Sustained disruption of narwhal habitat use andbehavior in the presence of Arctic killer whalesGreg A. Breeda,1, Cory J. D. Matthewsb, Marianne Marcouxb, Jeff W. Higdonc, Bernard LeBlancd, Stephen D. Petersene,Jack Orrb, Natalie R. Reinhartf, and Steven H. Fergusonb

aInstitute of Arctic Biology, University of Alaska, Fairbanks, AK 99775; bArctic Aquatic Research Division, Fisheries and Oceans Canada, Winnipeg, MB,Canada R3T 2N6; cHigdon Wildlife Consulting, Winnipeg, MB, Canada R3G 3C9; dFisheries Management, Fisheries and Oceans Canada, Quebec, QC, CanadaG1K 7Y7; eAssiniboine Park Zoo, Winnipeg, MB, Canada R3R 0B8; and fDepartment of Biological Sciences, University of Manitoba, Winnipeg, MB, CanadaR3T 2N2

Edited by James A. Estes, University of California, Santa Cruz, CA, and approved January 10, 2017 (received for review July 17, 2016)

Although predators influence behavior of prey, analyses of elec-tronic tracking data in marine environments rarely consider howpredators affect the behavior of tracked animals. We collectedan unprecedented dataset by synchronously tracking predator(killer whales, N = 1; representing a family group) and prey(narwhal, N = 7) via satellite telemetry in Admiralty Inlet, alarge fjord in the Eastern Canadian Arctic. Analyzing the move-ment data with a switching-state space model and a series ofmixed effects models, we show that the presence of killer whalesstrongly alters the behavior and distribution of narwhal. Whenkiller whales were present (within about 100 km), narwhal movedcloser to shore, where they were presumably less vulnerable.Under predation threat, narwhal movement patterns were morelikely to be transiting, whereas in the absence of threat, morelikely resident. Effects extended beyond discrete predatory eventsand persisted steadily for 10 d, the duration that killer whalesremained in Admiralty Inlet. Our findings have two key conse-quences. First, given current reductions in sea ice and increasesin Arctic killer whale sightings, killer whales have the poten-tial to reshape Arctic marine mammal distributions and behavior.Second and of more general importance, predators have thepotential to strongly affect movement behavior of tracked marineanimals. Understanding predator effects may be as or more impor-tant than relating movement behavior to resource distribution orbottom-up drivers traditionally included in analyses of marine ani-mal tracking data.

predator–prey dynamics | sea ice | biologging | climate change |trait-mediated effects

Consumptive effects (alternatively termed “density-mediatedeffects”) of predators on prey refer to the mortality incurred

when predators kill and consume prey during predation events.They can control prey populations and in certain circumstances,restructure ecosystems through trophic cascades (1–3). Noncon-sumptive effects (also termed “trait-mediated effects”) can sim-ilarly affect prey populations by altering species’ behavior andspace use under perceived or real predation risk, which areassociated with decreased fitness through loss of access tokey foraging areas, disrupted social structure, increased energyexpenditure and stress imposed by persistent vigilance andescape behaviors, and decreased reproductive success (3–7).Nonconsumptive effects are sublethal (8, 9), but because theycan impact many individuals in a population simultaneously, thecumulative effect may exceed consumptive effects (8, 10–12).

In terrestrial systems, movement data collected by electronictelemetry tracking tags have been used to clearly show thatcarnivores affect prey species’ use of space and habitat selec-tion (13–15), and these nonconsumptive effects can negativelyimpact population dynamics (10, 13, 11). When large enough,such effects have even been suggested to lead to trophic cascades(16, 17, 3). However, there is disagreement about whether non-consumptive effects can be strong enough to cause trophic cas-cades, even in well-studied exemplar systems (18–20).

Electronic tracking tags are also frequently used to track verte-brates in marine systems. Although there is evidence that marineanimals adjust their behavior under predation threat (21, 22, 12),few data or analyses exist showing how predators affect themovement of tracked marine animals. These data are lackingbecause marine environments are more difficult to observe andtracked animals often move over scales much larger than theirterrestrial counterparts, making it difficult to measure predatordensity in situations where tracking tags are deployed on prey.Instead, analyses have tended to focus on habitat preference,resource distribution, and the oceanographic controls of primaryproduction (23–25). If predators are affecting the behavior oftracked animals without being considered or recognized, partic-ularly in situations where exposure to predators is chronic, infer-ence about which habitats animals prefer could be biased. More-over, nonconsumptive effects of predators, such as lost foragingopportunities, could manifest as nutritional stress or starvationthat is incorrectly attributed to changes in productivity.

Here, we show using unprecedented telemetry data from syn-chronously tracked and interacting predator (killer whale Orci-nus orca) and prey (narwhal Monodon monoceros) collectedin the Eastern Canadian Arctic (ECA) that persistent inter-action with killer whales induces changes in both behaviorand habitat use of narwhal. Previous findings, in this system(26) and elsewhere (6, 27–32), have shown that killer whales eli-cit a variety of antipredator responses in other marine mamm-als. However, these earlier observations are generally limited to

Significance

Predators are widely understood to impact the structureand stability of ecosystems. In the Arctic, summer sea ice israpidly declining, degrading habitat for Arctic species, such aspolar bears and ringed seals, but also providing more accessto important predators, such as killer whales. Using datafrom concurrently tracked predator (killer whales) and prey(narwhal), we show that the presence of killer whales sig-nificantly changes the behavior and distribution of nar-whal. Because killer whales are effective predators of manymarine mammals, similar predator-induced changes would beexpected in the behavior of tracked animals in marine ecosys-tems worldwide. However, these effects are rarely consideredand may frequently go unrecognized.

Author contributions: C.J.D.M., J.W.H., S.D.P., and S.H.F. designed research; C.J.D.M., B.L.,J.O., and N.R.R. performed research; G.A.B. and M.M. analyzed data; and G.A.B. andC.J.D.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The data reported in this paper are available in Dataset S1.

1To whom correspondence should be addressed. Email: gabreed@alaska.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1611707114/-/DCSupplemental.

2628–2633 | PNAS | March 7, 2017 | vol. 114 | no. 10 www.pnas.org/cgi/doi/10.1073/pnas.1611707114

Dow

nloa

ded

by g

uest

on

June

27,

202

0

ECO

LOG

Y

Fig. 1. Map of all tracking data after sSSM fitting. Numbers indicate dayof killer whale tag deployment (the first point of every fifth day is num-bered to indicate days since deployment). Red and blue colors indicate sSSM-inferred behavior for narwhal—all seven narwhal tracks are plotted usingthe red/blue color code for behavioral state. Killer whale and narwhal tag-ging locations are indicated by yellow and cyan circles, respectively. Inferredbehavior is not shown for the tracked killer whale, which is plotted in green.

either (i) immediate antipredator responses made in conjunc-tion with direct observations on a small number of killer whalepredation events or (ii) simulated encounters, where behavior iselicited using killer whale vocalization playback. These methodslimit inferential scope to behaviors observed immediately proxi-mate to the predation or playback events. Unlike earlier studies,

Fig. 2. Behavioral time series for three tracked narwhal. Colors indicate behavioral state, with red indicating resident behavior and blue indicating inferredtransit. Black line segments indicate the movement vector at each time step for narwhal; orange lines indicate the distance the narwhal is from the taggedkiller whale at that time step, and green lines indicate the move displacement vector made by the killer whale at that time step (displayed only when<15 km from the tracked narwhal). Note the shift in narwhal behavior coinciding with killer whale departure from Admiralty Inlet. Time-series of alltracked narwhal included in SI Appendix, Fig. S2-2.

we show that behavioral changes extend beyond discrete preda-tion or attack events and that the mere presence of killer whalesin a system can cause relatively large and persistent changesin behavior and space use in prey species. Our data show thatchanges persist for the entire period of exposure of narwhal tokiller whales. Narwhal behavior quickly returned to normal afterkiller whales left the system. These dynamics, persistent changein a system (in this case, narwhal behavior) while an affectingagent (killer whales) is present followed by rapid recovery afterthe affecting agent is removed, suggest that killer whales act as apress disturbance when present.

Our findings also have relevance for the future of Arcticecosystems. Arctic summer sea ice cover is declining (33, 34),which is affecting lower trophic levels through increased pri-mary productivity, changes in plankton community structure, andaltered benthic–pelagic coupling (35–37). Sea ice loss also affectsice-dependent upper trophic-level species, such as ringed seals(Phoca hispida), bearded seals (Erignathus barbatus), and polarbears (Ursus maritimus), that use the ice as a platform to for-age and breed (38–43). In addition, summer sea ice historicallyserved as a barrier to many open water species. Ice degradationnow allows a number of marine mammal species, with limitedor no historical presence in the Arctic, regular summer access(44, 45). In the ECA, killer whales were historically limited tomore open segments and blocked from large areas, including allof Hudson Bay. Now, in areas killer whales had historical sum-mer presence, such as in Davis Strait and Lancaster Sound (46),they arrive earlier, leave later, and are more numerous (47, 48),whereas in regions of no historical presence, such as HudsonBay, they are observed annually (47). Thus, Arctic warming maybring new temporal and spatial predation threats, which havethe potential to reshape Arctic marine mammal distributions andbehavior.

Behavior of Narwhal Exposed to Killer Whale PredationTo understand how predation risk from killer whales affects preybehavior, Argos tracking tags were simultaneously deployed onkiller whales and narwhal in Admiralty Inlet on the northern-most

Breed et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2629

Dow

nloa

ded

by g

uest

on

June

27,

202

0

0 200 400 600 800

010

2030

4050

60 exposurepost−exposure

Water Depth (m)Fr

eque

ncy

0 5 10 15 20

050

100

150

Distance to Coastline (km)

Freq

uenc

y

Fig. 3. Histograms of (Upper) depth and (Lower) distance from shore for nar-whal during the exposure and postexposure periods. Histograms of individualnarwhal are available in SI Appendix, Supplementary Results and Figures.

end of Baffin Island in 2009. Admiralty Inlet is a long (300 km),wide (50 km), and deep (>1,000 m) fjord that functions as a semi-closed marine ecosystem (72.55◦ N, 86.28◦ W) (Fig. 1).

We tracked one killer whale beginning August 15, 2009, whicheffectively represented a group composed of 12–20 individualsassumed to be acting as a cohesive unit (SI Appendix, Supplemen-tary Results and Figures). The group departed Admiralty Inlet onAugust 28, 2009 at 0000 h. Seven narwhal were contemporane-ously tracked (tags deployed between August 15 and 18, 2009);predator and prey shared Admiralty Inlet for ∼10 d until thedeparture of the killer whale group. This departure affordeda natural experiment, allowing for comparison of behavior andspace use of narwhal under predation risk (exposure period)with their behavior immediately after departure of killer whales(postexposure period). Tracking data were first analyzed with abehavioral-state switching-state space model (sSSM) (49–51) fol-lowed by a series of mixed-effects models to estimate the effectof killer whales on narwhal behavior and habitat use.

State Space Model Fits. Narwhal movement behavior was dis-tinctly different from killer whale movement; sSSM fits easilydiscriminated two behavioral states in all narwhal tracks, indi-cating clear switches between transit (highly autocorrelated) andresident (negatively or nonautocorrelated) (more details are inref. 50) movement types. Killer whale movement, by contrast,was not discriminated into two clear states. This lack of cleardiscrimination is likely owing to a patrolling movement patternthat remained highly autocorrelated at all times (Figs. 1 and 2and SI Appendix, Fig. S2-2). sSSM-fitted tracks were used in allsubsequent analyses.

Fitted tracks from narwhal indicate that seven individuals didnot move as a cohesive social unit, although some individu-als occasionally swam near (within 1 km) each other for short

periods, consistent with fission–fusion dynamics of many smallcetaceans (52, 53). This dynamic is an important considerationin the analysis and interpretation, because movement patternsand reactions to killer whale presence were not correlated acrossindividuals via processes of group animal movement dynamics(54). Narwhal migrated out of Admiralty Inlet several weeksafter the tagged killer whale group, and departure was asyn-chronous across individuals.

Effect of Killer Whales on Narwhal Habitat Use. Habitat use stronglydiffered between the exposure and postexposure periods. Duringthe exposure period, narwhal were almost entirely constrainedto a narrow band of water directly adjacent to shore, with themost highly used region within 500 m of coastlines. During thepostexposure period, narwhal moved offshore, generally usingareas between 4 and 10 km from coastlines while avoiding areas<3 or >10 km from shore (Fig. 3). Apparent depth preferencemirrored this pattern, with narwhal predominately using areas<100-m deep during the exposure period and >400-m deep dur-ing the postexposure period (Figs. 3 and 4).

Mixed effects model fits indicate that general presence ofkiller whales (exposure vs. postexposure periods), distance tokiller whales, and inferred behavioral state all significantly pre-dicted distance from shore in single-parameter models, withexposure category being by far the most predictive (Table 1). Ofthe various multiparameter models fit, a simple additive modelincluding behavioral state, exposure category, and distance tokiller whales was most predictive, although a two-parametermodel including only exposure category and behavioral state fitnearly as well, suggesting little additional information in the pre-cise distance from the killer whales. Parameter estimates indi-cate that exposure resulted in habitat use very close to shore,that transit behavior was associated with being close to shore,and that, as distance to killer whales increased, narwhal tendedto move farther from shore. The results indicate that presence of

Fig. 4. Empirical probability density functions (epdfs) of habitat use bynarwhal as a function of distance from shore during the exposure andpostexposure periods mapped onto Admiralty Inlet visualizing large changein apparent habitat preference of narwhal when exposed to killer whales(epdfs shown as histograms in Fig. 3).

2630 | www.pnas.org/cgi/doi/10.1073/pnas.1611707114 Breed et al.

Dow

nloa

ded

by g

uest

on

June

27,

202

0

ECO

LOG

Y

Table 1. Mixed effects model fits predicting distance from shoreof tagged narwhal

Model: Dsh ∼ df AIC ∆AIC LR P

Ekw × B × Dkw 11 4,846 3 — —Ekw + B + Dkw 7 4,843* 0 4.4 0.036Ekw + B + Ekw × B 7 4,847 4 — —Dkw + Ekw + Dkw × Ekw 7 4,854 11 — —Dkw + B + Dkw × B 7 4,878 35 — —Ekw + B 6 4,845† 2 13.2 <0.001Dkw + Ekw 6 4,855 12 3.9 0.047Dkw + B 6 4,884 41 — —Ekw 5 4,857† 14 162.8 <0.001Dkw 5 4,918 75 101.6 <0.001B 5 4,959 116 60.8 <0.001Dsh ∼ 1 (null model) 4 5,017 174 — —

Models are organized by level of complexity. ∆Akaike Information Crite-rion (AIC) is relative to the best-fitting model; likelihood ratios and P valuescalculated from likelihood ratios are relative to the best model in the setof models one df lower in complexity. B, logit continuous sSSM inferredbehavioral state; Dkw , log distance between tagged narwhal and taggedkiller whale; Dsh, log distance to shore; Ekw , exposure category; LR, likeli-hood ratio. For parameter estimates of selected models see SI Appendix,Full Output of Mixed-Effects and GLMM Best-Fitting Models.*Overall best-fitting model.†Best-fitting model at each level of complexity as assessed by AIC scores.

killer whales in Admiralty Inlet strongly altered apparent habitatpreference, whereas the precise distance had only a small addi-tional predictive effect.

Effect of Killer Whales on Narwhal Movement Behavior. Gener-alized linear mixed models (GLMMs) also indicate a signifi-cant effect of killer whale presence on narwhal behavior. Insingle-parameter models, depth, distance to shore, exposure cat-egory, and distance to killer whales all significantly predictedbehavioral state compared with the null model (Table 2). Depthwas slightly favored over exposure category, but this effect islikely because killer whale exposure affects habitat selectionto such a degree that depth and exposure category becomecollinear. The best model overall included depth, exposure cat-egory, and an interaction between the two. Parameter esti-mates indicate that killer whale exposure considerably increasedthe probability of being in the transit state as did shallowerwater. The interaction parameter, however, indicated that nar-whal exposed to killer whales in deep water were more likely tobe in the transit state, despite the general pattern of deeper waterbeing associated with resident-type movement behavior.

Step lengths did not differ between exposure and postexpo-sure periods (SI Appendix, Fig. S2-4). Turn angles, however,tended to be straighter (more turns near 0◦) during the expo-sure period, whereas the postexposure period included moreturns near 180◦ (SI Appendix, Fig. S2-3). These differences werestatistically significant (P < 0.001) as assessed using a Watsontwo-sample test for circular data (55). Dive behavior was alsoaffected by the presence of killer whales, which caused narwhalto perform deeper dives about 10% more frequently and shortendives by about 25 s (14%). Although these differences weresmall, they were significant and could impact energetic expendi-ture and foraging opportunities more than the differences mightsuggest (56). Full results are available in SI Appendix, Supplemen-tary Results and Figures.

Discussion and ImplicationsKiller whale call playback experiments and direct observations ofattacks show immediate strong evasive responses to killer whalesin a variety of marine mammal prey (6, 27, 30, 31). Field notes,Inuit observations, and more recent scientific work describing

killer whale attacks on narwhal (26, 57–59) indicate that nar-whal similarly initiate evasive behaviors during or immediatelyafter killer whale attacks. However, in all of these reports, infer-ence about evasive response has been limited in temporal scopeto periods immediately proximate to observed attacks becauseof limited observations on predator behavior, prey behavior, orboth. Our findings, however, indicate that behavioral changes innarwhal extend beyond predation events, with altered behaviorand habitat use persisting steadily for the duration that killerwhales share habitat with their narwhal prey. These alteredbehaviors and habitat usage clearly represent nonconsumptivepredator effects.

In terrestrial habitats, telemetry data have provided convinc-ing evidence of strong nonconsumptive effects of predators onprey in a variety of species (8, 13–15). In open marine systems,however, data showing nonconsumptive or intimidation effects ofpredators on prey are lacking. Although great efforts have beeninvested in recent attempts to simultaneously track predator andprey in open ocean systems, they have mostly yielded mixed results(60–62). Before these efforts, standard survey methods had beenused to show changes in the distribution of prey species attributedto the presence of predatory sharks in a few important stud-ies of large marine vertebrates (21, 22, 63, 12, 64). Our findingsnotwithstanding, there remains a paucity of data clearly showingnonconsumptive or trait-mediated effects in open marine systems,although the importance of these effects is frequently evoked indiscussions of marine predator–prey relationships and the effectof predators on marine ecosystems (65–68).

Our findings have immediate relevance to the predator–preydynamic between narwhal and killer whales in the ECA andthe effect that increased exposure to killer whales might haveon Arctic marine mammals as sea ice degrades. Previous workhas estimated the potential consumptive predatory mortality ofkiller whales on key Arctic marine mammals and discussed thepotential for increased killer whale presence in the Arctic toaffect these species (47, 69). Nonconsumptive effects shouldalso be considered and could be strong. Even small changes in

Table 2. Mixed effects model fits predicting sSSM fit behavioralstate of tagged narwhal

Model: Bcat ∼ df AIC ∆AIC LR P

Dsh × Ekw × Dkw 9 108.8† 5.6 — —Dpth × Ekw + Ekw × Dkw 7 103.7† 1.1 — —Dsh × Ekw + Ekw × Dkw 7 106.1 3.5 — —Ekw × Dpth 5 102.6* 0 5.7 0.016Ekw + Dpth + Dkw 5 109.3 6.7 — —Ekw × Dsh 5 109.6 7 — —Ekw × Dkw 5 118.0 15.4 — —Ekw + Dpth 4 106.4† 3.8 5.5 0.019Ekw + Dsh 4 107.7 5.1 4.1 0.042Ekw + Dkw 4 122.3 19.7 — —Dpth 3 109.8† 7.2 35.6 <0.001Ekw 3 115.6 13 29.8 <0.001Dsh 3 120.9 18.3 24.4 <0.001Dkw 3 135.6 33 9.7 0.002Bcat ∼ 1 (null model) 2 143.4 40.8 — —

Models are organized by level of complexity. ∆Akaike Information Crite-rion (AIC) is relative to the best-fitting model; likelihood ratios and P valuescalculated from likelihood ratios are relative to the best model in the set ofmodels one df lower in complexity. Bcat , SSM inferred behavioral category;Dkw , log distance between tagged narwhal and tagged killer whale; Dpth,log water depth; Dsh, log distance to shore; Ekw , exposure category; LR, like-lihood ratio. For parameter estimates of selected models see SI Appendix,Full Output of Mixed-Effects and GLMM Best-Fitting Models.*Overall best-fitting model.†Best-fitting model at each level of complexity as assessed by AIC scores.

Breed et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2631

Dow

nloa

ded

by g

uest

on

June

27,

202

0

behavior caused by whale-watching vessels were associated withmodest but significant increases in energetic costs, but dispro-portionately larger losses in foraging opportunities (8). In addi-tion (56) showed that nonconsumptive effects could be moreimportant than consumptive effects in most predator–prey sys-tems analyzed.

The changes in narwhal habitat use and behavior that we doc-ument are likely representative of effects felt by all narwhalsummering in Admiralty Inlet [∼35,000 individuals; represent-ing about one-fifth of the world population (70, 71)]. Noncon-sumptive effects, even if small, across such a large populationcould have an appreciable impact at the population level andcould exceed consumptive mortality (8). Moreover, seals (ringed,harbor, and bearded), narwhal, and beluga, key prey species ofkiller whales in the ECA, are mesopredators, and there exists apotential for increased densities of killer whales to elicit struc-tural changes to Arctic ecosystems mediated through both con-sumptive and trait-mediated mesopredator effects (3, 72, 73).Such structural changes are more likely in light of the predicteddecline and possible loss of polar bears, the current apex preda-tor, as sea ice is lost (42, 43, 74–76).

Although changes in Arctic predator regimes are importantand widely relevant findings of our analysis, at least as impor-tant is the interpretation of marine animal telemetry data world-wide. Killer whales are globally distributed predators of marinemammals (77–79), and large predatory sharks are also presentacross large areas of the world’s oceans (80). In our study, infor-mation about the predator’s location was key to understandinghow movements of narwhal, as observed via satellite telemetry,were affected by predator threat. However, interpretation andanalysis of movement data collected in marine systems rarely, ifever, include knowledge of exact predator positions or even gen-eral predator densities or distributions.

Effects of predators likely receive less attention because mea-suring their impact on tracked animals remains logistically daunt-ing. There is also somewhat less acceptance that predators areimportant in marine ecosystems compared with their terres-trial counterparts, despite recent advances suggesting otherwise(66, 81–84). Not considering these effects may lead to incorrectinference about an animal’s biology and could be problematicwhere tracked data are used to formulate management advice.Use of tracking data for active marine management is becom-ing an important aspect of applied animal telemetry (23, 85–88).Researchers and managers using tracking data to infer preferredhabitat from which management policy is drawn need to carefully

consider how predators affect space use. Analyses that use track-ing data to estimate preferred habitat might erroneously inferareas that animals use as refugia from predators to be preferredareas for foraging, resting, or reproduction if they have no infor-mation about the presence and effect of apex predators on thedistribution and behavior of tracked animals.

In this system, killer whale predators share habitat withnarwhal and other potential Arctic prey species for 1–2 mo,and although this period is lengthening as sea ice degrades,most Arctic marine mammals are free from killer whale pre-dation for most of the year. In other areas, such as the NorthPacific and North Atlantic, killer whales and large sharks havea perennial presence, and small changes in their behavior,density, or habitat use could provoke large nonconsumptiveeffects on prey species (89). Observed nutritional stress, repro-ductive failure, and starvation are often attributed to changesin primary or secondary production (90–92). Such symptoms,however, would also be expected if apex predators increasedtheir presence in important foraging areas, such that forag-ing prey species avoided them in favor of more marginal habi-tat. Without knowledge or understanding of nonconsumptiveeffects of predators, conclusions about movement behavior orchanges in demographic parameters may be misleading or in-complete.

MethodssSSMs were used to estimate locations and infer behavior from noisy Argostracking data (49–51, 93). After sSSM fitting, we constructed a series ofmixed effects models and GLMMs to compare apparent habitat preferenceand sSSM-inferred behavioral state of tracked narwhal while exposed tokiller whales and during a postexposure period using an approach similarto that used in ref. 50. Habitat parameters were simply distance to nearestshoreline and water depth, which we expected to be strongly influencedby killer whales based on previous work (26). Using binned dive summarydata, we also compared differences in maximum dive depth, dive duration,and time at depth between exposure categories using GLMMs with Pois-son distribution and log link. Full methodological details are in SI Appendix,Methods.

ACKNOWLEDGMENTS. We thank Nataq Levi, Elly Chmelnitsky, GretchenFreund, Charlie Inuarak, Enookie and Michael Inuarak, and Sandie Black.Ikajutit Hunters and Trappers Organization (Arctic Bay, Nunavut) providedessential support and assistance. Funding was provided by Fisheries andOceans Canada, Ocean Tracking Network, International Governance Strat-egy, Oceans North, Nunavut Wildlife Management Board, World WildlifeFund Canada, ArcticNet, the Carlsberg Foundation, US National ScienceFoundation, Polar Continental Shelf Program, the University of Manitoba,and the University of Alaska.

1. Croll DA, Maron JL, Estes JA, Danner EM, Byrd G (2005) Introduced predators trans-form subarctic islands from grassland to tundra. Science 307:1959–1961.

2. Ripple WJ, et al. (2014) Status and ecological effects of the world’s largest carnivores.Science 343:1241484.

3. Ripple WJ, et al. (2016) What is a trophic cascade? Trends Ecol Evol 31:842–849.4. Boonstra R, Hik D, Singleton GR, Tinnikov A (1998) The impact of predator-induced

stress on the snowshoe hare cycle. Ecol Monogr 68:371–394.5. Clinchy M, Zanette L, Boonstra R, Wingfield JC, Smith JN (2004) Balancing food

and predator pressure induces chronic stress in songbirds. Proc Biol Sci 271:2473–2479.

6. Allen AN, Schanze JJ, Solow AR, Tyack PL (2014) Analysis of a Blainville’s beakedwhale’s movement response to playback of killer whale vocalizations. Mar MammSci 30:154–168.

7. Mukherjee S, Heithaus MR, Trexler JC, Ray-Mukherjee J, Vaudo J (2014) Perceived riskof predation affects reproductive life-history traits in Gambusia holbrooki, but not inHeterandria formosa. PLoS One 9:e88832.

8. Preisser EL, Bolnick DI, Benard MF (2005) Scared to death? The effects of intimidationand consumption in predator-prey interactions. Ecology 86:501–509.

9. Peckarsky BL, et al. (2008) Revisiting the classics: Considering nonconsumptive effectsin textbook examples of predator-prey interactions. Ecology 89:2416–2425.

10. Laundre JW, Hernandez L, Altendorf KB (2001) Wolves, elk, and bison: Reestab-lishing the “landscape of fear” in Yellowstone National Park, USA. Can J Zool 79:1401–1409.

11. Creel S, Christianson D, Liley S, Winnie JA (2007) Predation risk affects reproductivephysiology and demography of elk. Science 315:960–960.

12. Wirsing AJ, Heithaus MR, Frid A, Dill LM (2008) Seascapes of fear: Evaluating sublethalpredator effects experienced and generated by marine mammals. Mar Mamm Sci24:1–15.

13. Fortin D, et al. (2005) Wolves influence elk movements: Behavior shapes a trophiccascade in Yellowstone National Park. Ecology 86:1320–1330.

14. Laporte I, Muhly TB, Pitt JA, Alexander M, Musiani M (2010) Effects of wolves onelk and cattle behaviors: Implications for livestock production and wolf conservation.PLoS One 5:e11954.

15. Laundre JW (2010) Behavioral response races, predator–prey shell games, ecology offear, and patch use of pumas and their ungulate prey. Ecology 91:2995–3007.

16. Ripple WJ, Larsen EJ, Renkin RA, Smith DW (2001) Trophic cascades among wolves, elkand aspen on Yellowstone National Park’s northern range. Biol Conserv 102:227–234.

17. Ripple WJ, Beschta RL (2012) Trophic cascades in Yellowstone: The first 15 years afterwolf reintroduction. Biol Conserv 145:205–213.

18. Kauffman MJ, Brodie JF, Jules ES (2010) Are wolves saving Yellowstone’s aspen? Alandscape-level test of a behaviorally mediated trophic cascade. Ecology 91:2742–2755.

19. Middleton AD, et al. (2013) Animal migration amid shifting patterns of phe-nology and predation: Lessons from a Yellowstone elk herd. Ecology 94:1245–1256.

20. Middleton AD, et al. (2013) Linking anti-predator behaviour to prey demographyreveals limited risk effects of an actively hunting large carnivore. Ecol Lett 16:1023–1030.

21. Heithaus MR, Dill LM (2002) Food availability and tiger shark predation risk influencebottlenose dolphin habitat use. Ecology 83:480–491.

2632 | www.pnas.org/cgi/doi/10.1073/pnas.1611707114 Breed et al.

Dow

nloa

ded

by g

uest

on

June

27,

202

0

ECO

LOG

Y

22. Heithaus MR, et al. (2007) State-dependent risk-taking by green sea turtles medi-ates top-down effects of tiger shark intimidation in a marine ecosystem. J Anim Ecol76:837–844.

23. Block B, et al. (2011) Tagging of Pacific Pelagics: Tracking apex marine predator move-ments in a dynamic ocean. Nature 475:86–90.

24. Hussey NE, et al. (2015) Aquatic animal telemetry: A panoramic window into theunderwater world. Science 348:1255642.

25. Hays GC, et al. (2016) Key questions in marine megafauna movement ecology. TrendsEcol Evol 31:463–475.

26. Laidre KL, Heide-Jørgensen MP, Orr JR (2006) Reactions of narwhals, Monodon mono-ceros, to killer whale, Orcinus orca, attacks in the eastern Canadian Arctic. Can FieldNat 120:457–465.

27. Fish J, Vania J (1971) Killer whale, Orcinus orca, sounds repel white whales,Delphinapterus-leucas. Fish Bull 69:531–535.

28. Jefferson TA, Stacey PJ, Baird RW (1991) A review of killer whale interactions withother marine mammals: Predation to co-existence. Mamm Rev 21:151–180.

29. Pitman RL, Ballance LT, Mesnick SI, Chivers SJ (2001) Killer whale predation on spermwhales: Observations and implications. Mar Mamm Sci 17:494–507.

30. Cure C, et al. (2013) Responses of male sperm whales (Physeter macrocephalus) tokiller whale sounds: Implications for anti-predator strategies. Sci Rep 3:1579.

31. Cure C, et al. (2015) Predator sound playbacks reveal strong avoidance responses in afight strategist baleen whale. Mar Ecol Prog Ser 526:267–282.

32. Pitman RL, et al. (2016) Humpback whales interfering when mammal-eating killerwhales attack other species: Mobbing behavior and interspecific altruism? MarMamm Sci 33(1):7–58.

33. Stroeve J, Holland MM, Meier W, Scambos T, Serreze M (2007) Arctic sea ice decline:Faster than forecast. Geophys Res Lett 34:L09501.

34. Parkinson CL, Comiso JC (2013) On the 2012 record low Arctic sea ice cover:Combined impact of preconditioning and an August storm. Geophys Res Lett 40:1356–1361.

35. Arrigo KR, van Dijken G, Pabi S (2008) Impact of a shrinking Arctic ice cover on marineprimary production. Geophys Res Lett 35:L19603.

36. Arrigo KR, van Dijken GL (2015) Continued increases in Arctic Ocean primary produc-tion. Prog Oceanogr 136:60–70.

37. Barber DG, et al. (2015) Selected physical, biological and biogeochemical implicationsof a rapidly changing Arctic Marginal Ice Zone. Prog Oceanogr 139:122–150.

38. Kovacs KM, Lydersen C, Overland JE, Moore SE (2011) Impacts of changing sea-iceconditions on Arctic marine mammals. Mar Biodivers 41:181–194.

39. Luque SP, Ferguson SH, Breed GA (2014) Spatial behaviour of a keystone Arctic marinepredator and implications of climate warming in Hudson Bay. J Exp Mar Biol Ecol461:504–515.

40. Pilfold NW, McCall A, Derocher AE, Lunn NJ, Richardson E (2016) Migratory responseof polar bears to sea ice loss: To swim or not to swim. Ecography (Cop.) 40(1):189–199.

41. Stirling I, Lunn NJ, Iacozza J (1999) Long-term trends in the population ecologyof polar bears in western Hudson Bay in relation to climatic change. Arctic 52:294–306.

42. Molnar PK, Derocher AE, Thiemann GW, Lewis MA (2010) Predicting survival, repro-duction and abundance of polar bears under climate change. Biol Conserv 143:1612–1622.

43. Molnar PK, Derocher AE, Klanjscek T, Lewis MA (2011) Predicting climate changeimpacts on polar bear litter size. Nat Commun 2:186.

44. Moore SE, Huntington HP (2008) Arctic marine mammals and climate change: Impactsand resilience. Ecol Appl 18:S157–S165.

45. Wang M, Overland JE (2009) A sea ice free summer Arctic within 30 years? GeophysRes Lett 36:L07502.

46. Reeves RR, Mitchell E (1988) Killer whale sightings and takes by American pelagicwhalers in the North Atlantic. Rit Fiskideildar 11:7–23.

47. Higdon JW, Ferguson SH (2009) Loss of Arctic sea ice causing punctuated changein sightings of killer whales (Orcinus orca) over the past century. Ecol Appl 19:1365–1375.

48. Higdon JW, Hauser DDW, Ferguson SH (2012) Killer whales (Orcinus orca) in the Cana-dian Arctic: Distribution, prey items, group sizes, and seasonality. Mar Mamm Sci28:E93–E109.

49. Jonsen ID, Mills Flemming J, Myers RA (2005) Robust state-space modeling of animalmovement data. Ecology 86:2874–2880.

50. Breed GA, Jonsen ID, Myers RA, Bowen WD, Leonard ML (2009) Sex-specific, seasonalforaging tactics of adult grey seals (Halichoerus grypus) revealed by state-space anal-ysis. Ecology 90:3209–3221.

51. Jonsen ID, et al. (2013) State-space models for bio-loggers: A methodological roadmap. Deep Sea Res Part 2 Top Stud Oceanogr 88:34–46.

52. Connor RC, Wells RS, Mann J, Read AJ (2000) Cetacean Societies, eds Mann J,Connor RC, Tyack P, Whitehead H (Univ of Chicago Press, Chicago), pp 91–126.

53. Aureli F, et al. (2008) Fission-fusion dynamics. Curr Anthropol 49:627–654.54. Codling E, Pitchford J, Simpson S (2007) Group navigation and the “many-wrongs

principle” in models of animal movement. Ecology 88:1864–1870.55. Watson GS, Williams EJ (1956) On the construction of significance tests on the circle

and the sphere. Biometrika 43:344–352.56. Williams R, Lusseau D, Hammond PS (2006) Estimating relative energetic costs of

human disturbance to killer whales (Orcinus orca). Biol Conserv 133:301–311.57. Degerbol M, Freuchen P (1935) Mammals. Report of the Fifth Thule Expedition, 1921-

24. The Danish Expedition to Arctic North America in Charge of Knud Rasmussen, PhD(Gyldendalske Boghandel, Nordisk Forlag, Copenhagen), Vol 2.

58. Steltner H, Steltner S, Sergeant D (1984) Killer whales, Orcinus orca, prey on narwhals,Monodon monoceros an eyewitness account. Can Field Nat 98:458–462.

59. Campbell R, Yurick D, Snow N (1988) Predation on Narwhals, Monodon monoceros,by killer whales, Orcinus orca, in the Eastern Canadian Arctic. Can Field Nat 102:689–696.

60. Hayes SA, et al. (2013) Mobile receivers: Releasing the mooring to ‘see’ where fish go.Environ Biol Fishes 96:189–201.

61. Baker LL, et al. (2014) Probability of detecting marine predator-prey and speciesinteractions using novel hybrid acoustic transmitter-receiver tags. PLoS One 9:e98117.

62. Lidgard D, Bowen W, Jonsen I, Iverson S (2014) Predator-borne acoustic transceiversand GPS tracking reveal spatiotemporal patterns of encounters with acousticallytagged fish in the open ocean. Mar Ecol Prog Ser 501:157–168.

63. Wirsing AJ, Heithaus MR, Dill LM (2007) Living on the edge: Dugongs prefer to foragein microhabitats that allow escape from rather than avoidance of predators. AnimBehav 74:93–101.

64. Heithaus MR, Wirsing A, Dill L (2012) The ecological importance of intact top-predator populations: A synthesis of 15 years of research in a seagrass ecosystem.Mar Freshw Res 63:1039–1050.

65. Dill LM, Heithaus MR, Walters CJ (2003) Behaviorally mediated indirect interactionsin marine communities and their conservation implications. Ecology 84:1151–1157.

66. Heithaus M, Frid A, Wirsing A, Worm B (2008) Predicting ecological consequences ofmarine top predator declines. Trends Ecol Evol 23:202–210.

67. Ferretti F, Worm B, Britten GL, Heithaus MR, Lotze HK (2010) Patterns and ecosystemconsequences of shark declines in the ocean. Ecol Lett 13:1055–1071.

68. Swain DP, Benoıt HP, Hammill MO (2015) Spatial distribution of fishes in a northwestatlantic ecosystem in relation to risk of predation by a marine mammal. J Anim Ecol84:1286–1298.

69. Ferguson SH, Higdon JW, Westdal KH (2012) Prey items and predation behavior ofkiller whales (Orcinus orca) in Nunavut, Canada based on Inuit hunter interviews.Aquat Biosyst 8:1.

70. Heide-Jørgensen M, et al. (2010) Abundance of narwhals (Monodon monoceros) onthe hunting grounds in Greenland. J Mammal 91:1135–1151.

71. Doniol-Valcroze T, et al. (2015) Abundance Estimates of Narwhal Stocks in the Cana-dian High Arctic in 2013 (Fisheries and Oceans Canada, Ottawa), Fisheries and OceansCanada Canadian Science Advisory Secretariat Research Document 2015/060.

72. Prugh LR, et al. (2009) The rise of the mesopredator. Bioscience 59:779–791.73. Ritchie EG, Johnson CN (2009) Predator interactions, mesopredator release and bio-

diversity conservation. Ecol Lett 12:982–998.74. Laidre KL, et al. (2008) Quantifying the sensitivity of Arctic marine mammals to

climate-induced habitat change. Ecol Appl 18:S97–S125.75. Durner GM, et al. (2009) Predicting 21st-century polar bear habitat distribution from

global climate models. Ecol Monogr 79:25–58.76. Castro de la Guardia L, Derocher AE, Myers PG, Terwisscha van Scheltinga AD, Lunn NJ

(2013) Future sea ice conditions in Western Hudson Bay and consequences for polarbears in the 21st century. Glob Chang Biol 19:2675–2687.

77. Ford JK, et al. (2005) Killer whale attacks on minke whales: Prey capture andantipredator tactics. Mar Mamm Sci 21:603–618.

78. Ford JK, Reeves RR (2008) Fight or flight: Antipredator strategies of baleen whales.Mamm Rev 38:50–86.

79. Ferguson SH, Kingsley MCS, Higdon JW (2012) Killer whale (Orcinus orca) predationin a multi-prey system. Popul Ecol 54:31–41.

80. Baum JK, et al. (2003) Collapse and conservation of shark populations in the North-west Atlantic. Science 299:389–392.

81. Estes JA, Tinker MT, Williams TM, Doak DF (1998) Killer whale predation on sea otterslinking oceanic and nearshore ecosystems. Science 282:473–476.

82. Myers RA, Baum JK, Shepherd TD, Powers SP, Peterson CH (2007) Cascading effectsof the loss of apex predatory sharks from a coastal ocean. Science 315:1846–1850.

83. Baum JK, Worm B (2009) Cascading top-down effects of changing oceanic predatorabundances. J Anim Ecol 78:699–714.

84. Boyce DG, Frank KT, Worm B, Leggett WC (2015) Spatial patterns and predictors oftrophic control in marine ecosystems. Ecol Lett 18:1001–1011.

85. Maxwell SM, et al. (2011) Internesting behavior and movements of olive ridley seasurtles (Lepidochelys olivacea) in Gabon, Africa. PLoS One 6:e19905.

86. Zydelis R, et al. (2011) Dynamic habitat models: Using telemetry data to project fish-eries bycatch. Proc Biol Sci 278:3191–3200.

87. Graham RT, et al. (2012) Satellite tracking of manta rays highlights challenges to theirconservation. PLoS One 7:e36834.

88. Hazen EL, et al. (November 29, 2016) Whalewatch: A dynamic management tool forpredicting blue whale density in the California Current. J Appl Ecol, 10.1111/1365-2664.12820.

89. Moxely JH (2016) The abundance and behavioral ecology of Cape Cod gray sealsunder predation risk from white sharks. PhD thesis (Duke Univ, Durham, NC).

90. Merrick R, Loughlin T (1997) Foraging behavior of adult female and young-of-the-year Steller sea lions in Alaskan waters. Can J Zool 75:776–786.

91. Trites A, Donnelly C (2003) The decline of Steller sea lions Eumetopias jubatus inAlaska: A review of the nutritional stress hypothesis. Mamm Rev 33:3–28.

92. DeMaster DP, et al. (2006) The sequential megafaunal collapse hypothesis: Testingwith existing data. Prog Oceanogr 68:329–342.

93. Jonsen I (2016) Joint estimation over multiple individuals improves behavioural stateinference from animal movement data. Sci Rep 6:20625.

Breed et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2633

Dow

nloa

ded

by g

uest

on

June

27,

202

0