Ecology, 91(12), 2010, pp. 3563–3571 Ó 2010 by the Ecological Society of America

Field evidence for pervasive indirect effects of fishing on prey foraging behavior

1,2,5 3,4 1,3 ELIZABETH M. P. MADIN, STEVEN D. GAINES, AND ROBERT R. WARNER 1Department of Ecology, Evolution, and Marine Biology, University of California, Santa Barbara, California 93106 USA 2Department of Biological Sciences, Macquarie University, Sydney, New South Wales 2019 Australia 3Marine Science Institute, University of California, Santa Barbara, California 93106 USA 4Bren School of Environmental Science and Management, University of California, Santa Barbara, California 93106 USA

Abstract. The indirect, ecosystem-level consequences of ocean fishing, and particularly the mechanisms driving them, are poorly understood. Most studies focus on density-mediated trophic cascades, where removal of predators alternately causes increases and decreases in abundances of lower trophic levels. However, cascades could also be driven by where and when prey forage rather than solely by prey abundance. Over a large gradient of fishing intensity in the central Pacific’s remote northern Line Islands, including a nearly pristine, baseline coral reef system, we found that changes in predation risk elicit strong behavioral responses in foraging patterns across multiple prey fish species. These responses were observed as a function of both short-term (‘‘acute’’) risk and longer-term (‘‘chronic’’) risk, as well as when prey were exposed to model predators to isolate the effect of perceived predation risk from other potentially confounding factors. Compared to numerical prey responses, antipredator behavioral responses such as these can potentially have far greater net impacts (by occurring over entire assemblages) and operate over shorter temporal scales (with potentially instantaneous response times) in transmitting top-down effects. A rich body of literature exists on both the direct effects of human removal of predators from ecosystems and predators’ effects on prey behavior. Our results draw together these lines of research and provide the first empirical evidence that large-scale human removal of predators from a natural ecosystem indirectly alters prey behavior. These behavioral changes may, in turn, drive previously unsuspected alterations in reef food webs. Key words: behavior; coral reef; fishing; food web; herbivore; indirect effects; Line Islands, Central Pacific; marine; predator.

INTRODUCTION insignificant top-down effects on ecosystems. Alterna- Fishing has long affected marine systems globally tively, it may suggest that the effects of fishing can take different pathways through ecosystems. For example, (Jackson et al. 2001) but has intensified drastically in fishing could potentially have larger effects on prey recent decades (Berkes et al. 2006). Although the direct behavior than on prey density by altering the levels of impacts of fishing (such as declines of harvested species) risk prey face (Dill et al. 2003, Schmitz et al. 2004). are well documented (Myers and Worm 2003), their Small-scale behavioral responses by prey to varying indirect, ecosystem-level consequences are not well predation risk have been shown in other systems to understood. For example, in some systems predator modify community structure (Ripple et al. 2001, Dill et loss has been associated with alternating abundances or al. 2003, Preisser et al. 2005). Theory also suggests that biomasses of lower trophic level species, termed a fishing may indirectly drive comparable food web density-mediated trophic cascade. This simple expecta- alterations over large scales (Walters and Kitchell tion, while commonly observed (McClanahan and 2001). Because these responses can have far greater net Muthiga 1988, Babcock et al. 1999, Dill et al. 2003, impacts (by occurring simultaneously over entire assem- Myers et al. 2007), is not found everywhere (Strong blages) and operate over far shorter temporal scales 1992, Mumby et al. 2006, Newman et al. 2006, Sandin et (with potentially instantaneous response times; Lima al. 2008). The large variability in cascade occurrence and and Bednekoff 1999, Walters and Kitchell 2001) in magnitude could indicate that fishing frequently has transmitting top-down effects than numerical effects alone, it is critical that they be considered in studies of Manuscript received 23 November 2009; revised 3 March the indirect effects of marine predator removal. 2010; accepted 22 April 2010. Corresponding Editor: S. R. Across a wide diversity of aquatic and terrestrial Thorrold. systems, including marine, freshwater, and terrestrial 5 Present address: Department of Environmental Sciences, University of Technology Sydney, Sydney, New South Wales systems, prey are known to respond to predators by 2007 Australia. E-mail: [email protected] altering their activity levels and spatial habitat uses (see 3563 3564 ELIZABETH M. P. MADIN ET AL. Ecology, Vol. 91, No. 12 reviews by Lima and Dill [1990] and Dill et al. [2003] and risk driven by fishing: the bullethead parrotfish (Chloru- meta-analysis by Preisser et al. [2005]). Such responses rus sordidus), a large, mobile herbivore; the whitecheek reflect the inherent trade-off between obtaining food and surgeonfish (Acanthurus nigricans), a medium-bodied, avoiding predators (Lima and Dill 1990), as described mobile, herbivore; the blackbar damselfish (Plectrogly- by the ‘‘ecology of fear’’ model of Brown et al. (1999). phidodon dickii), a small, site-attached omnivore; and Nonetheless, ecological studies and models often ignore the bicolor chromis (Chromis margaritifer), a small, site- behavioral responses of prey (Peckarsky et al. 2008), attached zooplanktivore. We quantified excursions (i.e., even though many prey behaviors (e.g., grazing on distance or area of reef over which individuals move benthic algae) play key roles in structuring natural during 5-min observations; see Appendix C) as a key systems. There is a growing body of theoretical (Brown metric of prey behavior. Excursions provide access to et al. 1999, Walters and Kitchell 2001, Frid et al. 2008, food and, in some cases, mates, yet necessarily impose Orrock et al. 2008) and empirical (Ripple et al. 2001, risk, because the probability of encountering predators Heithaus and Dill 2002, Dill et al. 2003, Werner and scales with the magnitude of the excursion (Anholt and Peacor 2003, Schmitz et al. 2004, Preisser et al. 2005, Werner 1995). As such, they provide a useful means of Heithaus et al. 2008, Peckarsky et al. 2008, Stallings determining how these prey fishes titrate for resources 2008) evidence for nonconsumptive effects and resultant and safety (Brown and Kotler 2004). behaviorally mediated trophic cascades, whereby a predator affects the resources of its prey via changes in Predation risk the prey’s behavior. Behavioral responses can mediate Both predation risk and prey behavioral responses top-down effects by altering per capita foraging rates were quantified at 11 sites within the three atolls and/or the spatial distribution of foraging effort (Palmyra, N 5; Tabuaeran, N 3; and Kiritimati, N (Schmitz et al. 2004, Preisser et al. 2005), as observed 3). Sites were¼ located at least 1¼ km apart and ranged in the Rocky Mountains of North America following ¼from two to 10 m depth. Both of our key metrics of the reintroduction of wolves to areas in which they had predation risk are based on biomass, rather than previously been eradicated by hunting (Ripple et al. abundance, of piscivorous fishes. Biomass has been 2001). Importantly, this mechanism can act in concert demonstrated to be a more meaningful metric for with or independently of numerical changes in prey (Dill detecting the effects of fishing on coral reefs, particularly et al. 2003, Werner and Peacor 2003, Schmitz et al. for areas in which larger and less abundant fish are 2004). Peckarsky et al. (2008) recently reevaluated a targeted by fishing (Russ and Alcala 1998). Importantly, number of classic textbook examples of (purportedly this underlying metric incorporates the sizes, in addition density-mediated) predator–prey interactions and found to abundance, of predators facing prey fishes, an that nonconsumptive effects played a key role in driving important factor in determining perceived predation some of the observed patterns. This evidence, coupled risk. In order to determine whether results could with global declines of marine predators due to fishing, potentially be idiosyncratic to this biomass-based metric suggests that large-scale empirical exploration of the of risk, however, an alternative metric of risk based on behavioral consequences of fear-released systems is both predator abundances filtered by a size threshold was also vital and timely (Heithaus et al. 2008). We set out to test used (see Appendix D). All fishes known to be primarily whether fishing has the potential to invoke this piscivorous were included in our predator counts. mechanism in natural reef systems by determining if Individuals of Scomberoides lysan ,30 cm total length and how prey fishes alter their foraging behavior in (TL) were excluded from our counts because juveniles of response to fishing-induced predator loss. Here we this species are not primarily piscivores and feed instead quantitatively examine these behavioral consequences on fish scales torn from schooling fishes (Lieske and across the remote northern Line Islands archipelago of Myers 1994), and their location at the surface of the the central Pacific, where the landscape of risk differs water column renders them as non-interactors with the spatially as a consequence of variation in human fishing focal prey species. pressure. Key to testing the indirect, behavioral conse- Predation risk was quantified at two scales. Our quences of fishing is inclusion of intact, predator-rich measure of chronic predation risk provides an estimate food webs free from human fishing, an increasingly rare of the ambient predation risk faced by prey at a given commodity (Jackson et al. 2001). Palmyra Atoll, an site and can be thought of as risk integrated over time. incorporated Territory of the United States, experiences This metric was calculated at the site level and thus the virtually no harvesting of reef fishes, while Kiritimati method used did not vary between focal prey species. and Fanning Atolls, both of the Republic of Kiribati, Chronic risk was quantified by averaging estimates of have increasingly dense human populations and inten- biomass of all piscivorous fishes across at least 15 sities of fishing pressure (see Appendices A and B). replicate, haphazardly placed 60-m2 belt transects within each site. Ranked relative abundances of observed MATERIALS AND METHODS piscivore species found at each study atoll are given in We used four species of common non-predatory fishes Appendix K and their size probability distributions are to examine behavioral responses to reduced predation shown in Appendix J. Acute risk provides an estimate of December 2010 INDIRECT BEHAVIORAL EFFECTS OF FISHING 3565 the immediate temporal-scale risk to which each farthest markers and the two farthest markers perpen- observed individual was subject while under observa- dicular to that axis were measured. These distances were tion. Acute risk was quantified by recording all then used as the major (2a) and minor (2b) axes, piscivorous fishes that swam within a defined area respectively, of an ellipse, and its areal formula (Pab) (1.5–3 m radius, depending on prey species) surrounding was used to approximate the area of reef covered by the each focal prey fish. Acute predation risk was calculated individual. for the two focal mobile species (C. sordidus and A. For the site-attached omnivore P. dickii, excursion nigricans) as the biomass of predators that each focal area was determined by dividing the reef substrate into prey fish encountered over the 5-min observation period. four quadrants whose origin was at the center of the The number of predators encountered was calculated as focal individual’s home coral colony, then observing the the number of piscivores that came within the 3 m radius farthest distance in each quadrant that the individual sphere surrounding the focal prey individual at any time ventured from the home coral colony during the 5-min during the 5-min observation period. For the two focal observation period. At the conclusion of the period, four prey species whose site-attached nature rendered it markers were placed at these points within each of the possible (P. dickii and C. margaritifer), this metric also quadrants, and the distances between opposing pairs incorporated the amount of time that each predator measured. These distances were then used in the spent in the immediate area surrounding the focal previous formula to approximate the area of reef individual. For these species, the immediate area was covered by each individual. In order to standardize by defined as the 3-m side length square cube of water the area of shelter provided by the home coral colony surrounding the focal individual. In this instance, acute within this area, the area of the coral colony was risk was calculated as the biomass of predators measured (using its longest and perpendicular axes as multiplied by the duration of their visit that each focal the major and minor axes, respectively) and the ratio of prey fish encountered over the 5-min observation period. excursion area to shelter area calculated. For the Acute risk was measured only at Palmyra and Tabu- comparison of this species’ excursion size as a function aeran. Both surveys of predation risk and observations of chronic predation risk, all instances where large, of behavioral responses were conducted using a mix of finely branching Acropora spp. served as the individual’s snorkel and scuba apparatus, with both methods home coral colony were excluded because this coral employed at all atolls. morphology was only found at one atoll (Palmyra). For the site-attached zooplanktivore, C. margaritifer, Behavioral observations excursion distance was determined by estimating the Prey fish observations were conducted for 5 min, with focal individual’s distance in the water column from a ;3-min habituation period for each focal individual their home coral colony at regular (10-s) intervals over prior to observation. Behavioral observations of prey the observation period. The mean of these (30) estimates individuals followed approximately the protocol out- was then calculated and served as the excursion distance lined in Bellwood (1995). All observations were con- value for that individual. ducted between 08:00 and 16:30 hours. Observers remained between 3 and 5 m from focal individuals at Model predator trials all times during the 5-min observation period, except In order to isolate effects of perceived predation risk when the fish under observation swam toward the on prey species, additional focal individuals of P. dickii observer. In such cases, which occurred at all atolls in and C. margaritifer were observed when presented with approximately the same frequency, the observer re- model predators of three sizes (plastic sharks of 14, 25, mained motionless until the fish was again at least 3 m and 74 cm TL). Although these models were of the form away. Observation of an individual was discontinued of small sharks, our objective was simply to present an and data discarded if either the fish was lost or showed object that prey individuals would perceive as a any indication of observer interference (e.g., hiding, potential predation threat; objects of similar sizes but flight), although the latter (observer interference) was other shapes could be substituted. These model pisci- never observed. Observations were also discontinued if vores were utilized to mimic acute risk for only these two water visibility fell to ,5 m. species because it was impractical to conduct compara- For the two mobile focal prey species, excursion area ble in situ experiments with mobile grazers. was determined by dropping numbered subsurface Each of 15 focal individuals of each of these species markers (weights attached to corks with ;25 cm of was first observed for three minutes immediately prior to flagging tape) at regular (30-s) intervals along the offset presentation of the series of model predators. This time path of the focal fish over the observation period. period served as a control for each individual’s behavior Markers were placed ;5 m downstream of the focal under natural conditions and was contiguous with the individual, resulting in a record of the path that the model predator presentations. During this time, no individual took over the observation period offset from object was presented to the focal individual. At the end the actual path by 5 m. At the conclusion of each 5-min of the 3-min period, the focal individual was sequentially observation period, the distance between the two exposed to remotely operated small, medium, and large 3566 ELIZABETH M. P. MADIN ET AL. Ecology, Vol. 91, No. 12

FIG. 1. Prey excursion size in relation to chronic predation risk for four prey species of different functional groups. Excursions refer to the distance or area of reef over which prey individuals move during 5-min observations. Prey species are bullethead parrotfish (Chlorurus sordidus), whitecheek surgeonfish (Acanthurus nigricans), blackbar damselfish (Plectroglyphidodon dickii), and bicolor chromis (Chromis margaritifer). Points are site means (6 SE) within atolls in the remote northern Line Islands, central Pacific: Palmyra (circles); Tabuaeran (squares); and Kiritimati (triangles). model predators via an underwater pulley system, with behaviors with maintaining other necessary functions the order of model predator sizes varied haphazardly (foraging, reproduction, and other functions; Lima and among focal individuals. For C. margaritifer,the Dill 1990), while prey under low risk may freely engage excursion value presented for each model predator in in otherwise risky behaviors. Thus, risk should constrain Fig. 3 represents the mean of all observations of the upper bound of foraging excursions. To test this a excursion distance from shelter while the model predator priori prediction, we examined the slope of the upper was within 4 m (2 m on either side) of the focal bound of the relationship between predation risk and individual. For P. dickii, these values represent the ratio prey excursions (see Appendix E). of the maximum area over which the focal individual Across the Line Islands sites, predator biomass traveled to the shelter area provided by its home coral density (an indirect measure of fishing) is inversely colony (as described previously) during the 2-min period related to human population density (see Appendix F) in which the model predator was within 4 m (2 m on and is drastically lower in areas with intense fishing (Fig. either side) of the individual. The differences in 1, x-axes). Across the resulting gradient in predation observation methods reflect the different nature of these risk, we found strong negative associations between species’ behavior patterns (i.e., it was not possible to predator biomass density and the upper bound of the quantify behavior of these two species in the same distance or area of reef over which individuals travel manner, nor in the same manner as the mobile species’ (Fig. 1). The pattern of response was similar across a behavior). diverse suite of prey species. As expected, heavily fished sites under low chronic predation risk showed a wide RESULTS range of mean excursion sizes. By contrast, maximum Predation risk may constrain the actions prey will excursions dropped precipitously under even moderately take. Prey under high risk must balance limiting risky high predation risk. Maximum excursions were 65–86% December 2010 INDIRECT BEHAVIORAL EFFECTS OF FISHING 3567

FIG. 2. Prey excursion size in relation to acute predation risk for the four focal prey species. Solid lines show linear regression, and dashed lines show best-fit upper and lower 95% prediction intervals based on a negative log-likelihood optimization function. Points are values for individual prey where predation risk is measured by predator biomass for mobile prey species and predator biomass 3 duration for site-attached prey species. A shift parameter in the form of species-specific constants (a values) was determined by the mean body size of each focal prey species and thus its vulnerability to predators of different body and gape sizes; these values are described in Appendix E and are: 300 (C. sordidus); 250 (A. nigricans); 400 (P. dickii); and 300 (C. margaritifer). lower at high vs. low predation risk sites (Fig. 1). The H). The immediate presence of predators imposed upper excursion declines were strikingly nonlinear, with the bounds on prey risk-taking that scale strongly with bulk of the declines for all four species occurring among predator abundance and size. sites with predator biomass density ,40 g/m2. An Model piscivores elicited similar responses as prey alternative metric of predation risk based on abundance individuals responded to perceived acute predation risk with a size threshold produced patterns that were (Fig. 3). C. margaritifer displayed significant, graded qualitatively similar to the biomass metric (Appendix reductions in its excursion distance in the face of G). different sized predators relative to control periods Prey fishes also demonstrated striking reductions in (Fig. 3A; repeated-measures ANOVA; F3,80 10.165; excursion sizes in response to acute predation risk from two-tailed P , 0.001; N 22). Similarly, ¼P. dickii individual predator encounters (Fig. 2). This acute exhibited significant predator¼ avoidance behavior when response was similarly nonlinear for all species. In those faced with the largest size (74 cm TL) of model predator prey with significant declines in maximum excursion (C. (Fig. 3B; Welch two-sample paired t test; t 2.664; sordidus one-tailed P 0.057, P. dickii one-tailed P one-tailed P , 0.001; N 14). Reduced risk-taking¼À in 0.033, C. margaritifer one-tailed¼ P 0.006), the declines¼ this species was also observed¼ when presented with ¼ were as large as 90%. Again, qualitative concordance intermediate-sized predators (Fig. 3B), but the reduc- was also seen when these analyses were repeated using tions were not significant nor as clearly graded as those the abundance-based metric of predation risk (Appendix of the smaller, more vulnerable C. margaritifer. 3568 ELIZABETH M. P. MADIN ET AL. Ecology, Vol. 91, No. 12

fishing pressure, including a relatively intact baseline food web at Palmyra Atoll, our results provide evidence that human-driven, indirect behavioral effects occur in a natural, nonexperimental system. In so doing, we demonstrate that fishing has the potential to invoke the mechanism of risk-induced behavioral responses in coral reef ecosystems. Interestingly, the shape of the response between both chronic and acute risk and prey behavioral responses was clearly nonlinear. This nonlinearity suggests that thresholds of predation risk may exist that serve to suppress prey risk-taking, and that once these thresholds are exceeded prey behavior may assume a fundamentally different form—one of relatively uninhibited excursions for foraging, reproduction, and other activities. Evidence from coral reefs (Stallings 2008) and other systems (Dill et al. 2003, Werner and Peacor 2003, Schmitz et al. 2004, Preisser et al. 2005, Heithaus et al. 2008) demonstrates that behavioral responses like those shown here can trigger cascading demographic and community-wide changes. These findings, in conjunction with the results presented herein, provide compelling evidence that fishing, by removing predators and indirectly affecting prey foraging behavior, may ulti- mately lead to behaviorally mediated trophic cascades on coral reefs. Similarly, in contrast to our study of the effects of human-mediated predator loss, Ripple et al. (2001) and Creel et al. (2007) have shown that recovery of predator populations from previous human-induced FIG. 3. Prey excursion size in relation to acute perceived decline can have dramatic, cascading consequences on predation risk for the two site-attached focal prey species. Control period represents baseline behavior prior to initial both prey themselves and their food resources in model predator presentation. Model predator sizes (TL) are 14, terrestrial systems. The predation risk–safety trade-off 25, and 74 cm. Bars are means (6SE). faced by fishes in predator-rich habitats (Milinski and Heller 1978, Werner et al. 1983) provides a general DISCUSSION framework for understanding how such behavioral alterations should affect resource use. In particular, These observations of 629 individual prey fishes over the ‘‘foraging arena’’ concept (Walters and Kitchell four species and two temporal scales of risk showed 2001) provides a theoretical basis for understanding how strong and consistent behavioral responses to human- these small temporal- and spatial-scale responses may mediated predation risk. Fishing indirectly influenced scale up to have cascading impacts beyond the simple antipredator behavior of nontarget, prey fishes by behavioral responses themselves. This model suggests reducing predator densities and/or size. Many studies that fishing for predators may drive the redistribution of have demonstrated the loss of predator biomass prey per capita foraging effort from previously low- to resulting from fishing on coral reefs in general (Russ previously high-predation risk patches once predation and Alcala 1998, Friedlander and DeMartini 2002, risk is reduced or eliminated. Such changes would be Newman et al. 2006) and in the Line Islands in expected to alter the spatial distribution of prey particular (Stevenson et al. 2007, DeMartini et al. resources (e.g., benthic algae) even in the absence of 2008, Sandin et al. 2008), and others have shown that numerical changes in prey or their resources. The spatial and/or temporal variation in predation risk synthesis of Heithaus et al. (2008) and the examples affects prey behavior (Werner et al. 1983, Holbrook described within (particularly those from Shark Bay, and Schmitt 1988, Helfman 1989, Ripple et al. 2001, Western Australia; Heithaus and Dill 2002, 2006, Heithaus and Dill 2002, 2006, Biro et al. 2003, Heithaus Heithaus et al. 2007, Wirsing et al. 2007) strongly et al. 2007, Wirsing et al. 2007, Stallings 2008). Our support this expectation and provide compelling evi- study has demonstrated empirically that large-scale dence from diverse marine systems that risk effects are human removal (as opposed to reintroduction or indeed widespread and must be considered in order to recovery) of predators from a natural ecosystem generate accurate predictions of the ecological effects of indirectly affects prey behavior. By utilizing the unique marine top predator declines. Important cascading ‘‘natural experiment’’ of the Line Islands’ gradient of changes to community structure may thus be more December 2010 INDIRECT BEHAVIORAL EFFECTS OF FISHING 3569 complex than is often assumed and may not necessarily the scope of this paper, one hypothesis is that many be detected by the simple estimates of mean resource coral reefs’ macroherbivore guilds are dominated by abundance or biomass typically used to detect trophic behaviorally sophisticated herbivorous fishes, in con- cascades. trast to many other systems from which evidence of Alternative explanations for the observed spatial trophic cascades has emerged (e.g., temperate rocky patterns of the prey excursions—such as food availabil- reefs) whose macroherbivores consist primarily of ity, habitat complexity, prey body size (a proxy for invertebrates (e.g., urchins). vulnerability), and competition for food resources and/ Rarely, if ever, has the mechanism of prey behavioral or space—were all insufficient to explain the observed responses been considered in the design of fisheries patterns (see Appendix I). Furthermore, in accordance management policies (Dill et al. 2003, Heithaus et al. with the threat sensitivity hypothesis (Helfman 1989), 2008). Our findings suggest that fishing for top trophic prey exhibited marked and consistent responses to level species could potentially have widespread conse- perceived acute predation risk when that risk was quences for nontarget prey by affecting their movement isolated from the other factors by using model piscivores patterns and subsequently their food intake, reproduc- (Fig. 3). tive output, and growth rate that may go well beyond Likewise, prey vulnerability and resultant tenacity the direct consequences of reducing prey mortality. In (Brown and Kotler 2004) may help explain why the addition, by broadening the spatial extent of herbivore surgeonfish, A. nigricans, was the only prey species to grazing, fishing for reef predators could also potentially exhibit a nonsignificant reduction in maximum excur- affect the distribution ofareasuitableforcoral sion size in response to acute predation risk based on settlement (Mumby et al. 2007). Contrasting effects biomass. Although this species does not attain a may arise from policies that lead to increased predator maximum body size as great as that of the parrotfish densities, such as marine reserves or other fisheries C. sordidus, it is the only of the four study species to closures, stocking, or accidental escape from hatcheries. possess a physical antipredator defense (sharp, razor- Most ecosystems globally have already experienced like ‘‘scalpels’’ on the caudal peduncle). This defense substantial predator loss; few remain with undisturbed may reduce vulnerability to predation such that individuals are able to maintain slightly greater excur- food webs (Jackson et al. 2001). Although mechanisms sion sizes in the face of predation risk than those of the governing the structure and functioning of natural other three study species. systems may be understood to some extent from The qualitative concordance of patterns observed degraded and/or recovering ecosystems (e.g., marine using our biomass-based vs. an alternative predation reserves [Mumby et al. 2007] and predator [wolf] risk metric (see Appendices D, G, and H) suggests that reintroductions [Ripple et al. 2001]), the full complement the findings we present here are not idiosyncratic to the of ecologically important mechanisms (such as risk metric of predation risk we use. This comparison shows effects) may only be apparent through comparative that our biomass-based metric is likely to be sufficiently studies that include intact ecosystems (Heithaus et al. robust to account for potential biases caused by 2008). This notion provides yet another motivation for inclusion of all predators, even those too small to preserving the few that remain. potentially consume the focal prey species and/or ACKNOWLEDGMENTS individuals observed. The negligible contribution to This work was funded by the U.S. National Science the overall biomass estimate from small predators means Foundation, the U.S. Department of Homeland Security, the that they do not inordinately affect the total biomass National Geographic Society, the Marisla Foundation, the estimate, and their inclusion allows an estimate of Myers Oceanographic Trust, the Sigma Xi Scientific Research predation risk to be calculated based on easily obtain- Society, the Worster Family Trust, and University of Cal- able fish assemblage data with minimal assumptions ifornia–Santa Barbara Departments of Ecology, Evolution, and required (because all individuals are included, with no Marine Biology and Geography. We thank L. Carr, J. Lape, J. Madin, L. Max, J. Osborne, and M. Readdie for valuable need to make subjective determinations about species- assistance in the field and the staff of the Palmyra Research specific predation risk thresholds). Nonetheless, if Station, R. Bryden and staff, K. Andersen and staff, and detailed data become available regarding the level of Norwegian Cruise Lines for logistical field support. We thank threat posed by particular predatory taxa and/or sizes of S. Holbrook for helpful discussion and advice throughout, A. individuals to particular prey (for example, based on Friedlander and E. DeMartini for helpful advice on study species, and A. Stewart-Oaten, J. Madin, and J. de Leeuw for observed predation events or detailed gut content valuable statistical advice. S. Connolly, K. Lafferty, J. Madin, analyses for each pairwise predator and prey combina- P. Munday, and four anonymous reviewers provided valuable tion), then the predation risk metrics presented here comments. Collaborative fieldwork with E. Sala, S. Sandin, S. could be further refined to yield potentially greater Hamilton, S. Walsh, B. Ruttenberg, J. Smith, K. Lafferty, S. predictive power. Smriga, M. Roth, J. Shaw, B. Ramoy, and A. Kuris provided the basis for calculating scaling relationships for predator and Evidence for community-level trophic cascades has prey body dimensions underlying figures of Appendices G and been notably rare for coral reefs relative to other marine H. Finally, we thank the Palmyra Atoll Research Consortium, systems. While the reason(s) for this pattern are beyond The Nature Conservancy, the U.S. Fish and Wildlife Service, 3570 ELIZABETH M. P. MADIN ET AL. Ecology, Vol. 91, No. 12 the Kiribati Ministry of Environment, Lands and Agriculture and Caribbean, including the Red Sea. Harper Collins, Development, and the citizens of Tabuaeran and Kiritimati London, UK. Atolls for their support and cooperation. Lima, S. L., and P. A. Bednekoff. 1999. Temporal variation in danger drives antipredator behavior: the predation risk LITERATURE CITED allocation hypothesis. American Naturalist 153:649–659. Anholt, B. R., and E. E. Werner. 1995. Interaction between Lima, S. L., and L. M. Dill. 1990. Behavioral decisions made food availability and predation mortality mediated by under the risk of predation: a review and prospectus. adaptive behavior. Ecology 76:2230–2234. Canadian Journal of Zoology 68:619–640. Babcock, R. C., S. Kelly, N. T. Shears, J. W. Walker, and T. J. McClanahan, T. R., and N. A. Muthiga. 1988. Changes in Kenyan coral-reef community structure and function due to Willis. 1999. Changes in community structure in temperate exploitation. Hydrobiologia 166:269–276. marine reserves. Marine Ecology Progress Series 189:125– Milinski, M., and R. Heller. 1978. Influence of a predator on 134. optimal foraging behavior of sticklebacks (Gasterosteus Bellwood, D. R. 1995. Direct estimate of bioerosion by two aculeatus L.). Nature 275:642–644. parrotfish species, Chlorurus gibbus and C. sordidus, on the Mumby, P. J., et al. 2006. Fishing, trophic cascades, and the Great Barrier Reef, Australia. Marine Biology 121:419–429. process of grazing on coral reefs. Science 311:98–101. Berkes, F., et al. 2006. Ecology: globalization, roving bandits, Mumby, P. J., A. R. Harborne, J. Williams, C. V. Kappel, and marine resources. Science 311:1557–1558. D. R. Brumbaugh, F. Micheli, K. E. Holmes, C. P. Dahlgren, Biro, P. A., J. R. Post, and E. A. Parkinson. 2003. Population C. B. Paris, and P. G. Blackwell. 2007. Trophic cascade consequences of a predator-induced habitat shift by trout in facilitates coral recruitment in a marine reserve. Proceedings whole-lake experiments. Ecology 84:691–700. of the National Academy of Sciences USA 104:8362–8367. Brown, J. S., and B. P. Kotler. 2004. Hazardous duty pay and Myers, R. A., J. K. Baum, T. D. Shepherd, S. P. Powers, and the foraging cost of predation. Ecology Letters 7:999–1014. C. H. Peterson. 2007. Cascading effects of the loss of apex Brown, J. S., J. W. Laundre, and M. Gurung. 1999. The predatory sharks from a coastal ocean. Science 315:1846– ecology of fear: optimal foraging, game theory, and trophic 1850. interactions. Journal of Mammalogy 80:385–399. Myers, R. A., and B. Worm. 2003. Rapid worldwide depletion Creel, S., D. Christianson, S. Liley, and J. A. Winnie. 2007. of predatory fish communities. Nature 423:280–283. Predation risk affects reproductive physiology and demog- Newman, M. J. H., G. A. Paredes, E. Sala, and J. B. C. raphy of elk. Science 315:960. Jackson. 2006. Structure of Caribbean coral reef communi- DeMartini, E. E., A. M. Friedlander, S. A. Sandin, and E. Sala. ties across a large gradient of fish biomass. Ecology Letters 9: 2008. Differences in fish-assemblage structure between fished 1216–1227. and unfished atolls in the northern Line Islands, central Orrock, J. L., J. H. Grabowski, J. H. Pantel, S. D. Peacor, B. L. Pacific. Marine Ecology Progress Series 365:199–215. Peckarsky, A. Sih, and E. E. Werner. 2008. Consumptive and Dill, L. M., M. R. Heithaus, and C. J. Walters. 2003. nonconsumptive effects of predators on metacommunities of Behaviorally mediated indirect interactions in marine com- competing prey. Ecology 89:2426–2435. munities and their conservation implications. Ecology 84: Peckarsky, B. L., P. A. Abrams, D. I. Bolnick, L. M. Dill, J. H. 1151–1157. Grabowski, B. Luttbeg, J. L. Orrock, S. D. Peacor, E. L. Frid, A., G. G. Baker, and L. M. Dill. 2008. Do shark declines Preisser, O. J. Schmitz, and G. C. Trussell. 2008. Revisiting create fear-released systems? Oikos 117:191–201. the classics: considering nonconsumptive effects in textbook Friedlander, A. M., and E. E. DeMartini. 2002. Contrasts in examples of predator–prey interactions. Ecology 89:2416– density, size, and biomass of reef fishes between the 2425. northwestern and the main Hawaiian islands: the effects of Preisser, E. L., D. I. Bolnick, and M. F. Benard. 2005. Scared to fishing down apex predators. Marine Ecology Progress Series death? The effects of intimidation and consumption in 230:253–264. predator–prey interactions. Ecology 86:501–509. Heithaus, M. R., and L. M. Dill. 2002. Food availability and Ripple, W. J., E. J. Larsen, R. A. Renkin, and D. W. Smith. tiger shark predation risk influence bottlenose dolphin 2001. Trophic cascades among wolves, elk and aspen on habitat use. Ecology 83:480–491. Yellowstone National Park’s northern range. Biological Heithaus, M. R., and L. M. Dill. 2006. Does tiger shark Conservation 102:227–234. predation risk influence foraging habitat use by bottlenose Russ, G. R., and A. C. Alcala. 1998. Natural fishing dolphins at multiple spatial scales? Oikos 114:257–264. experiments in marine reserves 1983–1993: community and Heithaus, M. R., A. Frid, A. J. Wirsing, L. M. Dill, J. W. trophic responses. Coral Reefs 17:383–397. Fourqurean, D. Burkholder, J. Thomson, and L. Bejdero. Sandin, S. A., et al. 2008. Baselines and degradation of coral 2007. State-dependent risk-taking by green sea turtles reefs in the Northern Line Islands. PLoS ONE 3:e1548. mediates top-down effects of tiger shark intimidation in a Schmitz, O. J., V. Krivan, and O. Ovadia. 2004. Trophic marine ecosystem. Journal of Animal Ecology 76:837–844. cascades: the primacy of trait-mediated indirect interactions. Heithaus, M. R., A. Frid, A. J. Wirsing, and B. Worm. 2008. Ecology Letters 7:153–163. Predicting ecological consequences of marine top predator Stallings, C. D. 2008. Indirect effects of an exploited predator declines. Trends in Ecology and Evolution 23:202–210. on recruitment of coral-reef fishes. Ecology 89:2090–2095. Helfman, G. S. 1989. Threat-sensitive predator avoidance in Stevenson, C., L. S. Katz, F. Micheli, B. Block, K. W. Heiman, damselfish–trumpetfish interactions. Behavioral Ecology and C. Perle, K. Weng, R. Dunbar, and J. Witting. 2007. High Sociobiology 24:47–58. apex predator biomass on remote Pacific islands. Coral Reefs Holbrook, S. J., and R. J. Schmitt. 1988. The combined effects 26:47–51. of predation risk and food reward on patch selection. Strong, D. R. 1992. Are trophic cascades all wet? Differenti- Ecology 69:125–134. ation and donor-control in speciose ecosystems. Ecology 73: Jackson, J. B. C., et al. 2001. Historical overfishing and the 747–754. recent collapse of coastal ecosystems. Science 293:629–638. Walters, C., and J. F. Kitchell. 2001. Cultivation/depensation Lieske, E., and R. Myers. 1994. Coral reef fishes: Indo-Pacific effects on juvenile survival and recruitment: implications for December 2010 INDIRECT BEHAVIORAL EFFECTS OF FISHING 3571

the theory of fishing. Canadian Journal of Fisheries and mediated indirect interactions in ecological communities. Aquatic Sciences 58:39–50. Ecology 84:1083–1100. Werner, E. E., J. F. Gilliam, D. J. Hall, and G. G. Mittelbach. Wirsing, A. J., M. R. Heithaus, and L. M. Dill. 2007. Fear 1983. An experimental test of the effects of predation risk on factor: do dugongs (Dugong dugon) trade food for safety habitat use in fish. Ecology 64:1540–1548. from tiger sharks (Galeocerdo cuvier)? Oecologia 153:1031– Werner, E. E., and S. D. Peacor. 2003. A review of trait- 1040.

APPENDIX A The study system (Ecological Archives E091-250-A1).

APPENDIX B Location map of atolls and sites within atolls used in this study (Ecological Archives E091-250-A2).

APPENDIX C Study species and behavioral metrics (Ecological Archives E091-250-A3).

APPENDIX D Alternative predation risk metric (Ecological Archives E091-250-A4).

APPENDIX E Statistical analyses (Ecological Archives E091-250-A5).

APPENDIX F Human population density and piscivorous fish biomass at the atolls used in this study (Ecological Archives E091-250-A6).

APPENDIX G Prey excursion size in relation to alternative metric of chronic predation risk (Ecological Archives E091-250-A7).

APPENDIX H Prey excursion size in relation to alternative metric of acute predation risk (Ecological Archives E091-250-A8).

APPENDIX I Alternative explanations for patterns observed (Ecological Archives E091-250-A9).

APPENDIX J Predator size probability distributions for each study atoll (Ecological Archives E091-250-A10).

APPENDIX K Ranked relative abundances of observed piscivore species for each study atoll (Ecological Archives E091-250-A11).