Spatial ecology to strengthen invasive snake management on islands

Knowledge on the spatial ecology of invasive predators positively contributes to optimizing their management, especially when involving cryptic and secretive species, such as snakes. However, this information is lacking for most invasive snakes, particularly on islands, where they are known to cause severe ecological and socio-economic impacts. This research is focused on assessing the spatial ecology of the California kingsnake (Lampropeltis californiae) on Gran Canaria to strengthen management actions. We monitored 15 radio-tagged individuals once per day on 9–11 days per month from July 2020 to June 2021 to calculate the species' home range and describe annual activity patterns in the invaded range. To account for the species' diel activity during the emergence period, we additionally monitored snakes from January to May 2021 during three consecutive days per month in four different time intervals each day. We detected movement (consecutive detections at least 6 m apart) in 31.68% of the 1146 detections during the whole monitoring period. Movements most frequently detected were shorter than 100 m (82.24%), and among them the range 0–20 m was the most recurrent (27.03%). The mean distance of movement was 62.57 ± 62.62 m in 1–2 days. Average home range was 4.27 ± 5.35 ha—calculated with the Autocorrelated Kernel Density Estimator (AKDE) at 95%—and did not significantly vary with SVL nor sex. We detected an extremely low value of motion variance (0.76 ± 2.62 σ2m) compared to other studies, with a general inactivity period from November to February, January being the less active month of the year. Diel activity was higher during central and evening hours than during early morning and night. Our results should be useful to improve control programs for this invasive snake (e.g., trap placement and visual survey guidance) on Gran Canaria. Our research highlights the importance of gathering spatial information on invasive snakes to enhance control actions, which can contribute to the management of secretive invasive snakes worldwide.


Materials and methods
Experimental design: study area and sample size. Aiming to conduct a year-round monitoring of L. californiae spatial behavior, we located our study area in Amagro Mountain Natural Monument (NW Gran Canaria), invaded since 2010 24 , and where there were no anthropic infrastructures that would limit our monitoring or affect snake spatial behavior 35 .
In order to set our sample size, we had to adjust it to logistic and economic constraints, as well as to the number of snakes captured in Amagro and their weights. We only selected snakes captured in this area by the control staff between February and June 2020 to prevent homing behavioral effects 36 , and weighing over c. 180 gtransmitters (SI-2 9g, Holohil Systems, Ltd., Ontario, Canada) could only make up < 5% of animal weight 37 . We finally tagged 15 adult L. californiae individuals, measured their snout-to-vent length (SVL) with a measuring tape, and sexed them using a probe 38 . They comprised 6 males and 9 non-gravid females (mean ± SD SVL: 96.62 ± 11.05 cm; min-max SVL: 80-121 cm), all showing normal coloration 28,39 . All individuals were kept in captivity in appropriate conditions until surgical insertion of transmitters (all cases between July 4th-6th 2020), with a food intake the previous week. Surgery was carried out by an experienced veterinarian, following a protocol adapted from Melián 40 (see Supplementary Material 1). Veterinarian post-surgery care lasted for 48-72 h. All individuals survived surgery, so we released them back into the wild in their capture location or its vicinity. Field tracking. To estimate home range and describe the species phenology, we tracked individuals from July 2020 to June 2021. This one-year survey involved one field tracking session per month, each of 9-11 consecutive days, with c. 15 days between sessions, detecting all individuals once per day. Complementarily, to confirm activity during night-time, early morning or dusk, as suggested by Hubbs 28 , we analyzed diel activity following an experimental design adjusted from Abom et al. 41 . Thus, we detected individuals four times a day during three consecutive days (starting on the third day of tracking described above), only from January 2021 to May 2021-i.e., emergence period until the number of captures starts to decrease 25 . Each day we detected individuals at the beginning of the following periods: (1) early morning (7:30-10:30), (2) central hours (10:30-17:30), (3) evening (17:30-20:30), and (4) night (20:30-7:30) (see Abom et al. 41 ). Period duration is approximate, as we based session times on sunrise and sunset time each month.
To locate each individual exactly, a team of 2-3 people, each provided with a Biotrack three-element Yagi antenna coupled to a Biotrack Sika handheld receiver (Biotrack, UK), followed the signal until the 2-3 antennas coincided at the same spot. We recorded detections with a 5G GPS cell phone at c. 2 m precision, using the area aerial photograph in the QField App (The QField Project/OPENGIS.ch 2019). For each detection, we recorded whether each individual was on the surface (basking or moving) or sheltered.

Data analyses. Data preparation.
To estimate home range and describe the species phenology, we discarded the July 2020 data to avoid the collateral effects of surgery on movement behavior 42 . We defined movement as consecutive detections at least 6 m apart. An individual was assumed dead when it was detected in the same place for a long time (c. 1 month, excluding the brumation period, c. from November to February) until the end of the monitoring, assigning the date of the first detection at that place as the final detection date for subsequent analyses. Particularly for the diel activity analysis, we considered that an animal was active not only when it was located at least 6 m apart from the previous detection, but also when it was on the surface (informative of the opportunity to capture snakes for control purposes).  53 . We set window size and margin parameters to 7 and 3 telemetry detections (7 and 3 days in our case), respectively, based on the typical movements of individuals (snakes moved every c. 3 days) and our sampling regime, as suggested by Kranstauber et al. 52 . Analysis of motion variance informs about behavioral changes across a tracking period, detailing how straight a movement path is, as well as how much a path varies in speed, and the scale of movements in a time window 52 . Motion variance is high when animals are active and their path is irregular, but low when animals are inactive and/or follow a regular path 52 .
Effect of SVL, sex and month in the species spatial ecology. To analyze the influence of SVL and sex on home range and motion variance, we performed Spearman's correlation tests and Kruskal-Wallis tests 54 , respectively. We first checked home range and motion variance normality following the methods of Zuur et al. 55 , and the homogeneity of the variance for males and females with Levene's tests 56 . For motion variance, we additionally quantified individual variation with a repeatability estimation test 57 using the rptR v. 0.9.22 package 58 , and seasonal patterns using month as the explanatory variable in a Welch's heteroscedastic F Test with trimmed means (rate set at 0.1) and Winsorized variances 59 performed on the onewaytest package 60 .
We assessed snakes' diel activity with a Generalized Linear Mixed-Effects Model (GLMM) using a binomial error distribution for activity occurrence, including sex, month, and period of day as fixed factors, and the individual as a random factor. We also tested the interaction between month and period of day as it was our primary interest, but discarded it due to convergence issues. After verifying model assumptions-i.e., homogeneity of variance, normality and dispersion of residuals-with DHARMa v.0.4.3 package 61 , we retrieved the models' main effects using type-II Wald Chi square tests performed with ' Anova' function 62 , conducting post hoc analyses with the emmeans v.1.7.0 package 63 to obtain differences between each factor level.
We performed all analyses using R v. 4 Home range. We considered that all individuals' variograms were stable (Fig. 1). However, the effective sample size of four individuals was exceptionally low (< 10; 010, 590, 930, 950) (  Fig. 2). There was no correlation between home range and SVL (r S = -0.37, P = 0.323), nor significant difference between sexes ( χ 2 1 = 0.24, P = 0.624). Results from the GLMM revealed that diel activity was significantly affected by sex ( χ 2 1 = 4.26, P = 0.039), month ( χ 2 1 = 20.03, P < 0.005), and also period of day ( χ 2 1 = 29.51, P < 0.005) during the intense monitoring (January-May 2021). Females were more frequently active (basking/moving) than males (15.68% female detections and 8.55% male detections) (Fig. 4A). Post-hoc comparisons between months showed that January was the month with a significantly lower activity (P < 0.05 for all comparisons: Fig. 4B). Snakes were significantly more active during the central hours of the day and the evening than in the early morning and night (P < 0.05 for all comparisons; Fig. 4C).

Discussion
Our research provides essential information on the spatial ecology of the invasive L. californiae on Gran Canaria, which contributes to guide control actions for this pernicious invader. Home range data can be incorporated into designing control actions to manage invasive terrestrial vertebrates 10,12,65 , and are key to defining the density and distribution of traps 13 . On Gran Canaria, trap placement did not follow a specific density and distribution. Extracted from our home range calculation, to maximize the probability of a snake encountering a trap, these should not be separated by more than the minimum range detected, 104 m (diameter of a 0.85 ha circle, assuming a circle to be the most parsimonious design for a home range area; see Roy et al. 66 for this calculation). Since Table 1. Tracking summary by individual, showing the first day of tracking (Start monitoring date) and the last day we included in our analysis (Last fix date). Since we excluded July from the analyses, sample period represents the number of days between August 2020 and the last fix date that we used for each individual analysis. Battery life was calculated as the difference between the date the transmitters were implanted (4, 5 or 6 of July 2020) and the last day we received signal or the last monitoring day. We also show data points, movements (no. of movements longer than 6 m), and short movements (no. of 2 to 6 m movements) for each animal (July data between parentheses). We indicate movement mean ± SD (standard deviation) and maximum distance (m) (Euclidean) for each animal. We show the longest period (days) without movements we could ensure for each individual (Max days without movement). We also indicate overall mean ± SD for each parameter, except the dates.  68 , we recommend using grids or linear designs, particularly in biodiversity-sensitive areas of Gran Canaria or along the expansion front. The use of drift-fences can also increase the probability of capture 69 . Distance traveled in a day can be used to infer the width of control buffers to prevent invasion of adjacent areas. Given that most movements detected were smaller than 100 m, we recommend a containment buffer width of 100 m. These recommendations can be improved in the future by calculating trap distance using simulations 65 or the most cost-effective trap arrangement 70,71 . Control staff can also incorporate snake density information, as it could influence individual home range 12,35 . www.nature.com/scientificreports/ Lampropeltis californiae phenology on Gran Canaria is highly seasonal, with a brumation period from late November to early February, which coincides with information reported for the island/coastal regions of its native range 28 . The species is active the rest of the year, with an activity peak between March and May, presumably linked to the breeding season 30 . Each year, environmental managers have reinforced control actions by hiring extra personnel between March and August. Following our results, this reinforcement should be extended from early February to mid-November, increasing efforts before the breeding season and until the end of the activity period. From an ecological perspective, the average motion variance of L. californiae, a wide-foraging predator 26 (0.76 ± 2.62 σ 2 m, mean ± SD), is lower than in other active-searching or even ambush predators (mean ± SE for active predators: Ophiophagus hannah: 27.9 ± 0.6 m 72 , Boiga cyanea: 2.8 ± 0.8 σ 2 m 73 ; mean ± SE for an ambush predator: Python bivittatus: 2.66 ± 0.14 m 74 ). This seems to indicate that the species continuously forages until finding prey and then shelters and remains stationary for several days, lying dormant until it surfaces again to prey-which coincides with our findings that movements usually occurred in blocks of consecutive days. This result can be applied to management when a snake is detected but not captured, because the animal will probably remain in the same shelter/area for 2-3 days. Consequently, prospection of the surrounding area during that period could increase the probability of capture. In addition, as a notable amount of movements were short, intense visual surveys in the proximity of fresh snake tracks (e.g., scats or shed skins) for a similar amount of days can be also appropriate. Finally, it is worth noting that, since individuals were sheltered in most of the detections, the development of novel techniques to detect animals while immobile or sheltered is crucial to improve control success 67,[75][76][77] . Due to visual surveys being extremely time-and resource-consuming 15 for such a secretive snake, increasing detection on surface still requires further technological advances-e.g., remote sensing techniques 78 .
Diel activity analysis showed that females were more frequently active than males between January and May 2021, potentially due to their different behaviors during the breeding season (i.e., feeding, basking) 79 or the associated reproductive costs of breeding for females 80 . This strengthens our previous recommendation to reinforce control staff from mid-February onwards to increase the probability of capturing females before reproduction begins. We also determined that L. californiae is mainly active during central hours and the evening, although a certain proportion of activity occurred at night during normal weather (15.40% of all activity detected), and even in the rain. To link visual surveys to species activity pattern 81 , the former should be made during central hours and evenings.
The overall spatial parameters studied for L. californiae on Gran Canaria showed a wide variation. This disparity may be explained mainly by the effect of individual heterogeneity on spatial behavior, already demonstrated for other invasive snake species 82 . This spatial heterogeneity can be due to individual body condition affecting movement ecology 83 , individual personality influencing exploratory and defensive behaviors 84 or boldness and sociability 85 , as well as dispersive movements (such as may have happened with our individual 010). In addition, although some deaths are expected in this type of studies (e.g., due to tagging surgery, predators, health condition see 27,73,86 ), we noted a higher death rate than expected. This reduced our final sample size, and possibly skewed the results obtained in some comparisons, mainly regarding sexes. Finally, several ecological parameters can also influence animal spatial behavior (for instance density, prey and shelter availability, degree of anthropization, habitat type 12,35,87 ). We are however highly confident that our results provide robust knowledge on home Table 2. AKDE (Autocorrelated Kernel Density Estimators) 95% home range areas (ha) per individual showing the lower and upper confidence intervals (CI) at 95%. We also indicate sex, snout-vent length (SVL) (cm), number of data points, effective sample size (ESS; parameter based on the number of range crossings occurred during the study period) and the best fitted movement model per individual. Movement models are OU: correlated positions but uncorrelated velocities, OUF: correlated positions and correlated velocities, isotropic: movement in all directions (circular home range) and anisotropic: movement vary by direction (non-circular home range) 103,104 . OUf is a particular case of OUF where autocorrelated positions and velocities cannot be distinguished 103,104 . Discarded individuals are noted. We do not show information on dead 310 and 550, as we discarded them from all analyses.  www.nature.com/scientificreports/ range, phenology and activity patterns of L. californiae, which can be used to improve control action designs for the whole island of Gran Canaria and possibly elsewhere. Technological progress is still needed to facilitate the acquisition of reliable spatial ecology data for small, low-mobile and secretive snakes. GPS-based techniques for the study of animal movement have greatly advanced in recent years 88 , including for invasive snakes 89 . Nevertheless, this technology is still difficult to apply to cases like ours that involve a fossorial, less mobile, and small-bodied species 89,90 . To accumulate enough effective sample size for this animal, small but long-duration batteries are needed, a trade-off that still remains defying (see Mitchell et al. 91 ). Against this backdrop, the combined use of radiotelemetry and the novel AKDE home range estimation method allowed us to mitigate common telemetry data issues (e.g., irregular sampling regimes, data autocorrelation) and those deriving from snake behavior (e.g., spatial autocorrelation, small effective sample size), without compromising data accuracy. Therefore, this combination is a promising tool to unveil the spatial ecology of other secretive species whose studies may encounter similar methodological difficulties arising from their particular behavior 92,93 .
From a broader perspective, our study contributes to highlight the advantages of gaining information on spatial behavior in the management of invasive species. Spatial ecology studies have enabled managers to understand   11,65 , can be also useful to plan when and where to place traps 94 and conduct visual surveys 10 , determine sources of detection bias 15 , identify areas at risk 95 or manage local habitats to prevent spread of invasive species 96 . Therefore, we argue that efforts should be made to turn spatial behavior information into an essential tool for optimizing invasive species management.

Conclusions
The high ecological adaptability of L. californiae, its secretive behavior and unique propensity to survive 28 , its reproductive plasticity 97 , devastating ecological impacts 32,98 , and the climatic suitability of the archipelago 99 makes the strengthening of actions to control this invasion an urgent matter for Gran Canaria. Our research provides basic and applicable insights into the spatial ecology of L. californiae that can be directly incorporated into trapping and control action designs. In particular, traps should not lie more than 233 m apart and containment buffers should be less than 100 m. Moreover, control action reinforcement should be extended from early February to mid-November to increase captures (particularly of females), that is, before the breeding season starts and until the end of the activity period. Since L. californiae individuals move every 2-3 days and most movements are smaller than 100 m, intense prospection in the surroundings of a detected but uncaptured individual or of fresh tracks could increase the probability of capture. To link visual surveys to the species activity patterns, the former should be made during central hours and evenings. In addition, this study underlines the value of spatial ecology in the context of biological invasions. Such an approach can be an essential step in designing more effective control strategies, especially on other islands worldwide (e.g., Cozumel, Christmas or the Balearic Islands 19,100-102 ). Moreover, the combined use of AKDE method and radio tracking for less mobile small species like L. californiae is very appropriate. For a broader perspective, we appeal to the need to innovate and develop new technologies to improve management of invasive snakes, given their devastating impacts on many islands of the world.

Data availability
The datasets used and/or analyzed during this study are available from the corresponding author on reasonable request.