Methodology:

GPS tags were deployed from Inaccessible Island during brood-guard and from Nightingale Island during incubation, creche and brood-guard. Tags were attached to animals' feathers on the dorsal mid-line of the bird's lower back using overlapping layers of black waterproof tape and cyanoacrylate glue. Some GPS tags were deployed in combination with TDR devices. Birds were captured by hand or using a hand-held net at their nest site (when incubating or in brood-guard) or in the colony when feeding chicks (when in creche).

To reduce stress, only one bird per nest was tagged, with different birds captured during each breeding stage. Handling time was restricted to 15 minutes and birds were stained with temporary animal marker to allow them to be identified for recapture to retrieve loggers.

Loggers were programmed to sample positions every 0.75-3 minutes during brood-guard and creche. During incubation, various sampling rates were used as foraging trips were expected to last up to 3 weeks, and there were concerns about the batteries lasting for the entire trip. These ranged from 1 position every 10 minutes to every 5 minutes for 2 hours per day (between 00:00-01:00 h and 12:00-13:00 h). More information on sampling rates is in included in the pdf file 'Logger_settings.pdf'. All devices only recorded positions when the bird was at the water surface.

On Nightingale, equipped birds were recaptured when seen returning from the sea or at return to the nest. At Inaccessible Island, the tags downloaded data to the base station via a wireless UHF radio link, eliminating the need for tag retrieval.

Data collection:

GPS logger models used: Pathtrack nanoFix GEO on Nightingale, and Pathtrack nanoFix GEO+RT on Inaccessible (PathTrack, 54 x 19 x 15 mm, 24 g).

Data were downloaded and analysed using R (version 3.4.0, R Development Core Team).

Erroneous GPS locations were filtered by the speedfilter function (trip package, Sumner 2015) when speed exceeded 8km/h and duplicate positions removed. When devices recorded consecutive trips, data were separated into foraging trips, which were described by last and first recorded position back at the colony using QGIS (QGIS Development Team). To compensate for irregular gaps between positional fixes within the tracks, filtered data were subsequently processed using a continuous, time-correlated random walk model (crawl package, Johnson et al. 2008). Based on the simulation of 100 possible tracks the most likely or 'best fit'' path used by the penguin is generated.