Home range and dynamic space use reveals age-related differences in risk exposure for reintroduced parrots

Individual-level differences in animal spatial behaviour can lead to differential exposure to risk. We assessed the risk-exposure of a reintroduced population of kākā (Nestor meridionalis) in a fenced reserve in New Zealand by GPS tracking 10 individuals and comparing the proportion of each individual’s home range beyond the reserve’s fence in relation to age, sex, and fledging origin. To estimate dynamic space use, we used a sweeping window framework to estimate occurrence distributions from temporally overlapping snapshots. For each occurrence distribution, we calculated the proportion outside the reserve’s fence to assess temporal risk exposure, and the area, centroid and overlap to represent the behavioural pattern of space use. Home range area declined significantly and consistently with age, and the space use of juvenile kākā was more dynamic, particularly in relation to positional changes of space use. The wider- ranging and more dynamic behaviour of younger kākā resulted in consistently more time spent outside the reserve, which aligned with a higher number of incidental mortality observations. Quantifying both home range and dynamic space use is an effective approach to assess risk exposure, which can provide guidance for management interventions. We also emphasise the dynamic space use approach, which is flexible and can provide numerous insights towards a species’ spatial ecology.


| INTRODUCTION
Individual and group-level behavioral differences have been observed in a broad range of animal taxa, which represent intrapopulation diversity in physiological, behavioral, and ecological mechanisms (Bolnick et al., 2003;Dall et al., 2012;Stuber et al., 2022).These differences can manifest as the variable foraging strategies, spatial behaviors (Stuber et al., 2022), and problemsolving abilities of conspecific individuals (Loepelt et al., 2016), each of which influences population dynamics (Hanski, 1998).Extrinsic factors that influence behavior include resource availability and the density of competitors, predators, prey, or conspecifics, while intrinsic factors include age, sex, personality, and breeding status (Andreassen et al., 2002;Class et al., 2019;Clay et al., 2018;Lesmerises & St-Laurent, 2017;Loepelt et al., 2016;Reader & Laland, 2002;Stuber et al., 2022).In birds, age is considered to influence behaviors relating to problem-solving and foraging efficiency, with younger individuals being more behaviorally flexible, innovative, and able to solve novel problems (Loepelt et al., 2016;Sherratt & Morand-Ferron, 2018) while older individuals are typically more efficient foragers (Clay et al., 2018;Phillips et al., 2017).Age-related plasticity is evidenced by changes in behavior in relation to age, when there have not been selection pressures against certain behaviors (Class et al., 2019).
In a conservation context, individual-or group-level behavioral differences that lead to variability in space use may lead to differential exposure to risk, which can affect the population's or species' trajectory and persistence (Merrick & Koprowski, 2017).This is relevant for fullyfenced conservation reserves, which we consider to be discrete areas of habitat that undergo protection from threats and allow for the refuge and possible translocation of sensitive wildlife species that have been reduced or historically extirpated, even while the major causes of species decline persists in the wider region (Burns et al., 2012;Innes et al., 2019).Flighted species may "spill-over" from these reserves into the wider landscape (Tanentzap & Lloyd, 2017), and may encounter a fragmented or modified landscape containing a higher number of threats.The exposure to threats can impose a greater mortality risk outside the fence, which may be further increased if the species has lost predatoravoidance behaviors (Muralidhar et al., 2019).
Differences in spatial behaviors that lead to variable exposure to risk can be assessed through telemetry data and space-use analyses.Range distributions are a probabilistic representation of the space required by an animal over the long term (i.e., its home range), and ODs are a probabilistic representation of the space used by an animal during a defined period of observation (Alston et al., 2022;Fleming et al., 2015Fleming et al., , 2016)).We can, therefore, estimate long-term risk exposure by calculating the overlap of an individual's home range with areas of presumed risk, or risk exposure during a specified time period using an OD (Alston et al., 2022).While the home range approach provides valuable evidence and can indicate whether certain individuals or groups are likely to be at higher risk than others over long time scales, an animal's space use may not be stable, and resources or climate might drive the space use of certain individuals, groups, or even the entire population to change in position or size, leading to risk-exposure that changes dynamically.
Estimating space use dynamically using overlapping ODs can therefore provide information about the pattern of space use, and whether risk-exposure is likely to change throughout time.The term utilization distribution (UD) is often used to describe both range and ODs (Alston et al., 2022;Fleming et al., 2015), although to minimize ambiguity here we differentiate by using the term home range (HR) to denote a range distribution and OD to denote an occurrence distribution.
We aimed to assess risk exposure using range distributions and dynamic space use in a reintroduced parrot population in a fenced conservation reserve, and to highlight the utility of both approaches.K ak a (Nestor meridionalis) are large-brained, social parrots (Bond & Diamond, 2004;Iwaniuk et al., 2005) that are considered generalist, extractive foragers (Beggs & Wilson, 1987;Moorhouse, 1997).These traits suggest a propensity for exploration and innovation (Dunbar & Shultz, 2007;Sol et al., 2005), which has been observed particularly in juvenile k ak a (Bond & Diamond, 2004;Loepelt et al., 2016).Well understood threats to k ak a in wild populations are predominately the destruction of native habitat, predation by stoats (Mustela erminea) and competition for food resources by vespulid wasps and possums (Trichosurus vulpecula) (Beggs & Wilson, 1991;Greene et al., 2004;Moorhouse et al., 2003;O'Donnell & Rasch, 1991;Wilson et al., 1998).However, fragmented, semi-urban landscapes present novel threats that include human infrastructure, vehicles, cats, dogs, and traps, and toxins intended for pest species.In 2008, k ak a were translocated into the fully fenced semi-urban conservation reserve Orokonui Ecosanctuary (Te Korowai o Mihiwaka; 45 46 0 S, 170 36 0 E; hereafter Orokonui) in New Zealand, which is within a region that k ak a have been absent from for decades, and possibly more than a century (O'Donnell & Rasch, 1991).We recognized the continued presence of threats outside the reserve and hence attempted to understand to what extent movements outside the relative safety of the fenced reserve pose a risk to the birds, and whether this differs between certain groups.To assess whether analyses of home range and dynamic space use can address these questions, we GPStracked a subset of the k ak a of the Orokonui population.We assessed the k ak a's exposure to risk by estimating the proportion of their home range outside the Orokonui Ecosanctuary fence and described behavioral patterns by estimating their dynamic space use.The approach taken here has broad application for quantifying exposure to presumed risk and when this occurs, such as outside conservation reserves, within fishing grounds for marine species (Weinstein et al., 2017), and in areas with human infrastructure such as wind turbines (Bright et al., 2008).

| Ethical note
This study was conducted with approval from the University of Otago Animal Ethics Committee under the Animal Use Protocol AUP-18-237.We consulted with the Ng ai Tahu Research Consultation Committee and Kati Huirapa Runaka ki Puketeraki before undertaking the research, and all k ak a handling was undertaken by personnel trained by Department of Conservation staff with Level 3 bird banding permits.Free-ranging k ak a were captured in the reserve for tagging by luring them into a baited aviary used for soft-release of captive-raised individuals during population reinforcement translocations.All k ak a were weighed, subjected to a health check, and sexed by measuring the upper mandible (Moorhouse et al., 2008;Moorhouse & Greene, 1995).GPS devices were attached to k ak a using backpack harnesses with a weak-link (Karl & Clout, 1987), and we followed standard procedures for attaching devices and reduced handling time as much as possible to minimize stress to the k ak a. Nine of the GPS devices (Lotek PinPoint 350) were 18.4 g, and one (Lotek PinPoint 450) was 19.1 g.The weight of the k ak a ranged from 444 to 588 g and averaged 524 g, resulting in the devices averaging 3.55% of the k ak a's body mass, with a range from 3.1% to 4.3%.These transmitter weights are all less than previous tracking studies of k ak a that have observed no ill-effects (Powlesland et al., 2009;Recio et al., 2016).When possible, k ak a were recaptured to remove the GPS device when the battery was depleted, which was expected to be 5-6 months given the GPS fix rate and duration of VHF and UHF windows, and health checks were again conducted.No tags detached via the weak link before the batteries depleted, and there were no observed adverse health effects due to the GPS devices or their attachment to k ak a.

| Study site and semi-urban kaka reintroduction programs
Orokonui Ecosanctuary is a 307-ha sanctuary near the city of Dunedin (South Island, New Zealand) that is surrounded by a 1.9-m-high pest-resistance fence.Since the time the Orokonui fence was erected in 2007, the reintroduction of multiple taxa has led to positive biodiversity impacts within and beyond the fence (Aichele et al., 2021;Bogisch et al., 2016;Jarvie et al., 2016;Tanentzap & Lloyd, 2017).The vegetation within Orokonui is predominately comprised of regenerating kanukabroadleaf native forest with small areas of old-growth forest, as well as stands of exotic pine forest (predominately Pinus radiata) and mountain ash (Eucalyptus regnans).Surrounding the sanctuary is a fragmented landscape of old-growth and regenerating native forest fragments, as well as exotic conifer forest, working farms for grazing livestock, and semi-urban areas (Figure 1).At the time of the study, adjacent areas outside the fence were subject to continuous pest-control, targeting mustelids, possums (Trichosurus vulpecula), and rodents.Supplementary food is available to k ak a at feeding stations in Orokonui year-round as parrot-specific pellets and sugar water (Aichele et al., 2021).

| Data collection
To gather locational data, we fitted SWIFT fix GPS devices (PinPoint GPS VHF 350/450, 18.4 g/19.1 g, Lotek, NZ) to 10 k ak a, which were set to take a location every 3 h for the nine PinPoint 350 devices, and every 2 h for the single PinPoint 450 device due to a larger battery.The devices had VHF (very high-frequency radio) for locating animals, and UHF (ultra-high frequency radio) for downloading data remotely.We had a 4-h VHF window each day to locate the k ak a, and an 8-h UHF window each day to download the data.The UHF window was longer as it uses little battery and occasionally the data could be downloaded when individuals visited supplementary feeding stations, obviating the need for locating each individual via VHF.The GPS devices were distributed evenly between the sexes, with five females and five males, and across a range of ages from 1 to 10 years old (mean = 4.5 years).The tracked k ak a were a combination of Orokonui-fledged individuals, which are more akin to wild birds and have little human intervention besides monitoring nest success, and captive-raised individuals, which are raised in an aviary breeding facility by adult breeding pairs until they are independent enough for soft-release into Orokonui.K ak a are considered dependent on adults for food until roughly 5 months of age (Moorhouse & Greene, 1995).All but one k ak a had been in the sanctuary for a minimum of 5 months, with a range of time within the sanctuary of up to 10 years.The remaining k ak a was a 10-year-old male who was released (with his breeding partner) into Orokonui directly from a captive-breeding facility, and the environment was therefore novel to this individual.The GPS devices recorded location data, the number of satellites, and horizontal dilution of precision (HDOP) on the 2-or 3-h schedule listed above, as well as temperature and overall dynamic body acceleration (ODBA) on a separate schedule at 1-min intervals for all devices.All GPS locational data was projected to New Zealand Transverse Mercator (NZTM2000-EPSG:2193) before analysis.Before and during the study period (from late 2019 to early 2021) k ak a mortalities of tracked and non-tracked individuals were recorded during surveys by managers of the sanctuary, by mortality signals from the GPS devices in this study, or from VHF-only devices deployed separate to this study.K ak a were sent to Massey University, New Zealand, for post-mortem to determine cause of death, and we recorded the approximate time and location of the mortality where possible.

| Location filtering
To remove unrealistic locations that are unlikely to represent the true location but are instead due to GPS-related error, we used a filtering method which is based on the maximum observed speed of each individual (Shimada et al., 2012(Shimada et al., , 2016)).We calculated the maximum observed speed (Vmax) between successive locations for each individual, based on the locations from six or more satellites (average of 51.4% of locations were from ≥6 satellites per individual).As 95% of SWIFT-fix GPS locations are considered to be within 19.0 m from 6 satellites (Forrest et al., 2022), we expected these observed maximum speeds to reflect the maximum speed dictated by the particular individual's behavior, rather than due to measurement error.The individual V max values were then applied to all locations for that individual, and locations that had a speed exceeding the V max before and after the location were removed from further analyses.The location filtering analysis was conducted using the R package "SDLfilter" (Shimada et al., 2012).
The area that surrounds the Orokonui Ecosanctuary fence (shown as a red line) is a fragmented landscape comprised of oldgrowth and regenerating native forest fragments, which are adjacent to the sanctuary to the east and south, as well as a large area to the west.There are also smaller fragments of exotic conifers, as well as working farms for grazing livestock and semi-urban areas.Presumed threats in the forested areas outside the sanctuary include invasive mammalian predators such as stoats (Mustela erminea) as well as traps and toxins intended for pest species.In the human-modified areas threats also include infrastructure such as roads and powerlines, as well as suburban threats such as unhealthy sources of food, anthropogenic toxins (Sriram et al., 2018) and pathogens such as toxoplasmosis.The location of the study area within New Zealand is shown to the bottom left.

| Home range behavior and presumed risk exposure
To estimate each individual's home range, we used the weighted autocorrelated kernel density estimator (wAKDE) from the "ctmm" package (Calabrese et al., 2016;Fleming et al., 2015;Silva et al., 2022).We used the perturbative Hybrid REML (pHREML) approach to estimate the best fitting movement process, although the results with maximum likelihood (ML) were very similar.The selected movement models were a combination of Ornstein-Uhlenbeck (OU) and Ornstein-Uhlenbeck Foraging (OUF) models, which are both appropriate for estimating range distributions (Alston et al., 2022;Calabrese et al., 2016).For the risk exposure analyses we used the entire probability density of the range distribution (denoted as home range [HR]), as the HR quantifies the space that is expected to be used by individual over the longer term (Alston et al., 2022;Fleming et al., 2015).We can therefore sum cells of the discretized probability density to approximate the time the individual is expected to spend in a specified area.With this in mind, we calculated a proxy of presumed risk exposure by calculating the proportion of the HR that was outside the Orokonui Ecosanctuary fence, as the majority of threats to k ak a exist only outside of the fence.Presumed threats in the forested areas outside the sanctuary include invasive mammalian predators such as stoats (Mustela erminea) as well as traps and toxins intended for pest species.In the human-modified areas threats also include infrastructure such as roads and powerlines, as well as suburban threats such as unhealthy sources of food, anthropogenic toxins (Sriram et al., 2018) and pathogens such as toxoplasmosis.

| Comparison of home range area between individuals
To compare home range sizes between individuals and groups, we considered the area contained within the 50% contour (HR 50 ) to represent areas of core use, and within the 95% contour (HR 95 ) as the home range (Börger et al., 2008;Kie et al., 2010;Kranstauber et al., 2012;Stark et al., 2017).To identify differences between individual's space use, we assessed the influence of age, sex, and origin (Orokonui-fledged or captive-fledged) on home range area using a model-selection approach.We proposed candidate models including age, sex, or origin as the covariates for generalized linear models (GLM) with a Gamma distribution and logarithmic link function to account for the non-negative, continuous, and rightskewed data (Barto n, 2020; Burnham et al., 2011).Due to the low sample size of 10 individuals, we did not consider multiple covariates in a single model.Models were ranked by the corrected Akaike information criterion (AICc) due to the small sample size (Johnson & Omland, 2004), and model fit was assessed using a likelihood-ratio based on pseudo-R-squared for each model (Barto n, 2020) (Table 1).Dispersion (φ), Pearson residuals, and quantiles were checked for model fit (Dunn & Smyth, 2014).

| Dynamic space use
To assess how individual space use changed throughout time we employed a "sweeping window" framework.To achieve this, we estimated ODs using GPS locations from a temporal window of fixed width (e.g., a month of locations), which represents a snapshot of the animal's space use.This snapshot is moved across the full study period incrementally (e.g., by a day), resulting in temporally overlapping ODs (Figure 2).We refer to the approach of temporally overlapping ODs as dynamic space use, as it allowed us to identify when and how space use changed with respect to measures such as OD area, positional shifts of the OD centroid, or OD overlap across temporal snapshots.ODs are more appropriate than range distributions here, as ODs quantify the uncertainty within the animal's movement path, thereby estimating the probability of finding an animal in a particular location during a specified time period, rather than estimating long-term home range (Alston et al., 2022;Fleming et al., 2015Fleming et al., , 2016)).The sweeping window for space use is similar to the approach of Schlägel et al. (2019), who estimated interactions between individual animals.However, we focused on the dynamics of spatial behavior more broadly, and calculated summary statistics such as the area, centroid, and overlap of each snapshot OD, which we tracked throughout time and compared between individuals.Interpretation of the area of ODs should be made carefully however, as occurrence estimators quantify uncertainty only in the movement path and can be sensitive to the sampling interval and idiosyncrasies of an animal's movement behavior (Alston et al., 2022).A similar and useful approach outlining space use patterns for discrete, non-overlapping time periods is presented in Kranstauber et al. (2020).
Within the dynamic space use sweeping window framework there are two parameters, one that determines the number of locations in each temporal snapshot, which we call the window, and another we call the increment, which is the number of locations by which the window is moved along the track (Figure 2).We used a window containing 240 locations of the movement trajectory, which is $1 month at 3-h fix intervals, which was swept along at an increment of 6 locations, which is roughly 18 h.The recorded timestamp of each OD was midway between the first and final locations.The window and increment widths can be freely chosen and should depend on the frequency of telemetry locations and the questions being asked.Shorter window widths will capture changes in space use that are shorter in duration, but for identifying persistent changes in the pattern of space use the animal should ideally visit all its home range during the window's timespan.The only downside we see to a smaller increment is computation time, which should be as short as practicable.In this study, the fix-success rate (proportion of attempted GPS fixes that were successful) averaged 78.1% for all individuals, and therefore the actual window width will be wider than 1 month, and the increment may be >18 h.We kept the number of locations consistent rather than the period to ensure that space use was estimated for a consistent number of locations.A consistent period can also be used.In this case, we used dynamic Brownian bridge movement models (dBBMM) (Horne et al., 2007;Kranstauber et al., 2012) to estimate ODs separately for each temporal snapshot, although other occurrence estimators can also be used.The dBBMM method also requires the user to input a window and a margin size to calculate the likelihood, for which we used 15 and 3 locations, respectively.Although the window and increment parameters of the dynamic space use framework are similar to the window and margin parameters of the dBBMM, they are separate, and both need to be specified by the user.
To assess the pattern of space use changes throughout the tracking period we constructed a similarity matrix of Bhattacharyya's Affinity (BA), which represents the similarity between all the snapshot ODs for each individual (Fieberg & Kochanny, 2005;Kranstauber et al., 2020).This approach can be used to identify space-use patterns, such as in Kranstauber et al. (2020).We opted for BA as it considers the probability density of the OD (which was discretized into cells), it is computationally efficient, and in this case all snapshot ODs within each individual at least partly overlapped.If each individual's snapshot ODs did not overlap then earth mover's distance may be more F I G U R E 2 Concept plot of the dynamic space use method, based on data from this article (k ak a ID 08).Occurrence distributions (ODs) were estimated for temporally overlapping snapshots of a fixed window width of 240 locations, which was swept along the movement track at a fixed increment of 6 locations.For this individual it resulted in 169 temporally overlapping snapshots of space use.The k ak a's movement trajectory is represented by step lengths (black line-left y-axis), and the resulting OD 95 areas that were calculated for each temporal snapshot are shown by the blue line (right y-axis).Interpretation of the area of ODs should be taken carefully however, as occurrence estimators only quantify uncertainty in the movement path, and can be sensitive to the sampling interval and idiosyncrasies of the animal's movement behavior (Alston et al., 2022).Several example ODs (all on the same scale) are shown to illustrate the gradual changes of space use.In this example, the large increases in OD 95 were due to the inclusion of a new core area to the north in the central and right ODs.This is also reflected in the step lengths for the latter part of the tracking period, with a higher number of longer steps as this individual moved between the two core areas.
appropriate as it considers distance between nonoverlapping ODs (Kranstauber et al., 2016).To quantify expansion and contraction of space use for each individual throughout time we calculated the OD 95 area for each temporal snapshot, and to assess positional shifts in space use we calculated the location of the OD 95 centroid.To compare positional changes of space use between individuals and groups, we calculated the distance between successive centroid locations and took the overall average for each individual, which we termed space use drift.The centroid for multimodal ODs was taken to be the overall OD 95 centroid rather than the centroid of the largest OD 95 area, which considers the inclusion of new areas of space use but also prevents artificial "jumps" of the centroid if there were two modes of the OD that were similar in size but fluctuated as the largest in area.To compare space use drift between k ak a statistically, we fitted a generalized linear model with a Gamma distribution and logarithmic link function and conducted the same model diagnostics as above.

| Data collection, data filtering, and mortalities
The GPS unit data collection ranged from 111 to 168 days (mean ± SD = 147 ± 17 days), gathering between 725 and 1727 successful fixes for each individual (mean ± SD = 1087 ± 257 successful fixes).GPS fix-success rate of varied from 63.6% to 90.2% (mean ± SD = 78.1% ± 8.7%), resulting in a total of 10,868 successful location fixes for all individuals.The V max from 6 or more satellites ranged from 0.247 to 1.937 km/h between individuals (mean ± SD = 0.948 ± 0.58 km/h), resulting in between 0.23% to 3.29% (mean ± SD = 1.13% ± 0.93%) of locations being removed for each individual (Table S1).During an 18-month period before and including the tracking period, seven k ak a were recorded to have died.Two deaths occurred before tracking and five deaths occurred during the study period-of these five, two were individuals that were not part of this study but were being radio-tracked with VHF-only devices, and one was an individual that was part of this study.The GPS-tracked individual died within a month of deployment, so the data was excluded from the analyses in this article due to the short tracking period, and the GPS device was then redeployed on another individual, and the data for that individual is included in this study.The four mortalities of birds that were not being VHF-or GPS-tracked were discovered incidentally.Six of the k ak a that died were of known age and were all 3-years or younger with an average age of 1.8 years.Confirmed causes of mortality included brodifacoum poisoning and toxoplasmosis, and suspected but unconfirmed causes included electrocution on powerlines (Table S2).No mortalities or adverse effects were attributed to the GPS or other telemetry devices or the attachment harnesses.

| Static space use: Home range area and comparison between individuals
Individual home range area (HR 95 area) varied widely, from 0.34 to 9.92 km 2 , with a mean ± SD = 4.12 ± 3.83 km 2 (n = 10 individuals) (Figure 3 and Table S1).The home ranges of all k ak a were at least partly within the reserve, and three almost entirely, with the proportion of each k ak a's HR inside the Orokonui fence ranging from 0.11 to 0.98 (mean ± SD proportion of HR inside Orokonui = 0.59 ± 0.35) (Figure 3).There was no significant trend of HR 95 area in relation to sex or origin (Table 1 and Figure S4) but there was a significant negative correlation between HR 95 area and age (P = .001,R 2 = .73-Figure3c).The average home range size for k ak a 3 years and younger was 6.14 km 2 (n = 6), whereas the average home range size for k ak a 5 years and older was 1.09 km 2 (n = 4) (Figure 3b).As with home range area, the proportion of each individual's HR that was outside of the Orokonui Ecosanctuary fence also decreased significantly with age (P < .001,R 2 = .82-Figure3d).

| Dynamic space use
Space use was most variable for the two juvenile k ak a (1-year-old), with Bhattacharyya's Affinity (BA) reaching as low as 0.43 and 0.38, whereas the minimum BA of all remaining individuals was 0.76 (Figures 4 and S5).The large dissimilarities in BA for the juvenile k ak a were predominately due to positional changes of the OD 95 centroid rather than changes in OD 95 area.The area of every k ak a's OD 95 varied throughout the study period, although the risk exposure through time remained largely constant for most individuals (Figures S6 and S7).There was also some synchrony as to when OD 95 area varied within the male and female groups, with females expanding their OD 95 areas in September before contracting after October, and males expanding their OD 95 areas after October (Figure S6).
The centroids of the OD 95 drifted farthest on average for the two juvenile k ak a, with average drift distances ranging from 32.3 to 4.87 m between temporally overlapping ODs (F 1,8 = 38.52,P < .001,pseudo-R 2 = .81-Fig-ures 5 and S8).There was no discernible population-level trend of space-use drift that might have related to seasonality (Figure 5b).

| DISCUSSION
In this study, we used HR and dynamic space use analyses using temporally overlapping ODs to assess risk exposure in a reintroduced k ak a (Nestor meridionalis) population.Summarily, the HRs of younger individuals (<3 years) were much larger than that of older individuals, and the space use of juvenile k ak a (1-year-old) was more temporally variable, with significant shifts of the OD 95 centroid.We therefore infer that the larger home ranges and more exploratory behavior of younger individuals put these cohorts at higher risk than older individuals due to more time spent outside the reserve.These findings correlate with higher incidental mortality observations of younger individuals, of which the average known age was 1.8 years.All k ak a included the supplementary feeding stations within the core area of their HR, which suggests there may be a reliance on supplementary feeding stations that provide an anchoring effect, which may help to increase the safety of k ak a by reducing their time spent in higher risk areas.Additionally, as we caught individuals within the sanctuary, it is possible there are k ak a that have home ranges exclusively outside the sanctuary that were inaccessible for GPS device attachment.
The age-related differences in home range area and dynamics may be due to several factors.Namely, (1) younger k ak a may be more exploratory; (2) there may be intraspecific conflict between age groups leading to displacement; (3) there may be a selection pressure against large home ranges; and (4) familiarity rather than age may explain the trend.For (1), as shown in previous studies of wild k ak a and other bird species, older individuals are typically more neophobic and more efficient foragers, which can reduce the need for large home ranges relative to younger individuals (Bond & Diamond, 2004;Clay et al., 2018;Loepelt et al., 2016;Sherratt & Morand-Ferron, 2018).This trend may also be compounded by the constant availability of supplementary food from feeding stations, reducing a need to forage widely.The more dynamic space use in juvenile k ak a predominately arose from positional changes in the OD 95 centroid, which was due to the initial visitation of novel areas that were thereafter visited regularly, which suggests exploratory behavior followed by exploitation (Berger-Tal et al., 2014).This pattern may also indicate information gathering that may precede dispersal from their natal area (Andreassen et al., 2002;Ronce, 2007), and would provide valuable evidence of home range establishment, although this is difficult to test empirically (but see Maor-Cohen et al. (2021)).For (2) it is also possible there is agonistic behavior (intraspecific conflict), where older individuals are more dominant in areas with supplementary feeding stations, which displaces the younger individuals to the periphery, necessitating exploration of new areas.However, we did not directly observe agonistic behavior, all individuals still included the feeding stations in their core home ranges, and previous work has suggested that juvenile k ak a at Orokonui visit the feeding stations regularly (Aichele et al., 2021).This evidence suggests that there is no strong displacement of younger k ak a by older k ak a, at least not at feeding stations.For (3) another possibility is that individuals that have survived to adulthood have always had small home ranges, suggesting a negative selection pressure against large home ranges.Definitively addressing this question would require tracking for longer periods to assess whether home ranges begin to contract in younger individuals.From our data, younger individuals had consistently larger home ranges, which declined uniformly, rather than younger individuals having both small and large home ranges, which would undergo a filtering process against large home ranges as age increased (Class et al., 2019).For (4) we cannot definitively separate age and familiarity, which are inextricably linked but have relevance for conservation translocations, as ages may differ despite the environment being novel to all.Therefore, it is possible in our case that older individuals have smaller home ranges because they are more familiar with the area, and this is likely to be at least partly the case.However, the smallest home range in our sample was of a 10-year-old k ak a that was released into the sanctuary at the time of tracking, so the area was completely new to this individual.It should be noted though that this individual was raised in captivity, which might have led to a reliance on supplementary food, so results may differ for wild adults translocated into the sanctuary.Summarily, as younger k ak a still access supplementary food, the trend of larger and more dynamic space use in younger individuals appears to be predominately explained by exploratory behavior, which is likely linked to familiarity.The exploratory behavior is possibly preceding long-term home-range establishment, which would be indicated by a contraction and stability of space use.If that is the case, it is unknown whether the new, smaller home ranges would be inside or outside of the sanctuary, as there were areas of core use (HR 50 ) for several younger k ak a outside the sanctuary.
Despite clear age-related differences in home range area and patterns of dynamic space use, we acknowledge the sample size of 10 individuals from a single study site will not represent the full breadth of behavioral variability present in this or in other populations, and tracking more individuals may reveal more conclusive or even contradictory evidence.Given our timeline and funding, we prioritized fewer devices with a relatively high frequency of locations (i.e., GPS) to understand fine-scale space use, rather than solely VHF radiotracking devices that are less expensive and last longer, but do not have the consistency or frequency of locations to answer the questions we have addressed here.The duration of tracking also did not cover a full year and we may have missed seasonal variation outside of the tracking period.However, the coverage of sex and age groups provided valuable and convincing evidence toward age-related differences in spatial behavior, and the relatively finescale tracking data was important to provide sufficient resolution to estimate dynamic space use.
Although at least one other method has been developed to segment an animal's trajectory with a sweeping window framework (Schlägel et al., 2019), how an animal's space use changes gradually throughout time and whether this aligns with certain groups has otherwise been given little attention in the literature.We hope to address this by providing an easily applicable and intuitive dynamic space use estimation framework, and we highlight that different statistical space use estimators can be substituted for the dBBMM occurrence estimator, such as the generalized time-series Kriging approach which can accommodate multiple movement models (Fleming et al., 2016).Estimating both home ranges using a range estimator and dynamic space use using an occurrence estimator allowed us to improve our understanding of k ak a spatial behavior by assessing total space used and how that changed over time.Tracking the area, centroid, and overlap of the ODs allows for space use patterns to be identified as the ODs expand, contract, and shift in position, which can allow a multitude of questions about site fidelity, space-use patterns, dispersal, and exploration to be posed (Börger et al., 2008;Kranstauber et al., 2016Kranstauber et al., , 2020)).
Differences in spatial behavior could lead to increased risk for some taxa in conservation reserves, particularly when threats remain in the wider landscape.In this study, the presumed exposure to risk did not vary substantially within groups throughout time, but this may not be the case if the k ak a were tracked for longer, or for other taxa that have more predictable seasonal behavioral changes.Species that increase their area of space use or decrease their site fidelity to develop sufficient condition for breeding or hibernation, or to take advantage of ephemeral resources, could be exposed to greater risk as they venture beyond safe regions.Understanding which groups or individuals of a species are at the highest risk may assist management decision-making, particularly when planning translocations.

| MANAGEMENT IMPLICATIONS
Management of birds and other species that exhibit wideranging behavior in conservation reserves should consider differences between groups of individuals that may lead to differential exposure to risk beyond reserve boundaries.This will also be relevant to the planning of translocations, as individuals are likely to differ in their spatial behaviors (Stuber et al., 2022), which may relate to age.In that case it may be appropriate to translocate individuals from a range of ages if possible, although there are many factors that contribute to successful translocations, and these will vary by species (Miskelly & Powlesland, 2013;Spatz et al., 2023).It should be considered that animals of varying ages could respond differently to novel environments, and younger individuals may be more likely to explore.In fragmented landscapes that contain atypical features such as urban areas, infrastructure and exotic predatory species, reintroduced animal species will need to explore and adapt to survive, which may lead to innovations and a greater behavioral repertoire, improving their acquisition of resources and avoidance of threats.This study additionally highlights the key importance of appropriate management of threats outside conservation reserves, particularly for wide-ranging flighted species.For other risk-exposed taxa, we recommend quantifying both home ranges and dynamic space use to assess and understand the area requirements and space use patterns of individuals, and to determine when and where animals may be exposed to risk.

F
I G U R E 3 (a,b) 95% home range distributions (HR 95 ) for each individual estimated from GPS tracking and weighted autocorrelated kernel density estimation (wAKDE).Part figure (a) shows the HR 95 isopleths for individuals 3 years or younger (n = 6), and (b) shows the HR 95 isopleths of individuals 5 years or older (n = 4).The fence of the Orokonui Ecosanctuary is shown as the thicker black line in the center of each plot.(c) Home range area (HR 95 area) as a function of age for 10 k ak a GPS tracked at Orokonui Ecosanctuary, New Zealand.The solid line indicates the fit of a Generalized linear model using a Gamma distribution with log link with age as a predictor.The ribbon is the 95% confidence interval.Pseudo-R 2 based on likelihood ratio = .73.(d) Proportion of each individual k ak a's home range (HR) outside of the Orokonui Ecosanctuary fence, as a function of k ak a age.Generalized linear model using Gamma distribution with logarithmic link was fit with age as a predictor, ribbon is the 95% confidence interval.Pseudo-R 2 based on likelihood ratio = .82.The dashed line indicates when all of the HR would be outside of the sanctuary.F I G U R E 4 The left column represents the dynamic space use of an individual female k ak a of age 1 year, and the right column represents the dynamic space use of a 10-year-old male k ak a.The upper panels show the similarity matrices of Bhattacharyya's Affinity (BA) throughout the study period.The middle panels show the percentage difference in OD 95 area compared with each individual's average OD 95 area for the full study period.The lower panels show the drift of OD 95 centroid on the same scale, with the color relating to the time of year (purple to yellow), with the fence of Orokonui Ecosanctuary represented as the black outline.The lower panels also show the initial (purple with dashed line) and final (yellow with solid line) ODs for each individual.The scale bar is in 200 m increments and is 1000 m in total.

F
I G U R E 5 (a) Position of the UD 95 centroid for each individual k ak a throughout the study period, in relation to age (color-legend to the right).The fence of Orokonui Ecosanctuary is shown in black.The scale bar is in 200 m increments and is 1000 m in total.(b) Distance between UD 95 centroids from one space use snapshot to the next.There appear to be no discernible temporal trends in space use drift.