The influence of feeding station location on the space use and behavior of reintroduced 'alalā: Causes and consequences

Supplemental feeding is a common soft‐release strategy for increasing settlement, survival, and breeding in animals after translocation. However, supplemental feeding can also hinder natural patterns of dispersal or influence social interactions. Some drawbacks of feeding can be mitigated if feeding stations can function as management tools for directing animals' movement, for example, by guiding them toward wild food resources or prospective territories while directing them away from potential threats, for example, predator hotspots or urban/agricultural centers. However, these effects may not work equally across individuals or habitat types. For species such as 'alalā (Corvus hawaiiensis), Hawaii's last living corvid, being able to manipulate space use via supplemental feeders could improve reintroduction outcomes. We determined if and how feeding stations influence a released population of 'alalā by strategically moving feeders across varying distances and habitat types, while measuring how quickly birds discovered new feeders and how their space use changed. We found that 'alalā discovered feeders more rapidly in closed as opposed to open canopy habitat, and feeder movement influenced how far 'alalā ranged, especially in afternoon periods. Sex, social network position, and individual home range size did not predict feeder discovery. These insights offer lessons for using supplemental feeding when managing reintroductions.

in wild and recently reintroduced populations.However, supplemental feeding can also be potentially problematic for a number of reasons, each with varying levels of empirical support.Maintaining feeders can be financially costly; they can serve as disease vectors (Sorensen et al., 2014), they could skew diet toward nonnative sources, with potential impacts on the microbiome, or they may be ineffective if animals do not find or use provided food (Bannister et al., 2020).It has also been suggested that feeders may make animals more vulnerable to predation because they can influence antipredator behavior (e.g., vigilance; Lee et al., 2021), although there are few examples where increased predation risk has been confirmed (see discussion in Robb et al., 2008).Additionally, there is the potential that the presence of supplemental food interferes with spatial exploration and learning opportunities at the release site (Mertes et al., 2019).Translocation practitioners must therefore balance the pros and cons of supplemental feeding to determine where and when this management intervention should be employed.Exploring how the potential negative effects may be mitigated or offset could increase the benefits of supplemental feeding.
A little-explored application of supplemental feeding uses feeders as a tool for directing translocated or reintroduced animals toward or away from certain areas.For example, supplemental feeders could be used to guide animals toward wild food resources, prospective breeding territories, or conspecifics while directing them away from potential sources of threats such as predator hotspots, pollution sources (e.g., Stack et al., 2022), or urban or agricultural centers.Being able to direct movement could also artificially stimulate home range expansion (e.g., Wallace & Temple, 1987) or encourage dispersal away from the release site to make room for future release cohorts and prevent aggression from older, more competitive individuals.However, the effect of encouraging dispersal via means such as moving supplemental food stations has not been studied widely or experimentally.In fact, the majority of evidence is limited to studies on avian scavengers trialing or proposing release and feeding strategies (e.g., Stack et al., 2022;Wallace & Temple, 1987).Additionally, as a tool, moving feeders may not influence the space use of all individuals equally (Wallace & Temple, 1987) or work equally well across fragmented habitats, both of which would influence their effectiveness.
We studied the potential for influencing the movement of released animals with supplemental food by studying 'alal a, a territorial, critically endangered corvid species native to Hawai'i Island.'Alal a are an omnivorous species that act as the sole remaining seed disperser for a number of native Hawaiian forest plants (Banko et al., 2002;Culliney et al., 2012).Their only native predator is the 'io (Hawaiian hawk, Buteo solitarius), and it has been assumed that 'alal a are more vulnerable to their attacks in open habitat or forest with an open understory; however, their relationship with 'io historically is unclear (Banko et al., 2002;Greggor et al., 2021).After decades of decline and failed population reinforcement efforts in the 1990s, 'alal a went extinct in the wild in 2002.Conservation breeding, and reintroduction efforts continue to this day.However, recent releases have faced challenges such as 'io predation, poor body condition, and managing conspecific conflict during territorial establishment.As part of these efforts, being able to manipulate 'alal a space use via supplemental food locations would allow for a noninvasive post-release management method that could address some of these issues.
Although newly released 'alal a appear to increase their spatial exploration over time, even when supplemental food resources are static (Smetzer et al., 2021), there could still be considerable benefit to directing farther movement and encouraging dispersal at key periods in their social development.To realize this benefit, we need to know to what extent the feeders influence 'alal a space use and whether they can be used as a tool for directing movement.Additionally, by moving food into varying habitat types, we can also determine what constitutes a preferred feeder location (e.g., open canopy versus closed canopy, etc.) within a fragmented landscape that may differ in predation risk.Finally, by looking at individual-level predictors of feeder discovery and use, we can determine whether future management strategies may differ in their effectiveness across individuals.
Based on our observations of feeder use and the natural history of the species, we made several predictions about how 'alal a would respond when we moved half of the feeders away from the release site where they were anchored: (A) 'alal a will consume supplemental food from new locations but may be slower to find feeders moved farther away since their search strategy may be based on proximity to prior feeder locations; (B) birds will spend less time near the release site when half of the feeders are located farther away, since birds may be encouraged to visit and explore new areas of the forest when they go to find supplemental food; (C) there are predictive characteristics of individual birds that discover and use newly located food opportunities (e.g., faster to find food if more centrally located in social network or have a larger home range); (D) there are characteristics of the supplemental food locations that make their discovery and use more likely (e.g., feeders under cover are safer and therefore more likely to be used).
All birds were hatched and reared at breeding facilities on the islands of Hawai'i (Keauhou Bird Conservation Center) and Maui (Maui Bird Conservation Center) (see details in Greggor et al., 2018).In September and October of 2017, a total of 11 juvenile 'alal a were released into the Pu'u Maka'ala Natural Area Reserve (Table S1).All survived through the study period: July 16-Sept 9, 2018.At that time, all were still classified as juveniles (<3 years of age) and had not begun to demonstrate signs of territoriality.All birds were fitted with a VHF transmitter with a pulse rate of either 30 or 60 ppm (either a Holohill or JDJC unit), plastic color leg bands for visual identification, and a 125 khz RFID chip, which was embedded in one of the leg bands.Transmitters were attached as backpacks using spectra harness material (Bally Ribbon Mills) and small copper crimps.We used an R1000 telemetry receiver (Communication Specialists Inc.) and a three-element yagi antenna (Sirtrack, New Zealand) for tracking.
We provided fresh supplemental food daily for $8.5 h (morning feeding: 07:00-08:30; afternoon food removal: 15:00-16:30) in plastic feeders that 'alal a were trained to open prior to release.The food consisted of the same diet items and caloric value per bird as had been provided at the conservation breeding centers.Staff dressed in black costumes with face coverings when provisioning food, to avoid the birds associating humans with food.To avoid biasing birds' use of feeders based on provisioning order, we alternated each day between distributing experimental or static feeders first.When food was collected at the end of the day, we estimated the percent of food remaining per feeder, and recorded the total weight of food remaining.

| Data collection
We collected several types of data to help determine how feeder movements influenced birds' space use and whether there were differences in how and where birds responded to the new provisioning locations.We used both trail cameras and RFID readers at the feeders to determine visitation; we tracked the birds on foot to determine behavior; and we also conducted systematic VHF signal checks to gauge larger movement patterns.
Each feeder stand was outfitted with an RFID reader and motion-activated trail camera (Bushnell 14MP HD Trophy Cam) to determine when individual birds discovered new feeder locations.The RFID units were built with Eccel Technologies boards and antennas, housed in custom, 3-D-printed waterproof casing.Readers were set to the highest sensitivity (capturing visits <0.3 s long) and to collect data at all hours of the day.The RFID antenna was placed on the perch that birds used to access the feeder to capture when birds fed.Meanwhile, the trail camera also captured when birds spent time anywhere on the feeder stand, which they landed on before accessing the RFID perch (Figure 1).These two methods of data collection allowed for redundancy in feeding and arrival metrics in case either method failed due to weather or other factors.When designing the embedded perch, we conducted both in-person and trail camera pilot tests to confirm that the reader was reliably triggered by birds when they landed to open the feeder and that this movement would be picked up by the trail cameras.Multiple feeders also contained a spring scale, which allowed for the regular collection of bird weights via trail camera videos.
A release tracking team of 2-4 people monitored the birds daily throughout the study period.Staff used iPhone 6S units equipped with the Survey123 app (ESRI) for recording data.Staff recorded a survey and GPS point whenever they encountered and identified a bird in the forest or at a feeder.When multiple birds were present, a social observation survey was also taken, which noted the occurrence of multiple types of social behaviors, the individuals involved, and their roles.The behaviors recorded were adapted from previous studies that measured sociality and dominance across several corvid species (Greggor et al., 2016;Logan et al., 2013) (Table S2).Every bird was observed in person at least once every day, with the exception of a single day when a hurricane made observations impossible.Given the permitting requirements to observe each individual daily, we prioritized monitoring first at the feeders when food was provided in the morning, then tracking the birds using VHF signals to visually locate them in the forest.Given the density of the forest, visual relocations were within close F I G U R E 1 Screen shot of trail camera video on the supplemental feeding station, showing RFID antenna, feeding platform, and plastic feeder.
enough proximity of the birds to confirm their color band combination and take a reliable GPS point.Afternoon observations aimed to find birds away from the feeder locations, wherever that may be.
Finally, we also conducted routine VHF signal checks from a set location that had reliable signal reception.At three pre-determined times of day (06:45, 11:00, and 16:00), we recorded whether a signal was detected, the signal strength at full gain, and the bearing of the strongest signal for each individual, providing distance estimates before feeding, in the middle of the day, and after food had been removed.As the detection range of these transmitters is $750-800 m (determined by triangulation and observations from multiple observers), undetected birds were likely to be >750-800 m away, which was beyond the farthest feeder location.These distance estimates are subject to some error due to variations in topography, vegetation, and weather conditions, which could add noise to the data.However, these variables would not have varied in line with our hypotheses since our study design compared data across multiple weeks of controls and movement conditions.In addition, visual confirmation was used to validate a subset of these location estimates.

| Bird traits
We calculated social network metrics using the social observation data taken in the season prior to the movement of feeders (March 1-June 30, 2018).A weighted network was derived from all occurrences of undirected co-feeding and co-sitting behaviors using the asnipe package in R (Farine, 2013;R Core Team, 2021), capturing birds' propensity to forage and closely affiliate (Figure S1).Constructing the network from these behaviors is preferred over a network of simple spatial associations since they are indicative of higher social tolerance and attraction than might otherwise be present at communally attractive resources, such as food feeders.From this network, we calculated betweenness scores for each individual.Betweenness describes the extent to which an individual (node) connects subgroups and is often used as a social network metric for capturing the capacity for information flow (Sosa et al., 2021).
We calculated home range metrics for the time period (breeding season 2018) leading up to the experimental manipulation of feeders using an estimator that was robust to the coarse scale and irregular sampling profile of our tracking dataset.The difficulty of resighting each 'alal a using VHF location fixes likely ensured that the time interval between successive observations of each 'alal a (1-2 locations/day) was sufficiently long for most autocorrelation among observations to have decayed.However, autocorrelation is often still present even in VHF datasets and should not be overlooked (Harris et al., 1990;Silva et al., 2022).The small absolute sample size (n) of each 'alal a location dataset could also introduce bias into home range estimates.Therefore, we fitted variograms and continuous-time movement models (CTMMs) to VHF locations acquired from each reintroduced 'alal a using the ctmm package (Calabrese et al., 2016).This continuous-time stochastic movement framework: (1) properly accounts for the serial autocorrelation intrinsic to movement data; (2) explicitly accounts for telemetry error; (3) is robust to small sample sizes; (4) can handle irregularities and gaps in location datasets; and (5) appropriately estimates confidence intervals (for details on this method, see Calabrese et al., 2016;Fleming & Calabrese, 2017;Papageorgiou et al., 2021;Silva et al., 2022).Detailed methods and results of our statistical analysis of the spatial data used for calculating home ranges can be found in Supplementary Materials.

| Feeder movement
From the date of release, 'alal a had access to four feeders within 20 m of the release aviary.We began experimental manipulations of feeders approximately 9 months following release, when birds had formed home ranges near the release aviary but had not yet shown signs of territoriality.For 1 week at the onset of the study, the four feeders remained at their longstanding locations, 10-20 m south of the release aviary.Subsequently, we moved two of the four feeders along two perpendicular trajectories simultaneously, each week moving a greater distance than the prior week: 50, 100, and 150 m from the last used location on that route (Figure 2).We chose the two trajectories so that they differed in the amount of forest canopy cover, with the closed route traversing through undisturbed old growth forest and the open route following parallel to an old dirt road, bordered by old growth forest and areas of shorter, younger forest (Figure 2; closed canopy locations, mean = 80.0% ± 14.1% cover; open locations, mean = 40.0%± 11.4%).After 3 weeks of moves, we returned feeders to their original locations, and the process was repeated in the opposite direction.To determine whether the feeder locations differed in habitat characteristics other than canopy cover, we recorded the tree height and species composition of the canopy and midstory (according to the classification scheme of Jacobi, 1989) for the 5 m radius surrounding each feeder location.We also noted the number and species of fruiting plants in the 5 m plot and whether any of them were actively fruiting.Since 'alal a are omnivorous foragers with a wide diversity of fruit in their diet (Culliney et al., 2012), all fruiting plants were considered to be potential food.

| Analysis
All analyses were conducted in R (R Core Team, 2021) and the per-bird sample sizes for each analysis can be found in Table S3.We used separate datasets and a series of Generalized Linear Mixed Models (GLMMs) to investigate two lines of inquiry: (1) did moving feeders influence 'alal a space use, and (2) did characteristics of birds or feeder locations influence initial feeder discovery and their subsequent use?We scaled all numeric response variables with a mean of 0 and a standard deviation of 1 with the standardize package (Eager, 2017) to help facilitate model averaging.We then ran GLMMs using either the lme4 package (Bates et al., 2015) for models with binomial and negative binomial error distributions or the glmmTMB package (Brooks et al., 2017) for models with beta distributions.We used the MUMIn package (Barton, 2020) to conduct model selection based on AICc values.In cases where multiple submodels had a ΔAICc < 4 from the top model, we conducted model averaging (Burnham & Anderson, 2002;Grueber et al., 2011).We assessed model fit with the DHARMa package (Hartig, 2021) by evaluating plots of residual and predicted values for issues of dispersion and uniformity.With averaged models, we calculated parameter estimates and relative importance scores (RI) using the natural average method for all fixed effects and also checked for multicollinearity between fixed effects by calculating variance inflation factors (VIF) with the performance package (Lüdecke et al., 2021).

| 'Alal a space use
To get a baseline understanding of the influence of feeders on movement prior to the start of this study period, we compared home range size calculations from all visual relocations in the season prior to the study period (March 1-June 30, 2018) with a restricted dataset that excluded relocations <50 m from all feeders (see Supplemental Materials for more details).This comparison allowed us to assess the extent to which the feeders at the release site anchored birds to that area prior to feeder movement.
To understand whether moving feeders influenced birds' wider space use, we ran two models from data within the study period.First, we investigated birds' likelihood of being detected at the three daily signal checks (detected, Y = 1, N = 0) with a binomial Generalized Linear Mixed Model (GLMM).For birds that were in range, we then investigated factors that influenced the proximity of the birds to the survey location by analyzing their telemetry signal strength (i.e., number of bars detected on the receiver) with a negative binomial GLMM, which addressed issues of overdispersion in the data.For both models, we examined the influence of time of day (AM, Midday, or PM check), the farthest feeder distance (in 10s of meters from the feeder origin), and their interaction as fixed effects.To avoid pseudoreplication and artificially inflating degrees of freedom, we included Bird ID as a random effect for both models.

| 'Alal a feeder use
In order to determine if other habitat features confounded our open and closed canopy designations, we compared several aspects of the feeder locations along the open and closed canopy trajectories.We ran either two sample t-tests or Mann-Whitney U tests (after assessing normality with Shapiro Wilk tests) to examine differences in the canopy height, number of fruiting plant species (i.e., diversity), the total number of fruiting plants present, and the number of plants currently bearing fruit.Ties were broken in Mann-Whitney U tests by randomly jittering the data by 0.01, re-running the tests 10,000 times, and extracting mean test values.
To understand whether there are certain bird or habitat predictors of feeder use, we then ran two main analyses.First, we investigated predictors of feeder discovery.We analyzed whether the first arrival time of each bird at each feeder was influenced by feeder habitat classifications and bird traits (i.e., sex, betweenness in the social network, and home range size).Birds' arrival time was measured as the delay in minutes from feeder placement to bird visitation to the feeder, excluding time when food was not present.For the few cases (N = 6) where an individual never fed from a particular feeder, they were given the maximum possible time elapsed per study week.We ran a GLMM with a negative binomial error distribution, which included whether the habitat was open or closed, the distance of the feeder location from the baseline location of the feeder (origin), sex, home range size, and betweenness score (as calculated from the social network) as bird covariates.We included Bird ID as a random effect.
Second, we investigated predictors of longer-term feeder use by the whole cohort by examining the percentage of food left over in the feeders at the end of each day.Specifically, we used a GLMM with a beta error distribution and a logit link function (Douma & Weedon, 2019) to examine the influence of distance from the original feeders and the habitat type (open/closed) on how much food remained.We included Feeder ID as a random effect.

| 'Alal a space use
In the season prior to our manipulations, we found that the static feeders likely biased 'alal a space use.Specifically, we found that calculating home range sizes with the entire location dataset that included observations at the feeders led to a smaller group-level mean home range size (13.6 hm 2 , CI 7.4 low 23.0 high) compared to AKDEs calculated using the filtered dataset that excluded locations recorded within 50 m of the feeders (77.6 hm 2 , CI 44.1 low 126.6 high; Figure 3).
During the study period, 'alal a were less likely to be detected from the signal check location in the afternoon in weeks when the feeders were moved farther away, but there was no difference in detection probability in the morning or midday (Binomial GLMM, n = 1665, interaction Midday:Distance B = À0.20 ± 0.19, z = À1.06;PM:Distance, B = À0.54 ± 0.19, z = À2.88;RI = 1.00, Figure 4, Table 1).This same interaction between period of day and feeder movement was less relevant to predicting the signal strength of detected birds (RI = 0.49).Instead, 'alal a were more likely to have weaker signal strength as the day progressed (Negative Binomial GLMM, n = 1241; Midday, B = À0.03 ± 0.01, z = 2.90; PM, À0.05 ± 0.01, z = 4.84, RI = 1.00,Table S4), regardless of week, and to have weaker signals as the feeders were moved farther away (B = À0.06 ± 0.02, z = 3.61, RI = 1.00), regardless of time of day.Together, these suggest that all birds were in the vicinity of the baseline release site feeder location in the mornings, but some birds dispersed beyond signal range by the PM, especially when feeders were farther away.Meanwhile, those that did not disperse and stayed in telemetry range had weaker signals in the PM but did so regardless of feeder location.sample t-test, t = 5.97, df = 6.76, p < .001),reflecting an older forest structure.Both closed and open canopy feeder locations had a similar diversity (t = 2.02, df = 9.64, p-value = .07)and abundance (t = À0.57,df = 7.81, p-value = .59) of fruiting plants nearby (Figure S2).However, at the time of the census, there was a greater number of plants with ripe fruit in open feeder locations (Mann-Whitney U test, W = 2.52, p = .013),suggesting that the habitat types differed in their vegetation cover (possibly related to predation risk) and current fruit availability, but not in their year-round foraging potential.

| 'Alal a feeder use
'Alal a took longer to discover and use feeders that were located in open as opposed to closed canopy habitats (GLMM, N = 220 observations from 11 birds; B = 2.39e-04 ± 9.51e-05, z = 2.65, RI = 1.00; Figure 5), F I G U R E 4 Boxplots depicting the number of birds in telemetry range at each signal check by experimental week.Boxplots illustrate median values (black horizontal line), with 25th-75th percentile intervals (colored box area) and 95% CI (vertical lines).The feeders were all located at the origin site during weeks with black rectangles (baseline prior to manipulation of feeder location), and half of them spread farther out across the landscape in each subsequent week.The majority of birds were consistently present during the AM period (gray), which occurred right before feeding, but showed greater dispersion in the PM time periods (blue), especially as the feeders were moved farther away.
T A B L E 1 Model results for binomial GLMM examining effect of feeder distance on likelihood that birds were in VHF signal range.and those that were moved farther from the last known feeder (B = 2.86e-04 ± 9.51e-05, z = 3.00, RI = 1.00).The averaged model did not retain sex or home range size.Birds' social betweenness score was retained in the averaged model but had a low enough relative importance score (RI = 0.27) that it was deemed inconsequential for arrival times (Table S5).

Model
We did not detect any meaningful influence on longer term feeder use based on the percentage of food left over each day at each feeder.Specifically, there was no effect of the distance of the feeder from the last known feeder or the habitat type in the 5 m surrounding each feeder (i.e., canopy cover) on the amount of food consumed (Table S6).

| DISCUSSION
Post-release movements constitute an important component of translocation biology, and practitioners have tested numerous methods to prevent dispersal following release (Berger-Tal et al., 2020;Le Gouar et al., 2012;Swaisgood & Ruiz-Miranda, 2018).Supplemental food is often provided as part of a larger strategy to anchor released animals at the release site, which was one of the motivating factors for supplemental feeding for 'alal a.Our analyses indicate that 'alal a movement patterns were biased by provisioned food since home ranges were smaller when we included relocations near feeders (Figure 3).Prior to this study, these 'alal a had been spending progressively less time at the static feeders and ranging farther distances since release (Smetzer et al., 2021).Therefore, the extent to which feeders constrained home ranges is unclear.Presumably, had the feeders not been present or had they been spaced farther away, 'alal a would have ranged farther or had different home range centers.Without evidence, we cannot speculate whether greater movement would have been beneficial (e.g., facilitating discovery of new food sources or the social dispersion characteristic of the territorial system of adult birds) or disadvantageous (e.g., exposing them to greater risk of predation, poor foraging success, or anthropogenic landscapes).
There are also scenarios where being able to encourage directed movement by released populations of animals will have useful applications.For instance, influencing movement away from high-risk areas (e.g., anthropogenic threats (Stack et al., 2022) or predator hot-spots) or toward important resources (e.g., food or breeding opportunities) could be an important tool for translocation practitioners to actively manage post-release outcomes.We investigated whether moving supplemental feeders encouraged changes to space use in 'alal a.In doing so, we also explored whether the effectiveness of feeder movement as a tool depended upon the distance feeders moved, the habitat type at new locations, or individual bird traits, which might otherwise undermine the viability of future management strategies.We found that moving feeders influenced 'alal a space use, but only at certain times; placing feeders farther from the release site encouraged farther ranging beyond the feeder locations later in the day.'Alal a were slower to discover (or at least slower to use) feeders located in open habitat than closed habitat, even though these feeders should be easier to detect due to greater visibility in open habitats.No individual bird characteristics predicted feeder discovery.Together, these results suggest that feeders can be used for directing 'alal a movements in certain situations.
'Alal a space use changed as a result of feeder movement, especially in the afternoons.We saw greater afternoon dispersal (i.e., fewer signal detections) when feeders were farther away, in comparison to the control weeks where feeders were close by.Meanwhile, 'alal a were equally likely to be in signal range in the mornings, regardless of where feeders were located.The interaction between morning versus afternoon detections and feeder movement could be due to birds depleting the highquality food items in the nearby feeders first and then seeking the high-quality items in the farther feeders later in the day.Yet, more than the feeder contents alone seemed to drive movement.Undetected birds were likely spending time beyond the farthest feeder locations based on the detection range of the transmitter models we used ($750-800 m).Therefore, moving feeders appeared to have a positive influence on some birds' wider exploration of the habitat.Exploration after release into the wild can have many benefits, including facilitating learning to develop a cognitive map of the distribution of resources and making resource exploitation more efficient (Berger-Tal & Saltz, 2014;Lewis et al., 2021).Moving feeders on the landscape may therefore help tap into some of these benefits.
Despite the potential for exploration, it appears that not all individuals were equally influenced by feeder movement.A subset of birds stayed within the VHF detection range throughout the day.Those that stayed were more likely to have weaker signals as feeders were moved farther away, indicating that their local movements were still influenced by feeder location.However, for these individuals, more distant feeders did not appear to trigger long-range afternoon movements.Together, these space use results suggest there may have been two strategies within the cohort: (1) birds for which feeder movement encouraged wider range exploration in the afternoons; (2) birds that remained closer to feeders and modestly expanded their movements when feeders were moved farther away.In both cases, moving feeders increased the likelihood that birds would spend time away from the original cluster of feeders, but the effect of feeder movement on exploration was more pronounced for the group that left the area when feeders were moved.These differences between birds were not likely due to the onset of territoriality, since all birds were still younger than reproductive age and had not yet formed breeding pairs.
Feeder location also influenced how readily 'alal a used them.'Alal a were slower to discover (or at least slower to use) feeders located in open habitat, even though they would be more visible during overhead flights.These findings suggest either an active aversion to feeders in open landscapes, a preference for closed habitat feeders, or a more passive bias toward encountering feeders in areas of forest with more cover where they spend more time.This pattern was not due to fruit availability surrounding feeder locations, because the open habitats actually had more, not less, native fruit available during the time of the study.However, fruit is only one source of wild food resources for 'alal a, since they historically spent considerable time foraging on insects (Sakai et al., 1986).Because different abundances of insects may be present across the canopy types, it is possible 'alal a were drawn to the closed canopy areas in search of insect foraging opportunities.Alternatively, their delayed feeder use may reflect the avoidance of potentially risk-prone habitat as part of an evolved anti-predator strategy.Many prey species avoid landscape features associated with increased predation risk (i.e., they navigate a landscape of fear; Laundre et al., 2010), for instance, avoiding fragmented habitat to reduce their risk from aerial predators (e.g., Dinkins et al., 2014).Evidence across many species indicates that foraging behavior can compromise detection of predators and increase vulnerability, and that animals have evolved decision-making strategies to balance the risks and benefits of foraging in different habitat types (Creel, 2018;Lima & Dill, 1990).While we do not have the data to determine whether feeders in more open habitat actually exposed 'alal a to greater predation risk from 'io, evidence suggests that the act of opening the feeders reduces aerial vigilance in 'alal a (Lee et al., 2021), which could make them more vulnerable.Meanwhile, the team had previously documented 'io presence and 'alala-'io interactions at the study site in the months leading up to the start of the study (described in Greggor et al., 2021), suggesting that there was an ongoing predation risk in the area.A good test of habitat influences on 'alal a risk aversion would be to examine foraging decisions across habitat types at sites with varying levels of predation risk from 'io, that is, between sites that do and do not have predators present.
Despite their initial avoidance of open habitat feeders, we found that habitat type did not influence longer-term food consumption, as would be predicted if 'alal a adjusted their foraging behavior to reflect the level of predation risk.Perhaps risk-sensitive foraging early at the onset of daily foraging was masked by cumulative motivation to feed throughout the day (Bedoya-Perez et al., 2013).However, it is clear that once feeders were used by birds, they were readily incorporated into their foraging routines.The abandonment by 'alal a of their initial hesitancy highlights the attraction and risks that feeders may impose.These dynamics have implications for the future recovery of the species because predation by 'io was a major factor in recent and past reintroduction efforts (U.S. Fish and Wildlife Service, 2009; Greggor et al., 2021).Placing feeders in open habitat may be illadvised if 'io density is higher there because 'alal a are attracted to feed despite predation risk.Additionally, prioritizing future release sites that have a more closed canopy would offer better options for potentially safer feeder locations that birds would readily adopt.
Although 'alal a were slower to use feeders the farther they moved, all feeders were used within a few days, even at the farthest (150 m) movement.In other recently translocated bird species (e.g., white-shouldered fire-eye, Pyriglena leucoptera), dispersal to new foraging patches was limited beyond distances of 80-90 m, suggesting that many species have upper distance limits where feeder movement would fail to be effective (Awade et al., 2017).
Based on what we now know about the frequency of longer-distance movements in 'alal a (Smetzer et al., 2021), the distances we chose were all well within the ranging behavior of juvenile birds.Therefore, we likely did not move feeders far enough to truly test the limits of feeder movement, but we can at least conclude that juvenile 'alal a will find feeders relocated 150 m at a time relatively quickly.
No individual bird traits predicted feeder discovery.This is surprising, but it could be due to the characteristics we chose to analyze and the sample size of the cohort.Social information use has been suggested to act as a catalyst for foraging innovations and finding new resources in many species (Aplin, 2019;Slagsvold & Wiebe, 2011;Thornton & Malapert, 2009).Meanwhile, it is not uncommon for characteristics such as sex to influence postrelease movement and survival (Germano et al., 2017;Moehrenschlager & Macdonald, 2003).However, we did not find any effects of social network betweenness or sex on the speed with which birds found new feeders, suggesting that moving feeders do not appear to restrict food access to individuals based on these traits.While these results do not produce any management recommendations for feeder movement, there is still potential for these and other individual differences to predict outcomes in alternative management contexts and reintroduction strategies (Merrick & Koprowski, 2017).
To be an effective tool furthering conservation objectives, supplemental food must first be located and utilized by released animals (Bannister et al., 2020), so the selection of food items, presentation, and location of feeders are all important elements of this strategy that need to be considered and empirically evaluated.Based on our results, we can conclude that providing supplemental food altered space use, biasing their home ranges to include central feeder locations.However, feeders moved up to 150 m (and possibly more) will likely be discovered and used by juvenile 'alal a and this tactic may help encourage 'alal a to explore the landscape.This tactic is effective regardless of the bird's sex or position in the social network.Our findings also point to an important role for habitat, indicating that feeders will be more rapidly discovered and used in closed than open habitat, which also reveals a suspected but previously untested habitat preference.In future reintroductions, we recommend that open habitat be avoided when selecting feeder locations.This type of nuanced management may be readily applicable to other species that occupy heterogeneous landscapes with natural boundaries or encounter habitat features that might be connected to predation risk.Additionally, as this management tool is tailored to other species, the distance feeders can be moved to trigger dispersal may also need to be determined.With these details in mind, feeder movement could serve as a costeffective method for directing the movement patterns of released animals and as an additional benefit to using supplemental food.We recommend a comprehensive approach to supplemental feeding tactics, evaluating the efficacy of supplemental feeding as a function of variables such as individual characteristics, site selection for feeders, and the distance animals must traverse to locate new feeding locations.These variables should be explored to determine the overall efficacy of supplemental feeding as well as the role strategic feeder relocation can play in shaping the movement patterns of released animals to achieve beneficial outcomes, including survival and other metrics of translocation success.

F
I G U R E 2 A map of the study area depicting locations where supplemental food was provided.Two perpendicular trajectories (A-C, B-D) are depicted, with red symbols showing the first series of locations (moved 50, 100, and 150 m progressively from the last place a feeder had been located) and turquoise symbols showing the second series of feeder moves.A and C locations were in areas of high cover, meanwhile B and D locations had low relative canopy cover.Purple symbols indicate the original feeder locations prior to the experimental relocations.The white cross shows the location where the 3Â daily VHF signal checks took place.
Closed canopy feeders had taller canopies than the open ones(closed, mean 25.8 ± 7.4  m; open 6.3 ± 3.1 m; two F I G U R E 3 Map of the 95% mean home range utilization distributions (UD) derived from autocorrelated kernel density estimates for 'alal a location data for the 2018 breeding season.UDs with ±95% confidence intervals (CI) were generated using all location data for this time window (orange contours) as well as data filtered to exclude locations recorded within a 50 -m circular buffer around the feeders (blue contours).Also depicted is the group-level conditional distribution of encounters (CDE), shown as the solid orange polygon in the center.The plot shows individual and mean AKDE home range areas generated using the complete location dataset (blue symbols = male, orange = female).

F
I G U R E 5 Plot of raw feeder discovery times in hours since food was first provisioned, by canopy type.Each point represents a feeder discovery by an individual bird.Boxplots depict median values (black horizontal line), with 25th-75th percentile intervals (white box area) and 95% CI (vertical lines).
No submodels had an AICc within four of the top models, so we did not conduct model averaging and instead interpreted the full model (Model 1).The model output from Model 1 is listed in the lower half of this table.Bird ID was included as a random effect in all models. Note: