Estimating density of ocelots in the Atlantic Forest using spatial and closed capture–recapture models

Abstract

Abstract Monitoring variation in population features such as abundance and density is essential for evaluating and implementing conservation actions. Camera trapping can be important for assessing population status and trends and is increasingly used to generate density estimates through capture–recapture models. Moreover, success in using this technique can vary seasonally given shifting animal distributions and camera encounter rates. Notwithstanding these potential advantages, a gap still exists in our understanding of the performance of such models for estimating density of cryptic Neotropical terrestrial carnivores with low encounter rate probability with cameras. In addition, scanty information is available on how sampling design can affect the accuracy and precision of density estimates for Neotropical carnivores. We evaluate the performance of spatially explicit versus nonspatial capture–mark–recapture models for estimating densities and population size of ocelots (Leopardus pardalis) within an Atlantic Forest fragment in Brazil. We conducted two spatially concurrent surveys, a random camera-trap deployment covering the entire study area and a systematic camera-trap deployment in a small portion of the study area, where trails and unpaved roads were located. We obtained 244 photographs of ocelots in the Rio Doce State Park from April 2016 to November 2017, using 54-double camera stations spaced approximately 1.5 km apart (random placement) totaling 4,320 trap-nights and 15-double camera stations spaced from 0.3–10 km apart (systematic placement) totaling 1,200 trap-nights. Using the random placement design, ocelot density estimates were similar during the dry season, 14.0 individuals/km2 (± 5.6 SE, 6.6–30.0, 95% CI) and 13.78 individuals/km2 (± 4.25 SE, 5.4–22.1, 95% CI) from spatially explicit capture–recapture and nonspatial models, respectively. Using the systematic placement design spatially explicit models had smaller and less precise ocelot density estimates than nonspatial models during the dry season. Ocelot density was 12.4 individuals/100 km2 (± 5.0 SE, 5.8–26.7, 95% CI) and 19.9 individuals/km2 (± 5.2 SE, 9.7–30.1, 95% CI) from spatially explicit and nonspatial models, respectively. During the rainy season, we found the opposite pattern. Using the systematic placement design, spatial-explicit models had higher and less precise estimates than nonspatial models. Ocelot density was 24.6 individuals/100 km2 (± 13.9 SE, 8.7–69.4, 95% CI) and 11.89 individuals/km2 (± 3.93 SE, 4.19–19.59, 95% CI) from spatially explicit and nonspatial models, respectively. During the rainy season, we could not compare models using the random placement design due to limited number of recaptures to run nonspatial models. In addition, a single recapture yielded an imprecise population density estimate using spatial models (high SE and large 95% CIs), thus precluding any comparison between nonspatial and spatially explicit models. We demonstrate relative differences and similarities between the performance of spatially explicit and nonspatial capture–mark–recapture models for estimating density and population size of ocelots and highlight that both types of capture–recapture models differ in their estimation depending on the sampling design. We highlight that performance of camera surveys is contingent on placement design and that researchers need to be strategic in camera distribution according to study objectives and logistics. This point is especially relevant for cryptic or endangered species occurring at low densities and having low detection probability using traditional sampling methods.