Estimates on demographic parameters altogether with social factors are integral and can be very useful to assess the risks that a population may face in the future. Rescue operations may provide a unique opportunity to gather data on individuals of an area and thus provide population information. Animal rescue data provided by the Rio de Janeiro Botanical Garden fauna team were used to understand the structure of a population of Bradypus variegatus in an urban remnant of the Atlantic Forest (Tijuca National Park, PNT). This study aims to provide data on the abundance, density estimation, sex ratio, and occurrence of this population in the PNT. We rescued 44 sloths, four of whom were dead. The population density was estimated at 0.6 ind / ha, a low-density value compared to other urban remnants (8.5 to 12.5 ind / ha). Our model suggests a unstable and in decline population, which could be a delayed reflection of years of deforestation in the Atlantic forest. Although B. variegatus isn’t, yet, considered threatened due to their broad distribution, they can be locally extirpated due to population unfeasibility in forest remnants of Atlantic Forest regions, suggesting we should evaluate its threat levels at population level.

Data Collection

The RJBG Wildlife conservation team provided us with all data used in this study, and there were no active captures. The team works in technical cooperation with the Brazilian Institute of the Environment and Renewable Natural Resources (IBAMA), assisting the RJBG local vertebrate fauna when it is needed. They operate mostly in the park's common areas to protect both the animals and visitors when there is a call for injured animals or other situations that could offer risk. Rescued animals are measured and, except birds, receive a microchip for identification (Transponder Partners PA120 - PA140, Microchip Partners - Switzerland), being posteriorly relocated to the forest areas.

The animals were handled following Cassano et al. (2011) and Falconi et al. (2015) methods. We identified the sex of the individuals based on the presence/absence of the dorsal speculum (Hayssen, 2010). To classify their age classes, we considered the individuals’ HBL (head-body length) as follows:  juveniles, HBL < 40 cm; adults, HBL > 40 cm (Castro-Vásquez et al. 2010 apud Emmons & Feer, 1990 and Plese & Moreno, 2005). We used a 1.5 m measuring tape to measure the individuals’ body length, and a 300mm digital caliper (100.178BI Digimess, Brazil) to measure other individuals’ morphometry (e.g., size of the claws, tail, femur, humerus). We weighted the individuals using a digital weight scale – 15 kg (Economic line next – digital scale, Balmak, Brazil).

Individuals captured for the first time received a subcutaneous microchip associated with an ID number, as mentioned before. Sloth monitoring and individuals’ identification were performed through a remote reading of the animal’s ID microchip using a bamboo stalk, when needed, with a microchip reader attached to its extremity. After all procedures, the animals were released in a forest belt closer to TNP, known here as Atl. Forest – Cacti.

Data Analysis

We carried out our analyses in the software environment R 3.6.1 (R Development Core Team, 2019). We evaluated the differences in weight and HBL between sexes through the Mann-Whitney test since our data was not normally distributed (Shapiro-Wilk test, p > 0,05). We used a one-way analysis of variance (ANOVA One Way) to evaluate variations in weight per age class and correlation and polynomial regression tests to visualize the relationship between size and weight better. To evaluate seasonality, we applied the WO-Test (Webel-Ollech overall seasonality test) from the package seastests, which combines three other tests: QS-test, QS-R, and Kruskal-Wallis (Ollech, 2019).

Population estimation

We constructed a simple model (appendix) for the estimative of population size (N) based on a variation on detection probabilities (p) with time (similar to Model Mt, Otis et al. 1978). We then estimated the survival and recruitment rates between occasions to infer the population abundance. Hence, we implemented an open population model since a closed population is unlikely due to the survey's long timespan. We ran the model using the Bayesian MCMC framework (Kéry & Schaub, 2011) in which we could limit the posterior distribution of p (through prior definition) to lower values since we had few recaptures (appendix). We applied Data augmentation (Royle et al., 2007) for the estimation of N by transforming a closed population model into an occupancy model (Kéry & Schaub, 2011), which allowed us to increase the accuracy of the posterior distribution of the population size (N). Consequently, we run the occupancy model to estimate detection probability (p) and the probability of the inclusion of a member of the data augmented individual (Ω) to the population size (N) (Kéry & Schaub, 2011) (Table 1).

This model was implemented in software R3.1.2 (R Development Core Team, 2014), with the package R2jags (Plummer, 2012), which estimates the posterior distribution of the variables by performing Markov Chain Monte Carlo (MCMC) iterations. We run three chains with 15,000 iterations each, discarding the first 5,000 as burn-in. The convergence of the model for all chains was checked visually and with the Gelman-Rubin statistic (r ̂), in which values sr ̂<1.1 suggests convergence (Kéry, 2010). We also performed a model fitness using a Bayesian P-value (Zipkin et al., 2010), described in the supplementary material. Finally, we estimated the sexual ratio by the number of males per female (Soares & Carneiro, 2002).