Please do not cite

Abundance and stock identity of bottlenose dolphins along the Outer

Banks of North Carolina

Kim W. Urian, Nicholas School of the Environment and Earth Sciences, Duke University, 135 Duke Marine Lab
Road, Beaufort, NC, 28516 USA
John Durban, National Marine Mammal Laboratory, Alaska Fisheries Science Center, NOAA Fisheries, 7600 Sand
Point Way NE, Seattle, WA, 98115 USA
Danielle Waples, Nicholas School of the Environment and Earth Sciences, Duke University, 135 Duke Marine Lab
Road, Beaufort, NC, 28516 USA
Susan Barco, Virginia Marine Science Museum, 717 General Booth Blvd. Virginia Beach, VA, 23451 USA
Nan Bowles, North Carolina Maritime Museum, 315 Front Street, Beaufort, NC, 28516 USA
Rich Mallon-Day, Nags Head Dolphin Watch & Cape May Dolphin Survey, 40 Orchard Lane, Berwyn, PA,
19312 USA
Keith Rittmaster, North Carolina Maritime Museum, 315 Front Street, Beaufort, NC, 28516 USA
George Rountree, Christopher Newport University, 50 Shoe Lane, Newport News, VA, 23606 USA
Laela Sayigh, University of North Carolina-Wilmington, 5600 Marvin K. Moss Lane, Wilmington, NC 28409 USA
David Schofield, Marine Science Consortium Project, National Aquarium in Baltimore, Pier 3, 501 E. Pratt Street,
Baltimore, MD, 21202 USA
Victoria Thayer, Nicholas School of the Environment and Earth Sciences, Duke University, 135 Duke Marine Lab
Road, Beaufort, NC, 28516 USA
Rob Young, Coastal Carolina University, P.O. Box 261954, Conway, SC, 29528 USA
Andrew J. Read, Nicholas School of the Environment and Earth Sciences, Duke University, 135 Duke Marine Lab
Road, Beaufort, NC, 28516 USA

Contact e-mail: kurian@ec.rr.com

ABSTRACT

We describe the results of a photographic mark-recapture analysis of the abundance and stock identity of
bottlenose dolphins off the Outer Banks of North Carolina, USA during the winter of 2003. In our
estimation of abundance we adopted a Bayesian modeling approach to account for uncertainty resulting
from variance in capture probability across individuals and surveys, and to account for dependence
between surveys. We described the stock identity of bottlenose dolphins by matching individuals
photographed off the Outer Banks in winter to a subset of images from the Mid-Atlantic Bottlenose
Dolphin Catalog (MABDC), a large co-operative research project that includes images of dolphins from
New Jersey to Florida. We conducted surveys on nine days during February and March 2003. During
these surveys we encountered 34 groups of dolphins and took 2,907 photographs. Some photographs
contained the dorsal fin of more than one dolphin; we photographed 3,428 fins in total. Each dorsal fin
image was graded for both quality and distinctiveness, and we used only excellent and good quality images
of distinctive fins in our analysis. We constructed sighting histories of individual dolphins and used these
data to first estimate the abundance, N, of dolphins with reliable markings. We then re-scaled these
estimates to the overall number of dolphins, P, by using information on the proportion of individuals with
reliable markings. We estimate that 1,339 dolphins were present in our study area along the Outer Banks
during February and March 2003. Ninety-five percent of the posterior probability distribution
encompassed with this estimate of abundance ranged between 1,094 and 1,692. We identified dolphins
from the Northern Migratory and Northern North Carolina stocks but no dolphins from the Southern
North Carolina stock during our surveys. Thus, the southern boundaries of the Northern Migratory and
Northern North Carolina stocks may both lie somewhere between Beaufort and Wilmington, NC in
winter and, if so, the current management approach could be simplified by combining only these two
management units as the Winter Mixed stock and treating the Southern North Carolina stock as a separate,
year-round unit.

KEYWORDS: ABUNDANCE ESTIMATE, MARK-RECAPTURE; PHOTO-ID; MOVEMENTS;

BOTTLENOSE DOLPHIN

1
INTRODUCTION

Humans and bottlenose dolphins (Tursiops truncatus) have a long history of interaction off the Outer Banks
of North Carolina, USA. A fishery for bottlenose dolphins operated intermittently near Cape Hatteras
from 1797 until 1929 (Mead, 1975), in which dolphins were taken in large-mesh shore seines (True, 1885).
The fishery operated from November to May, when dolphins were plentiful enough to support the
operation (True, 1885). More than 2,000 dolphins were taken along the coast of North Carolina in some
winters (Mead, 1975; Reeves and Read, 2003). The main products of the fishery were oil, derived from
blubber and jaw fats, and leather (Mitchell, 1975). True (1891) noted that the fishermen of Hatteras were
of the opinion that the dolphins migrated northward in the spring and southward in the fall, although a
few remained in the vicinity of Hatteras throughout the summer. As noted by Mead (1975) “much of this
migration was within a hundred yards of the shore, providing the opportunity for a shore-based fishery.”

Today, the primary management concern regarding bottlenose dolphins along the North Carolina coast is
by-catch in coastal gill net fisheries. Assessment of the magnitude of this problem is complicated by: a
complex of gillnet fisheries; the seasonal movements and mixture of dolphin stocks along the coast; and
difficulty in estimating abundance and mortality for each stock. The best available scientific information
(Waring et al., 2002) suggests that three stocks of bottlenose dolphins are found in the coastal waters of
North Carolina: a Northern Migratory stock that migrates north as far as Long Island during the summer;
a Northern North Carolina stock that is present year round; and a Southern North Carolina stock that is
also present year-round (Fig. 1). The Northern Migratory stock moves south in the winter and is believed
to spend the winter off the coast of North Carolina. Current boundaries between these management units
are tentative and will likely change as more information becomes available (Garrison et al., 2003).
Nevertheless, this general pattern of stock structure is entirely consistent with the seasonal movements of
bottlenose dolphins reported by True from conversations with fishermen at Hatteras more than a century
ago.

Taking this hypothesized stock structure and patterns of seasonal movement into account, current
management measures for bottlenose dolphins in North Carolina are separated into summer (May -
October) and winter (November - April) periods. During the summer, only the Northern and Southern
NC stocks are present off the coast of North Carolina. In winter, dolphins from all three stocks are
believed to mix and are, therefore, vulnerable to by-catch in North Carolina gill net fisheries. It is during

2
this winter mixing period that most by-catches occur (Waring et al., 2002), so it is of considerable
importance to refine our knowledge of bottlenose dolphins off the coast of North Carolina during this
period.

There have been two prior estimates of the abundance of bottlenose dolphins off the coast of North
Carolina in winter. Analysis of an aerial line transect survey conducted off the coast of North Carolina
between January and March 1995 yielded an estimate of 4,734 (CV 0.49) dolphins (Garrison, 2002). This
survey extended from Cape Hatteras to the North Carolina - South Carolina border and out to 27 km
from shore. Dolphins observed further offshore were assumed to form part of a separate offshore stock
(Garrison, 2001; Torres et al., 2003). A more recent survey, conducted during January and February 2002,
generated an estimate of 16,913 (CV 0.23) (Garrison et al., 2003). This survey incorporated more
sophisticated analytical techniques and had a wider geographic coverage. The field methods used in 2002
were designed to test the assumption that all animals on the survey track line are detected. In addition, the
surveys extended north of Cape Hatteras to Cape Henry, Virginia (Garrison et al., 2003).

The potential biological removal (PBR), or allowable removal level for a stock of marine mammals,
authorized under the Marine Mammal Protection Act, is a direct function of abundance. The PBR for the
mixed winter stocks of bottlenose dolphins off North Carolina is 68; the most recent estimates of the
mortality of these dolphins in gill net fisheries is 151 per winter (Waring et al., 2002). The by-catch of
bottlenose dolphins exceeds PBR for several stocks of bottlenose dolphins along the Atlantic coast of the
U.S., so a Take Reduction Team was formed in 2001. In April 2003 this Team, mandated under the
Marine Mammal Protection Act, submitted a draft Take Reduction Plan to the Secretary of Commerce,
outlining conservation measures that will reduce the by-catch of bottlenose dolphins to below PBR for
each affected stock. The draft Take Reduction Plan includes a variety of measures designed to reduce the
by-catch of dolphins in North Carolina gill net fisheries.

In this paper, we report the results of a photographic mark-recapture analysis of the abundance and stock
structure of bottlenose dolphins in the waters off the Outer Banks of North Carolina derived from
dedicated surveys conducted during February and March 2003. In our estimation of abundance we
adopted a Bayesian modeling approach to account for uncertainty resulting from variance in capture
probability across individuals and surveys, and to account for dependence between surveys. We describe
the stock identity of bottlenose dolphins by matching individuals photographed off the Outer Banks in
winter to a subset of images from the Mid-Atlantic Bottlenose Dolphin Catalog (MABDC), a large co-
operative research project that includes images of dolphins from New Jersey to Florida (Urian et al., 1999).

3
Finally, we conduct an explicit test of the hypothesis that the dolphins found off the northern Outer Banks
during winter represent a mixture of the Northern Migratory, Northern North Carolina and Southern
North Carolina stocks. As noted above, the degree of mixing and boundaries between these stocks are
poorly understood; our efforts were designed to improve knowledge of the stock identity of dolphins in
this area.

METHODS

Field Effort
We conducted photographic surveys of bottlenose dolphins along the Outer Banks of North Carolina on
nine days in February and March 2003 (Table 1). The surveys were conducted from a 7m outboard-
powered, center-console research boat or a 13m, diesel-powered commercial fishing vessel. We surveyed
from Cape Hatteras to Ocracoke Inlet, a distance of approximately 50km (Fig. 2). We divided this area
into four zones of relatively equal size: from Cape Point to the Frisco Pier (Zone 1), from the Frisco Pier
to Hatteras Inlet (Zone 2), from Hatteras Inlet to the Ocracoke Woods (Zone 3) and from the Ocracoke
Woods to Ocracoke Inlet (Zone 4). We attempted to expend an equal amount of survey effort in each of
the four zones, but our daily survey routes were largely dependent on weather. We did not follow set
transect lines, but instead attempted to ensure that we encountered the largest number of dolphins, given
the prevailing sighting conditions. Our survey efforts were concentrated in near-shore habitat, from the
surf zone out to approximately 1000m from shore, where most dolphins were present.

At each encounter with dolphins, we recorded the position (using a GPS receiver) and a field estimate of
the number of dolphins in the group. One or two researchers took photographs of the dorsal fins of
dolphins using a Nikon digital camera or a Nikon camera with color slide film (both cameras were
equipped with 300mm lenses). We attempted to photograph every dolphin in each encounter, but in some
cases our photographic coverage of a group was incomplete because of inclement weather or because
dolphins were in waters too shallow to operate our vessels.

Photo-identification
We labeled each photographic slide and digital image with the date, encounter number and roll number.
Prior to photo-identification, we graded the dorsal fins in each slide and digital image for photographic
quality (Read et al., 2003 and Appendix 1). The photographic quality score was based on a weighted scale
that incorporated the following characteristics: focus and clarity, contrast, angle of the fin to the

4
photographer and visibility of the fin. Any image at an oblique angle or that did not show the entire
trailing edge of the dorsal fin from the tip to the posterior insertion was excluded from further analysis.
Excellent quality images received scores from 6-9; good quality images ranged from 10-12; and poor
quality images scored 13 and higher (see Appendix 1 for a full description of these methods).

We also scored the distinctiveness of marks on each dolphin’s dorsal fin; this was evaluated using the
patterns of nicks and notches on the fin. Dolphins with the most distinctive features (evident in even a
poor quality photograph) were scored 1; those with intermediate features (at least 2 distinguishing features
or 1 major feature) were scored 2; and animals with few or no distinctive characteristics were considered
unmarked and received a score of 3. The distinctiveness of each individual was evaluated independently
by two researchers; the researchers worked together, or consulted a tie-breaker, to resolve cases of
disagreement.

We examined all slides and digital images of dorsal fins considered to be of excellent or good photographic
quality (scores < 13) during photo-identification. By restricting the data set to excellent and good quality
images of distinctively marked individuals (distinctiveness scores of 1 or 2), we minimized subjectivity in
the matching process and reduced the chance of making incorrect identifications or missing them
altogether (Friday et al., 2000). We sorted images of individual dolphins based on the location of the most
prominent feature on their fin, gave each individual an identification number and placed its image in a
catalog. We compared each distinctive dolphin to every other marked individual before adding it to the
catalog, and every potential match of an individual dolphin from one encounter to another was verified by
a second researcher.

Abundance Estimation
We assumed that every dolphin using the study area had some probability of being photographed over the
course of the two-month period of field work, although this probability could vary across individuals and
individual dolphins might not have spent all their time in the surveyed areas. It is important to note that
our abundance estimates correspond to the number of animals that used the study area during the sampling
period, and we make no explicit assumption as to how this estimate relates to the total population to which
these animals belong. As described above, we compiled an identification history for each individual that
was deemed to be reliably marked, consisting of a row of 1's and 0's to indicate whether or not the
individual was identified on each of the nine surveys. We used these data to first estimate the abundance,
N, of dolphins with reliable markings. We then re-scaled these estimates to the overall number of

5
dolphins, P, by using information on the proportion of dorsal fins photographed on each survey that
included individuals with reliable markings.

The Rasch Model

To calculate N, we needed to estimate the number of individuals never identified, n0, by using information
gleaned from the pattern of re-identification of the n individuals that were identified in at least one of the
surveys. We used a stochastic (or probability) model to link these unknown parameters to the data to
make inference about their value. The model we employed was based on the Rasch model from the field
of educational testing (Rasch, 1960), which has been used in the context of abundance estimation by
Darroch et al. (1993), Agresti (1994) and, more recently, by Coull and Agresti (1999) and Fienberg et al.
(1999). This model was based on the assumption that the probability of identification varies both across
survey occasions and among individuals. If we let j = 1,…, N index the individuals, and k = 1,…, K index
the survey samples, the model has a matrix of N × K Bernoulli random variables yjk, such that

yjk = 1 if individual j is identified on survey k

yjk = 0 if otherwise

yjk ~ Bernoulli(pjk)

The probability pj,k, that individual j was identified on occasion k was modeled as a logistic function

parameterized by an overall mean level for the detection probability, µ, with departures from this overall
mean due to survey effects, αk, and individuals effects, θj:

log{pjk/(1- pjk)} = µ + θj + αk

In this situation, of course, the abundance N was unknown, and we only observed data for n rows of the
data matrix, where n was the number of individuals that were identified at least once. The remaining n0 (=
N – n) rows were all vectors of 0’s. The aim of our modeling was to estimate how many n0 rows of 0’s,
and associated individual effects, θ, there should be and, therefore, to estimate N, which was treated as an
unknown parameter in the model (e.g. Fienberg et al., 1999). Estimation was accomplished by modeling
the individual effects θ, for both observed and unobserved individuals, as random effects, drawn from the

6
same common probability distribution. Specifically, we assumed Normal distributions, centered on zero,
to describe the variability of these effects (e.g. Coull and Agresti, 1999; Fienberg et al., 1999):

θj ~ Normal(0, σ2θ)

where Normal(a, b) denotes a Normal distribution with mean a and variance b. By linking observed and
unobserved individual effects in this way, we used the distribution of catchability for the observed
individuals to predict the number of unseen individuals, and corresponding rows of zeros, that are
supported in the model. The survey effects were also modelled as random effects, centred on zero:

αk ~ Normal(0, σ2α)

Bayesian Hierarchical Formulation

To produce statements of uncertainty about model parameters in the form of full probability distributions,
we followed Fienberg et al. (1999) in adopting a Bayesian formulation to this model. Bayesian statistical
inference is advocated and used with increasing frequency for analyzing and communicating uncertainty in
ecological data analysis (e.g. Ellison, 1996; Wade, 2000). Bayesian inference begins by assigning prior
probability distributions to all unknown parameters in the model. This prior belief is then updated
conditional on the observed data, and inference about each parameter is based on the conditional or
“posterior” probability distribution (Gelman et al., 1995). In the absence of useful independent prior
information, we adopted vague priors for the parameters.

We adopted a discrete uniform distribution for N, where the lower bound was set to be the number of
individuals observed (n = 298) and the upper bound was set to be n + 2,000 (=2,298). Examination of
later results indicated that all the posterior probability for N was contained below this upper bound, and
therefore this prior was flat across the probably range of values. The number of unseen individuals n0
could then be derived from the logical relationship N – n. For the overall mean capture probability on the
logit scale, µ, we adopted a Normal prior centered at zero with a relatively large prior variance:

µ ~ Normal(0, 10)

7
With the survey and individual effects (α, θ) treated as random effects centered on zero (no effect), the
prior distributions for these parameters were hierarchically defined through the variance hyper-parameter
of their associated random effects distributions, σ2α and σ2θ , respectively. These variance hyper-
parameters were assigned separate inverse gamma prior densities that constrain values to be positive,
allowing non-zero effects to emerge as supported by the data:

-1
σ2α ~ Gamma (0.1, 0.1)
-1
σ2θ ~ Gamma (0.1, 0.1)

where Gamma(a, b) denotes a gamma distribution with mean a / b and variance a / b2

Inference Through MCMC

Once these prior distributions had been assigned, this Bayesian model could now be considered as a joint
probability distribution that described the relationships between all unknowns and the data. Inference was
then based on the conditional probability distribution of the unknowns given the data, the posterior
distribution. This multivariate distribution naturally incorporated uncertainty from one subset of random
variables or parameters into inferences for another subset. Inference on single parameters of interest
could then be obtained by integrating this joint distribution over all other parameters to obtain the
marginal posterior distribution of interest (Gelman et al., 1995). However, for multi-parameter problems,
this requires multi-dimensional integration, which can present practical difficulties. For the current model,
a closed form analysis of the posterior distribution was intractable, and therefore we followed Fienberg et
al. (1999) in the use of computer intensive Markov Chain Monte Carlo (MCMC) simulation methods
(Brooks, 1998). MCMC is a method of simulating random samples from any theoretical multivariate
distribution, and can therefore be used in Bayesian inference to approximate marginal posterior probability
distributions for individual parameters of interest. This posterior distribution not only indicates the most
likely values of the parameters, but also allows a crucial assessment of the uncertainty associated with each
of these parameters.

This Bayesian formulation offered the advantage that we could lay out all the pieces of the model and
modify exactly those parts that need adjustment to reflect the dependence in the data. Furthermore, when
fitting models using MCMC, additions and changes could be easily accommodated through modification
only to the conditional distributions of the parameters that were directly related to the new components.

8
Therefore, additional layers could be added to the model and thus directly included in this conditional
updating scheme, allowing uncertainty to be propagated across the hierarchical levels.

Hierarchical Extension 1: Mark-type rescaling

To produce estimates of overall abundance P, it was necessary to re-scale the estimate of the abundance of
marked dolphins, N, to include animals that were not reliably identifiable (i.e. those without long-lasting
identifiable markings, or whose markings were obscured by tassel barnacles (Xenobalanus). This required an
estimate of the proportion of individuals that were reliably marked. To calculate π, excellent and good
quality photographs were used to determine the number of photographs of dolphins with reliable marks
(Distinctiveness = 1 or 2) on each survey k (mk) from the total number of photographs of dolphins of all
distinctiveness levels on that survey (wk). The number with reliable markings was then treated as a
binomial sample from the total sample size on each survey, and the proportion of individuals that
possessed reliable markings (π) was estimated as the binomial probability:

mk ~ Binomial(π, wk)

The proportion π was initially assigned a vague beta prior distribution with probability mass equally spaced
between 0 and 1:

π ~ Beta (1,1)

where Beta(a, b) indicates a Beta distribution with mean c =a/(a+b), and variance, v =c (1-c)/(a+b+1).
Estimation of the posterior distribution for the rescaling proportion π was conducted in the same MCMC
run as estimation of the number of individuals N in the mark-recapture model. These two components
were then linked to form a single probability model, by defining the overall abundance P to be a function
of N and π :

P=N/π

By embedding this mark-type rescaling step into the full probability model with the mark-recapture
component, we incorporated the uncertainty from both the mark-recapture and mark-type rescaling
components directly into the inference about P.

9
Hierarchical Extension 2: Stratifying Individual Effects
The nine survey occasions were not evenly spread throughout the two-month survey period, but were
clumped into three relatively discrete time periods:

Period 1 Period 2 Period 3

1) 14-Feb-03 3) 04-Mar-03 6) 23-Mar-03
2) 19-Feb-03 4) 08-Mar-03 7) 24-Mar-03
5) 11-Mar-03 8) 25-Mar-03
9) 26-Mar-03

Therefore, we could not necessarily expect these surveys to be independent. Instead there may have been
dependent recaptures of individuals within periods, with perhaps a different rate of recapture between
surveys conducted in different periods. This could have been particularly important if animals were
moving in and out of the survey area during the overall study period. Additional dependency between
surveys may have also been introduced through variations in the geographical coverage across surveys
(Table 2). However, the basic Rasch model assumes independence between surveys. To relax this
assumption, we attempted to account for this possible dependence by including dependence structure
directly into the model formulation.

Following Fienberg et al. (1999) we modeled dependence between surveys through stratification of the
latent individual catchability distribution into multidimensional components, each defined by particular
survey occasions. Specifically, we defined three strata in this case (each corresponding to a survey period
as defined above), and we denoted θj 1 to underlie surveys 1-2; θj 2 for surveys 3-5 and θj 3 for surveys 6-9.

The prior distribution for (θj 1, θj,2, θj 3) was taken to be a multivariate Normal distribution N3(0, Σ). Rather

than being modeled through a single continuous latent distribution, the vector of individual effects, θ, was
now multidimensional, stratified into q = 3 strata. The multivariate Normal distribution assigned to θ
described the association between the vector of q = 3 continuous variates. Specifically, Σ was the co-
variance matrix of the order q*q, where the principal (left to right) diagonal element of this matrix was the
estimate for the variance of the individual catchability for strata q, and the off-diagonal values represented
co-variances between pairs of strata. For example, an individual with high catchability in strata 1 would
have high catchability in strata 3 if a positive co-variance existed.

10
In the multivariate case, the conjugate prior for the co-variance matrix Σ was taken as the inverse wishart
distribution, the multivariate generalization of the inverse gamma. The wishart is specified in terms of two
parameters. One is a degrees of freedom parameter, v, which must be equal to or exceed q if the prior is to
be proper. Larger values of v represent stronger belief, and we therefore adopted a value of v = q = 3 to
represent a vague prior. The other parameter is the scale matrix B of order q*q, which identifies the
structure of the multivariate data. A natural formulation for B is the identity matrix. The prior
distributions on N, µ and αk were the same as in the unidimensional Rasch model. The development of
the Bayesian MCMC framework for this multi-dimensional model was otherwise analogous to the simpler
unidimensional model, except that the inverse gamma conditional distribution for σ2θ was replaced by the
corresponding inverse wishart distribution for Σ.

Model selection and goodness-of-fit

To choose between these two possible models, we made a more formal comparison of model fit using the
Deviance Information Criteria (DIC), as suggested by Spiegelhalter et al. (2002). DIC is intended as a
generalization of Akaike's Information Criterion (AIC), but specifically designed for Bayesian hierarchical
models where the number of parameters is not clearly defined. This approach to model comparison
involves weighting a measure of fit, the deviance statistic, by a measure of model complexity to penalize
the fit depending on the number of free parameters in the model. The model with the smallest DIC is
estimated to be the model that would best predict a replicate dataset of the same structure as that currently
observed.

Although the DIC approach informs us of the relative fit of alternative models, it does not tell us how well
the selected model replicates the observed data. Therefore, we also performed a goodness-of-fit test as a
diagnostic model check. To assess model fit we used posterior predictive checks and Bayesian p-values
(Meng, 1994; Gelman et al., 1996). This approach was based on predicting a new set of data by sampling
directly from the model. We then compared our observed data with the predicted data from the model. If
the model adequately described the observed data, then the predicted data should be “similar” to the
observed data. This similarity was quantified though the use of Bayesian p-values, which applied a
discrepancy measure to compare the actual and predicted observations. We adopted absolute test
quantities as a discrepancy measure for both the observed data D(Y) and the predicted data D(Ynew), under
the model with expected values (p) determined by the parameter set comprising the model.

11
D(Y) = Σ Σ│yjk - pjk│

D(Ynew) =Σ Σ│ynewjk - pjk│

Since, from the Bayesian perspective, the parameters of the models had a distribution rather than a fixed
value, we obtained not a single discrepancy value for each data set, but a sample from the posterior
distribution of discrepancy values. Over the set of MCMC iterations, goodness of fit was thus measured
by comparing the calculated discrepancy values by recording the proportion of iterations that D(Y) was
greater than D(Ynew). If the distributions of discrepancy values for the observed and predicted data were
the same, then an optimal p-value of 0.5 would be obtained. In contrast, a p-value close to 0 or 1 would
indicate poor model fit, with the actual data in the tails of the predictive distribution.

Stock structure
To determine the stock identity of individual dolphins present off the Outer Banks in winter, we compared
images of their dorsal fins to a subset of the Mid-Atlantic Bottlenose Dolphin Photo-identification Catalog
(MABDC). Specifically, we compared our images to those taken by researchers in New Jersey, Virginia,
Roanoke Sound N.C., Pamlico Sound N.C., Beaufort N.C., Wilmington N.C, and Murrell’s Inlet, S.C. – a
total of 1,908 dolphins (Table 3). At the northernmost sites (New Jersey and Virginia) dolphins are
present only during summer and in Manteo survey effort is restricted to the summer months. At the other
sites, dolphins are present year-round and surveys are conducted throughout the year.

RESULTS

Field Effort
Our survey effort focused on the near-shore coastal habitat, from the surf zone out to approximately
1000m from shore, although most dolphin groups were encountered within 500m of the beach. Our
survey effort covered all four zones from Cape Hatteras to the Ocracoke Inlet (Fig. 2 and Table 2).
During our nine surveys we encountered 34 groups of dolphins (Fig. 2) and took 2,907 photographs
(Table 1).

Photo-identification
Some of the 2,907 photographs contained more than one dolphin dorsal fin; the total number of dorsal
fins photographed was 3,428. Each dorsal fin was graded for both quality and distinctiveness, thus one
photographic image could have multiple images of varying distinctiveness and quality. Of the 3,428

12
images of dorsal fins, 2,161 were of photographic quality scores less than 13 and thus met our criteria for
analysis. We considered 1,489 images of dorsal fins to be of excellent quality (scores 6-9) and 672 images
of dorsal fins to be of good quality (scores 10-12). From these 2,161 images of dorsal fins we identified
298 marked dolphins that met our criteria for distinctiveness. Sixty-five dolphins were assigned a
distinctiveness score of 1 and 233 dolphins were considered to have a distinctiveness score of 2. Forty-
seven of the 298 marked dolphins were photographed more than once during the course of the surveys: 34
of these were photographed on two different days, 10 on 3 days, and three dolphins were photographed
on four separate days.

Abundance Estimation
For both models we obtained a sample of 50,000 values from the parameter posterior distributions using
three different MCMC chains, after discarding the first 5,000 “burn-in” iterations prior to chain
convergence. The length of the burn-in was assessed using the method of Gelman and Rubin (1992), as
modified by Brooks and Gelman (1998), which is based on summary statistics comparing the variances
within and between the three different simulated chain sequences. Table 4 presents summary statistics for
the marginal posterior distributions of model parameters, for both models, that were estimated from the
MCMC sequences.

In the univariate Rasch model, the estimated variance in the individual effects ( σ2θ) was higher than the
variance for the survey effects (σ2α), representing substantial individual heterogeneity in the probability of
being identified. This heterogeneity resulted in a large number of estimated undetected individuals being
introduced into the model (n0 ~ 1,313), and the overall level of the capture probability on the logit scale,
µ, corresponded to a mean capture probability of only around 0.01. However, when dependence between
survey periods was incorporated into the multivariate Rasch model through stratification of the individual
effects, the estimated variances in individual effects were considerably smaller within the strata than was
the single variance estimate for the unidimensional model. This indicates that much of the variability in
the individual effects estimated in the univariate model can be attributed to variability in individual
catchability between strata, rather than between temporally adjacent surveys within strata. Additionally,
there was a generally negative estimated covariance in individual catchability of observed individuals
between pairs of strata (Table 5), indicating that individuals with high catchability in one strata typically
had low catchability in other strata. When these dependencies are accounted for directly in the model, we
estimate a considerably smaller number of undetected individuals being introduced into the model (n0 ~
519), and a corresponding higher overall level of the capture probability of around 0.04.

13
Driven by these differing estimates of n0, the models produced differing estimates of the number of
reliably marked individuals, N, with the estimate from the multivariate model representing only
approximately 50% of the estimate from the univariate model. However, for each model, the values
sampled in the MCMC sequence covered a wide ranged of possible values for N, reflecting the uncertainty
associated with this abundance estimation. The estimated proportion, π, of the animals with reliable
markings was approximately 61% (π = 0.61, 95% CI = 0.59-0.63). The estimate was the same in both
models (as the model structure for the mark-type re-scaling was identical in both) and therefore the
difference in the estimated abundance of reliably marked dolphins was also reflected in the posterior
estimate for total number of individuals, P. The median of the sampled values for total abundance was
2,643 (95% CI = 2,000–2,166) using the unidimensional Rasch model, compared to 1,339 (95% CI =
1,094–1,692) for the estimate from the multivariate model.

To choose between these two possible estimates, we made a more formal comparison of model fit using
the Deviance Information Criteria (DIC), as suggested by Spiegelhalter et al. (2002). The model with the
smallest DIC is estimated to be the model that would best predict a replicate dataset of the same structure
as that currently observed. In this case the DIC value for the unidimensional Rasch model was estimated
as 1,940, compared to the multivariate Rasch model with a lower DIC of 1,927. Therefore, we concluded
that the multivariate Rasch model, which incorporated the dependence between survey occasions, was a
better predictor of the observed data, and the associated abundance estimate of 1,339 (95% CI = 1,094–
1,692) appears to be the best estimate for the number of bottlenose dolphins present in our study area off
the Outer Banks of North Carolina during winter. Fig. 3 presents the full posterior probability distribution
for the estimated total abundance using this model, illustrating the uncertainty associated with this
estimate. Goodness-of-fit was assessed for the selected multivariate Rasch model based on a posterior
predictive p-value, which was obtained by comparing the distribution of the discrepancy measure for both
the observed data D(Y) and predicted data D(Ynew). The discrepancy for the observed data (posterior mean
= 541, SD = 15) was estimated to be very similar as in the predicted replications (posterior mean = 539,
SD = 26). The similarity of these two distributions was summarized by a posterior predictive p-value of
0.44, which implies that the realized discrepancy of the data is similar to what might occur under
replications under the model, indicating a satisfactory model fit (Gelman et al., 1996).

14
Stock structure

Of the 298 dolphins we considered to be ‘marked’ for our estimate of abundance, only 234 were of
sufficient distinctiveness to compare to the MABDC subset. We excluded 64 individuals because a
portion of their fins were obscured by tassel barnacles, thus obscuring the permanent markings necessary
to identify individuals over extended periods of time. We included these 64 individuals in our mark-
recapture estimation of abundance, because we were confident that we could re-identify them during our
two-month study period. We were not confident, however, that we could use these features to match
dolphins over time periods of years or decades.

To systematically compare each of the 234 dolphins to the MABDC subset, we made 450,526 pair-wise
comparisons. Most (156) dolphins were not matched to any other site, but we matched 78 dolphins (0.33)
to at least one other site. Matches were made to dolphins photographed at sites from New Jersey to
Beaufort, NC; matches were not made to the southern catalogs from Wilmington or Murrell’s Inlet.
Several individuals were matched to multiple sites. For example, two individuals were matched to the
catalogs of Virginia Beach, Pamlico Sound and Beaufort, and two other individuals were matched to
Roanoke Sound, Pamlico Sound and Beaufort. Fig. 4 shows the proportion of the dolphins we
photographed along the Outer Banks that were matched to other field sites, taking into account the size of
the catalog at each site. The highest proportion of matches were made to the catalogs of Virginia Beach
and Pamlico Sound.

To examine seasonal movement patterns, we examined the sighting histories of individuals we matched to
the MABDC and determined whether they were photographed at each site during winter or summer, as
defined by the current management regime (Winter = November - April; Summer = May - October).
Most of the dolphins we photographed along the Outer Banks during the winter of 2003 were identified
previously north of Cape Hatteras during summer or throughout the year in Pamlico Sound (Fig. 5). Some
dolphins appear to move back and forth between the Outer Banks and Beaufort during winter (Fig. 5), but
these animals do not move further south to Wilmington or Murrell’s Inlet. Thus, there appears to be a
high degree of exchange between the Outer Banks and Beaufort in the winter and a stock boundary
somewhere between Beaufort and Wilmington. It is important to note that no dolphins from the
Southern NC stock (defined as those dolphins south of Cape Lookout in summer) were photographed off
the Outer Banks in winter.

15
DISCUSSION

Our study of bottlenose dolphin abundance along the Outer Banks represents a non-standard application
of the mark-recapture approach. Our sample data were non-random and biased towards particular
individuals and sampling occasions, so we required models that can become complex (e.g. Chao et al.,
1992; Agresti, 1994; Coull and Agresti, 1999; Pledger, 2000). This can present challenges for quantifying
the uncertainty associated with abundance estimates (e.g. Coull and Agresti, 1999). Conventional
approaches for inference from mark-recapture models are typically oriented to find a single optimum
estimate, such as the maximum likelihood estimate, with statements of uncertainty generally derived from
assumptions about large sample normality (Buckland et al., 2000). However, inference of this type is not
necessarily the best suited for communicating uncertainty to both technical and non-technical users of
such population data (Wade, 2000). We present a Bayesian mark-recapture method for explicitly
communicating uncertainty about abundance, where capture probabilities vary across both individuals and
sampling occasions. The Bayesian method yields a full probability distribution for the number of animals,
rather than point estimates with associated errors, communicating both extent and shape of the associated
uncertainty. Although our analysis indicates a most probable abundance of approximately 1,340 dolphins,
the posterior distribution covers a wide range of possible values, with 95% of the posterior probability
encompassed with the abundance range between 1,094 and 1,692.

Our analysis also emphasizes the importance of customizing and selecting mark-recapture models in order
to adequately describe major sources of variation in capture probabilities (Pollock et al., 1990). Specifically,
by dividing the survey occasions into three strata of temporally adjacent surveys, we were able to be more
explicit about these sources of variation, and obtain better estimates. In particular, much of the variability
in the capture probability of individuals could be attributed to variability in individual catchability between
strata, rather than between temporally adjacent surveys within strata. Additionally, there was a generally
negative estimated covariance in individual catchability between pairs of strata, indicating that individuals
with high catchability in one strata typically had low catchability in another strata. Although this effect
may have been due in part to variations in the geographical coverage across surveys, it is also consistent
with the hypothesis that individuals may have been moving into and away from the study area during the
course of the study period, and were therefore typically only available for sampling within a single survey
strata. This emphasizes the importance of considering individual ranging patterns, not only to understand
this heterogeneity and its influence on the abundance estimate, but also to identifying appropriate
monitoring strategies and techniques for data analysis in the future.

16
Our abundance estimate refers to the estimated number of dolphins using the Outer Banks during a two-
month study period during the winter of 2003. The dolphins we identified were comprised of animals
from the Northern Migratory and Northern North Carolina stocks (Fig. 1). This includes dolphins that
spend the summer in New Jersey and Virginia (the Northern Migratory stock) and Roanoke and Pamlico
Sounds, NC (the Northern North Carolina stock). Many of the animals we identified off the Outer Banks
had been photographed during previous winters near Beaufort (Fig. 5), suggesting some long-shore
movement of dolphins from Cape Hatteras to Cape Lookout (near Beaufort) between November and
April. In addition, dolphins clearly move back and forth from Pamlico Sound into coastal waters, as we
identified many individuals that had been photographed previously in this adjacent estuarine area.

We did not identify any dolphins from the Southern North Carolina stock during our surveys, despite the
relatively large catalogs of dolphins from Wilmington, NC and Murrell’s Inlet, SC in the MABDC (Table
3). Dolphins from the Northern Migratory stock move as far south as Beaufort, but not to Wilmington, in
the range of the Southern North Carolina stock (K. Urian, unpublished data). Thus, the southern
boundaries of the Northern Migratory and Northern North Carolina stocks may both lie somewhere
between Beaufort and Wilmington, NC in winter. If so, the current management approach could be
simplified by combining only these two management units as the Winter Mixed stock and treating the
Southern North Carolina stock as a separate, year-round unit. The hypothesis that only the Migratory and
Northern North Carolina stocks overlap during winter (and that the Migratory stock does not move into
the range of the Southern North Carolina stock) should be tested with dedicated photo-identification
research conducted along the entire coast of North Carolina.

We can compare our estimates of the number of dolphins present along the Outer Banks during the
winter of 2003 with an abundance estimate generated from aerial line-transect surveys conducted during
the winter of 2002 (Garrison et al., 2003). These surveys, flown during January and February 2002, yielded
an estimate of 16,913 (CV 0.23) coastal dolphins for the entire North Carolina Winter Mixed management
unit. Density estimates obtained during the 2002 aerial survey were applied to an area of almost
38,000km2 (Garrison et al., 2003). Our study covered only a very small portion (50km2) of the total area
surveyed in 2002 and our results are not strictly comparable to density estimates derived from line-transect
surveys. Nevertheless, our observations suggest that a significant proportion of the Winter Mixed
management unit is found very close to shore between Cape Hatteras and Ocracoke Inlet. Our results
also suggest that the previous estimate of 4,734 (CV 0.49) dolphins present off the entire coast of North

17
Carolina during the winter of 1995 (Waring et al., 2002; Garrison, 2002) is likely to have been negatively
biased.

Our results are also consistent with the number of animals taken at Hatteras during the 19th century
fishery. The maximum number of dolphins taken in any single winter season (November to May) was
2,069 in 1886-1887 (Reeves and Read, 2003). During this season, 674 dolphins were taken during
February and March, the months in which we conducted our survey. If the population size of dolphins
today is similar to that in the 19th century, and patterns of habitat use and migrations have not changed
significantly, the Hatteras dolphin fishery likely had a significant effect on the stocks of bottlenose
dolphins using the Outer Banks during winter.

The density of bottlenose dolphins was extremely high along the Outer Banks during our winter field
season; approximately 1,340 dolphins used our 50km2 study area during February and March 2003. It was
not possible for us to calculate an instantaneous density of dolphins along the Outer Banks due to the
analytical approach we used, but it is clear that the density of dolphins was high and decreased as one
moved away from shore. Thirty of our 34 sightings occurred in water less than 5 m deep. Garrison et al.
(2003) also noted that the density of dolphin groups declined exponentially with increasing distance from
shore. Furthermore, mean group size of bottlenose dolphins also declined with distance from shore
during the 2002 winter aerial surveys (Garrison et al., 2003). Combining these results with our
observations, it is clear that the highest densities of dolphins along the Outer Banks during winter are
found just outside the surf zone. This raises questions regarding the efficacy of aerial surveys in the
habitat closest to shore. Coastal aerial surveys are typically flown perpendicular to the shoreline. Thus,
most sightings along the Outer Banks will be made either at the very start or at the end of a transect. The
potential effects of this heterogeneity of dolphin density on aerial survey methods should be examined.

The near-shore waters of the Outer Banks constitute an important habitat for many coastal bottlenose
dolphins during winter. The large number of dolphins in this area must require a high density of prey
species (Gannon and Waples, 2004). We also observed many other upper trophic level predators,
including piscivorous seabirds, fish and humpback whales (Megaptera novaeangliae), during our surveys. The
high biomass of fish in this area also attracts many commercial fishermen. Several important gill net
fisheries are conducted along the Outer Banks during winter (Steve et al., 2001), a number of which take
dolphins as by-catch (Waring et al., 2002). The spatial dispersion pattern of dolphins that we observed, in
which dolphins are clustered near shore, offers some possibility of segregating fishing activities from the

18
highest densities of animals. It might be possible, for example, to restrict fishing activities in waters close
to shore (e.g. within some distance of the beach). We do not yet know whether or not such an approach
would reduce the probability of by-catch, nor what the economic costs of such regulation might entail, but
this possibility merits further investigation.

Finally, we would like to note that many of the observations made by Frederick True (1891) hold true,
more than a century after he visited the village of Hatteras. Much of his record was derived from
conversations with fishermen and he recorded their opinions faithfully in his papers; almost all of these
conclusions are supported by our recent observations. As True noted, the density of dolphins at Hatteras
is highest from November to May and most dolphins can be found within ‘a hundred yards of the shore.’
The fishermen were correct in surmising that dolphins migrate northward in the spring and then return to
Hatteras in the fall, and that a few remain along the Outer Banks during summer. True noted that, in
winter, the trailing edges of many dorsal fins are obscured by barnacles of the genus Xenobalanus, which
greatly complicated our photo-identification efforts. Finally, True (1891) observed fetuses and young
calves of various sizes in the same schools and concluded that “One can readily imagine that during a
migration individuals from different localities would meet and journey together, and that the young in
quite different stages of development might be found in the same school.” Recent observations of the
reproductive seasonality of bottlenose dolphins in North Carolina support the concept of a seasonal
mixture of stocks with differing patterns of reproductive seasonality (Thayer et al., 2003).

ACKNOWLEDGEMENTS

This research was funded by the NC Sea Grant’s Fisheries Resource Grant Program as Project 02-EP-02,
for which Bill Foster was a co-principal investigator. We thank Dave Swanner and Richie Spears who
acted as boat skippers during our surveys. We appreciated the hospitality of Dave Swanner and Teach’s
Lair during our surveys in Hatteras. Jess Maher graded photographic images, labeled slides and helped to
maintain photographic catalogs; Lesley Thorne labeled slides. Contributions to the Virginia Beach portion
of the MABDC were supported by the Virginia Marine Science Museum Foundation Stranding Program.
The Southeast Fisheries Science Center (NOAA Fisheries) supported development of the MABDC from
1997 to 2001. We thank Paul Wade and Ben Wilson for very helpful advice with the experimental design
and analysis.

19
REFERENCES

Agresti, A. 1994. Simple capture-recapture models permitting unequal catchability and variable sampling
effort. Biometrics 50: 494-500.

Brooks, S. P. 1998. Markov Chain Monte Carlo Method and its application. The Statistician 47: 69-100.

Brooks, S. P. and Gelman, A. 1998. General methods for monitoring convergence of iterative simulations.
Journal of Computational and Graphical Statistics 7: 434-455.

Buckland, S. T., Goudie, I. B. J. and Borchers, D. L. 2000. Wildlife population assessment: Past
developments and future directions. Biometrics 56: 1-12.

Chao, A., Lee, S. M. and Jeng, S. L. 1992. Estimating population size for capture-recapture data when
capture probabilities vary by time and by individual animal. Biometrics 48: 201-216.

Coull, B. A. and Agresti, A. 1999. The use of mixed logit models to reflect heterogeneity in capture-
recapture studies. Biometrics 55: 294-301.

Darroch, J. N., Fienberg, S. E., Glonek, G. F. V. and Junker, B.W. 1993. A three-sample multiple-
recapture approach to census population estimation with heterogeneous catchability. Journal of the
American Statistical Association 88: 1137-1148.

Ellison, A. M. 1996. An introduction to Bayesian inference for ecological research and environmental
decision-making. Ecological Applications 6: 1036-1046.

Friday, N., Smith, T., Stevick P. and Allen J. 2000. Measurement of photographic quality and individual
distinctiveness for the photographic identification of humpback whales, Megaptera novaeangliae.
Marine Mammal Science 16: 355-374.

Fienberg, S. E., Johnson, M. S. and Junker, B. W. 1999. Classical multilevel and Bayesian approaches to
population size estimation using multiple lists. Journal of the Royal Statistical Society, Series A 162:
383-405.

Gannon, D. P. and Waples, D. M. 2004. Food habits of coastal bottlenose dolphins from the mid-Atlantic
coast of the U.S. Marine Mammal Science. In Press.

Garrison, L.P. 2001. Seeking a hiatus in sightings of bottlenose dolphins during summer and winter aerial
surveys. Unpublished document presented to the Bottlenose Dolphin Take Reduction Team.
Southeast Fisheries Science Center, Miami, FL.

20
Garrison, L.P. 2002. Update on winter abundance estimate for bottlenose dolphins, Tursiops truncatus, along
the Atlantic coast of the U.S. Unpublished document presented to the Bottlenose Dolphin Take
Reduction Team. Southeast Fisheries Science Center, Miami, FL.

Garrison, L. P., Rosel, P. E., Hohn, A., Baird, R. and Hoggard, W. 2003. Abundance of the coastal
morphotype of bottlenose dolphin, Tursiops truncatus, in U.S. continental shelf waters between New
Jersey and Florida during winter and summer 2002. Unpublished document presented to the
Bottlenose Dolphin Take Reduction Team. Southeast Fisheries Science Center, Miami, FL.

Gelman, A. and Rubin, D. B. 1992. Inference from iterative simulation using multiple sequences (with
discussion). Statistical Science 7: 457-511.

Gelman, A., Carlin, J. B., Stern H. S. and Rubin. D. B. 1995. Bayesian Data Analysis, London: Chapman
and Hall.

Gelman, A., Meng, X.L. and Stern, H.S. 1996. Posterior predictive assessment of model fitness via realized
discrepancies (with discussion). Statistica Sinica 6: 733-807.

Mead, J. G. 1975. Preliminary report on the former net fisheries for Tursiops truncatus in the western North
Atlantic. Journal of the Fisheries Research Board of Canada 32: 1155-1162.

Meng, X.L. 1994. Posterior predictive p-values. Annals of Statistics 22: 1142-1160.

Mitchell, E. 1975. Porpoise, dolphin and small whale fisheries of the world. Status and problems.
International Union for the Conservation of Nature Resource Monographs 3: 1-129.

Pledger, S. 2000. Unified maximum likelihood estimates for closed capture-recapture models using

mixtures. Biometrics 56: 434-442.

Pollock, K. H., Nichols, J. D., Brownie, C. and Hines, J. E. 1990. Statistical inference for capture-recapture

experiments. Wildlife Monographs: 107.

Rasch, G. 1960. Probabilistic Models for Some Intelligence and Attainment Tests, Chicago: University of
Chicago Press.

Read, A. J., Urian K. W., Wilson B. and Waples D. M. 2003. Abundance of bottlenose dolphins in the
bays, sounds and estuaries of North Carolina, USA. Marine Mammal Science 19: 59-73.

Reeves, R. R. and Read, A. J. 2003. Bottlenose dolphin, harbor porpoise, sperm whale and other toothed
cetaceans. Pp. 397-424 in: G.A Feldhammer, B.C. Thompson & J.A. Chapman (editors). Wild

21
 Mammals of North America: Biology, Management, and Conservation. Second Edition. The Johns
Hopkins Press, Baltimore.

Spiegelhalter, D. J., Best, N.G., Carlin, B.P. and van der Linde, A. 2002. Bayesian measures of model
complexity and fit (with discussion). Journal of the Royal Statistical Society, Series B 64: 583-640.

Steve, C., Gearhart, J., Borggaard, D., Sabo, L. and Hohn, A. A. 2001. Characterization of North Carolina
commercial fisheries with occasional interactions with marine mammals. NOAA Technical
Memorandum NMFS-SEFSC-458. 60 pp.

Thayer, V. G., Read A. J., Friedlaender A. S., Colby D. R., Hohn A. A., McLellan W. A., Pabst D. A. ,
Dearolf J. L., Bowles N. I., Russell J. R. and Rittmaster K. A. 2003. Reproductive seasonality of
bottlenose dolphins, Tursiops truncatus, in North Carolina. Marine Mammal Science 19: 617-629.

Torres, L. G., Rosel P. E., D’Agrosa C. and Read A. J. 2003. Improving management of overlapping
bottlenose dolphin ecotypes through spatial analysis and genetics. Marine Mammal Science 19:
502-514.

True, F. W. 1885. The porpoise fishery of Cape Hatteras. Field and Stream 24: 412-413.

True, F. W. 1891. Observations of the life history of the bottlenose porpoise. Proceedings of the U.S.
National Museum 13: 197-203.

Urian, K. W., Hohn A. A. and Hansen L. J. 1999. Status of the photo-identification catalog of
coastal bottlenose dolphins of the western North Atlantic: report of a workshop of catalog
contributors. NOAA Technical Memorandum NMFS-SEFSC-425.

Wade, P. R. 2000. Bayesian methods in conservation biology. Conservation Biology14: 1308-1316.

Waring, G. T. et al. 2002. U.S. Atlantic and Gulf of Mexico marine mammal stock assessments – 2002.
NOAA Technical Memorandum, NMFS-NE-169. Northeast Fisheries Science Center, Woods
Hole, MA.

22
Table 1. Number of encounters, number of dolphins and number of photographic images taken during
photographic mark-recapture surveys of bottlenose dolphins off the Outer Banks of North Carolina in
February and March 2003. Number of dolphins refers to field estimates of dolphins observed during
surveys.

Survey Hours Number of Number of Number of

Date (h) Encounters Dolphins Images

14-Feb-03 8:57 8 350 671

19-Feb-03 9:17 5 132 418
4-Mar-03 7:13 5 195 331
8-Mar-03 5:03 3 102 165
11-Mar-03 4:19 1 50 78
23-Mar-03 7:54 4 126 236
24-Mar-03 3:37 3 57 277
25-Mar-03 6:43 4 260 683
26-Mar-03 1:38 1 30 48

Total 54:41 34 1,302 2,907

23
Table 2. Temporal and geographic coverage of photographic mark-recapture surveys of bottlenose
dolphins off the Outer Banks of North Carolina in February and March 2003. See Figure 2 for definition
of Zones.

Date Zone 1 Zone 2 Zone 3 Zone 4

14-Feb-03 X
19-Feb-03 X X X
4-Mar-03 X X X
8-Mar-03 X X
11-Mar-03 X
23-Mar-03 X X X X
24-Mar-03 X
25-Mar-03 X X
26-Mar-03 X

24
Table 3. Photographic components of the Mid-Atlantic Bottlenose Dolphin Catalog compared with
images taken during photographic mark-recapture surveys of bottlenose dolphins off the Outer Banks of
North Carolina in February and March 2003.

Catalog
Field site size Period Contributor Affiliation

Cape May, NJ 32 1991-1996 R. Mallon-Day Cape May Dolphin Survey

Cape May, NJ 38 Sep-2002 K. Rittmaster NC Maritime Museum
Wallops Island, VA 21 1997-1998 D. Schofield Marine Science Consortium
Virginia Beach, VA 303 1989-1998 S. Barco Virginia Marine Science Museum
Roanoke Sound, NC 138 1997-1998 R. Mallon-Day Nags Head Dolphin Watch
Pamlico Sound, NC 358 1998-2002 A. Read Duke University Marine Laboratory
Beaufort, NC 559 1985-1998 K. Rittmaster NC Maritime Museum
Wilmington, NC 376 1991-2002 L. Sayigh/G. Rountree UNC-Wilmington
Murrell's Inlet, SC 85 1997-1999 R. Young Coastal Carolina University

25
Table 4. Estimates of model parameters from MCMC sequences when fitting the Bayesian Rasch models
to photographic capture-recapture data for bottlenose dolphins off the Outer Banks. Estimates are
presented for both the univariate Rasch (UVR) model and the multivariate Rasch model (MVR) with the
survey occasions partitioned into three strata. For both models, estimates are presented as medians (95%
probability interval) for the marginal posterior distributions for the overall mean level of the capture
probability on the logit scale, µ, the variance of the random effects distributions for survey effects, σ2α ,
the number of undetected individuals, n0, and the estimated abundance of reliably marked dolphins, N.
The single variance of the individual effects distribution, σ2θ, is presented for the UVR model, and the
variance of each of the three individual effects strata (Σ1,1, Σ2,2, Σ3,3) is presented for the MVR model.

Model µ σ2α σ2θ Σ1,1 Σ2,2 Σ3,3 n0 N

UVR -4.4 0.83 1.01 - - - 1313 1611

(-5.1, -3.6) (0.32, 2.94) (0.90, 1.20) (926, 1868) (1224, 2166)

MVR -3.3 0.82 - 0.15 0.18 0.13 519 817

(-4.0, -2.6) (0.31, 2.82) (0.08,0.30) (0.09,0.37) (0.07,0.25) (371, 732) (669,1030)

26
Table 5. The co-variance matrix for the catchability of observed individuals estimated under the
multivariate Rasch model. The matrix depicts the values for Σ, describing the association between the
vector of individual effects for the q = 3 strata of survey periods. The off-diagonal values represent co-
variances between pairs of strata.

Stratum 1 2 3

1 - -0.19 -0.21
(-0.75, 0.11) (-0.61, 0.04)
2 -0.19 -0.03
(-0.75, 0.11) - (-0.30, 0.17)
3 -0.21 -0.03 -
(-0.61, 0.04) (-0.30, 0.17)

27
Figure 1. Management units of coastal bottlenose dolphins along the Atlantic coast of the U.S. as defined
by NMFS (from Waring et al., 2002).

28
Figure 2. Study area for mark-recapture photo-identification surveys conducted off the Outer Banks of
North Carolina during February and March, 2003. Survey zones (1 to 4) are demarcated by lines and
identified by number. Dolphin sightings are indicated by circles.

35°10’N

75°45’W

29
Figure 3. Posterior probability distribution for the estimated number of dolphins, P, using near-shore
waters of the Outer Banks of North Carolina in February and March 2003.

3000
Probability Density

2000

1000

0
400 600 800 1000 1200 1400 1600 1800 2000 2200

Dolphin Abundance , P

30
 Figure 4. Proportion of matches between dolphins identified off the Outer Banks, North Carolina in
February and March, 2003, and other sites in the Mid-Atlantic Bottlenose Dolphin Catalog. The number
of individual dolphins represented in each catalog is listed below the name of the respective site. The star
indicates the location of our winter surveys along the Outer Banks. Proportions were calculated as:
[(Number of Matches) * 2]/(Catalog Size at Site X + Catalog Size of Outer Banks).

NJ
0 0.05 0.1 0.15 0.2 0.25

38°55’N
Cape May, NJ
(70)

Wallops Is. VA
Virginia (21)

Virginia Beach, VA
(303)

36°33’N
Roanoke Sound, NC
(138)
North
Carolina Pamlico Sound, NC
(358)

Beaufort, NC
(559)

SC Wilmington, NC
33°30’N (376)

Murrell's Inlet, SC
(85)

78°00’W 74°55’W

31
 Figure 5. The proportion of matches made to each site made during winter (blue) and summer (yellow)
for dolphins identified off the Outer Banks, North Carolina in February and March, 2003, and other sites
in the Mid-Atlantic Bottlenose Dolphin Catalog. Winter is comprised by the months of November to
April and Summer from May to October. The star indicates the location of our winter surveys along the
Outer Banks. Proportions were calculated as: number of matches from the Outer Banks at each site/234
(the total number of dorsal fins evaluated).

NJ 0.000 0.050 0.100 0.150 0.200 0.250

38°55’N
Cape May, NJ

Wallops Is. VA
Virginia
Virginia Beach

Roanoke Sound, NC
36°33’N

North Pamlico Sound, NC

Carolina
Beaufort, NC

Wilmington, NC
SC
33°30’N Murrell's Inlet

Summer Winter

78°00’W 74°55’W

32
 Appendix 1
Measurement of Photographic Quality and Dolphin Distinctiveness
for the Mid-Atlantic Bottlenose Dolphin Photo-ID Catalog
Kim Urian, Curator

OVERALL PHOTOGRAPHIC QUALITY

Overall Photographic Quality is based on the quality of the photograph independent of the distinctiveness of the fin.

The Overall Photographic Quality score is based on an evaluation and sum of the following characteristics (these scores are
absolute values, not a sliding scale):
• Focus/Clarity
Crispness or sharpness of the image. Lack of clarity may be caused by poor focus, excessive enlargement, poor
developing or motion blur; for digital images, poor resolution resulting in large pixels.

Based on the scale: 2 = excellent focus 4 = moderate focus 9 = poor focus, very blurry

• Contrast
Range of tones in the image. Images may display too much contrast or too little. Photographs with too much contrast
lose detail as small features wash out to white. Images with too little contrast lose the fin into the background and
features lack definition.

Based on the scale: 1 = ideal contrast 3= either excessive contrast or minimal contrast

• Angle
Angle of the fin to the camera.

Based on the scale: 1 = perpendicular to camera 2 = slight angle 8 = oblique angle

• Partial
A partial rating is given if so little of the fin is visible that the likelihood of re-identifying the dolphin is compromised
on that basis alone. Fins obscured by waves, Xenobalanus, or other dolphins, would be evaluated using this rating.

Based on the scale: 1 = the fin is fully visible, leading & trailing edge 8= the fin is partially obscured

• Proportion of the frame filled by the fin

An estimate of the percentage area the fin occupies relative to the total area of the frame.

Based on the scale: 1 = greater than 5%; subtle features are visible 5 = less than 1%; fin is very distant

To score Overall Photographic Quality, sum the scores for each characteristic:

6 - 9: Excellent quality => Q-1

10–12: Average quality => Q-2
>12 : Poor quality => Q-3

OVERALL DISTINCTIVENESS
Overall Distinctiveness is based on the amount of information contained on the fin; information content is drawn from
leading and trailing edge features, and pattern, marks, and scars.

D-1 - Very distinctive; features evident even in distant or poor quality photograph
D-2 - Average amount of information content: 2 features or 1 major feature are visible on the fin
D-3 - Not distinctive; very little information content in pattern, markings or leading and trailing edge features
These measurements are derived from:
Friday et al. 2000. Measurement of photographic quality and individual distinctiveness for the photographic identification of humpback whales,
Megaptera novaeangliae. Marine Mammal Science 16: 355-374.

33