Spatial use of non-breeding sites by adult GPS-tracked Ospreys Pandion haliaetus from Germany

Recent analyses of telemetry data on Ospreys Pandion haliaetus predominantly address migratory movements, whereas studies focusing on spatial use are rare, especially concerning the African non-breeding areas. We analysed GPS telemetry data of 15 adult Ospreys breeding in northeast Germany, assigned to 37 non-breeding events at non-breeding sites in Africa and south-western Europe. We calculated seasonal home ranges and investigated daily movements as well as overnight roosts. The females arrived in Africa about one month before the males. The home ranges of males correlated with those reported for their respective breeding seasons and were similar in size to those of the females. Half of the non-breeding sites were established on rivers, 36% on the coast and 14% on lakes, with no difference between sexes. Ospreys tracked for more than one year always returned to the same non-breeding sites. Daily home-range sizes varied during the non-breeding period, with the largest home ranges found in January, in parallel with the dry season. The average number of nights per overnight roost was lowest for river habitats and was related to the overall home-range size of the individuals. We highlight the importance of non-breeding sites for the survival of populations and suggest further studies to detect and mitigate threats to migratory bird species.


Introduction
Therefore, knowledge of the characteristics and use of non-breeding sites and migration routes is as crucial as that of breeding areas for the conservation of migratory species.
Breeding and non-breeding sites can be thousands of kilometres apart, and individuals from the same population can be located in widely separated regions during the non-breeding season, making it challenging to comprehensively monitor a population.For instance, ring recoveries and geolocators revealed a broad non-breeding range from central Europe to West Africa for Willow Warblers Phylloscopus trochilus breeding in Scandinavia (Hedenström and Pettersson 1987;Lerche-Jørgensen et al. 2017), or from West Africa to Namibia and Mozambique for Common Terns Sterna hirundo breeding in northern Germany (Piro and Schmitz-Ornés 2022).Some species, such as Lesser Spotted Eagles Clanga pomarina, Black Kites Milvus migrans, Eurasian Hobbies Falco subbuteo, and Red-crested Pochards Netta rufina are not stationary in their non-breeding areas and show considerable withinwinter movements (Keller et al. 2009;Meyburg and Meyburg 2009;Meyburg et al. 2015).The use of GPS telemetry enables researchers to track single individuals along their annual cycle and provides accurate spatiotemporal locations which allows the calculation of home ranges and the determination of ecological parameters on the individual level (e.g.Meyburg and Meyburg 2013;Tonra et al. 2019;Campion et al. 2020) as well as a better understanding of the genetic consequences of a large-ranged non-breeding distribution.
Raptors have been a group of choice for telemetry studies in recent decades, including the Osprey Pandion haliaetus (e.g.Alerstam et al. 2006;Vansteelant et al. 2015;Crawford and Long 2017;Anderwald et al. 2021).As most of these studies focus on migration, the migratory paths and spatial locations of the non-breeding sites are well-described, especially for Osprey breeding populations from North America and Europe (e.g.Hake et al. 2001;Martell et al. 2001;Bedrosian et al. 2015;Monti et al. 2018;Meyburg and Holte 2023).Mediterranean Osprey populations have also recently been the subject of telemetry studies (e.g.Monti et al. 2021;Montillo et al. 2022).However, analyses of telemetry data from long-distance migratory Ospreys focusing on spatial utilisation, especially on the non-breeding grounds, are still scarce.Washburn et al. (2014) reported that North American Ospreys stay at river systems, lakes and coastal habitats in different proportions during non-breeding seasons, similarly to Ospreys breeding in Scotland (Crawford and Long 2017).However, dynamics in spatial use and possible effects of the habitat type on movements within the non-breeding sites have not yet been reported in detail.
This study implemented a GPS telemetry data analysis to investigate the non-breeding sites and the spatial use of these sites by adult Ospreys breeding in northeastern Germany.We analysed home-range sizes, habitat types (river, lake, coast), daily movements and overnight roosts.Owing to the absence of parental duties, such as incubation, nest defence or feeding of offspring, which are unequally shared between the sexes during the breeding season (Poole 2019;Meyburg et al. 2023), we did not expect differences between the sexes in home-range sizes or in habitat selection during non-breeding.Because the proportion of water surfaces was assumed to differ between the habitat types, we expected a relationship between home-range sizes and habitat type.We expected dynamics in daily home ranges across the sojourn with larger home ranges at the beginning and at the end of the non-breeding season in terms of resettlement and reorientation.We hypothesised differences in the numbers of overnight roosts between habitat types because they were assumed to be linked to home-range sizes and to the number of potential roost trees which may vary between habitat types.

Data collection
Between 2006 and 2011, 17 adult Ospreys were trapped at breeding sites in the area around Lake Müritz in northeastern Germany (53°77′ N, 12°39′ E, Mecklenburg-Western Pomerania) using a dho-gaza (collapsible mist net: Zuberogoitia et al. 2008), with a live adult White-tailed Eagle Haliaeetus albicilla as a decoy.Each bird was fitted with a GPS/GSM 30g satellite transmitter (platform transmitter terminals, PTTs; Microwave Telemetry, Columbia, Maryland), with the transmitter attached to a Teflon harness.The PTTs recorded one GPS fix per hour between 2:00 and 23:00 UTC.
Before 2006, six other individuals had been equipped with a 30g Argos satellite transmitter (Microwave Telemetry, Columbia, Maryland).However, the temporal resolution and the spatial accuracy of the obtained data were not sufficient for home-range analyses.Therefore, the records of these birds were not considered in the spatial and statistical analyses here, but we present the locations of the non-breeding sites of these individuals.

Spatial and statistical analyses
We analysed the GPS data of 15 individual Ospreys, assigned to 37 non-breeding events in the period between arrival and departure at the respective non-breeding site in the given years.The term 'year' refers to the respective non-breeding season (October to March), although two calendar years are involved.Differentiation between migration and non-breeding season was conducted by visual inspection of the GPS fixes.Because the Ospreys navigated accurately to their non-breeding sites and did not show any post-or pre-migratory movements (in contrast to Western Marsh Harriers Circus aeruginosus: Strandberg et al. 2008), the beginning and the end of the non-breeding season was obvious.
Home-range areas were calculated using kernel density estimation (KDE) and minimal convex polygon (MCP) to keep the results comparable to those of other studies which used at least one of these home-range metrics.For both metrics, we calculated the home ranges at 95%, 80% and 50%.We determined three types of non-breeding habitat in accordance with Washburn et al. (2014) -coast, river and lake -and classified the non-breeding sites with respect to the main waterbody within the respective KDE50 home range by visual inspection of satellite imagery.A Fisher's exact test for count data was applied to the number of non-breeding habitats detected per sex (i.e. one count per individual) to test for differences in habitat selection between sexes.Using linear mixed-effects models (LMEs), the possible effects of habitat type and sex on home-range size were investigated.For each of the six calculated home-range metrics (KDE50, KDE80, KDE95, MCP50, MCP80 and MCP95), two models were fitted that included either the untransformed home ranges or the log10-transformed home ranges as the dependent variable.Since data from multiple non-breeding seasons was available in most of the individuals, the IDs of the individuals and the year were taken as random factors.In all modelling processes, the optimal model was identified by means of the Akaike information criterion (AIC) and the model with the lowest AIC value was selected.
The KDE95 home ranges of males at the non-breeding sites were tested as a function of those at the breeding sites (Meyburg et al. 2023), using an LME with the individuals' ID and the year as random factors.Females were not considered here because they have highly dynamic home ranges during the breeding season, with small home ranges during incubation and the first nestling stages and sometimes disproportionately large home ranges just after the young have fledged (Meyburg et al. 2023).
The data resolution of one record per hour was not appropriate for tracking an individual during a day in terms of identifying single foraging trips or estimating flight distances and durations, particularly when single data points are missing, such as because of transmission or positioning failure.To still investigate daily flight efforts, we calculated daily home ranges (for all six home-range metrics) for each individual when at least 10 data points per day were available (see Table 1 for number of fixes), although, with respect to the home-range analyses, the number of 10 data points is low.However, Ospreys are reported to be stationary during the non-breeding season, with limited ranges and only local movements (e.g.Prevost 1982;Washburn et al. 2014), andMitchell et al. (2019) have shown for KDE that priority habitats can be identified with an infrequent sampling strategy in terms of lower fix rates, which was assumed for MCP estimations as well.A Pearson's correlation test was applied to test for correlation between the mean of daily home ranges per non-breeding season and the corresponding total home ranges (KDE95) of individuals.Subsequently, we fitted LMEs on daily home ranges (KDE50, KDE80, KDE95, MCP50, MCP80, MCP95) with habitat type (coast, lake, river) and sex as explanatory variables as well as ID as a random factor.Daily home ranges had been log10-transformed before fitting the model and the model with the lowest AIC was selected as the best model.
Similarly, six generalised additive mixed models (GAMMs) were fitted to test for dynamics in daily home-range sizes throughout a non-breeding season.Daily home ranges were log10-transformed and used as the dependent variable.As the explanatory variable, the dates of data records were transformed into day of the year (DOY) format, centred on 31 Dec = 0 (30 Dec = −1; 1 Jan = 1, etc.), and a smoothing term was added.The individuals' IDs were taken as a random factor and the model with the lowest AIC was selected as the best model.The PTTs were solar powered and programmed not to record locations during the night.Therefore, a direct analysis of possible night activities or overnight roosts was not viable.Nevertheless, we were able to identify single overnight roosts by comparing the last location in the evening to the first in the subsequent morning.Sites, whereby the morning location lies within a buffer zone with a radius of 20 m around the last evening location and the bird was not in flight, were considered overnight roosts.Closely neighbouring roost sites with overlapping radii were taken as one overnight roost.The radius of 20 m was intended to account for inaccuracies in positioning by the PTTs and short-distance movements during the night, such as roost changes within one or between neighbouring trees, on the one hand, and to exclude larger movements, such as changes in roosting sites, on the other hand.Because the number of nights for which a determination of overnight roosts was possible (afterwards 'roost nights') varied between the individuals and years, we calculated the mean value of nights spent by an individual per identified overnight roost (number of roost nights divided by the number of roosts).We fitted four LMEs with the mean nights per overnight roost as the dependent variable, and habitat type, all six metrics of total home-range sizes (KDE50, KDE80, KDE95, MCP50, MCP80, and MCP95 of the total non-breeding season) and sex as explanatory variables as well as ID as a random factor.To account for possible effects of a low number of roost nights on the mean nights per roost, four data subsets were used (one for each model) based on the minimum number of roost nights: ≥1 (full dataset), ≥5, ≥10, and ≥30 roost nights.We log10-transformed both the mean nights per overnight roost and the home-range variables due to the best model performance (based on the AIC).Model selection was conducted for each of the four models by stepwise deletion of parameters (based on the AIC) without removing the random term.The four best models were compared based on their AIC values and a final model was selected.To assess the spatial distribution of overnight roosts, the proportion of detected roosts within the individual's KDE50 (as the core area at the non-breeding range) was calculated.
We used R 3.5.2(R Core Team 2018) to perform all spatial and statistical analyses.Satellite images were downloaded via the package 'OpenStreetMap' (Fellows 2019).

Results
The non-breeding sites of the tracked Ospreys from Germany were located in Central and West Africa (Senegal, The Gambia, Guinea-Bissau, Guinea, Mali, Burkina Faso, Ivory Coast, Nigeria and Chad), as well as in south-western Europe (Portugal and Spain) (Figure 1).One male did not travel directly to its non-breeding site but rested for about six weeks on the northern coast of Senegal before heading for the final non-breeding site in southern Senegal; this male was shot only a few days after arrival at the non-breeding site.Another male died two days after arrival at the non-breeding site in Guinea-Bissau due to unknown circumstances.The Ospreys arrived at the non-breeding sites on average on 30 September (females_ Africa = 25 September; males_Africa = 25 October; males_ Southwest Europe = 10 September).After an average duration of 186 days (females_Africa = 192; males_ Africa = 172; males_Southwest Europe = 200), individuals departed from the non-breeding sites on 5 April (females_ Africa = 5 April; males_Africa = 15 April; males_Southwest Europe = 30 March).
The Ospreys stayed at coast (36%), lake (14%) and river (50%) habitats (see Figure 2 for examples) during the non-breeding season (one count per individual).Males and females did not choose the habitat types differently (p = 0.767) (Table 2).All four males and five females for which data were available across more than 1 year returned to the same non-breeding sites in each year during the study.
The home-range sizes at the non-breeding sites varied between 2.38 and 611.24 km² (KDE95: mean = 117.93km², SD = 149.07km², median = 47.82 km²), excluding one female (ID81339) which showed a considerably large home range (857-1 148 km²) in all four tracked years (Table 2; Figure 3a).The best LME on home-range size (∆AIC = −17.44 to −1.45) included the log-transformed KDE95 home ranges as the dependent variable and showed no significant effect of habitat type or sex on home-range size (Table 3).The non-breeding home ranges of males were similar in size to those in the previous breeding season but different among individuals (Supplementary Table S1; Figure 3b).
The means of daily home ranges were not correlated with the total home ranges (p = 0.265, t = 1.132, df = 35, cor = 0.188).In the optimal model on daily home ranges (∆AIC = −794.12to −30.57), we found significant differences in daily KDE50 home ranges between the habitat types, with coastal non-breeding sites having smaller daily home ranges than lake and river habitats (Supplementary Table S2; Figure 4a).The optimal GAMM (∆AIC = −817.65 to −32.42) included the KDE50 home ranges and revealed a nonlinear relationship between DOY and size of the daily home ranges (Supplementary Tables S3  and S4).After arriving at the non-breeding sites, the daily home ranges decreased from early October to a minimum in mid-November.Afterwards, the daily home ranges increased to a maximum in early January and decreased again until early March (Figure 4b).
The mean number of roost nights was 61 (SD = 37.6, range = 1-133).The number of overnight roosts detected per individual ranged from 1 to 41 (mean = 9.2, SD = 7.59).The average number of nights that an individual spent per overnight roost ranged from 1 to 23.4 (mean = 7.7, SD = 5.1).All four best models on the mean number of nights per roost for the four different subsets included the log10-transformed KDE95 home ranges as a fixed effect but varied in the composition of the other fixed effects (Table 4).The model with the lowest AIC value considered the subset with at least 10 roost nights as well as the habitat types and the KDE50 and KDE95 home ranges as fixed effects (Table 4).The habitat type river and home-range size KDE95 negatively affected the mean nights per overnight roost, while the effects of KDE50 and sex were not significant (p > 0.05) (Figure 5; Supplementary Table S5).The results of the other models are presented in Supplementary Tables S6-S8.The proportion of detected overnight roosts that are located within the respective KDE50 home ranges was 69.8%.

Discussion
Most adult Ospreys from northeastern Germany sojourn in sub-Saharan West and Central Africa in the non-breeding season, and a few in the Iberian Peninsula, which is in line with other Osprey populations from most of Europe (Hake et al. 2001;Crawford and Long 2017;Monti et al. 2018;Montillo et al. 2022).In contrast, Väli and Sellis (2016) showed non-breeding sites of GPS-tracked Ospreys from Estonia which took the Eastern European-East African flyway in Central, southern and East Africa.The females arrived at the non-breeding sites approximately one month earlier than the males that stayed in Africa for the non-breeding season, which may be due to an earlier departure from the breeding sites by females (Poole 2019;Meyburg et al. 2023), as also reported for American Ospreys (Martell et al. 2001;Washburn et al. 2014).The earliest arrivals were recorded for males that stayed in south-western Europe for the non-breeding season, which were considered short-distance migrants.
In contrast to the findings of studies on American Ospreys, we found no differences between the sexes in terms of the geographic location of non-breeding sites.We found that the Ospreys visited the same non-breeding sites with high fidelity in subsequent years, which is in line with prior studies on Ospreys from Europe and America (e.g.Prevost 1982;Martell et al. 2001;Alerstam et al. 2006;Zwarts et al. 2009;Bedrosian et al. 2015).Although Ospreys are piscivorous and, thus, bound to aquatic ecosystems, they can exploit both freshwater and seawater habitats, which allows them to utilise a wide geographic range (Poole et al. 2002;Glass and Watts 2009).We found that half of the non-breeding sites in our study were river habitats, which were preferred by the birds over coastal and lake habitats in a similar proportion to New World Ospreys (Washburn et al. 2014), although we found no differences between the sexes.However, some river habitats were found to be near estuaries and in some cases, they also included coastal habitats.
Recent studies on other migratory raptors have shown that non-breeding sites can cover wide ranges on the individual level and can include 'within-winter movements' of long distances (Meyburg and Meyburg 2009;Trierweiler et al. 2013).In contrast, Ospreys are assumed to be stationary while in the non-breeding phase (Prevost 1982   to reduce the risk of mortality and to maximise energy conservation (Prevost 1982;Washburn 2014;Montillo et al. 2022).However, we found dynamic spatial use across the non-breeding period, which has not been detected in American Ospreys (Washburn et al. 2014).We suspect that the higher flight activity in the first days after arrival reflects the process of resettlement and reorientation at the non-breeding sites in terms of hunting and roosting areas.African Fish Eagle Haliaeetus vocifer (Prevost 1982;Zwarts et al. 2009).The largest daily home ranges, however, were detected in the period from mid-December to late January, which was unexpected, given that the Ospreys had been established at their non-breeding sites for several weeks at that time.We suggest that this increase of daily home ranges is a consequence of the dry season, which culminates in West Africa at this time (Knippertz and Fink 2009) and might coerce those birds staying at lake and river habitats to expand their foraging trips.Nevertheless, the Ospreys did not leave their non-breeding sites but remained stationary, indicating that Ospreys generally select non-breeding sites in sub-Saharan Africa that support them throughout the non-breeding season, including the dry season.This contrasts, for instance, to Montagu's Harriers Circus pygargus which shift their non-breeding sites to follow the highest prey availability (Trierweiler et al. 2013).Furthermore, we have shown that non-breeding home ranges of male Ospreys are similar in size to their respective breeding home ranges, which underlines the importance of the individual character of the birds rather than only energy conservation constraints.Prevost (1982) reported that Ospreys staying in Senegal and The Gambia during the non-breeding season had different daily movements depending on habitat.Coastal areas dominated by mangroves provided both fishing grounds and perching and roosting trees in close proximity, whereas in treeless areas Ospreys needed to cover longer distances between foraging and roosting sites.These results support our findings that daily home ranges are smallest in coastal areas, assuming that inland habitats along rivers and lakes are less rich in tree vegetation near the hunting grounds, which in turn might differ in prey fish abundances compared with in coastal areas, including estuaries.However, our result that the number of nights per roost were lowest in river habitats indicates that river habitats may provide a variety of potential roosting trees, albeit at a greater distance from the hunting grounds than in coastal areas.The mean number of nights per overnight roost was also negatively affected by the home-range size, which was expected because large home ranges are supposed to provide more roosting opportunities.In the final model, sex was not significant, unlike other models with poorer model performance, such as the model considering the full dataset (subset1) and the model considering at least 30 roost nights (subset4).Both these models show a higher number of nights per roost for males than for females.Our model selection process was strictly based on model performance as shown by the AIC value, although other model-selection approaches might have included other variables in the optimal model (e.g.Stoica and Selén 2004).
In Ospreys, as well as in some other long-lived migratory species, juveniles do not leave the non-breeding sites for spring migration after their first non-breeding season but stay in or close to these areas for at least 1 year until they reach full maturity (Prevost 1982;Newton 2008;Poole 2019).This highlights the importance of non-breeding sites for the survival of populations in migratory species.However, most of the non-breeding range of European Ospreys is lacking in adequate land protection and conservation policies (McDonald and Boucher 2011; Montillo et al. 2022), and the threats to Ospreys in the African non-breeding areas are poorly understood.Furthermore, migratory paths and stopover sites often do not include protected areas (Crawford and Long 2017), with mortality reported predominantly outside of protected areas (Montillo et al. 2022).Many studies, including this one, reported the death of individuals during migration or at the non-breeding sites (e.g.Prevost 1982;Hake et al. 2001;Bedrosian et al. 2015;Crawford and Long 2017;Anderwald et al. 2021;Montillo et al. 2022).More research is needed to identify and to mitigate the threats to migratory bird species, such as Ospreys.

Figure 1 :Figure 2 :
Figure 1: Non-breeding sites in West Africa and south-western Europe of Ospreys Pandion haliaetus breeding in Germany.Data of GPS-tracked birds (squares) were included in the analysis.Data of ARGOS-tracked birds (triangles) were not included because of their inaccuracy (see Methods).Numbers 1-3 refer to examples of habitat types (satellite imagery; see Figure 2)

Figure 3 :Figure 4 :
Figure 3: (a) Mean (±SD) home-range sizes (KDE95) of male (dark) and female (light) Ospreys Pandion haliaetus during the non-breeding season.(b) Correlation between breeding home range (Meyburg et al. 2023) and non-breeding home range of male Ospreys.Symbols reflect individuals

Table 1 :
Tracking period (arrival to departure) and number of total and daily fixes at the non-breeding sites of GPS-tracked Ospreys Pandion haliaetus during the non-breeding season

Table 3 :
Fixed effects of the linear mixed-effects model on home-range area (metric KDE95, log10-transformed) by sex and habitat type of non-breeding season Ospreys Pandion haliaetus

Table 4 :
Model comparison of four linear mixed-effects models with different data subsets after stepwise model selection (see Methods) on mean nights per overnight roost (log 10 -transformed) by habitat type, total home range (log 10 -transformed) and sex as fixed effects and ID as a random effect in Ospreys Pandion haliaetus during the non-breeding season.Subsets are based on the number of roost nights.Subset1: ≥1 roost nights (full dataset), subset2: ≥5, subset3: ≥10, subset4: ≥30.Bold font denotes significant variables in the best models.Random effect = standard deviation of the residuals (Kristensen et al. 2013)ver, this high estimate is likely biased by single outliers, as many individuals did have small home ranges (seeMontillo et al. 2022for details), similar to our findings.Stationarity during the non-breeding season has also been reported for other migrating bird species, such as the Common Redstart Phoenicurus phoenicurus(Kristensen et al. 2013), and is suggested