Introduction

Exploring factors affecting species distributions is important to better understand their biology and for decision-making processes in population monitoring, management, and conservation (Jonzén 2008). These factors include habitat features, such as resource availability and interactions with other species (Ferreguetti et al. 2016; Dias et al. 2019). Given the current human-induced biodiversity crisis worldwide, many studies have been focusing on understanding human-related factors that affect species distributions and may lead to their extinction, such as habitat loss and overexploitation (Rodrigues and Chiarello 2018; Dobbins et al. 2020).

The study of habitat relationships of some species is hampered by their low densities and/or elusive behavior, which is the case of armadillos (Xenarthra: Cingulata) (McDonough and Loughry 2008; Superina et al. 2014). They usually present low population densities and solitary habits and spend most of the time underground inside their burrows (McDonough and Loughry 2008). Although armadillos are subjected to many conservation challenges, few studies have focused on exploring their habitat requirements, and many species are still understudied (Superina et al. 2014; Superina and Abba 2020). For taxa such as armadillos, the evaluation of factors influencing species occurrence (or occupancy) can be suitable to assess habitat relationships, as it requires only presence-absence data while accounting for false absences caused by imperfect detection (MacKenzie et al. 2002). However, this methodological approach is not yet common in existing studies focused on this group (but see Ferreguetti et al. 2016; Rodrigues and Chiarello 2018).

The Brazilian three-banded armadillo, Tolypeutes tricinctus (Linnaeus, 1758) (Cingulata: Chlamyphoridae) (Gibb et al. 2016), is a threatened species endemic to northeastern Brazil (Miranda et al. 2014; Feijó et al. 2015; Reis et al. 2015; Santos et al. 2019b; Hannibal et al. 2021; Schetino et al. 2021). Tolypeutes tricinctus is remarkable for the ability to roll up its body into a ball as a defense mechanism (Online Resource 1 — Fig. S1); however, it also makes this species easy to hunt (Santos et al. 1994). Hunting and habitat loss are considered the main threats to T. tricinctus and may have led to its extinction from different localities (Miranda et al. 2014; Feijó et al. 2015; Reis et al. 2015). Consequently, it is thought to be currently restricted to less disturbed areas (Miranda et al. 2014). Tolypeutes tricinctus occurs in the Caatinga dry forests of northeastern Brazil and adjacent areas in the Cerrado savannas (Feijó et al. 2015). The areas in the Cerrado where T. tricinctus occurs have been increasingly converted into sugarcane and soybean plantations (Miranda et al. 2014). Meanwhile, the Caatinga is a historically neglected region for research and conservation initiatives (Santos et al. 2011) and has been experiencing an expansion of wind farms (Neri et al. 2019), which can impose negative impacts on wildlife, such as barrier effects to animal dispersal and increased levels of hunting and road-kills (Helldin et al. 2012; Dias et al. 2019, 2020).

Despite this dramatic scenario, only recently T. tricinctus has received conservation attention with the creation of a national action plan for its conservation (ICMBio 2014). In addition, little is still known about its biology. For instance, only recently the burrowing activity of Tolypeutes spp. and the T. tricinctus predation by native predators have been described in the literature (Attias et al. 2016; Magalhães et al. 2021). Most of the available information about T. tricinctus comes from unpublished works and field observations (e.g., Marinho-Filho et al. 1997), and no published study has systematically assessed its habitat relationships so far. The combination of restricted distribution, high threat level, low conservation attention, and lack of knowledge led T. tricinctus to be considered a priority species for research and conservation efforts (Superina et al. 2014).

Here, we provide the first systematic assessment of species-habitat relationships for T. tricinctus. We combined camera traps, active searches, and occupancy models to investigate how different environmental and human-related factors may influence its local distribution (Table 1). We worked with a recently discovered population of T. tricinctus in an unprotected area, where human activities, such as cattle raising, hunting, and wind power production are present.

Table 1 Covariates used to model occupancy (Ψ) and detection (p) probabilities of Tolypeutes tricinctus in a human-modified landscape in the Caatinga ecoregion, northeastern Brazil. The hypothesis, mean and range values, and the expected effect on occupancy and detection probabilities are given for each covariate. *Covariates used in the final model set after correlation tests. The plus ( +) and minus ( −) signals denote positive and negative effects, respectively. NE, not evaluated
Full size table

Materials and methods

Study area

This study was conducted in a private area of 53 km2 in the municipality of Brotas de Macaúbas, state of Bahia, in the Caatinga ecoregion of northeastern Brazil (Fig. 1). The climate is altitude tropical, with a warm rainy season (November–March) and cold dry season (April–October), and altitudes between 1000 and 1200 m. a. s. l. (SEI 2014). The soil types vary from Litholic Neosols (or Leptosols) at higher elevations and Red-yellow Latosols (or Ferralsols) at lower elevations (Online Resource 2). The native vegetation consists mainly of Caatinga dry forests and some small patches of Cerrado savannas and Rupestrian Grasslands (Fig. 1; Online Resource 3 — Figs. S4 and S5). Anthropized areas comprise wind farm infrastructure, roads, and former croplands in regeneration. The latter is characterized by zones of exposed soil partially covered by ruderal herbaceous and/or shrubby vegetation. In addition to habitat loss, hunting is performed by local people; however, T. tricinctus is not a frequent hunting target (R. A. Magalhães, unpublished data). The western portion of the area is partially within a biodiversity priority area (MMA 2017).

Fig. 1
figure 1

Location of sampling units or sites (black dots) used to sample for Tolypeutes tricinctus in the studied area (dashed polygon) in northeastern Brazil. Green is native bushy-arboreal vegetation. Pale brown is anthropized open areas, such as former croplands, and pink is other open areas, such as rocky outcrops. The shaded area is a biodiversity priority area. The insert shows the location of the study area (orange triangle) according to the Caatinga distribution (in gray) and Brazilian states. Biodiversity priority areas follow MMA (2017), and land cover and Caatinga distribution follow Projeto Mapbiomas (2021). Geographic Coordinate System and Datum: WGS84

Full size image

Data collection

We randomly distributed 24 sampling units (or sites) 1.5 km apart in the area. We installed one Bushnell® digital camera trap at each site approximately 30 cm above the ground and angled it slightly downwards to avoid missing closer records. All cameras were set to operate 24 h a day with a 1-min interval between 15-s-long videos. Sampling occurred between April and July 2019, totaling 90 days. In the middle of the sampling period, we visited all sites to check the cameras.

We conducted 30-min active searches for individuals of T. tricinctus and their signs (tracks and excavations; Online Resource 1 — Fig. S2) in a 10-m radius (63 m2) buffer surrounding each site during camera trap checking and removal, thus resulting in two active searches at each site. We defined this area due to logistic constraints, as it could be sampled in 30 min, thereby not compromising our field logistics. During each active search, we registered whether any evidence of T. tricinctus was found (1) or not (0). Active searches were conducted with the help of three previously identified specialists from local communities. They presented substantial knowledge of the local fauna, were able to track T. tricinctus individuals, and precisely described their signs. At each site, one of the three local specialists was responsible for performing the active searches. To avoid issues with false-positive detections and sign decay during active searches (Rhodes et al. 2011), we only considered fresh tracks and excavations that could be unambiguously attributed to T. tricinctus and erased all after camera trap installation and the first active search. This procedure prevented us from registering signs that could have been left before our sampling and from recounting the same signs during the second active search.

Occupancy modeling covariates

We selected covariates that could influence T. tricinctus occupancy probability based on the literature (Table 1). To assess the effect of habitat loss on T. tricinctus occupancy, we calculated the amount of native and anthropogenic habitat at each site (Regolin et al., 2021).

Using 30-m resolution land cover maps generated by Projeto Mapbiomas (2021) for the year 2019, we calculated the proportion of pixels of native and anthropogenic vegetation inside buffers of 1 ha, 10 ha, 20 ha, 50 ha, and 100 ha around each site (number of pixels/buffer area). We tested for correlation between pairs of buffers and because all were highly correlated (|r|> 0.70; Online Resource 4 — Table S2), we used only the scale that presented greater variation throughout the sites for occupancy analysis, which was 1 ha. We also measured the minimum distance from each site to biodiversity priority areas, wind turbines, permanent water sources, roads, and residences. All geoprocessing analyses were conducted in program QGIS v.3.16.7 (QGIS Development Team 2021). We also computed the detection rate of cattle (independent records/day) for each site. Independent records comprised those obtained at the same site with at least a 1-h interval. To evaluate the influence of soil sand content on T. tricinctus occupancy probability (Table 1), we collected soil samples (200 g) in depths of 0–20 cm and 20–40 cm at each site. We performed granulometric analysis using sieving and sedimentation techniques to assess the proportion of each soil particle size-class (in mass fraction): boulders and stones (> 2 mm), sand (2–0.05 mm), silt (0.05–0.002 mm), and clay (< 0.002 mm) (Donagema et al. 2011).

To evaluate the influence of methodological covariates on T. tricinctus detection probability, we used the type of method employed in each sampling occasion at each site as a categorical covariate (Table 1). We also expected detection to vary according to the sampling effort. For each active search, we fixed the sampling effort in 5 days for each occasion, i.e., we assumed that tracks and forage burrows recorded were left up to 5 days before each active search. During this period, they could remain identifiable before being naturally erased. The tracks and forage burrows of T. tricinctus would hardly remain identifiable for more than a few days because the area is very windy, and its soils are sandy and friable (Online Resource 2). For camera trapping, we computed the number of days the camera operated at each site and each 5-day occasion as an effort covariate. Only uncorrelated covariates (|r|≤ 0.70) were used in the analysis (Online Resource 4 — Table S3). As the distance to wind turbines was highly correlated with cattle detection rate and distance to residences, we removed wind turbines from the analysis. We also removed the cattle detection rate from the analysis because it was highly correlated with the proportion of native shrubby-arboreal vegetation, which indicates habitat loss, one of the main threats to T. tricinctus. Consequently, we used eight covariates to model T. tricinctus occupancy and detection probabilities (Table 1).

Data analysis

We combined detections into 20 5-day occasions (18 for camera-trapping and two for active searches) to compose detection histories for each site. Specifically, we recorded whether T. tricinctus was detected (1) or not (0) during each occasion. We used a single-season occupancy model for analysis (MacKenzie et al. 2002).

We built models representing our a priori hypotheses for occupancy (Ψ) and detection probabilities (p) (Table 1) and fit them in Program MARK (White and Burnham 1999). We adopted a “stepdown” strategy to build the models (Lebreton et al. 1992). Using an additive structure of the most parametrized model for Ψ, we first modeled p using only one covariate for each model. Then, using the most parsimonious model for p, we modeled Ψ using only one covariate for this parameter as well. We opted to run models with only one covariate due to our relatively small sample size, thus avoiding over-parameterization. The models were ranked according to the value of the Akaike information criterion adjusted for small sample sizes (AICc), for which the best models are those with the lowest values (Burnham and Anderson 2002). When the difference in AICc value between a given model and the best-supported model was less than two (ΔAICc < 2.00), we considered it as well supported by our data (Burnham and Anderson 2002). Consequently, all models with ΔAICc < 2.00 were identified as best supported. As more than one model presented ΔAICc < 2.00 for Ψ (model uncertainty), we calculated model-averaged estimates for this parameter (Burnham and Anderson 2002). Finally, we also checked for overdispersion using the most parametrized model for both p and Ψ (the full or global model) by applying the goodness-of-fit test developed for occupancy analyses (MacKenzie and Bailey 2004) in Program PRESENCE (Hines 2006).

Results

We obtained 10 records of T. tricinctus by camera trapping at 10 sites and detected the species by active searches at 11 of them, at six exclusively by this method. Tolypeutes tricinctus thus presented a naïve occupancy of 0.67. It was detected once at 11 sites, twice at four sites, and three times at one site.

The goodness-of-fit test revealed no evidence of overdispersion (χ2 = 910.87; ĉ = 1.00; P = 0.61). Our analysis showed uncertainty among Ψ model structures, four of which were better supported (AICc < 2.00; Table 2). Although the covariates distance to roads (β = 0.015), distance to the nearest residence (β =  − 0.004), and soil sand content (β =  − 20.000) were included in the most parsimonious models, the null model structure was supported by our data (ΔAICc = 0.00), predicting that none of these covariates had a strong influence on T. tricinctus occupancy probability, which was high and constant according to our model-averaged estimates (Ψ = 1.00; 95% CI = 0.95–1.00).

Table 2 Model selection results used to estimate occupancy (Ψ) and detection (p) probabilities of Tolypeutes tricinctus in a human-modified landscape in the Caatinga ecoregion, northeastern Brazil. Covariates: BPA, distance to nearest biodiversity priority area (m); effort, sampling effort (days); forest, the proportion of native shrubby-arboreal vegetation in a 1-ha buffer (pixels/area); method, if camera trapping (0) or active searches (1); residences, distance to the nearest residence (m); roads, distance to the nearest road (m); sand, soil sand content (g/g); water, distance to the nearest permanent water source (m); full or global, additive model structure including all covariates with expected effects on Ψ (BPA + forest + residences + roads + sand + water). The dot (.) means no covariate effect on model parameters
Full size table

The detection probability of T. tricinctus varied according to the sampling method (Table 2), being active searches the most efficient. The detection probability by active searches (p = 0.25; 95% CI = 0.15–0.39) was eight times higher than by camera trapping (p = 0.03; 95% CI = 0.01–0.05).

Discussion

We found the T. tricinctus occupancy probability close to one, indicating that this species potentially occupies all sites in the study area. This result highlights that, although T. tricinctus may have disappeared or become rare over much of its range (Miranda et al. 2014; Reis et al. 2015), it can still be widely distributed at localities where it is not subjected to intense hunting pressure and habitat loss, as discussed by Miranda et al. (2014). However, the occupancy probability we found might be overestimated due to the very low detection probability obtained by camera trapping (MacKenzie et al. 2002). The T. tricinctus detection rate by camera trapping in our study area (0.005 records/camera-day) is similar to that found elsewhere in the Caatinga (Campos et al. 2019) and higher than for T. matacus in the Pantanal wetlands (about 0.001 records/camera-day; Porfirio et al. 2014). Considering this apparent generally low detectability of Tolypeutes by camera traps, increasing sampling effort in future research with T. tricinctus should help to improve the precision and accuracy of occupancy estimates (MacKenzie et al. 2002). Nevertheless, even the T. tricinctus naïve occupancy in our study area was high when compared to those already found for other armadillos (Ferreguetti et al. 2016; Rodrigues and Chiarello 2018). Moreover, the high occupancy probability of T. tricinctus we found is supported by interview data from a rural community in our study area, whose residents consider this species abundant and widely distributed (R. A. Magalhães, unpublished data).

We recorded T. tricinctus in contrasting habitats, from forested sites either farther from roads and residences to sites closer to them. Tolypeutes tricinctus has been observed in different habitats, including croplands and managed forests in human-modified landscapes under moderate habitat loss and hunting intensity (Marinho-Filho et al. 1997; Bocchiglieri et al. 2010). However, previous studies have not systematically evaluated its distribution. Our results indicate that T. tricinctus can occupy less or more disturbed sites in such landscapes, thus supporting previous observations.

Although T. tricinctus is regarded as one of the most sensible armadillos to habitat alterations (Marinho-Filho and Reis 2008), we did not find effects of hunting and habitat loss on its occupancy probability. Hunting and habitat loss are considered the most important threats to T. tricinctus (Miranda et al. 2014; Reis et al. 2015). They have been shown to negatively affect the distribution of other armadillos (Ferreguetti et al. 2016; Rodrigues and Chiarello 2018). However, studies with T. matacus have indicated that this species can be common in areas where it is intensively hunted (Noss et al. 2008; Cuéllar 2008). In our study area, T. tricinctus is not a frequent hunting target (R. A. Magalhães, unpublished data), which likely explains why its distribution was not affected by proximity to roads and residences. Similarly, despite the existence of anthropogenic habitats, our studied landscape still holds a substantial amount of native vegetation (Fig. 1). Therefore, the current levels of hunting and habitat loss appear not to be severe enough to affect T. tricinctus distribution. Broader spatial and temporal scales may be required to detect such effects (Santos et al. 2019a).

The detection probability of T. tricinctus was eight times higher by active searches than by camera trapping. Although camera trapping is a non-intrusive method, whose efficiency does not vary significantly according to field conditions, it presents higher initial costs (Silveira et al. 2003). Active searches for signs are efficient to record species occurrence but depend heavily on environmental conditions and experienced personnel (Silveira et al. 2003). During 24 h of active searches, we detected T. tricinctus signs at 46% of the sites, whereas we spent ~ 46,100 camera-hours to detect the species at 42% of them. The costs with field personnel for both methods were equal, as they were applied simultaneously. However, camera trapping involved additional costs with equipment of more than 9000 USD and more time to process data. Active searches thus presented greater cost–benefit when compared to camera trapping to register T. tricinctus occurrence and can even improve occupancy estimates by increasing detection probabilities.

Our sampling design may have favored the detection of T. tricinctus by active searches when compared to camera trapping. First, camera traps were installed ~ 30 cm above the ground. Although angling them downwards may have favored detection (Rowcliffe et al. 2011), lower heights than we used (30 cm) might increase the detection of closer records. Second, active searches covered a 63-m2 buffer surrounding each site, expanding the sampling to outside the camera trap detection zone. This method also allowed us to record different signs of T. tricinctus. Finally, we counted on skilled people from local communities to conduct active searches, which may have also increased detection (Silveira et al. 2003). The inclusion of local specialists reinforces the importance of local ecological knowledge and engagement of local communities in research and conservation initiatives to increase their effectiveness.

In summary, we found T. tricinctus widely distributed in a human-modified landscape subjected to moderate levels of hunting and habitat loss, supporting previous field observations. We also found active searches more effective than camera trapping to detect T. tricinctus occurrence and that both methods can be successfully applied together. Thus, we recommend considering these differences in effectiveness for designing studies on T. tricinctus ecology. Based on our results, we support the engagement of local people in research and conservation projects for better outcomes.