Journal list menu
Adaptive resource management: Achieving functional eradication of invasive snakes to benefit avian conservation
Abstract
- Natural resource management often co-occurs with considerable uncertainty. One approach to mitigating uncertainty is through adaptive resource management (ARM), a specialized form of structured decision-making that modifies management decisions or actions through monitoring and implementation.
- Here, we present a case study on the attempted eradication of an invasive brown treesnake (Boiga irregularis) in a 5-ha enclosure on Guam with uncertainty in approach. We applied an ARM process across three field phases of snake removal and evaluated whether (1) eradication was achievable and (2) eradication was necessary to achieve an avian response. Field phases included the application of aerial toxic baits, toxicant baiting large mouse and birds, trapping with live mouse and bird lures and hand capture.
- We found that each removal technique improved control by either removing many individuals or targeting a subset of individuals that resisted prior control approaches. Although the effort did not result in eradication, the evaluation of identified indicators allowed for timely adjustments to removal using the ARM process.
- The snake removal efforts yielded an avian response in the treatment area after integrating live birds as snake lures, suggesting functional eradication of snakes may be possible. We also, however, observed a release of invasive rodents following snake control, with birds being more sensitive to the presence of snakes than rodents.
- Synthesis and applications. We suggest that using adaptive resource management to evaluate each phase of action in relation to established goals allowed us to measure outcomes and was successful in eliminating uncertainty in the application of control tools for wildlife conservation. We were able to create a documented and successful approach towards removing snakes inside a snake-exclusion barrier by following the ARM process.
1 INTRODUCTION
Natural resource management requires decision-making in the face of considerable uncertainty (Williams, 2009). Complex interactions between abiotic conditions, inter- and intra-specific interactions, behavioural plasticity, demographic fluctuations and other dynamic ecological processes make the full scope of uncertainty difficult to assess (Regan et al., 2002). Creating a formal and documented process by which recurrent decisions are made within ongoing or emerging uncertainty is the reason for the structure and paradigm behind adaptive resource management (ARM) (Lancia et al., 1996; Murphy & Weiland, 2014). ARM is a specialized form of structured decision-making that creates a documented, logical and transparent decision process to identify uncertainty while field-validating selected approaches (Williams et al., 2011). The strategic application of ARM allows for learning to occur during the management process, which over time can improve the efficacy of management approaches by integrating scientific information to mitigate prior uncertainty. The general goal of ARM is to predict system response to management actions based on known information relative to identified goals, revisit true response after action and revise the management strategy as appropriate. Each step is characterized by explicit statements and identified sources of uncertainty in advance of taking action to develop performance metrics or indicators, which provide a mechanism to address prior uncertainty or redesign approach through time.
As a major threat to global biodiversity, suppression or eradication of invasive organisms is often a desirable goal with documented benefits to native species (Clavero et al., 2009; Hogue & Breon, 2022). Indirect ecological effects or individual heterogeneity in the behaviour of invasive species contribute substantial uncertainty to their management (Cruz et al., 2015; Willson, 2017). With limited cases of successful widespread eradication of invasive species across large landscapes, many of the challenges with eradication programmes are tied to uncertainty. Uncertainty can be attenuated by optimizing approaches, understanding costs, improving ability to remove every individual or anticipating trophic responses to culling or removal efforts.
The challenges described above are particularly pertinent to the management of invasive reptiles, which are increasingly recognized as injurious species (Kraus, 2015). Introduced snakes, for example, have caused widespread native species declines (Dorcas et al., 2012; Piquet & López-Darias, 2021; Savidge, 1987). However, available control tools for snakes are limited, and successful eradications are rare and only at limited spatial scales (e.g. Campbell et al., 2012). Brown treesnakes (Boiga irregularis), an ecologically and economically costly invader to Guam (Guåhan), are generalists, arboreal, cryptic and nocturnal (Rodda & Savidge, 2007). Their introduction to Guam resulted in the extinction or extirpation of 9 of 11 native avian species, 2 of 3 bat species and reductions in the native lizard populations (Rodda & Fritts, 1992; Savidge, 1988). Despite decades of research into their control and removal, much remains unknown about accomplishing eradication for this species, as well as other snakes.
We evaluated the use of ARM for brown treesnake (hereafter snake) management under an eradication or native species recovery goal. Within our study, we applied the ARM approach to engage in an iterative learning process that interactively applied the exchange of data between research efforts in a small focal site and external snake management actions occurring elsewhere on Guam. We defined our effort under two broad goals: (1) attempt to eradicate snakes in an approach that would support landscape-scale eradication; and (2) determine whether avian recovery is achievable with or without eradication. Our work focused on the dominant tools that are currently applied in forested ecosystems on Guam: exclusion barriers, toxic baiting, trapping and hand removal. We herein describe the development of a removal approach that evaluates (1) the overall impact on the snake population; (2) the relative snake-dependent efficacy of tools; and (3) the bird and rodent responses to snake removal. We interpret our results relative to the phases of the ARM cycle, techniques that achieved desired goals, as well as identified limitations and additional sources of uncertainty identified during the ARM process.
2 MATERIALS AND METHODS
2.1 Study site and snake handling
This study took place in an enclosed population of snakes located in Northwest Field North (NWFN), Andersen Air Force Base, Guam, USA (13.639 N, 144.862 E). Guam is the largest and most southern island in the Mariana Island archipelago, with a tropical climate composed of a dry and wet season and peak rainfall occurring from July to October. The experimental snake enclosure is a 5-ha (224 × 224 m) plot of karst limestone forest dominated by invasive Leucaena leucocephala and bound by a barrier preventing snake ingress or egress (detailed in Rodda et al., 2007). Within the study site, we maintained 29 transects separated by 8-m intervals. Throughout the study period, we completed visual surveys of the entire study site 1–4 nights per week to count snakes, rodents and birds.
Prior to control, we censused the population from 24 October 2016 to 22 March 2017 (pre-treatment). All work was completed under U.S. Geological Survey Fort Collins Science Center Animal Care and Use protocols (FORT 2017-03). No permits were necessary to complete this work.
2.2 Adaptive resource management
We completed three field phases and one modelling phase, which were sequential in organization and scaled from least to most costly for implementation. We provide a general description of each tool and known sources of uncertainty when we began implementation of the tool in Table 1. A determination of inadequate response to yield timely eradication triggered the next field phase or sub-phase, which are described in temporal order below (Table 2). Each tool was added to the prior tool, such that no tool stopped being implemented if another tool was added. Briefly, Field Phase I was comprised of snake removal through the use of the Aerial Delivery System (ADS) (Table 1). ADS was initially chosen alone based on the potential to achieve large-scale results with minimal human labour if successful. Field Phase II was comprised of incorporating alternative baits (II-1) of larger size or different species (birds) and live lures (II-2, mice and birds) and was implemented to correct for potential prey selection bias (Table 1). ARM Field Phase III consisted of hand capture-based removal to ensure complete removal of every individual (Table 1) and determine density thresholds for prey response. The use of Phase III triggered a determination that landscape-scale eradication was not possible, requiring a statistical evaluation of the concept of functional eradication (Phase IV). We do not provide tool-specific cost estimates, as ground-based methodology cannot accurately predict aerial costs when tools are operationally implemented.
Study phase | Period of deployment | Control tool | Abbr. | Tool description | Expected benefits | Known sources of uncertainty |
---|---|---|---|---|---|---|
Field Phase I | 31 Mar 2017–4 Feb 2019 | Simulated Aerial Delivery System | (sADS) | The Aerial Delivery System (ADS) is an automated system in which mouse baits encapsulated in a cardboard ‘parachute’ are affixed with 80 mg of acetaminophen (lethal to snakes) and deployed from a helicopter based delivery unit (Siers et al., 2020). Here, ADS is simulated because we deployed baits by hand. Preliminary evaluations of the ADS suggest eradication may be possible, but treatment time would be decades (Nafus et al., 2022) | A broad-scale application of control that does not require ground access and, thus, an ADS-only-based eradication is likely to be most feasible | A potential reduction in willingness to consume carrion as snakes get larger; potential differences in food selections based on size or prey availability; and size-dependent dose responses (Savidge, 1988; Shivik & Clark, 1999; Siers et al., 2021) |
Field Phase I | 5 July 2019–20 Sept 2021 | Modified Aerial Delivery System | (mADS) | A modification of sADS that replaced 5% of baits with larger-sized mouse baits placed on the ground | Improved targeting of larger snakes that forage on the ground more frequently than smaller conspecifics | Unclear if larger baits will attract large snakes, and ground-based baits may be more often taken by non-targets |
Field Phase II-1 | 14 Sept 2020–25 Oct 2021 | Mixed species baiting in Bait Tubes | Bait Tube (BT) | Mouse and bird toxic carrion baits (i.e. baits affixed with an 80-mg acetaminophen tablet) placed inside polyvinyl chloride pipes (bait tubes) intended to exclude non-target species and hung from trees at a height of 1.5 m. We used a roughly equal number of mice (Mus musculus, small rodent, N = 201), rats (Rattus rattus, large rodent, N = 205), quail chicks (Coturnix spp., small bird, N = 206) and chicken chicks (Gallus gallus, large bird, N = 206). In prior studies, baits composed only of mice targeted 61% of the susceptible population and appear to have size-dependent effects that may be dependent on snake or bait size (Lardner et al., 2013; Shivik & Clark, 1999). Traditional baiting efforts for brown treesnake control and management have only used mice (Clark et al., 2018). However, more recent research in laboratory settings indicates that bird bait may target different snakes than mouse bait (Nafus et al., 2021) | Ability to target individuals with different food preferences. Also, potential integration into the ADS, maintaining an aerial delivery process | Degree of individual food preference or whether brown treesnake are generalists as individuals or a species |
Field Phase II-2 | 1 Mar 2021–23 Dec 2021 | Mixed Lure Live Trapping | Trap | Traps are modified minnow traps retrofitted with one-way entrances and a lure chamber (Rodda et al., 1999). Trap effects are size dependent with small snakes not attracted to mouse lures, but given sufficient effort, are suggested to be 100% effective at removing snakes >900 mm SVL (Rodda et al., 2007; Rodda & Fritts, 1992). As a tool, traps must be monitored regularly, have low capture relative to encounter rates, and have a higher cost of application than baiting (Clark et al., 2012; Yackel Adams et al., 2019). Lures included live mice (Mus musculus) or finches (Taeniopygia guttata) | Remove snakes that are not interested in carrion food attractants | Percentage of snakes not willing to consume carrion but willing to enter a trap in pursuit of live prey. Also, the degree of individual food preference variation and the number of snakes that might target one lure species but not the other |
Field Phase III | 24 Oct 2016–31 Dec 2022 | Hand capture by people | Hand | Surveys occurred 0–20 times per month during the period of deployment. We used two-person teams of trained observers that scanned opposite sides of transects, as detailed in Christy et al. (2010). We released captured snakes following the collection of morphological measurements until 24 January 2022, after which we euthanized snakes. Birds and rodents were counted during surveys to provide a measure of prey abundance across time | Allow for the capture of all size and age classes of brown treesnake, albeit with low capture probabilities that are influenced by habitat and other conditions (Christy et al., 2010; Rodda et al., 2007; Rodda & Fritts, 1992) | Whether every individual can truly be found and how long that would take in a given area |
Phase | Tool | Indicators of success | Evaluation approach |
---|---|---|---|
I |
Simulated Aerial Delivery System (ADS) toxic small mouse carrion |
Decline in snake population; Limited birth rate of snakes; Population near zero after 24 months of treatment. |
Monitoring individual mortality with telemetry after first two applications; Visual survey and trapping censusing to monitor declines; Modelling of data every 3 months. |
I |
Modified ADS toxic small and medium mouse carrion |
Improved removal of snakes >1000 mm snout–vent length (SVL, mm); Declining population of snakes; Reduced birth rate of snakes. |
Visual surveys and trapping censusing; Demographic evaluation through vital rate analysis and integrated projection model after 24 months of removal; Estimated time to achieve eradication. |
II-1 |
Bait Tubes toxic large mouse, rat, quail chick and chicken chick carrion |
100% removal of females >1050 mm SVL |
Assessment of whether females >1050 mm SVL were still detected after >2 treatment cycles (visual surveys); One month between visual survey review cycles. |
II-2 |
Live-lure trapping alive finches and mice |
100% removal of females >1050 mm SVL; Differential removal of snakes by bird- or mouse-lure traps. |
Assessment of whether females >1050 mm SVL were still detected after >2 treatment cycles (visual surveys and trapping) |
III-3 |
Hand Human removal |
100% removal of snakes |
Visual surveys; Weekly monitoring. |
2.2.1 ARM field phase I: ADS
Simulated ADS (sADS; Table 1) used toxic mouse carrion baits as the only removal approach. We manually distributed 594 standard ADS baits along an 8 × 10 m grid in the tree canopy and forest floor (refer to Nafus et al., 2022 for more detail). Following assessment indicating a need to improve the removal of ground foraging snakes (Nafus et al., 2020), we adjusted our baiting efforts to include larger-sized mouse baits placed on the ground (modified Aerial Delivery System; mADS; Table 1). The performance indicators and monitoring through population censusing were set to assess the likelihood that ADS could yield eradication and included population decline and a reduction in reproduction and/or recruitment (Table 2).
2.2.2 ARM Field Phase II-1: Mixed species baiting in ‘bait tubes’
To address uncertainty as to whether snake removal is affected by individual food preferences based on prey species or size, Phase II-1 implemented a mixed species effort that included mouse and bird carrion baits staged in bait tubes (BT; Table 1). Each baiting treatment was composed of three applications and occurred shortly after an mADS treatment. We included a 3-day no-baiting period between the first and second applications and a 4-day no-baiting period between the second and third applications. We completed six perimeter baiting applications using 78 bait tubes placed along the study site perimeter, inset by 5 m, along pre-cut transects. We subsequently conducted six interior baiting events in which the 78 bait tubes were distributed equally through the 5-ha plot. Performance indicators included the removal of reproductive females (Table 2), and inadequate success in removing snakes triggered lure integration (Field Phase II-2).
2.2.3 ARM Field Phase II-2: Mixed species live-lure trapping
To address uncertainty as to whether snake removal was affected by prey species or state (live or dead), we incorporated live mouse and bird lures as a removal tool (Trap; Table 1). We used traps as a sporadic censusing tool from 1 February 2017 to 6 November 2020, following intervals described in greater detail in Nafus et al. (2022). We transitioned to using traps as a lethal form of control from 1 March 2021 to 23 December 2021, during which we completed three trapping cycles. For trapping intervals 1–2, we placed 36 traps along the study site perimeter inset by 5 m, with bird lures in every third trap (N = 12) and mice in the remainder (N = 24). For the third interval, we saturated the landscape with traps along a 16 × 16 m grid and used finch lures in traps located in the middle of the grid (N = 13) and mouse lures for the remaining traps (N = 156). Performance indicators included 100% removal of reproductive females and comparative capture rates by mouse or finch lures (Table 2).
2.2.4 ARM Field Phase III: Visual surveys and hand removal
We hand-captured snakes during visual surveys (Hand; Table 1), which allowed for the capture of all snake sizes and age classes. Although they are logistically difficult and unfeasible at large spatial scales, understanding precisely what density level snakes are required to be at to observe a prey response was critical to addressing whether avian conservation goals are achievable. Visual surveys were therefore viewed as essential to remove every snake to clarify uncertainty surrounding demographic relationships to prey responses (functional eradication) and which control tools yielded those conditions (Phase IV—Analytical).
2.3 Statistical approach
We used R v. 3.6.1 (R Core Team, Vienna, Austria) and accepted significance at α ≤ 0.05 where applicable. We evaluated population and individual responses across all ARM phases to demonstrate the relative effects of different removal approaches on population responses and size-dependent effects or individual traits that affected heterogeneity in response to a given control approach. After Field Phase I, we elected to use performance indicators that were simple and did not require complex analysis to adjust to the next phase under the target of management objective 1. These analyses were not part of the ARM process but post-hoc evaluations across the entire effort to assist future decision-makers on when a specific control approach may be most optimal based on desired management outcomes and known response heterogeneity by snakes.
To compare size-based removal by treatment across the entire effort, we used a generalized linear regression with a normal distribution to measure the comparative size (log-transformed snake SVL) of snakes removed (1 [not detected]) versus not removed (0 [detected]), the size of removed snakes within treatment periods (sADS, mADS, bait tube, trap and hand capture) and their interaction as predictors. This analysis was completed to evaluate whether there were differences in the overall size of the removed snakes across treatments relative to non-removed snakes based on presence in the population during a given treatment approach.
To evaluate the effects of each treatment type on the probabilities of removal for susceptible-sized (≥850 mm SVL) snakes based on their individual traits, we used Cox proportional hazards models to compare the risk of failing to detect a snake during the treatment period after a specified monitoring interval ended. We used the ‘survival’ package in R to complete the Cox proportional hazards model (Therneau, 2014). We included the maximum size of individuals, their maximum BC, an interactive effect of size and condition and the treatment as predictors. To sequentially compare across treatments, we ran four hazard models identified as ‘ADS’, ‘BT’, ‘TRP’ and ‘HND,’ where each model built upon prior Field Phase (i.e. BT built on ADS, Table 3).
Phase | Model | Removal tool (TRT) includes |
---|---|---|
I: Aerial Delivery System (ADS) | surv(status, days) ~ SVL × BC + TRT | (1) no removal, (2) simulated ADS or (3) modified ADS (included 5% medium ground baits) |
II-1: Bait Tube (BT) | surv(status, days) ~ SVL × BC + TRT | (1) Phase I: ADS or (2) Phase II-1: ADS and BT |
II-2: Live-Lure Trapping (TRP) | surv(status, days) ~ SVL × BC + TRT | (1) Phase I and Phase II-1 or (2) Phase II-2: TRP |
III: Hand capture (HND) | surv(status, days) ~ SVL × BC + TRT | (1) Phase I and Phase II or (2) Phase III: HND |
For each model, in each interval, we included individuals that were detected during and after the specified interval. The monitoring period was the number of days between the first time that a snake was estimated to be ≥850 mm SVL within a treatment period subtracted from the last day the individual was detected in the treatment period. An individual that was not detected again after a treatment phase was assigned a status of 1 (e.g. a snake last detected during mADS received a 1 in the ADS model). An individual that was detected in a subsequent Field Phase was assigned a status of 0. For trapping and hand capture, we had known removals that we assigned a status of 1.
2.3.1 ARM phase IV (analytical): Functional prey response
We processed visual encounters during nocturnal surveys of the two prey taxonomic groups, rodents and birds, separately. To create an estimate of abundance, we used sightings per unit effort (SPUE) of prey by month as our metric. Monthly SPUE was calculated as the total number of counts across all survey days in a given month averaged by transect and divided by the sum total distance sampled across the same period cross all transects. We used a two-step process to identify the snake management action associated with a functional response in prey, as estimated by SPUE. In the first step, we used Akaike information criterion (AIC) approaches to identify whether snake density was correlated with prey SPUE and which, if any, specific snake size classes best fit prey SPUE. When evaluating size class, we divided the population into < or ≥ the specified SVL and included both classes as predictors. We evaluated five size groups, starting at the ‘susceptible’ (≥850 mm SVL) size class and increasing at 50-mm SVL increments up to ≥1050-mm SVL, as larger animals were uncommon throughout this study. We used Poisson generalized linear regressions with a log link and corrected prey SPUE to a whole number by multiplying by 100, where SPUE was the dependent variable and size-specific density (snakes/ha) was the predictor. For the second step, we used the snake size model identified through the above AIC approach in a segmented (piecewise) regression with one breakpoint using the ‘segmented’ package in R (Muggeo, 2022). The snake density that correlated to a change in the prey SPUE regression line was referenced against the date at which that density of snakes was achieved to identify the Field Phase that was occurring at the specified density.
3 RESULTS
Throughout the removal effort, we detected 267 unique snakes. Overall, the snake population declined from the 117 individuals present in October 2016 to an estimated final population size of 6 (95% decrease) by 31 December 2022, which was determined a failure to achieve eradication (Figure 1a). The number of snakes declined substantially following the first ADS application and then experienced two short-lived reproductive bursts, defined by an increase in the number of individuals <850 mm SVL but with no reproductive adult recovery (Figure 1b). The sADS and mADS tools yielded the greatest rate of removal per days of treatment (Table 4).

Phase | Tool | Effort unit | Effort | CPUE | Prop. Pop. | Days | Snake/day | Weaknesses | Strengths |
---|---|---|---|---|---|---|---|---|---|
I | sADS* | Bait No. | 8910 | 0.0120 | 0.55 | 45 | 2.38 |
Low targetability of larger snakes (>1000-mm snout–vent length [SVL])a; Less able to target ground foraging snakesa; Poor removal of snakes in good condition. |
Rapid removal of 50% of the populationa. |
I | mADS* | Bait No. | 11,893 | 0.0057 | 0.69 | 60 | 1.13 |
Ground baits rapidly consumed by non-target species; Negligible improvement in removal of snakes >1000-mm SVL; Poor removal of snakes in good condition. |
Yielded secondary decline after population stopped responding to sADSa. |
II-1 | BT M* | Bait No. | 302 | 0.0464 | 0.20 | 30 | 0.47 |
Carrion rat baits not consumed by snakesb; Requires more intensive groundwork; Snake suppression sufficient to yield rodent response creating competing prey availability and alternative invasive species issueb; Negligible improvement in the removal of snakes >1000-mm SVLb; Poor removal of snakes in good condition. |
Including bird baits appeared to remove individuals that had previously escaped removal by carrion baitingb. |
II-1 | BT R* | Bait No. | 301 | 0.0066 | 0.03 | 30 | 0.07 | ||
II-1 | BT C* | Bait No. | 294 | 0.0544 | 0.23 | 30 | 0.53 | ||
II-1 | BT Q* | Bait No. | 304 | 0.0493 | 0.21 | 30 | 0.50 | ||
II-2 | Mouse | Trap nights | 5748 | 0.0002 | 0.02 | 71 | 0.01 |
Substantive prey release creating alternate prey, population stopped responding to traps; Lingering bias against trapping the largest snakes. |
Improved removal of snakes in good condition; Bird traps successful in removing females >1050-mm SVL; Sufficient removal to observe avian response in treatment area. |
II-2 | Finch | Trap nights | 574 | 0.0087 | 0.09 | 71 | 0.07 | ||
III | Hand | Distance (km) | 926 | 0.0518 | 0.75 | 79 | 0.61 |
Extremely time intensive; Large lags in detecting some individuals (>12 months between detections). |
No further size or condition bias in removal. |
- Note: The process sequentially cycled through four removal tools: a simulated Aerial Delivery System (sADS) toxic mouse traditional baits; a modified ADS including larger ground-placed baits (mADS); mixed species toxic baiting (BT) including quail chicks (Q), chicken chicks (C), mice (M) and rats (R); trapping using live mouse (M) or finch (F) lures (TRAP); and hand captures during visual surveys (HAND). The effort unit reflects the unit type used to calculate total effort by tool relative to the number of snakes removed* by a given tool (CPUE). The proportion of the population (Prop. Pop.) represents the number of snakes removed* divided by the total number of snakes available for removal. Total days is the number of days a given control was on the landscape over the full treatment period. Snakes per day is the number of snakes removed* divided by the total days a tool was applied.
- * Estimated removal based on failure to detect during next treatment interval.
- a Nafus et al. (2020).
- b Nafus et al. (2023).
We did not detect 107 out of 194 (55%) snakes following sADS treatments. For mADS treatments, 68 of 98 snakes (69%) were not detected after baiting. A comparison of population size available for removal suggested similar sizes during sADS and mADS treatments (p = 0.13; see Appendix S1), but that mADS removed on average smaller snakes (p = 0.04; see Appendix S1). For the toxic mixed species bait tubing period, we failed to detect 18 of 71 (25%) snakes after treatment intervals. The average size of snakes available in the population had decreased (p = 0.001), while removed individuals tended towards being larger compared to ADS alone (p = 0.06; see Appendix S1). Across all lethal trapping events, we removed 6 of 59 (10%) snakes. The average size of snakes available remained smaller during trapping (p = 0.03), and removed individuals tended towards being larger compared to ADS alone (p = 0.06; see Appendix S1). We removed 55 of 61 (90%) snakes known to be present by hand capture, including only four previously captured snakes. The average size of snakes remained smaller (p = 0.05), but there was no difference in the size of removed snakes relative to ADS alone (p = 0.23, ST 1). Trapping had the lowest removal rate per days of effort (Table 4). Each tool appeared to correct for different biases in individual removal (Table 4). In general, across the entire population of snakes, those that were removed by trapping and toxic baiting averaged 964 ± 13 mm SVL (Figure 2) and were overall larger than those snakes that persisted (p < 001, ST 1).

Cox proportional hazards supported that susceptible snakes (≥850 mm SVL) were over two and three times more likely to be detected in the pre-treatment interval, as compared to the sADS (, p = 0.02) and mADS (, p < 0.001) phase periods, respectively (see Appendix S2). During the mixed species baiting, susceptible individuals were three times less likely to be detected after the treatment period as compared to both ADS treatments (, p < 0.001; see Appendix S2). Susceptible snakes had equal probabilities of removal during the trapping treatment as compared to all toxic baiting (, p = 0.20; see Appendix S2). Similarly, visual surveys yielded equal probabilities of removal for snakes susceptible to other treatments as all proceeding treatments (, p = 0.15; see Appendix S2). Across all treatments, sex had no effect on the post-treatment probability of detection for susceptible snakes (see Appendix S2). Susceptible snakes in better condition and of smaller sizes had greater probabilities of being detected post-treatment during ADS (BC: , p = 0.007; SVL: , p = 0.003; see Appendix S2) and mixed species baiting (BC: , p = 0.03; SVL: , p = 0.009; see Appendix S2) treatments. Size and condition positively interacted for both ADS (, p = 0.009; see Appendix S2) and mixed species baiting (, p = 0.03; see Appendix S2), supporting that the largest, best-condition snakes were most likely to be detected after toxic baiting treatment intervals ended. Trapped susceptible snakes tended to be in slightly worse condition (, p = 0.06; see Appendix S2), were smaller in size (, p = 0.02; see Appendix S2), but without a significant interactive effect between size and condition (, p = 0.07; see Appendix S2) compared to those that persisted. Condition had no relationship to hand-captured susceptible snakes (, p = 0.11; see Appendix S2), while hand-captured snakes tended to be smaller than those that persisted (, p = 0.06; see Appendix S2). There was no interaction between size and condition (, p = 0.15; see Appendix S2) on the probability of susceptible snakes surviving the hand capture treatment period.
3.1 Functional prey response following predator control
- If x ≤ 3.994: y = 6.037–1.526 × (x).
- If x > 3.994: y = 6.037–1.526 × (3.994) + (−1.526 + 1.559) × (x − 3.994).

- If x ≤ 2.0: y = 2.232–1.410 × (x).
- If x > 2.0: y = 2.232–1.410 × (2.0) + (−2.232 + 1.342) × (x − 2.0).
4 DISCUSSION
We found iterative modification of snake removal strategies based on evaluation of data from the prior phase to be an effective approach for maintaining a continuous decline of snakes in the treatment area. Throughout the full removal effort, the reproductive population of snakes did not recover, although eradication was not achieved. Our results did not support an approach for large-scale eradication without long-term or ground-intensive efforts. We did, however, determine that eradication may not be necessary to support initial avian recovery goals.
A matter of important consideration in programmatic assessments of invasive species control is not just the feasibility of eradication, but also its merits or necessity. Thus, understanding management feasibility is an important component in assessing whether goals and targets are potentially viable (Courtois et al., 2018). For the management of a novel situation—invasive reptile removal for native species recovery—there are numerous unknowns that could contribute to inefficient resource use. Our data indicate that eradication remains an elusive outcome for brown treesnakes, but also that such an outcome may not initially be necessary for species to rebound, and ‘functional’ eradication may be sufficient. Functional eradication is defined as reducing local population densities below levels that cause negative population growth in prioritized species (Green & Grosholz, 2021). Applying such a strategy for the conservation of endangered species would, however, likely require a better understanding of the direct numeric relationship between the density of the invader and the functional response of the target species, which can be defined as the density-impact relationship (Green & Grosholz, 2021). The density of snakes required to observe a response in birds was extremely low (2 snakes ≥900 mm SVL/ha), and comports with prior estimates based on vital rates and expert opinion for avian species on Guam (McElderry et al., 2022). We also did not achieve target densities or removals until we deployed live avian lures in our control programme. Therefore, specific tools may be important parts of the control programme, depending on the desired management outcome.
Guam is also considered prey depleted (Savidge, 1987; Wiewel et al., 2009; Wiles et al., 2003). Thus, research into the control and removal of brown treesnakes, including control tool efficacy, has primarily been completed after the loss of birds on Guam had already occurred. In areas with greater prey availability, control tools have been less successful in capturing and removing individuals (Gragg et al., 2007). This, combined with other studies indicating context-dependent behaviour (Nafus et al., 2020, 2021; Siers et al., 2024; Yackel Adams et al., 2019), indicates snakes may be more flexible about prey selection when resources are limited. This conclusion is also supported by our data, which indicate a direct relationship between body condition and the risk of removal using toxicants (carrion). These combined results also indicate that the order or sequence and approach to removal may affect outcomes. For example, had our programme been initiated with an application of all tools at once (prior to prey recovery), we may have achieved a much faster and more thorough removal than documented here using a slow scaling up of tools across time. Likewise, once birds have recovered, current control tools may become less effective based on an availability of accessible prey (Siers et al., 2024), limiting the viability of long-term suppression.
Brown treesnakes also have ontogenetic dietary shifts with rodents and birds added to the diet as individuals grow; thus, diet ontogeny may be important to consider when unmasking predator–prey population dynamics (Lardner et al., 2009; Savidge, 1988). Rodents, for example, showed stronger statistical relationships with snakes at the 950-mm SVL threshold, as compared to the 900-mm SVL size for birds. Additionally, the allowable density of snakes was lower for achieving a functional response in birds, arriving later in the removal effort and with more costly tools. The combination of these data indicates future uncertainties, including that data collected prior to rodent or avian recovery may not inform snake responses when birds are available as alternative prey. Studies on which tools continue to remain effective as population demography changes and when prey are abundant are likely critical towards reducing uncertainty surrounding snake control for avian recovery goals.
The specific mechanism of recovery in prey species may also be important to consider in evaluating outcomes relative to avian reintroduction, specifically. For example, avian SPUE changed abruptly in our study over a short period. Given the rapidity, avian SPUE changes likely resulted from increased immigration or improved survival of recent immigrants rather than reproduction. If avian functional response was driven by immigration, this would not necessarily predict an environment that was suitable for reintroduction, given the mechanistic dependence on reproduction and survival of fledglings for reintroduction goals. The fledging stage is estimated to have a high risk of predation by brown treesnakes for extant forest birds (Pollock et al., 2019). Thus, a functional response in extant birds may not reflect suitable conditions for the reintroduction of extirpated species.
An additional confounding factor in understanding interspecies dynamics is also our inability to differentiate between density dependent and individual effects of predator control on functional prey responses. Specific individuals can have disproportionate effects on native species that are not necessarily density dependent (Cruz et al., 2015). Therefore, the relationship between prey survival and snake density may not be linear. Low sample size, unsuitable study design, and lack of replication inhibit our ability to differentiate between these potential options. Direct studies of avian behaviour and reproduction in conjunction with snake control efforts would be necessary to further clarify these sources of uncertainty.
The variance in factors that predicted individual response to specific control tools supported a general idea that individual behavioural heterogeneity may inform both susceptibilities to control tools and the success of a suppression programme for a species recovery goal. For example, likely due to the ease of application and the scale at which ADS could be applied, the greatest per control day removal of snakes was by use of this tool. However, snakes in larger size classes have lower susceptibilities to acetaminophen intoxication and a reduced probability of mortality (Goetz et al., 2021; Siers et al., 2021), whereas carrion was apparently less attractive to snakes in good condition. The addition of live avian lures to the control programme was critical for removing good-condition individuals and achieving extant avian responses. Thus, carrion-based control programmes for snakes may be a relatively inexpensive approach to quickly remove a high number of individuals but may inadequately target snakes necessary to remove for avian conservation goals.
4.1 Management implications
A significant first goal for any invasive species management plan is to establish management objectives and performance indicators to monitor success. We found that a phased iteration of control tool evaluation under an adaptive management framework was informative in both removing uncertainty and identifying sources of future uncertainty related to setting broader conservation goals. We also determined that current control tools can suppress snakes sufficiently to achieve an avian response, but that large-scale snake eradication remains an elusive outcome.
AUTHOR CONTRIBUTIONS
Melia G. Nafus conceived the work. Melia G. Nafus and Scott M. Goetz designed the study. Melia G. Nafus and Amanda Reyes collected and analysed the data. Melia G. Nafus, Amanda Reyes and Thomas Fies drafted the initial manuscript. All authors contributed to revisions and approval.
ACKNOWLEDGEMENTS
We thank the biologists that have participated in the coordination and collection of these data. We thank S. Siers, R. Reed, D. Vice, and especially M. Hall and M. Mazurek for their invaluable feedback during development. We thank E. Paxton for reviewing and improving an earlier draft. This work was completed with funding support from the U.S. Office of Insular Affairs, U.S. Department of Navy (Agreement #N6112820MP001GS), and U.S. Marine Corps Forces, Pacific (Agreement # M2002117MPDP007). Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
Open Research
DATA AVAILABILITY STATEMENT
Data available via ScienceBase https://doi.org/10.5066/P9QRWKQB (Nafus et al., 2023).