Drivers and fitness consequences of dispersive migration in a pelagic seabird

Lay Summary Sex segregation, competition and differences in individual quality may drive dispersive migration in birds and affect their fitness. Atlantic puffins tracked for up to 6 years followed remarkably different migration routes, but individuals followed the same route every year. Although random dispersion and sex segregation could not explain the patterns observed, birds visiting the Mediterranean Sea foraged more and had a higher breeding success than birds remaining locally or visiting the Atlantic Ocean.


Energy costs
We used a model developed on murres from (Elliott et al., 2013)  sleeping (tucked leg) as a proxy for resting time (there is evidence that resting time at the colony is similar to resting on the surface (Elliott & Gaston, 2014)). We then converted our results to kJ/day and use the allometric equation developed for auks in (S. A. Shaffer, 2011) to adjust the results to a 370g puffin.

Laying date
Incubation generally lasts around 44 days (Harris and Wanless 2011) and is shared by parents alternating shifts while the other forages. Because of the difficulty of intensive direct observation in this subterranean nesting, easily disturbed species, we estimated laying date indirectly using saltwater immersion data to detect the start of incubation. For those pairs in which we tracked both birds, lay date was identified as the start of the first continuous 6 h daytime period in which one or other bird's immersion logger was dry (i.e. combined activity showed consistently that at least one parent was not on the water). This method also allowed us to detect failure of the first egg followed by relaying ( Figure S1). When only one bird of a pair was tracked, we estimated lay date as the start of the bird's first incubation shift. The accuracy of this method was verified using a subset of 5 nests which were checked daily with a burrowscope (Sextant Technology Ltd.) in 2012-2013 to determine Page | 3 precise laying date; its accuracy was ± 1.8 days. We calculated the birds' post-migration laying date for 89 of the 111 tracks in our dataset.
Figure S1 -Combined immersion activity for an example pair during the pre-laying and incubation period of 2013. Lay dates (defined as the start of a long (>6h) dry period) are in blue. The first egg was lost to predation the day after it was laid (pers. observation), the 2 nd egg successfully hatched. The end of incubation is indicated with a red line: in the first case after predation of the first egg, the second (hatching) 44 days after relaying (in this case presence of the chick was verified with a burrowscope). During these 44 days there was always an adult present on the colony, except for short bouts of maximum 1min.

Breeding success
We did not check nests directly during the 6-week chick-rearing period following incubation to minimise disturbance, except after 2012 when a burrowscope was available. Nests which failed during incubation (unsuccessful incubation, egg damaged or infertile) were identified using the same immersion data technique used to identify laying date: nests with an incubation period unusually short or long, i.e. outside of a 14-day window around the average 47 days estimated from our dataset, were deemed unsuccessful.
Page | 4 For the nests which manage to hatch a chick, we used a proxy for breeding success: the ability to hatch and rear a chick for at least 15 days (mortality is highest during the first few weeks; (Harris and Wanless 2011)). This was estimated by direct observations of the parents breeding beakfuls of sandeels to their burrow. During the chick-rearing season (late May to early July) we regularly observed burrows at dawn or dusk when feeding is most likely to occur (observation took place for a few hours several times a week during this whole period, each year). The birds were deemed to have a successful season if they were seen provisioning their chick at least 15 days apart (i.e. their chick reached at least 15 days, this is the lower threshold used in the standard method for this colony: Skomer Island seabird monitoring reports 2008-2014). Although 15 days was our cut-off value, in the most cases birds could be observed bring food to their chick for longer periods.

Sexing
We sexed 20 birds using DNA extracted from 5 breast feathers plucked during device retrieval (Avian Biotech, UK). 7 additional birds which could not be recaptured in 2014 were sexed behaviourally, using a conservative combination of at least 2 of 3 different measures. First, we compared bill size between partners, the main method used to sex puffins before DNA methods; females usually have a smaller beak than their partner (Corkhill 1972;Friars and Diamond 2011). Second we used repeated observations (conducted for 1-2h at least 1-2 times a week during the 2007-2013 breeding seasons) of typical male-behaviour at the colony (e.g. head-flicking display (Taylor 2011) and propensity to fight (Creelman and Storey 1991)) or copulation. Finally, when both members of a pair were tracked with a geolocator, we used the technique used to measure laying date to identify the bird which took the first incubation, which tends to be male in many other burrow-nesting seabirds (Warham 1990;Pinet et al. 2012). We used the DNA-sexed birds to validate these 3 methods. The accuracy obtained were 100% for the bill-size method (validated on 7 pairs), 100% for behavioural observations (validated on 10 pairs), and 100% for the first incubation shift method (validated in 6 pairs). However, to be conservative we only included the 7 birds for which we had results from 2 or 3 Page | 5 methods in our analysis of sex-differences (7 other birds for which we only had one of the 3 measurements were excluded). In one pair (which we did not include in our analysis), one of the three methods was in disagreement with the two others, which shows that none of these methods apart from DNA sexing are 100% reliable (despite our 100% validation), which is we only included birds for which we had 2 or 3 (agreeing) measures. In total we sexed 27 birds (13 males and 14 females), including 20 with DNA methods and 7 with behavioural methods.

Migration phenology -randomisation tests
We used randomisation tests to test whether the timings of major movements were more similar between birds than expected by chance. We defined major movements as entering and leaving the mid-west Atlantic (crossing the -20° meridian), and entering the Mediterranean Sea (crossing the Strait of Gibraltar). Because of the proximity to the spring equinox we could not collect data on leaving the Mediterranean Sea. For each bird the date of interest was extracted (e.g. entering the Atlantic) as well as the duration of stay in the Atlantic or in the Med (assumed in this case to be until the 28th February). A date was randomly sampled from any possible date during the 31 st July -28 th Feb window which would allow the bird to remain in the Atlantic or Mediterranean Sea for the same duration as its actual stay. For example, if a bird stayed for 100 days in the Atlantic, a date as randomly sampled from the 31 st July-20 th November window (20 th November = 28 th Feb -100 days).
We restricted the full window of dates to the 31 st July -28 th Feb, which corresponds to ~ 2 weeks after the end of breeding and before return to the colony, to allow for leaving/returning to the colony. This was done for all birds visiting the Atlantic and/or the Mediterranean Sea and the range of dates obtained was compared to the actual range. The process was repeated 10,000 times and a P-value calculated. If P < 0.05, the event was considered significantly more constrained than expected by chance.
Page   Table S2. Test results investigating differences between route types in proportions of time spent flying, foraging and sitting, and daily energy expenditure, for each month of the migration period, as presented on