Gap analysis of acoustic tracking data reveals spatial and temporal segregation in sympatric reef sharks

Background There are now a wide array of eld and laboratory techniques available for gaining insight into the movement and behaviour of sharks. Although acoustic telemetry may lack the ne-scale resolution of some satellite technologies, the low cost and longer battery life make it a powerful tool for investigating elasmobranch behaviour. Here, we develop a novel approach to analysing acoustic telemetry data, using detection gaps to infer movement patterns to and from monitored reef habitats, to investigate spatial and temporal segregation between two sympatric shark species in a large remote MPA. Methods A total of 102 grey reef sharks (Carcharhinus amblyrhynchos) and 76 silvertip sharks (Carcharhinus albimarginatus) were tted with long-term acoustic transmitters and tracked inside a large acoustic array of reef-based receivers in the British Indian Ocean Territory MPA, between 2014 and 2018. From the resulting dataset (768,081 detections), movements between receivers and recursive loops to the same receiver were identied. Using the durations of inter-receiver movements (i.e. detection gaps), individual behaviours were classied into ‘restricted’ or potential wider ‘out of range’ movements. Drivers of these movements were identied using network analysis, GLMMs and multi-model inference starting from an a priori set of explanatory variables. Results In general, silvertip sharks were more likely to undertake ‘out of range’ movements than grey reef sharks. ‘Out of range’ movements were more common at night compared to during the day, and during the wet season compared to the dry season. In addition, the difference in ‘out of range’ movements between the two species increased at night. These results suggest spatial and temporal segregation of movements between the two species. Conclusions We present a novel analysis of detection gaps from acoustic telemetry data to infer differential movement patterns and describe how species organise in space and time. Furthermore, this approach shows that acoustic telemetry gap analysis can be used for comparative studies, or combined, with other research techniques to better understand the functional role of sharks in reef ecosystems, moving towards more informed strategies for the conservation and management of the marine environment.

Network analyses of movements derived from acoustic telemetry are becoming more commonplace for exploring spatial and temporal movement patterns within acoustic detection data (20,23,33). To date, gap analysis has been used for studying the ecology of marine vertebrates, for example, using time periods between location xes from satellite telemetry to determine directed or restricted movements for optimal foraging in pinnipeds (34). However, typically, network analyses using acoustic telemetry often ignore the gaps between detections. These gaps can be informative for inferring the length of time taken between movements, and as a proxy of tortuosity, as the longer the duration between two points the greater the tortuosity of the movement is likely to be. Therefore, analysis of detection gaps from acoustic telemetry data can be useful for investigating coarse scale behaviour and its associated timings. For example, gaps might be used to inform the likelihood of elasmobranchs moving out of MPAs into unprotected waters, where they may be vulnerable to exploitation from commercial sheries (6). These gaps can also be used to estimate the timings of ontogenetic habitat shifts, when individuals begin leaving nursery areas for longer periods (35), as well as for more accurately determining the timings and thresholds to de ne residency events for spatial distribution and movement analyses (36).
In this paper we use data gathered across a large array of acoustic receivers monitoring reef-associated shark species throughout the BIOT MPA. Grey reef sharks (Carcharhinus amblyrhynchos) and silvertip sharks (Carcharhinus albimarginatus) were used as model species for this study as they are the most abundant large predator species in BIOT but also a target for illegal shing activity that continues to plague the MPA (37). In the absence of receivers in pelagic areas and su cient satellite telemetry data to make population level inferences, we aim to 1) analyse and assess the potential of detection gaps from acoustic time-series data to characterise 'restricted' reef-based movements, or movements 'out of range' of receivers, which could potentially indicate wider ranging, off-reef movements; 2) investigate the spatial and temporal segregation between sympatric elasmobranchs species in the BIOT MPA.

Acoustic tracking
Data collection was undertaken in the BIOT MPA between February 2013 and April 2019. Throughout the archipelago there have been situated up to 93 permanent and temporary acoustic receivers (VR2W, VR4-UWM, VR4G and VR2AR receivers, Vemco Inc, Nova Scotia, Canada), as con gured in Fig. 1. The BIOT MPA is characterised by numerous small islanded atolls with submerged banks and reefs, with depths of 1,000 m or more separating each atoll or reef system (38). As such, acoustic receivers in the BIOT MPA are mainly based on areas accessible to divers, such as coral reef systems, with few receivers covering the deep pelagic waters of the region. This array was initially started in 2013, expanded throughout subsequent years [for more information see Carlisle et al. (6)], covering a perimeter of 700 km and an area of 25,500 km 2 within the MPA, for the detection of acoustically tagged marine fauna. Of the 93 receivers, 82 are in depths 45 m or less. All receivers were situated far enough apart to avoid overlap in their detection range, with mean distance to closest receiver 2.15 km, with a range of 0.55 -4.57 km. The frequency distribution of inter-receiver distances can be found in Additional le 1: Figure S1.
Although range testing has not been undertaken for this array due to nancial and logistical constraints of vessel time in the BIOT MPA, other studies conducted around coral atolls in the Indian Ocean using the same or similar equipment have found detection ranges between 300 and 500 m (14,(39)(40)(41). Maps of receivers at the different reef systems within BIOT, and an estimated 500 m detection range, can be found in Additional le 1: Figure S2 We tagged 102 grey reef and 76 silvertip sharks with acoustic transmitters across nine different locations following the methodology described by Carlisle et al. (6). Of tagged grey reef sharks, 76 were female and 26 were male. Of the tagged silvertip sharks, 45 were female and 31 were male. As in previous studies (7,42), silvertip sharks (mean total length, TL = 123.56 cm ± S.D. 19.14) were on average slightly larger than grey reef sharks (mean TL = 119.15 cm ± S.D 18.07). Detailed metadata for each tagged individual can be found in Additional le; Table S1. Tags acoustically transmit a unique ID code with a nominal delay of 60-180 secs for the duration of their battery life (~10 years), providing a long-term time series of detection data. Receivers were downloaded and serviced annually at the same time each year (March-May).
Only complete years were included in the analysis, with the nal data set covering shark detections from 2014 -2018. Prior to analysis, data from the seamounts were removed. Seamounts in the BIOT MPA differ ecologically from the rest of the archipelago and recent analyses suggest that they are potentially important drivers of silvertip shark movement (unpublished data). While we acknowledge the ecological importance of seamounts, we only had acoustic equipment on two of these features in the south of the MPA. Despite the small number of receivers on these sites, a signi cant number of detections were produced (1,165,546 from 2014-2018) that could signi cantly impact our analyses. As such, in order to robustly investigate segregation of reef sharks in coral reef habitat, seamount data were excluded.

Movement classi cation
Using a movement network approach, a 'transition' was deemed to have occurred when an individual left one receiver and arrived at another in a new location (26). Alternatively, an animal might leave a receiver, move out of detection range and then return to the same location, a 'self-loop' in network parlance, but here called a 'recursion'. Classi cation of 'restricted' and 'out of range' activity was, therefore, inferred based on the duration of these two different types of movement (hereafter transitions and recursions) before being used as a binary response variable in subsequent models (Fig. 2).
All analyses were conducted in R version 3.6.0 (43). Classi cation of 'restricted' and 'out of range' movements was conducted using an optimal classi cation method, where similar data values are placed in the same class by minimising an objective measure of classi cation error (44). For recursions, detection gaps of less than six minutes (minimum of two detections) were removed from the data as an initial lter to ensure a recursion had taken place, rather than an animal had stayed in the same location but a detection had been missed. Time differences for recursive movements per species were log transformed to normalise the data, and the 'classIntervals' function in the classInt package (45) used to calculate thresholds between 'restricted' and 'out of range' movements. The Fisher algorithm was used, which determines thresholds by minimising the sum of absolute intra-class mean variance, as well as maximising inter-class mean variance (44,46). This resulted in a threshold of 91 minutes for grey reef sharks and 64 minutes for silvertip sharks for 'restricted' activity, beyond which it was assumed that the shark had conducted an 'out of range' movement.
Transitions were subject to a separate ltering process. Unlike recursions, no initial lter was required for transitions as the detection of an individual on one followed by another receiver is immediately indicative of a movement from one location to another. Temporal gaps in the detection data for any given pair of receivers were informed by both the distance and species-speci c minimum sustainable swim speeds (0.69 m/s for grey reef sharks and 0.73 m/s for silvertip sharks) (47). For example, the predicted transition duration of a direct movement of a shark between two receivers, without deviation, would be the ratio between distance and speed. As such, by rst calculating expected time for a transition using swim speeds and distance between receivers, the Relative Deviation from this Expected Time (RDET) between any pair of receivers was determined by dividing the expected transition time by the observed transition time. RDET values of > 1 were movements faster than expected, and values of < 1 slower/more tortuous than expected.
For transitions, log transformed RDET values were calculated for both species, and, as with recursions, the same optimal classi cation method for determining thresholds was used (44). Movement values of greater than the threshold value of 0.164 for grey reef sharks and 0.128 for silvertip sharks were determined as 'restricted', with values less than the thresholds determined as 'out of range'. Animals rarely travel in straight lines and often vary in their tortuousity depending on factors such as resource use, habitat quality, competition and predation (48-50). These thresholds of 0.164 and 0.128 are, therefore, very conservative to allow for a tortuous movement to occur and still be classi ed as 'restricted' in each species. Finally, recursions and transitions were combined so that every movement was categorised as a binary response (restricted = 0, out of range = 1) (Fig. 2).

Data analysis
To explore the in uence of explanatory variables on 'out of range' movements, an information theoretic approach was taken, which accounts for model selection uncertainty (51,52). In recent years, information theoretic approaches have become a staple for modelling ecological systems, particularly those where explanatory models describing the system may have similar complexity and t the data equally well, such as understanding the spatial distribution (53-55), behaviour (56, 57), and anthropogenic impact on survival of wildlife populations (58, 59). To limit exploratory analyses, and prevent model over tting, an a priori selection of variables and interactions based on previous research and theory was conducted (52, 60, 61). Explanatory variables included in the model were 'species', 'sex', 'size', 'season' (wet/dry) and 'diel period' (day/night) (7, 17, 62-64). As size had a non-normal distribution it was log transformed. The BIOT MPA is located near the equator and has a roughly 12-hour day/night cycle. As such, day was designated from 0700 to 1900 and night from 1900 to 0700 following sunrise and sunset times obtained from https://www.timeanddate.com. The MPA experiences distinct Indian Ocean wet and dry seasons with wet season running from October to March, and dry season from April to September (65). Seasonal variability is often greater than monthly variability in tropical ocean systems (66, 67), and, therefore we deemed season a more biologically relevant driver of shark movement.
All variables used in the model were assessed for multicollinearity. Multicollinearity, which occurs when predictors in a multiple regression are highly correlated (68), was assessed by producing a variance in ation factor (VIF) using the 'check_collinearity' function in the performance package in R (69). VIF measures the degree of multicollinearity in a regression model by providing an index of how much the variance of the model variables increases due to collinearity (70). No evidence of collinearity was found, with all variables having a VIF ≤ 1.05, less than the critical threshold of 5.0 (see Additional le 1: Table S2) (68, 71). As such, all a priori selected explanatory variables were included in the global model.
A global model was subsequently created using a generalised linear mixed model (GLMM) (family = binomial, link = logit) in the glmmTMB package (72). To explore putative spatial and temporal segregation between grey reef and silvertip sharks, 'species' was included as interaction term with all explanatory variables and individual ID as a random factor. As the likelihood of a movement between locations decays as a function of distance (D. Jacoby unpublished data), receiver location was also included as an independent random factor. Residuals of the global model were checked for heteroscedasticity, autocorrelation and errors were checked for binomial distribution using the functions 'resid', ' tted', and 'acf' from the stats package (43) and found free from autocorrelation and heteroscedasticity of residuals (Additional le 1: Figure S5).
To generate the model set from the global model, the 'dredge' function from the MuMIn package was used (73). Models in the set were ranked by Akaike Information Criterion (AIC) values. If no single parsimonious model results from the set and the weight of the best model is less than 0.9, model averaging is recommended over model comparison (61), and a con dence set, most likely to represent the system, selected. Only models with DAIC <2 were chosen for the con dence set (51,74). A model averaging approach was then undertaken on the con dence set (61) using the 'model.avg' function in the MuMIn package (73). This function calculates Akaike weights based on the con dence set (61). Model averaged estimates and con dence intervals (97.5%) for each explanatory variable and interactions were calculated (52). The relative importance of each predictor variable (relative to other variables in the con dence set) was calculated by summing Akaike weights for all con dence set models containing them (51).
Model averaged estimate values indicate the probability of observing an 'out of range' movement as the value for continuous predictor variables increase. Categorical predictor variables were compared to the categorical variable level used as the model baseline. Positive estimates indicate an increased probability of 'out of range' and a decreased probability of 'restricted' movements; negative estimates, the opposite. It is important to note that using this method it is possible that some levels of categorical predictors may display a high relative importance value but show no signi cant result in the model averaged estimates, as these are dependent on the baseline chosen (75).
Therefore, both the relative importance and model averaged estimate results should be considered in combination (56, 61). Marginal R 2 (R 2 m) and conditional R 2 (R 2 c) values were calculated, using 'r.squaredGLMM' in the MuMIn package (73,76), to assess variance in the xed effects and the combination of xed and random effects, respectively (76, 77).

Model cross-validation
To assess the predictive capabilities of our nal model, analysis was conducted on 80% of the data. Cross-validation of the model averaged estimate values from the con dence set of models was conducted on the remaining 20% of data as con rmatory analysis to quantify how well the selected model performed (52). The 'predict' function in the glmmTMB package was used to validate the expected outputs of the multi-model inference on the observed values from the reserved 20% of the data. Area under the receiver operating characteristic curve (AUC) values designate the probability that positive and negative instances are correctly classi ed (78). As such, AUC was calculated using the pROC package (79), as a threshold-independent method to check the robustness of the model. An AUC value of greater than 0.7 indicates better than random performance and that the model is a good representation of the system being evaluated (78, 80, 81).

Results
There were 768,081 detections, and movementstransitional and recursive, identi ed from 102 grey reef sharks and 76 silvertip sharks between 2014 -2018.

Model selection
Following the dredge of the global model, no single parsimonious model was found. The AIC of the null model was ranked 95 th of available model combinations (Additional le 1: Table S3), indicating that xed effects were an important inclusion to the model. The large number of models in the con dence set indicate uncertainty, with Akaike weights con rming that the top model is no more likely to be the best than any of the others (53) ( Table 1). The weight of the best model was 0.17 (Additional le 1: Table S3). As such, although model averaging and model comparison may lead to similar levels of accuracy (82), model averaging was used on this occasion. Six models had ΔAIC <2 and formed the con dence set for model averaging (Table 1) where model weights were recalculated (61). The results from model averaging are presented in Table 2. Five variables and three interactions had a relative importance greater than 0 and were deemed as important explanatory variables of 'out of range' movements (Table 2). Three variables and one interaction had a signi cant effect on 'out of range' movements (Table 2).

Model averaging
Species, diel period and season were signi cant predictors of 'out of range' movements. Silvertip sharks were overall more likely to undertake 'out of range' movements compared to grey reef sharks (p <0.001, Table 2) indicating spatial segregation between the species.
However, as both species still undertook regular 'out of range' and 'restricted' movements, this segregation is not discrete. 'Out of range' movements were more likely to occur at night than during the day (p <0.001, Table 2) (Fig. 3a), and during the wet season than the dry season (p <0.001, Table 2) (Fig. 3b). Interaction effects revealed temporal segregation of movements between the species with the difference in 'out of range' movements between silvertip sharks and grey reef sharks greater at night than during the day (p <0.001, Table   2) (Fig. 3a).

Discussion
Current analyses of acoustic telemetry data often neglect the timings and periodicity of detection gaps. In the absence of more accurate movement data for large numbers of animals, which can be prohibitively expensive using satellite telemetry, we have developed a new approach, which utilises gaps in detections from acoustic telemetry to infer presence or absence from regions of interest, such as coral reef systems. Prior research on patterns of movements between coral reef-associated elasmobranch species have been limited but primarily focus on differences in space use (17,83,84). To date, temporal aspects of segregation are rarely considered (85, 86). As seen here, the use of time dependent factors, such as detection gaps, has the potential, albeit at coarse scales, to identify both spatial and temporal differences in movement in sympatric shark species enhancing our understanding of the organisation and reorganisation of reef predator assemblages.
Grey reef and silvertip sharks in the BIOT MPA had signi cant differences in 'out of range' movements which can be inferred as wider, offreef movement activity. Overall, silvertip sharks were more likely to undertake these potential wider, 'out of range' movements than grey reef sharks. These results suggest spatial segregation between these species, with grey reef sharks, probabilistically, more likely to inhabit reefbased areas, while silvertip sharks were more mobile and conducted more widespread movements (Fig. 3). Our results indicate that although there is spatial segregation between the species, there is probable overlap between the two in the areas they reside. This supports evidence from isotope data which suggests resource partitioning between these two species in the BIOT MPA, with both species utilising both reef and pelagic areas for foraging, but with grey reef sharks obtaining 78% of their biomass from reef resources, but silvertip sharks only 60% (87). In addition, our results are consistent with work from the Great Barrier Reef suggesting grey reef sharks are true reef sharks, and silvertip sharks less reliant on reef-based resources in this region (88). Further, these results extend previous research describing variable patterns of movement and activity in both grey reef sharks and silvertip sharks globally (89-91). Locally, previous research in the BIOT MPA found that silvertip sharks had higher mobility, larger activity spaces, and lower reef residency compared to grey reef sharks, which had small activity spaces and rarely moved more than ve to ten kilometres from their tagging location (6). This study has shown that not only can gaps in detections be used to help interpret and support other types of data, but, even at coarse scales, can potentially be used to provide insight into the differences in spatio-temporal movements that underpin reef predator sympatry.
The diel variance in probable wider 'out of range' movements found in the study also indicates temporal segregation of movements between the two species, with silvertip sharks more likely to conduct 'out of range' movements at night, and grey reef sharks showing very little diel change in such movements (Fig 3a). At several other reef locations grey reef sharks are central place foragers (88, 89, 92), while silvertip sharks are more dispersive, and less site attached (88, 91). As foraging occurs at night, it may be that silvertip sharks move off-reef to forage, (i.e. increased 'out of range' movements), while grey reef sharks forage on-reef, which would support the differential patterns of resource use found between these species (87). Silvertip sharks may also occupy a greater spatial range at night during foraging, as has been seen in other reef shark species, such as whitetip reef sharks, Triaenodon obesus (93) and blacktip reef sharks, Carcharhinus melanopterus (89), whilst grey reef shark spatial range rarely changes. Alternatively, the reduction in detections at night, could be the avoidance of transient apex predators, such as tiger sharks, Galeocerdo cuvier, which are known to feed on reef-associated mesopredators, like silvertip sharks (94)(95)(96). Although few studies have been conducted on the movements of silvertip sharks, differing patterns from the literature on grey reef sharks suggest that diel movement patterns in this species may be population dependent, possibly driven by interspeci c interactions and/or local environmental conditions (39,89,90). For example, grey reef shark populations in Palau utilise shallower reef platforms for foraging during the night, with deeper, pelagic, waters exploited during the day (90). The same species around Palmyra Atoll in the Central Paci c Ocean, however, are detected on the forereef frequently during the day with far fewer detections at night (89), suggesting they move away from the reefs during these periods. Currently little is known about the foraging ecology of these two species in the BIOT MPA and further studies will be needed to investigate the apparent diel differences in 'out of range' movements between these two species in the region, as well as further research to investigate regional and population patterns of movement both within and between different reef shark species.
There was also seasonal variation in 'out of range' movements of both species (Fig 3b). Knowledge of seasonal patterns of elasmobranch movement in the region is important to aid enforcement efforts to prevent Illegal, Unreported and Unregulated (IUU) shing in the BIOT MPA (37). Like diel period, seasonal patterns of movement are common in multiple elasmobranch species. Such seasonality may be driven by mating in adults, ontogenetic expansion of home range, or foraging opportunities (64, 97, 98). Additionally, environmental conditions across the BIOT MPA are subject to considerable changes associated with the Indian Ocean monsoon current. In many species however, seasonal changes in movement may not be clear, with movements either unpredictable or subject to individual behavioural variability (98). Indeed, our model variance results and the low marginal R 2 values from our xed effects suggest that inter-individual variability also plays an important role in explaining the probability of 'out of range' movements (99). Inter-individual variability in movement behaviour is seen across shark species (100)(101)(102), and may be driven by differences in age class, sex, morphology, health and even personalities (103,104), as well as intra-speci c competition (92, 105), but currently is not fully understood (106).
Model results also indicated an increase in 'out of range' movements during the wet season. However, the reasons for this increase are unclear. As the wet season (October -March) coincides with the peak historic pelagic shing season in the BIOT MPA (107), this increase in movement may possibly be linked to a seasonal food resource, as previously hypothesised by Curnick et al. (87). Although food may be a primary driver of temporal changes in movement, factors such as thermoregulation and predator avoidance, and environmental variables such as sea surface temperature, salinity and current, have been found to also drive temporal changes in movement (91,108). Despite recent developments to our understanding of how environmental variables effect elasmobranch movement (108, 109), they are still relatively poorly understood, and investigations should be undertaken to further explore their impact on both spatial and temporal aspects of elasmobranch movement in this region.
Because acoustic telemetry only measures presence, array design and detection ranges can signi cantly impact results obtained using this technology (6). An inability to detect an animal could be due to the animal leaving the study area, or because it moved out of detection range (5), and consequently might result in some mis-designated movements in this study. Detection ranges for the region vary between 300 and 500 m (14,(39)(40)(41), and distances between receivers ranged, between 0.55 -4.57 km with mean distance to closest receiver 2.15 km, with minimal overlap between receivers (Additional le 1; Figure S1-4). We acknowledge that this could lead to signi cant 'blind spots', where a shark could be swimming and not detected rather than engaging in wider 'out of range' movements. We consider the impact of these, however, to be minimal for the following reasons; classi cation thresholds of movement were very conservative, giving considerable leeway for an animal to move around a reef area, detections to be missed, and still the movement be classed as 'restricted'; furthermore, neither of these species are able to rest motionless on the bottom, as they are required to ram ventilate (110), increasing the chances of them being detected on the same or additional receivers even when frequenting 'blind spots' within the array.
In addition, animals using the lagoons of these atolls would also not necessarily be detected on the array, and movements across the lagoons could also be mis-designated. However, lagoon use in these species tends to be minimal (42,111,112) and the isotope signatures obtained from these two species indicate they are not using lagoons for foraging (87). As such, in addition to our ndings complementing the work of Curnick et al. (87), we believe these potential issues should not have impacted our results signi cantly. However, we stress that this is an inference method being used in lieu of more accurate measurements for large numbers of free-ranging individuals and may not be suitable for regions with large gaps in receiver detection ranges, particularly when the study species may exhibit bouts of sedentary behaviour.
Although the consistent nature of our results with those obtained from stable isotope work (87) help to validate our methodological approach, further validation of our classi cations could be carried out using more accurate positional data. Although there were several silvertip sharks double tagged with both acoustic and PSAT tags, unfortunately the positional data from these satellite tags was not of a high enough resolution to fully con rm our claims. At low latitudes, the error associated with light-level geolocation estimates from PSAT tags can be very high (4). For example the geolocation error from a silvertip shark tagged with a PSAT in the BIOT MPA was estimated at 0.25° or 27.83 km at the equator (6). In addition, geolocation algorithms used to reconstruct positions from PSAT data only produce a single position per day, which limits their ability to investigate diel differences in location. Although not feasible in this study, the accuracy of this technique should be validated in future studies with more accurate positional data, such as those derived from Fastloc GPS tags (e.g. Smart Positioning Or Temperature transmitting tags).
There are no single 'silver bullet' research techniques for investigating elasmobranch ecology at an appropriate and meaningful spatial and temporal scale; each methodology has its limitations. However, by combining data from multiple approaches the limitations of any individual methodology can be overcome, and a more holistic understanding of elasmobranch ecology can be attained. Here we show that, in the absence of ner scale movement data beyond the boundaries of our acoustic detection ranges, gaps in detections can be used to support and extend conclusions about how variable behavioural strategies can in uence inter-speci c species organisation on coral reef systems. This approach can thus be used to greatly aid the overall understanding of the role of multi-species predator assemblages within coral reef ecosystems.

Conclusion
In the absence of accurate satellite tracking data, we have developed a novel methodological approach for analysing acoustic telemetry data, using absences in detections to classify 'restricted' and potential wider 'out of range' activity in two sympatric shark species. Our results highlight coarse diel and seasonal differences in the movements of grey reef and silvertip sharks in the BIOT MPA, potentially providing evidence of spatial and temporal segregation in these species. As apex or meso-predators, shark species play an important role in the ecological function of reef, pelagic and coastal ecosystems (25,29). Knowledge of how sympatric shark species interact with each other is therefore important to further determine their ecological function in coral reef ecosystems, which is currently poorly understood. Analysis of movement to investigate segregation in elasmobranchs has, to date, primarily focused on how spatial aspects of movement vary (17,83). However, temporal aspects of movement are also important for investigating interspeci c interactions. Despite potential limitations for the technique for some regions and species, this methodology has broad potential for investigating movement ecology of marine species in the absence of more accurate, expensive, technologies, or where receivers only cover a singular habitat. To our knowledge, this is the rst analysis of acoustic telemetry data to focus on temporal gap analysis to informing movement ecology and to infer both spatial and temporal segregation between two species. This is critical information for the management and conservation of threatened ecosystems such as coral reef systems (83), as well as aiding enforcement efforts of MPAs to mitigate against illegal shing activity that still threatens many protected reserves today. Availability of data and material The datasets and code for the analyses conducted during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.

Additional File Legend
Additional le 1.pdf Figure S1. Histogram of inter-receiver distances in the BIOT MPA. Figure S2. Map of Egmont reef with receiver locations indicated with red dots and a 500m metre buffer of the estimated receiving range around each receiver. Figure S3. Map of Peros Banos reef with receiver locations indicated with red dots and a 500m metre buffer of the estimated receiving range around each receiver. Figure S4. Map of Salomon with receiver locations indicated with red dots and a 500m metre buffer of the estimated receiving range around each receiver. Figure S5. Plots of model residuals to check for homoscedasticity and autocorrelation. Table S1. Metadata for tagged individuals analysed from 2014 to 2018. Table S2. Outputs from collinearity tests using 'check_collinearity' function in the 'performance' package. Table S3.
Model selection table following the dredge on the global model.