A re-examination of the timing of pupping for Steller sea lions Eumetopias jubatus breeding on two islands in Alaska

Steller sea lions are distributed from Japan to the California coast, USA, and population demographics vary spatially, with populations in some regions increasing while others are declining. To assess changes in population size, aerial surveys are conducted annually to quantify pup production. The timing of these surveys is critical for accurate population estimates, and survey windows were determined based on historical estimates of mean pupping date. We re assessed the timing of pupping for Steller sea lions at 2 breeding islands in the central Gulf of Alaska, USA (Marmot Island) and the eastern Aleutian Islands (Ugamak Island) for evidence of temporal shift. Using land-based counts of pups, we quantified mean pupping date and the duration of the pupping season between 2003 and 2013 and compared these data to historical mean pupping dates between 1977 and 1999. The mean pupping date of 9 June on Marmot Island was not significantly different than the mean pupping date on Ugamak Island, 8 June. On Marmot Island, mean pupping date differed by 3.7 ± 0.9 d between beaches; however, mean pupping date did not differ between beaches on Ugamak Island. On Ugamak Island, mean pupping date was significantly earlier than previously reported by 2.5 d, but this may be an artifact of the limited number of years available for comparison. On Marmot Island mean pupping date was not different from historical dates. On both islands, 94.2 ± 1.6% of the pups were born prior to the planned start of aerial surveys in Alaska (23 June). Our results demonstrate that although mean pupping date was variable and may have shifted earlier relative to historical data at Ugamak Island, the current timing of the aerial survey is suitable for obtaining peak pup counts for Steller sea lions in these regions.

Understanding the sources of variation in reproductive timing is critical because reproductive events, such as births, are often used as the baseline for other life-history studies (e.g.age-related survival, growth patterns) and for population assessments (Eberhardt et al. 1979, Bradshaw et al. 2000, Hastings et al. 2009).For example, for many species, counts of young are used to assess population size or track populations over time when direct counts of an entire population are unfeasible (Ling 1969, Eberhardt et al. 1979, Berkson & DeMaster 1985).In pinnipeds (seals, sea lions, and walrus), newborn pups spend a period of time on land prior to entering the water, making them an ideal population segment to census (Berkson & DeMaster 1985).If reproductive timing varies, then estimates of production can be falsely reduced or inflated, impacting measures of population size, which are vital for demographic studies and for development of successful management and conservation strategies.
Because their breeding locations are remote, distant from each other, and span a large geographic region, population counts of Steller sea lions Eumetopias jubatus in US waters are primarily conducted via aerial surveys and supplemented with some landor boat-based counts (Westlake et al. 1997, Snyder & Pitcher 2001, Fritz et al. 2016).Accurate indices of population size are critical for this species because population trends vary substantially across the species' range (Johnson & Fritz 2014).Steller sea lions are widely distributed throughout the Pacific Rim from Japan to California, USA (Kenyon & Rice 1961, Loughlin et al. 1984, NMFS 2008).The population is separated into 2 distinct population segments (DPS), eastern and western, based on genetic and demographic differences between regions, which are divided at 144°W longitude (Bickham et al. 1996, Loughlin 1997).
In 1990, after large declines in Steller sea lion numbers, the species was listed as threatened range-wide under the US Endangered Species Act (ESA).Following continued population declines in the western DPS, in 1997 the listing was changed to endangered for the western DPS (National Oceanic and Atmospheric Administration 1997).The eastern DPS has increased at a rate of approximately 3% yr −1 since the 1970s (Pitcher et al. 2007, Fritz et al. 2016) and was removed from the list of ESA-threatened species in 2013 (NMFS 2013).The decline in the western DPS in Alaska likely ended in 2003 and between 2003 and 2015, counts of both pups and non-pups (1 yr and older) have increased in most areas (Johnson & Fritz 2014, Fritz et al. 2016).Nonetheless, within the western DPS, sea lion numbers remain well below historic levels and regional variability in demographic trends remain significant (Johnson & Fritz 2014).
For Steller sea lion pup production estimates, aerial surveys in Alaska are generally conducted during a 3 wk period between late June and mid-July when most pups are ≤1 mo old (Pitcher et al. 2001, Fritz et al. 2016).The aerial survey window was determined based on historical measures of mean pupping date (Pitcher et al. 2001) with the goal of conducting counts near the peak of pupping (defined as the date of maximum pup counts) but before most pups begin entering the water (Gentry 1970, Sandegren 1970, Pitcher et al. 2001).The current timing of Steller sea lion aerial surveys in Alaska (preferred window of 23 June to 10 July) assumes that the timing of pupping has not changed since 1999 when it was last examined for Steller sea lions breeding in the USA (Pitcher et al. 2001).
The objective of this study was to reassess mean pupping date and determine if there is evidence for variability in the timing of pupping for Steller sea lions.We quantified mean pupping date at 2 breeding islands in the central Gulf of Alaska (Marmot Island) and the eastern Aleutian Islands (Ugamak Island) over 10 yr between 2003 and 2013.In addition, we assessed whether the current timing of aerial surveys occurs near the date of peak pupping, which is essential for accurate estimates of pup production.

Data collection
Data were collected between 2003 and 2013 by land-based observers at Marmot Island (Kodiak archipelago, Alaska, USA; 58°13.6'N, 151°47.8'W) and Ugamak Island (western side of Unimak Pass, Alaska; 54°13.5'N, 164°47.5'W; Fig. 1A, Table A1 in the Appendix).In 2006, the field season was suspended for the month of June on Marmot Island and canceled on Ugamak Island because of a research injunction, which resulted in insufficient data to model mean pupping date in that year.Between May/June and August each year, counts were conducted daily, weather permitting, at Beaches 4 (MB4) and 7 (MB7) on Marmot Island (Fig. 1B) and at 'Beach South' (UBS) on Ugamak Island (Fig. 1C).A second beach on Ugamak Island, 'Beach North' (UBN, Fig. 1C), was observed intermittently with a goal of at least one visitation per week.A detailed description of the data collection procedures can be found in Chumbley et al. (1997) and Wilson et al. (2012).
From cliff edges approximately 200 to 300 m above the rookeries, at least one count of sea lions by age-sex classes was conducted per day between 10:00 and 18:00 h (local time) using binoculars or spotting scopes.Pups were also counted hourly during a weekly dawn-to-dusk survey.Each count was assigned a confidence category based on the observers' assessment of completeness: 'complete' (ob -server was confident all sea lions present were counted); 'partial' (all visible sea lions were counted but some sea lions may not have been visible from the observation site); and 'complete partial' (a special category for UBN where the entire beach area cannot be observed from the counting location, indicating that every pup in the viewable region was counted).For partial counts, observer comments were used to identify and remove counts that were incomplete (e.g.'large proportion of pups in the surf not counted' and 'fog moved in; low confidence in total') in order to ensure anomalously low counts did not influence the model parameters.'Estimate' counts and any other counts where observer comments signified potentially poor data were not included in data analysis.When both count types occurred within a year at a beach, they were generally distributed throughout the survey period (e.g. in 2012, as shown in Fig. 2 below).Because of pup movements into and out of visible areas (or the water), partial counts were equal to or higher than complete counts within the same day on multiple occasions.In these cases, the highest total count was used for that day.

Pup count growth model and statistical analysis
A logistic growth model (Trites 1992, Pitcher et al. 2001) was fit to daily pup counts for each of the 4 beaches separately (MB4, MB7, UBN and UBS; Fig. 2).When modeling counts at each beach, annual variation in logistic parameters was accommodated with a random effect for each year.For beach i in year j, observed cumulative pup production on day d was modeled as follows: where N d is the total number of pups observed on day d, φ 1ij is the pup production asymptote, φ 2ij is the mean date of birth, φ 3ij is a scale parameter controlling birth synchrony, and ∈ ijd is an error term with N (0,σ 2 i ) distribution.For the φ 1 and φ 3 parameters we used the following parameterization for k = 1, 3; log φ kij = β ki + α kij , where β ki is the parameter intercept and α kij is a N (0,τ 2 ki ) random effect.The φ 2 parameters were modeled in the same fashion without the log transformation.To account for leap years, date was based on days since 1 May (i.e. 1 May = 0 and 2 May = 1).All calendar dates presented are calculated from a common year.Only pup counts prior to 4 July (MB4, MB7, and UBS) were used for analysis due to potential dispersal of females with pups and reduced sightability of pups as they mature and begin swimming (Raum-Suryan et al. 2004, Marine Mammal Laboratory unpubl. data).A longer survey period was required at MB4 in 2004 (until 8 July) due to a late season sampling gap and at UBN in all years (until 9 July) due to the reduced sampling effort and greater time gaps between samples.Mean pupping date (MPD) was defined as the date when pup count estimates reached one-half of the maximum pup count in the model (i.e. the asymptote).As in previous analyses for Steller sea lions (Pitcher et al. 2001), no adjustments were made for pup mortality, which resulted in MPD estimates being negatively biased (Trites 1992).For example, for northern fur seals Callorhinus ursinus, MPD was biased between 0.1 and 2.0 d earlier when pup mortality was not taken into account (Trites 1992, Kunisch 2011).To determine if the timing of pupping has changed since 1999, estimates of MPD were compared to those from Pitcher et al. (2001) for Marmot and Ugamak islands.The date by which 90% of the pups were born was a derived parameter from the model which was determined by calculating 90% of the asymptote pup count and identifying the date this value was reached.Similarly, the proportion of pups born prior to the start of the planned aerial survey (23 June) was a derived model parameter determined by calculating the number of pups born on 22 June and dividing that value by the maximum pup count from the model.Statistical analysis was conducted using R (R Core Team 2015).Parameters and random effects were estimated using Markov chain Monte Carlo (MCMC) via the 'rstan' package (Stan Development Team 2015).The MCMC was run for 30 000 iterations following a burn-in of 10 000 iterations which were discarded.The posterior mean and 95% credible intervals (CIs) were calculated for all parameters and derived parameters (e.g.island averages or differences between island parameters).Island MPD was calculated by averaging MPD for beaches at each island.No weighting factor was used because maximum pup count values were similar between beaches within an island.
To compare current MPD with historical data from Pitcher et al. ( 2001), we assumed a normal posterior distribution around the published MPD values and sampled 20 000 values in order to make comparisons between current and historical posterior distributions.The resulting yearly historical values were averaged for each island and compared to the current posterior distribution for MPD by island.Historical MPD at each island was considered significantly different from current MPD if the resulting 95% CIs of the posterior differences did not include zero.Summary data for MPD, date of 90% pupping, and proportion of pups born by the planned start of the aerial survey are reported as means ± posterior standard deviations (SD).

Data collection
The earliest start date for pup count surveys was 24 May in 2013 and the latest end date was 1 August in 2007 and 2011 (Table A1).The total number of daily pup counts used in the model and the proportion of complete counts varied among years, islands, and beaches (Table A1).At MB7, a late start in 2005 (first count on Day 41) and too few counts in 2003 and 2004 resulted in a poor model fit.At UBN, insufficient sampling, late starts, or large time gaps (up to 8 d) resulted in poor model fits for 2003, 2004, 2007, 2008, and 2013.As a result, these years were removed from the analysis.

Mean pupping date
MPD was not significantly different between Marmot (9 June) and Ugamak (8 June) islands when all   1.Model estimates for timing of pupping of Steller sea lions on 4 beaches in Alaska (MB4 and MB7 on Marmot Island, and UBS and UBN on Ugamak Island; see Fig. 1) between 2003 and 2013: mean pupping date (MPD), date of 90% pup production, the proportion of pups born by the planned start of the aerial survey on 23 June (Day 53), and maximum pup production estimated from the model (asymptotes).Estimates are shown as mean ± posterior SD.Dates are shown as number of days since 1 May in each year, with corresponding calendar dates determined based on a common year.Beach and island averages were calculated for all years pooled.*Significant difference between beaches within a year or, for pooled data, between beach averages on an island significantly later at MB7 than MB4 in 6 years with the largest differences occurring in 2010 (5.1 ± 0.8 d later;  2001).

Pupping season and aerial survey schedule
In all years, 90% of the pups were born by Day 49.8 ± 1.2 (19 June).This date was not significantly different between islands (Table 1; difference 0.9 d, CI −2.3 to 4.1 d) or between beaches on Ugamak Island (Table 1; difference 3.8 d, CI −1.6 to 9.2 d).The date of 90% pupping was significantly later at MB7 than at MB4 (Table 1; difference −4.1 d, CI −7.4 to −0.8 d), which is a reflection of the differences in MPD at these beaches.On Ugamak Island, annual estimates of the 90% pupping date differed only in 2012, the year when MPD at UBN was also later (Table 1).
Prior to the planned start of the aerial survey (23 June), the proportion of pups born at both islands was 0.942 ± 0.02, and this did not vary between island (Table 1; difference −0.01, CI −0.06 to 0.04).When beaches were considered separately, the proportion of pups born prior to the survey did not differ (Marmot difference 0.05, CI −0.003 to 0.10; Ugamak difference −0.05, CI: −0.12 to 0.02).For both islands, the proportion of pups born prior to the start of the survey ranged from 0.89 ± 0.02 (Marmot 2010) to 0.97 ± 0.005 (Marmot 2013) among years.

DISCUSSION
By conducting extensive observational research over a 10 yr period, data collected during this study were used to examine the timing of reproduction in a pinniped species that breeds in remote, hard-toaccess locations.The land-based counts of pups born at Marmot and Ugamak islands should be considered an index of pup production at these locations and not a census of total pup production (Merrick et al. 1988, Chumbley et al. 1997, Kirkwood et al. 2005).Factors that impact the visibility (e.g.fog, rain, heat haze) and the ability to sight pups (e.g.observation height, topography) vary significantly at each beach (Merrick et al. 1988, Chumbley et al. 1997), which likely explains the differing ratios of complete to partial counts (Table A1).Nevertheless, for this study it was not necessary to get complete counts of the pups born at each beach, only consistent counts of pups within the sampled regions to detect changes in pup numbers over the reproductive period.
Overall, we found that MPD did not differ between islands during the study period; however, there was some interannual variability.During 3 consecutive years of this study (2009 to 2011), sea lions at Marmot Island showed a pattern of slightly later pupping by 1.4 ± 0.7 d to 2.6 ± 0.6 d (95% CI: 1.4 to 3.8, 0.09 to 2.8 and 0.2 to 2.5 in 2009, 2010 and 2011 respectively).Breeding synchrony is the general pattern for many pinnipeds, though for species with wide-spread distributions, geographic or latitudinal variation has been found in the timing of reproduction (Boyd 1991, Wickens & York 1997).Marmot and Ugamak islands are near the center of the Steller sea lions large geographic range and are separated by less than 1000 km (Fig. 1), making geographic or latitudinal variation in MPD unlikely.Historically, a difference in MPD was not found between Marmot and Ugamak islands (Pitcher et al. 2001).The cause of the interannual variation between islands in this study is un known; however, for many pinniped species, including Antarctic fur seals Arctocephalus gazella and South American sea lions, a strong relationship has been found between annual pupping date and prey availability (Boyd 1996, Soto et al. 2004).
In contrast to the inter-island comparisons for MPD, when all years were combined there was a significant difference between beaches for MPD on Marmot Island.This difference was greater than 3 d, yet no difference was found between beaches on Ugamak Island.A similar pattern of later pupping by sea lions at MB7 was also described by Chumbley et al. (1997), but statistical comparisons were not made.At such a small geographic scale (< 5 km), it is unlikely that the later pupping date on MB7 could result from dissimilarities in prey availability.However, other factors that may influence pupping beach selection could play a role in the variation in pupping dates.For some pinniped species, older or prime-aged females give birth earlier in the season than younger females (Lunn & Boyd 1993, Gentry 1998, Boltnev & York 2001, Maniscalco et al. 2006, Hastings & Jemison 2015).This was reported for Steller sea lions that breed on Chiswell Island, Alaska, and the Forrester Island Complex, Alaska (Maniscalco et al. 2006, Hastings & Jemison 2015).For Antarctic fur seals it was suggested that, by returning earlier, older females could select the most suitable pupping sites (Lunn & Boyd 1993).If beach MB4 on Marmot Island is the most suitable site for pupping, it could potentially be occupied first by prime-aged females, leaving MB7 for young sea lions that pup at a later date.
Movements of mother−pup pairs between beaches could also impact our calculation of MPD leading to differences between beaches at Marmot Island.Steller sea lion pups can enter the water as early as 2 wk of age and disperse from a breeding beach with their mothers at 2 mo of age (Sandegren 1970, Merrick et al. 1988, Raum-Suryan et al. 2004).On Marmot Island, observations of permanently marked sea lions suggest that as the reproductive season progresses some sea lions move from MB4 to MB7 (Chumbley et al. 1997, Marine Mammal Laboratory unpubl. data).In addition, Chumbley et al. (1997) described movement of mother−pup pairs away from MB4 starting in mid-July and, in 1991, a late season survey (July 31) found that MB4 was almost completely abandoned, whereas MB7 counts remained unchanged (Chumbley et al. 1997).If female sea lions from MB4 or other beaches moved their pup to MB7 during the count period, this would result in an increase in daily pup counts unrelated to births at that beach, which would push the estimated MPD later.Similar small-scale movements by mother-pup sea lions pairs during the breeding season have been described for Steller sea lions at the Forrester Island Complex, Alaska (Hastings & Jemison 2015).
Based on comparisons with historical data (1977 to 1999), it appears that there has been no change for MPD at Marmot Island but a potential shift to earlier pupping at Ugamak Island, i.e. 3 d earlier than recorded by Pitcher et al. (2001).Unfortunately, only 3 yr of data were available for comparison from Ugamak Island (Pitcher et al. 2001).Given the interannual variability in MPD at Ugamak Island, which ranged over 1.5 d between 2003 and 2013, it is difficult to discern if there has been an actual shift in mean pupping over time or if our results are an artifact of the small historical sample size (3 yr) used for comparison.If a change in pupping date has occurred, with sea lions on Ugamak Island pupping earlier, it is important to consider whether this small shift is biologically important.The pupping season for Steller sea lions generally occurs over a 2 mo period (mid-May through mid-July) (Calkins & Pitcher 1982, Merrick et al. 1988) and Merrick et al. (1988) estimated that at least 90% of pups are born by the last week of June.This pattern is reflected in our data as pups were counted on the earliest obser-vation date (24 May 2013), and in all cases the date by which 90% of the pups were born occurred prior to the end of June (i.e.Day 60, Table 1).Therefore, although variability may exist in MPD between beaches, occasionally between islands, and even between our study and historical MPDs, there does not appear to be a significant shift in the overall pupping season for Steller sea lions at these 2 islands.
The existing variability we found in MPD, both temporal and spatial, could have an influence on estimates of population trends determined via aerial surveys (Calkins & Pitcher 1982, Fritz et al. 2008, 2013).However, when all beaches were combined, 94% of pups were born prior to the survey start (23 June) and this did not differ between islands or beaches within an island.Based on the beach maximum count estimates (asymptote values, Table 1), this results in the potential for on average only 13 to 18 pups to be born at each beach after the survey starts.In addition, the survey takes several weeks to complete, which would result in an even higher proportion of pups born as the survey progresses (Fritz et al. 2008(Fritz et al. , 2016)).Because the aerial surveys are scheduled to balance the trade-off between waiting for pups to be born and the increase in time pups spend in the water as they age, our results suggest that aerial surveys are still suitably timed to ensure the majority of pups are born when surveys are conducted.

CONCLUSIONS AND CONSERVATION IMPLICATIONS
Monitoring changes in the timing of reproduction is critical for population assessment and management of Steller sea lion stocks.We found that MPD is consistent between the 2 study islands, with some interannual variability.In addition, there may have been a slight shift to an earlier MPD from historical data at Ugamak Island but this is based on only 3 historical years for comparison.As a result, the current timing of the aerial survey is suitable to monitor changes in Steller sea lion pup production in these regions.It may be necessary to examine the timing of Steller sea lion births in other parts of Alaska (e.g.Central and Western Aleutian islands) to ensure that aerial surveys are properly timed in these regions.However, given that Pitcher et al. ( 2001) also re ported a MPD of 9 June on Medny Island (Commander Islands, Russia), we do not expect the timing of births in the rest of the Aleutian Islands between Ugamak and Medny islands to be significantly different.
The counts of Steller sea lions obtained during aerial surveys are considered a vital part of the continuing research guided by the Steller sea lion recovery plan (NMFS 2008).In addition, the recovery plan calls for focused research on sea lion population growth and reproduction to assess the relative impacts of threats to Steller sea lion population recovery (NMFS 2008).By identifying population level, multi-year patterns in MPD, researchers can now start to examine factors related to interannual and individual variation in this vital rate, such as impacts of environmental change, prey availability, maternal age, and reproductive history (e.g.Lunn et al. 1994, Bowen et al. 2003, Soto et al. 2004).An understanding of these relationships may help reduce the 'substantial uncertainty' that affects the ability to understand population-level threats re sponsible for the lack of recovery and, in some cases, continued regional declines within this population (NMFS 2008).
Fig. 1. (A) Steller sea lion research sites (circles) on Marmot and Ugamak islands (Alaska, USA).Marmot Island is located 45 km northeast of Kodiak Island in the Central Gulf of Alaska; Ugamak Island is on the western side of Unimak Pass in the Eastern Aleutian Islands.(B,C) Black lines show breeding beaches on each island: (B) Beach 4 (MB4) and Beach 7 (MB7) on Marmot Island; (C) Beach North (UBN) and Beach South (UBS) on Ugamak Island

216Fig. 2 .
Fig. 2. Example of the results of the logistic growth model for Steller sea lion pup production, based on counts in 2012, at Beach 4, Marmot Island (light gray) and Beach S, Ugamak Island (dark gray) (Alaska, USA).Shaded areas encompass the 95% CIs; points along the curve are live pup counts.Counts are separated between 'complete' and 'partial' counts' to show the variability in count types between locations (TableA1).Vertical dashed lines denote mean pupping dates: 8 June at Beach 4 (light gray); 7 June at Beach S (dark gray).The planned start of the aerial survey, June 23 (Day 53), is marked with an arrow

Table
Table 1; difference 1.3 d, CI −0.3 to 3.1 d).Differences among yearly estimates of MPD ranged between −0.4 and 2.6 d later at Marmot Island, with MPDs at Marmot Island significantly later than those at Ugamak Island in only 3 years(2009, 2010 and 2011).When beaches were examined separately, MPD was 3.7 ± 0.9 d later at MB7 than MB4 (Table1; CI −5.5 to −2.1 d), but there were no differences between beaches on Ugamak Island.On Marmot Island, annual estimates of MPD were