Study area
The study area is located in the north-eastern Italian Alps (Argentario range, in Val di Cembra and Valsugana; Autonomous Province of Trento), covers c. 16 km2 and ranges between 500 and 1,000 m a.s.l. The topography is generally mild, but steeper slopes (> 30°) occur in the northern portion. The climate is continental and characterized by a mean temperature of 1.0 °C in January and 21.0 °C in July, and a mean annual rainfall of 966 mm (average 2000–2018; https://www.meteotrentino.it). There is occasional snow cover between December and March, although the soil is mostly frozen at night. The study area is covered by 80.0% forest, mostly as relatively homogeneous secondary growth stands interspersed with small pastures. The forests are dominated by Pinus sylvestris with abundant shrub undergrowth, and by mixed stands of Fagus sylvatica, Picea abies and Abies alba and, to a lower extent, by Quercus petraea stands.
Roe deer is the primary large herbivore in the study area (6–9 individuals km−2; ref. values from Autonomous Province of Trento Wildlife Office). Adult roe deer do not have natural predators in this landscape, but young fawns may be predated by red fox (Vulpes vulpes). The fine-scale food selection of roe deer in the Alps has been described as mainly dependent on shrubs or regeneration of tree species as well as a diversity of herbaceous plants from the undergrowth from spring to fall, switching between items according to the temporal trends of availability48. In the winter time, roe deer strongly select for forested environments and opportunistically for supplemental food where available22.
Supplemental feeding management of roe deer is conducted at > 50 distinct feeding sites within the study area (FS; Supplementary Information S1: Fig. S1) and authorized year-round within a larger zone of c. 45 km2 (official authorization: “Autonomous Province of Trento order n. 2852/2013”). FS are typically shaped as wooden hopper dispensers that provide a continuous supply of corn accessible through a tray (Fig. 1). They have been deployed and provided continuously with food (at least in fall and winter) for many years (i.e., for longer that the average lifespan of roe deer in our study area). They are managed by private hunters for roe deer but are also attended sporadically by red deer (Cervus elaphus), as well as non-target mammals (Meles meles, Sciurus vulgaris, Apodemus sp., Microtus sp.) and birds (Garrulus glandarius, Columba palumbus).
Experimental design
We took advantage of roe deer use of a focal, identifiable resource—the FS—to design an in situ experimental manipulation of resource availability. We created three successive experimental phases based on the availability of this resource—pre-closure, closure and post-closure—by physically managing the accessibility of food at the FS. During the closure phase, access to forage at FS was transitorily restricted by placing wooden boards obstructing the tray; boards were then removed again in the post-closure phase (Fig. 1).
The experiment was conducted between January and April, when the use of high-nutritional supplemental feed (i.e., corn) by roe deer is the most intense17, for three consecutive winters (2017, 2018 and 2019). We implemented the experiment on 18 individuals, of which seven could be manipulated in two consecutive years—five individuals were recaptured and two collar deployments spanned two winters—leading to a total of 25 individual winter trajectories i.e., “animal-years” (21 adults: 15 females, 6 males; 4 yearlings: 2 females, 2 males; sample size n = 4, 11 and 10 in 2017, 2018 and 2019 respectively; see Supplementary Information S1 for details). Because roe deer captures at middle to low density in Alpine, heavily forested environments are rare events that have to rely on low-efficiency techniques such as box traps and because we had to account for stakeholder acceptance, repeating the experiment on single individuals in consecutive years allowed us to take full advantage of our sample.
Roe deer were captured using baited box traps (n = 21 capture events) or net drives (n = 2), and were fitted with GPS-GSM radio collars programmed to acquire hourly GPS locations for a year, after which they were released via a drop-off mechanism. Captures and marking were performed complying with ethical and welfare rules, under authorization of the Wildlife Committee of the Autonomous Province of Trento (Resolution of the Provincial Government n. 602, under approval of the Wildlife Committee of 20/09/2011, and successive integration approved on the 23/04/2015); all methods and experiments were carried out in accordance with the relevant guidelines and regulations. Radio-collared roe deer moved an average of 61.2 m per hour. This value of the average hourly movement distance (l) was subsequently utilized in the analyses described below.
For all captured animals, we assumed a post-capture response in ranging behaviour. We therefore considered the first re-visitation of the capture location as a likely sign of resettlement in the original range and we used this time as onset of the experimental pre-closure phase. Although not all the individuals were manipulated at the same time, we avoided interference between capture operations and FS manipulations, and between co-occurring different manipulation phases (i.e., ensuring that co-occurring manipulations occurred in separate areas).
During the pre-closure phase, we ensured a continuous supply of food at all managed FS—i.e., that were provisioned at least once in the month prior to the experiment—located within 500 m of each roe deer locations (known through twice-daily download of GSM-transmitted GPS relocations). At the end of the pre-closure phase, we identified the “manipulated” FS (M) for each individual as the managed FS with the largest number of locations within a radius (l) during this initial phase, and considered it as the FS to which an individual is most familiar. All other managed FS were considered as “alternate” (A) FS. During the closure phase, corn was made inaccessible at M for a duration of approximately 15 days, depending on personnel availabilities (min = 14.0 days, max = 18.1, mean = 15.5). M was then re-opened, thereby initiating the post-closure phase. During both pre- and post-closure phases, corn was available ad libitum at M. All A FS had corn available ad libitum throughout the duration of the experiment. To ensure a continuous supply of food during the experiment, field personnel visited and replenished the FS every third day. Across the experimental manipulations, we used a total of twelve distinct FS as M, and 23 distinct FS as A (mean = 4.04 A sites per animal-year, (sigma hspace{0.17em})= 1.43; of these, an average of 1.76, (sigma hspace{0.17em})= 1.13, were actually used by roe deer; see Supplementary Information S1: Table S1 for details on the identity of M and A for all animal-years). M sites were separate from A sites by an average distance of 702.5 m ((sigma hspace{0.17em})= 310.5), and M and used A sites by an averaged distance of 567.5 m ((sigma hspace{0.17em})= 235.7).
Data preparation
To ensure meaningful comparisons between animal-years, we homogenized the durations of each experimental phase to the minimum length of the closure phase in our sample (i.e., 14 days). Specifically, we truncated the movement data by removing initial excess positions for the pre-closure and closure phases, and terminal excess positions for the post-closure phase. GPS acquisition success was extremely high (99.57% during the experiment) and we did not interpolate missing fixes in the collected data.
The analyses of space-use and movement behaviour were based on spatially-explicit, raw movement trajectories. The analyses of resource use, instead, relied on spatially-implicit, state time series derived from the underlying movement data. To this end, we created an initial time series, for each animal-year, by intersecting the relocations with three spatial domains: vegetation (the matrix; V), manipulated FS (M) and alternate FS (A). We converted FS locations (M and A) into areas by buffering them. To investigate the sensitivity of buffer choice we considered six buffer sizes: l (i.e., 61.2 m) multiplied by 0.5, 1, 1.5, 2, 3 and 4. We associated all locations falling outside M and A to the state V. The three-state time series was then converted into three single-state presence/absence time series.
Preference for feeding sites
We calculated each individual’s preference for FS (({h}_{FS})) as the relative use of FS over natural vegetation during the pre-closure phase (i.e., the proportion of GPS fixes classified as either M or A). Because preference is considered to be temporally dynamic37, we chose to evaluate ({h}_{FS}) for each year separately in case individuals were manipulated in two separate winters. This reasoning allowed to account for the influence of individual condition and of the relative quality and quantity of vegetation resources on ({h}_{FS}). We included ({h}_{FS}) in all space-use, movement, and resource use analyses described below.
The variability of ({h}_{FS}) across animal-years was maximal when FS attendance was defined as a GPS location within a distance equal to the population mean hourly step length (l) i.e., 61.2 m from the FS (interquartile range = 0.278, mean = 0.343; Supplementary Information S2: Table S1). Accordingly, the results described below are based on this definition (see Supplementary Information S6 for a sensitivity analysis). At this scale, ({h}_{FS}) did not differ consistently between sex (mean for females = 0.346; mean for males = 0.336; t-test: p value = 0.901).
Analysis
We analysed how the experimental manipulation, and its interaction with both preference for FS and sex, affected roe deer space-use, movement behaviour, and resource use.
General modelling approach
We analysed the roe deer responses to the experiment using mixed effect models. The final fixed-effect structure was developed progressively, beginning with simple formulations and evaluating the consistency of our results to ascertain that our data could support more complex formulations. For example, regarding the analysis of home range size, we first fitted a simple function of the experimental phase (i.e., home range size ~ Phase), then evaluated a potential additive effect of preference for feeding sites (i.e., home range size ~ Phase + ({h}_{FS})), and then an interaction between the two covariates (i.e., home range size ~ Phase + ({h}_{FS})+ Phase:({h}_{FS})). We repeated this procedure when evaluating the effects of sex, and eventually, assessed the full fixed effect structure. We did not find irregularities in the behaviour of the nested models (i.e., marked changes in absolute parameter values or sign). In the full model, fixed effect terms were dropped when statistically non-significant (p value > 0.05). We considered “animal-year” as the sampling unit to account for the fact that an individual may respond independently to manipulations in different years. The choice of an “animal-year” random effect (instead of an “animal” random effect) did not qualitatively affect our results (Supplementary Information S8).
Space-use
We assessed the changes of home range and core area sizes (P1.1), and space-use overlap (P1.2, P3.1) between experimental phases. We calculated utilization distributions (UD)49 for each animal-year and experimental phase using a Gaussian kernel density estimation. After visual inspection, we chose to compute the UDs at a spatial resolution of 10 m and with a fixed bandwidth set to half the average hourly movement distance (i.e., l/2 = 30.6 m).
For home range and core area sizes, we calculated the area (in hectares) corresponding to the 95% and 50% UD contours, respectively, during each experimental phase (Phase; three levels; reference level: Pre-closure). We then analysed the log-transformed areas using a linear mixed-effect model (LMM) with five fixed effects: Phase, ({h}_{FS}), Sex (categorical predictor; reference level: Female), and two interaction terms (Phase:({h}_{FS}) and Phase:Sex). We included animal-year (ind) as random intercept.
We estimated the space-use overlaps for three pairs of UDs—pre- and post-closure, pre-closure and closure, and closure and post-closure (Contrast; three levels; reference level: Pre-/Closure)—using the volume of intersection statistic (VI)50. VI ranges from 0 (no overlap) to 1 (complete overlap). We analysed the logit-transformed overlaps using an LMM with Contrast, hFS, Sex, Contrast:hFS and Contrast:Sex as fixed effects, and ind as random intercept.
Movement behaviour
We investigated the movement responses of roe deer to the experiment (P1.3) by analysing the changes in hourly step length (Euclidean distance between two successive relocations) and turning angle ({theta }_{t}) (angle between two successive movement steps). We analysed the log-transformed step length, ({s}_{t}) and, because turning angles range between (-pi) and (pi), and were symmetric around 0, the logit-transformed absolute turning angle, ({varphi }_{t}=logleft(frac{left|{theta }_{t}right|}{1-left|{theta }_{t}right|}right)). We used LMMs with Phase, ({h}_{FS}), Sex, Phase:hFS and Phase:Sex as fixed effects, and ind as random intercept. Because step length was characterized by strong serial autocorrelation at short temporal lags and at circadian periodicities (a common pattern in animal movement trajectories51), we also included step length measured at lags 1, 2 and 24 h (i.e., ({s}_{t-1},{s}_{t-2}),({s}_{t-24})) as fixed effects to reduce the autocorrelation of the model residuals.
Resource use
To test whether the experiment led to a transitory change in resource use (P1.4a–b, P3.2), we fitted separate mixed-effect logistic regression models to the three single-state presence/absence time series (({u}_{M,t}), ({u}_{A,t}) and ({u}_{V,t})) using Phase, ({h}_{FS}), Sex, Phase:({h}_{FS}) and Phase:Sex as fixed effects, and ind as random intercept. The pre-closure level for Phase was dropped for ({u}_{V}) to avoid circularity (({h}_{FS}=1-{{stackrel{-}{u}}_{V,t}}_{Pre-closure})). We also included the response variables measured at lags 1, 2 and 24 h (e.g., ({u}_{M,t-1},{u}_{M,t-2}),({u}_{M,t-24})) as fixed effects to reduce the autocorrelation of the model residuals. However, for the sake of conciseness and clarity, we omitted these response lags when visualizing resource use predictions. Because the model results were consistent regardless of the inclusion of the response lags (Supplementary Information S5: Tables S1, S2), this decision had no impact on the interpretation. Two animal-years were excluded from the analyses of resource use due to the absence of suitable A-state: F4-2017 did not seem to have visited any alternate FS (A) prior to the experiment; and F16-2016 had two distinct, highly-used FS during pre-closure, but only the second most visited FS could be manipulated (due to stakeholder acceptance). While the use of A was more variable when including these two outliers, the general patterns remained unchanged (Supplementary Information S5: Tables S1, S3).
Software
All analyses were conducted in the R environment52. We used the packages adehabitatLT and adehabitatHR53 for the spatial analyses, fitted all mixed-effect models via Maximum Likelihood with the package lme454. We obtained the p-values for the fixed effects using afex55 and coefficients of determination using MuMin56.
Ethical statement
All experimental protocols and data collection were approved by the Wildlife Committee of the Autonomous Province of Trento (Resolution of the Provincial Government n. 602, under approval of the Wildlife Committee of 20/09/2011, and successive integration approved on the 23/04/2015). All experiments and methods were performed in accordance with relevant guideline and regulations.
Source: Ecology - nature.com
