All studies relate to only the period in which the birds were situated in the breeding area.
Transmitters and study area
The study is based on 29 breeding birds (19 males and 10 females) that were tagged with GPS loggers from the manufacturer e-obs GmbH (Grünwald, Germany) from 2012 to 2018 in an approximately 1000-km2 study area around Weimar (Federal State of Thuringia, Federal Republic of Germany, 50° 59' N, 11° 19′ E). These loggers, weighing 15–25 g, had a rechargeable battery charged by a solar panel as well as a GPS module and, among other sensors, a three-dimensional accelerometer (ACC). The loggers were programmed to determine and store the bird's location via GPS from approximately 1 h before sunrise to approximately 1 h after sunset. The timing of the fixes was determined by the state of charge of the battery. If the minimum charge level was reached, GPS fixes were made every 20 min. At medium charge, fixes were made at 5-min intervals. When the battery charge was very good, 12 of the transmitters (on 7 males and 5 females) were able to activate the GPS module continuously and record location fixes every second (so-called 1-Hz fixes). The integrated accelerometer was programmed to record at least 78 readings at a frequency of 20 Hz per axis every 5 min, providing information about the bird's movement and position. Since ACC measurements consume very little energy, they still provided data even when GPS tracking had been reduced or turned off due to low battery voltage.
This evaluation included a total of 11,473,174 GPS locations provided by the loggers from 2012 to 2018 during the birds’ stay in the breeding area. The abundance of Red Kites in the study area was approximately ten breeding pairs per 100 km2 (Pfeiffer 2012).
Breeding season phases
We calculated the time intervals of the individual breeding season phases per brood from the fledgling stage (usually determined by wing length during ringing) (Pfeiffer and Meyburg 2015). The breeding season phases, based on the hatching date (H) of the oldest fledgling, were calculated as follows: territory occupation and courtship: 1 March to H-34 days; incubation of the clutch: H-33 to H-1 days; small fledglings in the nest: H to H + 25 days (female broods and guards); large fledglings in the nest: H + 26 to H + 53 days (female increasingly participates in foraging); and postfledging-dependent period: H + 54 to H + 75 days.
Determination of the activity phase
To determine the morning onset of activity, we first determined the roosting site for the previous night and then searched for the first GPS fix that occurred more than 200 m from the roost site. This ensured the exclusion of both erroneous measurements due to inaccurate GPS locations and small-scale changes in the roost tree, which sometimes occur due to disturbances during the nocturnal rest period. Analysis of many individual data points showed that a GPS fix at this distance from the roost almost always represented the beginning of the first hunting flight. The last roost was considered to be the position of the first location fix of the day if it was less than 200 m away from the evening roost of the previous day or no further than 200 m from the eyrie. The beginning of activity was considered to be the midpoint in time between the last location at the roost and the first location that was more than 200 m away from the roost. The analysis included only those days on which GPS fixes were made during the period between the onset of activity and sunrise at 5-min intervals or more often.
Similarly, the end of activity was calculated as the midpoint in time between the last fix in the activity phase and the first location at the evening roost. Here, the last fix of the day was assumed to be the evening roost if it was less than 2 h before or after sunset. Additionally, it could not be more than 200 m from the next morning’s roost. Again, only those days on which there were consistent GPS fixes at least every 5 min from the end of activity until sunset were considered. Only data from years in which the respective birds successfully raised fledglings were used. For females, the period of incubation of the clutch and rearing of small young was not included in the evaluation, as females fly very little during this period. The transmitters were thus only partially charged, and consequently, there were only a few days with location fixes every 5 min around sunrise and sunset. As there were not enough fixes in the morning and evening hours from one transmitter, only the locations of 28 Red Kites (18 males and 10 females) were used for this part of the evaluation. In total, data from 2093 bird-days and 2478 bird-days were available for the determination of the onset of morning activity and the end of activity, respectively.
Determination of flight activity
A decision tree (Supplementary Information (SI) Fig. S1 for details) was used to determine if a bird was flying at a particular time. Primarily, the information from the acceleration (ACC) sensor (Nathan et al. 2012) and, if necessary, the data from the GPS module, were analysed. The former was significantly influenced by the earth’s gravity and allowed the formation of conclusions about the position of the logger and thus the bird’s posture. By assessing the pitch of the head–tail axis, it was immediately possible to tell whether the bird was sitting upright or was in a largely horizontal flight posture. There was a possibility of confusion only between gliding and incubating birds in the eyrie. In this case, the activities were differentiated by the distance from the nest and the time window of incubation. If a transmitter was in continuous GPS mode (1 Hz tracking), no ACC measurements were collected. In this case, however, the bird’s behaviour could be easily identified from the continuous GPS locations (distance to the previous location, speed, accuracy of the speed and flight altitude above the ground). For each hour in the activity phase and at 5-min intervals, whether the bird was in flight was determined. This allowed us to investigate at which times of the day kites fly and how frequently. In each case, the first location or measurement in the 5-min interval was used for the evaluation. For the 1-Hz fixes, only the fifth location was assessed, as it provided more accurate results than a single measurement or the first measurement of a 1-Hz sequence. Only those hours for which GPS fixes or ACC measurements existed for all 12 5-min intervals were included in the analysis. A total of 10,451 bird-days were included in the analysis, for which 12 measurements were available for at least 1 h.
Determination of the flight altitude
For the determination of the flight altitude, a total of 8,798,033 fixes from breeding birds that were in flight were analysed. The altitude data related to each fix, which were provided by the GPS modules in the transmitters, referred to the WGS84 ellipsoid. Since the terrain elevations determined via the Google Maps elevation service (Google Maps platform) for each GPS point referred to mean sea level, a corresponding correction for the geoid undulation was essential (Poessel et al. 2018). However, this was constant within the study area and was determined using 1:25,000-scale topographic maps from the Thuringian State Survey Office. The aboveground flight altitude was then determined from the adjusted GPS altitude minus the terrain elevation.
For all the loggers, initial tests were performed at a location with a known position and altitude before they were attached to the bird. The goal was to verify functionality and measurement accuracy. In addition, the ACCs were calibrated. The tests showed that the mean values of the altitude measurements were quite accurate and never deviated from the reference altitude by more than 1.6 m for all transmitters. However, the scatter was relatively high, and there were outliers of several 100 m. This was especially true for single location fixes or the first fix of a continuous 1-Hz measurement sequence; the standard deviation (SD) was 48 m. The scatter of the altitude readings for the 1-Hz sequences became progressively smaller with increasing temporal distance from the beginning of the sequence. For all test fixes from the 11th second after the start of a 1-Hz sequence, the SD was only 5 m (SI Fig. S8). Therefore, for the comparison of subgroups (e.g., sex, different breeding season phases) and associations with weather data, only the more accurate 1 Hz data from the 11th second (1 Hz–10 s data) were used for model construction. In addition to the content-related limitation of the value range (exclusion of nonpositive flight altitudes, which can arise due to measurement errors in the difference of the GPS altitude values adjusted for geoid undulation and terrain height), a sex-specific limitation of the upper value range was accounted for by considering that 1% of the highest flight altitudes were potentially errors and thus excluded. The total number of 1-Hz flight altitude data points used in the analysis was thus 6,676,742 fixes.
However, 1-Hz sequences occurred only when the transmitter battery was in a good state of charge, which in turn depended on how often the bird flew and whether the transmitter was exposed to sufficient sunlight. To determine the size of the deviations in the 1-Hz flight altitudes compared to data that were collected in good and bad weather conditions, an additional evaluation of a further data set was carried out; this evaluation considered only locations where the bird was flying and time intervals of at least 5 min. Such fixes were largely available from most birds except for a few females during the period of incubation of the clutch. After cleaning the data to remove computationally nonpositive flight altitudes and recognisable outliers, 319,739 altitude values were included in the data set.
To investigate the influence of precipitation, sunshine duration and wind speed on activity and flight altitude, these meteorological factors were additionally included in the data analysis. Weather data were obtained by querying the Environmental Data Automated Track Annotation system (Env-DATA) (Dodge et al 2013) in Movebank (Kays et al. 2021) for the respective data points, which provided data from the European Centre for Medium-range Weather Forecasts (European Centre for Medium-range Weather Forecast 2011). The weather parameters were average values over a model grid box of 0.75° and 3-h time steps. The reference area corresponding to the study area was approximately 53 × 83 km. Only wind speeds up to 10 m/s were included in the analysis since faster speeds were rare, and thus, the number of data points was too small for evaluation.
The processing of the raw data and the statistical analyses were performed with SAS 9.4 software (SAS Institute Inc., Cary, NC, USA). All statistical significance tests were carried out considering a five percent error probability. Due to the exploratory nature of the present study, correction for multiple testing was not foreseen.
Determination of daily activity start and activity end
The correlation between sunrise and activity start and between sunset and activity end was analysed using Pearson correlation and Spearman rank correlation analyses. A comparison of the start and end of observed activity, representing the temporal difference between sunrise and the first measured activity and the temporal difference between the last measured activity and sunset, respectively, was performed using a mixed linear regression model considering breeding season phase and sex as fixed factors, as well as the time of sunrise and sunset. Individual differences between animals were accounted for in the model by adding the animal as a random factor. Differences between subgroups (e.g., sexes) estimated from the regression model were expressed directly as temporal differences and represent the differences in estimated subgroup-specific daily activities.
Flight activity in breeding season phases
The comparison of the relative flight activity, which in each case represents the proportion of readings “in flight” among the available measurements, was performed using a Poisson regression model. In this model, “readings in flight”, which were represented by count data and were considered the dependent variable, was described by the linear combination of the independent explanatory variables of sex, breeding season phase, and hourly interval. Additionally, the interaction of breeding time phase and sex was analysed. The observed number of measurements “in flight” depended on the number of available measurements (i.e., “in flight” and “at rest”). Thus, from a content point of view, the proportion was modelled instead of the absolute number. This was realised by considering the different number of underlying measurements on the individual aggregation level as offsets. Several observations of an animal were included in the model (e.g., several breeding phases; several hourly intervals per breeding phase). Accordingly, modelling of repeated measures was performed; observations from different individuals were assumed to be independent, whereas multiple responses from the same subject were assumed to be correlated. The differences between subgroups (e.g., sexes) estimated from the Poisson regression model are expressed as relative risks (RRs) and represent the quotient of the estimated subgroup-specific flight activities.
Accuracy of GPS altitude measurements
The analysis of the test measurements, based on altitude differences, was primarily descriptive. The variation in the altitude difference related to a specific point in time was additionally determined as residual scatter in a linear model with a random logger effect to detect logger-specific measurement differences. There were no significant differences among loggers in the current study.
The comparison of flight heights was performed using a mixed log-linear regression model (log-transformation: natural logarithm) considering breeding success and sex as independent explanatory variables. In addition, the interaction of breeding success and sex was analysed. The fact that the data were dependent observations (multiple observations in different years per animal) was taken into account as a random effect (transmitter number and year). Data from nonbreeders were not included in the analysis, as these data were available for only one male. Differences between subgroups (e.g., sexes) estimated from the regression model are interpreted as ratios (due to log scaling).
Meteorological influencing factors
In the data analyses, the weather data related to precipitation, sunshine duration and wind speed were additionally considered, i.e., in addition to the primary analysis, meteorological influences were included in corresponding models as additional explanatory variables. The results provide information about the estimated effect size as well as the statistical significance of the meteorological factors. In addition, sex-specific models derived from a limited database comprising either males or females were constructed.