Study area
We performed this study at the San Martín Experimental Forest (39°41′S, 73°18′W), an 80-ha land owned by Universidad Austral de Chile, located 80 km north of Valdivia (southern Chile). Ambient temperatures range from 18 ± 1 °C in the austral summer (February) and 7 ± 1 °C in the winter (July). Mean annual precipitation is 2205 mm (ranging between 1500 and 3000 mm), concentrated in the austral fall (April–June) and winter (June–September) [20]. San Martín is a relict fragment of the Valdivian temperate rainforest [21], an ecosystem considered to be biologically unique and critically endangered [22, 23]. This experimental forest is a secondary-growth stand dominated by the evergreen trees Gevuina avellana, Drimys winteri, Laurelia sempervirens, Embothrium coccineum, Luma apiculata, Amomyrtus meli, altogether with sparse old Nothofagus spp. trees and the native bamboo Chusquea valdiviensis [24].
Data collection
We captured D. gliroides individuals at the study site using custom-made, single-door, Tomahawk-like wire-mesh live traps (26 × 13 × 13 cm). Traps were placed 1–2 m above ground level and baited with fresh banana slices [see 25, 26]. We set 144 traps in a web-style trapping grid, consisting in 12 radiating lines of 12 traps at 10 m intervals [27]. Trapping grids were monthly operated over ten consecutive days each month from December 2008 to April 2009 (total trapping effort = 3600 trapnights), avoiding full moon nights. Traps were checked daily at dawn and animals were released at the capture site after processing. We recorded body mass (using a Pesola® spring scale), standard morphometric measurements (total length, tail length, and nose tip to tail base length), sex (based on a visual inspection of the reproductive organs), and age (adult or sub-adult) for each captured individual. To differentiate adults from sub-adults we used the criteria of Salazar and Fontúrbel [16], being adults those individuals with body mass > 24 g and nose tip-to-tail length larger than tail length, and sub-adults those individuals with body mass ≤ 24 g and nose tip-to-tail length shorter than tail length. Capturing and handling procedures followed the guidelines of the American Society of Mammalogists [Sikes et al. 28], and were authorized by the Chilean Agriculture and Livestock Bureau (SAG; permits 161–2006 and 146–2008 granted to LMF).
During the austral summer (January to March) of 2009, 18 D. gliroides individuals were captured. Twelve of those individuals (three adult and one sub-adult males, and two adults and six sub-adults females) were fitted with telemetry transmitters (Biotrack® United Kingdom, cable-tie collars of 0.9 g, representing ≤5% of the body mass) and intensively tracked from dusk to dawn on a daily basis between 23-Jan and 12-Apr (we have not tracked individuals during rainy nights). Two teams (communicated with handy radios) simultaneously located each individual and registered the geographic coordinates (i.e., GPS position), bearing, time and signal strength to minimize data collection error. Noise and lights were kept to minimum to avoid disturbing the tracked animals. Before taking animal locations we conducted a field test, in which one person placed transmitters with known locations and then were tracked by other people (unaware of the real location) to assess estimation error, which was below 5 m.
Data analysis
We estimated actual locations of the tracked individuals by grouping bearings taken in the field in groups of three or four bearings taken within a 10-min interval, and separated at least by 20° [although we privileged groups with separations between 45° and 135° to reduce triangulation error; 29, 30]. We estimated locations using the software LOAS 4.0 (Ecological Software Solutions, Switzerland), considering a minimum of 30 min between relocations of the same individual to ensure data independence. Error ellipses average values ranged from 0.01 to 0.06 ha. Using the estimated locations, we calculated core areas and home range areas of the tracked D. gliroides individuals using the 50 and 90% kernel density estimators (KDE hereafter), respectively [15]. Core and home range areas were obtained with the adehabitatHR package [31] in R 3.5.0 [32], using a least square cross validation (LSCV) method for defining kernel bandwidth. In order to assess the robustness of the home range and core areas estimated in function of the number of telemetry locations used, we took the individual with the largest number of locations (individual M01 with 119 locations), and we made 45 location files with random subsets of 5 to 50 locations. Then we estimated home range and core areas for each location subset and compared them with the real values.
We compared body mass between males and females and between adults and sub-adults using generalized linear models (GLM) with a Gaussian error distribution. Likewise, we compared KDE home range and core area sizes between males and females, as well as between adults and sub-adults using GLMs. Then, we fitted additional GLM models to assess the relationship between home range and core area sizes with body mass and the number of locations; we assessed all individuals together first, and then we assessed males and females, and adults and sub-adults separately. Normality and variance structure were checked for each model fitted [33]. We also assessed the extent of overlap in core areas (i.e., KDE 50%) between adult males and females, and between adult and sub-adult individuals, using Cartographica 1.4.9 (ClueTrust, Reston VA) GIS software. For this purpose, we merged individual core areas for each group (i.e., males vs. females, adults vs. sub-adults) before conducting the overlap analysis to focus the comparison on the groups and not on the individuals.
Using estimated locations, we then calculated the distance between successive points within a 120-min period as a proxy of movement behavior. We considered three different activity time periods [following 19; based on camera-trap records]: (a) early activity from 19:00 to 01:00 h, (b) activity peak from 01:01 to 03:00 h, and (c) late activity from 03:01 to 07:00 h. Then, we estimated individual movement speed as the ratio between distance and time differences between consecutive locations (expressed in meters moved per minute). In order to determine if movement behavior changes in function of the activity period (early, peak and late) we performed a mixed-effects generalized linear model (GLMM) with a Gaussian error distribution, using movement speed as a response variable, the activity period as a fixed factor, and the tracked individual id as a random factor to account for inter-individual variability [34]. GLMM parameters and their significance were estimated using restricted maximum likelihood (REML) t-tests, with a Kenward-Roger approximation to degrees of freedom [35]. Values are presented as mean ± 1 SE unless otherwise indicated. GLM and GLMM results in the main text are presented using t- and P-values obtained, while estimates for each variable and their standard error are presented in the Supporting Information available online. GLMs and GLMMs were implemented in R 3.5.0 [32] using the packages lme4 [36], lmerTest [37], and pbkrtest [35], illustrations were developed using ggplot2 [38].