History of Mapping Systems. The boundaries for each of the relevant mapping systems were delineated considering various applications, leading to some marked differences in geography among units. The Flyways (i.e., Atlantic, Mississippi, Central, and Pacific) were devised primarily by the Service to help coordinate management of waterfowl populations across state and national boundaries (U.S. Fish and Wildlife Service 2014, 2015). We could find no direct accounts of the procedures followed to designate the boundaries of Flyway units, which follow political boundaries (sometimes at the county level). Given that the first use of administrative Flyways dates to 1947, we suspect the decisions were informed primarily by band returns of hunter-killed waterfowl (U.S. Fish and Wildlife Service 1984). Unlike Flyways that focused on waterfowl populations, habitat JVs were originally envisioned to help conserve waterfowl habitat (U.S. Fish and Wildlife Service 1986). The 11 habitat JVs that were established before 1999 considered only waterfowl habitat, whereas the nine JVs established later included all birds in their initial planning processes (Migratory Bird Joint Ventures 2016). Several JV boundaries were influenced by, or directly followed, BCR boundaries (see description below) or ecoregional boundaries that informed the development of BCR boundaries. Currently, all the JVs consider all bird species for their conservation planning and implementation efforts.

In contrast to Flyways and the first JVs that focused on a single non-raptor taxonomic group, the BCRs were developed by an international team in 1998 to reflect the current understanding of bird species distributions and requirements, as well as conservation challenges (Babcock et al. 1998). The BCR boundaries were derived from Commission for Environmental Cooperation (1997) level II, III, and IV ecoregions that were intended to map ecological regions of North America based on “enduring” components of ecosystems such as soil, landform, and major vegetation types. The stated purposes of BCRs included facilitating communication among bird conservation initiatives; systematically apportioning North America into conservation units; facilitating a regional approach to bird conservation; promoting new, expanded, or restructured partnerships; and identifying overlapping or conflicting conservation priorities. Lastly, the LCC Geographic Areas were formed in 2010 by considering BCRs, Omernik ecoregions, Freshwater Ecoregions, and existing national planning partnerships such as the JVs (U.S. Fish and Wildlife Service 2010). The goal for use of the LCC mapping system was conservation of both terrestrial and aquatic species, so some BCRs were split or combined to accommodate important watersheds.

Field Protocols. Golden Eagles were captured and tracked for various studies (Table 1). All eagles were fitted with transmitters mounted on custom-made backpacks and standard United States Geological Survey (U.S.G.S.) leg bands; some were also fitted with unique colored visual ID bands or patagial tags. Transmitters included Argos-only Platform Transmitting Terminals (PTTs), Global Positioning System (GPS)/Argos PTTs, and GPS/Global System for Mobile Communications (GSM) units from various manufacturers, which varied in shape, size, and mass (Table 1). Most transmitters were attached via variations of the “Y-harness” constructed of Teflon ribbon (Buehler et al. 1995). With harness material attached, transmitters typically weighed between 55 and 100 g, amounting to ≤3% of the mass of tagged eagles (Stahlecker et al. 2015). For additional details on methods specific to each study, see references in Table 1.

Table 1. 

Sources of data on Golden Eagle fixes.

Initial Post-processing of Fixes. Argos fixes represented only those in location classes 3, 2, and 1 (estimated error radii of <250 m, 250–500 m, and 500–1500 m respectively), unless the researcher had already screened the points and decided to retain lower-quality location classes (CLS [Collecte Localisation Satellites] 2011, McIntyre et al. 2008). If the Argos fixes had not been previously screened before submission to this study, we passed the higher-quality location class fixes through the Douglas Argos-filter algorithm (Douglas et al. 2012). Spikes with angles smaller than 15° and 25° were removed if their extension was farther than 2500 and 5000 m, respectively. We then passed fixes through several different velocity filters (20, 27.8, and 40 m/s), generated tracks for each filtered set, and selected which threshold to use based on visual inspection. At a minimum, we visually screened GPS fixes for obvious outliers, but some data sets received additional filtering and quality checks before being analyzed. To scale all observations to the same spatial scale, we then transferred both forms of eagle location data to a common, continent-wide 3-km grid and reported each fix as the center coordinates of a grid cell. Because datasets included fixes collected at various temporal scales (e.g., temporal frequencies from every 30 sec to ≥4 hr), we subsampled the data by selecting the first hourly fix per bird. Therefore, each dataset contributed no more than one fix per eagle per hour.

We assigned each fix to the relevant mapping unit in each of the four alternative administrative and ecological mapping systems. These four mapping systems divide North America into non-overlapping geographic regions, but the Flyways, JVs, and LCCs do not include all of North America. Our analytical approach required classification of all fixes to units. Therefore, for portions of North America not classified into a particular unit, we assigned fixes to either the appropriate Canadian province or Mexican state (for Flyways, because some provinces and states are assigned to unique Flyways), or to a combined unclassified Canada/Mexico. For example, a fix at coordinates of 45°12.405′N 105°46.722′W (NAD 83) would be assigned to the Badlands and Prairies BCR, the Central Flyway, Northern Great Plains JV, and Plains and Prairie Potholes LCC (Fig. 1). For each eagle, we then counted the subsampled fixes within each mapping system unit. The resulting four matrices (one for each mapping system) became the basis for further analyses, with each matrix having one row for each eagle, one column for each mapping unit, and the number of eagle fixes reported in each cell.

Cluster Analysis of Eagle Presence in Mapping System Units. We performed a cluster analysis to determine how the patterns of mapping-unit use by individual eagles conformed to preexisting administrative or ecological mapping systems. Individual eagles that used mapping units in similar proportions were grouped together regardless of total number of fixes per eagle because we transformed the data using a dissimilarity index that is unbiased by raw abundances (Legendre and De Cáceres 2013). For example, eagles that used the Atlantic Flyway exclusively would likely form one cluster, whereas eagles that used the Pacific and Central Flyways in similar proportions would form another cluster. We used a hierarchical clustering algorithm and performed separate analyses for each mapping system. Only counts of fixes per eagle within units informed the cluster analysis: the analyses did not consider the explicit spatial relationships among units (e.g., inter-unit distances and adjacency metrics). We transformed the summed eagle fixes per unit to a dissimilarity matrix using the Chao-Jaccard dissimilarity index (CommEcol package; Melo 2016) using R 3.2.2 (R Core Team 2015). The resulting dissimilarity matrix then formed the basis for a hierarchical agglomerative cluster using the flexible-beta linkage method, which sought to balance the development of compact, spherical clusters with optimizing similarity among cluster members (cluster package; Legendre and Legendre 1998, Maechler et al. 2015). Because cluster membership can be assigned to any number of clusters between one and the sample size, we determined the optimal number of clusters by visually inspecting plots of measures of cluster quality (average silhouette width or ASW, Hubert's gamma or HG, and point biserial correlation or PBC) and looking for the point(s) at which additional numbers of clusters failed to greatly improve clustering quality (WeightedCluster package; Studer 2013). After assigning each eagle to a movement cluster, we interpreted each cluster by examining which units were used by eagles in that cluster as well as by viewing maps of unit use by cluster. We also sought meaningful aggregations of mapping units. If multiple mapping units were represented in one cluster, or if several clusters were spatially contiguous and clearly nested within another cluster, we interpreted this as a suggestion that the mapping units could be combined into a single functional unit.

Summary of Data. We analyzed tracking data from 571 Golden Eagles, tracked for various periods between 1 August 1992 and 9 March 2016 (Fig. 2). Most of the data were from eagles tracked after 2010, although at least one eagle was being tracked at any given date during the study period, except for gaps from 10 May 1994 to 10 December 1995, 30 August 1996 to 24 July 1997, and 16 January 1999 to 21 July 1999. We considered data only from eagles tracked for 22 d or longer, which was sufficient time for at least one eagle to have crossed boundaries in each mapping system. Of these, we tagged 212 eagles (37%) in their nests as flightless young, 72 as fledged hatch-year eagles (13%), and 255 as after-hatch-year eagles (45%), and the age of 32 tagged eagles was reported as unknown (6%). Minimum distance moved by eagles in these four age classes averaged 10,409 km, 7846 km, 14,860 km, and 10,058 km respectively. Mean tracking duration per eagle was 419 d (95% CI: 387–451 d), with the longest tracking duration 3131 d. Eagle locations were as far south as 20.22°N and as far north as 70.48°N.

Figure 2. 

Telemetry fixes of 571 Golden Eagles tracked in North America between 1992 and 2016. Lines indicate unobserved straight-line paths between consecutive telemetry fixes.

Spatial Patterns Identified Through Cluster Analysis. The optimal number of clusters per mapping system varied from as few as five for Flyways to as many as eight for LCCs, implying that eagles could be classified into groups that shared distinctive movement behaviors based on their telemetry fixes (Table 2, Fig. 3). Distinguishing clusters representing migratory eagles from clusters containing mostly nonmigratory eagles was straightforward, because the higher maximum latitudes (i.e., most northerly locations) used by migratory eagles led to greater proportions of fixes in more northern mapping units. Mapping the percentage of fixes in each unit for each cluster provided additional clarification into the different uses of space by each group of eagles (see Fig. 4 for the example of the LCC mapping system).

Table 2. 

Suggested optimal number of clusters and interpretation of cluster meanings by administrative or ecological mapping system.

Figure 3. 

Example of quality statistics for clustering solutions (x-axis indicates number of clusters) for (A) Bird Conservation Regions (BCR), (B) Flyways, (C) Joint Ventures (JV), and (D) Landscape Conservation Cooperatives (LCC). Average silhouette width is ASW, Hubert's Gamma is HG, and PBC is Point Biserial Correlation. The peak or point of inflection of each curve suggests a number of clusters that balances cluster quality with cluster number, indicated by dashed vertical lines.

Figure 4. 

Percent of Golden Eagle telemetry fixes in relation to U.S. Fish and Wildlife Service Landscape Conservation Cooperative (LCC) units within movement clusters.

For all four systems considered, cluster analyses effectively separated the easternmost Golden Eagles (initially captured in winter or on migration in the Appalachians or Atlantic coastal regions) from their more western counterparts (Table 2). A second cluster of eastern eagles was identified in the JV system, which consisted of a few eagles tracked only briefly (<58 d) and therefore not observed to migrate north. Three of the remaining clusters in the Flyways were characterized by overlapping use of the Pacific, Central, and Mississippi Flyways as well as much of Canada, but a fifth cluster identified eagles that used only the Central Flyway, the Canadian provinces to the north, and the Mexican state of Coahuila. We found two additional clusters with similar geography in the BCR, JV, and LCC systems (Table 2). Each of these mapping systems contained a cluster of predominantly Great Basin or Intermountain West sedentary eagles, and another cluster of eagles that stayed primarily within the Coast Ranges or Central Valley of California. The BCR and LCC systems contained a Southern Rockies/Colorado Plateau/Four Corners regional cluster, a cluster combining the Northern Rockies with the Badlands/Prairie region, and a Northern Rockies/Alaskan migrant cluster. We found a Badlands/Prairie cluster in both the LCCs and JVs. The JV system further contained clusters of eagles that migrated from Alaska to the Intermountain West, those that migrated mostly from Arctic Canada throughout the western U.S.A., and sedentary eagles in the Mojave and Sonoran deserts. In contrast, in the LCC system, eagles using the Mojave and Sonoran deserts clustered with others using the Southern Rockies.

Adequacy of Administrative and Ecological Mapping Systems to Partition Eagle Fixes Among Units. Our set of eagle data did not conform well to any of the four mapping systems examined here, because eagles that were grouped into different clusters used many of the same geographic units. We could not devise a better grouping of units, primarily because only some eagles exhibited seasonal migratory behavior. However, our analysis highlighted several specific administrative units of relatively greater importance for conservation of Golden Eagles. For example, when considering LCCs, 92% of eagle fixes in the Northern Rockies cluster were within the Great Northern LCC, suggesting the presence of a resident population of Golden Eagles (Fig. 4). Additionally, the Great Northern LCC was the unit used second-most often by eagles in the Far Northern Migrant cluster, Plains and Prairies Potholes cluster, Southern Rockies cluster, and Great Basin cluster, containing 16%, 11%, 10%, and 6% respectively, of their fixes. Although Golden Eagles are not among the species currently designated as Target Species in the LCC's Strategic Conservation Framework (Chambers et al. 2013), conservation goals and targets for sage steppe and Rocky Mountain landscapes may potentially be expanded to incorporate this species, particularly with regard to wildlife habitat connectivity and energy development.

For some systems, the clustering exercise succeeded in illustrating logical groupings of eagles by combining their use of geography and migratory movements. Using all four mapping systems considered here, we could distinguish the group of eagles initially captured in the Appalachians and Atlantic coastal regions from all other groups, with several units used almost exclusively by eagles in that single Eastern cluster (Table 2). Most of the systems considered contained several clusters consistent with long-distance seasonal migrants, and other clusters that represented eagles that were more sedentary within a limited number of units. For example, the optimal clusters of LCC fixes identified a cluster of eastern long-distance migrants (Fig. 4E), two clusters of western long-distance migrants with different wintering areas (Fig. 4B, C), and five regionally focused clusters in the Great Basin, Northern Rockies, California, Desert, and Southern Rockies LCCs (Fig. 4A, D, F, G, H). The eagles in the five regional clusters appeared to be primarily sedentary or short-distance migrants. For each system, additional clusters of long-distance migrants may be identifiable by visualizing a larger number of clusters than the number suggested as a parsimonious solution.

Our analytical approach did not always depict the actual use of space by Golden Eagles, possibly because of the use of geographic boundaries that did not adequately demarcate Golden Eagle movements. For example, many eagles appeared to concentrate movements along the continental divide, a feature that forms part of the boundary between the Pacific and Central flyways. As a consequence, this likely reduces the effectiveness of this administrative boundary for conservation planning for eagles. Overlaying eagle tracks on cluster maps revealed other instances where eagles moved along or crossed boundaries, or where new boundaries might be useful. In the northern portion of the continent, eagle movements and space use tended to be directed along north-south vectors. This is not surprising, because many eagles depart northern-latitude ecosystems and overwinter in southern areas (Kochert et al. 2002). For example, eagles in both the Plains and Prairie Potholes Cluster and the Eastern Cluster used the Arctic LCC extensively, but the eagles in the western cluster were restricted to the Alaskan, Yukon, and Northwest Territories portion of the Arctic LCC (Fig. 5). Thus, alternative boundaries in the northern portions of the Golden Eagle's range may improve separation of eagle movement clusters into more mutually exclusive groups.

Figure 5. 

Inferred paths of Golden Eagles (gray lines) overlying the number of telemetry fixes of eagles within classified movement clusters in relation to U.S. Fish and Wildlife Service Landscape Conservation Cooperative (LCC) units.

Synthesis and Recommendations for Future Work. Our retrospective analyses allowed us to address our research questions with a level of rigor not possible before, but even with a large sample of tagged eagles, some age classes and capture locations were not well-represented by our data. We pooled data from 34 different studies, most of which deployed their transmitters in one to a few locations within a specific region. Furthermore, many studies focused on eagles of a particular age class, or deployed transmitters only during migration or winter. This resulted in initial uneven distributions of monitored eagles across the landscape (Fig. 6). Such data limitations may consequently limit our scope of inference for discerning continent-wide patterns of space use, especially if other groups of eagles not yet studied are as sedentary as those inhabiting the Californian Coastal Ranges and Central Valley, or as migratory as those in the east or Alaska. Therefore, our results will be helpful for directing strategic deployment of future survey and tracking efforts in underemphasized geographic regions, such as northwestern and north-central Canada, Mexico, and the central and midwestern U.S.A.

Figure 6. 

Numbers of Golden Eagles captured within geographic units for (A) Bird Conservation Regions (BCR), (B) Flyways, (C) Joint Ventures (JV), and (D) Landscape Conservation Cooperatives (LCC).

Regardless of biases in initial transmitter deployment locations, our analyses identified groups of both long-distant migrants and more sedentary eagles, as well as a consistent separation of eastern (and perhaps midwestern) from western Golden Eagles. Our results were consistent with a recent analysis of the genetic structure of North American Golden Eagles, which detected genetic differences between eastern and western Golden Eagles, and additional yet still unresolved population structure for eagles throughout the west (Doyle et al. 2016). Contrary to our expectations that eagle fixes would be best depicted by either the Flyways or the JVs, clusters of recognizable migrant or resident eagles were also detected when using the generalist bird- or ecoregion-focused systems of BCRs and LCCs. This suggests that an ecoregional mapping approach may be most useful when seeking to understand the space use and movements of a highly mobile and adaptable species such as the Golden Eagle. Further analyses of these data as spatially and temporally explicit eagle movement paths will bring greater insight into migration behavior, especially by identifying migration corridors and wintering areas, as well as factors influencing the timing and location of movements.