An active‐radio‐frequency‐identification system capable of identifying co‐locations and social‐structure: Validation with a wild free‐ranging animal

Behavioural events that are important for understanding sociobiology and movement ecology are often rare, transient and localised, but can occur at spatially distant sites e.g. territorial incursions and co‐locating individuals. Existing animal tracking technologies, capable of detecting such events, are limited by one or more of: battery life; data resolution; location accuracy; data security; ability to co‐locate individuals both spatially and temporally. Technology that at least partly resolves these limitations would be advantageous. European badgers (Meles meles L.), present a challenging test‐bed, with extra‐group paternity (apparent from genotyping) contradicting established views on rigid group territoriality with little social‐group mixing. In a proof of concept study we assess the utility of a fully automated active‐radio‐frequency‐identification (aRFID) system combining badger‐borne aRFID‐tags with static, wirelessly‐networked, aRFID‐detector base‐stations to record badger co‐locations at setts (burrows) and near notional border latrines. We summarise the time badgers spent co‐locating within and between social‐groups, applying network analysis to provide evidence of co‐location based community structure, at both these scales. The aRFID system co‐located animals within 31.5 m (adjustable) of base‐stations. Efficient radio transmission between aRFIDs and base‐stations enables a 20 g tag to last for 2–5 years (depending on transmission interval). Data security was high (data stored off tag), with remote access capability. Badgers spent most co‐location time with members of their own social‐groups at setts; remaining co‐location time was divided evenly between intra‐ and inter‐social‐group co‐locations near latrines and inter‐social‐group co‐locations at setts. Network analysis showed that 20–100% of tracked badgers engaged in inter‐social‐group mixing per week, with evidence of trans‐border super‐groups, that is, badgers frequently transgressed notional territorial borders. aRFID occupies a distinct niche amongst established tracking technologies. We validated the utility of aRFID to identify co‐locations, social‐structure and inter‐group mixing within a wild badger population, leading us to refute the conventional view that badgers (social‐groups) are territorial and to question management strategies, for controlling bovine TB, based on this model. Ultimately aRFID proved a versatile system capable of identifying social‐structure at the landscape scale, operating for years and suitable for use with a range of species.


| INTRODUCTION
Locating animals relative to one another (co-location) is fundamental to understanding sociobiology, gene-flow, dispersal patterns, and disease epidemiology, inter alia (Hansson, 1991;Kappeler, Barrett, Blumstein, & Clutton-Brock, 2013;Woodroffe et al., 2016), because co-location provides opportunities for animals to interact directly or indirectly. Such insights are also essential to designing effective wildlife management strategies (Carter et al., 2007;Woodroffe et al., 2016).
Conventional reliance on observation, or coarse-scale tracking technologies, can lead to misinterpretation of animal societies, especially when the study species is rare, elusive, cryptic and/or nocturnal, and thus less amenable to surveillance (Wilson & Delahay, 2001).
These issues are compounded further in high-density populations and in social-systems involving hierarchies. In such circumstances, monitoring the activities of a sufficient number of individuals, or diversity of socio-types is essential, because focusing on individuals that are prominent in society, or easily detectable/trappable, generates interpretive bias (e.g. Tinnesand et al., 2015).
That social organisation is often misconstrued is increasingly exposed by genetic pedigree, revealing patterns of hitherto unknown extra-pair/group paternity, the breeding contribution of non-territory holding floaters within populations and the extent to which unseen out-breeding maintains gene flow and averts inbreeding depression (e.g. Burke & Bruford, 1987;Clutton-Brock, 1989;Zack & Stutchbury, 1992). Important behavioural events may be transient, infrequent and therefore difficult to record without long-term monitoring at high temporal resolution. When the spatial scale of studies is restricted, rare, long distance animal movements (affording opportunities for landscape scale gene flow and disease spread) can go unobserved (Byrne et al., 2014).
These deficiencies have, in part, arisen through technological limitations, where (near) continuous and simultaneous tracking of two or more individuals is necessary to identify dynamic interactions (Doncaster, 1990). Furthermore, traditional tracking approaches, e.g. visual observation or radiotelemetry, risk perturbing the very behaviour under observation (Böhm, Palphramand, Newton-Cross, Hutchings, & White, 2008). Newer, non-tagging methods such as eDNA (environmental DNA; residual DNA from an organism remaining in its environment) and camera trapping (Powell, Ellwood, Kays, & Maran, In press) may be used to establish presence/absence at specific locations but eDNA cannot locate in time, camera trapping is restricted by field of view and nocturnal illumination, and both are limited by poor longevity (eDNA degrades; camera trap batteries deplete; memory cards fill with non-target triggers).
Tracking technologies are, however, increasingly facilitating higher resolution recording of animal movement patterns (Böhm, Hutchings, & White, 2009), and referencing of contacts spatially, to infer both with whom and where contacts occur (Woodroffe et al., 2016). Monitoring transgressions into neighbouring territories, and co-locations with neighbours at the edge of individual (or group) ranges, could potentially expose population-level connectivity.
Ability to co-locate is particularly important because co-location represents an animal's opportunity to mate, transmit disease (directly or indirectly), or otherwise socially interact, or ignore each other; these opportunities are key, even when co-location behaviour remains unknown. Established technologies, potentially capable of positioning animals (either alone or co-locating) with sufficient defined spatial accuracy to answer sociological questions, include: Very High Frequency radio tracking (VHF) and the digital equivalent "Coded tags" (e.g. www.Lotek.com); Global Positioning System (GPS tags); Passive Integrated Transponders (PITs); Proximity Tags (VHF, GPS and PIT reviewed by Ellwood, Wilson, and Addison (2007), Proximity tag utility reviewed by O'Mahony (2015)). These technologies are animal-borne (usually via collar/harness) with differing functionality that is either advantageous or disadvantageous depending on research goals (Table 1); none can identify behaviour or actual interaction.
If rare or transient events are to be recorded, tags need to be long-lived. Tag longevity is primarily determined by the battery capacity required to: (1) make any necessary radio transmissions (one or two-way); and (2) perform any necessary on-tag data-processing and storage. Battery capacity depends on battery type, temperature, and species-appropriate battery size, shape and weight. Because VHF and Coded tags are simple one-way transmitters, they are relatively twice as battery efficient as two-way transmitting Proximity tags. With 4. aRFID occupies a distinct niche amongst established tracking technologies. We validated the utility of aRFID to identify co-locations, social-structure and inter-group mixing within a wild badger population, leading us to refute the conventional view that badgers (social-groups) are territorial and to question management strategies, for controlling bovine TB, based on this model. Ultimately aRFID proved a versatile system capable of identifying social-structure at the landscape scale, operating for years and suitable for use with a range of species.

K E Y W O R D S
co-location, European badger, Meles meles, network analysis, RFID, social-group, sociobiology, territory, tracking, wildlife management GPS there are no transmission costs, but battery life benefits of this are outweighed by a heavy burden of on-tag data processing, giving GPS shorter longevity than VHF, Coded or PIT tags. Because PIT tags are energised externally they have an almost infinite life span, but only a very short detection range. Because of their small size and low weight (<10 mm, <1 g), PIT tags can often be injected subcutaneously.
While established technologies have various merits and deficits, ability to detect potentially transient, infrequent events, would benefit from an autonomous, continuously active, scalable, system with predictable spatial accuracy, increased tag longevity and high data security.
Here, we establish the functionality of a system that fulfils this niche: automated active-radio-frequency-identification (aRFID) tags coupled with static automated automated active-radio-frequencyidentification (aRFID)-detector equipped, wirelessly-networked, basestations (see Dyo et al., 2012) technical description, and Table 1 for comparison with established tracking technologies. This technology was adapted from a commercially-available security industry system, designed to protect valuable assets in small, defined areas (tens of metres) with good reliability (e.g. security of National Portrait Gallery exhibits, http://www.wavetrend.net).
We validated this aRFID system as an animal co-locating technology, capable of operating for years, via a short 13-week proof of concept study, tracking group-living European badgers (Meles meles) in a high-density UK population. Specifically, we investigated whether detailed aRFID data could corroborate established knowledge of the sociobiology of this model species, while providing new insights.
Mounting circumstantial evidence suggests hitherto unrealised connectivity between badger groups, refuting the established view that high-density badger populations are rigidly territorial. For instance, our study population (>40 badgers/km 2 ) exhibits 48% extra-group paternity (Annavi et al., 2014) and at any trapping session, c.19.8% of individuals are discovered making temporary inter-group visits (Macdonald, Newman, Buesching, & Johnson, 2008). Nevertheless, high-density badger populations appear sufficiently socially rigid that they confine bovine tuberculosis (Mycobacterium bovis; bTB) transmission such that culling-induced perturbation increases inter-group contagion (Macdonald, Woodroffe, & Riordan, 2015). Elsewhere, in lower density populations (c.1 badger/km 2 ), long distance movements occur both within and between groups: >5 km in Spain (Revilla & Palomares, 2002); >20 km in Ireland (Byrne et al., 2014), where bTB may be less constrained by group (Olea-Popelka et al., 2005). These observations highlight the sociobiological importance of developing a system that can detect events at high temporal resolution and is scalable to detect long distance movements.
By recording badger co-locations at a relatively few, important, fixed locations we reveal the extent to which:  Sigrist, Coppin, and Hermy (1999)  (1) badgers not only co-located with members of their own social-group -an "easy test," but also with extra-group members, a "hard test" (because such events may be transient and infrequent, or absent); (2) any inter-group co-locations occurred at setts or at "notional" territorial border latrines; (3) gender affected co-location patterns.
We then applied network analysis (Krause, Lusseau, & James, 2009), based on the time badgers spent co-locating at setts and latrines, to identify and validate badger communities algorithmically, and compared these with traditional definitions of social-group territories. putative social-groups. Badger territories were interpreted from biennial bait-marking (Delahay et al., 2006) and social-group affiliations derived from cage-trapping records using the formula described by Macdonald et al. (2008).
We placed base-stations at 10 badger setts affiliated with seven social-groups and at all 15 active shared border latrines conventionally believed to infer group-territory interfaces (established from contemporaneous bait-marking, and in use for ≥2 years preceding the study;

| Data, protocols and analyses
Detections were coded, post-hoc, as time intervals, using the "lubridate()" package (Grolemund & Wickham, 2011) in R Core Team (2014, and all detection combinations were compared iteratively, per base-station, per night, to identify overlaps in space and time. Dyadic overlaps indicated the "co-location" of badgers. In this way we calculated the duration of each co-location, in seconds. We fitted a fixed-effects normal-errors GLM model (R lm()) to explore the effects of dyadic social-group relationship, site-type, gender dyad, and week, on co-location duration per dyad (Box-Cox transformed to meet assumptions of normal errors and variance).
Total co-location time, per category, per week, was divided by the number of gender dyads of each type present in the relevant week, to control for any collar losses occurring, producing the variable "Colocation duration per dyad." Weeks were treated as levels of categorical time.
To evaluate active territorial defence via agonistic interaction we examined bite-wounding on badgers caught at the start and end of the study: scarring is visible for at least 6 months (Macdonald, Harmsen, Johnson, & Newman, 2004;C. Newman, unpubl. data).
The "igraph()" package (Csardi & Nepusz, 2006) was used to convert dyadic co-location data into network graphs, per week, using co-location duration as "Edge" values (Edges are lines connecting nodes on a network, the thickness of Edges equating here to dyadic strength of association between nodes (badgers)).
Networks were generated separately for co-locations that occurred at (1) setts, (2) latrines, and (3) setts and latrines combined (the "All" network). We applied the "Fast-Greedy" (F-G) community detection algorithm (Clauset, Newman, & Moore, 2004) to each network to estimate community structure blindly (resulting groups termed "Communities"), thus enabling direct comparison with our a priori definition of social-group composition based on trapping records.
We fitted fixed-effects normal-errors GLM models (R lm()) to explore the effect of site-type (sett or latrine) on: (1) the ratio, per week, of the number of badgers co-locating within:between social-groups (log transformed); (2) the proportion of badgers, per week, involved in inter-group co-locations (logit transformed).

| System performance
Compared to other technologies (Table 1), aRFID: (1) had a detection range of 31.5 m (similar to that to which Proximity tags can be adjusted); (2) co-located animals relative to fixed resources with a definable range accuracy; (3) with high data security (data stored off-tag); and (4) wireless data access. Critically, because signal transmission was one-way, coupled with the pre-defined short transmission range, these aRFIDtags would have a projected life span of 2-5 years (depending on transmission interval setting). The only aRFID-tag failures (10 tags) were mechanical, due to broken antennas preventing signal transmission.

| Patterns in time spent co-locating
There was evidence that the effect of site-type on co-location duration per dyad (from here "co-location duration") varied with week, and that the effect of dyadic social-group relationship on co-location duration varied with site-type (interaction terms, Table 2). Inspection of the week:site-type data reveals that the general pattern was for greater co-location durations at setts than latrines but that this was not the case in just two of 13 weeks. Therefore this interaction is unlikely to be biologically significant ( Figure S1). The dyadic social-group relationship:site-type interaction is clear in Figure 2 and was due to badgers from the same social-group being more likely to co-locate at the sett, as would be expected. There were significant main effects T A B L E 2 ANOVA describing the effect of gender dyad (GD), dyadic social-group relationship (DSGR), site-type (ST) and week on co-location duration per dyad of both site-type and dyadic social-group relationship on co-location duration, but none of gender-pair or week (Table 2).  (Figures 4 and S2-S13).
Importantly, the overall pattern in Figure 2 was consistent over time. Consequently, the distribution of co-locations across groups (see Figure 2), which underlies the networks patterns observed (e.g. Figure 4), are representative of all weeks (see Figures S2-S13 for remaining 12 weeks of networks), thus enabling us to draw general conclusions across weeks.

| Evidence for active territorial defence
No collared badgers acquired fresh bite wounds during the study, indicating that neither intra-nor inter-social-group co-locations resulted in sufficiently agonistic encounters to cause evident injury.

| Network analysis
The ratio of badgers co-locating within:between social-groups was consistently higher at setts than at latrines (Figure 3a In contrast, the communities identified by the latrine network were consistently larger and fewer, per week, than those at setts; exposing much greater population connectivity at latrines. Here, each community comprised a mixture of neighbouring social-groups (Figures 4b   and S2-S13).
Combining setts and latrines in a single network produced a comparable number of similarly composed communities to the communities and social-groups arising at setts alone, but with much greater inter-group connectivity (Figure 4c).

| DISCUSSION
Our proof of concept study validated the utility of aRFID communicating with autonomous base-stations for studying the sociobiology of a free-ranging wild animal. The system was capable of locating, and therefore co-locating, aRFID-instrumented animals reliably, within a few tens of metres of fixed resources/sites, and able to operate continuously for years. Relative to alternative technologies, this positions aRFID in a distinct tracking niche where tag longevity and the deliberate restriction of detection range enables infrequent and transient events to be detected within small defined areas.
In terms of badger sociobiology, the aRFID system identified social organisation consistent with existing knowledge , but with connectivity at the landscape scale that potentially extended social-group relationships beyond the status quo. Similarly, extra-group movements in low(er) density populations have led others to question whether badgers are truly territorial (Byrne et al., 2015;Revilla & Palomares, 2002). Our observations refute the archetypal territorial defence hypothesis widely proposed for badger social organisation (Kruuk, 1978), with implications for the transfer of social and genetic information and disease epidemiology.

| System performance
aRFID-tags are specifically engineered in combination with their basestations (they are not modified long-range transmitters) to: (1) have a short detection range, enabling precise tag location within small defined areas; (2) facilitate lightweight tags, with (3) low power requirements, allowing the system to operate (4) reliably, and (5) continuously, F I G U R E 4 Network patterns from week two co-locations at: (a) setts; (b) latrines; (c) setts and latrines combined ("All"). Node numbers, shapes and colours depict badger IDs, gender (square = female, circle = male), and social-group affiliation from trapping (light-blue = BP, dark-green = SH, light-green = BB, red = PO, gold = P, fawn = M2, orange = JH, dark-blue = W) respectively. Edge thickness is proportional to co-location duration. Edge colour indicates intra-(green) versus inter-social-group (red) co-locations. Black borders indicate network "community" estimates (from Fast-Greedy algorithm) (6) for many years. These features make aRFID particularly suitable for monitoring animals where they need to be located individually (simple presence/absence) or co-located communally, relative to fixed resources/sites, over long periods of time (years), at multiple sites (scalable to detect long-distance movements).
Such attributes are especially valuable not only for detecting rare, transient and spatially disperse events, such as those providing opportunities for disease transmission and mating, but also where disturbance or site access is restricted and repeated capture/sedation (to download data, swap collars or maintain identification marks) might perturb the population (Böhm et al., 2008). Such restrictions rule out non-tagging approaches such as camera trapping (marking required if not naturally patterned), or eDNA (poor temporal resolution; requires frequent site access); see Introduction for further limitations.

| Comparison to alternative technologies
Proximity tags are able to record far more co-locations per unit time, than aRFID, because they are not anchored to specific locations; however, our aRFIDs had a 5-year maximum lifetime, six times that of the Proximity tag equivalent lasting just 9 months (Drewe et al., 2012). aRFIDs will theoretically always last at least double the time of Proximity tags (Table 1). aRFIDs will maintain this relative advantage as battery and memory capacity improves across both technologies. Therefore, where co-location is critical, the choice is reduced to deciding which is most important: (1) recording co-location anywhere (but without knowing where), with Proximity tags, or; (2) limiting colocation to fixed resources/sites but for at least double the deployment time using aRFID; (This second choice might also include PIT tags [with almost infinite tag life] although detection range is a few centimetres).
VHF, Coded and GPS systems are generally not technically suited to co-location (GPS especially is too power hungry to generate locations at the temporal resolution required to co-locate animals for more than a few hours before batteries deplete) although it is possible to gain function by combining technologies. For example, GPS combined with data telemetry systems allows remote data download, while combining GPS with Proximity tags would spatially anchor co-locations.
Nevertheless, crucially, enhanced functionality increases power consumption, reducing deployment duration.
Another base-station tracking system, the "Trace Recorder," transmitted magnetic signals radially (3 m range) to be detected by receivertags on passing badgers (Kaneko, Suzuki, Atoda, Kanzaki, & Tomisawa, 1998). Such a system is capable of generating similar data to aRFID, but for much shorter time periods (3 months) due, as with GPS, to the high energy cost of processing and archiving data on-board a tag.

| Future refinements and developments
An important feature of our system was the wireless inter-connectivity of base-stations, potentially giving world-wide access. This allowed modification of our set-up to inform users of data quantities remotely via transmitted summaries, reducing the cost/disturbance of unnecessary visits to download data (remote download of full datasets is energetically inefficient (see Dyo et al., 2012). An advantage of transmitted data summaries is that, in the future, experimental design could become dynamic, rather than predetermined, for example, automatically switching on cameras to record transient events when aRFIDs are detected.
Another refinement would be to equip base-stations with directional antennas, giving elongated detection zones suited to monitoring territory borders. Furthermore, our aRFIDs were modified security tags; a bespoke aRFID designed for animals could be reduced by an order of magnitude to weigh 2 g, and so be carried by 40 g animals (tracking device <5% body weight; Kenward, 2001) without reduction in performance (A. Markham, unpubl. data).

| System validation: Co-location patterns in time
Badgers spent the greatest proportion of time co-locating with individuals from their own social-group (83.9%), predominantly at setts (68.9%); Figure 2. This "easy test" result was expected, due to coresidency, but important because it demonstrates that aRFID can corroborate known co-location patterns. Had our detection range been too great or small then our easy test would not have been fulfilled (communities resolved too large or too small, respectively), casting doubt on the other patterns observed. Contrary to convention, however, our "hard test" showed that the remaining co-locations (16.1%) occurred between badgers from different social-groups, with visits to neighbouring setts happening within all study weeks (Figures 4a and S2-S13). This contrasts with just four such events detected over 3 years using VHF tracking elsewhere (Böhm et al., 2008). And, because we instrumented about half of the local resident badger population, and logged data at a relatively few focal sites, these encounters between social-groups represent minima. It is therefore highly probable that, overall, total inter-social-group co-location time was greater, and could reflect levels of connectivity observed in lower density populations (Byrne et al., 2014). Furthermore, this pattern was consistent across weeks and unaffected by season, suggesting a stable pattern of inter-social-group connectivity. High and low density badger populations may therefore exhibit a similar lack of territoriality.
Co-locations by badgers away from the home-sett inferred a deliberate intention to encounter conspecifics, at least by sound and smell, because contacts could easily have been avoided in space and time.
This social tolerance was apparent for both intra-and inter-socialgroup co-locations. At latrines badgers exhibited similar levels of colocation with members of neighbouring social-groups and their own social-group members, irrespective of gender ( Figure 2). This suggests no tolerance bias based on own versus neighbouring group affiliation. This observation was supported by the total absence of bite wounding, where antagonism between neighbours would cause injuries (Macdonald et al., 2004).

| System validation: Network analysis
Network community estimates closely resembled social-group memberships derived from cage-trapping records at setts. For example, the membership of five out of seven social-groups (coloured nodes) were assigned to communities (black outlines) in accord with cage-trapped group affiliations in week 2 ( Figure 4a). Decisively, a single week's colocation data identified communities that took three cage-trapping rounds, spread over 5 months, to establish (Macdonald et al., 2009).
We acknowledge, however, that some measure of uncertainty could be informative, for example, placing some badgers as inter-community floaters.
Latrine-based communities were fewer and had more members ( Figure 4b) compared to those at setts (Figure 3a). This inferred "Super-groups" at the landscape scale (Evans, Macdonald, & Cheeseman, 1989), persisting throughout the study. In combination, these sett and latrine networks (Figure 4c) revealed far greater connectivity in this population than previously identified by cage-trapping alone (Macdonald et al., 2008). Again, this contrasts with the four such events detected over 3 years by Böhm et al. (2008) and with O'Mahony's (2015) finding that <1% of badger Proximity tag contacts involved members of different groups (but note this will at least partly have been a function of a lower population density in these studies, coupled with only very close proximity contacts being recorded, although these still do not infer social interaction).
Network analysis (Figure 4) revealed yet more inter-socialgroup connectivity than that based on time spent co-locating alone ( Figure 2), with between 20% and 100% of individuals involved in any given week. Badgers from different social-groups clearly did co-locate occasionally (our hard test) at each other's setts but co-located more frequently around border latrines, albeit for shorter periods of time (compare red lines connecting badgers/nodes in sett and latrine networks, Figures 4a,b and S2-S13). This pervasive inter-social-group connectivity, identified via co-location, clearly provides the opportunity -although, as with other technologies (except cameras), no proof -for actual interaction and further contradicts the traditional view of badger territoriality via active defence (Kruuk, 1978). This undermines reliance on bait-marking to determine badger socialstructure (Delahay et al., 2000). Importantly, this revised picture of badger society also countermands the model used in bTB management scenarios (Carter et al., 2007), a contention supported by recent work in Ireland (Byrne et al., 2015). Furthermore, recent evidence suggests that badger-to-cattle bTB transmission is rare (Donnelly & Nouvellet, 2013) and may at least partly follow an environmental route (Woodroffe et al., 2016). Coupled with these observations, our evidence of badger super-groups, through infrequent but regular colocation, presents a potential opportunity for disease transmission beyond traditional social-group boundaries, regardless of the frequency and mode of infection. We thus recommend aRFID as a versatile system capable of identifying social-structure at the landscape scale.