Migratory birds typically separate energetically demanding parts of the annual cycle like breeding, moult, and migration with some species engaging in so-called moult-migration. Moult-migration is known to occur in starlings from the northern breeding populations, however, little is known about the dynamics of this phenomenon and the costs and benefits for the involved individuals. Here, using state-of-the-art multi-sensor geolocators, we gathered information about the annual cycles of 10 starlings from two breeding sites in Latvia. We used a novel analytical approach based on atmospheric pressure measurements to reveal that all but one of the tracked individuals migrated to wintering sites in the British Isles. Tracking data exposed two separate migration strategies – (1) departure from the breeding grounds in mid-June soon after chick fledging with long stationary periods at moulting sites approx. 900 km westward (n = 5 of 10); (2) residing in close vicinity of the breeding sites up until the end of October (n = 5 of 10). Accelerometer data revealed significantly higher activity budgets during moult for the individuals exhibiting moult-migration. Furthermore, birds that underwent moult-migration arrived at the breeding sites in the following year on average 10 days later and showed significantly higher activity levels during the pre-breeding period compared to birds without moult-migration. Activity tracking also showed that 67% of all migratory flights were performed during the night, contradicting previous assumptions of starlings being predominantly diurnal migrants. Maximum recorded flight altitudes reached 2500 m.a.s.l. and the longest uninterrupted flight lasted 22.5 hours. Our results highlight energetic trade-offs of moult-migration in starling, but their downstream consequences remain to be tested.

In 2020, we set up two new study sites in Latvia to study Starlings. One in eastern Latvia in Zeltaleja (57.19°N 26.86°E) consisting of 100 nest boxes and the second in western Latvia near Lake Engure (57.27°N 23.10°E) consisting of two plots ca. 8 km apart with 100 nest boxes each. The distance between the two study sites was 230 km. During the breeding season from mid-April until mid-June all nest boxes were monitored at least once every four days and adult birds and chicks were ringed. Onset of egg laying at both study sites was highly synchronized among individuals with 90% of all recorded broods (n = 181) started between 19 and 29 April. No second brood was recorded at either of the two study sites. In 2020, we fitted 70 adult breeders (35 at each site; 34 males, 36 females) with multi-sensor archival data loggers (model: GDL3-PAM, Swiss Ornithological Institute,  Liechti et al. 2018), which were attached on the bird's back using nylon cord leg-loop harness. Adult birds were predominantly captured with traps inside the nest boxes. To avoid nest abandonment, the trapping and deployment of loggers were carried out during the late stages of the breeding between 17 and 26 May when the chicks were more than 10 days old. The geolocator including the harness on average weighed 1.90 ± 0.04 g, which represents less than 3% of the body mass of starling (80.4 ± 3.93 g, n = 107), falling under recommended ethical threshold (Barron et al. 2010, Brlík et al. 2020). The loggers accommodated sensors for ambient light intensity, atmospheric pressure, ambient temperature, and acceleration. The light sensor was equipped with a 7 mm long light guide and light intensity was measured every minute storing maximum values in 5 min intervals. Acceleration/activity was measured along the Z-axis for 3.2 seconds at 10 Hz frequency every 5 minutes. Atmospheric pressure and temperature recordings were set to 30-minute intervals (Briedis et al. 2020).

In 2021, we acquired nine full tracks (5 males, 4 females) and two incomplete tracks (1 male, 1 female). One of the incomplete tracks (male) had recorded data up until the 10th of December, thus, including information about the autumn migration and the first half of the wintering period, while the others (female) barometric pressure data were unusable due to malfunction of the sensor. Overall recapture rate for birds with loggers was 15.7% (11/70); 20% (7/35) in Engure, and 11% (4/35) in Zeltaleja. To see whether loggers had any impact on return rates we used Chi-square test comparing return rates of logger birds with those of control groups (n = 18; birds undergoing the same handling procedure except fitting the loggers). Results showed no significant difference between the two groups (χ² = 0.02, p = 0.88; males only: χ² = 4.77, p = 0.21; females only: χ² = 1.08, p = 0.84).

For distinguishing between movement and stationary phases during the annual cycle, we used accelerometer and atmospheric pressure recordings. Because starlings use flapping flight for migration (Rayner et al. 2001), we used the flapping flight classification from the R-package ‘pamlr’ (Dhanjal-Adams et al. 2022) and set the duration threshold to 1 h (equals to 12 consecutive readings of flapping activity at 5 min recording intervals). Thus, activity measures that were classified as flapping and lasted for at least 1 h were regarded as migratory flights (Briedis et al. 2020). Using 1 h as the cut-off threshold for the identification of migratory flights may have left some shorter duration flights unnoticed. However, lowering the threshold increases the risk of misidentifying extended commutes to and from roosting sites or lengthy murmurations (King and Sumpter 2012) as migration. From the classified migratory flights, we could invertedly identify stopover periods in-between migration flights. Longer periods exceeding 45 days were regarded as residency sites. We considered long stationary phases during winter months (November-February) as the main wintering grounds.

To derive geographic locations of stationary sites, we used atmospheric pressure as it is not expected to change rapidly when the bird is stationary (weather-related changes; Liechti et al. 2018, Sjöberg et al. 2018). Throughout the analyses, we followed the general procedure outlined in Nussbaumer et al. (2022a). The recorded pressure data were compared with surface-level atmospheric pressure data (ERA5 hourly surface-level pressure data) at 0.25x0.25-degree grid cells available at The Copernicus Climate Change Service (Hersbach et al. 2018). Based on pre-analyses of light data, we defined an area between 60°N, 45°N, 10°W and 30°E where all birds had resided throughout the annual cycle and further location estimates were rendered in this pre-defined area to save computational time. To avoid birds’ behaviour (e.g., flying, foraging, etc.) potentially influencing the recorded pressure data, only recordings during night (light recording = 0) and while the bird was not in motion (acceleration < 5 units on the arbitrary scale; max recorded value across all birds = 101) were used for location estimation. First, we excluded all the cells where recorded pressure data did not match the range of air pressure between lowest and highest elevation points within a cell at a given hour. To calculate this range, we combined remotely sensed elevation data at 90x90m resolution available from SRTM DEM Digital Elevation Database (Jarvis et al. 2008) and the ERA5 data. Second, to further refine possible locations we calculated the maximal distance starlings could have travelled from the previous stopover/residency site assuming flight duration as per flight classification analyses and maximum average ground speed of 110km h^-1 (we deliberately chose a very high ground speed value to account for potential effects of strong tailwind support). Consequently, we excluded all cells outside the potential flight range. Third, we derived longitudinal boundaries of the stopover/residency sites from recorded light intensity data using R-package, ‘GeoLight’ (Lisovski et al. 2012). Longitude, in contrast to latitude, estimates are more reliable throughout the year as they are not influenced by calibration and equinox times (Lisovski et al. 2012). After integrating these steps, we ran a correlation analysis of pressure recorded on the geolocator and ERA5 data in all remaining grid cells. Higher correlation coefficient here indicates a higher likelihood that a bird was residing at a given grid cell. Finally, we checked residuals between the recorded pressure data from geolocators with the ERA5 data from the cell with the highest correlation coefficient. This procedure allowed for detection of changes in roosting sites and potential migratory flights that were shorter than 1 hour. If such flights were discovered, we split the stationary period accordingly and rerun the analyses (see section Labelling tracks from the user manual for GeoPressureR, Nussbaumer et al. 2022a, 2022b PREPRINT). Final maps illustrate grid cells with correlation coefficients within the 95th percentile (pressure data) of the potential residency/stopover area.

Recorded light intensity data were used to identify sunrise and sunset times, which allowed us to distinguish between diurnal and nocturnal migratory flights. To better understand the migratory behaviour of starlings, this data was also used to compare departure and landing times to local sunrise and sunset. Also, individual migratory flight durations were compared with take-off times relative to local sunset. Further, for each individual, we calculated cumulative flight duration, migration distances, average ground speed and migration speed in summer/autumn and in spring. Migration distance was calculated as a great circle distance between the breeding site and the most likely location estimates across the annual cycle. Here, ‘migration speed’ (km day^-1) describes speed at which a bird travelled from breeding site to the main wintering grounds and back, including stopover time. ‘Ground speed’ (km h^-1) describes flight speed between individual stationary points along the migratory track.

Application of hypsometric equation to the atmospheric pressure measurements allowed us to calculate flight altitudes of each individual flight at 30 min temporal resolution (‘pamlr’ function calculate_altitude; Dhanjal-Adams et al. 2022). Flight altitudes were later compared between autumn and spring, as well as between diurnal and nocturnal flights using a two-tailed Student's t-test.

R-package ‘pamlr’ (Dhanjal-Adams et al. 2022)