Data Analysis
We developed two types of occupancy models, single-species single-season and co-occurrence models. The first ones estimate the probability of occupancy for each site while considering the probability of detection and assessing the effect of covariates on both probabilities (Mackenzie et al., 2006). For each species we developed detection histories for each site, considering 15 consecutive camera trap days as a sampling occasion, thus having four sampling occasions for MML and three for OL. For ungulates we included six additional sampling occasions corresponding to the transects.
For each species, we first modelled detection probability using covariates within the cell and the type of sampling, and selected the best model using AIC and AICw (Burnham & Anderson, 2002). We then evaluated separately the influence of each individual covariable on the probability of occupancy and then generated additive models combining the variables with the best response and with AIC < 2. For those variables in the best models we estimated 80% confidence intervals to assess whether β coefficients, which describe the effect, overlapped with zero (Burnham & Anderson, 2002). All models were adjusted using the package “Unmarked” from R (Fiske & Chandler, 2011). In this way, we assessed the most important variables affecting occupancy in each landscape, both for shared species and for species unique to each landscape. For the agouties and anteaters, we compared species from the same genus between landscapes, and we did the same for the two species of curassow, which have similar habits and are both subject to hunting by humans. To evaluate the interacting effects of landscape and body size in occupancy levels we conducted a tow-way ANOVA, splitting the species in two groups, larger or smaller than 30kg.
To determine the effects of one species on the probability of detection or occupancy of another species, we used co-occurrence occupancy models (Mackenzie et al., 2004). For each landscape we developed models for interactions between: Panthera onca (only MML), Puma concolor , Cuniculus paca, Dasyprocta punctata ( MML) and D. fuliginosa (OL). We used the parameterization\(\frac{\psi^{\text{Ba}}}{r^{\text{Ba}}}\) in the package RPresence for R (MacKenzie & Hines, 2020) and estimated the following probabilities:\(\psi^{A}\) probability of occupancy of the dominant species (A),\(\psi^{\text{BA}}\) probability of occupancy of the subordinate species when the dominant species is present, and \(\psi^{\text{Ba}}\) the probability of occupancy of the subordinate species when the dominant species is absent. We developed a set of models a priori , which assumed that the presence of the dominant species influenced the presence of the subordinate species (\(\psi^{\text{BA}}\ \neq\ \psi^{\text{Ba}}\)), and constrained models where the occupancy of the subordinate was independent of the presence of the dominant species (\(\psi^{\text{BA}}\ =\ \psi^{\text{Ba}}\)). We included those variables most important for each species derived from the single-species models and we run two models for each pair of species, switching the dominant species, as we assumed the effects were not symmetrical (\(\psi^{\text{BA}}\ \neq\ \psi^{\text{AB}}\)).
For the probabilities of detection, we estimated the following parameters: \(P^{A}\) probability of detecting the dominant species given the absence of the subordinate, \(P^{B}\) probability of detecting the subordinate given the absence of the dominant, \(r^{A}\) probability of detecting the dominant given that both are present, \(r^{\text{BA}}\)probability of detecting the subordinate given that both are present and the dominant is detected, \(r^{\text{Ba}}\) probability of detecting the subordinate given that both are present and the dominant is not detected. We built a set of models a priori , assuming that detection probabilities of each species were independent of the presence of the other species (\(p^{A}=r^{A}\) and\(p^{B}=r^{\text{BA}}=r^{\text{Ba}}\)), others in which only the subordinate species was influenced by the presence of the dominant one (\(p^{A}=r^{A}\) and \(p^{B}\neq r^{\text{BA}}=r^{\text{Ba}}\)) and models where each species was influenced by the presence and detection of the other species (\(p^{A}\neq r^{A}\) and\(p^{B}\neq r^{\text{BA}}\neq r^{\text{Ba}}\)).