Demographic consequences of migratory stopover: linking red knot survival to horseshoe crab spawning abundance

Transcription

1 Demographic consequences of migratory stopover: linking red knot survival to horseshoe crab spawning abundance CONOR P. MCGOWAN, 1,9, JAMES E. HINES, 1 JAMES D. NICHOLS, 1 JAMES E. LYONS, 2 DAVID R. SMITH, 3 KEVIN S. KALASZ, 4 LAWRENCE J. NILES, 5 AMANDA D. DEY, 6 NIGEL A. CLARK, 7 PHILIP W. ATKINSON, 7 CLIVE D. T. MINTON, 8 AND WILLIAM KENDALL 1 1 U.S. Geological Survey Patuxent Wildlife Research Center, Beech Forest Road. Laurel, Maryland USA 2 US Fish and Wildlife Service, Division of Migratory Bird Management, American Holly Drive, Laurel, Maryland USA 3 U.S. Geological Survey, Leetown Science Center, Leetown Road, Kearneysville, West Virginia USA 4 Delaware Division of Fish and Wildlife, 4876 Hay Point Landing Road, Smyrna, Delaware USA 5 Conserve Wildlife Foundation of New Jersey, 109 Market Lane, Greenwich, New Jersey USA 6 New Jersey Division of Fish and Wildlife, Endangered and Nongame Species Program, P.O. Box 420, 501 E. State Street, Trenton, New Jersey USA 7 British Trust for Ornithology, The Nunnery, Tetford, United Kingdom 8 Victoria Wader Study Group, 165 Dalgetty Road, Beaumaris, Melbourne, Victoria 3193 Australia Abstract. Understanding how events during one period of the annual cycle carry over to affect survival and other fitness components in other periods is essential to understanding migratory bird demography and conservation needs. Previous research has suggested that western Atlantic red knot (Calidris canutus rufa) populations are greatly affected by horseshoe crab (Limulus polyphemus) egg availability at Delaware Bay stopover sites during their spring northward migration. We present a mass-based multistate, capturerecapture/resighting model linking (1) red knot stopover mass gain to horseshoe crab spawning abundance and (2) subsequent apparent annual survival to mass state at the time of departure from the Delaware Bay stopover area. The model and analysis use capture-recapture/resighting data with over 16,000 individual captures and 13,000 resightings collected in Delaware Bay over a 12 year period from , and the results are used to evaluate the central management hypothesis that red knot populations can be influenced by horseshoe crab harvest regulations as part of a larger adaptive management effort. Model selection statistics showed support for a positive relationship between horseshoe crab spawning abundance during the stopover and the probability of red knots gaining mass (parameter coefficient from the top model ^b ¼ 1.71, SE ^ ¼ 0.46). Our analyses also supported the link between red knot mass and apparent annual survival, although average estimates for the two mass classes differed only slightly. The addition of arctic snow depth as a covariate influencing apparent survival improved the fit of the data to the models (parameter coefficient from the top model ^b ¼ 0.50, SE ^ ¼ 0.08). Our results indicate that managing horseshoe crab resources in the Delaware Bay has the potential to improve red knot population status. Key words: apparent survival; Calidris canutus; Limulus polyphemus; match/mismatch; migration carry-over effects; multistate models; robust design. Received 16 April 2011; accepted 16 April 2011; final version received 24 May 2011; published 22 June Corresponding Editor: D. P. C. Peters. Citation: McGowan, C. P., J. E. Hines, J. D. Nichols, J. E. Lyons, D. R. Smith, K. S. Kalasz, L. J. Niles, A. D. Dey, N. A. Clark, P. W. Atkinson, C. D. T. Minton, and W. Kendall Demographic consequences of migratory stopover: linking red knot survival to horseshoe crab spawning abundance. Ecosphere 2(6):art69. doi: /es Copyright: Ó 2011 McGowan et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits restricted use, distribution, and reproduction in any medium, provided the original author and sources are credited. 9 Present address: U.S. Geological Survey, Alabama Cooperative Fish and Wildlife Research Unit, School of Forestry and Wildlife Sciences, Auburn University, Auburn, Alabama USA. cmcgowan@usgs.gov v 1 June 2011 v Volume 2(6) v Article 69

2 INTRODUCTION Conditions at stopover sites may be major determinants of population viability of migratory species (Baker et al. 2004, Newton 2006, Calvert et al. 2009). Understanding how events during one period of the annual cycle carry over to affect survival and other fitness components in other periods is essential to understanding migratory bird demography and conservation needs (Calvert et al. 2009). However, evaluating carry-over effects and factors associated with stopover sites that affect bird survival is difficult. Birds must display some inter-annual stopover site fidelity, and researchers must have some capacity to resight or recapture birds as they pass through a stopover site within and across years. Typically migratory shorebird species using staging sites exhibit stopover site fidelity and can be recaptured and resighted across years, presenting opportunities to evaluate the effects of conditions and events at stopover sites on population dynamics (Morrison 1984). For many migratory species, the timing of migration corresponds closely to the dynamics of critical resources along migration pathways (Alerstam and Hedenström 1998). Timing of arrival and temporary residence at stopover sites must coincide with (match) resource availability, especially for species that show fidelity to sites with ephemeral resources used to fuel longdistance migration (Lyons et al. 2008a, Gillings et al. 2009, Niles et al. 2009). If timing of migration and availability of resources do not coincide (are mismatched), migrating animals may incur fitness consequences. Migrating birds face many energetic and temporal constraints during migration, and the joint dynamics of resource availability and migration timing may play a key role in influencing annual survival and reproduction (Alerstam and Hedenström 1998). These dynamics are relevant to the match/ mismatch hypothesis which postulates that timing of food resource availability can dictate productivity, population growth and abundance in animal populations (Cushing 1990, Fortier et al. 1995, Durant et al. 2007). The match/mismatch hypothesis is of increasing relevance in ecology and conservation because of the potential temporal mismatches created by phenological changes related to global climate change (Durant et al. 2007), and the hypothesis has been shown to be relevant to the dynamics of horseshoe crab spawning and shorebird migrations (Smith et al. 2011). The red knot (Calidris canutus) is a shorebird species that exhibits long distance annual migrations from tropical and temperate wintering grounds to high Arctic breeding grounds, making stops en route at staging sites to refuel and rest before continuing migration. The rufa subspecies, found in the Western Atlantic, migrates from three known wintering areas (Tierra del Fuego in Argentina and Chile, Maranhão in northern Brazil, and the southeastern United States) to the Canadian Arctic with notable stopover in the Delaware Bay on the Atlantic coast of the United States (Harrington 2001, Niles et al. 2008). The rufa subspecies population has exhibited dramatic population declines over the last 15 years (Baker et al. 2004, Morrison et al. 2004, Niles et al. 2009). One hypothesis proposed to explain the decline is over-harvesting of horseshoe crabs (Limulus polyphemus) in the mid-atlantic states, which may affect red knots stopping over in that region (Baker et al. 2004, Niles et al. 2008, 2009). Horseshoe crab eggs serve as the primary food source of knots passing through Delaware Bay (Castro and Myers 1993, Atkinson et al. 2007, Gillings et al. 2007). Much has been written about the relationship between horseshoe crabs and red knots in Delaware Bay, but efforts to quantify that relationship and its effects on population dynamics of either species have been frustrated by data availability and analytical limitations (Baker et al. 2004, Atkinson et al. 2007, Gillings et al. 2007, Niles et al. 2009, Smith et al. 2009). Understanding the relationship between horseshoe crab populations in Delaware Bay and migrating red knots is critical to the conservation and management of both crabs and knots (McGowan et al. 2009, McGowan et al. 2011). In this paper, we develop a capture-recapture/ resighting model tailored to the sampling design used for red knots in Delaware Bay since Specifically, the data are viewed as coming from a robust design (Pollock 1982) with multiple v 2 June 2011 v Volume 2(6) v Article 69

3 secondary sampling occasions (; weekly) within each stopover season ( primary periods). Birds were characterized as being in one of two body mass classes or states (Arnason 1972, Brownie et al. 1993), and our analyses focus on transition probabilities between mass states during the stopover period. We also evaluate apparent annual survival probabilities of birds as a function of mass at departure from Delaware Bay. By apparent survival we simply mean that the complement includes both death and permanent emigration, although we suspect that permanent emigration is small in our case. The model uses two types of encounter data: (1) captures of birds, for which mass state is known, and (2) resighting data based on field-readable marks (color bands or engraved leg flags), for which mass state is unknown. This type of model should have wide application to migration stopover capture-recapture/resighting data sets, especially for other shorebird species that annually congregate at staging and stopover sites. The model is used to focus on two hypothesized key relationships that link red knot population dynamics to the Delaware Bay system: (1) the relationship between horseshoe crab spawning during the shorebird stopover and red knot mass dynamics, and (2) the relationship between red knot departure mass and subsequent apparent survival. Both relationships are critically important to understanding the potential carry-over effects of the Delaware Bay stopover site on red knot population dynamics, and to the hypothesized ability to manage red knot population dynamics and abundance through horseshoe crab harvest management. METHODS Study area and field methods The Delaware Bay is an estuary on the Atlantic coast of North America (; N, ; W), located between the states of New Jersey and Delaware at the mouth of the Delaware River. For 12 years from 1997 through 2008, we trapped red knots during the stopover period ( primarily the month of May) at numerous beaches, harbors and inlets distributed throughout Delaware and New Jersey where red knots are known to occur. Researchers have been using cannon nets (Thompson and DeLong 1967, Clark 1986) to capture and band shorebirds in the Delaware Bay estuary since In that time, they have trapped and marked with USFWS individually numbered metal bands thousands of red knots and large numbers of other shorebird species. In 2003, researchers modified the banding program to include use of field readable, individually coded, alpha-numeric flags (Clark et al. 2005). As a result there are two types of encounter data for marked individuals: physical captures in cannon nets, and field resightings of individual birds. Individual identification number, location, and time were recorded for both types of encounters. Mass and size measurements were collected for birds that were physically captured. Red knots are also trapped and banded at other sites throughout the flyway (e.g., Virginia migratory stopover sites, Argentina wintering sites), and those birds were available for capture or resighting in the Delaware Bay. We only used data that were collected (by trapping or resighting birds) in the Delaware Bay during the stopover; information on original banding and subsequent encounters at other locations was not considered in this analysis. Statistical analyses Capture-recapture/resighting models. To address questions about ecological interactions during stopover, we used a robust design (Pollock 1982, Kendall et al. 1995, 1997) approach to multistate modeling (Arnason 1972, 1973, Hestbeck et al. 1991, Brownie et al. 1993, Schwarz et al. 1993, Lebreton et al. 2009). We tailored a multistate, robust design model to both the research questions of interest and the sampling design used at Delaware Bay. We used two body mass classes to define bird states:,180 g (light) and 180 g (heavy). Use of 180 g as a threshold for defining state was based on the work of Baker et al. (2004), who reported that apparent survival rate of knots in the Delaware Bay changed dramatically at this threshold. One hundred and eighty grams is widely viewed as the critical mass that red knots must obtain in order to survive the flight to the arctic breeding grounds (e.g., Niles et al. 2009). We used the 180 g threshold in our analyses to conform to accepted convention, to make our work relevant to ongoing conservation and v 3 June 2011 v Volume 2(6) v Article 69

4 research, and to test its validity. We split the stopover season into three unequal secondary sampling periods: 1 19 May, May, and 26 May 5 June. Animals could transition between mass states within the stopover season (i.e., between periods 1 2, and 2 3) and across years (i.e., secondary period 3 of year t to secondary period 1 of year t þ 1). We created secondary sampling periods such that approximately one third of all captures between 1997 and 2008 occurred in each of the three periods. For each secondary sampling period, we had two pieces of data in individual encounter histories: (1) captured or not (with mass state assigned to birds that were captured), and (2) resighted or not (with mass state unknown). For example, consider the following encounter history for a two year period: 00 L1 H1 01H001 This example history represents two successive years of capture and resighting data for one individual. The data for each year are grouped into three paired entries per year (first entry pertains to captures and the second to resightings) to represent the three sampling periods. In each encounter history, L (light) indicates that a bird was captured below the 180 g threshold, H (heavy) indicates that a bird was captured above the 180 g threshold, and 1 (which is only possible in the second column of the doublet) indicates that a bird was resighted. A 0 in the first column of a doublet indicates the bird was not captured; a 0 in the second column indicates that a bird was not resighted. In the example encounter history above, the individual was first caught in the light mass class during secondary period 2 of year 1, and also was resighted during period two. In period 3 of year 1, the bird was caught and found to be in the heavy mass class and was also resighted during this period as well. In year 2, the bird was resighted, but not caught, in period 1. It was then caught and found to be in the heavy mass class in period 2, year 2. Finally, it was resighted, but not recaptured, during period 3 of year 2. Use of the robust design with multistate models is not well-developed in the literature. Initial applications used separate analyses of closed and open model data (Nichols and Coffman 1999, Coffman et al. 2001) and more recent applications used joint likelihoods containing both closed and open model components (Nichols et al. 2000, Kendall et al. 2003, Skvarla et al. 2004, White et al. 2006, Lebreton et al. 2009). The primary periods of the robust design in our study were the May-June stopover periods of each year; each primary period includes three secondary periods (i.e., three capture/recapture resighting occasions). We could not use the traditional robust design which assumes closure within primary periods because red knots may arrive and depart the stopover site at any of the secondary periods. We thus required an open robust design (Schwarz and Stobo 1997, Kendall and Bjorkland 2001). Migration stopover behavior of red knots dictated that we constrain the open robust design to a single entry and a single departure each year (primary period). Departure probabilities required no special modeling, but our use of entry probabilities each year suggested the use of a superpopulation approach (Crosbie and Manly 1985, Schwarz and Arnason 1996) to model data from the secondary sampling occasions of each year. By superpopulation approach, we simply mean that we consider some total number of birds associated with the stopover site each year, and that these birds enter and depart at various times, such that the number of birds actually present at the stopover site at any secondary period is typically less than the superpopulation size. The superpopulation concept applies to each stopover season and includes all birds that are present in Delaware Bay during at least one period of the stopover season (i.e., the flyway population). Although the use of the superpopulation approach in migration stopover studies has been suggested a number of times (Nichols 1996, Nichols and Kaiser 1999, Efford 2005, Kendall 2006), it has only recently been used for analysis of data from stopover sites (Pledger et al. 2009). The modeling framework that we developed estimates the following seven parameters: p s i;t ¼ state-specific (mass state s; s ¼ L, H ) probability that a bird of mass class s that is present in Delaware Bay during secondary period i (i ¼ 1, 2, 3) of primary period (year) t is captured at that time; r s i;t ¼ state-specific (mass state s; s ¼ L, H ) v 4 June 2011 v Volume 2(6) v Article 69

5 probability that a marked bird of mass class s that is present in Delaware Bay during secondary period i (i ¼ 1, 2, 3) of primary period (year) t is resighted at that time; b s i;t ¼ state-specific (mass state s; s ¼ L, H ) probability that a bird of mass class s that is alive at the beginning of the stopover period i in year t, and that has not entered the Delaware Bay system prior to period i, enters the system in period i, and is available for capture and resighting in period i (i ¼ 1, 2, 3); d s i;t ¼ state-specific (mass state s; s ¼ L, H ) probability that a bird of mass class s in period i of year t that entered the Delaware Bay system prior to period i, departs the system to migrate north between sample periods i and i þ 1(i¼1, 2, 3), where d s 3;t ¼ 1 (i.e., all birds depart in the final period); w ss 0 i;t ¼ state-specific (mass state s; s ¼ L, H ) probability that a bird of mass class s in period i of year t that has not departed Delaware Bay before period i þ 1 of year t is in mass class s 0 at period i þ 1, for i ¼ 1, 2 (note that s may equal s 0, indicating no mass state change); w ss 0 3;t ¼ state-specific (mass state s; s ¼ L, H ) probability that a bird of mass class s in period 3 of year t that survives until the stopover period of year t þ 1 is in mass class s 0 at the period of entry to the Delaware Bay system in year t þ 1; S s t ¼ state-specific (mass state s; s ¼ L, H ) apparent annual survival probability; probability that a bird that is present in Delaware Bay at some secondary period of primary period t and departs the Bay in mass class s, is alive and has not permanently emigrated from the Bay in the stopover season of year t þ 1. These parameters are then used to model the encounter history data. For example, consider the example history presented above, 00 L1 H1 01H0 01. The probability associated with this history can be written in terms of the model parameters as: Prð00 L1 H1 01H001j unmarked bird of mass L captured in period 2 of year 1Þ ¼ r L 2;1 ð1 dl 2;1 ÞwLH 2;1 ph 3;1 rh 3;1 dh 3;1 SH 1 3 ½w HL 3;1 bl 1;2 ð1 pl 1;2 ÞrL 1;2 ð1 dl 1;2 ÞwLH 1;2 þð1 w HL 3;1 ÞbH 1;2 ð1 ph 1;2 ÞrH 1;2 ð1 dh 1;2Þð1 whl 1;2 ÞŠ 3 p H 2;2 ð1 rh 2;2 Þð1 dh 2;2 Þ½wHL 2;2 ð1 pl 3;2 ÞrL 3;2 dl 3;2 þð1 w HL 2;2 Þð1 ph 3;2 ÞrH 3;2 dh 3;2 Š The expression starts with (is conditioned on) the initial capture of the bird, and at this encounter the mass class is known ( L in period 2). The rest of the history is then modeled using the defined parameters, starting with the immediate resighting in period 2. The first line of the expression contains no uncertainty, as the bird was caught in the light mass class in period 2 and resighted, did not depart/migrate, was then caught again in the heavy mass class in period 3 and resighted, departed Delaware Bay, and finally survived until year 2. The second line incorporates the uncertainty associated with mass state at entry to the system in year 2. The bird was known to have entered in period 1, as it was resighted, but because it was not captured then, its mass state is not known. Regardless of mass state at entry, the bird made the transition to the heavy state in period 2, where it was captured but not resighted. It was resighted again in period 3 of year 2 and then departed, but mass state is again not known for period 3. A probability was constructed in this manner for the observed encounter history data for each bird, j, Pr(h j ). The joint likelihood for the entire data set is then proportional to the product of these probabilities: n o n o n Lð w ss 0 i;t ; b s i;t ; d s i;t o ; S s t n o Y ri;t s J j h j Þ } n ; p s i;t j¼1 Prðh j Þ; o ; where J is the total number of individual encounter histories in the analysis. Maximum likelihood estimates of the parameters were obtained numerically. We were especially interested in the possible relationship between horseshoe crab covariates (e.g., abundance of spawning crabs during the stopover period) and bird mass transitions. Covariate relationships were incorporated directly into our modeling framework using linear-logistic models, e.g.: v 5 June 2011 v Volume 2(6) v Article 69

6 w LH i;t ¼ eb 0þb 1 X t 1 þ e b ; 0þb 1 X t where X t is the crab covariate for year t, and b 0 and b 1 are intercept and slope parameters to be estimated. Details of covariate analyses and competing models are described in Candidate models and model selection. Our model requires the usual capture-recapture assumptions of homogeneity of probabilities (e.g., survival, mass transition, capture, resighting, entry, departure) for birds in a particular state at a specific point in time. We further assumed that birds behaved independently relative to these modeled processes. Some assumptions were specific to the Delaware Bay system and sampling program. For example, survival probability was assumed to approach 1 during the stopover period. This assumption is reasonable considering the high annual survival rate for this species: with an annual survival of 0.90, the survival rate during any two week period (i.e., the approximate length of stopover in Delaware Bay; Gillings et al. 2009) is approximately (assuming constant weekly survival rates throughout the year). In our model, birds could enter and depart Delaware Bay during any of the 3 periods in either mass state. A single entry and departure were permitted each year. For example, birds were assumed not to enter, then depart and then reenter the system in the same year. Prevailing thought, supported by anecdotal observations (N. Clark, personal communication) and some analytical studies (Gillings et al. 2009), asserts that the birds typically depart Delaware Bay en masse on or around May 28th (Niles et al. 2009). Our open robust design model does not require this assumption and allows birds to enter and depart the Bay at any time during the stopover season. Once a bird was captured and marked, the model development assumed that it would enter the Delaware Bay system each year it remained alive and had not permanently emigrated. Random temporary emigration (unavailability of a bird for one year) does not bias survival estimates, with the exception that capture and resighting probabilities now include the probability of a bird not being present for detection in a given year (Kendall et al. 1997). Year-, period-, and statespecificity were possible for all model parameters, so the modeling framework was extremely general and flexible. Candidate models and model selection. We developed a set of candidate models that incorporated various hypotheses about year-, period-, and state-specificity of all model parameters (Appendix). Some of these hypotheses were of primary ecological interest (e.g., state-specificity of apparent annual survival, with a prediction of higher survival for birds departing Delaware Bay heavy as opposed to light), whereas others were simply focused on parsimonious modeling of the available data. Goodness-of-fit tests are not available for robust design models, so we followed the usual approach of including very general models (including as many sources of variation as likely to be supported by the data) in the model set (e.g., Williams et al. 2002). To assess potential consequences of overdispersion in our data and lack of fit of our models, we conducted a ĉ sensitivity analysis (increasing the ĉ value to 1.5, 2.0 and 3.0) to see if the model selection results changed (Appendix). We followed a sequential approach to model selection (see below). We make no claim about optimality of such an approach but simply note that it represents a pragmatic attempt to deal with models such as these that include a large number of potential parameters (e.g., for similar approach see Franklin et al. 2004, Anthony et al. 2006). Initially, we determined the best model structure for parameters associated with the sampling process (capture probabilities, resighting probabilities) and for ecological parameters not under direct investigation (entry/arrival and departure probabilities). We compared various combinations of state-, time-, and period-dependency of capture and resighting probabilities, while keeping other parameters as general as possible (yearand mass-specific apparent survival probabilities, period-specific mass transition probabilities constant across years, mass- and period-specific entry and departure probabilities). Although not the primary focus of our analyses, we note that there were a priori hypotheses about the statespecificity of capture and resighting probabilities. Specifically, we hypothesized that mass would be relevant to detection, with lighter birds having higher capture, and probably resighting, probabilities than heavier birds. Netting and capture of v 6 June 2011 v Volume 2(6) v Article 69

7 birds is typically focused on feeding beaches, leading to the prediction that light mass birds might predominate in such locations. Much resighting also takes place at feeding areas, and thus we expected to find similar patterns as with capture probabilities. We settled on capture probabilities that were state- plus year-dependent (i.e., capture probabilities varied across years, but in a parallel manner for the 2 mass classes) and resighting probabilities that depended on state, period and year with interactions. We set resighting probabilities for years before 2003 equal to zero, because field-readable bands were not used prior to that season. We modeled mass gain of birds as a function of two horseshoe crab covariates. Delaware Bay horseshoe crabs spawn predominately in May and June. The proportion of the population that spawns during May, the period believed to be most critical for red knots, varies each year as a function of water temperatures and wave generating storms (Smith and Michels 2006). Our first crab covariate (HSC1) therefore was the estimated number of female crabs spawning during May (i.e., estimated female crab abundance in year t 3 estimated proportion of female crabs that spawned during May). HSC1 was based on two sources of data: (1) the annual Virginia Polytechnic Institute and State University offshore trawl survey (Hata and Hallerman 2009), which was conducted during September-November, and (2) a beach-based spawning survey, conducted during May-June (Smith et al. 2002, Smith and Michels 2006). We used total female crab abundance each year (HSC2) as an alternative horseshoe crab covariate, estimated using the offshore trawl survey data (Hata and Hallerman 2009). We used trawl survey data from the fall of each year because we believe that the trawl survey samples the post-spawning population as crabs exit Delaware Bay and return to the Atlantic Ocean for the winter months. Distinct models with covariates HSC1 or HSC2 were included in the candidate model set to investigate the importance of both abundance and timing of spawning for red knot mass gain (and subsequent apparent annual survival). Based on the match/mismatch hypothesis (Cushing 1990), we predicted that models with the HSC1 covariate would receive more support than models with HSC2 if red knot mass dynamics are driven not only by total female horseshoe crab abundance, but also by the temporal match of spawning activity with red knot stopover activity. We also assessed support for models that restricted the covariate structure on transitions from heavy to light states. Of.180 observed transitions in the 12 year study, we observed only four within-season transitions from heavy to light mass state, as well as seven cases where birds remained in the heavy state between periods. Given the paucity of within-season transition data for birds first observed in the heavy state, we evaluated models in which transition from the heavy state was constant over years (i.e., w HL i;t ¼ w HL i;: ; thus whh i;t ¼ w HH i;: ). The transition probabilities for 1997 and 1998 were modeled as year-specific parameters, rather than as a function of female horseshoe crab covariates, because we do not have spawning or trawl survey data from those years (Smith and Michels 2006, Hata and Hallerman 2009). We evaluated the effect of arctic weather conditions at the time of red knot arrival on the breeding grounds on apparent annual survival. Morrison (2006) and Morrison et al. (2007) reported that ( predicted) body condition upon arrival in the arctic impacted apparent survival for European red knots (C. c. islandica) breeding at Ellesmere Island, Canada. Morrison et al. (2007) concluded that birds leaving stopover sites in Iceland at low mass had higher mortality upon arrival on the breeding grounds if there were extensive snow cover and cold temperatures. Snow-related mortality also may impact rufa red knots of the Delaware Bay. We tested the hypothesis that negative effects of small fuel reserves when departing Delaware Bay are exacerbated by deep snow cover early in the breeding season; we expected to see lowest apparent survival of light birds in years of greater snow depth. We also expected that heavy birds would be unaffected, or less affected, by arctic snow depth. We used data from the Canadian National Climate Data and Information Archive (Environment Canada 2009; National Climate Data Archive, available online: hhttp://climate. weatheroffice.ec.gc.ca/i) from four weather stations distributed throughout the red knot breeding range (Resolute, Cambridge Bay, Taloyoak, v 7 June 2011 v Volume 2(6) v Article 69

8 and Coral Harbour) to acquire two arctic weather covariates. We chose these four stations because they were within the published estimates of red knot breeding range (Harrington 2001, Niles et al. 2008) and had data spanning all years of our capture-recapture study. Additional weather stations throughout the Nunavut Territory were not used because they were either outside the published red knot breeding range or they did not have data for all years of our study. For the first covariate in our arctic conditions survival models, we used the mean snow depth on 10 June, and for the second we used the number of days with mean temperature below 08C during 1 15 June across all four sites. The 10 June date, approximately 7 days after peak arrival of knots in most years, was chosen to represent conditions in the arctic breeding grounds in the days immediately following red knot arrival. The 180 g mass threshold is the mass hypothesized to lead to high red knot survival during the flight to the Arctic and several days following arrival (needed to retransform the intestinal tract), providing a buffer against poor arrival conditions in the arctic (A. Baker and N. Clark, personal communication). The interval 1 15 June covers the arrival and acclimation period. We used days below 08C as a second covariate to test the hypothesis that days with freezing temperatures would lead to fewer terrestrial invertebrate prey items, and because we thought that temperature data would be less subject than snow, for example, to local variation (e.g., drifting snow). We included models with no structure on apparent survival (null models), state-dependent apparent survival models, snow- or temperature-dependent survival models, and snowor temperature- and state-dependent apparent survival models (additive models and models with interaction terms) (Appendix). We used an AIC c information theoretic approach to compare the relative appropriateness of candidate models to the data (Akaike 1973, Burnham and Anderson 2002). All statistical analyses were done using a modified version of the program MSSURVIV (Hines 1994). RESULTS We had 16,151 individual birds in our data set, with a total of 16,609 captures (13,475 light mass birds and 3,097 heavy birds) and 13,992 resightings (Table 1). Birds captured and weighed multiple times within a stopover season (year) provide critical information about mass transitions. We had a total of 188 within-season recaptures in the data set after partitioning captures into the three sampling periods. Of these recaptures, 133 were of individuals that did not change mass states, but remained light or heavy. Of the 55 within-season recaptures that document transitions between mass states, four of those were transitions from heavy to light. The best supported model based on AIC c selection criteria modeled apparent annual survival as dependent on the snow covariate; light to heavy transition probabilities as dependent on female horseshoe crab spawning during the stopover (HSC1) and on within-season period (with interaction terms); and heavy to light transition probabilities as a function of withinseason period only (Table 2). The second best supported model described apparent annual survival as a function of mass state and arctic snow depth (additive terms), and the third ranked model described apparent survival as a function of mass state, arctic snow depth, and the interaction of mass state and arctic snow depth. The second and third ranked models were within 2.5 AIC c units of the best model, which indicates substantial uncertainty among the top three models. The top three models together comprised over 98% of the AIC c weight. The uncertainty among the top three models suggests the use of model averaging of parameter estimates across those three models (Burnham and Anderson 2002). Models with no state structure or snow covariates on apparent survival rates (e.g., null survival models) received no support in the model selection analysis, and models without the HSC1 covariate on the light to heavy transition probability also received no support. Our model selection results show strong support for state-dependent capture probabilities. Model-averaged mean capture probability (ˆ p) was for light birds and for heavy birds. According to model selection results, capture probabilities also varied by year (Table 2). We also found some evidence to support statedependent differences in resighting probability. The mean period-specific resighting probability (ˆ r) was for light birds and for heavy v 8 June 2011 v Volume 2(6) v Article 69

9 Table 1. Number of individuals caught in each mass class (heavy and light) and number of resights by year and period in Delaware Bay, USA. Period Year Heavy Light Resights Heavy Light Resights Heavy Light Resights NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA Total Table 2. Model selection for candidate models of apparent annual survival (S) and transition probabilities (w) between light (L, mass, 180 g) and heavy (H, mass 180 g) mass states for red knots (Calidrius canutus rufa) in Delaware Bay, USA. Apparent annual survival was modeled as a function of mass state ( s ) at time of departure and arctic snow cover ( snow ). Transition probabilities were modeled as functions of spawning abundance (HSC1) or total abundance (HSC2) of female horseshoe crabs (Limulus polyphemus) and migration period within each year ( per ). Model S w(l-h ) w(h-l) kà 2LogLik AIC c D AIC c w i 1 snow HSC13per per s þ snow HSC13per per s 3 snow HSC13per per s 3 snow HSC1 þ per per snow per per } s 3 snow HSC1 3 per s 3 snow HSC1 3 per HSC1 3 per s 3 snow per per The model structure for all 8 top models for the parameters p (recapture probability), d (departure probability from the migratory stopover site), r (resighting probability), and b (arrival probability to the stopover sight) was p(s þ t) d(s 3 per) r(s 3 t 3 per) b(s 3 per); t indicates a year-specific parameter. à Number of parameters in the model. Akaike model weights. } The. parameterization indicates that there were no covariates or sources of variation for w(h-l) in this model. birds. The model selection analysis supported year and period effects on resighting probability, and the estimates showed substantial variation across periods (e.g., ˆ r H 1 ¼ 0.058, ˆ r H 2 ¼ 0.533, ˆ r H 3 ¼ 0.469). Resighting effort typically changes across the season, depending on the number of volunteers and field workers. There was also high variability in estimated resighting rates between states within some periods (e.g., ˆ r H 1 ¼ 0.058, ˆ r L 1 ¼ 0.378, for ). Apparent survival estimates indicate that heavy birds had a slightly higher average apparent survival probability than light birds. The model-averaged mean annual apparent survival rate estimate for heavy birds was ( SE ^ ¼ 0.030), and that for light birds was ( SE ^ ¼ 0.036). During most years, apparent survival rates were similar and high, but in some years they diverged (e.g., 1997, 1998, 2005; Fig. 1). The model-averaged point estimate of light bird apparent survival was lower than that for heavy birds in 6 of the 11 years. For example, the apparent survival rate estimate was for heavy birds and only for light birds. The b coefficients associated with the mass transition probabilities indicated that from peri- v 9 June 2011 v Volume 2(6) v Article 69

10 Fig. 1. Model-averaged estimates of annual survival of red knots in heavy (solid squares) and light (solid circles) mass classes (695% C.I. in thinner dashed or thick light gray lines) at time of departure from the Delaware Bay stopover site related to average snow depth (solid dark gray line) from four weather stations at time of arrival in the Canadian arctic breeding grounds, ods two to three, the probability of transition from light to heavy was positively correlated with female horseshoe crab spawning abundance in May (HSC1; Fig. 2, Table 3). The HSC1 coefficients for the light to heavy transition rate between the first and second periods were close to 0, providing no evidence of an effect of spawning crabs on knot mass gain early in the season (Fig. 2, Table 3). Removing the horseshoe crab relationship from the heavy to light transition probabilities in favor of a simpler single parameter (this was done in response to very limited data; see the Methods section) improved the model. Models with the HSC2 covariate (total female abundance without accounting for annual spawning activity) did not receive support in the model selection. Models that did not include any horseshoe crab abundance covariate received little or no support in the model selection analysis (Table 2). The b parameters for the Arctic snow depth covariate indicate a significant positive effect of snow depth in the Arctic on red knot annual survival (Table 3), the opposite of what we predicted. Both light and heavy birds were estimated to have higher probabilities of survival in years with more snow (Fig. 1), and the effect of snow depth was stronger for light birds than for heavy birds (Table 3, Fig. 1). Models with number of days with mean temperature below 08C as the arctic conditions covariate, instead of snow depth, received no support compared to v 10 June 2011 v Volume 2(6) v Article 69

11 Fig. 2. Model-averaged estimates of transition probabilities (695% C.I.) from light to heavy mass classes (period 1 2 transitions represented by X s and period 2 3 transitions represented by triangles) for red knots stopping over in the Delaware Bay plotted against (A) total estimated female horseshoe crab abundance (HSC2) and (B) estimated female spawning abundance during the month of May (HSC1), from Hata and Hallerman (2009) and S. Michels (Delaware Division of Fish and Wildlife, personal communication), other models in our model set. We assessed the robustness of our inferences based on model selection to possible lack of fit by computing model selection statistics using overdispersion values (ĉ) ranging from 1 (good fit of most general model) to 3 (fairly poor fit). Model rank order varied to some degree across the different values of ĉ, but overall inferences changed very little (Appendix: Tables A1 A5). The effect of snow on survival was included in all models with QAIC c weights.0.01, regardless of ĉ. Mass state was an important determinant of survival in 3 of the top 5 models for all values of ĉ, although the summed model weights for models including state was reduced at higher values of ĉ. Similarly, seasonal abundance of horseshoe crabs was an important determinant of mass gain in at least 3 of the top 5 models for all ĉ values except ĉ ¼ 3 (2 models). Summed weights of models containing the crab abundance covariate declined as ĉ increased, but were.0.4 except for ĉ of 2.5 and 3.0. In conclusion, larger values of ĉ led to more support for simpler models, but support for the most important covariate rela- v 11 June 2011 v Volume 2(6) v Article 69

12 Table 3. Coefficient estimates for covariate relationships corresponding to the most supported models linking red knot survival (S) to arctic snow depth (snow) by mass state (L, H ), and transition probabilities (w) between light (L, mass, 180 g) and heavy (H, mass 180 g) mass states for red knots (Calidrius canutus rufa) in Delaware Bay, USA, to female horseshoe crab spawning abundance (HSC1). Model 1, w i ¼ 0.54 Model 2,à w i ¼ 0.20 Model 3, w i ¼ 0.18 Relationship Coefficient SE Coefficient SE Coefficient SE w LH 1(HSC1) w LH 2(HSC1) S L (snow) S H (snow) S(snow)w(L-H(HSC1 3 per), H-L( per))p(s þ t)d(s 3 per)r(s 3 t 3 per )b(s 3 per). à S(s þ snow)w(l-h(hsc1 3 per), H-L( per))p(s þ t)d(s 3 per)r(s 3 t 3 per )b(s 3 per). S(s 3 snow)w(l-h(hsc1 3 per), H-L( per))p(s þ t)d(s 3 per)r(s 3 t 3 per )b(s 3 per). tionships did not vanish even for large ĉ. DISCUSSION Our results show evidence of two key relationships that provide a direct link between red knot demography and spawning abundance of female horseshoe crabs in Delaware Bay. The probability that a bird arriving at Delaware Bay weighing,180 g will become a heavy bird (.180 g) was positively related to estimated female crab abundance on spawning beaches during migration stopover. We also provide evidence that apparent annual survival of red knots was dependent on arctic snow conditions upon arrival on the breeding grounds and, to a lesser extent, body mass when departing Delaware Bay. Red knot refueling during stopover Our analyses provided strong evidence that horseshoe crab spawning during the shorebird stopover (i.e., during May) influenced mass gain of red knots on Delaware Bay (Castro and Myers 1993, Atkinson et al. 2007). Furthermore, our results provided strong evidence that the timing of horseshoe crab spawning, not simply crab abundance, is important to red knot refueling during stopover (Fig. 2A, B). We observed very few instances of within season mass loss (downward transitions) in our data set. These downward transitions could represent a mass loss due to handling effect (Barter and Minton 1998). Despite the limited data on mass transitions, we found evidence of a positive relationship between abundance of horseshoe crabs spawning in May (HSC1) and the probability that a light bird becomes a heavy bird. The evidence was strong for transitions between periods 2 and 3; light-heavy transitions between periods 1 and 2 showed virtually no relationship with number of crabs spawning in May. Horseshoe crab spawning typically peaks during the second half of May, and is expected to be most important for late season mass gain. Total horseshoe crab abundance (HSC2) received little support as a predictor of mass state transition. Our evidence for the relevance of timing of spawning supports match/mismatch ideas, and emphasizes the importance of interactions of environmental variation and crab abundance for refueling of red knots in Delaware Bay. If spawning is delayed, even with relatively high total crab abundance, the probability that a light bird will add enough mass to become a heavy bird before departure may be lower, as was the case in 2008 when a mid-may Nor eastern storm caused delayed spawning (Fig. 2 A, B; Hata and Hallerman 2009). There have been several previous studies examining red knot mass gain during Delaware Bay stopover (Atkinson et al. 2007), and others looking at red knot use of horseshoe crab eggs during stopover (e.g., Karpanty et al. 2006, Gillings et al. 2007); however, our study represents the first direct test of the hypothesis that horseshoe crab abundance influences red knot mass gain and demography. Red knot annual survival We report apparent survival estimates, the complements of which confound annual mortality with permanent emigration (Williams et al. 2002). Apparent survival estimates are thus biased low with respect to true survival (Williams et al. 2002), although we suspect this bias to v 12 June 2011 v Volume 2(6) v Article 69

13 be small in our estimates because of the staging behavior exhibited by red knots in Delaware Bay and because of the longer term nature of our data set. Our estimates (ˆ S 0.92) are relatively high compared to other published estimates of annual survival of red knots (Harrington 2001, Baker et al. 2004, Morrison 2006, van Gils et al. 2006). For example, in the Baker et al. (2004) study, mean annual survival was approximately 0.86 (Table 2b in Baker et al. 2004). Our year-specific apparent survival rate estimates fell within the range of estimates for groups of years presented by Baker et al. (2004: Table 2b). In addition, our lowest estimates occurred in 1998, during the period of reported substantial population decline and low estimated apparent survival for the Tierra del Fuego population (Baker et al. 2004). However, given recent substantial reported population declines (Morrison et al. 2004, Niles et al. 2008), our survival estimates were somewhat larger and more stable than we had expected. Peak counts of knots from shorebird aerial surveys in Delaware Bay dropped from ;50,000 in to ;24,000 in 2009 (Niles et al. 2008; K. Kalasz, unpublished data), and previous studies have concluded that much of that decline was due to decreased adult survival caused by poor stopover conditions in Delaware Bay (Baker et al. 2004, Niles et al. 2009). Changes in count numbers from aerial surveys can be caused by changes in actual abundance, permanent or temporary emigration, or count effectiveness and timing. McGowan et al. (2011) used a full two species population model to simulate population change of horseshoe crabs and red knots using reported horseshoe crab harvest from and estimated red knot apparent survival and mass relationships from this paper. McGowan et al. (2011) demonstrated that the apparent survival estimates reported herein are potentially consistent (conditional on values of other vital rates [e.g., juvenile survival, reproductive rate], the values of which are not known) with a projected median red knot population decline of.40%. Our results suggest that arctic conditions (or large-scale correlates of arctic condition) are predictors of apparent annual survival probability for red knots, and have stronger effects on light birds than on heavy birds. Unexpectedly, our analyses indicate that snow depth in the breeding grounds has a positive influence on apparent survival probability for both heavy and light birds, the opposite of what we predicted based on Morrison (2006) and Morrison et al. (2007). There are many possible explanations of this unexpected result. For example, it is possible that birds may skip breeding in years with heavy snow after arriving in the arctic, assessing local and their own energetic conditions and deciding to forgo breeding until conditions are better. It is possible that birds that forgo the risk and effort of breeding might have higher annual survival (Williams 1966). Alternatively, the arctic breeding grounds are essentially deserts (Noy-Meir 1973), and snow may determine annual moisture and water in the environment, thus driving terrestrial invertebrate production and therefore red knot food availability in the breeding grounds. More snow results in more food for breeding red knots and thus leads to higher survival rates. A third possibility is that red knot survival is tied to lemming cycles in the arctic, which are in turn closely linked to snow depth. In years of low snow, when lemming populations crash arctic predators may rely more heavily on migrant birds as prey items (Lindström and Hornfeldt 1994). Other hypotheses to explain this pattern are possible, and this unexpected result requires more research. It is important to note that mean arctic snow depth as measured at four weather stations may not be a comprehensive measure of snow conditions across the breeding range of rufa red knots (Yang and Woo 1999). Furthermore, significant uncertainty exists regarding the breeding range of rufa red knots (Niles et al. 2008). Using alternative weather stations to develop covariates for our analyses may have led to different results, and the patterns that we detected require further research. Our analyses provided some support for the hypothesis that body mass at departure from Delaware Bay affects apparent annual survival of red knots. Two of the top three supported models in our AIC analyses, with a combined model weight of 38%, included effects of body mass (state) on apparent survival (Table 2, Appendix). However, the estimated magnitude of the effect of mass state on survival was not as large as anticipated, based on previous analyses (Baker et al. 2004). The difference between our estimates of state-specific apparent annual survival and those of previous researchers (e.g., Baker et al. 2004) v 13 June 2011 v Volume 2(6) v Article 69