Polar Bioi (2011) 34: 1065-1084 DOl 1O.1007/s00300-011-0967-4
Projected status of the Pacific walrus (Odobenus rosmarus divergens) in the twenty-first century
Chadwick V. Jay' Bruce G. Marcot • David C. Douglas
Received: 2 September 20101 Revised: 24 January 20111 Accepted: 2 February 20111 Published online: 2 March 2011 © US Government 2011
Abstract Extensive andrapid losses of sea ice in the Arctic the end of the century. The summed probabilities of vulner have raised conservation concerns for the Pacific walrus able, rare, and extirpated (p(v,r,e)) increased from a current (Odobenus rosmarus divergens), a large pinniped inhabit level of 10% in 2004 to 22% by 2050 and 40% by 2095. ing arctic and subarctic continental shelf waters of the The degree of uncertainty in walrus outcomes increased Chukchi and Bering seas. We developed a Bayesian monotonically over future periods. In the model, sea ice network model to integrate potential effects of changing habitat (partiCularly for summer/fall) and harvest levels had environmental conditions and anthropogenic stressors on the greatest influenceon future population outcomes. Other the future status of the Pacific walrus population at four potential stressors had much smaller influences on walrus periods through the twenty-first century. The model frame outcomes, mostly because of uncertainty in their future work allowed for inclusion of various sources and levels of states and our current poor understanding of their mecha knowledge, and representation of structural and parameter nistic influenceon walrus abundance. uncertainties. Walrus outcome probabilities through the century reflected a clear trend of worsening conditions for Keywords Status· Walrus· Odobenus· the subspecies. From the current observation period to the Bayesian network· Sea ice end of century, the greatest change in walrus outcome prob abilities was a progressive decrease in the outcome state of robust and a concomitant increase in the outcome state of Introduction vulnerable. The probabilities of rare and extirpated states each progressively increased but remained <10% through Climate warming and reductions in sea ice habitats present a particularly difficultchallenge to the conservation of pola,r marine mammals. Arctic-wide generalizations on species Electronic supplementary material The online version of this specific effects of climate change are difficult to make article (doi: 1 0.1 007/s00300-0 11-0967 -4) contains supplementary because the sensitivity and response of marinemammals to material, which is available to authorized users. environmental changes in the Arctic are complex and can vary among and within species (Laidre et al. 2008). The C. V. Jay (�) US Geological Survey, Alaska Science Center, Pacific walrus (Odobenus rosmarus divergens) is a large 4210 University Drive, Anchorage, AK 99508, USA pinniped inhabiting arctic and subarctic continental shelf e-mail: [email protected] waters of the Chukchi and Bering seas. Concerns over the future of the Pacific walrus, especially under the stress of B. G. Marcot US Department of Agriculture, Forest Service, climate change, prompted the US Fish and Wildlife Service Pacific Northwest Research Station, Portland Forestry Sciences to initiate a status review in September, 2009, to determine Laboratory, 620 SW Main, Suite 400, Portland, OR 97205, USA whether listing the subspecies as threatened or endangered is warranted under the Endangered Species Act (US Fish D. C. Douglas US Geological Survey, Alaska Science Center, and Wildlife Service 2009). To aid this endeavor, and to 3100 National Park Road, Juneau, AK 99801, USA focus new research and monitoring efforts, we synthesized
� Springer 1066 Polar BioI (2011) 34:1065-1084
Fig. 1 Study area used for pro jecting Pacific walrus status in the twenty-first century
studies and understanding of the Pacific walrus into a In spring, most adult female and young walrusesfollow knowledge-based model to evaluate possible future influ the receding ice pack northward to summer in the Chukchi ences of anthropogenic stressors and changing environmen Sea, while many of the adult males move toward coastal tal conditions on the Pacificwalrus population. areas to summer in nearshore areas of the Bering Sea and The Pacificwalrus is currently distributed in Russian and northern coast of Chukotka. In autumn, female and young Alaskan waters, and ranges in the north from the eastern walruses migrate with the developing sea ice southward East Siberian Sea to the western Beaufort Sea, and in the into the Bering Sea, where they are joined by the males in south from eastern Kamchatka to Bristol Bay (Fay 1982) late autumn and winter (Fay 1982). 1). (Fig. In winter, the entire Pacific walrus population Walruses are very gregarious and occur in groups of up resides in the Bering Sea. Breeding occurs in January and to "-'500walruses (Fay 1985; Speckman et al. 2(10). Group February. Leks are formed where breeding males display sizes of hauled out walruses tend to be larger when they are and vocalize from water alongside groups of females on shore than on ice (Fay 1985). Seasonal dynamics of sea hauled out on ice, and copulation occurs in water (Fay et al. ice cover in the Chukchi and Bering seas allow walruses to 1984). Most calving occurs in April-June (15-16 month exploit a wide area of the continental shelf during the year. pregnancy), and mothers care for and nurse their newborn Adult male walruses have dorsal inflatable pharyngeal calves on the ice (Fay 1985). Little is known of ice prefer pouches, which allow them to sleep at the surface in open ences for breeding and calving activities; however, wal water for extended periods (Fay 1960) between foraging ruses require ice floes large enough to support their weight trips from land in summer (e.g. Jay et al. 2(01). (Fay 1982; Simpkins et al. 2003). Within its range, Pacific Reduced summer sea ice over the continental shelf in the walruses typically occur in areas of unconsolidated ice, Chukchi Sea in the past decade has resulted in increased open leads, and thin ice where they can create breathing use of land haul-outs by adult females and young during holes, and they avoid areas with very high concentrations ice-free periods (Jay and Fischbach 2008; Kavry et al. of thick ice, such as in the Chukchi Sea in winter (Bums 2(08). This was particularly evident in the summer and et al. 1980, 1981; Fay 1982). fall of 2007 and 2009. During these years, thousands of
� Springer Polar BioI (2011) 34:1065-1084 1067
walruses hauled out along the coast of northwesternAlaska, surveys to contain walruses were unable to be surveyed and tens of thousands of walruses hauled out along the (Speckman et al. 201 coast of northern Chukotka when ice disappeared from the Pacificwalruses forage on the seafloorof the continental shelf. These events led to the trampling and death of hun shelves (Fay and Bums 1 Jay et aL where ben- dreds of walruses in Alaska and thousands in Russia (calves thic production is high (Grebmeier et aL Bluhm and are p A. Koc!h,ev, ChukotTTh"'RO,pers. comm. 2009). An unusu snails, clams, marine birds, and seals (Fay 1982; Sheffield ally high number of walruses hauled out and high levels of and Grebmeier but their diet most frequently con mortality occurred on the shores of Wrangel IsI known human activities (incidental takes) are estimated at b; Lalande et al. 2007; Bluhm and Gradinger about 3 walruses per year (US Fish and Wildlife Service Such shifts could result in reduced abundance of benthic 2Ul Thus, the greatest humall-caused direct mortality of prey for walruses. walruses is from subsistence harvest. In addition to warming the climate, increased levels of Factors regulating walrus popUlation dynamics are atmospheric CO2 have led to increased CO2 loading in the poorly understood, largely because of limited information oceans (The Royal Society Carbon dioxide dis on walrus vital rates and how vital rates change with popu solved in sea water forms carbonic acid, which decreases lation status (Fay Chivers 1 Garlich-Miller et aL the amount of calcium carbonate available to ma.rineinver The primary non-human predators of walruses are tebrates to construct shells or exoskeletons, including those polar bears (Ursus maritimus) and killer whales (Orcinus of benthic prey for walruses such as clams and snails. Our orca), and younger walruses are most vulnerable (Fay current understanding of climate-induced ocean acidificat ion on biological systems is rudimentary, and its long-term Surveys conducted at 5-year intervals between 1975 and consequences on marine ecosystems remain speculative 1990 produced Pacific walrus abundance estimates ranging (Orr et al. 2005; Guinotte and Fabl"J from 201,039 to 234,020 walmses; however, the estimates Sea ice forms a physical barrier to anthropogenic activi have unknown biases and unknown or large variances, and ties such as shipping, resource development, and commer cannot be used for detecting trends in population size (Hills cial fishing.Future summer sea ice losses are likely to lead and Gilbert 1994; Gilbert US Fish and Wildlife Ser to increased ship traffic in the Chukchi and Bering seas, vice 201 A survey conducted in 2006 produced an unbi including traffic from the transport of goods and from ased estimate of 129,000 individuals within the surveyed petroleum exploration and development (Arctic Council area of the population's early spring range, but the estimate Of the potential pollution sources from shipping, had low precision (95% CI: 55,000 to 507,000 individuals). perhaps the most significant threat to Arctic ecosystems is The estimate is considered to be less than the total popula the release of oil through accidental or illegal discharge tion size because some areas that were known from past with immediate � Springer 1068 Polar BioI (2011) 34:1065-1084 Council 2009). The high level of uncertainty associated The main steps we followed in developing the BN model with the interactions between complex physical, economic, were to: (1) create an influence diagram (i.e. a causal web) social, and political environments leads to great difficultyin depicting relationships among key factors that may affect predicting levels of Arctic marine shipping activities in the walrus distribution and abundance, (2) develop an initial future (Arctic Council 2009). Regulations in the US Arctic BN model using the influencediagram and comments solic Fishery Management Plan currently prohibit commercial ited from walrus experts, (3) revise the model after expert fisheries in the Beaufort and Chukchi seas (National Oce peer review, and (4) test the behavior of the model, using anic and Atmospheric Administration 2009), but sea ice sensitivity analyses on model subsets, and reconcile any losses could lead to the expansion of fisheries into more unrealistic residual model behavior. northernareas in the future. Prior to developing the influence diagram, we indepen Assessing walrus population status in the future requires dently solicited viewpoints from two Russian and two US a method that can accommodate both the complexity of walrus experts on potential walrus population stressors. We environmental and anthropogenic stressors that may affect used their comments to substantiate key factors and interre the future distribution and abundance of walruses, and the lationships that were represented in the final influence dia dearth of data on walrus vital rates and demographic gram and subsequent BN model. There was substantial response to such stressors. Here, we developed a Bayesian agreement among the experts in the main factors that were network (BN) model to represent linkages between poten identifiedas important to walrus status in the future. tial stressors and walrus responses in a probabilistic frame To guide our development of the initial BN model from work to evaluate potential outcomes of the walrus the influence diagram, we solicited comments on walrus population through the twenty-first century. The method . ecology and potential walrus population stressors from allowed us to combine various forms of information and Alaska Native elders during in-person meetings in Savoonga, uncertainties, including empirical data, modeled sea ice a prominent Siberian Yupik walrus hunting community on projections, and expert judgment in lieu of missing data St. Lawrence Island, Alaska. Subsequently, we conducted (e.g. Choy et al. 2009). We used the BN model to assess peer-review interviews with two US walrus experts to solicit potential walrus population outcomes at one past, one near concerns or recommendations on the BN model structure present, and four future periods. including the represented variables, linkages among vari ables, underlying conditional probabilities, model perfor mance, and sources of data by which to establish values of Materials and methods the model inputs. We documented the content andour recon ciliation of each peer review, including subsequent revisions Our study area (Fig. 1) was bounded by the edge of the con we made to the BN model structure and probabilitytables. tinental shelf (lSO-m isobath) and the range of the Pacific walrus as delineated by Fay (1982). The Chukchi and Ber Overall model structure ing seas within the study area span 709,000 and 934,000 km2, respectively. In general, our BN model consisted of input, intermediate, and output nodes. Intermediate and output nodes are condi Initial model development tional upon their direct antecedents or "parent nodes;" input nodes are said to be "parentless." Other nodes that are, in ® We developed a BN model, using the software Netica turn, directly linked from a given node are referred to as (Norsys, Inc.) and following the guidelines in Marcot et al. "child nodes." Input nodes consisted of anthropogenic stress (2006), to depict potential effects of environmental condi ors and environmental conditions, including sea ice condi tions and anthropogenic stressors on the future status of the tions as influenced by climate change. Intermediate nodes Pacificwalrus population. A BN model is a set of variables, were calculations or summaries of environmental and stres referred to as "nodes", linked by probabilities (Jensen and sor effects and walrus response, expressed by conditional Nielsen 2007). Our BN model is structured to evaluate the probabilitiesof multiple influences.Output nodes culminated net effectsof multiple stressors at various periods. It is nei the collective effectof all conditions and stressors, expressed ther a popUlation viability analysis nor a time-dynamic as conditional probabilitiesof walrus population outcomes. demography model because each period is treated indepen We structured our BN model with three seasonal sub dently. Instead, the BN model is a synthesis of expert models representing summer/fall (July through November), knowledge, existing studies and sea ice projections by cli winter (December through March), and spring (April mate models, and as such provides a comprehensive and through June) (Fig. 2), because walruses build energy flexible framework that can be refined as new information reserves in summer and fall, breed in winter, and give birth becomes available. in spring. This enabled us to specificallyparameterize input � Springer Polar Biol (2011) 34: 1065-1084 1069 Season: Summer! Fall (July - November) I ChukdliSeaicacowi'{%} 1 I: .".,,,'"F��P ��otblli , i " l 8±7 ) I / \ \ \ \ \ \ \ ' \ \,\ \ \ \ I AJl.e;eKon atll.sr.daroce strestc� !cwstreSSOf'S &&as moderate.'yjow &ressors sufficient both sufficientone sae modsrats/yhigh st"eSSWS seas hignsiressors itl$uffi<:iert both 1.61 :0.84 2.3 ± 0.15 Al.1-eeason walrus ouk:ome �rn ��:=- \ ::erable ��;�,,_/_�"-'1 j i ext!:';lated 640*' ;i:-' -;-:j 2.28±12 I Fig. 2 Influence diagram of a Bayesian network model to project "abundance stressors", and the spring submode! included a sea ice Pacific walrus status in the twenty-first century. This is the submodel "birthing platform" node between "shelf ice availability" and "abun diagram for the summer/fall season; similar submodels pertained to dance stressors." Shown here is the model set to calculate outcome winter and spring seasons . The winter submodel included a sea ice probabilities for period 2095 as depicted in Fig . -' "breeding environment" node between "shelf ice availability" and values (such as sea ice conditions and harvests) by seasons 1999-2008 (hereafter referred as 1984 and 2004, respec that encompass different a.'1d important walrus life history tively), and four future periods of 2020-2029, 2045-2054, events. All seasonal submodels had similar nodes � Springer 1070 Polar Bioi (2011) 34:1065-1084 IntergovernmentalPanel on Climate Change (IPCC) green We derived sea ice projections for the four future periods house gas (GHG) forcing scenarios: AlB and A2 (IPCC (2025, 2050, 2075, and 2095) from combinations of two 2010). These two GHG scenarios represent alternative CMIP3 GCM ensembles andthe two GHG scenarios (AlB paths of resource development and resulting GHG emis and A2) (Douglas 2010). The first ensemble of GCMs was sions. comprised of 18 GCMs (he.(eafter referredas the GCM_18 Except where noted below for the sea ice and harvest set, Online Resource 1: Table 4) that had monthly twenty and incidental take variables, values of input nodes and first century sea ice projections derived under each of the conditional probabilities were assigned by author Jay two GHG scenarios available in the CMIP3 data archive (walrus specialist) after considering opinions and com (Meehl et al. 2007). We selected the second ensemble of ments from other walrus experts, formal model reviews, GCMs from the GCM_18 ensemble based on each model's reports, and the scientific literature (Online Resource 1: ability to simulate the extent and seasonality of sea ice dur Tables 1-3). Values of all inputs were estimated under each ing the past 30 years of satellite observations (hereafter combination of the three seasons, six periods, and two referred as the GCM_SD2 set) and was comprised of 11 GHG scenarios. Many of the intermediate and output con GCMs for the Bering Sea and 10 GCMs for the Chukchi ditional probabilities address walrus life history responses Sea (Online Resource 1: Table 4). that are poorly understood, and in these cases, we partially We averaged monthly values of ice extent from the accounted for these uncertainties by assigning moderate to observation periods and monthly projections of ice extent high levels of spread in the conditional probabilities. from each GCM model, within season and year, then aver aged the within-season averages within each decadal Input and intermediate nodes period. For each GCM, we summed the number of ice-free months within season and year then averaged the within Input nodes, parameterized for each season, included sea season sums within each decadal period. For each BN ice projections from the GCMs, indirect climate effects on model run of future periods, we assigned a probability to benthic prey production, anthropogenic stressors including each state within a sea ice input node that was equal to the benthic perturbations fromresource utilization, effectsfrom proportion of models that projected a sea ice value within ship and air traffic and human settlements, and human the state's interval. This allowed us to capture the variabil caused mortalities from subsistence harvest and incidental ity (i.e. uncertainty) of ice projections among GCMs mortalities from fishing, industry, and research activities directly into the BN model. (Fig. 2, Table 1, and Online Resource 1: Tables 1 and 2). We calculated average sea ice extent separately for the Sea ice input nodes for each seasonal submode1 included Chukchi and Bering seas in each season as a metric to three variables: (1) percentage area of the Chukchi Sea cov describe the potential range of seasonal walrusmovements ered in ice ("Chukchi Sea ice cover"), (2) percentage area and occupancy in the two seas ("suitable ice extent") 2, 1, of the Bering Sea covered in ice ("Bering Sea ice cover"), (Fig. Table and Online Resource 1: Table 1). There is and (3) number of ice-freemonths in the Chukchi and Ber considerable uncertaintyas to the lower threshold of sea ice ing seas collectively ("ice-free months"). In this way, we needed to sustain walruses offshore. From radio-tracking expressed both the spatial and temporal extent of sea ice studies, walruses have been observed using very sparse, across the two seas and among seasons. These variables remnant ice during summer in the Chukchi Sea (Jay and represented two important aspects of sea ice relative to its Fischbach 2008), and during these conditions, walruses availability to walruses for hauling out. One is the spatial may need to travel greater distances to reach favorable for extent of sea ice available to walruses on which to rest and aging areas. In contrast, walruses are not able to penetrate initiate foraging trips across their seasonal range. The sec and effectively utilize areas with very high ice concentra ond is the duration of time that no sea ice occurs over the tions (Fay 1982), such as current conditions in winter in the entire continental shelf (number of ice-free months), and Chukchi Sea where ice concentrations of >90% commonly hence, the duration of time walruses must use terrestrial occur (Douglas 2010). Although we recognize that total ice haul-outs. extent can be comprised of many ice concentrations, we Detailed descriptions and methods for deriving the sea used ice extent in the two seas as a proxy for the availability ice inputs are presented in Douglas (2010) and briefly of suitable ice habitat for walruses. We considered very described here. We derived sea ice data for the two obser high or very low ice extent to be less-suitable walrus habi vation periods (1984 and 2004) from satellite observations tat. We used suitable sea ice extent fromall three seasons to of monthly sea ice concentration obtained from the indicate the annual potential range of walrus movements National Snow and Ice Data Center (NSIDC) final data and occupancy. We also used suitable ice extent with "ice archives for 1979-2007 (Cavalieri et al. 1996) and from free months" to inform seasonal "shelf ice availability" in preliminary sea ice archives for 2008 (Meier et al. 2006). the model. 1;) Springer Polar Bioi (2011) 34: 1065-1084 1071 Table 1 Title ar:ddescription of input, intennediate,and output nodes used in a Bayesian network model of Pacinc walrus status (unless oLh.erwise indicated, the same states are used for all three seasons in the model; see Online Resources 1: Table 1 for more detailed node descriptions) Node title Node descriptior: States Input nodes Ice-free months Mean number of months within a 0.0 to 0.5, 0.5 to 2.0, 2.0 season wiLh. no sea ice to support to 3.5, 3.5 to 5.0 (Summer/Fall) walruses for hauling out over the 0.0 to 0.5, 0.5 to 2.0, 2.0 continental shelf cf the to 3.0 (Spring) Chukchi and Bering Seas 0.0 to 0.5, 0.5 to 2.0, 2.0 to 4.0 (Winter) Chukchi Sea ice cover Extent of sea ice in the Chukchi Sea, 90--100%, 70--90%, 30-70%, expressed as a percentage of the 10-30%.0-10% Chukchi Sea shelf within the study area Bering Sea ice cover Extent of sea ice in the Bering Sea, 90--100%, 70-90%, 30-70%, expressed as a percentage of the Bering 10-30%, 0--10% Sea shelf within the study area Climate change on benthos Cumulative impact of various factors Positive, neutral, negative related to climate change on the production of benthic prey. Reduced sea ice and ocean acidification are assumed to potentially have the greatest influence on ber:thicprey production Resource utilization Impact of benthic prey production from Positive, neutral, negative activities that can perturb the seafloor from extraction of natural resources, such as from commercial fishing and oil and gas development Ship and air traffic Amount of ship and air traffic from commercial Low, moderate, high shipping, tourism, and fishing, and oil and gas development HumaTlsettlements Density of humans along the coasts of Alaska and Russia Low, medium, high Subsistence harvest Number of walruses killed by Native subsister:cehunting Low, moderate, high, very high in Russia and Alaska Incider:tal takes Namber of wa!rJses kiiled from illegal Low, moderate, high, very high activities and incidentally from fishing, industry, and research activities in Russia and Alaska Intermediate nodes Suitable ice eXIent Potential range of walms movements Sufficient both seas. in the Chukchi and Bering sufficientone sea. Seas as a function of the nodes "Chukchi Sea ice insufficient both seas cover" and "Bering Sea ice cover:' Abundance stressors S tressors to the abundance of the Pacific Low stressors, moderately walms population as a function low stressors, moderately of the nodes "body condition" high st"essors)high stressors and "total mortality" (and "breeding envircnment" in the winter submode!, and "birthing platfmm"in t.�e spring submode!) Shelf ice availability Avaiiability of sea ice to walruses for hauling out during the Excellent, good, fair, poer season as a function of ti'1e nodes "ice-free months" and "suitable ice extent" Benthic prey abundance Abundance of benthic prey as a function of the nodes High, moderate. low "climate change on benthos", "resource utilization", ac,d "oil spills" Energy expenditure Energy expended by walruses on foraging and swiII1Il1ing as Low, medium, high a function of the nodes "shelf ice availability" and "benthic prey abundallce" Disease and parasites Incider:ce of disease and parasites in the walrus population as a Low, moderate, high fJnction of the node "shelf ice availability" � Springer 1072 Polar Bioi (2011) 34:1065-1084 Table 1 contiuned Node title Node description States Oil spills Regularity and severity of hydrocarbons Low, moderate, high released into the water as a function of the node "ship and air traffic" Body condition Amount of body reserves possessed by animals High, medium, low in the population, particularly in the form of fat and muscle, as a function of the nodes "energy expenditure", "disease and parasites", and "oil spills" Predation and associated mortality Number of walruses killed by predators Low, moderate, high (excluding humans), which are primarily polar bears and killer whales, as a function of the node "shelf ice availability" Haul-out disturbance Level ofdisturbances to hauled out walruses on ice, Low, moderate, high and particularly, on terrestrial haul-outs as a function of the nodes "ship and air traffic", "human settlements", and "human-caused direct mortality" Crowding Number of walruses at a haul-out as a function of Low, moderate, high the node "shelf ice availability" Crowding and disturbance Intensity of a disturbance on a haul-out as a function of Low, medium, high the nodes "crowding" and "haul-out disturbance" Human-caused direct mortality Total number of walruses directly killed by humans Low to moderate, high, very high in Russia and Alaska as a function of the nodes "subsistence harvest" and "incidental takes" Total mortality Total number of walruses killed as a function of Low, moderate, high the nodes "predation and associated mortality", "crowding and disturbance" and "human-caused direct mortality" Breeding environment Adequacy of ice habitat for breeding as a function of Superior, adequate, inferior (node in winter submodel only) the node "shelf ice availability" Birthing platform (node in Adequacy of ice habitat for birthing, nursing, Superior, adequate, inferior spring submodel only) and providing protection to newborncalves during severe storms as a function of the node "shelf ice availability" Output nodes All-season suitable ice extent Overall suitable ice extent throughout the year, Sufficient both seas, sufficient one sea, which reflects the potential range and occupancy insufficient both seas of walrus movements in the Chukchi and Bering seas, as a function of "suitable ice extent" in summer/fall, winter, and spring All-season abundance stressors Overall stressors on walrus abundance throughout Low stressors, moderately low the year as a function of "abundance stressors" stressors, moderately high in summer/fall, winter, and spring stressors, high stressors All-season walrus outcome Walrus population overall outcome as a function of Robust, persistent, vulnerable, the nodes "all-season suitable ice extent" rare, extirpated and "all-season abundance stressors" In the model, we linked lower shelf ice availability to Kavry et al. 2008; Kochnev et al. 2(08). In the model, we greater walrus predation (particularly from polar bears), also linked lower shelf ice availability to greater amounts of walrus crowding, and incidence of disease and parasites in energy walruses expend on foraging and on swimming to the population, by increasing the size and concentrating the preferred feeding grounds. This might be especially true locations of walrus haul-outs. High levels of crowding and when walruses (particularly females and young) areforced disturbances at terrestrial haul-outs (and perhaps large haul to use terrestrial haul�outs when ice is completely unavail outs on ice floes) can lead to mortalities (particularly of able over the shelf during summer and must swim to and juvenile walruses) from intraspecific trampling events (e.g. from offshore prey patches between resting periods on � Springer 1074 Polar BioI (2011) 34: 1065-1084 to locally adverse conditions of anthropogenic stressors stressor could be altered by improving (or worsening) its and environmental conditions resulting in a population level of stress. We qualitatively compared walrus popula with greatly reduced abundance and occupancy that is tion outcomes generated from the influenceruns to the nor more or less restricted to isolated pockets; mative model outcomes. • extirpated the Pacific walrus population is absent through all, or nearly all, of the Chukchi and Bering Sea region. Results Model outcomes, sensitivity analysis, and influenceruns Mode! outcomes We compared model outcomes, consisting of probabilities Only small differences in all-season walrus outcomes of the five walrus population response states, among each resulted from sea ice projections derived from combina combination of period, GCM set, and GHG scenario to tions of the GCM_l8 and GCM_SD2 GCM sets and the evaluate differencesand trends. These resuits were based on AlB and A2 GHG scenarios (Online Resource 1: Table 5). our modeled best estimates of walrus outcomes and are Walms outcomes were nearly the same with the somewhat referred to here as "normative" outcomes under each com narrower spread sea ice projections from the GCM_SD2 bination. The distribution of probabilities among outcome subset as with the spread of projectionsfrom the GCM_lS states indicates the degree of unce11ainty in projected out set of models. The high similarity in walrus outcomes comes. Total uncertainty would be represented by uniform between the two GCM sets was probably due to the high probabilities among an outcome states, total certainty by similarity in the central tendency of sea ice input values 100% probability in one state and 0% probability in all between the two sets (Douglas and from how sea ice other states, and intermediate levels of certainty by the dis values were binned into discrete ranges in the EN modeL tribution of probabilities among multiple states. Walrus outcomes under the A2 GHG scenario were also We conducted sensitivity analyses of the BN model to qualitatively similar to those under the AlE GHG scenario. determine the sensitivity of outcomes to the input nodes. Hereafter, for simplicity, we refer to outcomes from model Sensitivity was calculated in the modeling shell Netica® as nms using only the GCM_SD2 models and AlB GHG entropy reduction (reduction in the disorder or variation)of scenario. the walrus outcome node relative to the information repre VIalms outcomes were similar between the two observa sented in the input nodes (see Marcot et al. 2006 for method tion periods (1984 and 2004). From the current observation and equation). Sensitivity tests indicated how much of the period to the end of century, the greatest change in walrus variation in the selected node is explained by each of the outcome probabilities was a progressive decrease in the other nodes considered and is conducted by setting ali nodes outcome state of robust and a concomitant increase in the to their default conditions. In our model, we set all prior outcome state of vulnerableo The probabilities of rare and probabilities of the input nodes for the sensitivity analyses to extirpated states each progressively increased at a similar uniform distributions, reflecting total uncertainty among rate, but remained Table 2 Groups of input nodes and their forced state used in influence runs of a Bayesian network model of Pacific walrus status Influence runs Input nodes and their forced state in each run (A) Influence of overall anthropogenic stressors Compare results on the response variables from influence runs IA1 and IA2, with those from the "normative" run. The differencewill be the potentially mitigating or adverse influence of maximal protection guidelines on anthropogenic stressors (other than climate change) (IAI) Influence of minimal anthropogenic stressors RUsummer/fall, RUwinter, and low harvest, under projected climate change effects RU spring --* positive SATsummer/fall, SA Twinter, SATspring --* low HUMsummer/fall, HUMwinter, HUMspring --* low HVSTsummer/fal1, HVSTwinter, HVSTspring --* low (IA2)Influence of maximal anthropogenic stressors RUsummer/fall, RUwinter, and very high harvest, under projected climate change effects RUspring --* negative SATsummer/fall, SATspring --* high SATwinter --* moderate HUMsummer/fall, HUMwinter, HUMspring --* high HVSTsummer/fal1, HVSTspring --* very high HVSTwinter --* low (B) Influence of harvest When compared with influence runs IA1 and IA2 above, influence runs m1-m3 would reveal the incremental positive influence of regUlating harvest only (and not resource utilization, ship and airtraffic, and human settlements) (m1) Influence of low harvest HVSTsummer/fall, HVSTwinter, HVSTspring --* low (m2) Influence of high harvest HVSTsummer/fal1, HVSTspring --* high HVSTwinter --* low (m3) Influence of very high harvest HVSTsummer/fal1, HVSTspring --* very high HVSTwinter --* low (C) Influence of ship and air traffic (IC1) Influence of minimal ship and airtraffic SATsummer/fall, SATwinter, SATspring --* low (IC2) Influence of maximal ship and air traffic SATsummer/fal1, SATspring --* high SATwinter --* moderate (D) Influence of climate change on benthos (ID1) Influence of positive effects of CCBsummer/fall, CCBwinter, climate change on benthos CCBspring --* positive (ID2)Influence of negative effects CCBsummer/fal1, CCBwinter, of climate change on benthos CCBspring --* negative � Springer 1076 Polar Bioi (2011) 34:1065-1084 Table 2 continued Influence runs Input nodes and their forced state in each run (E) Influence of sea ice Period 2025 2050 2075 2095 (lEI) Influence IceMsummer/fall-+ 0 to 0.5 IceMsummer/fall -+ 0.5 to 2 IceMsummer/fall -+ 0.5 to 2 IceMsummer/fall-+ 0.5 to 2 of maximal sea ice IceMwinter -+ 0 to 0.5 IceMwinter -+ 0 to 0.5 IceMwinter -+ 0 to 0.5 IceMwinter-+ 0 to 0.5 (the assigned state for IceMspring -+ 0 to 0.5 IceMspring -+ 0 to 0.5 IceMspring -+ 0 to 0.5 IceMspring -+ 0 to 0.5 each input variable IceCsummer/fall -+ 30 to 70 IceCsummer/fall-+ 30 to 70 IceCsummer/fall -+ 30 to 70 IceCsummer/fall -+ 10 to 30 was based on the IceCwinter-+ 90 to 100 IceCwinter-+ 90 to 100 IceCwinter -+ 90 to 100 IceCwinter -+ 90 to 100 value from the IceCspring-+ 90 to 100 IceCspring -+ 90 to 100 IceCspring -+ 90 to 100 IceCspring-+ 90 to 100 GCM_SD2 AIB IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 model that projected IceBwinter-+ 70 to 90 IceBwinter-+ 30 to 70 IceBwinter -+ 30 to 70 IceBwinter -+ 30 to 70 the highest sea ice IceBspring -+ 30 to 70 IceBspring -+ 30 to 70 IceBspring -+ 30 to 70 IceBspring-+ 30 to 70 condition for that variable in the indicated season and period) (1E2)Influence of IceMsummer/fall-+ 2 to 3.5 IceMsummer/fall-+ 3.5 to 5 IceMsummer/fall-+ 3.5 to 5 IceMsummer/fall-+ 3.5 to 5 minimal sea ice IceMwinter-+ 0 to 0.5 IceMwinter -+ 0 to 0.5 IceMwinter -+ 0 to 0.5 IceMwinter -+ 0 to 0.5 (the assigned state for IceMspring-+ 0 to 0.5 IceMspring -+ 0 to 0.5 IceMspring-+ 0 to 0.5 IceMspring -+ 0 to 0.5 each input variable IceCsummer/fall-+ 10 to 30 IceCsummer/fall-+ 0 to 10 IceCsummer/fall-+ 0 to 10 IceCsummer/fall-+ 0 to 10 was based on the IceCwinter-+ 90 to 100 IceCwinter-+ 70 to 90 IceCwinter -+ 70 to 90 ICeCwinter -+ 30 to 70 value from the IceCspring -+ 90 to 100 IceCspring -+ 70 to 90 IceCspring -+ 70 to 90 IceCspring-+ 70 to 90 GCM_SD2 AIB IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 IceBsummer/fall-+ 0 to 10 model that projected IceBwinter-+ 30 to 70 IceBwinter -+ 10 to 30 IceBwinter -+ 0 to 10 IceBwinter -+ 0 to 10 the lowest sea ice IceBspring -+ 10 to 30 IceBspring -+ 0 to 10 IceBspring -+ 0 to 10 IceBspring -+ 0 to 10 condition for that variable in the indicated season and period) RU Resource utilization, SAT Ship and airtraffic, HU Human settlements, HVST Subsistence harvest, CCB Climate change on benthos, IceM Ice freemonths, IeeC Chukchi Sea ice cover, IeeB Bering Sea ice cover, summerIJallsummer/fall submodel node, winter winter submodel node, spring spring submodel node 100 In our BN model, changes in walrus outcomes (Fig. 3) 90 � reflected changes in the parent nodes: all-season suitable .. 80 e ice extent and all-season abundance stressors (Fig. 2). All 0 JO season suitable ice extent was derived from suitable ice 50 60 extent in each of the three seasons. Suitable ice extent grew .I'" so i: increasingly insufficientin both seas for walrus movements 'C 40 and occupancy through end of century in all three seasons, 30 ...� '" but most sharply in summer/fall (Fig. 4). In summer/fall, ... 20 a.. suitable ice extent was determined from sea ice extent in 2 10 0 the Chukchi Sea. In winter and spring, probabilities of 1S1l4 2004 202S 21l5O 2075 2095 insufficientice extent in both seas occurred when there was Obs.""'" Projected some probability of very high ice extent in the Chukchi Timepericd Sea, a condition making it difficultfor walruses to occupy Fig. 3 Probabilities of walrus outcomes projected from a Bayesian the area, together with some probability of low ice extent in network model of Pacific walrus status for 6 periods and using sea ice the Bering Sea. Projected sea ice extent in the Chukchi Sea projections from the GCM_SD2 set and AlB GHG emission scenario never reached low levels in winter or spring through the ("normative" run) end of century. All-season abundance stressors were influenced by vari low even at the end of the century; hence, uncertainty ables linked to shelf ice availability, which was determined among the model outcomes was spread primarily across the by both suitable ice extent and number of ice-free months robust, persistent, and vulnerable states. (Fig. 2). In a trend similar to that of suitable ice extent, � Springer Polar Bioi (2011) 34:1065-1084 1077 Fig. 4 Probabilities of "suitable A 100% ice extent" (a) and "abundance 90% stressors" (b) by future periods BO% > 70% and seasons from a Bayesian .t! network model of Pacific walrus 60% lim .c status using sea ice projections 50% from the GCM_SD2 set and ...f 40% AlB GHG emission scenario 30% 20% 10% 0% 2025 2050 2075 2095 2025 2050 2075 2095 2025 2050 2075 2095 Summer/fall Winter Spring ! • Insufficient both seas ," Sufficient one sea III Sufficientboth seas I 8100% 90% BO% � 70% 60% li ! SO% ...f 40% 30% 20% 10% 0% 2025 2050 2075 2095 2025 2050 2075 2095 2025 2050 2075 2095 Summer/fall Winter Spring I IIIHigh stressors III Mod high stressors *" Mod low stressors III low stressor. I stressors on walrus abundance increased through end of cen walrus outcomes to those from the normative model run, tury in all seasons, but most notably in summer/fall (Fig. 4). suggested that sea ice stressors had substantial influence Abundance stressors increased only slightly in winter. on walrus outcomes (Fig. 5, Table 2). Setting sea ice habi tat (number of ice-free months and ice extent) to GCM Sensitivity analyses projected maximal sea ice conditions at each future period, while leaving all other input nodes unchanged at their nor Sensitivity analyses suggested that the inherent probability mative values, resulted in projected P(v,r,e) values 6-24% structure (i.e. not specified for any particular scenario or below the normative values through end of century. In period) of the BN model renders the all-season walrus out contrast, setting sea ice habitat to GCM-projected minimal come to generally being more sensitive to sea ice and har sea ice conditions resulted in projected P(v,r,e) values 1- vest levels than to other factors (Table 3). Walrus outcome 19% above the normative values through end of century. was somewhat more sensitive to ice-free months in winter The slight decrease in P(v,r,e) in 2095 under the minimal and spring than in summer/fall. This is likely due to the sea ice influence run was due to improved sea ice condi added "breeding environment" and "birthing platform" tions in the Chukchi Sea from decreased sea ice cover in nodes in the winter and spring submodels, respectively, winter. which affected abundance stressors and provided an addi The influence runs suggested that "climate change on tional path for ice-free months to influencethe outcomes. A benthos" and "ship and air traffic" had negligible influence full analysis of sensitivity of the outcomes to every node in on walrus outcomes (Fig. 5, Table 2). That is, minimizing the BN model (Online Resource 1: Table 6) suggested that and maximizing the values of these stressors resulted in factors in each season played a role in determining walrus very little differencefrom the normative values of P(v,r,e). outcomes, with a greater overall sensitivity to all-season Minimizing harvest had little positive influenceon wal abundance stressors than to all-season suitable ice extent. rus outcomes compared to the normative P(v,r,e) values (Fig. 5), but high and very high harvest had progressively Influenceruns higher adverse influenceon walrus outcomes. High harvest led to an increase in P(v,r,e) of 13-17% over normative val Influence runs of the BN model, in which we varied ues through end of century. Very high harvest resulted in an selected stressors to extreme values and compared the increase in P(v,r,e) of 3�35% over normative values. � Springer 1078 Polar Bioi (2011) 34:1065-1084 Table 3 Results of sensitivity analysis results of the Bayesian network model, sorted by season. Entropy reduction refers to the degree to which the finalall-season walrus outcome probabilities are sensitive to each input node in the model, listed here in decreasing order of effect SummerlFall Winter Spring Node name Entropy Node name Entropy Node name Entropy reduction reduction reduction Incidental takes 0.00856 Ice-freemonths 0.01417 Ice-free months 0.01832 Subsistence harvest 0.00856 Incidental takes 0.00574 Subsistence harvest 0.00426 Bering Sea ice cover 0.00567 Subsistence harvest 0.00574 Incidental takes 0.00426 Ice-free months 0.00382 Bering Sea ice cover 0.00516 Bering Sea ice cover 0.00381 Chukchi Sea ice cover 0.00287 Chukchi Sea ice cover 0.00256 Chukchi Sea ice cover 0.00187 Ship and airtraffic 0.00127 Ship and air traffic 0.00101 Ship and air traffic 0.00085 Human settlements 0.00009 Human settlements 0.00006 Human settlements 0.00004 Climate change on 0 Climate change on benthos 0 Climate change on benthos 0 benthos Resource utilization 0 Resource utilization 0 Resource utilization 0 Similar to the harvest influenceruns, minimizing overall Sensitivity analyses and influence runs indicated that anthropogenic stressors (which included harvest, Table 2) sea ice habitat and harvest could have the greatest influ had little positive influence on walrus outcomes compared ence on futurewalrus outcomes. The progressive increase to the normative walrus outcomes, but maximizing anthro over time in probabilities of vulnerable, rare, and extirpa pogenic stressors imparted a high degree of adverse influ tion suggests that perhaps there will be no thresholds that ence. Maximizing overall anthropogenic stressors resulted trigger accelerated population change, given that future in an increase in P(v,r,e) of 35-45% over normative values sea ice change adheres to the GCM projections and that through end of century. future harvest rates are proportional to walrus population In summary, the influence runs suggest that changes in size. sea ice habitat and harvest could have the greatest influence Sea ice projections indicate that ice-free conditions over on future walrus outcomes. In the normative run, harvest the entire shelf will occur during August, September, and was set to a constant state through the end of century (mod October (during summer/fall) by the end of the century erate for spring and summer/fall and low for winter); there and will be accompanied by an earlier sea ice melt in spring fore, most of the increase in P(v,r,e) through end of century and delayed sea ice freeze-up in late fall (Douglas 2(10). in the normative run was due to projected declines in sea We assumed that the geographic location of breeding and ice habitat. Since the harvest influenceruns were identical birthing canshift without consequence to the population, but to the normative run but with varying levels of harvest set if this assumption proves false, then adverse influences on tings, the influence of harvest could be largely additive to walruses may be greater than indicated by our BN model. the influence of sea ice on future walrus outcomes. Of the In our model, sea ice habitat decreased most in summer/ anthropogenic stressors that might be mitigated (other than fall (Fig. 4) due to decreased sea ice extent in the Chukchi climate change), harvest had the greatest influenceon wal Sea and increased number of ice-free months over the con rus outcomes. tinental shelf. The decreased sea ice habitat was linked to decreased body condition from poorer ice availability and increased total mortality from increased crowding and dis Discussion turbance mortalities on the haul-outs. This, together with the moderate setting of harvest we specifiedthrough end of Dominant effectsof sea ice and harvest century for summer/fall, resulted in the summer/fall season having the greatest influence on all-season walrus out The projected decreases in the walrus outcome states of comes. Abundance stressors projected for spring were less robust and persistent fromthe current observation period to than those projected for summer/fall because sea ice condi the end of century (and concomitant increases in the out tions in spring are not projected to degrade as greatly as in come states of vulnerable, rare, and extirpation) reflected a summer/fall. Sea ice and other conditions through end of trend of worsening conditions for the Pacific walrus. The century in winter (including low harvest level settings) had trends in outcomes were accompanied by increasing uncer only small adverse influences on all-season walrus out tainty among the robust, persistent, and vulnerable states. comes. � Springer Polar BioI (201l) 34:1065-1084 1079 Fig. Summed probabilities of 5 - Climate on benthos Ship and air vulnerable, rare, and extirpated "C 0.8 0.8 derived from "influenceruns" of Jg tV a Bayesian network model of ...c. 0 .7 0.7 Pacific walrus status . Influence � 0.6 0.6 runs consisted of individually 1I>... varying specified input condi- 0 0.5 0.5 tions (e.g. sea ice, harvest, and ...tD 0.4 other human activities) while tV..... OA holding other input conditions 0.3 0.3 constant at their normative lev- .Q.!!! 0.2 els for that period, and using sea ....tV 0.2 ice projections from the Q) c 0.1 0.1 GCM_SD2 set and AlB GHG ::! emission scenario. Harvest influ- 0.0 .2:- 0.0 ence runs included two levels of 0- 2 2050 2025 2050 2075 20 5 2075 2095 2095 increased input stressors (see Table 2) Sea ice Harvest "C 0.8 Q) 0.8 iii 0.7 ...c. 0.7 � 0.6 1I> 0.6 ... 0.5 0 0.5 ...tD ...tV OA OA JJi 0.3 0.3 .Q IY 0.2 � 0.2 Q) 0.1 0.1 .5::l > ...... 0.0 0.0 0- 2025 2050 2025 2050 2075 2095 2075 2095 .- Overall anthropogenic "C 0.8 �tV ....c. 0.7 � n:- 0.0 2025 2050 2075 2095 Time Period We incorporated uncertainty of modeled sea ice projec than indicated by our model's normative run. Our minimum tions into our BN model runs by having each GCM model sea ice influencerun used sea ice input values that represent provide an observation for each run under each combination a composite of the most extreme minimal sea ice projections of period, GeM set, and GHG scenario. However, summer (Table 2). For the early twenty-first century, these projec Arctic sea ice extent could decline more rapidly than fore tions portrayed a trajectory of ice loss that was actually casted by many GeMs (Stroeve et al. 2007; Wang and faster than observed, and for subsequent decades, it would Overland 2(09). If sea ice continues to decline more rapidly not be unreasonable to consider the results of the minimum in the future than is projected by most GCMs, changes in the sea ice influence run (Fig. 5) as exemplifying a worst-case status of the walrus population could occur more rapidly trajectory of ice loss under the AlB forcing scenario. � Springer 1080 PolarBioi (2011) 34:1065-1084 Results of the influenceruns suggest that although mini uncertain (Grebmeier et al. 2010). Earlier sea ice melt and mizing harvest from current levels may have little positive reductions in sea ice extent have the potential to reduce effecton population outcomes in the future, harvests of high benthic production and increase pelagic consumption in and very high levels (above the moderate harvest levels set Arctic marine ecosystems and thus result in decreased food in the normative conditions) could add significantly to the supply to walrus benthic prey; however, dependencies on adverse effectsof future sea ice conditions. If in the future, regional conditions make detailed biological responses the walrus population declines, but a constant number of difficult to predict (Piepenburg 2005; Grebmeier et al. walruses continue to be harvested, the level of stress by the 2006a, b; Lalande et al. 2007; Bluhm and Gradinger 2(08). harvest would effectively increase; as exemplified in our In addition to the potentially large effects that reduced high or very high harvest influence runs (Fig. 5). Also, sea ice may have on lowering production of walrus prey, increased access to walruses on terrestrial haul-out sites in increasing levels of dissolved CO2 in the world's oceans Alaska and Russia during progressively longer ice-free and associated increase in ocean acidification could ulti periods in summer/fall could result in higher harvest levels mately have profound adverse effects on shell-producing and disturbances, leading to higher mortality levels. The net organisms (Bates et al. 2(09). Since walruses feed largely effect of harvest on walrus outcomes in the future will on clams, this phenomenon could have larger effectson the likely depend on the level of stress imparted by future sea walrus outcomes than we projected. Forecasts are uncertain ice change and the ability of walruses to cope with those for saturation levels of carbonate minerals in the Chukchi changes. The potential influence of very high harvest on Sea and broader Arctic Ocean because of challenges in pre walrus outcome in our model is consistent with the dicting future changes in sea ice cover, temperature, stratifi observed response in walrus distribution and abundance cation and nutrient supply, inputs of freshwater and afterhigh levels of harvestin past centuries. Very high lev terrestrial organic carbon, and complex physical and bio els of commercial harvests of Pacific walruses in late nine logical feedbacks in the region (Bates et al. 2009). teenth through mid-twentieth century led to depleted population levels and dramatic reductions in population Paleobiogeography and the potential of walruses to adapt range (Fay et al. 1989; Garlich-Miller et al. 2006). to future stressors Implications of uncertainties of the marine ecosystem Biogeographical studies can provide important contextual information to management and conservation issues and are The negligible influence that "ship and air traffic" and "cli particularly relevant in the Arctic where rapid environmental mate change on benthos" had on walrus outcomes in the change is projected to occur (Murray 2(08). The extant BN model was likely due to the low influencethese factors Pacificand Atlantic walrus (0. r. divergens and 0. r. rosma imparted on body condition, combined with the relatively rus), and at least 20 fossil species within 14 genera and three low influence of body condition on seasonal abundance subfamilies, are members of the monophyletic Odobenidae. stressors. The probabilities we assigned to these relation Phylogenetic and stratigraphic data suggest that odobenids ships in the model (Online Resource 1: Tables 2 and 3) refl first evolved in the North Pacific sometime before 18 Ma ected our current poor understanding and uncertainty of (late early Miocene) (Kohno et al. 1995; Demere et al. their future conditions and the mechanisms and processes 2003). These small-bodied archaic walruses gave rise during by which they will affect walrus prey abundance and body the middle and late Miocene to larger-bodied, but non condition and their ultimate ramifications on walrus vital tusked lineages of imagotariine and dusignathine walruses rates and abundance. However, the relatively low influence (Demere et al. 2003; Amason et al. 2(06). More recent these factors had on walrus outcomes in our model does not records of walrus fossils from Japan and California suggest reflect certain knowledge that these factors are trivial. that the lineage containing the modem walrus evolved in the Rather, their low influence reflects our uncertainty in the North Pacific near the Miocene-Pliocene boundary about magnitude and timing in which the factors might be 5-7 Ma. An early, large-bodied, tusked member of this expressed and the degree of stress that the factors could odobenine lineage dispersed northward from the North impart on the Pacificwalrus population. Pacific during a pre-glacial event of the early Pliocene While greater.areas of open water and a longer growing through the Bering Strait perhaps 4-5 Ma, into the ice-free season from changes in sea ice conditions in the future Arctic Ocean, and eventually into the North Atlantic in an could lead to increased primary production in Arctic waters east-to-west direction (Kohno et al. 1995; Demere et al. (Arrigo et al. 2008), the biological processes that govern 2(03). These Pliocene North Atlantic walruses went extinct pelagic and benthic productivity at regional scales are com without any descendants (Kohno and Ray 20(8). plex, and therefore, the effect of future sea ice losses on It is hypothesized that Odobenus evolved in the North ecosystem structure in the Chukchi and Bering seas is Pacific during the early Pleistocene and apparently moved � Springer PolarBioi (2011) 34:1065-1084 1081 northward during interglacials and southward during gla sources of data and expert knowledge, including the paucity cials (Hili'ingtonand Beard however, they were peri of data on waL.rus vital rates and demographic response to odically isolated in the North Pacificfrom the Arctic Ocean stressors, and to represent structural and parameter uncer and North Atlantic from the cyclical closing of the Bering tainties explicitly through probabilities of influence. Such Strait from glacio-eustatic oscillations (Davies 1958; Dem an approach is an appropriate means of evaluating potential ere et al. Since the start of the current ice age about future cumulative effects of multiple stressors, p Demere et al. The evolution of the extant O. rosma- within the next few years from the IPCC 5th assessment rus Pacific and Atlantic subspecies sometime during the report and can be incorporated into the BN model. Pleistocene probably arose from the splitting of the species' The energetic costs associated with modified walrus former Holarctic range from extensive sea ice in the Cana behaviors in response to environmental changes are pres dian Arctic during an early glacial phase (Kohno et al. ently unYJlown. New research to address this information 1995; Demere et al. Thus, the extant Pacific and gap, such as studies of walrus bioenergetics relative to sea Atlantic walruses have apparently been exposed to repeated ice availability and distribution of benthic prey, could be opening and closure of the Bering Strait and glacial cycles coupled with models of walrus population dynamics to bet (Augustin et aL with potentially ice-free marginal ter understili,d and quantify the linkages between energy seas during interglacial periods (Davies since the expenditure, body condition, (Fay 1 in the absence of sea ice, and the wide taxo- nomic range of walrus prey (Fay ! suggests plasticity We used a EN model to represent linkages between poten in the walrus diet. These characteristics may help walruses tial stressors and walrus responses in a probabilistic frame withstand some of the future sea ice-related stressors they work to evaluate potential outcomes of the walrus may encounter in the face of rapid environmental change. population through the twenty-first century. Model out However, the degree to 'vVmch their potential ecological comes reflected a clear trend of worsening conditions for flexibility would also provide a buffer against additional the Pacific walms. From the current observation period to anthropogenic-related stressors in the future cannot be the end of century, the greatest change in walrus outcome gauged by the historic or prehistoric record alone. probabilities was a progressive decrease in the outcome state of robust and a concomitant increase in the outcome Next steps in continuing research and model refinement state of vulnerable. The probabilities of rare and extirpated states each progressively increased but remained <10% Climate Chili'l.ge and the rapid loss of sea ice raise concerns through the end of the century. The summed probabilities for conservation of the Pacific walrus and other arctic spe of vulnerable, rare, and extirpated (P(v,r,e)) increased from cies (e.g. Amstrup et al. Forecasting the response of a currentlevel of 10% in 2004 to 22% by 2050 and 40% by arctic species to their complex and rapidly changing envi 2095. The degree of uncertainty in walrus outcomes ronment requires consideration of a number of potential increased monotonically over future periods. stressors, often with very limited information. This study Sensitivity analyses and influence runs from our BN was a first step in developing a comprehensive framework model indicated that sea ice habitat and harvest could have to integrate the V � Springer 1082 Polar BioI (2011) 34:1065-1084 Results of the influence runs suggested that, of the References anthropogenic stressors that might be mitigated (other than climate change), harvest had the greatest influence on wal Adam PJ, Berta A (2002) Evolution of prey capture strategies and diet in rus outcomes. High and very high harvest levels could add the Pinnipedimorpha (Mammalia, Camivora) . Oryctos 4:83-107 Amsttup SC, Marcot BG, Douglas DC (2008) A Bayesian network significantly to the projected adverse effects of future sea modeling approach to forecasting the 21st century worldwide sta ice conditions. Minimizing harvest fromcurrent levels may tus of polar bears . In: DeWeaver ET, Bitz CM, Tremblay L-B have minor positive effect on population outcomes in the (eds) Arctic sea ice decline: observations, projections, mecha future; however, improvements to multiple environmental nisms, and implications, vol 180. Geophysical Monograph Series, Washington DC, pp 213-268 factors and anthropogenic stressors together could provide Arctic Council (2009) Arctic marine shipping assessment 2009 report greater benefit. Amason U, Gullberg A, Janke A, Kullberg M, Lehman N, Petrov EA, The negligible influence that "ship and air traffic" and Vainola R (2006) Pinniped phylogeny and a new hypothesis for "climate change on benthos" had on walrus outcomes their origin and dispersal . Mol Phylogenet EvoI 41:345-354 Arrigo KR, van Dijken G, Pabi S (2008) Impact of a shrinking Arctic mostly reflectsuncertainty in future states of these variables ice cover on marine primary production. Geophys Res Lett 35: 1- and our current poor understanding of the processes and 6. doi: 1O.!02912008GL035028 mechanisms by which changes in ship traffic, reduced sea Augustin L, Barbante C, Barnes PRF, Barnola JM, Bigler M, Castel ice, and atmospheric carbon loading in the oceans may lano E, Cattani 0, Chappellaz J, Dahl-Jensen D, Delmonte B, Dreyfus G, Durand G, Falourd S, Fischer H, Fliickiger J, Hansson affect walrus prey abundance and its ultimate ramifications ME, Huybrechts P, Jugie G, Johnsen SJ, Jouzel J, Kaufmann P, on walrus vital rates and abundance. The relatively low Kipfstuhl J, Lambert F, Lipenkov VY, Littot GC, Longinelli A, influence that these factors had on walrus outcome in our Lorrain R, Maggi V, Masson-Delmotte V, Miller H, Mulvaney R, model does not reflectcertain knowledge that these factors Oerlemans J, Oerter H, Orombelli G, Parrenin F, Peel DA, Petit J are trivial. Rather, their low influence reflects our uncer R, Raynaud D, Ritz C, Ruth U, Schwander J, Siegenthaler U, Sou chez R, Stauffer B, Steffensen JP, Stenni B, Stocker TF,Tabacco tainty in the magnitude and timing in which the factors IE,Udisti R, Wal RSW, Broeke M, Weiss J, Wilhelms F, Winther might be expressed and the degree of stress that the factors J- G, Wolff EW, Zucchelli M (2004) Eight glacial cycles from an could impart on the Pacificwalrus population. Antarctic ice core . Nature 429:623-628 Bates NR, Mathis JT, Cooper LW (2009) Ocean acidification and bio This study is a first step in developing a comprehensive logically induced seasonality of carbonate mineral saturation framework to incorporate the various complex linkages of states in the western Arctic Ocean. J Geophys Res 114. environmental and anthropogenic stressors that may affect doi: 10.102912008jc004862 the future distribution and abundance of walruses. This Bluhm BA, Gradinger R (2008) Regional variability in food availabil ity for arctic marine mammals . Ecol Appl 18:S77-S96 . framework also allowed for the integration of various doi: 10. 1 890/06-0562. 1 sources of data and expert knowledge, including the paucity Burns n, Shapiro LH, Fay FH (1980) The relationships of marine of data on walrus vital rates and demographic response to mammal distributions, densities, and activities to sea ice condi these stressors. Future efforts to build on our current work tions . U.S. Dept. Commerce, NOAA, Environmental assessment of the Alaskan continental shelf. Final report of principal investi should include monitoring the response of walruses to gators, vol IT. Biological Studies changing arctic conditions, research on specific aspects of Burns n, Shapiro LH, Fay FH (1981) Ice as marine mammal habitat in walrus life history pertaining to climate change effects on the Bering Sea. In: Hood DW, Calder JA (eds) The easternBering the benthos and prey production, and development of wal Sea shelf: oceanography and resources. University Washington Press, Seattle, pp 781-797 rus bioenergetics models relative to changes in sea ice Cavalieri D, Parkinson C, G10ersen P, Zwal1y HJ (1996) Sea ice con availability and benthic prey coupled with models of walrus centrations from Nimbus-7 SMMR and DMSP SSMII passive population dynamics. microwave data, 1979-2007. National Snow and Ice Data Center . Digital media, Boulder (updated 2008) Chivers S (1999) Biological indices for monitoring population status Acknowledgments Funding for this project was provided by the US of walrus evaluated with an individual-based modeL Garner Geological Survey. We aregrateful for the helpful comments and sug In: GW, AmsttupSC, Laake JL, Manly BFJ, McDonald LL, Rober gestions from S. Amsttup, A. Bo1tunov, J. Gar1ich-Miller, B. Kelly, ston DG (eds) Marine mammal survey and assessment methods . and A. Kochnev at various stages in the development of this study. A.A. Balkema, Rotterdam, pp 239-247 L. Kava, G. Noongwook, C. Pungowiyi, R. Toolie, and C. Waghiyi of Choy SL, O'Leary R, Mengersen K (2009) Elicitation by design in Savoonga generously shared knowledge of walruses and other wildlife ecology: using expert opinion to inform priors for Bayesian statis in the St. Lawrence Island area. T. Demere kindly provided comments tical models. Ecology 90:265-277 and improved our discussion of walrus paleobiogeography . We thank Costa DP, Gales NJ (2003) Energetics of a benthic diver: seasonal for US Fish and Wildlife Service, Marine Mammals Management Office, aging ecology of the Australian sea lion, Neophoca cinerea. Ecol for providing walrus harvest data. Mention of trade names is for Monogr 73:27-43 descriptive purposes only and does not constitute endorsement by the Davies JL (1958) Pleistocene geography and the distribution of north federal government . ern pinnipeds. Ecology 39:97-113 Demere TA, Berta A, Adam PJ (2003) Pinnipedimorph evolutionary Conflictof interest The authors declare that they have no conflict of biogeography. Bull Am Mus Nat Hist 279:32-76 interest . � Springer Polar BioI (201l) 34:1065-1084 1083 Douglas DC (2010) Arctic sea ice decline: projecled changes in timing Jensen FV,Nielsen TD (2007) Bayesian networks and decision graphs, and extent of sea ice in the Bering and Chukchi Seas . US Geolog 2nd edn. Springer, New York ical survey open-fliereport 1176 Kavry VI, Boltunov AN, Nikiforov VV (2008) New coastal hau/outs of Fay PH (1960) Structureand fu nction ofthe pharyngeal pouches of the wa1"Uses (Odobenus rosmarus) - response to the climate changes. walrus (Odobenus rosmarus L.). Mammalia 24:361-371 Paper presented at the International Conference of Marine Mam Fay PH (1982) Ecology and biology of the Pacific walrus, Odobenus mals of the Holarctic V, October 14--18,20 08. Odessa, Ukraine rosmarus divergens HEger, vol 74. US Department of the Interior . Koehnev AA, Kryukova NV, Pereverzev AA, Ivanov Dr (2008) Coast Fish and Wildlife Service, Washington al haulouts of the Pacific walruses (Odobenus rosmarus diver Fay PH (1985) Odobenus rosmarus. Mammalian species No . 238 . The gens) in Anadyr Gulf (Bering Sea), 2007 . Paper presented at the Pur.erica.l1Society of Mammalogists InternationalConf erence of Marine �,;lammals cf the HolarcticV. Fay PH,Bowlby CE (994) The hill'"Vest of Pacific walms, 1931-1989. October 14--18, 2008 . Odessa, Ukraine USFWS R7 MMM Technical Report 94-2 . US Fish and Wildlife Kohno N, Ray CE (2008) Pliocene walruses from the Yorktown For Service, Marine Mammals Management, Anchorage, AK mation of Virginia and North Carolina, and a systematic revision Fay FH, Bums JJ ([988) Maximal fe eding depth of walruses . Arctic of the Nol'Lh Atlantic Pliocene walruses, vol 14. Virginia Museum 41:239-240 of Natural HIstory Special Publication, Martinsville, pp 39-80 Fay FH, Ray GC, Kibal 'chich AA (1984) Time and location of mating Kohno N, Tomida Y, Hasegawa Y, Furusawa H (1995) Pliocene and associated behavior of the Pacific walrus . In: Fay Fri, Fedo tusked odobenids (MillTilllalia: Carnivora) in the western North seev GA (eds) Soviet-American cooperative research on marine Pacific,and their paleobiogeography . Bulletin of the National Sci mammals, vol I-Pinnipeds . NOAA Techical report NMFS 12 . ence Museum Series C (Geology and Paleontology) 21:111 -131 pp 89-99 Laidre KL, Stirling I, Lowry LF, Wiig 0, Heide-Jorgensen MP, Fergu Fay FR, Kelly BP, Sease JL (1989) Managing the exploitation of Paci son SH (2008) Quantifying the sensitivity of Arctic ma.,,"ine mam ficwalruses: a tragedy of delayed response and poor corrmmnica mals to climate-induced habitat change . Ecol Appl 18:S 97-S 125 . tlon . Mar MaInm Sci 5:1-16 doi: 1 89CJ/06-G5-2.6.i Fischbach AS, Monson DH, Jay CV (2009) Enumeration of Pacific Lalande C, Grebmeier JM, Wassmann P, Cooper LW, Flint MV, walms carcasses on beaches of Lhe Chukchi Sea in Alaska follow Sergeeva VM (2007) Export fluxes of biogenic matter in the pres ing a mortality event, September 2009. US Geological survey ence and absence of seasonal sea ice cover in the Chukchi Sea . open-file report 2009-1291 Cont Shelf Res 27:2051-2065. doi:! 10 16ij.csr.20C'7.05 GG5 Garlich-Miller JL, Quakenbush LT, Bromaghin IF (2006) Trends in Marcot BG, Steventon JD, Suthedarrd GD, �'i1cCann RK (2006) age structure and productivity of Pacificwalruses harvested irrthe Guidelines for developing and :lpdating Bayesian belief networks Bering Strait region of Alaska, 1952-2002 . Mar Marum Sci applied to ecological modeling and conservation. Can J For Res 22:880-896 36:3063-3074 Gilbert JR (1999) Review of previous Pacific walrus sur,eys to devel Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell op improved survey designs. In: Garner GW, Amstrup SC, Laake JFB, Stouffer RJ, Taylor KE (2007) The WCRP eMIP3 multi IL, Manly BFJ, McDonald LL, Roberston DG (eds) Marine mam medel dataset-a new era in climate change research. Bull Am mal survey and assessment methods . A.A. Balkema, Rotterdam, Meteoroi Soc 88:138 3-1394. doi: lO.l17S/L � Springer 1084 Polar BioI (2011) 34:1065-1084 Simpkins MA, Hiruki- Raring LM, Sheffield G, Grebmeier JM, Bengtson US Fish and Wildlife Service (2009) Endangered and threatened JL (2003) Habitat selection by ice-associated pinnipeds near St. wildlife and plants ; 90-Day findingon a petition to list the Pacific Lawrence Island, Alaska in March 2001. Polar Bioi 26:577-586 Walrus as threatened or endangered. Federal register, vol 74, No . Speckman SG, Chemook VI, Bum DM, Udevitz MS, Kochnev AA, 174, Thursday, September 10, 2009, Proposed Rules, FWS-R7- Vasilev A, Jay CV, Lisovsky A, Fischbach AS, Benter RB (2010) ES-2009-0051 Results and evaluation of a survey to estimate Pacificwalrus pop US Fish and Wildlife Service (2010) Pacific walrus (Odobenus rosm ulation size, 2006. Mar Mamm Sci. doi: 10.I III(j.1748-7692. arus divergens). Alaska marine mammal stock assessments, 20 1 0.004 19.x 2009. NOAA technical memorandum NMFS-AFSC-206 Stroeve J, Holland MM, Meier W, Scambos T, Serreze M (2007) Wang M, Overland IE (2009) A sea ice free summer Arctic within Arctic sea ice decline: faster than forecast. Geophys Res Lett 30 years ? Geophys Res Lett 36:L07502. doi: 10.102912009GL 34:L09501. doi:09510.0 1029!02007GL029703 037820 The Royal Society (2005) Ocean acidificationdue to increasing atmo spheric carbon dioxide. The Royal Society, London Springer