Introduction

Recently efforts have been made to use multi-scale habitat assessments to determine the best scale at which different environmental factors impact animals across landscapes (Mayor et al. 2009; McGarigal et al. 2016). Among the numerous analytical methods developed to study multi-scale habitat selection, resource selection functions (RSFs) obtained through generalized linear models (GLMs) prevail in the literature (McGarigal et al. 2016). RSFs can be subdivided into point, step, path, and home range, depending on the data type collected (e.g., transects vs. telemetry) (Zeller et al. 2012). Point selection functions (PSFs) apply to data gathered at point locations and compare habitat features between where an individual was observed vs. absent, or with random sites generated to represent habitat availability (Zeller et al. 2012; Dias et al. 2017). This method is suited to assess habitat selection—especially at the local scale—for resting, feeding and latrine sites (Zeller et al. 2012; Carvalho et al. 2014). Animals make daily choices to optimize their foraging, marking, sleeping, and breeding activities. Dispersal may also occur at least once in their lifetime and can take several months or even years. These daily and seasonal activities entail habitat selection at different scales (Mayor et al. 2009). While choosing a place to rest or scent mark occurs at the local scale, mating and dispersing imply choices at the home range and regional scales, respectively (e.g., Johnson 1980). The importance of scale is even more evident for species occurring in the same area (sympatric) and habitat (syntopic), thus likely competing for local resources like food, resting and marking sites (Carvalho et al. 2014; Buesching and Jordan 2022).

Latrines are microsites that are repeatedly used by animals to defecate, urinate, and deposit several glandular secretions (Buesching and Jordan 2022). Therefore, they represent efficient communication hubs, as the scent marks deposited provide information about the health and social and reproductive status of individuals, which stabilizes social structures and ensures reproductive success (Barrientos 2006; Espírito-Santo et al. 2007). The use of latrines is crucial in carnivores due to their sedentary lifestyle, often large home ranges, and low densities, where the chances of encounter with conspecifics are limited (Darden et al. 2008; Buesching and Jordan 2022). Latrines are frequently located at conspicuous sites spread across landscapes and have been reported in carnivores such as ocelots (Leopardus pardalis, King et al. 2017), northern raccoons (Procyon lotor, Hirsch et al. 2014) and swift foxes (Vulpes velox, Darden et al. 2008). Latrines vary in amount of faeces deposited, location and spatial distribution (Palomares 1993; Green et al. 2015). Some species place their latrines at territorial boundaries or inside core areas (Balestrieri et al. 2011; Hirsch et al. 2014). Others like the social European badgers (Meles meles) may locally use a hinterland approach, whereby they disperse their faeces throughout their entire home range (Roper et al. 1993; Kilshaw et al. 2009).

Among small carnivores, genets (Genetta spp.; Viverridae) are solitary, territorial and live at low densities (Carvalho et al. 2014, 2018), thus encounters between individuals are scarce and limited mostly to boundary areas (Espírito-Santo et al. 2007). Therefore, indirect scent communication in genets is paramount, and so they use latrines regularly (Livet and Roeder 1987; Roberts et al. 2007). Here, we focused on common small-spotted genets (Genetta genetta) and Cape genets (Genetta tigrina), which live in sympatry across the southwestern coast of South Africa (Carvalho et al. 2016b; Widdows et al. 2016). They are of similar size, have similar diet and activity patterns, and share latrines at home range borders (Carvalho et al. 2016b; Widdows et al. 2016). So far, only one study by Espírito-Santo et al. (2007) investigated the factors affecting the placement of common genet latrine sites in Portugal, although the role of scale was not discussed. Other studies focused on the defecation rate at latrines by the same species in Spain (Barrientos 2006) or highlighted their importance in dietary studies on Cape genets (Roberts et al. 2007), but latrine site selection per se was not investigated.

Hence, most research conducted to date on carnivore latrines focused on their social and communication significance, year-round intensity of use (temporal aspect) and secondarily as a source of scats for dietary studies. Yet, we do not know how small carnivores—and particularly genets—select latrine sites. Moreover, very little attention has been focused on which habitat features play a role during latrine site selection. Failing to describe an optimal scale may lead to poor model performance and ultimately to the conservation of wrong habitat features (McGarigal et al. 2016). However, despite the cognizance that scale plays a preponderant role in habitat selection, many studies still do not integrate multi-scale approaches (McGarigal et al. 2016).

To fill this gap, our work is the first using multi-scale modelling to investigate latrine site selection by sympatric carnivores. The main goal was to determine what factors drive latrine site selection at different spatial scales: (1) fine scale (20 m plot); (2) landscape scale (100 and 500 m); and finally, (3) scale-independent (distance to crucial features). To achieve that, we used conditional (match-control design) logistic regression to compare latrine sites (observed) vs. random locations (Zeller et al. 2012; Dias et al. 2017). We used genets as proxy for small carnivores, as done elsewhere (Carvalho et al. 2014, 2016a, 2018). Moreover, detailed information on home range size, daily movements, resting locations, and latrine sites has been consistently gathered in our study area. This information enables a more effective application of a multi-scale analysis framework.

We hypothesised that genets select latrine sites that: (i) maximize the spread, longevity and/or detection of scents to communicate efficiently with conspecifics (Roeder 1980; Barrientos 2006); (ii) offer protection against predators (Popp et al. 2007; Carvalho et al. 2014); and (iii) are located near food and resting sites for an effective foraging strategy (Palomares 1993; Espírito-Santo et al. 2007). We therefore predicted that, at the fine (micro-habitat) scale, genets select sites that (i) are elevated to efficiently spread scent (Prediction 1a) or (ii) are well protected from harsh weather (e.g., heavy rains) and animals that feed on scat contents or trample latrines, hence ensuring a higher scent longevity (Prediction 1b) (Larivière and Calzada 2001; Rodgers et al. 2015; King et al. 2017). At the fine scale (latrine use) and landscape scale (movements to and from the latrine), we expected genets to place latrines where their use is safe due to the presence of shrub and tree cover that reduces predation risk (Prediction 2) (Popp et al. 2007; Carvalho et al. 2014). We also predicted that genet latrines would be predominantly found in riverine corridors (Prediction 3) (Palomares 1993; Espírito-Santo et al. 2007). They have been described as natural highways for many species (Shirk et al. 2014; Carvalho et al. 2016a), but also boast a high amount of food resources and micro-habitats favourable for resting (Buesching and Jordan 2022). Finally, we also anticipated genet latrines to respond to important scale-independent elements in the landscape, such as the proximity of (i) other latrines to allow marking even if some are destroyed (Prediction 4a) (Espírito-Santo et al. 2007; Buesching and Jordan 2022); (ii) dirt roads where marking behaviour along their edges has been recorded (Carvalho et al. 2011), which may increase the chances of scent-mark detection due to frequent animal passages (Prediction 4b) (Buesching and Jordan 2022); and (iii) water sources as a crucial element in semi-arid landscapes (Prediction 4c) (Carvalho et al. 2016a). Lack of suitable latrine sites in landscapes may disrupt genet olfactory communication, but also the access to a territory that provides food, resting and breeding sites, and mates (Buesching and Jordan 2022). This, in turn, will ultimately compromise their survival. Our models predict the location of latrine clusters and the scale at which these studied factors operate. These results provide insightful guidelines for management practices that will help maintain latrine availability across landscapes.

Materials and methods

Study area

This study was conducted at the Great Fish River Nature Reserve (GFRNR) in the Eastern Cape Province of South Africa (Fig. 1a). It is located between 33°04′ and 33°09′ S and 26°37′ and 26°49′ E. The reserve covers a total area of 45 500 ha, and it falls within the Albany Thicket Biome, with the Great Fish Thicket being the dominant vegetation type (Hoare et al. 2006). The vegetation at GFRNR is dominated by very dense and impenetrable thicket ca. 2 m high with characteristic evergreen sclerophyllous shrub species such as Portulacaria afra (Jacq.) and bushclumps composed of Rhus spp., Scutia myrtina (Kurz.) and numerous other tall bush species, separated by vast patches of karoo shrublets that are < 50 cm in height (Evans et al. 1997). The climate is semi-arid, with warm summers (December–February), when maximum daily temperatures often exceed 35 °C (mean monthly maximum of 39.6 °C in January). Winters (June–August) are cold, with minimum night-time temperatures often below 0 °C (mean monthly minimum of 2.1 °C in July) (Hoare et al. 2006). Rainfall is variable, with peaks generally occurring in October and March. The mean annual rainfall is 435 mm.

Fig. 1
figure 1

Study area showing: a location of the Great Fish River Nature Reserve (GFRNR) and the different latrines found; b example of walking transects to search for latrines in a riverine corridor and the surrounding habitat; c a main latrine and two generated random points used in the landscape scale analysis (100 m and 500 m buffers); d main latrine and two generated random points used in the fine scale analysis (20 m buffer)

Latrine surveys and monitoring

Parallel line transect walk surveys were carried out between May and August 2017 by a team of three field biologists (one was a small carnivore expert), 10–20 m apart from each other (Fig. 1b), following the method described by Stewart et al. (2002). Different structures favoured by genets to place their latrines were searched, such as burrows, tree hollows, termite mounds, etc. Based on previous studies, most latrines were found near riverine areas (Do Linh San, pers. obs. 2005–2016; Roberts et al. 2007). Therefore, we initiated our surveys using riverine corridors as the baseline reference. Nonetheless, to avoid any sort of bias toward riverine corridors, similar line transects were conducted while covering a similar distance—whenever possible—through the other main habitats (Fig. 1b). Additionally, the habitats outside riverine areas were also sampled by line transects and genet latrines were recorded while monitoring resting sites used by radio-collared genets for a complementary project (F. Carvalho, pers. obs. 2017–2019). A site was deemed as a latrine site if five or more genet scats appeared in the area (1.06 ± 2.01 m2 on average (± SD), n = 53), following recommendations by Espírito-Santo et al. (2007). However, the presence of temporary defecation sites (with < five scats) was also recorded. Genet scats were identified by examining their location, shape, size and colour following descriptions by Roberts et al. (2007), as well as their characteristic smell compared to that of faeces of other small carnivores occurring in the reserve. The coordinates (accuracy: ± 4–6 m) of all latrines and temporary defecation sites were obtained with a GPS device (Model “etrex”, Garmin, Olathe, Kansas, USA).

Latrines were then monitored over a period of 1 year (from September 2017 to August 2018) for side dietary and behavioural studies using camera traps and involving monthly collection of scats. We kept our impact residual (a couple of minutes on average) while monitoring the latrines, always during daytime. The collection of the scats was done wearing gloves and only latrines that had fresh scats were handled. Following methods described for European badger and common genet latrines (Roper et al. 1986; Espírito-Santo et al. 2007), we grouped genet latrines into three distinct categories based on their intensity of use over the 1-year period. Sites used less than 3 months were considered temporary defecation sites. Secondary latrines were used during 4–7 months, while main latrines were used for more than 8 months. We pooled the last two categories, since their features were similar, and their use was much higher than that of the temporary defecation sites.

Fine scale, landscape scale, and scale-independent variables

We considered 29 predictors related to land cover, landscape configuration, local vegetation, topography, and human disturbance (Table 1). These were selected based on previous studies describing the selection of latrine and resting sites, habitat, and movements patterns by genets (e.g., Palomares and Delibes 1994; Espírito-Santo et al. 2007; Roberts et al. 2007; Galantinho and Mira 2009; Camps and Alldredge 2013; Carvalho et al. 2011, 2014, 2016a, b, 2018). Genets, like other small carnivores, perceive their surroundings at different scales, which reflects the constant decisions they make while foraging and the interrelations between landscape and local habitat characteristics (Borthagaray et al. 2014). Accordingly, to describe genet latrine sites, several scales were selected following the work of Johnson (1980). For the fine scale (4th order, i.e., specific resources within a habitat patch), we used a buffer of 20 m around each latrine and two randomly generated, unused sites (Fig. 1d). The chosen radius was based on previous studies on genet resting sites (Carvalho et al. 2014). To reflect genet core and home range areas at the landscape scale (3rd and 2nd orders, i.e., specific habitats inside the home range and the full home range), we used buffers of 100 and 500 m, respectively (Espírito-Santo et al. 2007; Carvalho et al. 2014). Finally, we also characterized genet latrines (and random sites) by considering their distance (scale-independent, including partially 2nd and 1st orders) to focal landscape features such as dirt roads, habitat ecotones, water resources, main latrines and temporary marking points (Carvalho et al. 2016a).

Table 1 List and description of explanatory (independent) variables for each scale used to model the probability of latrine site selection vs. random sites by sympatric genets

The fine scale variables were recorded by two observers using visual estimation. We visited the main latrines and random sites during five field sessions of 3–5 days each during the winter months (June and July) of 2018 to determine latrine characteristics and record micro-habitat variables (Table 1, Table S1, Fig. 1a). Two random points were generated, using the software QGIS 3.16 Hannover (QGIS Development Team 2020) (details in Fig. 2), in the surroundings of the latrine site but constrained by a minimal distance of 20 m between them (Fig. 1d). The number of random points and the distance among them were a compromise between adequately sampling the available area and the need to limit overlap (< 20%) among sampling plots (Fig. 1d) (Carvalho et al. 2016a; Dias et al. 2017). We recorded the type of latrine structures (e.g., burrow, hollow tree, termite mound, rock, etc.; Fig. 3). When relevant, an Abney level was used to measure the inclination of both the latrine entrance and the slope of surrounding area, while a compass (Model DP6, Recta AG, Cham, Switzerland) was used to measure the orientation of the entrance and the slope direction (aspect). Using a similar approach to Lagesse and Thondhlana (2016), vegetation cover was assessed by visual estimation inside four quadrants of the plot (20 m buffer) surrounding the latrine (or random site); the overall cover (%) was then calculated as the average of the four independent values. For example, understory cover has been described as inversely proportional to predation risk and thus allows a safe use of latrines, where genets can be concealed while scent marking (Popp et al. 2007; Carvalho et al. 2014). Moreover, genets often use the vegetation strata vertically due to their climbing skills (e.g., Carvalho et al. 2014). Therefore, three vegetation cover heights were considered (Table 1). Vegetation above 1.5 m refers mostly to the canopy (trees) and allows a safe escape from non-climbing predators. To determine the micro-habitat variables such as the height of, and distance to, nearest trees and/or shrubs, we used a 10 m tape for short measurements, while a range finder (Aculon AL11, Nikon) was used for longer distances.

Fig. 2
figure 2

Multi-scale spatial methodological framework including the definition of the response variable (used vs. random); covariates at different spatial scales; and the steps undertaken to build the best average models to describe the probability of latrine site selection by genets

Fig. 3
figure 3

Commonly observed fine scale structures where genet latrines were placed in the GFRNR: a burrow; b ground hole; c termite mound; d rocks; e hollow tree trunk; f hollow tree branch

At the landscape scale, all variables were extracted from GIS layers using QGIS 3.16 Hannover. We considered different landscape scale levels, to account for different ways animals perceive the landscape (Borthagaray et al. 2014; Carvalho et al. 2016a). The 100 m scale was small enough to reflect the immediate surroundings perceived by an animal when following a path. Besides, latrines are not placed at random in landscapes (Espírito-Santo et al. 2007); instead, they tend to be deposited along territory borders, while others cluster inside the genets’ core areas (Stewart et al. 2002; This study). A parallel study on latrine use, through camera traps, confirmed that genets visit several latrines regularly (Xhobani 2024). The 500 m scale was used to reflect the wider area available to animals when making movement decisions. It matches about twice the average movement step length (250.0 ± 19.5 m, G. genetta (n = 1763); 167.0 ± 30.0 m, G. tigrina (n = 1339) between successive locations estimated during the radio-tracking sessions, corresponding to about one hour of foraging activity (Carvalho et al. 2016a; F. Carvalho, unpublished data). This scale also allowed for several latrine clusters inside a genet home range to be included in the used vs. availability context (Espírito-Santo et al. 2007). Wider areas could have been tested to approach the actual average home range size of genets at the study area (MCP95, 375.0 ± 18.8 ha, G. genetta (n = 12); 204.2 ± 30.6 ha, G. tigrina (n = 6)). However, using a 1 km buffer or larger would result in overlaps among latrines and random sites well above 20% and therefore systematically obtaining similar variable values and decreasing models’ accuracy and inference (Barbet-Massin et al. 2012).

The landscape of the study area was characterised using a digital land cover map (2010), which was updated at the 1:10,000 scale using recent aerial Google Maps images (https://www.google.com/maps) and a combination of field checks. We considered a minimum mapping unit of 0.5 ha, and the land cover classes were adapted from the vegetation map developed by Nyamugama and Kakembo (2015). Accordingly, landscape composition at both 100 and 500 m buffers includes the percentage of each land cover class considered (Table 1; Fig. 2). Previous knowledge acquired on genet latrine placement revealed the preference for ecotone-based habitats (Espírito-Santo et al. 2007; Carvalho et al. 2011). Consequently, some edge density metrics associated with forest habitats and dirt roads (Table 1) were obtained inside buffers using the plugin LecoS (Jung 2016) (details in Fig. 2). Topography around sites was assessed through the Topographic Ruggedness Index (TRI), which was calculated using a 12.5 m resolution digital elevation model (https://search.asf.alaska.edu/). This index reflects the variation in three-dimensional orientation of grid cells within a neighbourhood, effectively capturing variability in slope and aspect (terrain heterogeneity) in one measure (Sappington et al. 2007). Finally, distance to potentially important landscape features (scale-independent) that were not assessed through the previous scales were also considered (Table 1). These metrics were used to estimate whether latrines were influenced by the spatial location of point (e.g., water sources, main latrines, and temporary marking points) and linear landscape features (e.g., riverine corridors, habitat edges). Moreover, because latrines do not occur at random, we accounted for spatial autocorrelation by including the distance to the nearest latrines (i.e., scent-marking points) into the models (Espírito-Santo et al. 2007; Carvalho et al. 2014; Dias et al. 2017).

Data analysis

We performed a multi-scale modelling approach using several steps to determine the main descriptors that influence the probability of latrine site selection by genets in the Albany Thicket landscape, as described in detail below and summarised in Fig. 2.

Step 1—match-control design analysis: defining the response variable and data structuring

Analysis followed a match-control design framework (Fig. 2), using a binomial response variable coding a used latrine (1) vs. two random sites (0), thereby creating a group “stratum”. This modelling approach is based on conditional logistic regression to compare multi-scale descriptors (Table 1) around used latrines with random unused sites and has been used in different ecological applications (e.g., Carvalho et al. 2016a; Thorne and Ford 2022). Accordingly, considering different scales, we developed several a priori model sets representing different categories of latrine site features and landscape surroundings. The best landscape predictors influencing latrine site selection by genets between the 100 and 500 m scales were selected based on the lower AICc (Akaike’s Information Criterion, corrected for small sample sizes; Burnham and Anderson 2002), obtained through the univariate scaling analysis for each variable (e.g., Carvalho et al. 2016a).

Step 2—calculating the multi-scale descriptors and pruning analysis

Prior to statistical analyses, descriptors were transformed to approach normality and to reduce the influence of extreme values, using either the angular transformation for proportional variables (%) or the logarithmic transformation otherwise (Legendre and Legendre 1998) (Table 1). All variables were then standardized to zero mean and unit variance, to allow comparability of effect sizes (Schielzeth 2010). Collinearity among variables (including categorical ones) was investigated using generalized variance inflation factors (GVIF), and one variable from each pair correlated at > 3 was discarded from further analyses (Zuur et al. 2009). We retained the variable with the highest biological meaning based on previous studies on genet latrine and habitat selection (Espírito-Santo et al. 2007; Carvalho et al. 2014, 2016a). In addition, for univariate models, only variables that performed better (lower AICc) than the null model (model containing only the group “stratum”) were retained for further steps in our analysis (Holland et al. 2019; Thorne and Ford 2022).

Step 3—selection of the best multivariate model for each scale analysed

We obtained a set of candidate multivariate models representing all possible combinations of variables within each working scale hypothesis. We then ranked the fitted models according to their AICc and we considered as equally suitable those models within 2 AICc units from the best model (AICc < 2) (Burnham and Anderson 2002). When more than one model matched this criterion, we applied a model-averaging procedure (Burnham and Anderson 2002). We then estimated the 95% coefficient confidence intervals (95% CI). For every set of best models, we considered the variables included in those average models whose 95% CI did not include the zero as highly supportive variables (i.e., the impact on the dependent variable was positive or negative) and those that included zero slightly as moderate support (Cumming 2009). Model fit was assessed with the pseudo R-squared of Tjur (2009), and model discrimination ability was assessed with the area under the remote operating characteristic curve (AUC; Fielding and Bell 1997). The autocorrelation in model residuals for each scale was assessed with Moran’s I correlation coefficient, which helps detect spatial patterns in the response variable (Bivand and Wong 2018).

Step 4—best descriptors (marginal effects) of the probability of latrine site selection by genets for each scale

We used the best average model for each different scale analysed, to quantify and plot the marginal effects of the best descriptors with a high and moderate support for the probability of latrine site selection by genets.

Analyses were performed using the “mclogit” package (Elff 2022) for model fitting, the “MuMIn” package (Barton 2022) for model averaging, the “modEvA” package (Barbosa et al. 2022) for R-squared and AUC calculation, and the package “spdep” to check for spatial autocorrelation (Bivand and Wong 2018), all within the R statistical platform, version 3.6.1 (R Core Team 2019).

Results

Characteristics of genet latrines

Overall, 87 latrine sites were surveyed in the GFRNR, including 23 main latrines (26.4%), 30 secondary latrines (34.5%) and 34 temporary defecation sites (39.1%). These latrines were mostly placed in animal burrows (39.1%), followed by ground holes (18.4%), on bare ground (16.1%), rocky substrate (8.0%), termite mounds (9.2%), trees (4.6%), and in anthropogenic structures (4.6%) (Table S2, Fig. 3 for examples). Burrow latrines had a mean (± SD) entrance height of 0.51 ± 0.20 m, a width of 0.57 ± 0.20 m and a depth of 0.76 ± 0.31 m (Table S3). Rocky latrines (crevices (Fig. 3d), rock fragments, rocky ridges) were located at a mean height of 1.46 ± 0.59 m, had a width of 0.56 ± 0.22 m and a depth of 0.63 ± 0.94 m (Table S3). Termite mound latrines (Fig. 3b) had a mean height of 0.70 ± 0.34 m. Owing to irregular shapes and sizes of termite mounds, we presented the average values of both their smaller and larger widths, 0.6 ± 0.28 m, and 0.82 ± 0.18 m, respectively (Table S3). Finally, most bare ground latrines occurred near linear edge habitats, such as riverine corridors and dirt roads.

From the 42 latrines monitored with camera traps, 24 (57.1%) were used only by the common small-spotted genet, while 18 (42.9%) were used by both genet species. No single latrine was used exclusively by the Cape genet. Eleven of the remaining main and secondary latrines had no cameras and therefore the identity of the visiting species remained unknown.

Fine scale predictors of genet latrine site selection probability

Univariate analysis provided strong support (ΔAICc > 10) for significant differences between latrines and random sites for two variables (out of 12) (Table S1). Latrine sites had a higher availability of latrine structures and a higher latrine slope inclination than random sites (Table S1). From the original variables at the fine scale, only three (Latrine structure availability, Latrine slope inclination and Maximum height of nearest bush), with a significant support higher than the null model, were retained in subsequent multivariate analysis (Table S1).

In multivariate analysis, two models were less than two AICc units from the best supported model among 8 candidate models. The average model provided strong support (w + = 1.00) for the positive influence of latrine structure availability on the probability of latrine site selection by genets (Table 2). A positive influence of the latrine slope inclination was also detected, although with a moderate support (w + = 0.73), as the 95% CI just slightly included the zero (Table 2). The Tjur R-squared of this model was 0.35 and the AUC was 0.84. No significant spatial autocorrelation, as assessed by Moran’s I coefficient, was obtained in full model residuals at the fine scale (Table S4).

Table 2 Average models describing the effects of explanatory variables on the selection of latrines sites by genets at different spatial scales: fine scale and landscape scale. Model averaging is based on the confidence set of models at ΔAICc < 2 from the best model. For each variable, we show the standardized regression coefficient (β), unconditional standard error (SE), 95% confidence interval of coefficient estimate (CI), and relative importance (w+)

Landscape scale predictors of genet latrine site selection probability

Overall, support was much stronger for variables measured in 100 m than 500 m buffers, though the opposite was observed for the Topographic Ruggedness Index and the presence of water sources (Table S1). At the 100 m buffer scale, genets selected areas with a lower proportion of Bushclump Karroid Thicket, but with a higher proportion of both Acacia and Combretum riverine forests to place their latrines, when compared to random sites (Table S1). Riverine related forest edge density variables were also significantly higher at latrines compared to random points (Table S1). All scale-independent variables showed a strong support (Table S1). Latrines tended to be much closer to riverine and edge habitats compared to random sites. The same pattern was also obtained for nearest main latrines and temporary defecation sites. Finally, latrines were also placed on average nearer to dirt roads compared to random sites (Table S1).

For the multivariate analysis at the landscape scale, only 4 out of 32 candidate models were at less than two AICc units from the most supported model. The average model showed a strong support (w + = 1.00) for the negative effect of the Bushclump Karroid Thicket cover and the Topographic Ruggedness Index at the 100 m and 500 m scale, respectively, on genet latrine site selection probability (Table 2). In contrast, the average model provided a strong support (w + = 1.00) for the positive influence of Riverine Combretum Forest cover at the 100 m scale on genet latrine site selection probability (Table 2). The average model also showed a moderate support (w + = 0.44) for the positive effect of dirt road density at the 100 m scale, on genet latrine site selection probability, although the 95% CI interval just slightly included the zero (Table 2). The Tjur R-squared of this model was 0.42 and the AUC was 0.80. No significant spatial autocorrelation, as assessed by Moran’s I coefficient, was obtained in full model residuals at the landscape scale (Table S4). For the scale-independent multivariate analysis, only 2 out of the 8 candidate models, were selected for a ΔAICc < 2 from the most supported model. There was a strong support (w + = 1.00) for the negative effects of distance to the nearest riverine corridor on genet latrine site selection probability (Table 2). Moreover, the average model also showed a moderate support (w + = 1.00) for the negative influence of the distance to the nearest main latrine on genet latrine site selection probability, although the 95% CI interval slightly included zero (Table 2). The Tjur R-squared of this model was 0.77 and the AUC was 0.97. No significant spatial autocorrelation, as assessed by Moran’s I coefficient, was obtained in full model residuals at the scale-independent level (Table S4).

Best descriptors (marginal effects) of the probability of latrine site selection by genets

The best average model, developed at the fine scale, indicates that the probability of latrine site selection by genets primarily depends on the availability of latrine-type structures, and a minimum of two latrine-type structures are needed to increase the chance of a latrine placement to 0.8 (Fig. 4a). Secondarily, a latrine access slope inclination above 20° increases to some extent the chances (> 0.5) of those structures being used as latrines by genets (Fig. 4b). For the landscape scale, our best average models provide clear evidence that the probability of latrine site selection decreases considerably with a higher cover of Bushclump Karroid Thicket (> 60%) (Fig. 4c). Genet latrine site selection probability also declines substantially for values of the Topographic Ruggedness Index above 2, and above 6 the chances are almost null (Fig. 4d). Strong positive effects on the probability of latrine site selection were revealed by the proportion of Riverine Combretum Forest cover. Indeed, above just 10% of riverine cover, the chances increase to near 0.9 (Fig. 4e). To a lesser extent, dirt road density at the 100 m scale seemed to increase the probability for a genet to select a latrine in the area; above 25 m of dirt roads, the odds are good (> 0.6) but reach a plateau above 75 m (Fig. 4f). Finally, for the scale-independent best average model, a sharp decrease in genet latrine site selection probability over 50 m away from riverine corridors, was strongly supported (Fig. 4g). A similar pattern, although with a moderate support, was also obtained for the influence of the proximity of main latrines on the probability of latrine site selection (Fig. 4h). Within 100 m from a main latrine, the probability to place another latrine is very high (> 0.8) (Fig. 4h).

Fig. 4
figure 4figure 4

Probability of latrine site selection (with 95% CI) by genets predicted from the average models at different scales: fine scale (a, b); landscape scale including macro-scale (c, d, e, f) and scale-independent (g, h). For scale, the response to each focal explanatory variable was extracted from the model by maintaining all the other independent variables at their median values (marginal effects). Models were fitted on transformed and standardized data but are here depicted on the original scale. *Variables with a moderate support

Discussion

Effect of the spatial scale on the probability of latrine site selection by genets

The selection of latrines by common small-spotted genet and Cape genet was significantly explained by environmental features that acted at different spatial scales. At the fine scale, the availability of micro-structures like burrows, holes, and termite mounds were crucial for the placement of genet latrines. In addition, a higher slope inclination seemed to increase the chances of latrine placement. For the landscape scale, effects were stronger within a 100 m buffer, reinforcing that genets’ decisions are based on their immediate surroundings (e.g., Carvalho et al. 2016a). Only the Topographic Ruggedness Index was significantly stronger inside the 500 m buffer, which might reflect the shortage of burrows and holes made by fossorial animals in rugged areas. Scale-independent predictors confirmed the crucial value of riverine forests as latrine hotspots, supporting the importance of ecotones for small carnivore marking behaviour (Palomares 1993; Buesching and Jordan 2022). We also observed that latrines tend to cluster, allowing genets to keep an area marked regardless of some latrines being destroyed (Barja and List 2006). Our work pioneers the application of a multi-scale framework to assess environmental factors influencing latrine site selection by genets across the landscape, focused on the hierarchical spatial 4th order (Johnson 1980). This approach proved pivotal in uncovering previously overlooked fine-scale predictors, essential for explaining latrine habitat selection.

How are latrine sites selected by sympatric genets?

The environmental factors driving the selection of latrine sites by genets were similar to those for other resources (e.g., resting sites) inside genet home ranges elsewhere (Palomares and Delibes 1994; Espírito-Santo et al. 2007; Carvalho et al. 2014, 2016a, b, 2018). At the fine scale, genets selected conspicuous sites that were either elevated—termite mounds, hollow branches, rocky ridges (Prediction 1a)—or well protected—burrows, ground holes, rock crevices, hollow trunks (Prediction 1b) (Fig. 3)—to place their latrines and avoided areas with low vegetation cover (Prediction 2, fine scale). Similar latrine structures have been described for genet latrines (Espírito-Santo et al. 2007) and those of other small carnivores elsewhere (Ruibal et al. 2011; Buesching and Jordan 2022). We hypothesize that using a wide range of micro-structures provides an efficient diffusion of scent marks, a higher longevity, and/or a higher probability of detection. For example, latrines located on bare ground may not satisfy the first two criteria, but they may be located near well-used animal trails or dirt roads, hence increasing detection probability (Carvalho et al. 2011). Genets are flexible and use the structure types that are more available, being mostly old trees in cork oak woodlands in Portugal (Espírito-Santo et al. 2007), and rocks in France and central Spain (Livet and Roeder 1987; Barrientos 2006). In our study, genets seemed to follow the same strategy and used more burrows and ground holes, and less trees (Table S2). This behaviour reflects the high abundance of burrows and holes made by fossorial species like the aardvark (Orycteropus afer), Cape porcupine (Hystrix africaeaustralis), honey badger (Mellivora capensis) and springhare (Pedetes capensis), among others. Latrines at conspicuous sites might also indicate their use as landmarks to help in orientation (Buesching and Jordan 2022). This hypothesis is equally supported by their placement along edge habitats like riverine corridors (Prediction 3) and, to a lesser extent, along dirt roads (Prediction 4b) (Fig. 4e, f, and g) (Palomares 1993; Carvalho et al. 2011). The access route to the latrines was characterized by a higher slope inclination when compared to random locations. Steeper slopes (mostly in burrows) help maximize olfactory marking by allowing the accumulation of scats. They also contribute to the protection of the latrines from heavy rains and/or direct sun radiation and hinder the removal by animals like coprophagous beetles (Rodgers et al. 2015; Buesching and Jordan 2022).

Latrine placement decreased in Bushclump Karroid Thicket areas (BKT) within the 100 m buffer (Fig. 4c), which suggests that genets avoid marking in areas where predation risk is higher (Prediction 2, landscape scale; Buesching and Jordan 2022). Indeed, BKT patches include smaller karroid thickets and scattered bigger bushclumps, which creates opportunities for predators like black-backed jackals (Canis mesomelas) and caracals (Caracal caracal) to ambush and kill genets. In addition, BKT areas do not provide trees for a safe escape. In fact, four out of our 21 radio-collared genets were killed by jackals in BKT patches (F. Carvalho, unpublished data) and dietary studies at GFRNR confirmed that jackals eat genets (Do Linh San et al. 2009).

As predicted, most latrines were located along riverine corridors, more specifically in Riverine Combretum Forests (Prediction 3, Fig. 4e). Moreover, our best scale-independent model showed that the probability of latrine site selection by genets was higher (> 0.9) within 100 m from riverine corridors (Fig. 4g). These corridors boast a high availability of potential latrine structures, mostly burrows, hollow trunks, and fallen tree branches (Fig. 3). They also provide water and a high abundance of preys (e.g., small mammals and insects) for small carnivores (Palomares 1993; Madikiza et al. 2010; Sikade 2017), and buffer extreme weather conditions (Shirk et al. 2014). Riverine forests are genets’ olfactory marking hotspots, where several latrines are clustered along them (Palomares 1993; Espírito-Santo et al. 2007). This cluster configuration is moderately supported by our scale-independent model for the distance to the nearest main latrines (Fig. 4h) (Prediction 4a). Latrine clusters might help save energy while foraging, as they are often located close to resting sites (Espírito-Santo et al. 2007). The spread of latrines along ecotones also suggests that latrines may play a role in territory defence (Piñeiro and Barja 2015). In fact, riverine habitats are intensively used as movement corridors by animals, increasing the chances of encountering conspecifics, prey, and predators (Shirk et al. 2014). However, the concealment offered by dense vegetation cover, coupled with a high number of evasion routes (trails), provides safety while marking (Bist et al. 2021). The regular use of different latrines might also be a strategy to maintain several active scents to confuse predators (Rathbun and Cowley 2008). Finally, using several latrines can also be related to reducing parasite transmission at otherwise heavily used latrines (Buesching and Jordan 2022).

Distance to permanent water sources was not supported by our scale-independent model (Prediction 4c); this might be a consequence of genets being able to get enough water in riverine corridors (Palomares 1993). Finally, the negative effect of the TRI at the 500 m scale might reflect the low availability of burrows and holes made by fossorial animals in rugged areas. Indeed, in those areas the few latrines we found were in rocks. The same pattern was described for rusty-spotted genets (G. maculata) at Telperion Nature Reserve (South Africa), where walking transects were conducted in both riverine and rocky outcrops habitats, but all latrines were found in the riverine areas (Blomsterberg 2016).

Scope and limitations

Most multi-scale studies on habitat selection and suitability by carnivores focused on wider-ranging species at broader scales and were based on presence-only data modelling (e.g., Maxent) (e.g., Mateo-Sánchez et al. 2013; Khosravi et al. 2019). We applied a multi-scale modelling framework on a particular life trait—latrine site selection—of two sympatric small carnivores. We then adapted a match-control design to compare each observed latrine site with two randomly generated points (PSF) and described the best environmental predictors explaining the probability of latrine site selection (Fig. 2). Although seldom used (compared to Maxent analysis), this method has been suggested as a valid alternative (McGarigal et al. 2016; Thorne and Ford 2022). We also focused our analysis on a fine scale (20 m buffer), reflecting latrine site selection by genets inside different habitat types (4th order selection, Johnson 1980). Thus, our method is effective because it emphasized factors that are crucial for marking behaviour at the fine scale, and which would have been ignored at broader scales. This reinforces the need for a multi-scale assessment of small carnivore habitat selection inside home range habitats (Larroque et al. 2017).

Nevertheless, we had to face some limitations. First, habitat selection studies recommend measuring driving factors at spatial and temporal scales (McGarigal et al. 2016; King et al. 2017). We only included the spatial scale, but we considered latrines used by both species covering all seasons to enable a more robust analysis. Consequently, all behaviours (patrolling, breeding, etc.) related to olfactory marking were included. Second, we based our approach on detailed spatial orders of selection, from the 2nd to the 4th (Johnson 1980). Moreover, we tested different scales (20 m, 100 m, 500 m, and scale-independent), but not broader scales. Using values higher than 500 m would result in obtaining variables that would not change their values proportionally among latrines and random points. Assessing effects at wider ranges would be more appropriate in heterogenous landscapes with a high human influence (Vergara et al. 2016). Third, at the fine scale some resources are homogeneously distributed and abundant in natural areas. Thus, differences in shrub and tree cover might not be significant between latrine sites and random points (McGarigal et al. 2016). However, these resources were crucial for latrine site selection, as they allow a safe use of the latrines by decreasing predation risk, and their shortage compromises the marking behaviour of carnivores across landscapes. Unfortunately, remote sensing tools cannot measure these local variables systematically. Thus, most studies have disregarded local variables due to the extensive field work effort needed to sample them across large areas. Nevertheless, this precludes extrapolating these results across landscapes.

Management implications

Our results have important implications for the conservation management of protected areas (PAs) and surroundings. Outside PAs, the impacts from intensive agriculture and livestock pressure threatens specific habitats that enhance latrine placement, such as riverine forests. Conserving animal species is only possible if we also maintain healthy populations outside PAs (Khosravi et al. 2019). Accordingly, to promote small carnivore persistence across landscapes without disrupting their complex olfactory marking system, conservation efforts should focus on preserving riverine forests. This is especially relevant in more open and thus more hostile matrices (Carvalho et al. 2016a). Considering our findings that most latrines were located within 100 m from riverine corridors, it is recommended that management measures focus on preserving natural vegetation within a minimum distance of 50 m on each side of these corridors. Concurrently, owners of land surrounding PAs should avoid using heavy machinery that might destroy important latrine structures such as burrows, termite mounds or remove old trees. The information gathered here applies to genets and other small carnivores such as wildcats (Felis sylvestris) (Piñeiro and Barja 2015), mongoose species (Palomares 1993; Jordan et al. 2007), pine martens (Martes martes) (Barja et al. 2011), and ringtails (Bassariscus astutus) (Barja and List 2006). Nevertheless, future research should also target areas outside PAs, accounting for different scales and targeting habitats more favourable for latrine placement like riverine forests. Moreover, future studies should address measures to conserve the above-mentioned fossorial species, as they are the main providers of latrine structures (burrows) for our focal genet species.

Conclusions

Our work joins a very small set of studies that use a multi-scale approach focussing on local scale elements such as latrine sites for small carnivores. We showed that latrine site selection by genets is driven by variables acting at different scales. Our analysis at the fine scale revealed that burrows and holes are the favourite structures used as latrines and suggested that the latrine access slope is important to preserve and maximize scent transmission. The best models at the landscape scale (including scale-independent) highlighted the crucial importance of riverine corridors as hotspots for latrine placement (Palomares 1993; Espírito-Santo et al. 2007), but also the avoidance of open areas where predation risk is higher (Popp et al. 2007). Protecting riverine forests not only preserves latrines sites, but also resting sites, prey availability, water sources, and important movement corridors (Espírito-Santo et al. 2007; Shirk et al. 2014; Carvalho et al. 2014, 2016a). Unravelling the drivers of latrine site selection at multiple scales by animals with a complex communication system is crucial to understand their social structure and to devise management practices that ensure the persistence of populations across landscapes.