Content uploaded by J. Christopher Haney
Author content
All content in this area was uploaded by J. Christopher Haney on May 28, 2019
Content may be subject to copyright.
MARINE ECOLOGY PROGRESS SERIES
Mar Ecol Prog Ser
Vol. 513: 239–252, 2014
doi: 10.3354/meps10839 Published October 22
INTRODUCTION
Large oil spills into the coastal ocean can kill hun-
dreds of thousands of birds (Piatt & Ford 1996, Tan et
al. 2008, Munilla et al. 2011). Petroleum exposure
alters feather microstructure (Jenssen 1994, O’Hara
& Morandin 2010), compressing plumage so that it
loses its buoyancy, insulating function, and flight
capability (Leighton 1993). Physiological health of
birds is further impaired by oil-induced diseases
(Briggs et al. 1996), including hemolytic anemia,
ulcerations, cachexia, and aspergillosis (Balseiro et
al. 2005). Birds contaminated at sea thereby die due
to drowning, hypothermia, starvation, or dehydration.
The largest accidental release of petroleum into
marine waters in history (Camilli et al. 2012, McNutt
et al. 2012), the Deepwater Horizon MC 252 blowout
was an unprecedented perturbation to the northern
Gulf of Mexico. The cumulative slick area exceeded
100 000 km2(Norse & Amos 2010, Garcia-Pineda et
al. 2013). Oiling impacts extended from the pelagic
ocean seaward of the continental slope, across the
© The authors 2014. Open Access under Creative Commons by
Attribution Licence. Use, distribution and reproduction are un -
restricted. Authors and original publication must be credited.
Publisher: Inter-Research · www.int-res.com
*Corresponding author: chaney@defenders.org
Bird mortality from the Deepwater Horizon oil spill.
II. Carcass sampling and exposure probability in the
coastal Gulf of Mexico
J. Christopher Haney1,4,*, Harold J. Geiger2, Jeffrey W. Short3
1Terra Mar Applied Sciences LLC, 123 West Nye Lane, Suite 129, Carson City, Nevada 89706, USA
2St. Hubert Research Group, 222 Seward, Suite 205, Juneau, Alaska 99801, USA
3JWS Consulting LLC, 19315 Glacier Highway, Juneau, Alaska 99801, USA
4Present address: Defenders of Wildlife, 1130 17th St. NW, Washington, DC 20036, USA
ABSTRACT: Two separate approaches, a carcass sampling model and an exposure probability
model, provided estimates of bird mortalities of 600 000 and 800 000, respectively, from the 2010
Deepwater Horizon MC 252 oil spill in coastal waters of the Gulf of Mexico. Monte Carlo simula-
tion of parameter uncertainty led to respective 95% uncertainty intervals of 320 000 to 1200 000
and 160 000 to 1900 000. Carcass sampling relied on expansion factors multiplied by counts of bird
carcasses retrieved in shoreline surveys, whereas exposure probability estimated bird deaths as a
product of estimated coastal bird density, average oil slick size, slick duration, and proportionate
mortality due to oiling. The low proportion of small-sized carcasses recovered, compared with
considerably higher proportions of small live birds in coastal Gulf habitats, indicate an especially
low probability of recovery for small birds after oil spills at sea. Most mortality affected 4 species:
laughing gull Leucophaeus atricilla (32% of the northern Gulf of Mexico population killed), royal
tern Thalasseus maximus (15%), northern gannet Morus bassanus (8%) and brown pelican Pele-
canus occidentalis (12%). Declines in laughing gulls were confirmed by ~60% reductions in
National Audubon Society Christmas Bird Count data for 2010−2013 along the Gulf coast. Popu-
lation-level effects in apex predators of this magnitude likely had effects on prey populations that
warrant careful assessment.
KEY WORDS: Avian mortality · Exposure probability · Carcass sampling · Oil spill · Deepwater
Horizon · Gulf of Mexico · Coastal habitat · Christmas Bird Count · Monte Carlo simulation
O
PEN
PEN
A
CCESS
CCESS
Mar Ecol Prog Ser 513: 239–252, 2014
continental shelf, and into remote brackish estuaries
along the Gulf coastline (Peterson et al. 2012, Michel
et al. 2013).
Avian mortality is one direct, immediate measure
of ecological impact caused by a marine oil spill.
Because it is impossible to observe each seabird
death from acute oil exposure, total mortality is usu-
ally inferred from shoreline carcass surveys and
probability-based expansion factors to account for
birds that are killed but not collected (e.g. an Oiled
Seabird Mortality Model; Wiese & Robertson 2004).
These factors account for birds that die but disappear
before arriving on shore (Wiese 2003, Munilla et al.
2011) and those on shore that go undetected by spill
responders (Van Pelt & Piatt 1995, Byrd et al. 2009).
Alternatively, seabird mortality can be estimated
from the numbers of birds present and vulnerable to
lethal exposure (Wilhelm et al. 2007, Haney et al.
2014, this volume).
Our objective here is to estimate coastal seabird
mortality from acute oil exposure during the Deep-
water Horizon spill. We compare mortality estimates
derived from both a carcass sampling model and an
exposure probability model. The carcass sampling
model includes expansion factors for lost, missed,
and unobservable bird carcasses. The exposure pro -
bability model is based on surveys of aerial seabird
densities above coastal waters of the northern Gulf of
Mexico, estimates of the size of the Deepwater Hori-
zon oil slick, and an estimate of the proportion of
oiled seabirds that subsequently died. The 2 ap -
proaches are based on largely independent data,
providing an unusual opportunity to compare alter-
native models of estimating avian mortality after a
large marine oil spill.
MATERIALS AND METHODS
Study area and modeling domain
Spatially, we limited the scope of investigation to
coastal waters within 40 km offshore of the Gulf coast
(see Fig. 1 in Haney et al. 2014), thus delineating an
effective catchment area (sensu Wiese & Robertson
2004). We chose 40 km as the likely maximum
distance from shore inhabited routinely by coastal
seabirds.
The Deepwater Horizon discharged oil into the
Gulf from the day when the casing was breached on
20 April 2010 until the well was capped on 15 July
2010. Oil first appeared within 40 km of the coast on
28 April 2010. We thus considered the acute mortal-
ity phase to last for 95 d, until 31 July 2010, to account
for bird mortality from contact with lingering surface
oil (Aeppli et al. 2012). Avian mortality clearly con -
tinued after the well was capped (based on wildlife
collection reports) because the ratio of dead-to-live
bird recoveries increased in late July 2010 (Belanger
et al. 2010, Antonio et al. 2011).
Bird carcass tallies by size category and location
We obtained carcass counts retrieved in all habitats
during the Deepwater Horizon incident and archived
by spill response authorities (www.fws.gov/ home/
dhoilspill/ collectionreports.html, table dated 12 May
2011; accessed 22 March 2013). Using a total of 2121
carcasses identified as known oiled, we organized
these counts by species and body size. Live moribund
oiled birds were not counted as carcasses because
they could reach shorelines independent of winds
and currents (e.g. Stienen et al. 2004).
To account for variation in detectability with body
size during spill response (Seys et al. 2001, Ford &
Zafonte 2009), each bird species and all individual
birds were assigned to one of 3 size categories: large
(≥500 g), medium (300−499 g), or small (10−299 g).
We excluded 92 bird carcasses not considered as
coastal on the basis of primary habitat used by
each species, life history traits, and migratory habits.
Species of estuarine categorization included rails,
galli nules, certain waders, and species that occur
in terrestrial vegetation types.
Species that feed in the neritic zone and associated
shorelines, including shorebirds and other seasonal
migrants that rely upon beach or wide mudflats dur-
ing the time of year of the spill, were broadly desig-
nated as coastal for purposes of oil exposure. Birds
listed in the recovery archives as ‘other’ (a total of 31)
and ‘unknown’ (a total of 51) were categorized as
coastal and medium sized. Other birds tallied un -
known to species were assigned to the same size
category and habitat as the most commonly re -
covered species in that taxon (see Table S1 in the
Supplement at www.int-res. com/ articles/ suppl/ m513
p239_ supp. pdf). Dead birds assigned to the coastal
category but retrieved offshore (a total of 25 car-
casses recovered before they reached shore; see
www.fws.gov/home/ dhoilspill/ collectionreports. html),
were also subtrac ted from the carcass tally. A total of
2004 bird carcasses constituted the observations that
represented the recoveries of birds from the coastal
habitat, comprising 128 small-, 1383 medium- and
493 large-bodied bird carcasses.
240
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal 241
Carcass sampling model
From a sampling perspective, carcass counts can
be thought of as the realization of repeated stochastic
sampling steps, with each step reducing the number
of carcasses ultimately available for recovery (e.g.
Seys et al. 2001, Wiese & Robertson 2004). The
unknown number of bird deaths in each body size
category from the Deepwater Horizon spill was de -
noted as Nj, and the number of carcasses counted as
xj, with the index jindicating body size category (1:
large; 2: medium; 3: small). Within each body size
category, a series of 5 stochastic steps reduced the car -
casses available to be recovered by some proportion.
We denote ras the probability that a carcass was
transported and deposited on shore following a bird’s
death from oil, lthe probability that a carcass per-
sisted on shore until it could be found, hthe pro -
portion of habitat sampled for bird carcasses, kthe
proportion of the total carcass deposition period sys-
tematically sampled with a fixed interval between
successive sampling events, and dthe probability
that a carcass would be detected by searchers.
Parameters r, l, h, and kwere assumed to be the same
for all bird body size categories. Parameter dwas
assumed to vary by size category, with d1, d2, and d3
the probabilities of detection that correspond to x1,
x2, and x3. For convenience, we let pjdenote the com-
bined probability that a bird from size category jthat
died as a result of oil spill exposure was eventually
recovered. Also, we used the caret notation to indi-
cate between an estimate (e.g, p
ˆj) of a parameter and
we leave off the caret to denote either the true value
or else the parameter as a random variable (in the
Bayesian sense, e.g. pj). Then, in a sampling context
(Thompson 2002), the estimated number of total
birds killed within size category jis
(1)
The grand total number of birds estimated killed is
the sum of each N
ˆj.
Parameter selection
Carcasses adrift will vary in their final destination
as a consequence of daily changes in wind speed and
direction. Shelf currents in the northern and eastern
Gulf of Mexico are dominated by wind and fresh-
water driven flows (Barker 2011). Surface oil reached
Gulf shorelines only during strong onshore winds
(MacFadyen et al. 2011). We therefore estimated car-
cass drift from wind data, because neritic currents
from the Mississippi and Atchafalaya River discharges
are primarily alongshore to the west in response to
Coriolis forcing.
The probability rthat a carcass was transported to
shore depends on its initial position offshore, the car-
cass drift velocity towards a shoreline, and the prob-
ability of remaining afloat en route. We assumed the
initial carcass position was proportional to the prod-
uct of the relative area densities of seabirds and sur-
face oil as a function of distance from shoreline, aver-
aged over the 95 d when seabird carcasses were
presumed present within 40 km of the shoreline. We
assumed that coastal seabird density declined from a
maximum at the shoreline to some negligibly low
value 40 km offshore, proportional to 1 − z/40, where
zindicates the distance from shoreline in km. Daily
oil slick size was calculated from shape files that
were synthesized in the Experimental Marine Pollu-
tion Surveillance Daily Composite Products (www.
ssd. noaa.gov/PS/MPS/deepwater.html; accessed 11
Au gust 2013). Spatial depictions of the Deepwater
Horizon oil slick in these products were based prima-
rily on satellite sensors augmented with oil spill tra-
jectory models and other ancillary data (see ’Parame-
ter selection for exposure probability model’ in the
Supplement). The initial location of seabird carcasses
was then assumed to be proportional to the product
of the risk of oiling and the assumed seabird density
of 1 − z/40 (Table S2 in the Supplement).
We estimated shoreward carcass drift velocity from
observations of wind velocities archived for Station
BURL1, Southwest Pass, Louisiana, a C-MAN Station
maintained by the National Data Buoy Center. We
assumed a 2% coupling of wind and seabird carcass
drift speeds (Seys et al. 2001, Castege et al. 2007),
and a Coriolis deflection for seabird carcass drift of
18° to the east (Poulain et al. 2009). Shorelines where
seabird carcasses were deposited were as sumed to
face either southward (80%) or eastward (20%). Sea-
bird carcass drift speeds were computed as the
respective northward and westward components of
the drift velocity vector, and the average values of
these velocities vNand vWwere determined. The
overall average drift velocity, v--, of 4.1 km d−1 was
computed as v-= 0.8 v-
N+ 0.2 v-
W(Table S3 in the Sup-
plement).
We assumed carcasses lost buoyancy from decom-
position at an instantaneous rate of 1.00 d−1 (see
‘Computation of transport probability to shorelines’
in the Supplement), based on results reported by
Ford et al. (1991) and Wiese (2003) after accounting
for temperature differences and for likely effects of
ˆˆˆˆˆ ˆ ˆ
Nx
rlhkd
x
p
j
j
j
j
j
==
Mar Ecol Prog Ser 513: 239–252, 2014
scavengers that increase carcass decomposition rates.
The probability that a carcass would remain afloat
after a time tat sea was computed as exp[−(1.00 t)],
where t= z/4.1 km d−1. The probability rthat a car-
cass was transported to shore was computed as a
numerical approximation of the integral of the prod-
uct of this exponential function and the distribution
of initial carcass positions at sea as described above
(Table S4 in the Supplement).
Based on these assumptions regarding carcass losses
at sea, <1% of the carcasses would remain available
for shoreline deposition after 4 d. Inte grating these
time-dependent losses with the initial distribution of
carcasses led to an estimated average proportion of
carcasses reaching a shoreline of approximately rˆ =
0.057 (Table S4). Most of this transport probability
arises from carcasses within 15 km of the shore,
beyond which the average transport time of ~4 d or
more leads to nearly complete removal of carcasses
from the sea surface. The parameter rwas assigned a
beta probability distribution with the mean centered
at 0.057 to model uncertainty in this parameter.
To estimate the probability of carcass persistence
on a beach prior to search and recovery, l, we as -
sumed that a carcass survey was conducted every
3 d, as was originally planned (USDOI 2011). We
took the daily probability of carcass persistence to be
0.50 to account for removals by scavengers or burial
by wind- or wave-driven sediment transport, a rea-
sonable approxi mation based on other studies (Page
et al. 1990, Seys et al. 2001). We assumed that the
probability of a carcass persisting ndays was 0.50n
(n ≤3). Probability of carcass persistence, l, within
this deposition interval was thus the average of 1/2,
1/4 and 1/8, or 0.292. We assigned this variable a
beta distribution with a mean centered at 0.292 to
model uncertainty.
We assumed that all Gulf shorelines exposed to
neritic waters of the Gulf of Mexico and suitable for
carcass deposition were searched, so h= 1.00. We
also assumed that 90% of the shoreline surveys for
carcasses were separated by intervals of 3 d so k=
0.90. We assigned this variable a beta distribution
with the mean centered at 0.90 to model uncertainty.
Estimates of searcher efficiency (dj) were based on
Ford et al. (2013). We used a value of 0.424 (= d
ˆ1) as
the estimate of searcher efficiency for large bird car-
casses. Using the average of the 2 values for searcher
efficiency reported in Table 2 from Ford et al. (2013)
for small bird carcasses, we assumed d
ˆ3= 0.078.
Finally, we assumed the average of the values for
small and large bird carcasses, d
ˆ2= 0.251, as the
searcher efficiency for medium-sized carcasses. These
searcher efficiency parameters were also assigned
beta distributions, with the means centered at the
values listed above.
Parameter uncertainty
The beta probability distributions assigned to each
unknown parameter represent our attempt to organ-
ize and display what we do and do not know about
the parameter’s location (Silver 2012), and to assess
what would have happened had we used alternate
assumptions about the parameter values. To ex -
plain how the variances of these distributions were
derived, let qrepresent any of the carcass model
parameters. Because these are beta distributions for
parameters, the associated distribution for qhas
hyperparameters aqand bq. The mean of the beta dis-
tribution is given by aq/(aq+ bq) and the variance is
also a function of aqand bq(Casella & Berger 2002).
We linked all of the beta distributions together, for all
of the parameters, through a single common hyper-
parameter, u, such that aq= uqˆand bq= u(1 − qˆ), for
q
ˆrepresenting the estimate of r, l, k, and each value
of dj. Without this step we would have had twelve
individual hyperparameters to manipulate (2 hyper-
parameters for each beta distribution). Note that u
can be factored out of both numerator and denomina-
tor of the mean, so the mean is independent of u.
However, the variance is strongly affected by u: the
larger the value of u, the smaller the variance of each
parameter.
The value of hyperparameter uwas chosen so that
distributions of all 6 model parameters were as con-
sistent as possible with uncertainty in parameters
described in the literature and with estimates from
this and other oil spills (Table 1). Rather than develop
even more complexity to express the uncertainty in
u, we simply set the value of uat 200 after acknowl-
edging that the distribution of Nis sensitive to this
choice. We established the value of uat 200 by
iterative fitting. That is, we varied u, examined the
central 95% interval for each of the variables, and
then readjusted uuntil all 6 probability intervals
closely approximated the 95% interval for parameter
values that we considered plausible for likely alter-
nate parameter assumptions (Table 1). Later, we dou-
bled and halved this value of uto assess the sensiti -
vity of the analysis to our choice of u(and to the
assumed variance of the beta distributions, which u
controls).
After setting the value of uat 200, we evaluated the
overall uncertainty in our estimates by making re -
242
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal
peated random draws from the beta
distributions of each carcass model
para meter and then calculating Nj
and summing over j. The Monte Carlo
distribution for Nwas developed by
taking 1 million random draws for
each parameter (r, l, h, k, dj), then
devel oping 1 million values of Nby
the repeated use of Eq. (1) and sum-
mation over j.
Exposure probability model
Bird deaths from a spill can also be
estimated using bird density (D), the
proportionate mortality due to oiling
(M), and the spatial extent of the oil
(A). We assume that birds become
contaminated with oil in direct pro-
portion to the product of the spatial
extent of oiling on the water and the
density of seabirds within that spatial
extent. The number of affected birds in contact with
the oil multiplied by the proportionate mortality (bird
deaths/oil-exposed bird) should then approximate N,
the number of bird deaths, so N= ADM (e.g. Wilhelm
et al. 2007, Haney et al. 2014).
Estimating seabird mortality as the product of oil
slick area, seabird density and proportionate mor -
tality due to oiling presumes exposure popu lation
size equal to AD for an exposure period suf ficient
to result in proportionate mortality M. This simple
relation for estimating mortality due to oil exposure
must be modified to account for exposure of new
populations of birds owing to oil slick movement
and to replacement of birds killed by oil exposure
through immigration to the slick area, especially given
the ~3 mo persistence of the Deepwater Horizon
oil slick.
To account for the total population of birds exposed
to oil, we introduce an effective exposure renewal-
period parameter, P, after which the population of
birds exposed to oil is effectively renewed. Denoting
the duration of the oil spill as T, there are T/Psuch
exposure renewal periods. If the area of the oil slick
during the ith exposure renewal period is denoted as
Ai, the number of birds killed during that period is
Ni= AiDiMi, where Diand Miare the bird density
and proportionate mortality during that period. The
total number of birds killed, N, is:
(2)
If Diand Miare assumed constant over the duration
of the oil spill T, it can readily be shown that
reduces to A
-
(T/P), where A
-is the average oil slick
area over the duration of the spill (Haney et al. 2014).
Then the number of birds killed is simply:
(3)
Parameter selection
We set value of the period Pto 1 d based in part on
information from satellite imagery for oil spill per -
sistence and recurrence in coastal waters (see the
Supplement for more detailed justifications of para -
meter values). We estimated the average extent of
the surface oil slick (A
-) as 3600 km2for coastal waters
of the northern Gulf of Mexico spill zone based on
methods described above for the carcass sampling
model. To simulate the uncertainty in A
-, this para -
meter was treated as a random variable (in the
Bayesian sense), and assigned a gamma dis tribution
with shape parameter of 36 and a scale parameter of
100 (i.e. mean and standard deviation approx. 3600
and 600 km2d−1, respectively). A gamma distribution
was chosen because it has an inherent lower bound
of zero but is otherwise similar to the normal distribu-
tion with the parameter values chosen. Two standard
deviations constituted ca. 33% of the mean, which
reflects uncertainty associated with the overestima-
NADM
TP
ii i
=Σ1
NADM
T
P
=⎛
⎝
⎜⎞
⎠
⎟
Σ1
TPAi
243
Parameter Assigned Mean of 2.5th−97.5th Median expan-
distribution distribution quantiles sion factor
Carcass sampling model
rBeta 0.057 0.029−0.093 18
lBeta 0.292 0.231−0.356 3.5
kBeta 0.900 0.855−0.938 1.1
d1Beta 0.424 0.357−0.493 2.4
d2Beta 0.251 0.194−0.313 4
d3Beta 0.078 0.045−0.119 13
Exposure probability model
A
-Gamma 3600 2520−4780 NA
DPoisson 5.8 2−11 NA
MBeta 0.4 0.177−0.659 NA
Table 1. Key features for the probability distributions used to simulate the
number of coastal bird deaths caused by the Deepwater Horizon oil spill. The
following parameters constitute the random variables for the carcass sampling
model: r: probability of a bird carcass transported to shore following death;
l: probability of the bird carcass persisting between survey periods; k: tem -
poral fraction of the spill that was sampled; dj: probability of detection of a
carcass (j= 1, 2 or 3 denoting large [≥500 g], medium [300−499 g] or small
[10−299 g] carcasses, respectively). For the exposure probability model:
A
-: average oil slick size (km2); D: bird density (birds km−2); M: proportionate
mortality for birds
Mar Ecol Prog Ser 513: 239–252, 2014
tion from pixel distortion and analyst misclassifi -
cation during satellite image as sessment for oil
presence (e.g. Haney et al. 2014).
We assumed that the mean density of birds, D, was
5.8 birds km−2 based on McFarlane & Lester (2005),
who gave a range of 3.6 to 9.4 birds km−2. We
assumed the density of seabirds in the coastal Gulf
of Mexico remained at approximately this level be -
cause of several movement processes available for
repopulating the spill zone. Birds in this area had
flight speeds that facilitated influx to any part of the
spill zone in less than 24 h. Also, aerial foragers (spe-
cies that use surface plunging, aerial dipping, aerial
pursuit, skimming and hydroplaning; Nelson 1979)
accounted for 96% of all species in the coastal sea-
bird community in the Gulf, whereas more sedentary
divers and other primarily surface foragers (a group
mutually exclusive from aerial foragers, and that
included loons, sea ducks, and phalaropes) are a
negligible fraction (Johnson 2011). Fi nally, new birds
arrived continuously into the region during the inci-
dent period via ongoing seasonal migrations. Seabird
abundance in the Gulf of Mexico increases 17% from
spring to summer (Peake 1996), and certain taxa (e.g.
storm-petrels, shearwaters, and some terns) become
4 to 5 times more abundant (Hess & Ribic 2000). This
random variable Dwas assigned a Poisson distribu-
tion (see Clarke et al. 2003, Oppel et al. 2012) with
mean and variance of 5.8 birds km2to reflect the
observed range of seabird density in coastal waters
(McFarlane & Lester 2005).
We selected a value of 0.40 (i.e. 40%) as an esti-
mate of the proportionate mortality, M, in part based
on the observed proportionate mortality in the oiled
birds retrieved during the Deepwater Horizon (justi-
fication detailed in the Supplement). We also consid-
ered the proportionate mortality of aerial seabird
species reported from other oil spills. Proportionate
mortalities from 5% to near 90% from oil exposure
are reported for marine birds, with rates of 22 to 89%
(median 61%) listed for 13 aerially foraging seabird
groups (Camphuysen & Heubeck 2001). This para -
meter was assigned a beta distribution with para -
meters a= 6 and b= 9 (mean of 0.40 and standard
deviation of 0.12) so that most values in the distribu-
tion were below the median value of reported esti-
mates noted above.
Parameter uncertainty
As in the carcass sampling approach, we used a
Bayesian notion of probability to organize and weight
plausible assumptions about alternative para meter
values and to reflect our knowledge of para meter lo-
cation based on available literature and observations
from the Deepwater Horizon spill. A Monte Carlo
distribution for the unknown number of bird deaths
estimated by the exposure probability model was
then developed by taking 1 million rep licates from the
distribution of the parameters (Table 1), then generat-
ing 1 million values using Eq. (3).
RESULTS
Estimates of total bird deaths
The carcass sampling and exposure probability
approaches produced similar estimates for the num-
ber of bird deaths. Dividing the number of recov-
ered carcasses by the estimated expansion factors
(Table 2) gave an estimated 600 000 bird deaths.
Alternatively, assuming an average slick size of
3600 km2, a proportionate mortality of 0.40 bird
deaths per exposed bird, and a bird density of
5.8 birds km−2, we estimated slightly over 8000 bird
deaths d−1, on average. Expanding for all 95 d having
an observed oil slick within 40 km of the coast, the
number of bird deaths was estimated as approx.
800 000.
Notably, with the carcass sampling approach,
expansion factors ranged from slightly under 200 to
over 800 unobserved carcasses per each recovered
carcass (Table 2), depending on size category. The
largest expansion factor was for birds with small
body sizes, the size category with the fewest number
of carcasses found (6.4% of all recoveries). Numbers
of bird deaths in the small body size category were
ultimately estimated to be slightly over 15% of the
total.
The Monte Carlo distribution of the number of bird
deaths from carcass sampling had a mean of 630 000
and a median of 590 000, indicating the distribution
was skewed slightly toward larger values (Fig. 1).
Approx. 95% of the simulated probability covers the
interval from 320 000 to 1 200 000 bird deaths (short-
est interval), whereas 80% of the simulated proba -
bility covers the interval from 390 000 to 910 000.
The probability that the number of bird deaths
exceeded 400 000 is approx. 89%, and the probabil-
ity that the number of bird deaths exceeded 500 000
is approx. 69%, based on the carcass sampling prob-
ability model.
The Monte Carlo distribution for the carcass sam-
pling distribution of bird deaths was sensitive to the
244
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal
value of the hyperparameter (u = 200). A change to
this hyperparameter changes the variances of all of
the distributions of the parameters for the carcass
survey together. Halving this hyperparameter to 100
increased the variance, flattened the distribution
somewhat, and shifted the distribution toward larger
values. This change raised the standard deviation of
the rparameter from 0.016 to 0.023, the lparameter
from 0.032 to 0.045, and the kparameter from 0.021
to 0.030. The standard deviation of the Monte Carlo
distribution of bird deaths increased from approx.
220 000 to 400 000, the medians increased from
590 000 to 620 000, and the upper 97.5th percentile
increased from 1.2 million to slightly over 1.8 million.
Halving the hyperparameter again to 50 flattened the
distribution of the total bird deaths even more, with
the 97.5th percentile being greater than 3 million bird
deaths. In contrast, doubling uto 400 reduced the
standard deviation of the rparameter from 0.016
to 0.012, the lparameter from 0.032 to 0.023, the k
parameter from 0.021 to 0.015, and the 97.5th per-
centile of the distribution to 920 000 — values that we
judged to be too small to reflect the actual uncer-
tainty in the parameter values. We concluded that u=
200 produced distributions of the parameters that
were the most consistent with reasonable alternate
assumptions about the parameter values.
The Monte Carlo distribution for the number of
bird deaths using exposure probability was flatter,
and centered over slightly higher values when com-
pared to the distributions of estimates from the car-
cass sampling approach. The median of this distri -
bution was approx. 700 000 whereas the mean was
approx. 800000, again skewed toward larger values
(Fig. 2). The 95% uncertainty interval covers 160 000
to 1 900 000 bird deaths, while the 80% interval
covers 300000 to 1 400000. The probability that the
number of bird deaths exceeded 400 000 is approx.
81%, and the probability that the bird deaths ex -
ceeded 500 000 is approx. 72%.
Influence of carcass size
Relative to their occurrence when surveyed while
alive in coastal Gulf habitats, small-bodied carcasses
were markedly under-represented in the tally of
dead oiled birds found during the Deepwater Hori-
zon spill (Fig. 3). As defined here, small-bodied birds
account for about 34% of the communities in coastal
habitats (Johnson 2011). Conversely, both medium-
and large-bodied birds made up a disproportionately
large proportion of the carcass counts relative to their
occurrence in coastal habitat.
By implementing a searcher parameter (d) specific
to body size, the size distribution of bird-death esti-
mates more closely matched the live proportions
observed in the small-bodied bird category (Fig. 3).
In addition, by using the size-group searcher effi-
ciency parameters we eliminated entirely the dispro-
portionately high representation of large-bodied
birds in the carcass counts relative to their live occur-
rence in coastal Gulf habitats. However, the searcher
efficiency adjustments did not reduce a similar dis-
crepancy in representation seen for the medium size
birds (Fig. 3).
245
Parameter Large birds Medium birds Small birds
(≥500 g) (300−499 g) (10−299 g)
pˆ j0.00635 0.00376 0.00117
1/pˆ j157 266 856
95% uncertainty 90−330 150−570 440−2100
interval for 1/pˆ j
N
ˆj80000 400000 100000
Table 2. Estimates of the combined probability of final
shoreline recovery for large-, medium-, and small-bodied
seabirds (denoted pˆ j) in a carcass-sampling model used to
estimate mortality in coastal waters of the Gulf of Mexico
during the Deepwater Horizon spill. The sampling-based
expansion factor for the observed number of carcasses in
the ith size category is the multiplicative inverse (1/pˆ j) of
the probability of recovery for each size class. N
ˆjdenotes a
sampling-based estimate of the number of birds killed for
body size j. Observed carcass counts were 493 for large
birds (j= 1), 1383 for medium birds (j = 2), and 128 for small
birds (j= 3). All estimates are rounded (mortality estimates
do not necessarily sum exactly due to rounding)
Carcass survey-based bird deaths (x10 6)
Frequency (x10 3)
0.2 0.4 0.6 0.8 1.0 1.2 1.4
0
20
40
60
80
100
Fig. 1. Monte Carlo distribution (1 million simulation repli-
cates) of the number of coastal bird deaths from the Deep-
water Horizon oil spill based on a carcass survey approach
to estimating total bird mortality. Median for the simulated
distribution = 590 000 bird deaths; mean = 630000
Mar Ecol Prog Ser 513: 239–252, 2014
DISCUSSION
Our results indicate that bird losses from the Deep-
water Horizon oil spill almost certainly numbered
into the hundreds of thousands in coastal waters of
the northern Gulf of Mexico. Although each estima-
tion approach was different conceptually —and used
largely independent data — both showed that the most
likely coastal bird mortality was approx. 700 000.
Carcass sampling model
Carcass sampling expanded approx. 2000 bird
carcasses recovered from the coastal zone into an
estimate of nearly 600 000 bird deaths —a very large
statistical extrapolation. Even so, those few observed
carcasses constitute the actual, observable evidence
analyzed in the conventional manner used for avian
mortality assessments after oil spills (e.g. Castege et
al. 2007, Munilla et al. 2011). What was less conven-
tional was using probability to generate numerical
results that include a probability distribution of alter-
nate assumptions about how to statistically expand
such carcass counts to account for realistic processes
that re duced the numbers recovered from shoreline
surveys.
The counts of recovered carcasses are usually far
less than the number of birds killed by an oil spill
(Seys et al. 2001, Ford 2006). Low carcass counts or
small recovery probabilities alone do not necessarily
lead to bias. However, in this case, because carcass
recovery was low and expansion factors were large,
any source of bias would be highly leveraged. For
example, if some unmeasured process (e.g. carcass
burning) removed even a small number of carcasses
en route to shore, each missing carcass would create
bias in our estimate proportional to the large expan-
sion factors.
We have tried to understand and account for all
recovery processes. On balance we suspect that we
were more likely to have underestimated than to
have overestimated bird mortality. Nevertheless, if
birds tended to fly towards coastlines after exposure,
and then later died near or on shore, then we would
have underestimated rand overestimated mortality.
If the actual decline of seabird density with distance
from shore is better described as a negative exponen-
tial function, then our assumption of a linear decline
also would lead to overestimation of mortality.
Cleanup tactics during spill response may have led
to underestimation by intercepting and destroying
carcasses before they could be retrieved. Between
28 April and 19 July 2010, 376 controlled burns at sea
consumed 3.5 to 4.9 × 107l of oil, a volume approxi-
mately equivalent to one Exxon Valdez spill. Although
burns were avoided if live apex predators were pres-
ent (Allen et al. 2011), floating carcasses would have
been destroyed since ignitions took place where con-
vergences had concentrated buoyant material (FISG
2010). Skimming operations removed half of the oil
volume removed by burning (FISG 2010), which may
also have removed carcasses.
Carcass loss at sea from biological factors also
reduces shoreline deposition. Bird consumption by
tiger sharks Galeocerdo cuvier or other higher tro -
phic-level consumers (Kaufman 2012) in the Gulf
exacerbates carcass loss (Wiese 2003). Such losses
246
Exposure probability-based bird deaths (x10 6)
Frequency (x10 3)
0 0.5 1.0 1.5 2.0 2.5
0
20
40
60
80
100
Fig. 2. Monte Carlo distribution (1 million simulation repli-
cates) of the number of coastal bird deaths from the Deep-
water Horizon oil spill based on an exposure probability
model over 95 d. Median number of bird deaths ≈700 000
Fig. 3. Comparison of the proportions of birds that were sur-
veyed live in coastal Gulf habitat (Johnson 2011) to the pro-
portions recovered as carcasses and estimated killed during
the Deepwater Horizon spill. Bird size categories: small:
10−299 g; medium: 300−499 g; large: ≥500 g
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal
lower the estimate of mortality if, as in this study, no
correction is applied for in toto ingestion of seabird
carcasses.
Our estimate of 5.7% probability for carcass trans-
port to shorelines is lower than values reported in
most previous studies. Piatt & Ford (1996) reported a
probability of 12%. A summary by Munilla et al.
(2011) gave a mean of 17.1% based on 37 experimen-
tal studies with 7040 combined block and carcass
releases. Block recoveries overstate deposition as
blocks do not decompose and weather like carcasses
(Wiese & Jones 2001). Using a block-to-carcass cor-
rection provided by Munilla et al. (2011), the differ-
ence between values from the meta-analysis above
and our study narrows to 11.2% versus 5.7%, respec-
tively. Our estimate of carcass deposition was never-
theless higher than reported for another Atlantic
Ocean oil spill (as low as 0.8% in the Prestige spill;
Castege et al. 2007).
Although we describe the carcass movement with a
simple model, the actual movement process was
quite complex. Early in the spill, oil was steered away
from the coast (Dietrich et al. 2012). Response author-
ities opened locks on Mississippi River canals to re -
direct the discharge and prevent or delay Deepwater
Horizon oil from reaching the ecologically fragile
coastal marshes. Under weak winds and effects of
sea surface slope, high buoyancy-driven outflow
(Falcini et al. 2012) blocked much of the surface oil
from reaching most Gulf shorelines until late May
2010. Except along the immediate shoreline, wind
forcing played a negligible role in large-scale oil
transport (Huntley et al. 2011) until early July when
Tropical Storm Alex transited the southern Gulf
(Dietrich et al. 2012, Le Hénaff et al. 2012). Oil was
then driven shoreward by the storm surge (Pugliese
Carratelli et al. 2011). The extent of oiled shoreline
subsequently tripled (Boufadel et al. 2014), and
recovery of bird carcasses accelerated just before the
well was capped (Belanger et al. 2010).
We assumed that all shorelines affected by this spill
were searched (i.e. h= 1.00). Responders searched
6841 km of the Gulf’s beach, wetland, and man-made
shorelines during the Deepwater Horizon incident,
and some oiling was observed on 1705 km (Owens et
al. 2011a,b). If some shorelines were missed, or if
they received less attention by searchers, both of
which seem likely given the geographic extent of the
Deepwater Horizon spill and the complexity of shore-
lines, then fewer carcasses would be recovered, ulti-
mately leading to an underestimate of the bird losses.
Our assumed daily carcass persistence proportions
of ~50% on beaches the first few days after deposi-
tion is similar to values reported from other oil spills.
Nevertheless, as few as 10% of carcasses may
remain 2 to 4 d after reaching shore (Ford 2006, Byrd
et al. 2009, Ford & Zafonte 2009). Due to the Gulf’s
many scavengers (e.g. gulls, crows, raptors, raccoons,
mink, crabs), shoreline persistence may have been
even lower than we assumed. Given search intervals
by Deepwater Horizon responders that ranged up to
12 d (USDOI 2011), our assumed 3 d interval would,
at least on some occasions, lead to an overestimate of
carcass persistence, ultimately under estimating bird
deaths. Carcass persistence was not derived in situ,
however, so this parameter had to be approximated
from values based on previous oil spills.
Exposure probability model
Despite a simpler conceptual approach and fewer
parameters, the exposure probability model gave an
even wider uncertainty interval for the bird deaths
estimated in coastal waters (Fig. 2). Here again, we
may have made incorrect assumptions about the val-
ues of the parameters that led to estimates that erred
in either direction.
We assumed that 60% of coastal birds exposed to
Deepwater Horizon oil avoided a lethal dosage. The
lethal exposure proportion we used falls near the
lower end of the interval of values used to character-
ize bird mortality from other oil spills (e.g. 23 to
100%; WIW 2001, Robertson et al. 2006, Wilhelm et
al. 2007, Fifield et al. 2009). Assuming that warm
Gulf waters may be less likely to induce thermoregu-
latory stress from exposure, we projected that all live
oiled birds survived (e.g. see Selman et al. 2012). In
cooler seas, live oiled birds are typically included in
the tally of mortality (cf. Page et al. 1990, Castege et
al. 2007, Munilla et al. 2011). To the extent that we
underestimated the lethal dose, we would have
underestimated the total bird deaths with the expo-
sure probability approach. We could not, for exam-
ple, consider all of the ways that oil could lead to
mortality, such as via ingestion and inhalation. Any
error in Eq. (3) would propagate linearly. If, for exam-
ple, 60% of exposed birds died rather than our
assumed 40%, then our estimate of total mortality
would underestimate by a factor of 60/40 or 1.5.
We note that few large coastal spills have occurred
in warm subtropical seas where the seabird commu-
nities comprise high proportions of aerially foraging
species. Future study aimed at documenting seabird
exposure and behavior around oil slicks in warm
oceans are needed (e.g. see Watson et al. 2009).
247
Mar Ecol Prog Ser 513: 239–252, 2014
Delayed mortality of oiled birds may have occurred
at a higher rate than what we assumed.
We can think of several ways we could have
over estimated the bird mortality through errors in
measurement, incorrect assumptions, or oversimplifi-
cation. The most obvious potential source for overes-
timate might be our assumption that the bird density
returned to a baseline level of 5.8 birds km−2 due to
emigration and bird flux after a period of just 1 d (but
see Fifield et al. 2009). Numbers of birds subject to
risk at the time of a spill are often hard to quantify for
assessments of spill mortality (e.g. French McCay &
Rowe 2004). The bird density that we used for coastal
Gulf waters was typical for seasons when the Deep-
water Horizon incident occurred (McFarlane &
Lester 2005), although it was derived from outside
the immediate spill zone. We attempted to account
for uncertainty in this parameter with an assigned
probability distribution, but we have no way to assess
our success at estimating the background density or
the time it took after mortality for the density to
return to the background level because there are no
alternative estimates for bird density readily avail-
able.
Most importantly, due to limits imposed by image
coverage or sensor sensitivity (Leifer et al. 2012,
Lindsley & Long 2012) and high application of chem-
ical dispersants (e.g. Allan et al. 2012), satellite delin-
eation of the exposure risk presented to birds is
incomplete (Haney et al. 2014). Underestimation of
contamination risk on the water would also lead to
underestimation bias in the bird mortality using the
exposure probability approach.
Importance of body size
Our study is among very few to make substantial
adjustments to mortality estimation based on carcass
size following a large marine oil spill. An adjustment
for bird body size was used in the relatively small MV
Kure spill (1.7 × 104l) in Humboldt Bay, California
(Ford et al. 2013). Smaller birds are more difficult
to detect under field conditions, whether alive (Bar-
braud & Thiebot 2009) or dead (Smallwood 2007).
Such adjustments are necessary to account for the
large disparity observed between relatively low pro-
portions of small bodied carcasses recovered com-
pared with live small birds detected in Gulf surveys
(Fig. 3).
Proportionate mortalities from oil spills do vary by
body size (Page et al. 1990), although differences
among species are not well understood (Evans &
Keijl 1993). Exposures to very small oil dosages cause
loss of plumage integrity (O’Hara & Morandin 2010),
lowering insulation and raising mortality (Jenssen
1994). Because thermal conductance scales allomet-
rically to body size (Aschoff 1981), insulation is more
easily compromised in smaller-bodied species, and
a greater representation of small birds might be
expected in the tally of mortality. Instead, however,
only 128 small-bodied carcasses were retrieved dur-
ing the Deepwater Horizon spill (6.4% of the total),
despite smaller birds making up a third of avian com-
munities in the coastal Gulf (Johnson 2011).
Small-bodied birds may have beach persistence
times only 5 to 13% those of large- and medium-bod-
ied birds (Seys et al. 2001), as small birds are more
readily carried off by scavengers (Ford & Zafonte
2009, Ponce et al. 2010), whereas large, heavy birds
must usually be scavenged in situ. In addition,
smaller bird carcasses can be scavenged by con-
sumers having a wider range of body sizes, including
small-bodied predators. Without correcting for body-
size, especially where bird communities contain a
large proportion of small-bodied birds as in the Gulf
of Mexico (Johnson 2011), bird losses in marine oil
spills could be substantially underestimated.
Ecological implications from spill mortality
We focused solely on near-term, acute mortality
from the Deepwater Horizon spill, even though ob -
servations of oiled birds continued for at least 1 yr
(Henkel et al. 2012). We did not include estimates for
indirect (Velando et al. 2005) or chronic population
effects (Irons et al. 2000), either of which can increase
the avian mortality (Pérez et al. 2008). For example,
oil cleanup and other response activities reduce bird
survival (Burger & Tsipoura 1998).
Despite a focus here on near-term effects, high
avian mortality in the Deepwater Horizon spill likely
had population-level repercussions. We estimated
mor tality in 4 seabird species using the size-specific
expansion factors (1/pjlisted in Table 2) and numbers
of carcasses found for that species (listed in Supple-
ment 1). Projected losses reached or exceeded 24 000
birds in all 4 species and reached 232 000 for laugh-
ing gulls (Leu cophaeus atricilla, Table 3). In 3 species,
estimated losses reached or exceeded 12% of the
total population estimated present in the northern
Gulf of Mexico.
Mortality of laughing gulls was ~32% of the regional
Gulf population (Table 3). National Audubon Society
Christmas Bird Count (CBC) data from Florida, Ala-
248
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal
bama, Mississippi, Louisiana, and Texas (NAS 2010)
corroborate this decline. CBC data reveal that after
the Deepwater Horizon incident, laughing gulls in
the Gulf region declined by 61 to 64% between win-
ter 2010 and winters 2011, 2012, and 2013. Given
adequate coverage and sample sizes (only for laugh-
ing gull did CBC survey effort exceed 10 birds h−1
among the 4 species examined), the NAS CBC
could provide an alternative to evaluate pop ulation-
level impacts from large marine oil spills in North
America.
In addition to carcass sampling and exposure prob-
ability models, differences in breeding colony atten-
dance before and after a spill are another means to
corroborate avian mortality (e.g. Piatt & Ford 1996).
Without due consideration of the population age
class and vital rates affected (Votier et al. 2008), oil
spill impacts to seabirds can be masked. We there-
fore encourage more research into the demographic
consequences to the Gulf's coastal birds from this
large marine spill.
Given both its chronic and acute characteristics,
the Deepwater Horizon disaster was a major pertur-
bation to the northern Gulf of Mexico (Peterson et al.
2012). Our estimates do not address delayed mortal-
ity from changes to ecosystem function (e.g. Peterson
2001) or consequences to food webs from reductions
in marine predators caused by this oil spill. Both top-
ics await further study.
Acknowledgements. This study was funded jointly by The
Murray Firm and by Cossich, Sumich, Parsiola and Taylor
LLC. Findings in this manuscript reflect those of the
authors only; interpretations do not reflect positions that
may be held by any organization, entity, or other interest.
All content analyzed and reported here was available in
the public domain. No data, information, do cuments, find-
ings or any other proprietary content protected by any
confidentiality restriction or agreement, including those
pertaining to the Natural Resource Damage Assessment
conducted for the Deepwater Horizon Mississippi Canyon
252 oil spill under 61 Fed. Reg. 440, the Oil Pollution Act
of 1990, were consulted or otherwise used in preparation
of this manuscript. We thank Steven C. Heinl, Charles H.
Pe terson, Terrance J. Quinn II, Robert B. Spies, and 4
anonymous reviewers for their comments on earlier ver-
sions of this paper. Xinxian Zhang provided advice on sta-
tistical computing. Public access to historical data from the
Christmas Bird Count is provided courtesy of the National
Audubon Society and its many volunteer contributors (see
www.christmasbirdcount. org).
LITERATURE CITED
Aeppli C, Carmichael CA, Nelson RK, Lemkau KL and
others (2012) Oil weathering after the Deepwater Hori-
zon disaster led to the formation of oxygenated residues.
Environ Sci Technol 46: 8799−8807
Allan SE, Smith BW, Anderson KA (2012) Impact of the
Deepwater Horizon oil spill on bioavailable polycyclic
aromatic hydrocarbons in Gulf of Mexico coastal waters.
Environ Sci Technol 46: 2033−2039
Allen AA, Jaeger D, Mabile NJ, Costanzo D (2011) The use
of controlled burning during the Gulf of Mexico Deep-
water Horizon MC-252 oil spill response. Int Oil Spill
249
Species Size Population exposed (×103ind.) Estimated mortality
cat. Sub- Adults Total (×103ind.) %
adults Br. Non-Br.
Northern gannet Morus bassanus Large 170 234 32 436 35 8a
Brown pelican Pelecanus occidentalis Large 99 68b34 201 24 12
Laughing gull Leucophaeus atricilla Medium 238 260 232c730 232 32
Royal tern Thalasseus maximus Medium 81 79d40 200 26 13
aApproximately 50% if considering only the fraction of North American gannets projected as over-wintering in the Gulf of
Mexico (Montevecchi et al. 2012)
bShields (2002), updated by Louisiana Department of Wildlife and Fisheries (www.wlf.louisiana.gov/news/29112; accessed
16 June 2013)
cMinimum 40% of non-residents (all ages) estimated to be in the Gulf from northeastern USA (see Belant & Dolbeer 1993b)
dBuckley & Buckley (2002), updated by Florida Fish and Wildlife Conservation Commission (www.myfwc.com/bba/;
accessed 9 June 2013)
Table 3. Estimated population size and mortality from the Deepwater Horizon oil spill for 4 coastal bird species. Breeding num-
bers refer to the US Gulf of Mexico only (except northern gannet, for which breeding numbers pertain to entire Canadian pop-
ulation; Chardine et al. 2013). Sub-adults are numbers in age classes prior to age at first breeding, estimated with life tables
(e.g. Rattiste 2004) that accounted for longevity and age-specific survival (Belant & Dolbeer 1993a, Burger 1996, Buckley &
Buckley 2002, Mowbray 2002, Shields 2002). Unless otherwise specified, non-breeders were assumed to constitute 50 % of
breeders (Votier et al. 2008), not including subadults. Total populations assumed to be vulnerable to the spill were adjusted for
any non-residents and non-breeding adults (Dolbeer & Bernhardt 2003). Spill mortality was estimated using size-specific ex-
pansion factors (Table 2) applied to the carcass counts of each species (see Table S1 in the Supplement at www.int-res.com/
articles/suppl/m513p239_supp.pdf). Size categories: large: ≥500 g; medium: 300−499 g
Mar Ecol Prog Ser 513: 239–252, 2014
250
Conf Proc, Vol 2011, No. 1, American Petroleum Insti-
tute, Washington DC
Antonio FJ, Mendes RS, Thomaz SM (2011) Identifying and
modeling patterns of tetrapod vertebrate mortality rates
in the Gulf of Mexico oil spill. Aquat Toxicol 105: 177−179
Aschoff J (1981) Thermal conductance in mammals and
birds: its dependence on body size and circadian phase.
Comp Biochem Physiol 69: 611−619
Balseiro A, Espí A, Márquez I, Pérez V, Ferreras MC, García
Marín JF, Prieto JM (2005) Pathological features in mar-
ine birds affected by the Prestige’s oil spill in the north of
Spain. J Wildl Dis 41: 371−378
Barbraud C, Thiebot JB (2009) On the importance of esti-
mating detection probabilities from at-sea surveys of fly-
ing seabirds. J Avian Biol 40: 584−590
Barker CH (2011) A statistical outlook for the Deepwater
Horizon oil spill. In: Liu Y, MacFadyen A, Ji ZG, Weis-
berg RH (eds) Monitoring and modeling the Deepwater
Horizon oil spill: a record-breaking enterprise. Geophys
Monogr Ser 195. American Geophysical Union, Wash-
ington, DC, p 237−244
Belanger M, Tan L, Askin N, Wittnich C (2010) Chronologic
effects of the Deepwater Horizon Gulf of Mexico oil spill
on regional seabird casualties. J Mar Anim Ecol 3: 10−14
Belant JL, Dolbeer RA (1993a) Population status of nesting
laughing gulls in the United States, 1977–1991. Am Birds
47: 220−224
Belant JL, Dolbeer RA (1993b) Migration and dispersal of
laughing gulls in the United States. J Field Ornithol 64:
557−565
Boufadel MC, Abdollahi-Nasab A, Geng X, Galt J, Torlapati
J (2014) Simulation of the landfall of the Deepwater Hori-
zon oil on the shorelines of the Gulf of Mexico. Environ
Sci Technol 48:9496−9505
Briggs KT, Yoshida SH, Gershwin ME (1996) The influence
of petrochemicals and stress on the immune system of
seabirds. Regul Toxicol Pharmacol 23: 145−155
Buckley PA, Buckley FG (2002) Royal tern (Thalasseus
maxi mus). In: Poole A (ed) The birds of North America
online. Cornell Lab of Ornithology, Ithaca, New York.
(http: //bna.birds.cornell.edu/bna/species/700; accessed
9 June 2013)
Burger J (1996) Laughing gull (Leucophaeus atricilla). In:
Poole A (ed) The birds of North America online. Cornell
Lab of Ornithology, Ithaca, New York. (http: // bna. birds.
cornell.edu/bna/species/225; accessed 9 June 2013)
Burger J, Tsipoura N (1998) Experimental oiling of san -
derlings (Calidris alba): behavior and weight changes.
Environ Toxicol Chem 17: 1154−1158
Byrd GV, Reynolds JH, Flint PL (2009) Persistence rates and
detection probabilities of bird carcasses on beaches of
Unalaska Island, Alaska, following the wreck of the M/V
Selendang Ayu. Mar Ornithol 37: 197−204
Camilli R, Di Iorio D, Bowen A, Reddy CM and others (2012)
Acoustic measurement of the Deepwater Horizon Ma -
con do well flow rate. Proc Natl Acad Sci USA 109:
20235−20239
Camphuysen CJ, Heubeck M (2001) Marine oil pollution
and beached bird surveys: the development of a sensi-
tive monitoring instrument. Environ Pollut 112: 443−461
Casella G, Berger RL (2002) Statistical inference, 2nd edn.
Wadsworth and Brooks/Cole, Pacific Grove, CA
Castege I, Lalanne Y, Gouriou V, Hemery G and others
(2007) Estimating actual seabirds mortality at sea and
relationship with oil spills: lessons from the Prestige oil
spill in Aquitaine (France). Ardeola 54: 289−307
Chardine JW, Rail JF, Wilhelm S (2013) Population dyna -
mics of northern gannets in North America, 1984 –2009.
J Field Ornithol 84: 187−192
Clarke ED, Spear LB, McCracken ML, Marques FCC,
Borchers DL, Buckland ST, Ainley DG (2003) Validating
the use of generalized additive models and at-sea
surveys to estimate size and temporal trends of seabird
populations. J Appl Ecol 40: 278−292
Dietrich JC, Trahan CJ, Howard MT, Fleming JG and others
(2012) Surface trajectories of oil transport along the
northern coastline of the Gulf of Mexico. Cont Shelf Res
41: 17−47
Dolbeer RA, Bernhardt GE (2003) Age-specific reproduction
in female laughing gulls (Larus atricilla). Auk 120:
531−535
Evans MI, Keijl GO (1993) Impact of Gulf War oil spills on
the wader populations of the Saudi Arabian Gulf coast.
Sandgrouse 15: 85−105
Falcini F, Jerolmack DJ, Buongiorno Nardelli B (2012) Mis-
sissippi River and sea surface height effects on oil slick
migration. PLoS ONE 7: e36037
Fifield DA, Baker KD, Byrne R, Robertson GJ and others
(2009) Modelling seabird oil spill mortality using flight
and swim behaviour. Environ Studies Res Funds Rep
186, Canadian Wildlife Service, Dartmouth, NS
FISG (Federal Interagency Solutions Group) (2010) Oil budget
calculator, Deepwater Horizon: a report to the National
Incident Command. Tech Doc Oil Budget Science &
Engineering Team (http:// www. restorethegulf. gov/ sites/
default/files/documents/pdf/OilBudgetCalc_Full_HQ-
Print_111110.pdf; accessed 17 Sep 2014)
Ford RG (2006) Using beached bird monitoring data for sea-
bird damage assessment: the importance of search inter-
val. Mar Ornithol 34: 91−98
Ford RG, Zafonte MA (2009) Scavenging of seabird car-
casses at oil spill sites in California and Oregon. Mar
Ornithol 37: 205−211
Ford RG, Casey JL, Hewitt CH, Lewis DB, Varoujean DH,
Warrick DR, Williams WA (1991) Seabird mortality
resulting from the Nestucca oil spill incident, winter
1988-89. Report for Washington Department of Wildlife.
Ecological Consulting, Portland, OR
Ford RG, Sharp BE, Zafonte M (2013) The efficiency of
searchers recovering seabirds and waterfowl killed in
the 1997 M/V Kure oil spill in northern California. Cali-
fornia Dept of Fish & Game, Sacramento, CA (https: //
nrm. dfg.ca.gov/FileHandler.ashx? DocumentID= 20097;
accessed 8 Apr 2013)
French McCay D, Rowe JJ (2004) Evaluation of bird impacts
on historical oil spill cases using the SIMAP oil spill
model. In: Proc 27th Arct Mar Oil Spill Program (AMOP)
Tech Semin, Ottawa, p 421−452
Garcia-Pineda O, MacDonald IR, Li X, Jackson CR, Pichel
WG (2013) Oil spill mapping and measurement in the
Gulf of Mexico with textural classifier neural network
algorithm (TCNNA). IEEE J Selected Topics Appl Earth
Observ Remote Sens 6: 2517−2525
Haney JC, Geiger HJ, Short JW (2014) Bird mortality from
the Deepwater Horizon oil spill. I. Exposure probability
in the offshore Gulf of Mexico. Mar Ecol Prog Ser 513:
225–237
Henkel JR, Sigel BJ, Taylor CM (2012) Large-scale impacts
of the Deepwater Horizon oil spill: can local disturbance
affect distant ecosystems through migratory shorebirds?
Bioscience 62: 676−685
Hess NA, Ribic CA (2000) Seabird ecology. In: Davis RW,
Evans WE, Würsig B (eds) Cetaceans, sea turtles and
seabirds in the northern Gulf of Mexico: distribution,
Haney et al.: Bird mortality from Deepwater Horizon. II. Coastal
abundance and habitat associations. Vol II: Technical
Report. Prepared by Texas A&M University, Galves -
ton, TX; and National Marine Fisheries Service, New
Orleans, LA. OCS Study MMS 2000-003, p 275−316
Huntley HS, Lipphardt BL, Kirwan AD (2011) Surface drift
predictions of the Deepwater Horizon spill: the La -
grangian perspective. In: Liu Y, MacFadyen A, Ji ZG,
Weisberg RH (eds) Monitoring and modeling the Deep-
water Horizon oil spill: a record-breaking enterprise.
Geophys Monogr Ser 195. American Geophysical Union,
Washington, DC, p 179−195
Irons DB, Kendall SJ, Erickson WP, McDonald LL, Lance BK
(2000) Nine years after the Exxon Valdez oil spill: effects
on marine bird populations in Prince William Sound,
Alaska. Condor 102: 723−737
Jenssen BM (1994) Review article: effects of oil pollution,
chemically treated oil, and cleaning on thermal balance
of birds. Environ Pollut 86: 207−215
Johnson EI (2011) Audubon coastal bird survey: first annual
report, July 2011. National Audubon Society, Baton
Rouge, LA
Kaufman R (2012) Sharks eating songbirds in Gulf of Me xi -
co (http: //news.nationalgeographic.com/news/ 2012/ 01/
120113-sharks-songbirds-gulf-of-mexico-animals-fish-
science/; accessed 27 Feb 2013)
Le Hénaff M, Kourafalou VH, Paris CB, Helgers J, Aman
ZM, Hogan PJ, Srinivasan A (2012) Surface evolution of
the Deepwater Horizon oil spill patch: combined effects
of circulation and wind-induced drift. Environ Sci Tech-
nol 46: 7267−7273
Leifer I, Lehr WJ, Simecek-Beatty D, Bradley E and others
(2012) State of the art satellite and airborn marine oil spill
remote sensing: application to the BP Deepwater Hori-
zon oil spill. Remote Sens Environ 124: 185−209
Leighton FA (1993) The toxicity of petroleum oils to birds.
Environ Rev 1: 92−103
Lindsley RD, Long DG (2012) Mapping surface oil
extent from the Deepwater Horizon oil spill using
ASCAT backscatter. IEEE Trans Geosci Rem Sens 50:
2534−2541
MacFadyen A, Watabayashi GY, Barker CH, Beegle-Krause
CJ (2011) Tactical modeling of surface oil transport dur-
ing the Deepwater Horizon spill response. In: Liu Y,
MacFadyen A, Ji ZG, Weisberg RH (eds) Monitoring
and modeling the Deepwater Horizon oil spill: a record-
breaking enterprise. Geophys Monogr Ser 195. Ameri-
can Geophysical Union, Washington, DC, p 167−178
McFarlane RW, Lester LJ (2005) Determination of nearshore
seabird density on the upper Texas coast. Final report.
State Energy Conservation Office, Austin, TX (www.
seco. cpa. state.tx.us/re/docs/re_wind_projects-bird.pdf;
accessed 22 Feb 2013)
McNutt MK, Camilli R, Crone TJ, Guthrie GD and others
(2012) Review of flow rate estimates of the Deepwater
Ho rizon oil spill. Proc Natl Acad Sci USA 109:
20260−20267
Michel J, Owens EH, Zengel S, Graham A and others (2013)
Extent and degree of shoreline oiling: Deepwater Hori-
zon oil spill, Gulf of Mexico, USA. PLoS ONE 8: e65087
Montevecchi W, Fifield D, Burke C, Garthe S, Hedd A, Rail
JF, Robertson G (2012) Tracking long-distance migration
to assess marine pollution impact. Biol Lett 8: 218−221
Mowbray TB (2002) Northern gannet (Morus bassanus). In:
Poole A (ed) The birds of North America online. Cornell
Lab of Ornithology, Ithaca, NY (http: //bna. birds. cornell.
edu/bna/species/693; accessed 9 June 2013)
Munilla I, Arcos JM, Oro D, Álvarez D, Leyenda PM, Ve -
lando A (2011) Mass mortality of seabirds in the after-
math of the Prestige oil spill. Ecosphere 2: art83
NAS (National Audubon Society) (2010) The Christmas bird
count historical results [Online] (http: //netapp. audubon.
org/CBCObservation/Historical/ResultsBySpecies.aspx?1;
accessed 12 Sep 2013)
Nelson B (1979) Seabirds: their biology and ecology. A&W
Publishers, New York, NY
Norse EA, Amos J (2010) Impacts, perception, and policy
implications of the Deepwater Horizon oil and gas disas-
ter. Environ Law Report 40: 11058−11073
O’Hara PD, Morandin LA (2010) Effects of sheens associated
with offshore oil and gas development on the feather
microstructure of pelagic seabirds. Mar Pollut Bull 60:
672−678
Oppel S, Meirinho A, Ramirez I, Gardner B, O’Connell A,
Louzao M (2012) Comparison of five modelling tech-
niques to predict the spatial distribution and abundance
of seabirds. Biol Conserv 156: 94−194
Owens E, Taylor E, Michel J (2011a) The Deepwater Hori-
zon-Macondo 2010 Shoreline Cleanup Assessment
Technique (SCAT) program. GSA Abstracts with Pro-
grams 43: 11
Owens EH, Santner R, Cocklan-Vendl M, Michel J, Reimer
PD, Stong B (2011b) Shoreline treatment during the
Deepwater Horizon-Macondo response. Int Oil Spill
Conf 271: 1−9
Page GW, Carter HR, Ford RG (1990) Numbers of seabirds
killed or debilitated by the 1986 Apex Houston oil spill in
central California. Stud Avian Biol 14: 164−174
Peake DE (1996) Bird surveys. In: Davis RW, Fargion GS
(eds) Distribution and abundance of cetaceans in the
north-central and western Gulf of Mexico, Vol II. Techni-
cal Report, OCS Study MMS 96-0027. US Department of
the Interior, New Orleans, LA, p 271−304
Pérez C, Velando A, Munilla I, López-Alonso M, Oro D
(2008) Monitoring polycyclic aromatic hydrocarbon pol-
lution in the marine environment after the Prestige oil
spill by means of seabird blood analysis. Environ Sci
Technol 42: 707−713
Peterson CH (2001) The ‘Exxon Valdez’ oil spill in Alaska:
acute, indirect and chronic effects on the ecosystem. Adv
Mar Biol 39: 1−103
Peterson CH, Anderson SS, Cherr GN, Ambrose RF and
others (2012) A tale of two spills: novel science and policy
implications of an emerging new oil spill model. Bio-
science 62: 461−469
Piatt JF, Ford RG (1996) How many seabirds were killed by
the Exxon Valdez oil spill? Proc Exxon Valdez Oil Spill
Symp, Anchorage, AK. Am Fish Soc Symp 18:712−719
Ponce C, Alonso JC, Argandoña G, García Fernández A,
Carrasco M (2010) Carcass removal by scavengers and
search accuracy affect bird mortality estimates at power
lines. Anim Conserv 13: 603−612
Poulain PM, Gerin R, Mauri E (2009) Wind effects on
drogued and undrogued drifters in the Eastern Mediter-
ranean. J Atmos Ocean Technol 26: 1144−1156
Pugliese Carratelli E, Dentale F, Reale F (2011) On the
effects of wave-induced drift and dispersion in the Deep-
water Horizon oil spill. In: Liu Y, MacFadyen A, Ji ZG,
Weisberg RH (eds) Monitoring and modeling the Deep-
water Horizon oil spill: a record-breaking enterprise.
Geophys Monogr Ser 195. American Geophysical Union,
Washington, DC, p 197−204
Rattiste K (2004) Reproductive success in presenescent com-
mon gulls (Larus canus): the importance of the last year
of life. Proc R Soc Lond Ser B Biol Sci 271: 2059−2064
251
Mar Ecol Prog Ser 513: 239–252, 2014
Robertson GJ, Ryan PC, Dussureault J, Turner BC, Wilhelm
SI, Power K (2006) Composition of beached marine birds
from an oiling event in southeastern Newfoundland and
Labrador, November 2004. Mar Ornithol 34: 141−146
Selman W, Hess TJ, Salyers B, Salyers C (2012) Short-term
response of brown pelicans (Pelecanus occidentalis) to
oil spill rehabilitation and translocation. Southeast Nat
11: G1−G16
Seys J, Offringa H, Van Waeyenberge J, Meire P, Kuijken E
(2001) Numbers of beached bird corpses and mortality of
seabirds, how do they relate: a North Sea study in a
wider context. In: Seys J (author) Het gebruik van zee-
en kustvogelgegevens ter ondersteuning van het beleid
en beheer van de Belgische kustwateren. PhD thesis,
Universiteit Gent, p 78−96
Shields M (2002) Brown pelican (Pelecanus occidentalis). In:
Poole A (ed) The birds of North America online. Cornell
Lab of Ornithology, Ithaca, NY (http: //bna.birds.cornell.
edu/ bna/species/609; accessed 16 June 2013)
Silver N (2012) The signal and the noise: why so many pre-
dictions fail—but some don’t. Penguin Press, New York,
NY
Smallwood KS (2007) Estimating wind turbine-caused bird
mortality. J Wildl Manag 71: 2781−2791
Stienen EWM, Haelters J, Kerckhof F, Van Waeyenberge J
(2004) Three colours of black: seabird strandings in
Belgium during the Tricolor incident. Atlas Seab 6:
129−146
Tan L, Belanger M, Wittnich C (2008) Revising the correla-
tion between estimated seabird mortality and oil spill
size. J Mar Anim Ecol 3: 20−26
Thompson SK (2002) Sampling, 2nd edn. Wiley, New York,
NY
USDOI (United States Department of Interior) (2011) Work
plan (bird study #1C): beached carcass persistence study.
Final report, 7 June 2011 (www.doi.gov/deepwaterhorizon/
adminrecord/upload/BIRD_Telemetry_Carcass_Drift_
1D_corrected-signature-page-final-executed-plan.pdf;
accessed 10 March 2013)
Van Pelt TI, Piatt JF (1995) Deposition and persistence of
beachcast seabird carcasses. Mar Pollut Bull 30: 794−802
Velando A, Munilla I, Leyenda PM (2005) Short-term indi-
rect effects of the ‘Prestige’ oil spill on European shags:
changes in availability of prey. Mar Ecol Prog Ser 302:
263−274
Votier SC, Birkhead TR, Oro D, Trinder M and others (2008)
Recruitment and survival of immature seabirds in rela-
tion to oil spills and climate variability. J Anim Ecol 77:
974−983
Watson JEM, Joseph LN, Watson AWT (2009) A rapid
assessment of the impacts of the Montara oil leak on
birds, cetaceans and marine reptiles. Report to Dept
Environment, Water, Heritage and Arts, Canberra (www.
environment. gov.au/coasts/publications/pubs/montara-
rapid-survey. pdf; accessed 12 Sep 2013)
Wiese FK (2003) Sinking rates of dead birds: improving esti-
mates of seabird mortality due to oiling. Mar Ornithol 31:
65−70
Wiese FK, Jones IL (2001) Experimental support for a new
drift block design to assess seabird mortality from oil pol-
lution. Auk 118: 1062−1068
Wiese FK, Robertson GJ (2004) Assessing seabird mortality
from chronic oil discharges at sea. J Wildl Manag 68:
627−638
Wilhelm SI, Robertson GJ, Ryan PC, Schneider DC (2007)
Comparing an estimate of seabirds at risk to a mortality
estimate from the November 2004 Terra Nova FPSO oil
spill. Mar Pollut Bull 54: 537−544
WIW (Wildlife Injury Workgroup) (2001) Estimate of total
acute mortality to birds resulting from the Chalk Point oil
spill, Swanson Creek, Maryland, April 7, 2000. Final
report. NOAA, Silver Springs, MD (www. darrp. noaa.
gov/ northeast/ chalk_point/pdf/ cp2048.pdf; accessed 8
March 2013)
252
Editorial responsibility: Jacob González-Solís,
Barcelona, Spain
Submitted: October 16, 2013; Accepted: April 30, 2014
Proofs received from author(s): September 30, 2014
