Long-term tracking reveals a dynamic crocodylian social system

https://doi.org/10.1016/j.anbehav.2023.02.015 0003-3472/© 2023 The Authors. Published by Elsevie license (http://creativecommons.org/licenses/by/4.0/) Animal social systems are inherently dynamic, with individuals moderating how they associate with conspecifics according to spatiotemporal shifts in population demography and resource availability. Understanding such variation is important not only to further our knowledge of a species' ecology but also to gain insights into the factors influencing the evolution of animal social systems. Using a 10-year acoustic telemetry data set containing the movements and co-occurrences of 166 tagged individuals, we investigated how time of year, individual sex and maturity status affect the social organization and connectivity of a wild population of estuarine crocodiles, Crocodylus porosus. We found that our tagged population of crocodiles displayed social structure, where individuals segregated spatially into distinct communities along 120 km of river and estuary. The social organization and structure of these communities were temporally dynamic, with association rates and the connectedness of individuals peaking during the dry season before disintegrating prior to the onset of the wet season. The formation of communities was found to coincide with an increase in the frequency of co-occurrence events between mature and matureeimmature dyads prior to the onset of the mating season. Together these findings indicate that estuarine crocodiles have a structured social system, where the proximity to the mating season and an individual's maturity status dictate how they associate with conspecifics. © 2023 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/).

All species must navigate an environment that contains conspecifics regardless of their degree of gregariousness. To reduce competition and aggression, individuals may spatially segregate themselves from conspecifics (Galezo et al., 2017;Wielgus & Bunnell, 1994), while others may attempt to maximize their spatial overlap with conspecifics to locate resources and avoid predation (Carter et al., 2009;Peignier et al., 2019). These nonrandom patterns in spatial organization form the spatial structure of most animal populations (Peres-Neto & Legendre, 2010) which underpins the foundation on which social behaviours (i.e. social organization, social structure, mating and care systems) develop and evolve (Kappeler, 2019). For those individuals that share the same or neighboring territories, individuals can further mitigate potential consequences of social conflict (i.e. competition over resources, attacks, coercion, etc.) by selectively associating with or avoiding particular conspecifics (Muller et al., 2018;Piza-Roca et al., 2018;Robitaille et al., 2021;Strickland et al., 2018). The content, quality and patterning of these associations form the social structure of the population (Hinde, 1976). In contrast to a population's spatial structure, the social structure directly relates the association patterns of individuals to their fitness consequences. For example, increased familiarity with conspecifics has been shown to improve the fitness of individuals by decreasing the costs associated with territorial defence and increasing reproductive success (Beletsky & Orians, 1989;Siracusa et al., 2019Siracusa et al., , 2021. These association patterns can therefore not only influence an individual's physiology and behaviour (Muller et al., 2018;Rimbach et al., 2022), but also moderate the transfer of genes within populations (Gardner et al., 2013).
Rather than being static, the social structure of a population is often dynamic in response to temporal changes in the spatial distribution of resources and conspecifics (Brent et al., 2013;Smith et al., 2016;. These include seasonal shifts in resource abundance, predation risk and an individual's reproductive state (Brown et al., 2008;Jordaan et al., 2021;Lantz & Karubian, 2017). The relationship between conspecifics will also shift as individuals grow and reach sexual maturity (Berman, 1982) and become more competent in their social environment (Kulahci et al., 2018;Taborsky & Oliveira, 2012). As such, the social organization (i.e. group size and composition) and structure of populations are continually rewired as new social ties form and old ones collapse following shifts in the population's demography (Shizuka & Johnson, 2020). Understanding this inherent temporal dynamism can assist in understanding the drivers and consequences of animal social systems. Although important, this inherent temporal dynamism is often not considered (Cantor et al., 2021) because it requires the simultaneous monitoring of large cohorts of individuals over extended periods of time. Information obtained from such monitoring, however, is crucial for assessing how the social system of the population responds to seasonal variation in resources and individual motivations to associate with conspecifics (Farine, 2018).
Advances in animal-tracking technologies, such as the reduction in size and cost, and the extension of the battery life, have increased our capacity to remotely monitor the behaviours and movements of large cohorts of individuals across multiple years (Smith & Pinter-Wollman, 2020). As a result, we are increasingly seeing the adoption of these technologies to examine the social behaviours of a diverse range of taxa (Firth et al., 2017;Godfrey et al., 2014;Robitaille et al., 2021). For aquatic and semiaquatic species, passive acoustic telemetry (where individuals are tracked remotely using coded tags and an array of fixed hydrophone receivers) has become a powerful tool to quantify the social structure and dynamics of populations within a natural setting which have previously not been possible using traditional approaches Armansin et al., 2016;Jacoby et al., 2016Jacoby et al., , 2021Papastamatiou et al., 2020).
While typically described as solitary, crocodylians display considerable interspecific variation in gregariousness, with some species (e.g. estuarine crocodile, Crocodylus porosus) displaying apparent intolerance towards conspecifics, while others (e.g. American alligator, Alligator mississippiensis) form large-scale breeding and basking aggregations (Lang, 1987). Estuarine crocodiles are the largest and most widely distributed crocodylian, found in tropical and subtropical rivers and estuaries from the east coast of India throughout Southeast Asia to northern Australia (Webb & Manolis, 1989). They are considered to be the most aggressive and least social crocodylian (Lang, 1987), with associations between conspecifics potentially leading to severe injury (i.e. loss of limbs and tail and lacerations) or death (Webb & Manolis, 1989). Despite this perceived intolerance towards conspecifics, some estuarine crocodile populations have been shown to exhibit nonrandom spatial structure, where individuals actively form and maintain spatial overlaps with conspecifics for up to 5 years (Baker et al., 2021). Furthermore, crocodylians have been shown to vary how they associate with conspecifics according to the time of year and life history stage, with the performance of social displays and agonistic interactions with conspecifics peaking during the dry season (Boucher et al., 2021;Brien et al., 2013;Gallagher et al., 2018;Staton & Dixon, 1975). Together this suggests that estuarine crocodile populations have an underlying social structure, where relationships shift both within and across years according to resource availability and an individual's social status. However, while described as living within size-based dominance hierarchies (Lang, 1987;Messel et al., 1981;Messel & Vorlicek, 1986), no studies have formally examined the social system (i.e. the social organization and structure) of estuarine crocodiles or other crocodylians beyond the interactions between conspecifics. By understanding the social organization and structure of estuarine crocodiles, not only will we gain insights into the social systems of crocodylians, but it will also provide insights into how the social systems of nongroup-living species respond to temporal shifts in resources within their environment.
In this study, we used a combination of implanted acoustic transmitters and an array of fixed acoustic receivers to quantify the social organization and structure of a wild population of estuarine crocodiles. This technology allowed us to track 166 individuals of various size and age classes concurrently over a 10-year period, as they moved throughout 120 km of river and estuarine habitat. Specifically, we were interested in where and when tagged crocodiles co-occurred within the detection fields of our array of fixed acoustic receivers. Our study aimed to (1) describe the social organization and structure of estuarine crocodiles, (2) assess whether associations between conspecifics were nonrandom and consistent through time, and (3) examine how time of year and the demography of the population influence the association patterns of individuals. We hypothesized that (1) crocodiles would display nonrandom social structuring, with individuals maintaining associations with conspecifics through time, and (2) the social structure of crocodiles would be dynamic, with both the time of year and demography of the population influencing the social structure of the population.

Crocodile Capture and Tracking
An array of hydrophone receivers (VR2-W, Innovasea, Bedford, NS, Canada, innovasea.com) were placed along the Wenlock River, Cape York, Australia to detect the presence of coded acoustic tags (Fig. 1a). Each receiver was attached to a concrete anchor and float and placed 2e20 m from the riverbank and approximately 1e6 m below the water surface. Receivers were placed between 8 and 102 km 'adopted middle thread distance' (AMTD, the distance measured along the middle of a watercourse that a specific point in the watercourse is from the watercourse's mouth or junction with the main watercourse) and were spaced on average 4.70 ± 4.37 km apart (mean ± SD; 0.38e19.6 km).
To capture crocodiles, traps were deployed along a 47 km stretch of the Wenlock River between 2008 and 2019 following the methods described in Campbell et al. (2010), Campbell et al. (2013), Campbell et al. (2015) and Dwyer et al. (2015) (Fig. 1b). Traps were set between August and September each year and positioned within macrotidal brackish and nontidal freshwater environments. Small crocodiles (<2 m total body length, TL) were captured by hand using spotlighting and a snare. Once restrained, the animal's sex, TL and maturity status were recorded, with female crocodiles >2 m TL and male crocodiles >3.3 m TL classified as mature (Webb & Manolis, 1989), and smaller individuals identified as immatures. A local anaesthetic (lignocaine, Troy Laboratories, Glendenning, NSW, Australia) was injected and a coded acoustic transmitter (V13T-6x, N ¼ 6, or V16T-6x, N ¼ 160, Innovasea) inserted into a pocket created between the dermis and muscle behind the left forelimb (Franklin et al., 2009). The pulse transmission rate of the transmitters was set to transmit every 90e120 s. A projected battery life of 5e10 years and small size (diameter 13e16 mm) permitted the recording of movements from across a wide size range of individuals (0.86e4.64 m TL) across multiple years.

Ethical Note
Crocodile capture and tagging procedures were carried out with approval from The University of Queensland Animal Ethics Committee (SIB/302/08/ARC, SBS/204/11/ARC/AUST ZOO (NF), SBS/215/ 14/AUST ZOO/ARC, SBS/287/17/ARC, SBS/137/20) and under Queensland Government Scientific Purposes permits (WISP05268508, WISP13189313, WA0008255). The process of removing crocodiles from traps to their eventual release took up to 60 min, with crocodiles released at the point of capture. All crocodiles survived the capture and tagging process and resumed normal behaviour within 24 h of release.

Defining Social Associations
Associations between tagged conspecifics were identified following a 'gambit of the group' approach (Whitehead & Dufault, 1999), where individuals were considered to be associating if they were observed co-occurring at fixed acoustic receivers at the same time. The detection radius of each receiver was approximately 400 m (Baker et al., 2021), and as the river width was rarely >250 m, it was unlikely for tagged crocodiles to pass by a receiver without being detected. As estuarine crocodiles have been observed reacting to conspecifics up to 1 km away (Lindner, 2004), and as crocodylian vocalizations have been recorded travelling up to 500 m under water and up to 2 km in air (Dinets, 2011(Dinets, , 2013, we assumed that crocodiles were likely to be aware of each other's presence when co-occurring at acoustic receivers. Associations at acoustic receivers were identified using a variable time window approach. For each focal individual we placed a set sampling period either side of their observed detections forming an 'influence zone'. Every tagged conspecific observed at the same acoustic receiver within this 'influence zone' was assigned as associating with the focal individual. To prevent overestimating the associations between conspecifics, when multiple detections of a focal individual had overlapping influence zones, we pooled the intervals forming the variable time window. The sampling period of the variable time window was set at 4 min, which represents the maximum duration between two consecutive transmissions from a single acoustic transmitter. To prevent underestimating the association rate due to low sample size, only data obtained between August 2010 and August 2020 were included in our analysis when the number of crocodiles in the array was >50 individuals representing approximately 70% of the population . To determine how time of year influences (1) the number of detections and (2) the number of crocodiles detected per receiver, we used generalized linear mixed-effects models (GLMM) with a Poisson distribution using the lme4 (Bates et al., 2015) package in R (R Core Team, 2021). The number of detections and number of crocodiles detected per receiver were the response variables, with month as the predictor variable. To account for shifts in the number of tagged crocodiles in the acoustic array, study year was included as a random effect (Appendix Fig. A1).

Crocodile Social Organization
To gain insights into social organization at the population level, we examined how group size (based on the number of tagged individuals detected at receivers) and composition (based on an individual's maturity status) varied according to time of year. These seasonal shifts were investigated by constructing a GLMM (Poisson distribution), with the number of co-occurrences observed at acoustic receivers as the response variable, maturity status combination (i.e. matureemature, matureeimmature, immatureeimmature) and month as the predictor variables. Crocodiles were recorded as mature once their estimated length was greater than 2 m TL for females and 3.3 m TL for males (Webb & Manolis, 1989). As the telemetry technology used enabled tagged animals to be tracked across multiple years without recapture (maximum ¼ 10 years), it was necessary to adjust the body size/ maturity status of individuals for those years that individuals were tracked but not measured. To do this, the TL of individuals was estimated by adding the mean population growth rate of 7.39 cm/ year to their TL for each year they were tracked (Baker et al., 2019). To account for the number of crocodiles detected at receivers potentially influencing the number of co-occurrences observed, the number of crocodiles detected at receivers (log-transformed) was included as a covariate. Both receiver ID and study year were included in the model as random effects to account for potential differences between acoustic hydrophones or study years.
Next, to gain insights into social organization at the individual level, we examined how the proportion of time that individuals were observed associating varied according to time of year and crocodile maturity status. To achieve this, we first determined the duration of co-occurrence events as the time period in which all co-occurring conspecifics were present. We then determined the duration of time that individuals were observed at our acoustic receiver stations using the RunResidenceExtraction function in the VTrack (Campbell et al., 2012) R package. From this, the proportion of time that individuals were observed associating with conspecifics (time associating with conspecifics/total time detected) was determined for each month that an individual was detected within our acoustic array. We then constructed a GLMM (binomial distribution, logit link), with the proportion of time an individual spent associating with conspecifics on the acoustic array as the response variable, maturity status and month as the predictor variables, and study year and crocodile ID as random effects.

Crocodile Social Structure
To investigate temporal changes in crocodile social structure, we generated monthly social networks for our tagged crocodiles using the igraph (Csardi & Nepusz, 2006) package in R. We then used the simple ratio index (SRI) to measure the strength of associations between conspecifics (Hoppitt & Farine, 2018). Tagged conspecifics that were present within a crocodile's monthly home range (95% utilization distribution) but did not co-occur at receivers were assigned an SRI of 0 to prevent overinflating the absence of pairwise associations for these monthly dyads. Crocodile home range estimates were generated using a least-cost utilization distribution following the methods described in Baker et al. (2021). A minimum of five unique detections was required to generate home range estimates. Any crocodiles for which we did not have the data to generate at least one home range estimate were excluded from the analysis.
To describe the social structure of crocodiles, we determined the transitivity and modularity of the observed monthly social networks. Transitivity is a measure of how closed trios tend to be and captures the degree of clustering and connectivity in the network, with values ranging between zero (no triadic closure) and one (all trios are connected; Farine & Whitehead, 2015). Modularity is a measure of how well a population can be separated into distinct communities (Whitehead, 2008), with values ranging between zero (random association) and one (no associations between communities). To determine modularity, the tagged population was first divided into communities along the river system using the eigenvector-based modularity method (Newman, 2006). This allowed us to identify groups of individuals that were more densely connected in the network by calculating the leading non-negative eigenvector of the modularity matrix of the graph (Newman, 2004(Newman, , 2006. The modularity of these communities was then calculated for each month as the difference between the proportion of associations within communities and their expected proportion. Modularity (Q) values greater than ca. 0.3 were then used as a threshold to identify whether strong divisions between communities were present (Newman, 2004). The transitivity, community assignment and modularity of the networks were all calculated using the igraph R package. Seasonal differences in crocodile social structure were compared using GLMMs (Gaussian distribution), with either transitivity or modularity as the response variable, month as the predictor variable and study year as the random effect.

Does Crocodile Social Structure Differ from Random?
To determine whether individuals display distinct preferences or aversions towards conspecifics, we compared our observed association patterns to that of a spatially explicit null model (RW model). Unlike traditional permutation techniques (Farine, 2017), spatially explicit null models allow social associations to be decoupled from space use by randomizing movement trajectories within (rather than between) individuals (Peignier et al., 2019;Spiegel et al., 2016). In our null model, the number of detections and time between detections matched the original data set but each crocodile's location along a movement trajectory was randomized albeit constrained by their probability of being observed at acoustic receivers and the maximum distance they could travel between detections based on their observed maximum speed (0e6.23 km/h). Owing to computational limitations, we ran 50 iterations of the null model and then determined the social environment and associations of each of the simulated data sets following the methods described above. Following Whitehead et al. (2005) and Carter et al. (2013), we determined whether the social structure of the population was different from chance by calculating the coefficient of variation (CV) and the proportion of nonzero association indices of the observed and simulated data sets. A significantly higher observed CV to that of the simulated data sets indicates the presence of long-term preferences between conspecifics, while a significantly lower proportion of nonzero association indices indicates the presence of long-term avoidances (Whitehead et al., 2005). Here, a population displays nonrandom social structuring if the CV and/or the proportion of nonzero association indices differ significantly (P < 0.05) from the expected distribution.

Stability of Associations Through Time
To examine whether crocodiles maintained associations with tagged conspecifics across multiple years, we calculated the lagged association rate (LAR) using a monthly sampling period with the asnipe (Farine, 2013) R package. The LAR determines the probability of two individuals associating at time t when they have been observed associating together in the past (Whitehead, 1995). To determine the influence of maturity status on the stability of associations, we calculated the LAR for each combination of maturity status (i.e. mature/mature, mature/immature) based on the estimated maturity status of individuals for a given year. To further determine whether conspecific sex influenced the stability of associations between mature dyads, the LARs for same-and oppositesex dyads were calculated separately. LAR analyses were restricted to a maximum duration of 5 years as this has previously been found to be the period in which 75% of acoustic tags successfully transmit within this population (Baker et al., 2021;Appendix Fig. A1). To determine whether crocodiles were actively maintaining their preferences or aversions to tagged conspecifics through time, we compared the observed LAR with that of a null model. This null model was based on 50 random walk simulations using the methods described above. We would expect the observed LAR to be greater than the null model if individuals were actively maintaining associations with tagged conspecifics, but less than the null model if individuals were actively avoiding conspecifics.

RESULTS
Between August 2010 and August 2020, 5 458 680 detections from 166 tagged C. porosus were recorded on the Wenlock River acoustic receiver array. Time of year significantly influenced the detectability of tagged individuals, with detection frequency (GLMM: c 2 11 ¼ 611 226, P < 0.01) and number of crocodiles detected per receiver (GLMM: c 2 11 ¼ 223.34, P < 0.01) at their lowest during March and April, and at their highest during September ( Fig. 2a and b).

Crocodile Social Organization and Structure
From August 2010 to August 2020, we identified 43 589 cooccurrence events involving 159 tagged crocodiles (males: N ¼ 101, 0.56e4.67 m TL; females: N ¼ 58, 0.84e3.23 m TL). Cooccurrence events were observed throughout the extent of our acoustic array, with most of these events occurring within the trapping area (Fig. 1). A significant positive correlation was observed between the number of crocodiles detected and the number of co-occurrences per month (GLMM: c 2 1 ¼ 8687.67, P < 0.01). Crocodiles were observed spending 10 ± 16% of their time associating with tagged conspecifics, with associations typically lasting for 12.5 ± 26.4 min (mean ± SD; range 1 se16 h). Cooccurrence events were primarily composed of pairs of conspecifics (92.9%); however, groups of up to five individuals were also recorded (N ¼ 6; Appendix Fig. A3).

Ontogeny influences social structure and stability
Maturity status was found to have a major influence on the composition and stability of associations between tagged conspecifics, with the majority (60.4%) of associations occurring between mature and immature individuals. There was a significant interaction between maturity status and time of year when predicting the number of co-occurrences per receiver (GLMM: c 2 22 ¼ 913.14, P < 0.01; Fig. 4a) and the duration of co-occurrences between conspecifics (GLMM: c 2 22 ¼ 15 428, P < 0.01; Fig. 4b). For immature crocodiles, both the number of associations and the proportion of time associating with conspecifics remained stable throughout the year. In contrast, associations involving at least one mature individual displayed a distinct cyclic trend, which peaked at the onset of the mating season in August for both the number of associates and the proportion of time associating with tagged conspecifics (Fig. 4).
The associations of matureeimmature dyads were found to be stable across years, with individuals actively maintaining associations with conspecifics for up to 5 years despite periodically intercepting the null model during the wet season (Fig. 5a). In contrast, mature dyads were less stable across years: the probability of individuals reassociating decayed successively across years (maximum ¼ 2.5 years) and did not differ from our null model ( Fig. 5b and c). The sex composition of mature dyads was also found to influence their temporal stability, with same-sex mature dyads (maleemale, femaleefemale) exhibiting stable associations between conspecifics within (but not across) years (Fig. 5b) and mixed sex (maleefemale) mature dyads found to be high cyclic with seasons (Fig. 5c). Dyads containing only immature crocodiles had the least stable associations between conspecifics, with associations not differing from the null model and persisting for ca. 1.5 years only (Fig. 5d).

Crocodile social structure is seasonal
Time of year significantly influenced both the transitivity (GLMM: c 2 11 ¼ 117.22, P < 0.01) and the modularity (GLMM: c 2 11 ¼ 79.34, P < 0.01) of the crocodile social network. A distinct cyclic pattern was present in the transitivity of the network, with transitivity peaking during October towards the end of the mating season and prior to the onset of the Austral wet season (Fig. 6a). The modularity of the network also increased throughout the year from May onwards (Fig. 6b) but disintegrated during JanuaryeApril where our least-squares estimates of modularity dropped below 0.2 indicating homogeneity in the network during this time (Fig. 6b).

DISCUSSION
Using a combination of acoustic telemetry and social network analyses, we discovered that the social system of a population of estuarine crocodiles is structured. In addition to having consistent spatial overlaps with conspecifics between months and across years (Baker et al., 2021), crocodiles displayed both long-term preferences and aversions within areas of spatial overlap towards tagged conspecifics. Furthermore, our results suggest that crocodile social systems are dynamic, where social connectedness and the formation of communities are influenced not only by the time of year but also by the sex and maturity status of individuals. These findings add to a growing body of literature demonstrating the prevalence of structured social systems in species typically considered to be solitary Clark et al., 2012;Elbroch & Quigley, 2016;Mourier et al., 2012).
Supporting our hypothesis, we found that the social structure of estuarine crocodiles was temporally dynamic, forming spatially segregating communities in which association rate transitivity and modularity varied between the wet and dry seasons. Association rate, transitivity and modularity were found to peak during the dry season, before disintegrating between November and December prior to the onset of the wet season in January . This is consistent with previous studies on crocodylians, which also found that the spatial overlap between conspecifics, along with the performance of social displays and agonistic behaviour towards conspecifics, was greatest during the dry season (Baker et al., 2021;Boucher et al., 2021;Staton & Dixon, 1975). Given that we observed the decrease in associations prior to the onset of the wet season, this suggests that the observed cyclical patterns were due to the aggregation of conspecifics rather than a drop in acoustic receiver performance . Similar trends have been observed in the associations of other seasonally reproductive species, with rhesus macaques, Macaca mulatta, Tasmanian devils, Sarchophilus harrisii, and bottlenose dolphins, Tursiops aduncus, displaying more densely connected and clustered social structures during the mating than the nonmating season (Brent et al., 2013;Hamede et al., 2009;Smith et al., 2016). Further research is required to examine how seasonal shifts in the mating system influence the social structure of estuarine crocodiles. For instance, what is occurring during the wet season when the social system of crocodiles appears to disintegrate? While it has been shown that mature female crocodiles leave the river to build and defend nests between October and May (Baker et al., 2019), it is unclear what male crocodiles are doing during this period when they depart our acoustic array (Campbell et al., 2013;Grosell et al., 2020). As seen in other regions, mature males may be travelling beyond the extent of our acoustic array to capitalize on seasonal resource pulses (Gallagher et al., 2018;Grigg & Kirshner, 2015), raising further questions as to whether individuals display fidelity to these feeding areas and the potential of forming social ties. Regardless of what is influencing this broad scale seasonality in social structure, our findings demonstrate that crocodiles inhabit a temporally dynamic social system where individuals actively modulate where and when they associate with conspecifics throughout the year.
While communities have been well described within groupliving species such as dolphins (Zanardo et al., 2018), sea lions (Wolf et al., 2007) and kangaroos (Best et al., 2013), the presence of communities in solitary species have only more recently been observed (e.g. black-tipped reef sharks, Carcharhinus melanopterus, (d) Figure 5. Lagged association rate (LAR) estimating the probability that tagged estuarine crocodiles will continue to associate with tagged conspecifics across extended periods of time. LAR estimates were created separately for (a) matureeimmature dyads (b) mature same-sex (i.e. maleemale, femaleefemale), dyads, (c) mature maleefemale dyads and (d) immature dyads. The black line represents the observed LAR and the red line represents the null LAR with the red fill representing ± SD of the null LAR. Grey bars indicate the wet season. Mourier et al., 2012;white sharks, Carcharodon carcharias, Anderson et al., 2021). Despite the observed seasonality, we found that estuarine crocodiles formed spatially segregated communities along the river system. Furthermore, the formation of these communities was found to begin between May and June, prior to the onset of the crocodiles' mating season (August and November). Familiarity with conspecifics has been shown to decrease the costs associated with territorial defence and agonistic interactions, increasing individual fitness through the formation of stable 'neighbourhoods' (Siracusa et al., 2017(Siracusa et al., , 2019(Siracusa et al., , 2021. Indeed, we found that individuals maintained associations with specific conspecifics for up to 5 years. This suggests not only that familiarity is present between conspecifics, but also that community membership may be stable across years. Thus, an increase in associations between conspecifics and the formation of communities prior to the breeding season may enable individuals to refamiliarize themselves with conspecifics and decrease the costs of territorial defence during the mating season. Further studies are required to examine the mating systems of crocodiles, in particular how an individual's social status and familiarity with conspecifics influence its reproductive success, and how this in turn influences the social dynamics of the population. This, however, raises the question of what is causing the observed community delineation. As the territories of male estuarine crocodiles on the Wenlock River are more or less fixed in space (Baker et al., 2021), one potential explanation is that these communities may be forming around the territories of dominant males. This is consistent with previous studies of estuarine crocodiles and other crocodylians, with dominant male 'boss crocs' forming yearround territories in which they control access to resources (i.e. females, food, basking areas) through the select exclusion of conspecifics (Lang, 1987;Messel et al., 1981;Webb & Messel, 1978). Alternatively, the delineation of these communities may have been due to shifts in habitat type and/or structure. Individual habitat specialization and how resources are distributed throughout the environment can influence and drive the association patterns of conspecifics (Cantor & Farine, 2018;Peignier et al., 2019;Sheppard et al., 2021). For instance, habitat-driven community divisions are common within delphinid societies (Lusseau et al., 2006;Wiszniewski et al., 2009;Zanardo et al., 2018), with differences in water depth, benthic habitat and prey assemblages influencing community differentiation. While substantial variation is present in the salinity, vegetation structure, topography, benthic substrate and prey assemblages throughout the extent of our acoustic array Grosell et al., 2020;Hanson et al., 2015;Herbert et al., 1995), further study is needed to understand how spatial variation in habitat type, resource abundance and the distribution of large adult males influences the formation, composition and spatial distribution of estuarine crocodile communities.
We found that the temporal stability of associations between dyads changed depending on the maturity status combination of the pair. Matureeimmature dyads not only had a greater association rate than matureemature or immatureeimmature dyads, but they also had the most stable associations through time (for up to 5 years). This is consistent with previous work which found that mature individuals display greater tolerance towards immature conspecifics as they are more able to outcompete them for food or basking resources and they do not represent potential competitors for mates (Brien et al., 2008;Kay, 2004;Teichroeb et al., 2014). In contrast, dyads that were the same maturity status displayed less temporally stable associations. While immature dyads displayed consistent levels of association throughout the year, the associations within dyads were inconsistent and only persisted for up to 18 months. In comparison, while mature dyads displayed a more seasonal association pattern, these dyads (particularly maleefemale dyads) actively maintained these associations for 2e3 years. The seasonality present in dyads involving mature individuals (including matureeimmature dyads) corresponds to known periods of crocodile absence in the study system. As discussed, while further research is required, these results suggest that the movement and presence of mature individuals may be influencing the observed seasonality in the social structure of crocodiles.
Acoustic telemetry relies on an array of spatially fixed acoustic receivers to detect and monitor the movements and behaviours of tagged individuals. As such, associations between tagged crocodiles are only identified when both individuals are in the water and within the detection field of a receiver, and any associations with nontagged individuals are not represented. While we tried to minimize this issue by deploying an extensive array of receivers and tagging a large proportion of the Wenlock River crocodile population (ca. 70% of individuals), it is likely that we have underestimated the association rate between conspecifics and therefore only revealed a component of the social structure in this population. However, rather than detracting from the findings of this study, this instead suggests that the social systems of estuarine crocodiles may be even more complex than that described here. Further studies into the social behaviours of estuarine crocodiles using complementary approaches (e.g. proximity tags, direct observations) would build upon the findings of this study and provide further insights into the social systems of crocodylians.
Through the combination of acoustic telemetry and social network analyses, we provide new insights into the social organization and structure of estuarine crocodiles. Our findings highlight that rather than being asocial and intolerant of conspecifics, estuarine crocodiles instead are flexible in how they associate with conspecifics depending on their maturity status and the time of year. These results add to a growing body of literature demonstrating the presence of complex social systems in nongroup-living species Clark et al., 2012;Elbroch & Quigley, 2016;Mourier et al., 2012), highlighting that species do not have to be group living to be social, and that nongroup-living species are not necessarily asocial.

Data Availability
Data were sourced from the Acoustic Animal Tracking Database (https://animaltracking.aodn.org.au) of the Integrated Marine Observing System (IMOS, www.imos.org.au). IMOS is a national collaborative research infrastructure supported by the Australian Government. The database is a centralized acoustic telemetry data repository maintained by the IMOS Animal Tracking Facility and the Australian Ocean Data Network (AODN, https://portal.aodn.org.au/). The processed data and all R code supporting the analyses reported in this article are available from GitHub: (https://github.com/ Cameron-J-Baker/Baker_et_al_2023_Ani_Behav).

Declaration of Interest
We declare no competing interests.