Ocelot (Leopardus pardalis) Density in Central Amazonia

Ocelots (Leopardus pardalis) are presumed to be the most abundant of the wild cats throughout their distribution range and to play an important role in the dynamics of sympatric small-felid populations. However, ocelot ecological information is limited, particularly for the Amazon. We conducted three camera-trap surveys during three consecutive dry seasons to estimate ocelot density in Amanã Reserve, Central Amazonia, Brazil. We implemented a spatial capture-recapture (SCR) model that shared detection parameters among surveys. A total effort of 7020 camera-trap days resulted in 93 independent ocelot records. The estimate of ocelot density in Amanã Reserve (24.84 ± SE 6.27 ocelots per 100 km2) was lower than at other sites in the Amazon and also lower than that expected from a correlation of density with latitude and rainfall. We also discuss the importance of using common parameters for survey scenarios with low recapture rates. This is the first density estimate for ocelots in the Brazilian Amazon, which is an important stronghold for the species.

We considered Mh the most biologically plausible amongst the candidate models because the ocelot is a territorial species, resulting in unequal access to the camera trap grid by different individuals (Tobler and Powell, 2012). There is no biological reason to believe that detections of ocelots in the surveyed area vary with time (seasonality effect) or behavior (as bait had no significant effect on photographic rate). Therefore, we reported results from model Mh using the jackknife estimator (Noss et al., 2012).
We divided the population size estimate generated under model Mh by the estimated effective sampled area of the camera-trap survey to estimate density.
The effective sampled area is usually calculated by adding a buffer around each camera-trap station (Karanth and Nichols, 1998;Silver et al., 2004), as the area used by captured individuals is certainly larger than the area covered by the trapping grid. In theory, the width of the buffer is related to the home range size of the target species in the study area. In the absence of home range size data, the mean of the maximum distance moved (MMDM) by all individuals photographed at more than one camera-trap station is used as an approximation of the home range diameter. Most studies of ocelot have used the ½ MMDM buffer to estimate effective sampled area (Di Bitetti et al., 2006;Maffei et al., 2005;Trolle and Kéry, 2003). However, some studies have demonstrated that full MMDM buffer may be a better proxy of home range size than ½ MMDM (Dillon and Kelly, 2008;Maffei and Noss, 2008).
We combined individual movements from all surveys to calculate MMDM and shared this value across surveys. As for the spatial capture-recapture analyses, by combining movement information from all surveys we assumed that there is no significant variation on home range sizes across surveys. As recommended by Dillon and Kelly (2007) for sparse data, we included zero-distance movements (animals captured multiple times but always at the same trap) in the MMDM calculation. For the sake of comparison with previous studies, we reported densities using both buffer widths (full and ½ MMDM) as estimates of effective sampled area. As for the spatial capture-recapture analysis, we used a habitat mask to exclude water surface areas. The standard error for density estimates followed (Karanth and Nichols, 1998).

Results
The mean maximum distance moved by individuals captured more than once (MMDM) was 2919.7 m (range: 0 -10812 m, SD = 3304.7, N = 15). The effective surveyed area with MMDM buffer was 276.3 km 2 for the first and third survey and 281.3 km 2 for the second survey. We report the summary of population size and density estimates for different effective sampled areas in S1 Table. S1 Spatial capture-recapture model with independent estimation of parameters for each survey Instead of sharing parameters across surveys, we specified a model that independently estimates parameters and density for each survey separately (independent model), using the same procedure described in the Methods section.
The estimates for the independent model had larger confidence interval for two of the three surveys (S1 Fig.). Whereas density estimates for the shared parameters model vary from 20.8 to 25.4 ocelots per 100km 2 , estimates for the independent model vary from 10.4 to 73.0 ocelots per km 2 (S2 Table). This higher variability of the density estimates for the independent model is mostly due to differences in the movement parameter estimates across surveys (Tobler and Powell, 2012). S1 Figure. Ocelot density estimate and 95% confidence interval for the spatial capture-