The Characteristics and Variation of the Golden Eagle Aquila chrysaetos Home Range

Abstract

Satellite tracking allows for novel investigations into golden eagle home range characteristics. Understanding home range characteristics is important for conservation and for assessing the potential impact of landscape changes from forest planting, wind farms, etc. Small sample sizes, inconsistent definitions and methods restricted several previous studies. Our study involved 69 resident tagged eagles with over one year of data across five Scottish regions. Home range size was estimated from 95% isopleth contours extracted from Utilisation Distributions. Above a small threshold, estimated range size was not affected by the number of records but at least one year of data is required, largely because of the breeding and non-breeding seasonal differences. There were no significant range size differences between birds tagged as range holders and those previously tagged as nestlings. Across four regions, with considerable intra-regional variation, planar 95% isopleths did not differ (medians, km²): Argyll 58.9, Northwest Highlands 61.7, Northeast Highlands 89.3, South of Scotland 91.9. Ranges in the isolated Outer Hebrides region were exceptionally small, at 24.0 km². Estimated range area was usually reduced to 70–80% of the planar area when restricted to usable habitat, as estimated by the Golden Eagle Topography (GET) model. Applying measures of known unsuitable habitat (closed-canopy commercial forest and wind turbines) further reduced usable open land. Loss of otherwise suitable habitat was substantially due to commercial forest. Larger ranges had larger extents of suitable habitat (according to GET), with no apparent optimum of preferred GET habitat. Range size was not different across a year between the sexes. Breeding ranges were smaller, and females’ breeding ranges were much smaller than those of males, but larger than males’ ranges in the non-breeding season. Breeding attempt duration was probably also influential. Our study provides novel insights into golden eagle home range characteristics and can guide further research and practical applications.

1. Introduction

Understanding the characteristics of home ranges is important for golden eagle Aquila chrysaetos conservation and reintroduction programmes (e.g., [1]), but also for assessing the potential impact of anthropogenic landscape changes from, for example, planting of forests and the construction of wind farms (e.g., [2]). Like many large raptors the golden eagle is vulnerable to human influences, such as land-use changes and intentional disturbance, or persecution, which can have a negative impact. Predictions of future populations and the consequences of habitat and land management changes must include discussions of what needs to be within or excluded from home ranges.

GPS satellite tracking has allowed novel investigations into the characteristics of golden eagle home ranges. Previous analyses of the home range are, however, fraught with difficulties, for two reasons. First, ‘range’, ‘home range’ and ‘territory’ are sometimes used interchangeably and without clear definition and, second, measures of the size and shape of home ranges utilise a range of techniques that can lead to very different results.

In Scotland, breeding pairs of golden eagles occupy home ranges year-round and, usually, until death. Parts of the home range are actively defended seasonally as an exclusive territory [3]. Territory is sometimes used interchangeably with home range. To avoid confusion, territory should be confined to ‘a defended area’ [4], even if exclusive use and a limited overlap of neighbouring boundaries may imply defence or, alternatively, mutually respected boundaries. A territory is invariably much smaller than the home range. Aggressive defence depends on a bird’s immediate underlying physiological state (the immediate motivation), so it varies over time (even from hour to hour) and position in a home range. Territory descriptions should be restricted to spaces delimited by intraspecific agonistic interactions, which means those spaces are extremely difficult to define. Moreover, it is unclear what role golden eagles’ undulating flight displays may serve in the defence or advertisement of a territory, especially when they are not limited to locations close to presumptive adjoining boundaries [5]. Analyses should ideally, therefore, consider the home range, although they may infer territoriality if there is limited overlap between neighbouring home ranges.

Home ranges include aspects of the local environment that an individual becomes familiar with and learns physical locations within the area that are important for its survival. An occupier of a home range forms a cognitive map and individuals that are more successful in spatial mapping their home range show increased survival [6]. Since an animal’s requirements can differ from those of other individuals and also change over time, a measurement of a home range is only ever a snapshot of an individual’s behaviour and descriptions of the typical home range of a species can only be indicative [7,8,9].

Unfortunately, different techniques have been used historically to measure home range size and to map out their shapes. There is an implicit assumption in many studies that sufficient data have been used but how many data are needed to estimate a golden eagle’s home range? In some studies, this problem is glossed over while others consider it in more detail. Most use “statistical” methods to validate their results. An underlying premise is that more time or fixes, which may not be equivalent, should provide estimates that plateau once the true range extent is reached. This accords with using measures of variance across time, or across an accumulated number of records, which is based on a fundamental assumption that home range extent estimates become more asymptotic (stable with lower variance) with more data accumulation.

In other modelling involving home range, this issue was not addressed explicitly but it was noted that the breeding season home range size increased with increasing number of tracking days [10]. It was also noted that, in contrast to home range size, core area size was not influenced by the number of tracking days. Hence, this metric was considered as a potentially influential covariate in models (even if, biologically, it has much less relevance) [10]. Conclusions in this study are complicated by the use of data from several different age and sex classes.

Few papers have considered both time and number of records as to when a reliable home range estimate may have been reached, with a notable exception [11]. However, a possible problem with this paper [11] is that the time period considered as to when the invariant/low variation plateau is reached is short. It also has many internal contradictions which are either noted or ignored. This paper examined the prospect that, within 30 days of a territorial bird being tagged (during the non-breeding season), there may be sufficient data to estimate the bird’s home range. But the study seems to be examining how many days/data are needed to estimate home range use within this restricted window of 30 days [11]. Nonetheless, this is a useful study as it considered, from tag records, both accumulated time and accumulated records when, given data supply from different tag models and duty cycles, these two metrics can be different in the underlying data richness. All else being equal, one could envisage that a data-rich tag output could provide for the asymptotic ‘truth’ earlier than a data-poor tag output. Alternatively, when all else may not be equal (e.g., there are seasonal changes), it is the longitudinal (time) contribution that would be more important than the number of records per se.

Unfortunately, another study [12] is also flawed; first, by using the outdated minimum convex polygon (MCP) method to characterise a home range. Second, the study had no recourse to data beyond the longevity of the tail-mounted tags. As these tags were shed due to moult of the tail feathers, or birds otherwise dropping/removing the tags, there was no prospect of any analysis of home range use beyond this right-censorship caused by the tagging method. Nevertheless, most importantly this study, like [11], constrains the estimation of home range by an unambiguous right side (temporal) data censorship. Hence, it only estimates home range extent within a curtailed time window that may be inadequate as an estimate from a longer time period (and/or more records), and especially given results from other studies with more longitudinal data.

Data from ≤2 seasons or ≤2 years were considered for 10 eagles (eight male and two female) and used the Brownian bridge movement model (BBMM) to delineate home ranges (99% isopleths) [13]. This paper is the most in-depth consideration of estimation of golden eagle home ranges, but it is not easy to extract how long or how many data were needed to estimate a home range consistently. One implication from this study is that it takes more than one year of data to characterise an eagle range’s extent and it is unclear if the use of 99% isopleths, rather than 95% isopleths, contributes to this conclusion. Citing [13], it has been assumed that at least one year of telemetry data were necessary to obtain a good approximation of an ‘adult’ golden eagle home range [14].

A Swedish study [15] did not consider whether their records were sufficient to properly characterise a home range, using the biased random bridge approach, although they do assume that they were comparable by using a similar number of records and/or time. However, this presumption may be problematic, as they note that birds abandoned their home range if they failed. This study only considered the breeding season (March to August, inclusive) with six locations a day for each bird, between 04:00 and 18:00 h, but with a constraint that there must be ≥ 100 data points in a single month. This constraint is an implicit assumption that ≥ 100 data points in a single month was adequate to characterise a home range estimate, in any month. Although it is unclear if the use of a month with ≥ 100 fixes adequately and consistently estimates home range use, this seems secondary to this study’s objectives. The recognition that some threshold is relevant (i.e., ≥100 fixes) indicates that the study was aware of the issue.

Unfortunately, answering the question about how much data are needed is complicated by multiple confounding issues. Seasonality is such an issue. Modern GPS tags have solar powered batteries, which means that the location fix rate is dependent on the amount of charge, and, in Scotland as in other higher-latitude regions, there are marked differences in daylength across the year, even without accounting for the prevalence of sunshine in different months (see also [16]). For example, in Scotland there are approximately 7 h of daylight on 21 December and 17.5 h on 21 June (https://www.timeanddate.com/sun/@2638360; accessed 26 June 2024). Consequently, the number of daytime records will be greater during the breeding season, even if the fix rate is constant; at a constant fix rate there would 2.5 times more records on 21 June than 21 December. In reality, a lower sun and a shorter day, combined with less sunshine, will exaggerate this difference. If there are seasonal differences in range size it may be difficult to disentangle the effects of the number of records on the range size.

Many raptors are central place foragers during the breeding season as they need to bring prey back to the nest to feed offspring or, before offspring can thermoregulate, there is a tendency for parental role division with females more likely to incubate and brood young offspring, and hence males bringing food for the female too [3]. Golden eagle nest sites will not have the same focus outside of the breeding season [3]. Hence, an important aspect of range size is the potential for seasonal differences, particularly as the nest site is much less focally important to the pair outside of the breeding season.

It follows logically that, if range use is more extensive in, for example, the non-breeding season, then estimates of home range size derived only from the breeding season [15] will not be representative and data from both seasons are necessary [13,14]. If in this example, however, range use in the non-breeding season encompasses use in the breeding season, then data from only the non-breeding season may be adequate to characterise range size. Nevertheless, if seasonal differences exist, this casts further doubt that home range extent can be properly characterised using data restricted to time periods shorter than a breeding and/or a non-breeding season (e.g., [11,12]).

Golden eagle range size has been found to be smaller in the breeding season in northern temperate regions [13,17,18]. Range extent (MCPs) in southern Idaho, USA, increased from spring through summer to the fall ([19], n = 4 birds). Winter home ranges were larger than those in summer [10], but these were migratory birds in eastern North America. This seasonal pattern in resident birds was reversed in desert golden eagles, where high spring and summer heat occur during the breeding season [20].

Breeding status may also contribute to the magnitude of seasonal differences. If a pair does not attempt to breed, there is no requirement to remain close to the nest during the breeding season. A Kruskal-Wallis test was used [2] to determine if there were differences in range sizes with different reproductive outcomes. Unfortunately, the results of these tests are not reported, and all the study subjects had surprisingly poor breeding outcomes [2]. It has been suggested that breeding season ranges were larger when a pair did not breed [18,20], and it has also been suggested that breeding attempts and breeding success almost certainly influenced the range size of tracked eagles [10].

Finally, are there differences in the ranges of males and females in the same range? Sexual differences in home ranges do not appear to have been examined thoroughly, at least with a reasonable sample size. Two studies [2,15] found no difference between the sexes, but these papers referred only to the breeding season and are not independent studies, as they used the same, or a very similar, dataset. It is perhaps surprising that no sexual difference was found, as it is more expected in the breeding season (their study period), due to different parental roles between the sexes. However, it seems that all birds were unsuccessful breeders or did not lay eggs, which could explain the absence of a sexual difference.

It was recommended that males should be preferentially tracked [13] but this was not based on an analysis, which would have been difficult when eight of the ten tracked birds were male [13]. The presumption seems to have been that because females are more likely to spend many weeks at the nest, then males, as the main food provider during the breeding period, will range further.

To address the gaps and discrepancies in previous studies, and provide additional evaluations, our study used a sample of 69 GPS-tagged golden eagles in Scotland, each with at least a year of tag data, to ask six questions about golden eagle home ranges:

• How many data are needed to obtain a reliable estimate of range size?
• Are there differences between range characteristics of birds tagged as range holders and those originally tagged as nestlings who subsequently settle on a home range? (Birds tagged as range holders will tend to be older).
• What are the range sizes in absolute and usable terms and are they correlated or is there an optimum range size in usable terms (so there is a consistent usable range size across different absolute range sizes)?
• Do range sizes show marked regional differences?
• Are there seasonal differences in range sizes?
• Does range size differ between the sexes?

2. Methods

2.1. Study Area and Species

Scotland covers c. 80,000 km² on the northwestern limit of Europe and hosts over 500 resident golden eagle pairs occupying home ranges in Scotland’s uplands [21], which are also used by hundreds of non-territorial birds [22,23,24]. These uplands vary in geology, vegetation, topography, and climatic influences [24,25,26].

Climatically, situated on the northeastern edge of the Atlantic Ocean, the west of Scotland is subjected more to the North Atlantic Current and is wetter and windier, with more equitable seasonal changes in weather, and, hence, more oceanic. The east is drier, with greater seasonal change in weather and, climatically, more continental [3,24,25,26].

The contrasting oceanic/continental influences tend to produce upland vegetations that in the east are only found at higher altitudes but may occur at sea level in the west. The preferred open habitats are dry or wet heathland and peatland dominated by heather Calluna vulgaris and relatives in the east, with graminoids, sedges, and deeper peatland more common in the west [25].

Irrespective of vegetation, airspace is a critical habitat for soaring birds like golden eagles [27] and topography generates much of this habitat by affecting the available wind resources for efficient movement. A combination of altitude, slope, and distance from ridges robustly surrogates for orographic wind energy availability and is highly influential in eagle movements and distribution (Golden Eagle Topography (GET) model [22,28]).

Golden eagle biology in Scotland has been described elsewhere [3,29]. The density of occupied ranges in some parts of the Northeast Highlands (Figure 1) remains below the likely carrying capacity because of persistent illegal persecution and association with intensive management for driven shoots of red grouse Lagopus lagopus scotica (e.g., [23]). In the Northeast Highlands the diet tends to be dominated by red grouse and mountain hare Lepus timidus with a greater diversity of species taken in western regions, such as Argyll and the Northwest Highlands, and including seabirds in coastal ranges of the Outer Hebrides [3,30]. The South of Scotland supports relatively few occupied ranges [21], but numbers are expanding through recent translocations of birds from further north. Breeding productivity is usually highest regionally in the Northeast Highlands [3,21]. This is probably not because of a more specialised diet, but rather that the abundance of preferred prey is highest there [30].

2.2. Tag Data

Previous Scottish studies have described in detail the methods used to tag and sex nestling and range holding golden eagles and the absence of discernible adverse effects (e.g., [23,28]). All tags in our study were solar-charged transmitters, with most models manufactured by MTI (Microwave Telemetry Inc., Columbia, MD, USA) as 70 g GPS/Argos or GPS/GSM PTTs. The tags deployed in the Outer Hebrides included OrniTrack 50 g GPS/GSM models (Ornitela, Vilnius, Lithuania).

We had location data from 104 individuals; 38 birds who had transmitters fitted when already holding a range (Holder) and 66 birds who had transmitters fitted prior to fledging but subsequently settled in a range (Settled). As we intended to examine seasonal differences, we applied a preliminary filter and restricted our analyses to 69 tags with at least one year of tracking records (30 Holder and 39 Settled) (Figure 1).

The Holder tags were present in two regional groups: Argyll (n = 15) and Outer Hebrides (n = 15). The Settled tags were present in five regions: Argyll (n = 10), Northeast Highlands (n = 16), Northwest Highlands (n = 3), Outer Hebrides (n = 3) and the South of Scotland (n = 7) (Figure 1). In twelve presumptive ranges both the male and female were tagged, giving 57 with ≥ one tagged bird: Argyll (n = 20), Northeast Highlands (n = 16), Northwest Highlands (n = 3), Outer Hebrides (n = 12) and South of Scotland (n = 6). Tag details are in

Table 1. Tags used in analyses. The table is split into Holder tags (birds fitted with a tag when a range holding bird: n = 30) and Settled (birds fitted with a tag prior to fledging and subsequently settled: n = 39). There are three record dates: First—the first date of a record (notionally the fledged date for a settled bird: Settled) or the day after a tag was fitted to a range holding bird (Holder); Last—the last date for which data were available at the time of the analyses; and Settled—the date a bird tracked since fledging settled [31]. Status—the current status of the tag; tags that are no longer providing data are in italics and the reason for the end of data transmission is given (Died = a natural death, Malfunction = a tag that failed technically, Dropped = a tag that was shed from a bird through harness failure, and SNMF = a tag which stopped suddenly with no obvious sign of a technical malfunction, often associated with a likely persecution event [23]). There are two numbers of Tracked Days which are identical for Holder tags, and HR is the number of tracked days when a bird held a range. For Region (Figure 1): Argyll = Argyll, OH = Outer Hebrides, NE = Northeast Highlands, NW = Northwest Highlands and SOS = South of Scotland.

		Record Date				Tracked Days
TagID	Sex	First	Last	Settled	Status	All	HR	Region
Holder
102	F	30/11/15	04/07/21		Dropped	2043	2043	Argyll
103	M	29/01/15	10/05/24		Still tracking	3389	3389	Argyll
814	M	30/11/17	03/02/24		Died	2256	2256	Argyll
815	F	14/03/17	13/05/24		Still tracking	2617	2617	Argyll
816	M	01/03/17	12/05/24		Still tracking	2629	2629	Argyll
817	M	26/01/17	28/09/21		Dropped	1706	1706	Argyll
818	F	08/02/17	12/05/24		Still tracking	2650	2650	Argyll
991	M	09/02/18	13/05/24		Still tracking	2285	2285	Argyll
992	F	25/01/19	13/02/20		Died	384	384	Argyll
993	M	18/01/19	02/12/21		Malfunction	1049	1049	Argyll
994	F	06/12/18	02/04/24		Still tracking	1944	1944	Argyll
995	M	11/01/19	12/02/20		Died	397	397	Argyll
999	M	05/03/21	09/11/22		Still tracking	614	614	Argyll
1157	F	07/02/21	13/05/24		Still tracking	1191	1191	Argyll
99000	M	14/02/19	30/06/21		Died	867	867	Argyll
990	F	22/01/22	12/05/24		Still tracking	841	841	OH
1152	M	10/11/21	03/02/24		Malfunction	815	815	OH
1154	M	19/01/22	12/05/24		Still tracking	844	844	OH
1155	M	19/01/22	28/10/23		Still tracking	647	647	OH
226090	F	23/01/23	12/05/24		Still tracking	475	475	OH
226091	F	28/01/23	23/05/24		Still tracking	481	481	OH
226092	M	28/01/23	23/05/24		Still tracking	481	481	OH
226095	F	29/11/22	22/05/24		Still tracking	540	540	OH
226096	M	24/01/23	23/05/24		Still tracking	485	485	OH
226097	M	25/11/22	20/05/24		Still tracking	542	542	OH
226099	M	24/01/23	23/05/24		Still tracking	485	485	OH
226100	F	23/01/23	24/05/24		Still tracking	487	487	OH
226102	F	26/11/22	24/05/24		Still tracking	545	545	OH
226103	M	30/11/22	23/05/24		Still tracking	540	540	OH
226104	F	25/11/22	23/05/24		Still tracking	545	545	OH
Settled
100	M	01/08/14	13/05/24	22/03/19	Still tracking	3573	1879	Argyll
582	F	01/08/16	12/05/24	02/03/21	Still tracking	2841	1167	Argyll
584	M	01/08/15	12/05/24	16/10/21	Still tracking	3207	939	Argyll
928	M	01/08/17	02/02/24	29/04/21	Still tracking	2376	1009	Argyll
932	M	01/08/17	27/12/22	07/10/19	Malfunction	1974	1177	Argyll
933	F	01/08/19	18/01/24	20/04/22	Still tracking	1631	638	Argyll
1031	M	01/08/18	22/04/23	06/09/21	Malfunction	1725	593	Argyll
1091	F	01/08/19	23/01/24	10/09/22	Still tracking	1636	500	Argyll
1097	M?	01/08/19	03/02/24	31/08/22	Still tracking	1647	521	Argyll
120196	M	01/08/12	21/03/21	10/04/15	Malfunction	3154	2172	Argyll
334	M	01/08/16	27/05/20	25/09/18	Malfunction	1395	610	NE
660	M	01/08/17	18/03/24	06/11/18	Still tracking	2421	1959	NE
1025	M	01/08/18	14/09/22	10/09/21	Still tracking	1505	369	NE
1026	M	01/08/18	06/07/23	08/12/21	SNMF	1800	575	NE
1080	M	01/08/22	03/02/24	28/09/22	Still tracking	551	493	NE
1083	F	01/08/20	25/03/24	11/02/23	Still tracking	1332	408	NE
1162	F	01/08/20	25/03/24	23/03/22	Still tracking	1332	733	NE
1164	M	01/08/21	25/03/24	13/03/23	Still tracking	967	378	NE
57109	M	01/08/10	01/07/20	21/03/15	Dropped	3622	1929	NE
89251	F	01/08/11	04/09/16	23/11/14	Malfunction	1861	651	NE
129005	M	01/08/13	21/03/24	26/01/16	Still tracking	3885	2977	NE
129012	M	01/08/13	07/12/20	07/04/15	Malfunction	2685	2071	NE
148632	F	01/08/15	04/07/18	08/02/17	SNMF	1068	511	NE
148635	F	01/08/15	29/10/18	06/03/17	Malfunction	1185	602	NE
148637	M	01/08/15	21/03/24	12/03/21	Still tracking	3155	1105	NE
148639	F	01/08/15	21/01/23	06/04/17	Dropped	2730	2116	NE
809	F	01/08/16	11/04/22	24/05/20	Died	2079	687	NW
89279	F	01/08/11	05/06/18	14/02/16	Malfunction	2500	842	NW
129008	F	01/08/14	21/03/24	06/11/16	Still tracking	3520	2692	NW
1071	F	01/08/19	13/05/24	07/05/23	Still tracking	1747	372	OH
1072	F	01/08/19	12/05/24	27/03/23	Still tracking	1746	412	OH
129006	M	01/08/13	21/03/24	15/01/19	Still tracking	3885	1892	OH
57595	F	01/08/18	09/04/24	16/02/22	Still tracking	2078	783	SOS
57597	F	01/08/18	10/04/24	31/12/21	Still tracking	2079	831	SOS
57598	F	01/08/21	10/04/24	20/12/22	Still tracking	983	477	SOS
57599	M	01/08/19	10/04/24	26/07/21	Still tracking	1714	989	SOS
57602	M	01/08/21	10/04/24	28/09/22	Still tracking	983	560	SOS
84135	F	01/08/10	16/09/23	03/10/11	Still tracking	4794	4366	SOS
234165	M	01/08/18	10/04/24	16/10/21	Still tracking	2079	907	SOS

.

2.3. Data Pre-Processing

Range sizes were estimated using records between dawn and dusk, nighttime roost sites were excluded by filtering out all records between dusk and dawn. Dusk and dawn times were obtained for each location and date using the R suncalc library ([32] version 0.5.1).

Sample sizes were adjusted by applying a time filter rather than sub-sampling at a fixed number of records. The time filter helps to reduce spatial autocorrelation. Record time stamps were converted to decimal values between 0 and 24, rounding to the nearest fraction equivalent to sampling every 3, 6, 10, 20, 30, 45, 60 and 120 min. Duplicate records, which had the same date and rounded time stamp, were filtered, leaving a single unique record for each x-minute period. The processed data had five fields (x_, y_, t_, id, month), in which x_ and y_ are coordinates in the Ordnance Survey (OS) map projection (epsg = 27,700), t_ is the rounded time stamp, and id is the tag id and month of the record. These data were saved as rds files (a tidyverse equivalent of a csv file, but one which retains metadata). The rds file was read and piped to the amt make_track function to create a track, which is the basic building block of the amt package. The track was assigned the OS map projection.

If the number of records is not a significant predictor of range size it would be appropriate to select a single time filter for subsequent analyses. Restricting records to, for example, one record per 30 min, helps overcome two potential problems. First, records close in time could be spatially autocorrelated, and this may be a problem during the highest rates of data transmission when records could be separated by one minute. Spatial autocorrelation that is problematic for some animals, such as relatively slow-moving mammals, is less of a problem for a golden eagle, which can easily move from one end of its range to the other in 10–30 min (see also [14]). Second, restricting records to a longer period helps overcome differences in record frequencies between winter and summer when the fix rate is dependent on the state of the solar charged batteries, as noted earlier. Failing to account for differences in record frequency could bias the estimates to the summer period.

2.4. Home Range Estimation

Home range sizes were estimated from track files using the hr_akde function from the amt library (Animal Movement Tools version 0.2.1.0, [33]). This takes account of possible autocorrelation between sequential records [34]. Home range isopleths have 95% confidence intervals but, for almost all ranges, they were effectively identical, and indistinguishable when mapped. We used the planar area of the central estimate as our measure of simple range size. Home range size was estimated from isopleth contours extracted from the Utilisation Distribution (UD) of record locations at the 95% level (total home range extent [35]).

Home range sizes were estimated for four time periods: monthly; the breeding season (March to August inclusive); the non-breeding season (all other months) and all data.

2.5. How Much Data Are Needed

Our first question (Section 1) has two interpretations. First, how many records are needed and, second, how much time is needed? If there is seasonality in the range size, the starting month has implications for the estimated range size and at least 12 months of data will be needed to cover seasonal effects, if they exist. Because the number of records varies between months, we selected December (shortest days) and June (longest days) as the best months to investigate the effect of sample size on the range size, within a month. Restricting this analysis to these extreme single months overcomes the potential confounding problem of seasonality in both range size and sample size.

If there is seasonality this will be seen as a regular cycle in the home range size estimated monthly. Software limitations mean that at least five records are needed to estimate a home range, but we considered this too low and used 100 records, approximately three per day in a month. Analyses failed computationally if the home range was tiny; this was only a problem for some females in May when they were presumably on, or near, the nest. We checked the spatial extent of the records prior to estimating the home range, and if it was less than 1 km we assigned an arbitrary area of 50 ha for the 95% isopleth.

2.6. Extent of Good Habitat within a Home Range

UD isopleths may include habitat that is little used by golden eagles because, for example, it is topographically unsuitable, it is a water body or it is commercial closed-canopy forest. Including unused, or little used, habitat within the estimate of a range size is misleading, particularly when comparing range sizes. Water bodies >0.25 ha were excluded from all measures of range area. We used the GET model [22,28] to deal with three known unsuitable habitats within a home range: topographically unsuitable [22], commercial closed-canopy forest (e.g., [36]) and wind turbines (e.g., [28,37,38]). The GET model predicts space use by golden eagles as a topographically based surrogate for the availability of orographic winds, which have repeatedly been found as influential in habitat selection studies of golden eagles and other large facultative/obligate soaring raptors. ‘GET scores’ range from 1 to 10 and a GET 6 score is a switch point in preference, so that GET 6+ indicates increasingly preferred habitat. All 50 m pixels in Scotland have GET scores. Land within 500 m of an operational wind turbine or covered by forest (as defined in the woodland region class in the Ordnance Survey Vector Map 2020 dataset (https://osdatahub.os.uk/downloads/open/VectorMapDistrict; accessed 28 May 2024) was masked out of the GET model and their pixel values set to 0. Hence, the useable range size was estimated from the area of 50 m pixels with a GET score of 6 or more after the exclusion of water bodies, commercial closed canopy and wind turbines buffered to 500 m around their locations.

Question 3 (Section 1) asked if there is an optimum range size in terms of usable habitat. If this is true, we should expect this area, within a range, to reach an asymptote even if the overall range size continues to increase. This might be expected in, for example, areas with a large extent of closed-canopy commercial forestry.

2.7. Statistical Analyses

All statistical analyses were undertaken using R ([39], version 4.2.3). Initial summary statistics used core R commands and the tidyverse library [40].

Mixed models were used to investigate if range size was a function of various factors. Tag ID and region were random factors with random intercepts. Region was added because it was known, a priori, that ranges on the Western Isles were very small compared with the other regions. Also, within a region, many ranges were adjacent and therefore not spatially independent. Both range size and the number of records were log transformed for the analyses because both were right skewed. Model summaries and comparisons used the sjPlot library [41].

Question 1 (Section 1), on how many data are needed to obtain a reliable estimate of range size, was initially investigated separately for December and June. The class of tag (Holder or Settled) was included as a fixed factor. If the number of records was a significant predictor of range size in these months, a single sampling period appropriate to the shape of the response curve would be selected for subsequent analyses. If number of records was not a significant predictor of range size in these months, a single time interval with a maximum of one sample every 30 min would be used. This is a compromise sampling interval between reducing potential spatial autocorrelation and maximising the sample size. In a 12-hour day, there would be a maximum of 24 records, or ~720 per month.

We examined the influence of the month and tag class (Holder or Settled) on the range area (question 2: Section 1). These were also mixed models with tag ID and region as random factors with random intercepts. Month was specified as a factor. We found the best model by comparing a null model (random factors only) with other models containing one or two predictors using model AICs.

Simple summary statistics and correlations were used to answer question 3 (Section 1). Range size normally refers to the 95% isopleth from the UD, but this includes unused habitat such as lochs, the sea, commercial forests, etc. Usable golden eagle habitat (defined here as land with a GET value of at least 6 after excluding other unsuitable habitat: Section 2.6) will be a subset of the land within the 95% isopleth. We tabulated range areas as the area within the 95% isopleth and the area which is usable golden eagle habitat.

Analysis of variance (R aov function and the Tukey multiple comparisons of means test (95% family-wise confidence level)) was used to identify pair-wise differences between regions (question 4: Section 1). The Shapiro-Wilk normality test was used to test if the assumptions about residuals were valid.

Questions 5 and 6 (Section 1), are range sizes different between sexes and seasons, were answered using mixed models, with tag and region entered as random factors and sex and season as fixed factors. Because it is possible that males and females may behave differently between seasons, we included an interaction term between sex and season.

3. Results

There was a clear annual cycle of records per month, with fewer outside of the breeding season, particularly in December and January (Figure 2). There were also annual cycles in the 95% range size (Figure 3). However, the two cycles were not in perfect synchrony.

The logarithm₁₀ of the 95% isopleth area was not significantly related to the logarithm₁₀ of the number of records in either December or June (

Table 2. Model summaries and comparisons for the relationships between the logarithm of the 95% isopleth area and the logarithm of the number of records in December or June.

	June			December
Predictors	Estimates	CI	p	Estimates	CI	p
(Intercept)	8.24	7.78–8.71	<0.001	8.17	7.11–9.22	<0.001
records [log]	−0.03	−0.06–0.01	0.128	0.08	−0.10–0.26	0.380
Random Effects
σ2	0.19			0.16
τ00	0.39 Tag			0.38 Tag
	0.18 region			0.15 region
ICC	0.75			0.77
N	63 Tag			26 Tag
	5 region			3 region
Observations	1446			339
Marginal R2/Conditional R2	0.001/0.750			0.001/0.770

). Therefore, over the range of values tested here, there was no significant change in the home range size as the number of records increased. The number of tags available for testing was smaller in December because of the small number of records for some tags. The Inter Class Correlation (ICC: Table 2) measures how much of the variation in the logarithm of the range area, not attributed to the main effect (log₁₀ (number of records)), is accounted for by the tag and region random effects. About 50% of the variation was due to differences between tags (e.g., 0.39/0.76) and about 25% was due to differences between regions (0.15/0.76) (Table 2). Overall, 75% of the stochastic variation was accounted for by the random factors.

As the number of records was not a significant predictor of range size in either month, subsequent analyses used a single time interval with a maximum of one sample every 30 min.

The effect of two fixed factors, class or month, on range size was modelled, plus a null model (Model 0) which included only the random factors. Models 1 and 2 had single fixed predictors, while Model 3 had both (

Table 3. Model summaries and comparisons for the relationships between the logarithm of the 95% isopleth area and the class of tag and the month. Null model—Model 0; tag class—Model 1; month—Model 2; tag class x month—Model 3. Statistically significant results are emboldened.

	Model 0		Model 1		Model 2		Model 3
AIC	3419.9		3420.0		3096.8		3097.8
Predictors	Estimates CI	p	Estimates CI	p	Estimates CI	p	Estimates CI	p
Intercept	8.45		8.61		8.87		8.99
	8.05–8.8	<0.001	8.11–9.12	<0.001	8.40–9.34	<0.001	8.44–9.55	<0.001
class [S]			−0.23				−0.18
			−0.53–0.06	0.119			−0.48–0.12	0.247
Feb					−0.09		−0.09
					−0.27–0.09	0.323	−0.27–0.09	0.325
Mar					−0.29		−0.29
					−0.47–−0.11	0.001	−0.46–−0.11	0.001
Apr					−0.75		−0.75
					−0.92–−0.57	<0.001	−0.92–−0.57	<0.001
May					−0.81		−0.81
					−0.98–−0.64	<0.001	−0.98–−0.63	<0.001
June					−0.76		−0.76
					−0.93–−0.59	<0.001	−0.93–−0.58	<0.001
July					−0.48		−0.48
					−0.66–−0.31	<0.001	−0.66–−0.31	<0.001
Aug					−0.2		−0.2
					−0.37–−0.02	0.027	−0.37–−0.02	0.029
Sept					−0.07		−0.07
					−0.25–0.11	0.430	−0.24–0.11	0.447
Oct					−0.05		−0.05
					−0.23–0.14	0.605	−0.23–0.14	0.618
Nov					−0.06		−0.06
					−0.25–0.14	0.574	−0.25–0.14	0.581
Dec					−0.04		−0.04
					−0.27–0.19	0.743	−0.27–0.19	0.749
Random Effects
σ2	0.46		0.46		0.37		0.37
τ00	0.17 Tag		0.17 Tag		0.18 Tag		0.18 Tag
	0.18 region		0.25 region		0.23 region		0.29 region
ICC	0.44		0.48		0.53		0.56
N	65 Tag		65 Tag		65 Tag		65 Tag
	5 region		5 region		5 region		5 region
Observations	1588		1588		1588		1588
Marginal R2/Conditional R2	0.000/0.438		0.015/0.487		0.108/0.579		0.115/0.612

).

Tag class by itself, Model 1, was not a significant predictor of home range size. Month, by itself (Model 2) was a significant predictor of home range size and had the smallest AIC, although the difference was small compared with Model 3. However, as there was no significant difference between Models 2 and 3 (analysis of variance) and Model 2 was simpler, Model 2 was selected as the best model.

Models 1–3 answer question 2 about possible differences between the range sizes of birds tagged as range holders and those originally tagged as nestlings who subsequently settle on a home range. The coefficient of −0.23 in Model 1 suggests that holder ranges were smaller; however, the difference was insignificant as indicated by the coefficient’s confidence intervals which include 0 (Table 3). It was appropriate, therefore, to combine Holder and Settled tag data in analyses.

The reference month in Model 2 is January and coefficients for the other months are relative to the range size in January. Range sizes in March to September were significantly smaller than in January (Table 3, Figure 3) suggesting that breeding season ranges were smaller than non-breeding season ranges. About 21% of the variation was due to differences between tags (e.g., 0.18/0.84) and about 35% was due to differences between regions (0.29/0.84) (Table 3). Overall, 56% of the stochastic variation was accounted for by the random factors.

On question 3 (Section 1), range size normally refers to the area within the 95% isopleth from the UD, but this boundary will include underused or unused habitat such as freshwater or seawater lochs, commercial forests, or wind farms. GET 6+ habitat is a surrogate for the land potentially preferred by golden eagles topographically, and inherently excludes water bodies. Not all GET 6+ predicted habitat is ‘available’; however, because golden eagles in Scotland avoid closed-canopy commercial forest and the vicinity of wind turbines (Section 2.6). By excluding these features, we created an ‘open land’ area, in addition to the GET6+ (topographically preferred) area, within the 95% isopleth of ranges. This allowed us to examine how much of the 95% isopleth (full planar) range size estimate was composed of topographically preferred habitat (GET6+, excluding water bodies too) and available open land (after exclusion of commercial forests and the vicinity of wind turbines (

Table 4. Mean 95% isopleth areas (ha) ± standard deviations by region of the planar area, the area of GET 6+ habitat, the proportion of GET 6+ habitat with respect to the planar area and the proportion of GET 6+ habitat with respect to open land. Our study involved five regions in Scotland (Figure 1).

Region	Planar Area	Area of GET 6+	GET 6+/ Planar Area	GET 6+/ Open Land Area
Argyll	7190 ± (4738)	5141 ± (3315)	0.738 ± (0.172)	0.885 ± (0.101)
NE	8871 ± (6267)	6697 ± (3925)	0.813 ± (0.118)	0.873 ± (0.092)
NW Highlands	6335 ± (3139)	4842 ± (2678)	0.740 ± (0.127)	0.781 ± (0.142)
South of Scotland	10,443 ± (5537)	7636 ± (4423)	0.730 ± (0.118)	0.933 ± (0.031)
Outer Hebrides	2439 ± (940)	862 ± (393)	0.393 ± (0.208)	0.419 ± (0.207)

).

Standard deviations were large for all regions, suggesting that there was considerable variation in range size within a region. The most obvious observation was the very small size of the ranges on the Outer Hebrides. Another clear observation was that the proportion of open land that was GET 6+ was large in all regions, except for the Outer Hebrides.

There was no evidence for an asymptote when the area of GET 6+ habitat was plotted against total range size (Figure 4). It appeared that larger ranges had, in general, more suitable golden eagle habitat, but at a relatively constant proportion between 70% and 80% of the planar area within the 95% isopleth. Even after accounting for land unlikely to be used by golden eagles, the proportions of good golden eagle habitat remained relatively constant and high at ~80–90%, with the exception again being the Outer Hebrides (Table 4). The exceptional nature of the Outer Hebrides in this regard was not because ranges there had less open land, i.e., through more commercial conifer forest and/or wind turbines. Rather, the opposite was true and the driver was the low relatively low proportion of GET6+ habitat (40%).

The proportion of topographically preferred GET 6+ habitat lost to commercial forests in home ranges was, on average, ~9%, although there were large differences between regions (mean percent loss and maximum percent loss): Argyll (13.5%, 34.9%); Northeast Highlands (5.6%, 15.5%); Northwest Highlands (3.6%, 6.9%); South of Scotland (19.2%, 32.8%) and Outer Hebrides (0.2%, 0.5%). The rank order of losses is unsurprising as both Argyll and the South of Scotland have extensive areas of commercial forests while on the Outer Hebrides climatic conditions mean that tree growth is low, and plantations are relatively few.

The loss of GET6+ habitat was far greater due to commercial forests than 500 m buffered wind turbines (

Table 5. Loss of topographically preferred golden eagle habitat (GET 6+) by region and cause. Regions (n tags in parentheses): A = Argyll, NE = Northeast Highlands, NW = Northwest Highlands, SOS = South of Scotland, OH = Outer Hebrides. Land (ha) = the total area of land (excludes water bodies) within the 95% isopleths and GET 6+ (ha) = the area of land with a GET value of 6 or more. Lost = the total area (ha) of GET 6+ land lost to turbines and commercial forests. Turbines = the area (ha) lost to turbines (500 m buffer around turbines) and Forests = the area (ha) lost to closed-canopy commercial forests. Percentage loss values show the contributions of the two land uses to the total lost. For example, overall, 49,329 ha of preferred eagle habitat (14.2%) was lost from the 346,666 ha of preferred habitat within the 95% isopleths; 94.7% of this loss was a result of commercial forests.

		GET 6+ Habitat Area
Region	Land	GET 6+	Lost	% Loss (Total)	Turbines	% Loss (Turbines)	Forests	% Loss (Forests)
A (25)	170,549	148,466	25,074	16.9	710	2.8	24,364	97.2
NE (16)	117,101	97,292	8845	9.1	1915	21.7	6930	78.3
NW (3)	18,918	15,134	607	4.0	0	0.0	607	100.0
SOS (7)	76,192	71,111	14,715	20.7	0	0.0	14,715	100.0
OH (18)	37,754	14,664	87	0.6	0	0.0	87	100.0
All	420,515	346,666	49,329	14.2	2625	5.3	46,704	94.7

). Across all regions and ranges, the loss of preferred habitat was 14%, and this loss was mostly attributable to commercial forests (95%). In three regions, there were no preferred habitat losses due to wind turbines (Table 5), but in two regions there were seven ranges where both land uses affected GET6+ availability (four in Argyll, three in Northeast Scotland). In these Argyll ranges, 19% of the GET6+ loss was due to turbines, and 81% to commercial forests. In the Northeast Scotland ranges, there was a 31% loss to turbines, and 69% to forests.

Question 4 (Section 1), do range sizes show marked regional differences, was assessed using a single factor analysis of variance of the range size (estimated using all data). There was a significant difference in range size across the regions (F = 12.53, 4 df, p < 0.001, Shapiro-Wilk normality test w = 0.9796, p = 0.370). The Tukey HSD identified three pairwise significant differences, so that the Outer Hebrides ranges were significantly smaller, with the exception of the Northwest Highlands (p = 0.071). Range sizes of the other four regions did not differ significantly from each other. The median range sizes were slightly smaller than the mean areas (Table 4) suggesting a right skew in range size: Outer Hebrides, 2395 ha; Argyll, 5890 ha; Northwest Highlands, 6173 ha; Northeast Highlands, 8931 ha and South of Scotland, 9188 ha.

Superficially, when both partners were tagged, home ranges could be slightly larger in males (Figure 5). Nevertheless, questions 5 and 6 (Section 1) are answered in more detail by the summarised mixed models (

Table 6. Model summaries and comparisons for the relationships between the logarithm of the 95% isopleth area and the class of tag and the month. Statistically significant results are emboldened.

	Model 0		Model 1		Model 2		Model 3		Model 4
AIC	3419.9		3418.6		3229.5		3228.3		3206.5
Predictors	Estimates CI	p	Estimates CI	p	Estimates CI	p	Estimates CI	p	Estimates CI	p
Intercept	8.45	<0.001	8.35	<0.001	8.31	<0.001	8.22	<0.001	8.16	<0.001
	8.05–8.85		7.93–8.77		7.87–8.75		7.75–8.68		7.69–8.62
sex [M]			0.21	0.063			0.21	0.071	0.32	0.006
			−0.01–0.44				−0.02–0.43		0.09–0.55
season [NB]					0.49	<0.001	0.46	<0.001	0.69	<0.001
					0.43–0.56		0.43–0.56		0.59–0.80
sex × season									−0.34−0.48–−0.20	<0.001
Random Effects
σ2	0.46		0.46		0.4		0.4		0.4
τ00	0.17 Tag		0.17 Tag		0.18 Tag		0.17 Tag		0.17 Tag
	0.18 region		0.19 region		0.22 region		0.24 region		0.25 region
ICC	0.44		0.44		0.5		0.5		0.51
N	5 region		5 region		5 region		5 region		5 region
	65 Tag		65 Tag		65 Tag		65 Tag		65 Tag
Observations	1588		1588		1588		1588		1588
Marginal R2/Conditional R2	0.000/0.438		0.013/0.447		0.066/0.532		0.078/0.541		0.083/0.553

). Model 4, with the interaction term, was clearly the best model as measured by the AIC. Model 1, with sex as the only fixed factor, was not significantly different from the null model (Chisq = 3.38, p = 0.066) while Model 2, with season as the only fixed factor, was highly significantly different from the null model (Chisq = 192.47, p < 0.001). Similarly, Model 3, with both fixed factors, was not significantly different from the model containing season only (Chisq = 3.18, p = 0.075). However, Model 4, with an interaction term, was highly significantly different from Model 3, with no interaction term (Chisq = 23.83, p < 0.001).

The best model was, therefore, the model with an interaction between sex and season (see also Figure 6). Overall, there was a significant difference between the sexes (p = 0.006) and between seasons (p < 0.001). However, there was also a significant interaction between sex and season (p < 0.001). Breeding season ranges were smaller for females, sometimes much smaller. Non-breeding season ranges were larger, but the difference was more marked in females.

4. Discussion

We first discuss the six questions originally posed (Section 1).

4.1. Question 1: How Many Data Are Needed to Obtain a Reliable Estimate of Range Size?

Given Scotland’s latitude and our study’s solar-charged tags, it was not surprising that the number of monthly location records was lowest in winter. It was surprising, however, that range size was not related to the number of records used to estimate it. This was probably due, in part, to the overriding influence of seasonal changes. The need to incorporate the major differences between range estimates in the breeding and non-breeding seasons was important. Hence, at least one year of data are needed to characterise the range use of resident golden eagles. Our analyses agreed with those of [13], and as followed by [14].

An implication of our finding is that studies based on lower time periods may not have involved representative estimates of range extent (e.g., [2,12,15]). This conclusion applies only to residents, and not to migratory eagles (e.g., [10]). For migratory eagle studies that can only consider breeding season ranges, our other findings (see later: Section 4.6) also highlight the need for knowledge of birds’ sex and (likely) breeding attempt duration to reliably estimate home range.

4.2. Question 2: Are There Differences between the Range Characteristics of Birds Tagged as Range Holders and Those Originally Tagged as Nestlings Who Subsequently Settle on a Home Range?

Birds tagged as holders were probably older than settled birds, in part because settlement on a range in settled tags could be detected at a known and often young age [31]. Nonetheless, and despite regional differences in our sample of Holder tags (Figure 1), there was no significant difference in the 95% isopleth ranges between the two classes.

4.3. Question 3: What Are the Range Sizes in Absolute and Usable Terms and Are They Correlated or Is There an Optimum Range Size in Usable Terms?

Estimated planar range size by 95% isopleths was highly variable (minimum 1066 ha (tag 1072)–maximum 23,349 ha (tag 1031); mean 6414 ha, median 4692 ha; excluding the Outer Hebrides mean 7743 ha, median 6033 ha). As an absolute estimate 95% isopleths will contain both suitable and unsuitable habitat within their limits and so do not represent the ‘usable’ range. Hence, this variability could have been due to varying extents of suitable habitat within planar range limits. Alternatively, the extent of usable habitat could reach a plateau as planar range size increases (“enough is enough”).

We found that the large variation in planar range estimates was not due to them being larger to accommodate more useable suitable habitat (as estimated by the GET model: GET6+ scores), because there was a significant correlation between 95% isopleth estimates and the extent of preferred topographical habitat (GET 6+) within them. This relationship was not changed after further unsuitable terrestrial habitat (closed-canopy commercial forest and a 500 m buffer around wind turbines) was also excluded as unsuitable. There was no maximum limit of habitat deemed as suitable which planar ranges incorporated, so there was apparently no optimum range size from the metrics we used to describe ‘usable’. Rather, the larger the absolute planar range estimate was, the more ostensibly suitable habitat it incorporated. Planar 95% ranges were not larger because occupants needed to compensate for a lower useable content through more commercial forest, for example.

In most regions, on average 70–80% of the 95% isopleth (planar) range was usable according to GET (topographically preferred habitat) (Table 4). In the Outer Hebrides, not only was planar range size much smaller than in other regions (Section 4.4) but the proportion of the planar range deemed suitable was also even smaller at 40% (70–80% elsewhere). The mean usable range area in this closed sub-population [42,43] was therefore exceptionally small at 8.6 km² (although, see Section 4.4). Mean figures for the usable range area (without excluding commercial forest or wind turbine buffers) for other Scottish regions were (km²): 51.4, 67.0, 48.4 and 76.4 (Table 4). A practical implication, when ranges in the Outer Hebrides are tightly packed (Figure 5) and thereby heavily constrained by neighbouring ranges [24], is that the loss of preferred habitat here may have proportionately greater adverse impacts on range breeding productivity and/or persistence.

The loss of potentially useable habitat (GET6+) to two forms of land development was variable across regions but was predominantly due to commercial forests with wind turbines being less impactful, even in ranges that were subject to both. Across all regions, 95% of the loss was due to commercial forests. In the most heavily affected regions (Argyll and South of Scotland), the loss of otherwise useable habitat to forests was 17% and 21%, respectively.

4.4. Question 4: Do Range Sizes Show Marked Regional Differences?

There were some differences in range size between regions, but these were low compared to variation within a region, and only the Outer Hebrides had significantly smaller ranges. Albeit without significant differences, the largest ranges were in Northeast Highlands and South of Scotland. The relatively large range size in the South of Scotland may be because this sub-population [42] is in the process of expansion through reinforcement using translocations from further north [44]. Indeed, our initial screening of ranges with ≥ one year of data and our analytical cut-off date necessarily excluded several other new ranges, which continue to be occupied [44]. As more translocated birds settle on ranges, this is liable to decrease the range size, as new settlers’ ranges squeeze the limits of existing ranges.

The finding for the Northeast Highlands may be similar through the relatively higher availability of unoccupied suitable habitat (e.g., [21]), so with more space, birds probably exploit this opportunity by having larger ranges. The lower density of occupied ranges in this region is a consequence of persecution in some parts (e.g., [23]). If persecution is reduced, as apparent in one part of the region with consequent expansion in home ranges [45], and if the increasing numbers of settled ranges elsewhere in the region (suggested by the present study) are indicative of wider change, then as more ranges are occupied their size could decrease.

In the absence of formal analyses, we speculate that density dependence is at least partially involved in determining range size in Scotland. The uniquely small ranges in the Outer Hebrides are probably the best indicator, where the sub-population is closed [42,43]. In this archipelago almost all known historical ranges are occupied and over the last two decades many new ranges have been located in areas not previously occupied [21,46]. This densely occupied landscape is in contrast to, for example, the more thinly occupied landscape of Northeast Scotland [21], where there is a greater availability of space for further ranges, and the range size is larger.

However, there may be additional reasons for the small Outer Hebrides ranges other than heightened density dependence. (Argyll, for example, may also be approaching carrying capacity, e.g., [21] but has much larger ranges than Outer Hebrides.) One possibility is that the Outer Hebrides has no mammalian predators/scavengers (e.g., red fox Vulpes vulpes or European badger Meles meles), and this may reduce competition for prey and carrion, as well as making safer otherwise accessible nest sites. The island of Mull in the Inner Hebrides, part of a much larger sub-population with the mainland [42], is also absent of such mammals and, here too, the range size appears smaller [18]. Moreover, eagles in the Outer Hebrides are probably also less reliant on the typically vital resource of orographic wind uplift as proxied by the topographically based GET model. The archipelago is exceptionally windy due to its close proximity to the northeastern Atlantic Ocean and the North Atlantic Current, and this oceanic resource should make more land suitable than proxied by the GET model [22]. Hence, range size may be correspondingly smaller, and this is perhaps particularly indicative of why Outer Hebrides planar ranges had the lowest regional proportion of GET6+ (topographically preferred) habitat. This argument is consistent with the only substantive part of Scotland where the GET model predictions were poor was in the flat peatlands of the northernmost Outer Hebridean island [22], an area heavily used by dispersing eagles [42].

Food supply has been frequently expected or shown as an important influence on home range size (e.g., [47,48,49]), even if experimentally it has not always been confirmed [50]. The notion is that more food = more restricted foraging activity = smaller range size. We did not have estimates of food supply across our large sample of ranges. However, it seems likely at a regional scale that this factor was not a major driver of range size in our study. Breeding productivity (fledglings per occupied range/pair) can be a proxy for food supply (e.g., [3]) and it is consistently higher over decades in Northeast Scotland than in Argyll and, particularly, Outer Hebrides [3,21,24]. Yet, there was no difference in range size between Northeast Scotland and Argyll, and Outer Hebrides had much smaller ranges. This is not to suggest that food supply did not contribute to range size at a local level.

4.5. Question 5: Are There Seasonal Differences in Range Sizes?

From previous research on northern temperate residents [13,17,18], seasonal differences were expected, and this was confirmed through our larger sample size. Non-breeding ranges were significantly larger than breeding ranges. A contributor to this difference was the higher level of more movement activity centred at and around the nest during the breeding season, probably through a heightened influence of central place foraging.

4.6. Question 6: Does Range Size Differ between the Sexes?

Prior to our study, there was little information on how range use may differ between the sexes, despite some reasonable assumptions, such as potential differences during the breeding season due to the different parental roles, e.g., [13]. Overall, across a year, we found no significant difference in range sizes between the sexes, even if in some cases males appeared to have slightly larger ranges in paired birds. Therefore, our results showed that if GPS-tag data are collected over at least a year, estimates of range size should not be strongly influenced by the tagged bird’s sex.

There were significant sex differences according to seasonal range use, however. Females had smaller ranges during the breeding season (as would be expected from parental role division) but they had larger ranges in the non-breeding season. The sex difference in the breeding season is presumably a consequence of females spending more time attending eggs and young chicks, and males being the main food supplier for the female and young chicks [3].

It should follow that the range size of females should be higher, and disparity with males lower, when there is less or no reason for females to be caring for reproductive contents in the nest. Therefore, the duration of a breeding attempt should affect the breeding range extent, particularly for females, but males too. The use of more areas away from the nest should be particularly evident in pairs which either fail to lay eggs or fail early in incubation, and thereby decrease prospects for sex differences. This would explain why, with universal early breeding failure, together with some range abandonments afterwards, studies of breeding range use in Sweden found no sex difference [2,15]. Others have either suggested or strongly suspected that the duration of a breeding attempt was influential on breeding season range use [10,18,20].

In our study, involving 69 tagged birds across often remote parts of Scotland, we did not know the breeding status for every range and every year for which we had range use data. Hence, we could not include this metric in formal analyses. We do know, nevertheless, that our sample contained range use data when breeding failed early in the season and conversely when breeding was successful in the production of fledgling(s). From the data across several years, for some ranges, the contrast was apparent between breeding range extent in a poor season (early failure) and in a successful season (fledged young). In our study, albeit subjectively and expected, the duration of a breeding attempt clearly affected range use during the breeding season. This reflects the “undoubted”, but similarly subjective, observation of [10].

4.7. Other Inferences, and Conclusions

We did not undertake a statistical analysis of overlap between neighbouring ranges’ outer limits. It was subjectively apparent, nonetheless, that there was minimal overlap (e.g., Figure 5). Minimal or no overlap would infer territorial defence between neighbours or mutually respected boundaries. Even so, dispersing eagles routinely intrude into occupied ranges [28], likely on a seasonal basis [51].

As for all birds, knowledge of the home range characteristics of golden eagles is fundamental to understanding their biology and conservation. Using a relatively large sample of golden eagle home ranges obtained from GPS-telemetry, we have advanced knowledge of several characteristics.

Research on northern temperate resident golden eagles should aim to use at least one year of GPS tag data to obtain representative estimates of home range. Fundamentally, this is because range use varies seasonally, with larger ranges in the non-breeding season. Anything less temporally may compromise research conclusions. Studies restricted to one season (particularly the breeding season) also need to be aware that there are sex differences in range use, which may be exacerbated in the breeding season by a longer duration of the breeding attempt. If studying only the non-breeding season there can also be a sex difference, although if data from at least one year are used there is no substantive difference between the range use of males and females. Nevertheless, the preferential tagging of one sex is not recommended as it may distort how males and females can behave differently through the annual cycle.

Home ranges can vary substantially even within a region and between neighbouring pairs. The causes of this variation remain uncertain, largely because gathering data on the ecology and behaviour of golden eagles is still a challenge, despite the advantages gained from GPS-telemetry. We proffer that, at least in Scotland, density dependence may deserve increased attention along with other more commonly considered drivers, such as food supply. More research on this is required, however, and in Scotland the several unique features of the Outer Hebrides should provide insights, including further comparison with ranges elsewhere.

Our study also illustrated that simple planar range estimates (in our case 95% UD isopleths) can exaggerate how much usable habitat is encompassed. Hence, these simple range estimates may err from what is relevant to range occupants. Using a few measures of suitable/unsuitable habitats, we found that usable habitat within planar ranges was markedly lower (if again variable). This should be expected, but it is perhaps a consideration which future studies of home range should pay more attention to. We found that there was no upper limit to how much usable habitat was contained within the 95% isopleths, because planar range size was positively correlated with usable range size. This indicates that ranges were not larger because their occupants needed to incorporate more usable habitat, and so can be dismissed as a driver underlying planar range size variation.

Finally, practically, in our extensive sample of ranges, commercial forest was conclusively a greater source of habitat loss than wind turbines. This is probably not because the development of wind turbines is uncommon in Scotland [38]. The construction of wind farms is relatively recent in Scotland compared to commercial afforestation, with the wind energy industry gaining pace since the 1990s, preceded by decades of exotic conifers being planted in large dense coups for crops. Correspondingly, however, the adverse effects of forest expansion have been known for longer [3,36] than those for wind farms [38]. Different planning systems and their oversight may also play a role. Both land uses continue to expand and need proper planning. Our results and experience involving both land uses suggest that planning could be improved in forestry, but wind farm planning scrutiny should not be correspondingly relaxed. Several regions have proposed expansions of both land uses (notably the fragile, recently reinforced south of Scotland sub-population). In the Outer Hebrides, climatically, there will be little or no prospect of commercial forestry proposals, but for the same climatic reasons there are several wind farm proposals that should require especial planning scrutiny.