Study sites- Cozumel Island is located in the Mexican Caribbean Sea, 17.5 km east of the coast of the state of Quintana Roo, in the Yucatán Peninsula, between the extreme coordinates 20°16′18.2′′ to 20°35′32.8′′N; 86°43′23.3′′ to 87°01′31.1′′W (Fig. 1). It has an area of 478 km2 (Cuarón 2009), and the dominant vegetation type on most of the island is semi-evergreen tropical forest. However, there are also other vegetation types present, including deciduous low tropical forest, secondary vegetation, chit palm forest, halophilus sand dune vegetation and mangroves (Segrado et al. 2008; McFadden et al. 2010). It is the island in Mexico with the highest number of endemic vertebrate taxa (31; Cuarón 2009; McFadden et al. 2010).
The island 's population has expanded significantly during the last decade, from 70,000 to approximately 88,500 inhabitants, the majority of whom live in the city of San Miguel (INEGI 2020). Tourism has become the main economic activity, and this has a considerable impact on the transformation of the landscape inside the island (Palafox et al., 2015; López-Contreras et al. 2021).
In order to obtain information on the occurrence of pygmy raccoons on Cozumel, we conducted photo-trapping sampling at 15 study sites (Fig. 1) during 2014, 2015 and 2016. Then, throughout 2021 and 2022, 3 of these sites were re-surveyed intensively (sites 1, 4 and 15). At each study site, up to 20 camera traps stations were placed (minimum 6 and an average of 13 by site). Each site was sampled for at least 20 days between January to June 2014, August to December 2015 and January to November 2016. Then, between July to December of 2021 and from January of April 2022, only three sites were sampled, with at least 10 cameras each for at least 30 days.
We used Long Range IR Model E2 and Black Flash Model E3 (Cuddeback®) camera traps, spaced 400–600 meters apart and 40–60 cm above the ground on tree trunks. Throughout each survey period, they were scheduled to operate continuously during each survey period and take three photos and one 20-second video for each capture event. To increase the detectability of the species, each camera was baited with a can of sardines perforated and nailed to the ground to prevent animals removing the bait.
Occupancy Models- Using the recording data for the different species at the different survey stations, we used multi-season occupancy models (MacKenzie et al. 2003), to examine whether different habitat types and the presence or absence of potential predators (feral dogs) and competitors (dwarf coati and opossum), altered the occupancy probability (Ψ) and detectability (p) of P. pygmaeus at different sites and between survey periods. While the survey duration varied from year to year, we standardized camera survey effort by limiting our analyses to the first 20 days of photo-trapping for 2014–2015 and to the first 60 days of photo-trapping for 2016 and 2021–2022.
We calculated the total number of independent photographic records of each species for each sampling station based on each survey (considering as independent, those records obtained in the same camera trap, with at least one hour of interval between them). For the construction of the multi-season occupancy models, we utilized PRESENCE 2.13 software (Hines 2006) based on a total of 682 independent records obtained from more than 5000 camera trap days.
For the multi-season models we considered two additional parameters, ε(t) and γ(t). These parameters are, respectively, the probability that a species was not present at a site, or occupied a site, between seasons t and t + 1. Moreover, multi-season models contain four different parameterizations alternatives (Mackenzie et al. 2017); we used parameterization number four, which is the simplest alternative, and given that our analyses were limited to probability of detection, gamma and epsilon were modelled as constants. Based on the Akaike’s Information Criterion (AIC) we selected the model with the best fit to the data in each analysis.
We considered environmental variables (categorical) to have the potential to influence probabilities of detection at cameras sites. These were five types of habitat based on INEGI (2013): mangrove (mn), coastal dune vegetation (dv), chit palm forest (cp), medium tropical forest (mt) and low tropical forest (lt). And also three levels of anthropogenic disturbance: low disturbance (Ld), medium disturbance (Md) and high disturbance (Hd), based on the proximity to roads and transformed areas. Of the 15 study sites studied, 5 were classified as Ld (sites 2, 5, 11, 12, 14), 9 as Md (1, 3, 4, 6, 7, 9, 10, 13 and 15) and one as Hd (site 8 located next to the Cozumel landfill).
Activity patterns- To determine the activity patterns of each species, we considered all records of each species at each survey station that were at least one hour apart. Using the criteria proposed by Gómez et al. (2005), the precise time of each record was obtained to classify each activity pattern as diurnal, nocturnal, mostly diurnal, mostly nocturnal, crepuscular or cathemeral. Using this data, we plotted the distributions of activity patterns for each species and calculated the degree of overlap between species pairs (Δ) with 95% confidence intervals derived from by 1000 bootstrap parametric interactions. Δ values range from 0 to 1, where 0 indicates completely distinct temporal activity and non-overlapping, and 1, indicates complete overlap between the species being compared species. The Watson-Wheeler test was used to determine whether mean hour of activity varied significantly between species pairwise. To implement the temporal analyses, we used the “Overlap” and “Circular” packages in R environment and languages (Jammalamadaka et al. 2001; Ridout and Linkie 2009).