The study was conducted on Ventotene (Italy), an approximately 1.5 km2 island () in the Tyrrhenian Sea (40°47′ N, 13°24′ E), which is visited by songbirds during migration. A ringing station has been active on the island since 1988. We caught garden warblers on 10 May (n = 18) and 11 May 2009 (n = 2), two consecutive days with constant weather conditions, which did not change until all birds had left the island (). In fact, 8–13 May was the main period of broad-front garden warbler migration along the Pontinian islands (including Ventotene) in 2009.
Map of Ventotene with main roads in black, red dots indicate one stationary (tower) and six mobile telemetry locations, the inlay shows Italy with a black arrow indicating the location of Ventotene (scale bar, 200 m).
Table 1. Weather data from Ponza (40 km NE of Ventotene) and number of tagged and leaving birds (2 birds that lost their transmitters were excluded; period during which all birds left the island in italic; abbreviations: ENE, east northeast; ESE, east southeast; (more ...)
Following the European ringing standards described in Bairlein (1995)
, we scored their subcutaneous fat on a 0–8 scale, the size of the pectoral muscles on a 0–3 scale, and measured body mass, tarsus and the third primary. We selected 10 lean (fat scores of 0 or 1, muscle scores 1 or 2) and 10 fat (fat scores 3 or 4, muscle scores 2 or 3) garden warblers and glued a Holohil LB-2 or LB-2N transmitter (battery life two weeks, weight 0.37 g) on their backs (skin glue, Sauer GmbH, Lobbach, Germany). The transmitters represented 1.9–2.7% of the body mass of the birds (lean birds range: 13.9–16.4 g; fat birds range: 15.4–19.1 g). We alternated tagging of fat and lean birds, so that capture time of day did not differ between lean and fat birds (Mann–Whitney U
= 29.5, p
= 0.12). Birds were released after attaching the transmitter, and their presence on the island monitored during day and night on a 2 h basis for the first 48 h, on a 4 h basis until the last animal had left, and a 9–12 h basis for five more days. Telemetry was conducted with one stationary receiver located at a tower and a mobile receiver at six other locations around the island (see and the electronic supplementary material). In total, we conducted 923 scans to measure the absence or presence of radio-tagged individuals and unambiguously determined the presence of birds remaining on the island or their absence after departure to resume migration. Two garden warblers (one lean and one fat) lost their transmitters (continuous constant signal from a constant direction) and were excluded from the dataset. The signals of 14 birds were recorded during all scans, the signals of three birds were missed during one scan each, and the signal of one bird was missed during three subsequent scans, but then the signals of all these birds were detectable again until departure. Each frequency was checked for at least 72 h after the signal had disappeared, to exclude the possibility that the bird was only temporarily not detectable.
Weather data were obtained from Ponza, a neighbouring island 40 km northeast of Ventotene ().
Statistical analyses were conducted with Systat
12 and Mark
(White & Burnham 1999
). Telemetry data were analysed with the ‘nest success model’ for radio-telemetric data implemented in Mark
. Use of the nest success model was justified because of the high detection probability of radio-tagged birds (only six misses, see above). We (i) constructed a full model with survival probability depending on fat class, time (i.e. survival probability changes between different re-sightings) and the interaction between these two factors (
[fat class * time]). We compared this full model with all other possible reduced models that considered (ii) only variation in survival probability over time independent of fat class (
[time]), (iii) only variation in fat class independent on time (
[fat class]), and (iv) a model constant survival model independent of fat class and time (
[.]). The best model was selected according to the corrected Akaike information criterion (AICc) implemented in Mark
, and the model fit was tested against the fit of the next best model using the likelihood ratio test implemented in Mark
. Because our data were left-truncated, i.e. we did not know how long each bird had already been present on the island before capture, we estimated total stopover duration using the life expectancy formula S
) as suggested by Efford (2005
; see also the electronic supplementary material for the calculations and the conversion of S
into hours). Data are presented as means ± 95% CI.