Key migration numbers and 3-day mean location data of Common, Pallid and Alpine Swifts from the Levant

We captured Common, Pallid and Alpine Swifts from breeding colonies at three different locations in Israel (Ramat-Gan, Mount Sodom and Haifa) during the breeding season in the years 2017 to 2020. Using a full-body harness, we attached 29 multisensory light-level geolocation (ms-GLS) loggers on Common Swift, 20 ms-GLS loggers on Pallid Swifts and 10 ms-GLS loggers on Alpine Swifts. Additionally, we equipped another 10 Pallid Swifts with light-level geolocation loggers that had the ability to remotely download the data by radio (rd-GLS). Of these, we recovered data from 35% (10/29) of the ms-GLS loggers on Common Swifts, 25% (5/20) on Pallid Swifts, and 80% (8/10) on Alpine Swifts. For the rd-GLS loggers, we successfully downloaded data from 70% (7/10) of the Pallid Swifts. We were able to use data over the full annual cycle from 10 loggers for Common Swifts, 10 for Pallid Swifts and five for Alpine Swifts. Additionally, we had data only until the arrival at the non-breeding range for one Alpine Swift and two Pallid Swifts. One of the Pallid Swifts (20MX) provided data over two annual cycles, of which we only used the first year in the analysis. In addition to attaching the GLS on the swifts, we measured their wing length and weight, and collected feather samples for molecular sexing (Griffiths et al., 1998) that was done by the Karnieli Vet Ltd. laboratory.To determine the spatiotemporal location (latitude and longitude) each day, we used twilight data from the light-level sensor of the geolocation loggers. For the calibration of the seven rd-GLS loggers from Pallid Swifts, we used the package ‘SGAT’ (Sumner et al. 2009; Lisovski and Hahn 2012), while we used the package ‘FLightR’ (Rakhimberdiev et al. 2017) for the calibration of the other ms-GLS loggers (Lisovski et al. 2020). To identify stationary periods of relative geographic stasis over at least 3 days, e.g. stopover and stationary behaviour at non-breeding sites, and to differentiate these periods from directional migratory flights, we analysed the twilight data with the ‘ChangeLight’ function from the package ‘SGAT’ (Sumner et al. 2009; Lisovski and Hahn 2012). Here, we used a quantile of 0.85 and merged sites within the distance of the maximal error. To calibrate this maximal error, we identified the precision of the locations for the period during which birds were in the breeding range (i.e. after capture and before departure for autumn migration). For the ms-GLS loggers, the maximal error was 102 km. For the rd-GLS loggers the maximal error was larger, 264 km. Due to the low precision of spatial location, particularly from the rd-GLS loggers, we used 3-day mean locations for analyses (Åkesson et al. 2012; Thorup et al. 2017). Furthermore, we excluded the spatiotemporal locations 10 days on either side of the equinox times (Lisovski et al. 2020).See also the attached R code on the statistical analysis of the migration numbers. Please run renv::restore() to install the exact package versions.