within that time step. Thus, the probability of being detected in time step t
with the continuous radial hazard k(r)=cr
where c>0, d>2is
gtðr0;r1Þ¼1exp xt
where x
is the perpendicular distance of the animal from the line at time t,r
is the radial
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 6/15
distance of the animal from the observer at the beginning of the time step, r
is the radial dis-
tance at the end of the time step, b=d1, sb¼cGð0:5bÞGð0:5Þ
2Gð0:5dÞand I
is the cumulative distribution
function of a Beta random variable evaluated at rwith parameters (b, 0.5). Here, we use a time
step of 0.05s duration. We will also consider different sizes of the encounter region by changing
the shape parameter (b= 2, 3, 5) and the lookahead distance (l= 0, 5, 10, 15, 20, 25, 30). Finally,
we will investigate the effect that truncating the data at the analysis stage has on bias by trun-
cating at distances where the true detection function appoximately equals 0.05, 0.1, 0.2, 0.4, 0.6
and 0.8.
For each animal speed and movement model, we simulated one thousand surveys consisting
of five hundred transects each. We did not consider any model selection: we fit a hazard-rate
model only. The simulated data were analysed using the Multiple-Covariate Distance Sampling
(MCDS) engine that is included in the software Distance [23]. The computer code used to per-
form this simulation study is included as online supporting material (S1 Code).
2 Simulation Results
Fig. 2 gives the bias estimated from the simulation study (with hazard-rate shape parameter
b= 2) for each relative animal speed and movement model. The bias caused in strip transect
sampling, calculated from the analytic expression in Section 1, with the observer searching as
Fig 2. Percentage bias in the abundance estimate for strip and line transects. Bias for strip transect
sampling (solid line) calculated from the analytic expression given and for line transects (hazard-rate b=2)by
averaging over 1000 simulations for linear movement (dashed line) and home range movement (dotted line)
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 7/15
far ahead as to the side, is also included in the figure for comparison. The strip transect bias in-
creases linearly when animal speed is less than observer speed, and then increases nonlinearly
and rapidly thereafter. Bias in line transect sampling under both movement models behaves
similarly as animal speed increases, but the bias at each speed is less than in strip transects. For
strip transect sampling, bias reaches nearly 10% when animal speed is just 30% of observer
speed, while for our simulation of line transect sampling under linear animal movement, bias
does not reach 10% until animal speed is 80% of observer speed. However, for this scenario, at
very large animal speeds (150% 200% observer speed) the bias in line transect sampling is
similar to or exceeds that in strip transect sampling.
We now consider these results. First, the difference in bias between linear movement and
home range movement confirms that bias is related closely with the range of perpendicular dis-
tances from the line that an animal can travel through. At smaller speeds, the animals move
through a similar range of perpendicular distances under each movement model which ex-
plains why bias is similar between the two. However, at larger speeds, animals under the linear
movement model travel through a large range of perpendicular distances compared with the
home range model where animals are constrained to a local territory; hence, linear movement
causes more bias at these speeds. This shows that bias from animal movement is related to the
distance an animal of interest may travelthe smaller that distance, the less the bias. Second,
these results show that any difference in the bias in strip and line transect sampling is due to
the detection process as line transect sampling is a generalisation of strip transect sampling
where detection of animals is uncertain.
2.1 Detection Function. Fig. 3 shows that as speed increases, the estimated detection
probability falls off more rapidly and the shoulder (range of distances from the line for which
probability of detecting an animal is one) narrows. This can also be seen in the distribution of
observed distances in Fig. 4. The effect on bias of changing the shape parameter of the detection
function is shown in Fig. 5. Here, we discuss how the bias in the estimated detection function
occurs, how it relates to bias in abundance estimates, and how the shape parameter of the func-
tion affects bias.
Bias in the estimated detection function is a result of animal movement and the shape of the
true detection function. As speed increases, the number of detections close to the line increases
rapidly since more animals are able to travel very close to the line during the survey and so be
detected. The number of detections at distances further from the line does not increase as rap-
idly since some of the animals travelling toward the line that are detected replace those that re-
main undetected by travelling away from the line. This does not occur very near to the line
since there the probability of detection is close to one and so the animals travelling both toward
and away from the line are detected. The rapid fall of detections as distance from the line in-
creases is the reason the detection function falls more rapidly at greater speeds. This bias in the
estimated detection function generates positive bias in the abundance estimate as the probabili-
ty of detection is underestimated.
Nevertheless, bias in line transect estimates of abundance is less than that for strip transects
when animal speed is less than 200% of observer speed. This is due to the detection process. In
strip transects, animals entering the strip are detected at the boundary at which they enter
since detection probability is one inside the strip and zero outside. In line transects, the animals
that would be detected at the boundary in strip transects are unlikely to be detected as they ap-
proach from a distance with small detection probability. Furthermore, as detection probability
falls off gradually from the line some of these animals, if detected, may replace those that
moved further from the line that would, if immobile, have been recorded. So, the reduction in
the number of detections by the detection process reduces bias by a greater amount than the
bias in the estimated detection function augments it when animal speeds are slow. Conversely,
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 8/15
at animal speeds close to 200% of observer speed, bias in line transect sampling becomes simi-
lar to that in strip transect sampling due to the biased estimated detection function. At these
speeds, the detection function falls off very rapidly and many animals are assumed to be unde-
tected resulting in very high bias in abundance estimates which may exceed the bias in strip
transect sampling.
Bias is also related to the shape of the detection function (Fig. 5). At slow animal speeds,
bias is similar for all shape parameters considered (b= 2, 3, 5) which is unsurprising as the bias
arises from too many detections close to the observer and all three functions here have similar
shapes in this region (they all approximately equal one). At animal speeds exceeding observer
speed, the hazard rate with shape parameter b= 5 consistently produces smaller bias than the
other two shape parameters (b= 2, 3). This may be due to the assumed detection function
being isotropic: animals overtaking the observer are less likely to be detected ahead of the ob-
server when b= 5 than b= 2, 3 since the shoulder of the detection function is narrower. How-
ever, here we not only change the shape of the detection function but also the probability of
detection and so the bias may be a result of both effects.
2.2 Lookahead Distance. Fig. 6 shows that bias, at each speed, increases toward a plateau
as lookahead distance increases for a fixed detection function. The bias reaches a plateau since
the probability of detection at a large lookahead distance is small and so further increasing
Fig 3. Estimated detection function for line transect sampling for increasing animal speeds. Estimated
detection function averaged over 1000 simulations for line transect sampling (hazard-rate b= 2) with the
linear animal movement model at animal speeds 0% (black solid line), 20% (red dashed line), 40% (blue
dotted line), 80% (green dotdash line), 150%(purple longdash line) and 200% (grey twodash line)
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 9/15
lookahead distance results in a smaller increase in detectionsand so bias increases by less.
Also we see for small animal speeds, the bias increases slowly with lookahead distance, while
for animal speeds that exceed observer speed, bias increases rapidly. As for strip transect sam-
pling, this rapid increase is due to animals overtaking the observer from behind. For larger loo-
kahead distances, such animals remain in the encounter region for longer and are thus more
likely to be detected. Non-zero bias when the observer searches abeam (lookahead distance is
zero) is another consequence of animal speed exceeding observer speed, when animals travel
slower than the observer no bias occurs.
2.3 Truncation Distance. When analysing line transect data, it is common to truncate the
recorded distances at some truncation distance to prevent outliers unduly influencing the
abundance estimate [1] (pp. 103108). We explored the effect of truncation distance on bias
with the linear animal movement model and b= 2 hazard-rate. We assumed that observers
were actively searching out to distances beyond the truncation distance, and recorded distances
of detected animals from the line when they were first detected. Thus, truncation was assumed
to be applied by the analyst, not in the field.
Results indicate little if any relationship between bias and truncation distance (Fig. 7). This
is unsurprising as bias is caused by too many detections near to the line which truncation does
not affect.
Fig 4. Frequency of detections at each perpendicular distance from the line. Estimated histogram of
recorded distances averaged over 1000 simulations for line transect sampling (hazard-rate b= 2) with linear
animal movement at animal speeds 0% (black solid), 20% (red dashed) 80%(green dotted) and 150% (purple
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 10 / 15
Animal movement independent of the observer generates bias in abundance estimates from
line transect surveys. The degree of bias depends both on animal movement and dectectability.
For a given mean animal speed, the analytical strip transect expression provides an upper
bound for this bias across all possible detection functions and movement types. The expression
shows bias to be reduced when the observer searches further to the side and a shorter distance
ahead (smaller lookahead). Also, the result shows the significant bias caused when animals
travel faster than the observer [1] (pp. 131). To reduce bias in strip transect sampling, observers
should record only those animals that were initially inside the strip and ignore those animals
that enter from the side or behind [24]. As line transects are often much wider than strip tran-
sects and detection of animals is uncertain [25], it is not possible to exclude animals that enter
the strip from the side, but often animals that overtake the observer can (and should)
be excluded.
Bias in line transect sampling for most animal speeds is considerably less than that of strip
transect sampling. The magnitude of the bias depends on the behaviour of the surveyed animal:
those that quickly travel long distances cause more bias than those that travel slowly or remain
within a small area [16]. Once again, survey procedure can reduce bias in line transect surveys
by searching further to the side and a shorter distance ahead; however, this may increase the
Fig 5. Percentage bias in abundance estimate for different shape parameter values. Percentage bias in
abundance estimate for linear movement against animal speed as a percentage of observer speed for shape
parameters b= 2 (black solid), b= 3 (blue dashed) and b= 5 (red dotted) with percentage bias for a strip
transect (grey solid) estimated from 1000 simulations included for comparison.
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 11 / 15
chances of failing to spot an animal before it moves in response to the observer [26] (pp. 129
131). If the animals to be surveyed tend to flee from the observer or be attracted, searching fur-
ther along the line will reduce the bias caused by this responsive movement [1] (pp. 32). Clear-
ly, this opposes our advice concerning independent animal movement. Often the bias caused
by independent animal movement will be small compared to that caused by movement in re-
sponse to the observer [27].
How to reduce bias from both independent and responsive movement will depend on the
type of survey and species. For aircraft surveys, animal speed is typically slow relative to observ-
er speed [28], and search effort is generally concentrated to the side rather than ahead. Thus in-
dependent animal movement is unlikely to cause bias. For shipboard surveys, observers tend to
concentrate more effort on searching ahead rather than to the side. This has the advantage that
animals are less likely to respond to the ship before detection, but greater bias will be generated
by independent animal movement. Therefore ship speed should be such that bias from inde-
pendent animal movement is low. For fast-moving animals, such as seabirds in flight, this may
not be an option; then methods for correcting for movement should be considered. For exam-
ple, seabirds on the sea might be surveyed, then a correction adopted based on the proportion
of time birds are in flight. If seabirds dont respond to the ship, then one strategy to eliminate
bias is to simply record distances of detected birds when they pass abeam; if they do not pass
abeam while visible, then they are not recorded [1] (pp. 198203). Corrections have also been
Fig 6. Percentage bias in abundance estimate against distance observer looks ahead. Percentage bias
against distance observer looks ahead with hazard rate b= 2 and animal speeds 50% (red dotted), 100%
(blue dashed) and 150% (black solid) of observer speed
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 12 / 15
developed for seabirds in flight [15]. For terrestrial surveys of birds, surveying perched birds,
and separately estimating the proportion of time birds are in flight, can be effective [29]. For
surveys on foot of ground or arboreal animals, an assessment should be made of whether it is
feasible for observers to travel sufficiently fast without compromising other assumptions, to en-
sure bias from independent animal movement is low. If not, corrections for animal movement
should be considered.
Once the survey is completed there is, at present, no clear way to reduce bias in abundance
estimates using the standard methods. For example, truncating the data [1] (pp. 103108) has
no discernible effect on the bias.
Here we have shown that bias can reach up to 10% for animal speeds of 50% 80% of ob-
server speed. This is useful for providing an idea of the severity of violating the assumptions of
distance sampling. However, the quantities of bias presented here are dependent on the type of
movement and survey protocol assumed. The expression for bias in strip transect sampling as-
sumes a linear movement of animals, which is unrealistic and causes greater bias compared to
other movement types [10]. Similarly, within the line transect simulation study we have not
considered how variable animal speed or heterogeneous home range radius may affect bias.
Furthermore, we only considered how the shape of the detection function affected bias by vary-
ing the shape parameter, which not only determines the functions shape, but also the average
probability of detection; it may be more informative to describe the relationship between
Fig 7. Percentage bias in abundance estimate for different truncation distances. Percentage bias in
abundance estimate averaged over 1000 simulations for hazard-rate b= 2 against truncation distance
applied by the analyst for animal speeds 20% (red solid), 80% (green dashed) and 150% (purple dotdash)
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 13 / 15
detection function shape and bias for a fixed probability of detection. Also, in this study only
the hazard-rate detection function was fit to the simulated data, while in practice other detec-
tion functions may be chosen by model selection [1] (pp. 110), which may affect bias.
In conclusion, independent animal movement can cause substantial bias in abundance esti-
mates and its possible effects should be considered in line transect surveys. It arises from both
the detection process inherent in line transect sampling and from the movement characteristics
of the surveyed animal. There are, at present, no analytical methods to deal with this reality;
however, bias is reduced by searching further perpendicularly to the line, searching less ahead
and ignoring animals that overtake the observer.
Supporting Information
S1 Code. Computer Code for Simulation Study.
The manuscript was peer-reviewed by two helpful and insightful anonymous reviewers.
Author Contributions
Conceived and designed the experiments: RG STB LT. Performed the experiments: RG. Ana-
lyzed the data: RG STB LT. Contributed reagents/materials/analysis tools: RG STB LT. Wrote
the manuscript: RG STB LT.
1. Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers DL, Thomas L. Introduction to Distance
Sampling. Oxford University Press, Oxford. 2001.
2. Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers DL, Thomas L. Advanced Distance
Sampling. Oxford University Press, Oxford. 2004.
3. Borchers DL, Laake JL, Southwell C, Paxton CGM. Accommodating unmodeled heterogeneity in dou-
ble-observer distance sampling surveys. Biometrics. 2006; 62: 372378. doi: 10.1111/j.1541-0420.
2005.00493.x PMID: 16918901
4. Buckland ST, Laake JL, Borchers DL. Double-observer line transect methods: levels of independence.
Biometrics. 2010; 66: 169177 doi: 10.1111/j.1541-0420.2009.01239.x PMID: 19432793.
5. Marques TA. Predicting and correcting bias caused by measurement error in line transect sampling
using multiplicative error models. Biometrics. 2004; 60: 757763. doi: 10.1111/j.0006-341X.2004.
00226.x PMID: 15339299
6. Borchers DL, Marques TA, Gunnlaugsson T, Jupp PE. Estimating distance sampling detection func-
tions when distances are measured with errors. Journal of Agricultural, Biological, and Environmental
Statistics. 2010; 15: 346361 doi: 10.1007/s13253-010-0021-y
7. Gurarie E, Ovaskainen O. Towards a general formalization of encounter rates in ecology. Theoretical
Ecology. 2012; 6: 189202 doi: 10.1007/s12080-012-0170-4
8. Skellam JG. The mathematical foundations underlying the use of line transects in animal ecology. Bio-
metrics. 1958; 14(3): 385400. doi: 10.2307/2527881
9. Yapp WB. The theory of line transects. Bird Study. 1956; 3(2): 93104. doi: 10.1080/
10. Hutchinson JMC, Waser PM. Use, misuse and extensions of the ideal gasmodels of animal encoun-
ter. Biological Reviews. 2007; 82: 335359. doi: 10.1111/j.1469-185X.2007.00014.x PMID: 17624958
11. Palka DL, Hammond PS. Accounting for responsive movement in line transect estimates of abun-
dance. Canadian Journal of Fisheries, and Aquatic Sciences. 2001; 58(4): 777787. doi: 10.1139/f01-
12. Smith GEJ. Aspects of line transect sampling when the target population moves. Biometrics. 1979; 35
(1): 325329. doi: 10.2307/2529953
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 14 / 15
13. Buckland ST, Turnock BJ. A robust line transect method. Biometrics. 1992; 48: 901909. doi: 10.2307/
14. Watson RA, Carlos GM, Samoilys MA.Bias introduced by non-random movement of fish in visual tran-
sect surveys. Ecological Modelling. 1995; 77: 205214. doi: 10.1016/0304-3800(93)E0085-H
15. Spear L, Nadav N, Ainley D. Estimating absolute densities of flying seabirds using analyses of relative
movement. The Auk. 1992; 109(2): 359389. doi: 10.2307/4088211
16. Hiby AR. The effect of random whale movement on density estimates obtained from whale sighting sur-
veys. Report of the International Whaling Commission. 1982; 32: 791793.
17. Hiby AR. Results of a hazard rate model relevant to experiments on the 1984/85 IDCR minke whale as-
sessment cruise. Report of the International Whaling Commission. 1986; 36: 497498.
18. Cañadas A, Hammond PS. Model-based abundance estimates for bottlenose dolphins off southern
Spain: implications for conservation and manangement. Journal of Cetacean Research and Manage-
ment. 2006; 8(1): 1327.
19. Strindberg S, Buckland ST, Thomas L. Design of distance sampling surveys and Geographic Informa-
tion Systems. In: Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers DL, Thomas L, editors.
Advanced Distance Sampling. Oxford University Press; 2004. pp. 190228.
20. Schweder T. Point process models for line transect experiments. In: Barba JR, Brodeau F, Romier G,
Van Cutsem B, editors. Recent Developments in Statistics. North-Holland Publishing Company;
1977. pp. 221242.
21. Hayes RJ, Buckland ST. Radial distance models for the line transect method. Biometrics. 1983; 39:
2942. doi: 10.2307/2530804
22. Schweder T. Independent observer experiments to estimate the detection function in line transect sur-
veys of whales. Report of the International Whaling Commission. 1990; 40: 349355.
23. Thomas L, Buckland ST, Rexstad EA, Laake JL, Strindberg S, Hedley SL, et al. Distance Software:de-
sign and analysis of distance sampling surveys for estimating population size. Journal of Applied Ecolo-
gy. 2010; 47(1): 514. doi: 10.1111/j.1365-2664.2009.01737.x PMID: 20383262
24. Granholm SL. Bias in density estimations due to movement of birds. Condor. 1983: 243248.
25. Burnham KP, Anderson DR, Laake JL. Efficiency and bias in strip and line transect sampling. The Jour-
nal of Wildlife Management. 1985; 49: 10121018. doi: 10.2307/3801387
26. Burnham KP, Anderson DR, Laake JL. Estimation of density from line transect sampling of biological
populations. Wildlife monographs. 1980: 3202.
27. Turnock BJ, Quinn TJ. The effect of responsive movement on abundance estimation using line transect
sampling. Biometrics. 1991; 47: 701715. doi: 10.2307/2532156
28. Certain G, Bretagnolle V. Monitoring seabirds populations in marine ecosystems: the use of strip-tran-
sect aerial surveys. Remote sensing of environment. 2008; 112: 33143322. doi: 10.1016/j.rse.2008.
29. Marsden SJ. Estimation of parrot and hornbill densities using a point count distance sampling method.
Ibis. 1999; 141: 377390.
The Effect of Animal Movement on Line Transect Estimates of Abundance
PLOS ONE | DOI:10.1371/journal.pone.0121333 March 23, 2015 15 / 15
... Other relevant problems when studying fish populations in farming ponds are not directly derived from hydroacoustics, but they are inherent to the sampling of wild animal populations in their natural habitats. Several authors have highlighted that the sampling has a direct or indirect effect on fish detectability and, ultimately, on its abundance or biomass estimation (Brehmer et al., 2004;Ona et al., 2007;MacNeil et al., 2008;Guillard et al., 2010;Kulbicki et al., 2010;Bozec et al., 2011;Kaartvedt et al., 2012;Glennie et al., 2015;Prato et al., 2017;Pais and Cabral, 2018;Brehmer et al., 2019). Specifically, two types of bias can occur in acoustic explorations: 1) positive bias, when fish move toward the transducer which leads to an overestimation of abundance and/or biomass; 2) negative bias, when fish move in the opposite direction of the transducer (avoidance), which leads to an underestimation of density and biomass. ...
... There is extensive literature confirming that the use of methods which incorporate the detection function of individual fish in the sampling (P) can significantly increase the reliability of the results obtained in fish population studies (Buckland et al., 2001;Williams et al., 2002;Draštik and Kubečka, 2005;Hjellvik et al., 2008;Marques et al., 2010;Cox et al., 2011;Martella et al., 2012;Marques et al., 2013;Glennie et al., 2015;Wheeland and Rose, 2015;Pais and Cabral, 2018). The detectability of populations can vary considerably depending on the circumstances, which leads to issues in the interpretation of the raw data obtained. ...
The use of hydroacoustics is currently being studied and developed as a promising non-intrusive methodology to monitor and manage fish stocks in aquaculture farms. The main objective of this study was to develop an acoustic method for the estimation of fish density and biomass in inland aquaculture farms and test the accuracy and precision of the estimates with real data provided by the company. The study was conducted in sea bass (Dicentrarchus labrax) production ponds located in Seville (Southern Spain). A Simrad EK60 echosounder with two split-beam circular transducers operating simultaneously at 200 kHz was used for hydroacoustic surveys. Two different hydroacoustic designs were considered: central trajectories and zigzag trajectories. The accuracy and precision of the estimates were examined in order to select the best sampling design. Due to a non-homogeneous fish distribution in the pond caused by the avoidance behaviour, as a response to the sampling disturbance presented by fish, acoustic density and biomass were corrected by applying sampling theory according to the probability of fish detection. When density and biomass were corrected, the estimates became highly accurate and precise with respect to real data, which confirms that the proposed method is adequate. Similarly, acoustic estimates of fish weight were highly in agreement with real data, due to the use of specific equations developed “in situ” for the study. Although no significant differences were recorded in the density and biomass estimates with regard to the trajectory used (central vs. zigzag), it was observed that the most accurate agreement and precision were always obtained in central trajectories. Therefore, central design is proposed as the most appropriate design for hydroacoustic measurements in inland ponds. The results obtained in this study provide estimates of density and biomass that accurately match the real data, supporting the use of hydroacoustics as a potentially valid tool to manage inland aquaculture farms.
... In good conditions schools are likely to be detected before any significant response to the vessel occurs. Movement of animals (independent of the vessel) will generate some upward bias in estimates (Glennie et al., 2015), and although this bias may not be negligible, incorporation of an explicit animal movement model into the distance sampling methodology allows for bias correction (Glennie et al., 2017). In addition, although it can be difficult to estimate group size and species proportions in mixed-species groups, as noted above, aerial photographs of a sample of schools are used to quantify and correct for bias. ...
Full-text available
A twelve-year hiatus in fishery-independent marine mammal surveys in the eastern tropical Pacific Ocean (ETP), combined with a mandate to monitor dolphin stock status under international agreements and the need for reliable stock status information to set dolphin bycatch limits in the tuna purse-seine fishery, has renewed debate about how best to assess and monitor ETP dolphin stock status. The high cost of replicating previous ship-based surveys has intensified this debate. In this review, transect methods for estimating animal abundance from dedicated research surveys are considered, with a focus on both contemporary and potential methods suitable for surveying large areas for dolphin species that can form large, multi-species aggregations. Covered in this review are potential improvements to the previous ship-based survey methodology, other ship-based methods, alternative approaches based on high-resolution imagery and passive acoustics, and combinations of ship-based and alternative approaches. It is concluded that for immediate management needs, ship-based surveys, with some suggested modifications to improve precision, are the only reliable option despite their high cost. However, it is recommended that a top research priority should be development of composite methods. Pilot studies on the use of high-resolution imagery and passive acoustics for development of indices of relative abundance to be used in composite methods should be part of any future ship-based survey efforts.
... In either case, the number of acoustic events detected is quantified, the distance to each detected event is estimated [either using time difference of arrival (TDOA) information or auxiliary methods such as simultaneous tag deployments, visual surveys, or propagation modeling], and the distribution of these distances is used to estimate the detection function. Distance sampling requires that animals be detected at their initial location (i.e., there is no animal movement) so that density estimates are not biased by animals moving into the survey area or by moving toward or away from the hydrophone (either randomly or in response to the survey platform) (Buckland et al., 2001;Glennie et al., 2015). For a vessel-based line-transect survey, the vessel is typically moving sufficiently quickly (faster than the animals) to ensure that animal movement relative to the ship is trivially small. ...
A single-hydrophone ocean glider was deployed within a cabled hydrophone array to demonstrate a framework for estimating population density of fin whales ( Balaenoptera physalus) from a passive acoustic glider. The array was used to estimate tracks of acoustically active whales. These tracks became detection trials to model the detection function for glider-recorded 360-s windows containing fin whale 20-Hz pulses using a generalized additive model. Detection probability was dependent on both horizontal distance and low-frequency glider flow noise. At the median 40-Hz spectral level of 97 dB re 1 μPa ² /Hz, detection probability was near one at horizontal distance zero with an effective detection radius of 17.1 km [coefficient of variation (CV) = 0.13]. Using estimates of acoustic availability and acoustically active group size from tagged and tracked fin whales, respectively, density of fin whales was estimated as 1.8 whales per 1000 km ² (CV = 0.55). A plot sampling density estimate for the same area and time, estimated from array data alone, was 1.3 whales per 1000 km ² (CV = 0.51). While the presented density estimates are from a small demonstration experiment and should be used with caution, the framework presented here advances our understanding of the potential use of gliders for cetacean density estimation.
... The first issue relates to survey effort and animal movement. Typically, survey effort is defined as line length implications for (1) how survey effort is defined in the density estimator and (2) whether animal movement may be problematic or not (Glennie, 2015). If a platform moves considerably slower than the study animals (or is stationary), any bias caused by animal movement can be avoided by either counting cues (e.g. ...
Technical Report
Full-text available
Literature review prepared to improve the understanding of how autonomous vehicles can be used to monitor marine animals, particularly during activities of the oil and gas industry.
... While we recommend surveying at low altitudes, we also encourage future users of aerial platforms to be aware of reactive movement induced by the aerial platform (Mulero-Pázmány et al., 2017). Such reactive movement can bias density estimates if not accounted for (Buckland et al., 2005;Glennie et al., 2015Glennie et al., , 2021. Perhaps, the easiest way to avoid double counting on neighboring transects is to space transects far enough apart to completely avoid this error (e.g. ...
Full-text available
Aerial vehicles equipped with infrared thermal sensors facilitate quick density estimates of wildlife, but detection error can arise from the thermal sensor and viewer of the infrared video. We reviewed published research to determine how commonly these sources of error have been assessed in studies using infrared video from aerial platforms to sample wildlife. The number of annual articles pertaining to aerial sampling using infrared thermography has increased drastically since 2018, but past studies inconsistently assessed sources of imperfect detection. We illustrate the importance of accounting for some of these types of error in a case study on white-tailed deer Odocoileus virginianus in Indiana, USA, using a simple double-observer approach. In our case study, we found evidence of false negatives associated with the viewer of infrared video. Additionally , we found that concordance between the detections of two viewers increased when using a red-green-blue camera paired with the infrared thermal sensor, when altitude decreased and when more stringent criteria were used to classify thermal signatures as deer. We encourage future managers and ecologists recording infrared video from aerial platforms to use double-observer methods to account for viewer-induced false negatives when video is manually viewed by humans. We also recommend combining infrared video with red-green-blue video to reduce false positives, applying stringent verification standards to detections in infrared and red-green-blue video and collecting data at lower altitudes over snow when needed.
... Il est nécessaire que tous les individus de la zone soient détectés ou que la proportion d'individus manqués soit correctement estimée, ce qui limite en général la taille de la zone. D'autre part, les mouvements des individus doivent être lents par rapport à celui des observateurs et les distances doivent être mesurées sans erreur (Borchers et al., 2002;Buckland et al., 2004;Buckland et al., 2001;Buckland et al., 2015;Glennie et al., 2015;Hedley and Buckland, 2004;Thomas et al., 2010). Dans le cas du recensement des équidés, il n'était pas possible de recomposer des transects à partir des sources de données, car nous ne disposions pas de zones dans lesquelles le nombre d'équidés vivants était connu de manière précise. ...
Malgré l’existence d’une base nationale d’identification des équidés centralisant les informations sur les équidés, leurs propriétaires et les détenteurs, les caractéristiques démographiques de la population équine (taille et localisation spatiale) sont mal connues, rendant difficiles le suivi, l’évaluation et la modélisation des événements sanitaires en filière équine et limitant l’efficacité des dispositifs de surveillance et de lutte. Cette situation est liée à l’organisation particulière de filière avec une mixité de particuliers et de professionnels, à un respect peu rigoureux de la réglementation et à l’absence d’obligation de déclaration de la localisation individuelle des équidés. La filière est également composée de multiples organismes et acteurs professionnels, engendrant une fragmentation des informations dans différentes bases de données non ou peu interconnectées. En investiguant la qualité des différentes sources de données disponibles et les habitudes de détention d’un échantillon de propriétaires et/ou détenteurs d’équidés, nous avons développé une méthodologie d’exploitation conjointe de ces données. Celle-ci nous a permis de réaliser au moyen d’une approche bayésienne, les premières estimations de la répartition spatiale des équidés en France à l’échelle de la commune, incluant les équidés détenus par des non-professionnels. La pertinence et la cohérence de ces résultats prometteurs seront évaluées prochainement en les confrontant aux recensements de terrain effectués dans certaines zones par les contrôleurs de l’Institut Français du Cheval et de l’équitation.
... Abundance estimates from visual aerial surveys of cetaceans are typically negatively biased due to availability and perception bias, and unmodelled heterogeneity in detection probability (Marsh and Sinclair, 1989;Buckland et al., 2015). Additionally, measurement errors (e.g., species misidentification, or mis-measurement of declination angles and school sizes), and animal movement in response to the observer, or independently of the observer, can cause biases in abundance estimates Glennie et al., 2015). Using aerial-line transect surveys with a double-observer platform and markrecapture distance sampling analyses, we were able to derive relatively precise estimates with CVs close to ≤ 20% (Buckland et al., 2001), while correcting for availability and perception bias. ...
Full-text available
Conservation management of wildlife species should be underpinned by knowledge of their distribution and abundance, as well as impacts of human activities on their populations and habitats. Common dolphins ( Delphinus delphis ) are subject to incidental capture in a range of Australia’s commercial fisheries including gill netting, purse seining and mid-water trawling. The impact these fishery interactions have on common dolphin populations is uncertain, as estimates of abundance are lacking, particularly for the segments of the populations at risk of bycatch and in greater need of protection. Here we used double-observer platform aerial surveys and mark-recapture distance sampling methods to estimate the abundance of common dolphins in 2011 over an area of 42,438 km ² in central South Australia, where incidental mortality of common dolphins due to fisheries bycatch is the highest. We also used the potential biological removal (PBR) method to estimate sustainable levels of human-caused mortality for this segment of the population. The estimated abundance of common dolphins was 21,733 (CV = 0.25; 95% CI = 13,809–34,203) in austral summer/autumn and 26,504 in winter/spring (CV = 0.19; 95% CI = 19,488–36,046). Annual PBR estimates, assuming a conservative maximum population growth rate of R max = 0.02 and a recovery factor of F r = 0.5 for species of unknown conservation status, ranged from 189 (summer/autumn) to 239 dolphins (winter/spring), and from 378 (summer/autumn) to 478 dolphins (winter/spring) with an R max = 0.04. Our results indicate that common dolphins are an abundant dolphin species in waters over the central South Australian continental shelf (up to 100 m deep). Based on the 2011 abundance estimates of this species, the highest estimated bycatch of common dolphins (423 mortalities in 2004/05) in the southern Australian region exceeded the precautionary PBR estimates for this population segment. Recent bycatch levels appear to be below PBR estimates, but low observer coverage and underreporting of dolphin mortalities by fishers means that estimates of dolphin bycatch rates are not robust. The effects of cumulative human impacts on common dolphins are not well understood, and thus we recommend a precautionary management approach to manage common dolphin bycatch based on local abundance estimates.
... In a sense, the issue is comparable to biases caused by random movement of animals before detection where animals are not detected at their original snapshot location but at some distance from it. Glennie et al. (2015) showed that for most animal speeds, the movement of animals causes larger biases for strip transects (a type of PS) than for line transects (a type of DS). Hence, we assume that for our case study, the higher estimates for PS compared to DS may have been caused by the uncertainty in localization and the resulting random error in distance measurements. ...
Full-text available
Various methods for estimating animal density from visual data, including distance sampling (DS) and spatially explicit capture-recapture (SECR), have recently been adapted for estimating call density using passive acoustic monitoring (PAM) data, e.g., recordings of animal calls. Here we summarize three methods available for passive acoustic density estimation: plot sampling, DS, and SECR. The first two require distances from the sensors to calling animals (which are obtained by triangulating calls matched among sensors), but SECR only requires matching (not localizing) calls among sensors. We compare via simulation what biases can arise when assumptions underlying these methods are violated. We use insights gleaned from the simulation to compare the performance of the methods when applied to a case study: bowhead whale call data collected from arrays of directional acoustic sensors at five sites in the Beaufort Sea during the fall migration 2007–2014. Call detections were manually extracted from the recordings by human observers simultaneously scanning spectrograms of recordings from a given site. The large discrepancies between estimates derived using SECR and the other two methods were likely caused primarily by the manual detection procedure leading to non-independent detections among sensors, while errors in estimated distances between detected calls and sensors also contributed to the observed patterns. Our study is among the first to provide a direct comparison of the three methods applied to PAM data and highlights the importance that all assumptions of an analysis method need to be met for correct inference.
... The CV of the dive-based density estimate is also lower than any of the other approaches. At this time, given the uncertainty in estimating an acoustic detection function for dives and the potential for bias caused by animal movement during a dive (Glennie et al., 2015), the dive cue-count method cannot be considered reliable. A full consideration of variation in dive lengths is also needed (instead of simply using a mean dive duration). ...
The population density of Cuvier's beaked whales is estimated acoustically with drifting near-surface hydrophone recorders in the Catalina Basin. Three empirical approaches (trial-based, distance-sampling, and spatially explicit capture-recapture) are used to estimate the probability of detecting the echolocation pulses as a function of range. These detection functions are used with two point-transect methods (snapshot and dive-cue) to estimate density. Measurement errors result in a small range of density estimates (3.9–5.4 whales per 1000 km²). Use of multiple approaches and methods allows comparison of the required information and assumptions of each. The distance-sampling approach with snapshot-based density estimates has the most stringent assumptions but would be the easiest to implement for large scale surveys of beaked whale density. Alternative approaches to estimating detection functions help validate this approach. The dive cue method of density estimation has promise, but additional work is needed to understand the potential bias caused by animal movement during a dive. Empirical methods are a viable alternative to the theoretical acoustic modeling approaches that have been used previously to estimate beaked whale density.
Standard line transect estimation is not robust to departures from the assumptions that animals do not move in response to the observer before detection and that all animals on or near the trackline are detected. We propose field methodology and analysis methods for which estimates are robust to the first assumption and do not require the second. The approach is applicable only to surveys in which two observation platforms are available, such that conventional line transect field methods, with effort concentrated close to the observer, are carried out from one, and a wider area is scanned from the second, possibly with low probability of detection. An example is a ship or terrestrial line transect survey together with a helicopter searching a wider area ahead of the observer. In the case of ship surveys, both observation platforms might be on the same vessel; observers might scan with hand-held binoculars or the naked eye on the first platform, and an observer with powerful tripod-mounted binoculars, screened from the view of the other observers, might search from the second.
Line transect estimators of abundance are biased when responsive movement in relation to the transect line occurs. We present and compare two approaches to correct estimators with auxiliary information on movement: a correction factor approach and a decomposition approach. Movement and detection of animals were simulated by computer. The decomposition approach is preferred over the correction factor approach. The methods are illustrated with survey data of Dall's porpoise in the Gulf of Alaska.
A model is proposed for estimating animal population densities using line transect methods when the intrusion of an observer causes the animals to move before being detected. This may occur for populations of small mammals or birds. The model is based on the concept used by many recent authors that the probability of observing an animal is a function of its perpendicular distance from the path followed by the observer. Estimation requires both the observed distances of the animals from the transect and the extent and pattern of their motion, the latter being more difficult to obtain. Some special cases are discussed as well as some possibilities for future development.
Line-transect estimation of wildlife populations can be based on either right-angle distances or radial distances. Estimation based on radial distances requires more restrictive assumptions about the mechanism by which observations are generated. We examine this mechanism and show that the mean sighting angle may frequently be larger than predicted by Hayne's fixed-distance model or by Eberhardt's variable-distance model. The modified and generalized Hayne's models are also investigated. An alternative approach, based on hazard-rate analysis, is suggested, but we conclude that a nonparametric right-angle-distance model will generally result in more reliable estimation than any radial-distance model.
A method is developed to account for effects of animal movement in response to sighting platforms in line transect density estimates using data on animal orientation. Models of expected distributions of animal orientation show that presence of responsive movement is determined by the ratio of animal sightings with angles of orientation in the third quadrant relative to the first quadrant. The distance at which response began is estimated using logistic generalized additive models of the relationship between radial distance and orientation. Density corrected for responsive movement is estimated by applying the Buckland and Turnock two-team analysis method to data poststratified into regions "close" to and "far" from (beyond the distance that responsive movement began) the observation platform instead of the original stratification by observation team. For data collected in the North Atlantic, white-sided dolphins, harbor porpoises, and minke whales responded by avoiding the survey ship, and white-beaked dolphins were attracted to the ship. For these populations, our method to correct for responsive movement gave significantly higher estimates, from 1.4 to 2.7 times the uncorrected estimates.