Introduction

One of the major characteristics of birds is the widespread phenomenon of migration (Newton 2008). Migratory behaviour of birds has been studied intensively over many decades, with a variety of tools employed such as visual observations (Gatter 2000; LWVT/SOVON 2002), marking by rings (Wernham et al. 2002), systematic catches within standardised programmes (Berthold and Schlenker 1975; Hüppop and Hüppop 2011), physiological measurements in the field (Bairlein 1991; Atkinson et al. 2007) and radar studies (Lack and Varley 1945; Bruderer 2003; Gauthreaux and Belser 2003). However, the application of satellite telemetry, and more recently also data loggers, has revolutionised our understanding of bird migration as has been shown for bird taxa such as geese (Green et al. 2002), swans (Pennycuick et al. 1999), raptors (Ueta et al. 2000), shearwaters (Shaffer et al. 2006) and very recently small landbirds (Bächler et al. 2010) and even songbirds (Stutchbury et al. 2009).

A central question in studies of migratory behaviour is why some populations or individuals of the same species migrate, while others tend to stay during winter rather close to their breeding sites (Berthold 1996; Griswold et al. 2010). Migration itself involves an additional energetic cost of travelling, as well as likely increased risks of mortality during the journey, and so it is presumed that for migration behaviour to evolve, these costs are exceeded by benefits of moving to more favourable regions. In contrast, staying near the breeding place in winter will reduce migration costs but may involve higher costs for a wide variety of reasons, especially because of reduced food availability, and also adverse climatic conditions at higher latitudes in winter (Newton 2008).

While observing migrating birds over land is very challenging even if using radar technology, this is even more difficult at sea when birds are out of sight or reach. Thus, our understanding of the migration of seabirds is mainly based on ring recoveries or on systematic watches of migration along the coast (Camphuysen and van Dijk 1983; Meltofte and Faldborg 1987) and on very few offshore islands or platforms (Hüppop et al. 2010). While this so-called seawatching has advanced our understanding of the timing of migration and migration intensity, major gaps still exist as such studies are rare and confined to days with good weather, we do not know much from beyond the coast, and we lack information on individual behaviour. This last point is essential for understanding ecological and evolutionary processes.

The northern gannet (Morus bassanus) is a seabird that visits a wide spectrum of winter regions from close to their breeding sites (e.g. in the North Sea for eastern Scottish populations) to several hundreds of kilometres south, as has been demonstrated by ring recoveries (Nelson 2002). A recent study using data loggers has improved our knowledge of gannet migratory and wintering behaviour substantially; Kubetzki et al. (2009) showed that individual northern gannets from the Bass Rock, east Scotland, exhibit considerably different migration strategies. While some birds stayed in the North Sea, that is, relatively close to their breeding site, others flew as far south as the Atlantic shelf of NW Africa, a direct distance of 3,200–4,700 km from their breeding colony. A high proportion of birds employed an intermediate strategy and wintered in the Bay of Biscay. Gannets from other European colonies also show extensive variation in winter home ranges of individual birds as well as some evidence of differences among colonies (Fort et al. 2012).

Deploying data loggers on gannets permits, besides information on the geographic position of the bird, measurements of the amounts of time spent flying each day, and of the temperature of the water in which birds are resting. We take flight time as a proxy for migration and foraging effort; thus, these data may shed light on costs and benefits of wintering in different regions. Although birds were wintering along a gradient from north to south rather than in distinct areas only, we grouped their main wintering grounds into three regions. While hypothesising that the net benefit of each wintering area is equal, we analyse our data in relation to this hypothesis and discuss possible advantages and disadvantages for northern gannets selecting different wintering locations.

Materials and methods

Study site and field work

Chick-rearing adult northern gannets on the Bass Rock, Firth of Forth, east Scotland (56.078°N, 2.639°W) were equipped with geolocation loggers over three winter seasons. Fifteen devices were deployed in August 2002 and 26 devices on different individuals in August 2003. Thirteen were retrieved in April and May 2003 (first season; recapture rate = 87 %) and 21 in April and May 2004 (second season; recapture rate = 81 %). For more details, see Kubetzki et al. (2009). Not all of the 34 loggers recovered worked throughout the whole period of deployment. Combined for both years, 65 % of the devices worked until 30 November, 32 % until 31 January and 18 % until 31 March. Logger failure was found to be a result of excessive power drain by one component and condensation within the logger from battery discharge. So there is no reason to think that logger failure related in any way to the location or behaviour of individual birds on which the loggers were deployed. We deployed 30 new loggers in August 2008. We retrieved 25 loggers in the subsequent spring and summer (recapture rate = 83 %) of which 23 loggers had full data sets (light plus temperature) for the winter 2008–2009.

For almost all birds recaptured after the second season of deployment, and in 2009, sex was determined from blood samples using standard molecular methods (Griffiths et al. 1998). The work conformed to the legal requirements of the UK and was carried out under the BTO permit 2282 and Home Office licence.

Data loggers

The geolocation data loggers (GeoLT; Earth & Ocean Technologies, Kiel, Germany) were housed in a pressure-tight seawater-resistant casing (diameter, 14 mm; length, 45 mm; 8.2 g, equivalent to ca. 0.3 % of the bird body mass) and attached to a custom-built leg band. The main sensor of the device was a light sensor, allowing geographic position to be calculated from day length and time of local midday and midnight (Wilson et al. 1992, Kubetzki et al. 2009). In addition, the GeoLT recorded ambient temperatures (i.e. air or sea surface temperature, depending on its position) every 120 s throughout deployment. This was done by a temperature sensor with a measuring range from 0 to 32 °C, a resolution of 0.125 K, an accuracy of 0.2 K and a 90 % response time of <4 min. The loggers deployed in 2008 were slightly shorter (length 38 m) and had more reliable batteries, but they recorded essentially the same data.

Data analysis

Geographic positions were calculated from light curves using MultiTrace Geolocation from Jensen Software Systems (Laboe, Germany). For more details, see Kubetzki et al. (2009). Since our focus was on migration and wintering areas, analysis was restricted to the period from colony departure to colony return. Colony attendance was derived from logger temperature profiles, since they differ between birds on land and those in air or on water (for details see Wilson et al. 1995; Garthe et al. 2003). The activity of the bird at sea (swimming or flying) could be deduced from records of thermal fluctuations (following Wilson et al. 1995; Garthe et al. 1999). When the temperature remained constant and relatively low, the bird was considered to be floating (i.e. resting, but this could also include periods when birds were actively swimming) on the water surface with the logger submerged and indicating sea surface temperature (SST). For convenience, this behaviour is called ‘swimming’ throughout the rest of the manuscript. When the temperature record varied slightly within a higher range and within the range recorded for air temperature at that latitude, the bird was considered to be flying. Flight activity was analysed for full-days (24 h). As analyses were time-consuming, only a subsample of more or less equally spaced days were selected for analysis, but with a focus to fully cover the respective periods (migration and wintering) in 2002–2003 and 2003–2004. For the 2008–2009 data set, two core periods for migration and wintering were analysed.

For loggers deployed in 2003 and 2004, percentage of time spent flying was compared between 8 birds migrating quickly and directly to their wintering areas in W Africa and 4 birds that stayed in the North Sea. Data for the African birds were analysed for every second or third day when they were actively on their southward migration, as derived from the positional analyses (Kubetzki et al. 2009). Comparative data for the resident North Sea birds were taken from the main gannet migration period defined by the birds travelling to Africa, 15–31 October, for every other day. For each individual of both groups of birds, mostly 8–9 days of activity were analysed, depending on the extent of the migratory period and on the duration of the working periods of the data loggers. A similar analysis was carried out for loggers deployed in 2008, but using every day from 21 to 30 October—within the migration window—so that 10 days of data were analysed, for 11 birds that eventually wintered off West Africa, 6 birds that wintered between the English Channel and the Bay of Biscay (since in 2008–2009 no birds remained in winter within the North Sea), and a further 4 birds that wintered off Portugal and Gibraltar.

Percentage of time spent flying was also compared between birds in their winter region. For loggers deployed in 2003 and 2004, data were analysed for 9 days (ca. every third day) from 2 December to 30 December, except for shorter periods when the logger ceased working. Sample sizes comprised 4 individuals for the North Sea/English Channel, 5 individuals for the Bay of Biscay/Celtic Sea and 10 for the waters off West Africa. For loggers deployed in 2008, data were analysed for each day from 1 to 10 January since some birds were still moving south as late as early December in 2008 and because data could be recorded throughout the whole equipment period. We compared 11 birds that eventually wintered off West Africa, and 6 birds that wintered between the English Channel and the Bay of Biscay (since in 2008–2009 no birds remained in winter within the North Sea).

Sea surface temperatures (SSTs) were measured in the middle of the inactive period of the bird overnight. Temperature signals were stable overnight except for some short periods when the birds were flying. Those extended swimming periods gave best results as the temperature sensor in the device was fully adapted to the SST. For loggers deployed in 2003 and 2004, data were analysed for each fourth day from 1 to 31 December, except for shorter periods when the logger ceased working. For loggers deployed in 2008, SST was measured in the same way for each night from 1 to 10 January 2009.

Thermal conductance

There are no published data on the thermoregulatory costs of swimming gannets, and a complex thermal modelling including heat losses and gains by conduction, convection and radiation (Walsberg 1988; Fort et al. 2009) is beyond the scope of this article. To get at least an estimate, the thermal conductance (C) was measured on two carcasses by the warming constant technique (Morrison and Tietz 1957; de Vries and van Eerden 1995). C’s obtained by cooling curve analysis compare reasonably well over a large range of body masses with C’s that were mainly obtained by gas exchange experiments (de Vries and van Eerden 1995). Both gannets were beached, but their plumage was in good condition. Nevertheless, their low body mass indicated that they were emaciated (1,694 and 1,860 g, respectively, vs. 2,300–3,600 g in healthy gannets, Nelson 2002). The carcasses were stored deep-frozen in plastic bags until the measurements began. After thawing, the carcasses were kept in a room with constant temperature (about 1 °C) for 24 h. Then, they were transferred to a climate chamber (24 °C) where they were fixed in a swimming posture in a water tank of 30 l. Air, water and core (stomach) temperatures (T a, T w and T c, respectively) were measured every 30 s with Hobo-UAA-002 temperature loggers (Onset Computer Corporation, Bourne, USA; accuracy 0.47 K, resolution 0.1 K). Since all loggers were in exact accordance when kept at the same temperature and since only temperature differences were to be measured, the accuracy of only 0.47 K can be neglected. The difference between T a and T w was always smaller than 0.5 K. Warming constants were obtained from the slope of log(T c–T w) versus time. C (J × K−1 × g−1 × h−1) was calculated by multiplying the warming constant with the specific heat of animal tissue (3.35 J × K−1 × g−1; Calder and King 1974). C’s were calculated over time intervals of 4.5 h. The measured C values are certainly too high, since both gannets used for the measurements were emaciated. However, because the plumage of both birds remained in good condition and entirely down-dry until the end of the measurements, we believe that its insulation remained unaffected. We thus corrected the measured C’s by multiplying them with the respective carcass mass and dividing the result by the body mass of a ‘standard northern gannet’ (3,000 g, Nelson 2002), although adult gannets in winter may be lighter—measurements do not exist. We ignored thermoregulatory costs during flight because in a bird species of this size, one can assume that these are totally or greatly substituted by activity generated heat (e.g. Humphries and Careau 2011).

Statistical analysis

Tests were conducted using Generalised Linear Mixed Models (GLMMs; Faraway 2006). These tests were applied to account for multiple measurements of individuals. GLMMs with Gaussian error distribution were performed using the LME4 package (Bates et al. 2008) of the open source software package R 2.13.0 (R Development Core Team 2008). Data on flight percentages were first arcsin-transformed and then log-transformed to match a normal distribution of data and residuals.

For 2003 and 2004 loggers, before starting the comparisons between regions as the main focus of the paper, we tested:

  1. (a)

    The effect of year (2002–2003 and 2003–2004; all 18 individuals over both winters.); with the response variable = percentage of time spent flying, with predictor = year and random effect = region + individual. The result was not significant (χ2 = 0.143, p = 0.705).

  2. (b)

    The effect of sex (male and female; 10 individuals with sex determination from the second year 2003–2004); with the response variable = percentage of time spent flying, predictor = sex and random effect = region + individual. The result was not significant (χ2 = 0.984, p = 0.321).

As both year and sex were not significant, the different response variables as described in the hypotheses (i.e. percentage of time spent flying and SST) were tested afterwards, each time with predictor = region and random effect = individual. Values given are mean ± SE.

Date of return to the colony was derived from the temperature measurements by the logger (see above). Return date was defined as the first night a bird spent at least part of the night in the breeding colony. Due to poorer logger performance, return dates could only be inferred for 2 birds in spring 2003 and 4 birds in spring 2004, while the sample size for spring 2009 was much larger (n = 25 individuals).

Results

Flight times

During the time of migration in 2002–2003 and 2003–2004, birds migrating to West Africa were flying significantly more (26.1 ± 2.2 % of 24-h day) than birds staying in the North Sea at the same time (17.6 ± 2.4 % of 24-h day; χ2 = 4.455, p = 0.035, GLMM; Fig. 1). In 2008–2009, differences in flight activity were again high. Birds migrating to West Africa flew significantly more (19.2 ± 1.6 % of 24-h day) than birds staying in areas from the English Channel to the Bay of Biscay (12.1 ± 1.8 % of 24-h day; χ2 = 5.948, p = 0.015, GLMM). In all years, migrating birds showed very little flight activity during the hours of darkness (Fig. 2).

Fig. 1
figure 1

Percentages of 24-h-day spent flying in four northern gannets staying in the North Sea/English Channel (bird ID 1–4; at the same time of peak migration of gannets to West Africa: 15–31 Oct) and eight northern gannets migrating to West Africa (bird ID 10–17)

Full size image
Fig. 2
figure 2

Diel rhythm in flight activity of eight northern gannets migrating to West Africa (lower chart) and four northern gannets staying in the North Sea/English Channel (upper chart; at the same time of peak migration of gannets to West Africa: 15–31 Oct)

Full size image

During winter 2002–2003 and 2003–2004, though the birds were flying more in the Bay of Biscay (16.2 ± 1.3 % of 24-h-day) and North Sea (11.5 ± 1.5 % of 24-h-day) than in West Africa (10.9 ± 0.4 % of 24-h-day), the differences were not statistically significant (χ2 = 0.984, p = 0.321, GLMM; Fig. 3). In 2008–2009, birds wintering in West Africa in early January were flying 9.4 ± 1.6 % of 24-h-day while birds wintering from the English Channel to the Bay of Biscay at the same period of year flew 4.6 ± 1.3 % of 24-h-day; this difference was significant (χ2 = 4.310, p = 0.038, GLMM). In all years, resident birds showed very little flight activity during the hours of darkness.

Fig. 3
figure 3

Percentages of 24-h-day spent flying by northern gannets in their winter region. Bird IDs 1–4 refer to gannets wintering in the North Sea and English Channel, bird IDs 5–9 to gannets wintering in the Bay of Biscay and the Celtic Sea, and bird IDs 10–19 to gannets wintering off West Africa

Full size image

Expressed as a percentage of daylight time, flight activity in winter 2002–2003 and 2003–2004 was significantly different between the three regions (χ2 = 7.140, p = 0.008, GLMM; Fig. 4). Bay of Biscay birds flew more of the daylight period (44.2 ± 3.6 % of daylight hour time), followed by North Sea birds (35.2 ± 4.9 % of daylight hour time). West African birds flew much less (23.7 ± 0.9 %). Differences between the North Sea and the Bay of Biscay were not significant (χ2 = 2.863, p = 0.091, GLMM) but in both regions birds showed significantly more flight activity than in West Africa (North Sea vs. West Africa: χ2 = 7.068, p = 0.008, GLMM; Bay of Biscay vs. West Africa: χ2 = 23.253, p < 0.0001, GLMM). In 2008–2009, birds wintering off West Africa in early January flew on average 17.7 ± 2.5 % of daylight time, while those wintering further north flew 9.6 ± 3.1 % of daylight time, the difference being significant (χ2 = 4.408, p = 0.036, GLMM). Both these groups flew much less in 2008–2009 than birds monitored in 2002–2003 and 2003–2004.

Fig. 4
figure 4

Percentages of daylight time spent flying by northern gannets in their winter region. Bird IDs 1–4 refer to gannets wintering in the North Sea and English Channel, bird IDs 5–9 to gannets wintering in the Bay of Biscay and the Celtic Sea, and bird IDs 10–19 to gannets wintering off West Africa

Full size image

Thermostatic costs

Data for 2002–2003 and 2003–2004 showed that sea surface temperature differed strongly and significantly between the three regions (χ2 = 57.036, p < 0.0001, GLMM; Fig. 5). West Africa exhibited the highest SSTs (16.1 ± 0.2 °C), followed by the Bay of Biscay (12.7 ± 0.2 °C) and the North Sea (9.7 ± 0.4 °C). In 2008–2009, birds wintering off West Africa had early January mean SST averaging 16.4 ± 0.1 °C. Birds wintering off Gibraltar or off Portugal averaged 12.3 ± 0.8 °C, and birds wintering in the English Channel and Celtic Sea averaged 8.8 ± 0.6 °C. Differences between the three regions (χ2 = 52.777, p < 0.0001, GLMM) as well as between all pairs of regions were highly significant and considerably affect the energy expenditure of gannets. Thermal conductances corrected for the body mass of a ‘standard gannet’ were 0.86 and 1.00 J × K−1 × g−1 × h−1, respectively, when swimming on water. Thus, using a mean C of 0.93 J × K−1 × g−1 × h−1, assuming a core temperature of 39 °C (SG unpubl. data) and a mean basal metabolic rate (BMR) of 17.5 J g−1 h−1 (Bryant and Furness 1995), all water temperatures below a critical temperature of 20.2 °C will increase thermostatic costs. The estimated average metabolic rate, that is, for SSTs below 20.2 °C thermal conductance × (core temperature–SST), of a resting gannet on water will be 21.3 J g−1 h−1 or 1.22 × BMR off West Africa, 24.5 J g−1 h−1 or 1.40 × BMR in the Bay of Biscay, and 27.3 J g−1 h−1 or 1.56 × BMR in the North Sea. Thus, BMR plus thermoregulatory costs for swimming gannets are 28 % higher in the North Sea and 15 % higher in the Bay of Biscay, respectively, compared to gannets off West Africa. According to the allometric relations for aquatic birds (measurements do not include gannets) by de Vries and van Eerden (1995), C for a 3,000 g bird in swimming position is predicted to be 0.84 J × K−1 × g−1 × h−1, which is slightly less than our estimates.

Fig. 5
figure 5

Sea surface temperature (SST) exhibited by northern gannets in their winter region. Bird IDs 1–4 refer to gannets wintering in the North Sea and English Channel, bird IDs 5–9 to gannets wintering in the Bay of Biscay and the Celtic Sea, and bird IDs 10–19 to gannets wintering off West Africa

Full size image

Arrival and departure times in the colony

Arrival dates at Bass Rock differed significantly between birds from the three wintering regions: Channel and Celtic Sea (median = 7 March, n = 3), Biscay to Gibraltar (12 March, n = 8) and West Africa (19 March, n = 12; t = 2.127, p < 0.045, GLM). Though one could think that sex differences in migratory behaviour could have influenced the results, the tendency was the same when only considering males (sample size for females was even lower): Median dates were 7 March (n = 3 birds), 14 March (n = 6) and 19 March (n = 4), respectively. The differences were not significant (t = 1.007, p = 0.036, GLM), which could likely be due to the small sample size. Similarly to arrival date, birds heading farthest south were also leaving significantly earlier in autumn: Channel and Celtic Sea 6 October (n = 3), Biscay to Gibraltar 9 October (n = 8) and West Africa 4 October (n = 12; t = −2.778, p = 0.011, GLM).

Discussion

Flight activity of gannets during winter was low. Although birds in the different regions face strongly varying environmental conditions (water temperature, air temperature and wind regime), these percentages were only partly significantly different, perhaps suggesting that the wintertime foraging effort of gannets is not much different between regions. Most flight activity occurs in the morning and the evening and so seems to coincide with the expected foraging pattern (Garthe et al. 2003). Individual variation, when averaged over all days in winter, was low too; individuals in 2002–2003 and 2003–2004 showed mean values ranging from 8.5 to 19.5 %. Variability increased when looking at an individual daily level, with percentages of time flying varying between 4.0 and 30.0 % of the 24-h cycle from day to day. In winter 2008–2009, 5 of the 6 birds wintering in the English Channel/Bay of Biscay region were not flying at all on single days in the early January sample period while this was not the case for the eleven birds wintering off West Africa at the same time. During the breeding season, the relative amount of flight activity is higher (means: 22 %, Garthe et al. 2003; 25 %; Lewis et al. 2004; when assuming that each individual stays 50 % of the time at sea and 50 % in the colony guarding the nest and the egg/chick), also suggesting that foraging in the non-breeding season, when birds are able to stay in their foraging areas, is not a particularly demanding process. Nocturnal flight activity was negligible in winter. This can easily be explained by foraging patterns, as northern gannets have been shown to forage only during daylight periods and are considered visual hunters (Cunningham 1866; Garthe et al. 1999, 2003; Hamer et al. 2000). Day length may act as a severely limiting factor in foraging seabirds. Green et al. (2005) studied the foraging behaviour of Macaroni penguins (Eudyptes chrysolophus) year-round. They observed that the birds dive predominantly during daylight hours at all times of the year but appear to be more constrained by daylight during the short winter days. Several diving variables including dive duration, dive rate and amount of time spent diving were significantly related to day length, and these associations were stronger during winter than summer. In another study, Daunt et al. (2006) showed that seasonal patterns of foraging time in European shags (Phalacrocorax aristotelis) appear to be driven by a combination of light levels and weather conditions and may be linked to the availability of the shag’s principal prey fish species. They found no evidence that shags dispersed south in winter to increase potential foraging time. When expressing flight activity as a proportion of the daylight period (in contrast to the 24-h day used before) in this study, we found significant differences between regions. In 2002–2003 and 2003–2004, gannets wintering in the North Sea flew on average slightly more than a third of the daylight period, gannets wintering in the Bay of Biscay almost half of the daylight period. Although this still leaves many daylight hours for resting on average, individual daily values reached up to 81 % in the Bay of Biscay and 88 % in the North Sea but never higher than 48 % off West Africa. This becomes even more obvious when flight activity is expressed as the percentage of days when gannets spent more than 50 % of the daylight period flying: 33 % of all days in the Bay of Biscay, 31 % in the North Sea and 0 % off West Africa. It may be assumed that gannets can face strong time constraints on certain days, for example, with severe weather and accompanying bad visibility, in their northern wintering areas, whereas on many days rather little time is required to obtain sufficient food. Interestingly, in 2008–2009, the amount of flight activity in the northern part of the wintering area was considerably less than in the data from 2002 to 03 and 2003–2004, which might either suggest that feeding conditions were better in 2008–2009 so required less activity by the birds or that adverse weather conditions prevented the birds from feeding. In contrast, percentage of time flying was almost identical for birds wintering in West Africa.

In 2002–2003 and 2003–2004, birds migrating to West Africa flew on average 48 % longer per day than birds staying in the North Sea during the same time of year. From the devices used in this study, it is not possible to distinguish between long-distance flights and foraging flights, but energetic costs for both flying modes are thought to be equally high (Enstipp et al. 2006). As flight activity is estimated to be 3.5 times more costly than resting at sea (= swimming) in northern gannets (Enstipp et al. 2006), birds that winter off West Africa would have higher costs due to elevated activity for two periods of 3–4 weeks each when they are on migration. This would result in an increase of the daily energy requirement of a gannet migrating to West Africa by 15 % as compared to a gannet staying in the North Sea because of the higher amount of flying. In 2008–2009, birds that eventually spent the winter off West Africa flew 58 % longer per day than birds commuting only to regions from the English Channel to the Bay of Biscay at the same time of year. But as total flight time was lower in 2008–2009 than in 2002–2003 and 2003–2004, the increase in daily energy requirement was (only) 14 % for the birds migrating much further south in that winter. The lack of nocturnal flight activity indicates that migration is not an energetic constraint for the birds investigated, in contrast to recordings of gannets crossing the North Atlantic in just a few days (Fifield 2011).

By far the largest differences between the different winter regions are in the sea surface temperatures (SSTs). The waters of the Atlantic off West Africa are considerably warmer than in the Bay of Biscay (3.4 °C warmer on average) that again is much warmer than in the North Sea (3.0 °C warmer on average). Although no direct measurements exist for metabolic rates of northern gannets as a function of water temperature, it is obvious that individuals staying in the cold North Sea water face much higher costs than birds wintering further south, particularly since it has been argued that members of the Sulidae family in general are better adapted to sub-tropical climates than to cold-temperate ones (Birt-Friesen et al. 1989). The coldest SST analysed for an individual gannet wintering in the North Sea was 7.0 °C. According to our estimates, this results in high energetic costs of 29.8 J g−1 h−1 or 1.70 × BMR for a gannet resting on water in still air. Despite the deeper ventral plumage (20–23 mm compared to 9–11 mm on the back, our own measurements on a museum specimen, cf. Fort et al. 2009 for measurements on auks), heat loss to (still) water is much higher than to (still) air since the thermal conductivity of water is 24 times higher than that of air. In the north-eastern Atlantic, air temperatures at the sea surface are generally close to water temperatures (Cayan 1980) and hence close to the conditions in the measurement chamber. However, heat loss through forced convection by winds (wind chill) may have considerable impact on the heat balance (Walsberg 1988; Fort et al. 2009). But since surface wind speeds in winter are on average highest in the North Sea and lowest off West Africa (http://www.ssmi.com/amsr/amsr_browse.html), this generally corroborates our hypothesis of lower thermostatic costs for birds wintering further south. Humphreys et al. (2007) also point out the high costs of resting on (cold) water and suggest that migration to milder latitudes, following breeding, will provide enhanced benefits, particularly to seabirds such as black-legged kittiwakes (Rissa tridactyla) that rest on the sea surface during darkness, as did all of our gannets. However, waterbirds might partly be able to counterbalance their elevated thermostatic costs while resting on water by the heat increment of feeding, that is, heat generated as a by-product of digestion. Enstipp et al. (2008) estimated that substitution from the heat increment of feeding might reduce the daily thermoregulatory costs of double-crested cormorants (Phalacrocorax auritus) wintering in coastal British Columbia by about 38 %.

We cannot calculate accurately whether the reduced thermoregulatory costs of resting in warmer water are more or less than the costs of migratory flight for the birds moving to warmer climates, since data from geolocator data loggers are too crude to estimate distances covered by daily flights with sufficient accuracy and we cannot assume that gannets follow consistently straight lines while on migration. But energetic costs for migration flight and for thermoregulation must tend to cancel each other out in terms of lifetime energetics. On the other hand, it was shown that, quite in contrast to the conspecifics wintering further north, none of the birds wintering off West Africa spent more than half of the day flying. If we take flight time as a proxy for the foraging effort, it becomes obvious that the risk of days with bad feeding conditions is much smaller off West Africa. On the other hand, birds wintering further north arrived earlier in the breeding colony and were thus able to occupy better nest sites than their conspecific further south. Unfortunately, another major factor, food availability, could not be quantified in our large-scale study. Overall, however, it seems that there was no clear optimal region for gannets in the years in which we deployed loggers and that the net cost-benefit may be rather similar for all areas. This may explain why all of these disparate wintering regions were used by good numbers of Bass Rock gannets that individually follow quite different wintering strategies.