Correlating tropical climate with survival of an Arctic-breeding, trans-equatorial migrant seabird

: Extreme climate can negatively affect survival through increased physiological demands or by reducing prey availability. This can have significant population-level consequences for organisms with low reproductive rates, such as seabirds. As an Arctic-breeding trans-equatorial migrant, Sabine’s gull ( Xema sabini ) is exposed to a profound variety of climate regimes during the year. Therefore, its annual survival may be affected by broad-scale teleconnection patterns that influence regional climate variability. We used Program MARK to estimate apparent survival and resighting probabilities from 2007˗2013 for adult Sabine’s gulls breeding at a High Arctic colony. We then combined capture-mark-recapture data for the High Arctic colony with those previously published from a Low Arctic colony (1998˗2002) to examine influences of climate variability on survival. Mean ± SE apparent survival estimated for the High Arctic colony was 0.90 ± 0.03, similar to that previously reported for the Low Arctic colony. We found a negative relationship between survival and the Tropical/Northern Hemisphere pattern, an atmospheric mode that is associated with the Pacific jet stream. Our study suggests that although Sabine’s gull survival was generally high and relatively constant over time, adult mortality may increase during years of extreme climate events in regions far beyond their Arctic breeding grounds.

D r a f D r a f t Abstract: Extreme climate can negatively affect survival through increased physiological demands or by reducing prey availability. This can have significant population-level consequences for organisms with low reproductive rates, such as seabirds. As an Arctic-breeding trans-equatorial migrant, Sabine's gull (Xema sabini) is exposed to a profound variety of climate regimes during the year. Therefore, its annual survival may be affected by broad-scale teleconnection patterns that influence regional climate variability. We used Program MARK to estimate apparent survival and resighting probabilities from 2007˗2013 for adult Sabine's gulls breeding at a High Arctic colony. We then combined capture-mark-recapture data for the High Arctic colony with those previously published from a Low Arctic colony (1998˗2002) to examine influences of climate variability on survival. Mean ± SE apparent survival estimated for the High Arctic colony was 0.90 ± 0.03, similar to that previously reported for the Low Arctic colony. We found a negative relationship between survival and the Tropical/Northern Hemisphere pattern, an atmospheric mode that is associated with the Pacific jet stream. Our study suggests that although Sabine's gull survival was generally high and relatively constant over time, adult mortality may increase during years of extreme climate events in regions far beyond their Arctic breeding grounds.
D r a f t

Introduction
Warming climate, increasing climate variability, and extreme weather events may all have negative effects on migrant birds (Stenseth et al. 2002, Frederiksen et al. 2008, particularly for those travelling long distances (Both et al. 2009). For seabirds, effects are typically manifested either by changes in marine food availability, changes in physiological stressors, or a mismatch in the timing of migration relative to peak food supplies (e.g., Schreiber 2001;Sandvik et al. 2005), and can have carryover effects across years (Déscamps et al. 2010). Although there is much variation among species and effects, generally poor marine habitat conditions in the winter or on migration have negative influences on adult or juvenile survival (e.g., Thompson and Ollason 2001;Sandvik et al. 2005), whereas poor conditions on the breeding grounds tend to result in reduced reproductive effort (number of breeders, egg size, or clutch size) or breeding success (e.g., Gaston et al. 2005;Moe et al. 2009). Over longer periods, climate trends can influence seabird populations, and thus understanding seabird migration and population dynamics in relation to climate patterns is essential for understanding current trends and predicting future effects on species (e.g., Trathan et al. 2007;Irons et al. 2008).
Sabine's gull (Xema sabini) is a small, ~190-g larid seabird that has a patchy, circumpolar breeding distribution, nesting primarily on small Arctic islands and in coastal wetland areas (Day et al. 2001). It is the only Arctic-breeding gull with a trans-equatorial migration, making a return trip of more than 28,000 km annually to overwinter at highly productive upwelling zones off the coasts of Peru and southern Africa (Stenhouse et al. 2012;Davis et al. 2016). Limited census data for Sabine's gulls suggest that their numbers are stable or increasing in parts of their breeding range, but declining in others (Raven and Dickson 2006; D r a f t Johnston and Pepper 2009;Mallory et al. 2012). To date, survival of adult Sabine's gulls has only been reported for a Low Arctic colony at East Bay Migratory Bird Sanctuary (Southampton Island, Nunavut) and was relatively constant and high at 0.89 ± 0.02 from 1998˗2002 (Stenhouse and Robertson 2005). However, the short duration of that study precluded the ability to detect temporal variability in survival, or assess influences of environmental variables.
To further our understanding of Sabine's gull population dynamics, apparent survival and resighting probabilities were estimated for a High Arctic colony, and compared to the Low Arctic colony at East Bay (Stenhouse and Robertson 2005). Furthermore, data for both colonies were combined to test for an influence of climate variability on apparent survival of adult Sabine's gulls breeding in the Canadian Arctic. Because we were interested in adult survival, likely affected by winter or migration habitat conditions, we used three broad-scale climate indices as proxies for climate variability in winter habitats: the North Atlantic Oscillation, the Southern Oscillation Index, and the Tropical/Northern Hemisphere pattern. We expected that the phases of these indices that are correlated with warmer water and a consequent reduced marine productivity (e.g., Hays et al. 2005;Behrenfeld et al. 2006;Irons et al. 2008;Beaugrand et al. 2015) would also be correlated with lower survival in Sabine's gulls. Data from this study constitute baseline demographic information on Sabine's gulls, which we hope leads to identification of factors affecting survival of long-distance migrants breeding in the Arctic.

Study sites
The High Arctic site used in this study was Nasaruvaalik Island (75° 49' N, 96° 18' W; Fig. 1), one of several islands found within Queens Channel, Nunavut, Canada. This area is D r a f t considered important breeding and foraging habitat for many marine birds (Mallory and Gilchrist 2003), largely due to the occurrence of polynyas, which are areas of persistent open water surrounded by sea ice. The 3-× 1-km gravel island has old beach ridges and scattered patches of purple saxifrage (Saxifraga oppositifolia), moss, and lichen (Mallory et al. 2012). Sabine's gulls (16˗31 pairs, over eight years of monitoring) nest within Arctic tern (Sterna paradisaea) colonies (~600 pairs in total) at either end of the island (Mallory and Gilchrist 2003;Mallory et al. 2012; S.E. Davis, unpublished data).
To obtain a longer time series for the purpose of evaluating effects of climate on survival, Nasaruvaalik Island data were combined with previously published data from a Low Arctic colony at East Bay Migratory Bird Sanctuary (64° 01' N, 81° 47' W; Fig. 1

), Southampton
Island, Nunavut (Stenhouse and Robertson 2005). Southampton Island is much larger than Nasaruvaalik Island and can essentially be considered a mainland site, of which the East Bay sanctuary covers approximately 1200 km 2 . For a more detailed description of the East Bay study site, see Stenhouse et al. (2001) and Stenhouse and Robertson (2005).

Capturing and resighting adults
Banding of Sabine's gull adults on Nasaruvaalik Island began in 2007, with additional adults banded in each successive year until 2012. Adults were captured during incubation (starting about late June to early July for ~20˗23 days) using a spring-loaded bow-net (Bub 1991) or a handheld CO 2 -powered net gun (Edwards and Gilchrist 2011). They were banded with a numbered US Fish and Wildlife Service steel band, as well as a unique combination of Darvic color bands (typically three) on their tarsi for easy identification from a distance. In 2008, three birds had a small geolocator attached to their steel leg band, In 2010-2011, some birds (n=33) D r a f t also had a small geolocator device attached on Darvic bands in some years representing ≤ 2% body mass (Davis et al. 2016); these birds had very high return rates (100% for 2008 birds, 92% resighted overall) and we did not observe any obvious, deleterious effects of the tracking equipment.
Subsequent resightings of adults were obtained each year until 2013. Resightings were recorded daily by two or three observers from about mid-June to mid-August during the rest of the breeding season in the year a gull was captured, and in subsequent years, either from viewing blinds (~150 m away from the colony), or by searching for nests on foot. Typically, 4˗5 hr were spent each day (weather permitting) scanning Sabine's gulls around the island to determine their unique color band combination and confirm resightings.
Similar methods were employed at East Bay from 1998-2002. After 2002, however, banding and resighting effort at East Bay has been intermittent (<10 days per season), and following a series of preliminary analyses, data after 2002 were deemed to be too sparse to be used in this study. Therefore, the analysis takes into account the gap (2003)(2004)(2005)(2006) from the end of the Stenhouse and Robertson (2005) study to the beginning of the Nasaruvaalik Island study.

Survival analyses
Apparent survival (ϕ), i.e. true survival confounded by permanent emigration, and encounter (p) probabilities were estimated using traditional capture-mark-recapture (CMR) methods in Program MARK (hereafter 'MARK'; White and Burnham 1999). Encounter histories displaying initial captures and resightings for each bird were generated using the RMark (Laake 2014) package in R 3.2.1 (R Core Team 2014). To account for the time gap between studies in the combined analysis, all years outside the study period for each colony were treated as 'zero' D r a f t within the encounter history for each individual (i.e. every individual received a 'zero' each year from 2003˗2013 for East Bay, and from 1998-2006 for Nasaruvaalik Island; reduced m-array summaries for encounters are provided in Supplementary Material, Tables S1 and S2). All unobserved island-year combinations were assigned to a single dummy ϕ parameter, with the associated p fixed to zero. The dummy ϕ and respective fixed p were not counted as estimable parameters during model selection.
The Cormack-Jolly-Seber model, an open-population model where apparent survival and recapture probabilities vary over time (Lebreton et al. 1992), was used as the global model for Nasaruvaalik Island, as well as the previous East Bay study. A preliminary analysis of the Nasaruvaalik Island data was conducted to determine the best-fitting model structure of encounter probability to be used in the combined analysis. The model structure of resighting probability for East Bay was retained from the best model selected by Stenhouse and Robertson (2005), which was constant recapture rates across years, except for the final year which was known a priori to have a reduced resighting effort. Once the best resighting parameterization was determined, it was denoted as p study for this analysis.
For analysis of both sites combined, the starting model included annual variation in survival (i.e. ϕ time ), a model for which the amount of time-dependent process variation (σ 2 ) in survival probabilities for each colony using variance component analysis can be determined (Burnham et al. 1987). Note that because there was no overlap in time between the two studies, effects of colony and time are confounded.
An information-theoretic approach to model selection was employed, using Akaike's Information Criterion adjusted for small sample size (AIC c ), to choose the best overall model.
Finally, to determine the amount of variation explained by a particular covariate (R 2 ), we used D r a f t model deviances to calculate a fixed effect model based coefficient of determination (equation 6 in Grosbois et al. 2008).

Effects of climate variability on survival
Effects of climate variability on survival were assessed by incorporating broad-scale climate indices as covariates in the models. A comprehensive list of indices can be found through the National Oceanographic and Atmospheric Administration (NOAA; available at http://www.esrl.noaa.gov/psd/data/climateindices/list/). Selection of climate indices was based primarily on recent geolocator (GLS; Global Location Sensing) tracking data showing Sabine's gull migratory routes and wintering areas. The majority (93%, n = 25) of the tagged gulls at Nasaruvaalik Island migrated west towards Alaska, and then followed the Pacific coast south to wintering areas off of Peru (Davis et al. 2016). The remaining 7% travelled through the Atlantic to winter off of South Africa (Davis et al. 2016). Tracking data have not yet been obtained for Sabine's gulls at East Bay (although some birds travelling to Nasaruvaalik Island pass very close to East Bay during spring migration; Davis et al. 2016). However, considering the more southern and eastern location of the East Bay colony, it is possible that they could migrate to Atlantic waters as do a few of the Nasaruvaalik birds and similar to Sabine's gulls breeding in Greenland (Stenhouse et al. 2012). Therefore, we thought it prudent to consider both Pacific and Atlantic indices.
Because multi-collinearity among covariates can interfere with parameter estimation and model selection results (Grosbois et al. 2008), Pearson correlations were evaluated to further reduce the number of indices considered by excluding indices which were strongly correlated (arbitrarily r > 0.35) to multiple other indices in the set. Ultimately, given the large non-breeding D r a f t distribution of Sabine's gulls spanning two ocean basins, three indices were tested (details in Table 1): the principal component-based North Atlantic Oscillation (NAO; Fig. 2A), Tropical/Northern Hemisphere pattern (TNH; Fig. 2B), and Southern Oscillation Index (SOI; Fig. 2C). For each index, winter values were used (Dec-Mar for all except TNH, for which only Dec-Feb data were available). Results are presented ± SE unless otherwise indicated.

Goodness-of-fit
Program UCARE was used to test for violations of Cormack-Jolly-Seber model assumptions (Choquet et al. 2009), which looks for evidence of heterogeneity among individuals (e.g., due to transiency or trap effects; Pradel et al. 1997Pradel et al. , 2005 in recapture and survival probabilities, respectively. The variance inflation factor (ĉ) is a measure of the overall fit of a model and can be used to account for overdispersion. This was estimated for the global model with either the median ĉ method or the parametric bootstrap, using 1000 simulations and taking the ratio of observed to expected deviance (Anderson et al. 1994, White et al. 2001). If ĉ > 1, model selection was based on Quasi-Akaike Information Criterion (QAICc), which assigns further penalties for the number of parameters in a model (Anderson et al. 1994).

Goodness-of-fit
There was evidence of heterogeneity among individuals in recapture or survival probabilities for Nasaruvaalik Island (UCARE combined test: χ 2 13 = 23.5, p = 0.04). Upon examination of individual test components, this was driven by heterogeneity in the timing of bird D r a f t sightings in subsequent years as a function of when they were first captured (TEST2.CT: Z = -2.5, p = 0.01) (trap happiness; Choquet et al. 2009). This effect was entirely driven by a single cell in the final year of the study, so possible trap effects were not pervasive and were addressed with general adjustments for lack of model fit. Median ĉ was estimated at 1.67 for the Nasaruvaalik Island data, and was used to adjust and model selection criteria and parameter estimates. In Stenhouse and Robertson (2005), results from RELEASE (TEST 2 + TEST 3: χ 2 5 < 0.1, p > 0.99) and an estimated ĉ of 0.49 using the bootstrap method both indicated underdispersion (i.e. less variation than expected based on the global model) of the East Bay data. Because the median ĉ method did not perform well for the combined-site analysis, the bootstrap method was used and model selection criteria and parameter estimates were adjusted by a variance inflation factor (ĉ = 1.34) calculated from the global model ϕ time , p study . D r a f t

Combined analysis and effect of climate variability
In the combined analysis, a colony effect on survival was not detected (∆QAIC c > 2; Table 2). Variance component analysis of apparent survival estimates from the global model (ϕ time , study ) indicated that process variance (σ 2 = 0.001, 95% CI: -0.001 to 0.015) accounted for only a small proportion of the total annual variance (process + sampling variance) in the data.
Taking into account sampling error, mean annual survival for both colonies was 0.92 ± 0.02.
In spite of limited evidence of temporal variation, the model with TNH had considerable support and was the best model overall, with ~17.5 times more support than the null model (Table 2). This model describes a negative relationship between TNH and survival (β = -1.08 ± 0.39, 95% CI: -1.84 to -0.32) and explained 33.5% of the variation in survival (Fig. 3). Models including NOA and SOI were not supported (Table 2).

Apparent survival on Nasaruvaalik Island
This study presents the first estimate of apparent survival (~0.90) for a High Arctic colony of breeding Sabine's gulls. Our estimate was in the upper range of those reported for other Arctic-breeding gull species (Stenhouse et al. 2004, Allard et al. 2010 and was similar to that reported for the Low Arctic colony of Sabine's gulls at East Bay (Stenhouse and Robertson 2005). Similar to other colonial nesting seabirds, Sabine's gulls have strong fidelity to their breeding grounds, even if they experienced breeding failure in previous years (Stenhouse and Robertson 2005). Furthermore, the High Arctic colony is relatively small (16˗31 pairs), with nearly all banded birds resighted each year. Therefore, the high annual survival rate is not D r a f t unexpected. We were not able to directly test for a colony effect on survival due to confounding effects of time. Our results indicate that survival was constant over time for the Nasaruvaalik Island colony, but it is difficult to detect temporal variability in short-term survival studies.
Indeed, models with fewer parameters (e.g., constant models) tend to be better-supported (Anderson et al. 1994).
Attachment of tracking devices, such as GLS, may influence survival probabilities and other demographic parameters for seabirds (Igual et al. 2005, Quillfeldt et

Effects of climate on survival
From the analysis combining both colonies, we found evidence of a negative relationship between anomalous winter climate and apparent survival of adult Sabine's gulls. The best supported model suggested that annual survival probabilities were generally high, although they declined non-linearly as TNH moved from negative to positive, and dropped in 2008/09, in relation to a notably high Tropical/Northern Hemisphere pattern in that year. A similar pattern was observed for adult razorbills (Alca torda), which maintained high annual survival probabilities in all years except 1997/98 when Labrador Current SSTs were particularly high (Lavers et al. 2008). Likewise, annual survival of Cassin's auklets (Ptychoramphus aleuticus) D r a f t was lowest during 1997/98, coinciding with an El Niño event in the Pacific (Bertram et al. 2005, Morrison et al. 2011). These patterns suggest, for some long-lived seabirds, that the capacity to maintain consistently high survival probabilities is diminished during pronounced shifts in broadscale weather patterns.
Although adult mortality can be a direct consequence of extreme weather events (Frederiksen et al. 2008;Mallory et al. 2009), effects of regional climate on seabird population dynamics tend to manifest indirectly via changes in food availability (Durant et al. 2004;Sandvik et al. 2005;Gaston 2011;Hovinen et al. 2014). A reliable food source leading up to the breeding season is important for maintaining body condition, and thus influences adult survival and reproductive effort (Monaghan et al. 1989;Oro et al. 2004;Davis et al. 2005;Harding et al. 2011). Sabine's gulls are surface foragers that feed primarily on various zooplankton species (e.g., amphipods; Duffy 1983; Day et al. 2001). Climatic conditions may have negatively influenced Sabine's gull survival indirectly by limiting access to or reducing their primary food source at some point during the non-breeding season (see Vincent et al. 2002;Dorresteijn et al. 2012). Unfortunately, with our data we are unable to pinpoint where or when this might occur.
For example, we found lower Sabine's gull survival with higher TNH indices, opposite to what we predicted, because high TNH usually is associated with cooler sea surface temperature conditions in the Pacific (Barnston et al. 1991), which should mean higher marine productivity (e.g., Hays et al. 2005). However, recent findings suggest that high, positive TNH is associated with stronger development of warm Pacific Blob conditions in the northeastern Pacific (Liang et al. 2017), in the staging and migratory route of Sabine's gulls in our study. The late arrival at the colony and below-average reproductive effort (clutch size) and hatching success in 2009 (Mallory et al. 2012) would be consistent with this interpretation.
D r a f t Spatial variability in climate regimes throughout the range of habitats encountered by long-distance migratory species presents obvious challenges for investigating influences of climate on adult survival (Grosbois et al. 2008 (Davis et al. 2016). Exposure to food supplies, predation, and weather conditions should differ markedly on these routes. More tracking work is needed to determine the proportion of gulls going to each location so we can be more confident in selecting climate indices to test.
Although the model with TNH was best-supported among the given set of models, it is possible that a different index, or a suite of indices and/or local climate variables, better explains variability in Sabine's gull survival estimates. However, testing effects of interactions requires high statistical power that is beyond the present data. We also wanted to avoid finding spurious results by taking an "all possible models" approach (as described by Anderson et al. 2001).
Nevertheless, findings from this study add to growing evidence that variability in climate influences the survival of long-lived seabirds (e.g., Grosbois and Thompson 2005, Sandvik et al. 2005, Lavers et al. 2008, Dorresteijn et al. 2012, Sydeman et al. 2012, Genovart et al. 2013, Hovinen et al. 2014, and highlights the role of global climate in affecting Arctic-breeding species, notably those that are long-distance migrants (Both et al. 2009  D r a f t    349x249mm (300 x 300 DPI)