Data were collected on the German island of Helgoland (1 km2; 54°11′N, 07°55′E). The distance to the nearest coast in eastern to southern directions is c. 50 km, while birds heading to the north-west face about 750 km of sea (Fig. 1). We trapped northern wheatears during spring 2008 and spring 2009 from the beginning of April to the end of May, the main spring migration period (Dierschke & Delingat 2001). Birds were ringed, aged, sexed and measured according to Svensson (1992) and weighed to the nearest 0·2 g. Fat stores and size of the breast muscle were scored according to Kaiser (1993) and Bairlein (1994). Wing length was used to identify the subspecies: males and females with wing lengths exceeding 102 and 97 mm, respectively, were treated as belonging to the leucorhoa subspecies (Svensson 1992). Ten in 2008 and 21 in 2009 of the overall 278 birds trapped were radio tagged. Leucorhoa wheatears were caught, ringed and radio tagged under licence of the Ministry for Agriculture, the Environment and Rural Areas, Schleswig-Holstein, Germany.
Radiotransmitters were constructed by the Swiss Ornithological Institute in cooperation with the University of Applied Sciences Bern, Switzerland (Naef-Daenzer et al. 2005). Radiotransmitters, including battery and harness, weighed 0·8 g. The transmitters were attached to the leucorhoa wheatears using a Rappole-type harness made from 0·5-mm elastic cord (Rappole & Tipton 1990). Length of leg-loops was adjusted individually to birds (Naef-Daenzer 2007). Because the lowest body mass of the leucorhoa wheatears involved was 21·3 g (mean 29·7 g, SD 5·1 g, n = 31), the mass of the radiotransmitter represents <3·8% (mean: 2·7%) of the bird’s body mass. The relative load was therefore below the recommended 5% limit (Cochran 1980; Caccamise & Hedin 1985). Potential adverse effects on the birds’ behaviour are insubstantial (e.g. Naef-Daenzer, Widmer & Nuber 2001; Rae et al. 2010), and the increase in flight costs is small (Irvine, Leckie & Redpath 2007); though, drag (Bowlin et al. 2010) as well as energy expenditure (Barron, Brawn & Weatherhead 2010) increases. Transmitter life was about 30 days.
Using Yagi 3EL2 hand-held antennas (Vårgårda, Sweden) in combination with YAESU FT-290RII receivers, the detection range was c. 12–15 km. The detection range was determined by using a radiotransmitter on a ferry leaving the island and locating its position every few seconds by means of a GPS device. As birds likely fly at an altitude exceeding that of the test transmitter, this represents a minimum tracking distance. The detection range of the radios is affected by the orientation of the transmitter antenna, if pointing towards the receiver the detection range was 8 km, if oriented perpendicularly towards Helgoland 15 km. We tracked departing birds for on average 18 min (SD 7·5 min, n = 26). As a northern wheatear’s airspeed is about 47 km h−1 (Bruderer & Boldt 2001), the duration of tracking also indicated a mean detection range of 14 km (SD 6 km), which coincided well with the detection range obtained by the ferry experiment. We tracked birds every night continuously from sunset till early morning or until departure. The locations of the birds on the island were estimated by triangulation mostly from two to three mobile observers or by a single mobile observer localizing the bird from different positions on the island. During each departure event, birds were radiotracked from the ‘Oberland’ cliff (c. 50 m above sea level and highest area of the island) mostly by one mobile observer. For departing birds, we recorded directions until loss of signal. According to the bearings, birds departed in a straight line from the island. We used last recorded direction before loss of signal as the departure direction. The bearing accuracy was determined by blindfolded tests with fixed radiotransmitters. The average bearing error was 3° (SD 5°, n = 49); see also Schmaljohann et al. (2011). Flight altitude cannot be determined with the radiotelemetry settings used in this study. As the island is very small, a potential parallax error in direction estimates is small compared to the bearing accuracy of hand-held antennas (Kenward 2001). Set-off distance between bird and observer was far <500 m (see above); the parallax error in respect to a tracking distance of 15 km would be <2°.
We compared fuel loads rather than body masses because leucorhoa wheatears differ substantially in size (range of wing lengths in this study 97·5–109 mm). We estimated lean body mass from wing length using a linear regression based on 220 ‘lean’ northern wheatears with fat score <2 and muscle score <2 caught on Helgoland in the years 1998–2002 and 2008:
- (eqn 1)
(linear regression: n = 220, F1,218 = 95·07, adj-R2 = 0·30, P <0·0001).
Departure fuel load was calculated for each individual as:
- (eqn 2)
We used body mass at capture as departure body mass for ten birds that departed on the night of their capture day. We estimated departure body mass for the other 21 leucorhoa wheatears by remote weighing or by modelling. Our method of remote weighing was to place bowls supplied with mealworms ad libitum on electric balances so that the body mass of individually ringed leucorhoa wheatears could be read repeatedly over time (Schmaljohann & Dierschke 2005). We defined a bird’s departure body mass as its body mass on the evening of departure (after 7 pm). This procedure succeeded for four birds. Departure body mass and fuel deposition rate, as the relative body mass increase over time (Schmaljohann & Dierschke 2005), of the other 17 had to be modelled. Mean minimum stopover duration of these 17 birds was 2·6 days (SD 2·3 days). From intensive research with northern wheatears on Helgoland (Dierschke & Delingat 2001; Dierschke, Mendel & Schmaljohann 2005; Schmaljohann & Dierschke 2005; Delingat et al. 2009; Schmaljohann et al. 2011), we know that the average fuel deposition rate is about 0·09 day−1 proportionally to bird’s lean body mass (see Appendix S1 in Supporting Information).
To estimate the effect of wind on departure decision, we used wind data from the National Oceanic and Atmospheric Administration (NOAA, Boulder, CO, USA; available http://www.cdc.noaa.gov/cdc/data.ncep.reanalysis.derived.html; Kalnay et al. 1996). These data included four different pressure levels (1000, 925, 850, and 700 mbar) representing roughly four altitude intervals (ground level – 445, 445–1145, 1145–2375 and 2375–4000 m). Midnight wind direction and wind speed correlated significantly between adjacent altitudes during the main study period (April – May 2009: Rc–c > 0·84 and RS > 0·76, all P < 0·0001). We included midnight wind data for our analyses, because departure time in every case was closest to the 0 a.m. wind data. The individual wind profit was calculated at the four different altitudinal intervals on departure night following Liechit, Hedenström & Alerstam (1994) and Erni et al. (2002) as:
- (eqn 3)
To assess the effect of the wind profit on the flight range of each bird, we first estimated the bird’s potential nocturnal flight duration. This quantity is a function of departure fuel load:
- (eqn 4)
following Delingat, Bairlein & Hedenström (2008). Bird’s flight duration was, however, restricted to the time remaining between departure and sunrise and not until fuel load was depleted, because northern wheatears are nocturnal migrants (Schmaljohann et al. 2011). This nocturnal flight duration was then multiplied by species air speed (13 m s−1 = 47 km h−1; Bruderer & Boldt 2001), which is the flight range in still air restricted for the period between departure and sunrise (restricted flight rangei). To account for the wind profit experienced, we added the nocturnal flight durationi times the wind profiti to the restricted flight rangei in still air:
- (eqn 5)
To consider potential long nonstop flights across the Atlantic (cf. Thorup, Ortvad & Rabøl 2006), we estimated birds’ maximum flight ranges based on the assumption that they would experience the same wind condition along their route as they did during their departure, whereby total flight duration was defined as birds’ flight duration until fuel loads were depleted, see eqn 4:
- (eqn 6)
We used sun’s elevation at departure rather than time elapsed after sunset to describe departure at night because sun elevation is a physical trait that has important effects on orientation cues (Muheim, Moore & Phillips 2006) and because the time associated with a particular elevation changes over the season.
Statistics were calculated using the statistical software package R (R Development Core Team 2010). Uniformity of directions was tested with the Rayleigh test of uniformity (Batschelet 1981; Jammalamadaka & SenGupta 2001). Circular–linear correlations were calculated following the methods described by Jammalamadaka & SenGupta (2001). The P-value for a circular–linear correlation was approximated by a randomization test. For each circular and linear variable, random samples with replacement were drawn, and the circular–linear correlation coefficient of these values was estimated. We used 10 000 random replications in each case. The number of such replicates that have correlation coefficient larger than that associated with the original data set, divided by the total number of replications, provides a robust estimate of the corresponding P-value (Crawley 2005). Values of sun’s elevation were log-transformed in all linear regression models so that residual analyses did not show any serious deviation from normal distribution. All appropriate statistical analyses remained significant when the time after sunset instead of the sun’s elevation was used as a predictor.