Fieldwork and material
Birds were captured and ringed at a single study site in Valais, Switzerland (46.33 N, 7.43 E; 1800–2100 m above sea level) during the breeding season, i.e. in April–June 2015–2020. Captures were performed with 2.5-m high mistnets placed among potential foraging grounds or parallel to forest edges. Birds were sexed from plumage coloration and age — either second calendar year (2cy) or adult (>2cy) — determined based on the presence of a moult limit in the greater coverts 41.
We used four types of loggers to record ring ouzel locations: simple geolocators (hereafter GL; model GDL2, Swiss Ornithological Institute (SOI), Switzerland); remote-download geolocators (hereafter also termed GL; model GDL-uTag, SOI, Switzerland); multi-sensor loggers (hereafter MSL; model GDL3-PAM, SOI, Switzerland) and GPS loggers (GPS; model nanoFix-GEO, PathTrack Ltd, UK). In addition to light intensity, the deployed MSL measured acceleration and atmospheric pressure at 5-min intervals see 21 for details. GPS were programmed to record position once a week. All types of loggers were fixed on the birds using a leg-loop harness, made of elastic rubber or inelastic threaded nylon as concerns GL and MSL, and Teflon ribbon for GPS. The different types of loggers (see details in the Supplementary Table S1) weighted at most 2.6% of the mean (± SD) body mass as measured from captured birds (males: 95.1 ± 5.1 g, n = 191; females: 100.8 ± 8.9, n = 91). The permit for bird capturing was delivered by the Swiss Federal Office for the Environment (F044-0799) and fitting of tracking devices was authorized by the Swiss Federal Food Safety and Veterinary Office, with all study protocols approved by the responsible ethics committee. Capturing and tagging were performed following all relevant guidelines and regulations of the abovementioned federal offices.
We equipped a total of 59 individuals with 62 GL or MSL (three individuals were equipped twice) as well as 15 individuals with GPS between 2015–2019 (see Supplementary Table S1). Only seven out of the 62 GL/MSL were retrieved by recapture of the tagged bird, while data from another four GL could be downloaded remotely in the field. Two additional GL-tagged individuals had lost their logger at the time of recapture. We thus retrieved data from, in total, 5 MSL and 6 GL. For MSL, data was complete (over one year) except for one device that had stopped recording as early as February in the year following tagging. Regarding GL, intense shading prevented data exploitation for two of them. Shading by feathers or the surrounding habitat may indeed strongly bias the measurements of sunrise or sunset times (hereafter twilights) and lead to spurious localizations. We additionally retrieved two out of the 15 GPS by recapture but both had malfunctioned, with locations available for only one GPS for just a month after deployment.
On subsequent years following ringing, we resighted 33.9% (20/59) of the individuals equipped with GL and MSL, and 20% (3/15) of the GPS-tagged birds, to be compared with 29.9% (64/214) of the ring ouzels that had only been colour-ring marked at the study site and served as a control group. As assessed with Bayesian Cormack-Jolly-Seber models from visual resightings following 42, apparent survival rates of GL- and MSL-tagged birds did not differ from the control group (β = 0.27, 95% CI: -0.92 to 1.66), while we evidenced a detrimental effect of the slightly heavier GPS loggers (β = -1.74, 95% CI: -3.32 to -0.42).
Analyses
All analyses were performed with the software R version 3.6.2 43 using the packages TwGeos 44, GeoLight 45, SGAT 46 and PAMLr 47, following the general framework described in Lisovski, et al. 48. Starting with data from the five MSL, we classified bird behaviour into four categories of activity (no activity, low activity, high activity and migration) based on acceleration measures, using the algorithm from the classifyFLAP function in PAMLr. We defined migratory flights as those equal or longer than 30 min, which corresponds to at least six consecutive readings with ascertained flight activity. Based on this data, we defined the migratory schedule and separated the annual cycle into four periods: post-breeding, autumn migration, non-breeding (i.e. overwintering) and spring migration (the locations during reproduction being irrelevant here). The post-breeding period started on the day of the first nocturnal flight in June or July and lasted up to the autumn migration departure, which was defined as the first true migratory flight after August 1st. We assumed that birds had reached their non-breeding residence area as soon as they had stayed for at least two weeks in a row at the same place after October 1st. Spring migration started with the first ascertained migratory flight in March.
In a second step, we converted readings of atmospheric pressure into m above sea level (hereafter m asl) using the function altitudeCALC in the PAMLr package, which is based on the hypsometric equation that assumes standard atmospheric conditions 21,49. Hence, estimates of altitude are rather precise but can be biased by natural variations in atmospheric pressure, i.e. influenced by the so-called «high- and low-pressure areas». Such shifts in pressure are however fairly slow and minor (maximum of 2 hPa h-1) so that they would not generate abrupt changes in estimated altitude 21. Furthermore, daily fluctuations in atmospheric pressure, called atmospheric tides, reach at most 3 hPa in the tropics 50, potentially inducing a maximal daily altitudinal deviation of only ca 30 m for a given location. We summarized the altitude information as the median and range (minimum to maximum) for each of the four periods of the annual cycle, treating readings during migratory bouts separately.
Finally, we derived geographic positions of the nine birds for which light-intensity data was available and of sufficient quality. We first defined twilights using TwGeos and then categorized those into residency and movement periods. For MSL, this distinction was based on the migratory flights that were identified as described previously. We considered only periods of eight consecutive days without migratory flight as true stopovers, given the noise in the data and thus the need of longer periods to estimate accurate locations. For GL, the distinction was done using the function changeLight in GeoLight, again setting a threshold of eight days for distinguishing a stopover. We used «in-habitat» calibration of the sun elevation angles (zero and median) for parameterizing the error distribution around the twilight times 45, i.e. using as a reference the period during which a bird was for sure present at its breeding site. We then modeled the migration trajectory as well as stopover and residency locations using SGAT. We chose a grouped Estelle model, where estimates within residency periods are grouped together to increase spatial precision 48. We forced residency periods to occur on land only, whereas movement was not constrained spatially but flight speed assumed to follow a gamma distribution (β = 2.2, SD = 0.08). The starting point of each trajectory track was fixed at the very breeding location, as was the end point, except for the individual whose logger stopped recording in the middle of winter. To fit the Estelle model, we first drew 1’000 initial samples using a ‘modifiedGamma’ model (i.e. relaxed model, allowing negative errors on twilight times), tuned it 5 times with 300 iterations using a ‘Gamma’ distribution. We shall here report median estimates ± 95% credible intervals (CI; based on 2.5 and 97.5% quantiles) from a final run with 2’000 iterations to ensure convergence.