Terrain-mediated biotic interactions shape puma ecology: insights on density, space use, and activity patterns in southeastern Peru

Introduction

The puma (Puma concolor) is the most widely distributed large carnivore in the Americas, occupying a diverse range of habitats from Canada to Patagonia (Quiroga et al., 2016). Despite this ecological flexibility and some degree of resilience to anthropogenic pressures (Guerisoli et al., 2019; Nanni et al., 2023; Quiroga et al., 2016), the species has lost more than 32% of its historical range (Wolf and Ripple, 2017). Habitat loss and fragmentation, prey depletion, and human persecution continue to drive local declines across parts of the species’ range (Nielsen et al., 2015). Although pumas are currently listed as Least Concern globally, they are considered Near Threatened in several South American countries, including Peru (Nielsen et al., 2015). At the same time, puma ecology remains poorly documented across much of tropical South America, particularly in dense forest habitats where direct observation is difficult and rigorous density estimates remain scarce (Kelly et al., 2008; Nielsen et al., 2015).

This knowledge gap is especially important in Amazonian landscapes undergoing rapid anthropogenic change. Tropical forests experience some of the highest rates of agricultural expansion globally (Song et al., 2018), and land-use change is a major driver of carnivore population declines and local extirpations (Nanni et al., 2023). In Madre de Dios, southeastern Peru, the construction of the Interoceanic Highway opened access to previously remote forests, facilitating settlement expansion, agriculture, logging, and increased hunting pressure in some areas (Chávez Michaelsen et al., 2013; Naughton‐Treves et al., 2003). Such changes may affect pumas directly through habitat modification and persecution, and indirectly through changes in prey communities or interactions with other large carnivores. Understanding puma ecology in this region is therefore important not only for monitoring a widespread carnivore, but also for evaluating the functioning of Amazonian predator-prey systems in a changing landscape.

Recent camera-trap studies from the same broader Las Piedras landscape indicate that Amazonian carnivores and disturbance-sensitive mammals can show nuanced responses to moderate human disturbance. Ocelots, for example, may alter their temporal niche in response to anthropogenic impacts without necessarily changing their spatial distribution (Zwicker and Gardner, 2024), while persecuted or disturbance-sensitive mammals can serve as indicators of moderate human disturbance in the Peruvian Amazon (Zwicker et al., 2025). Together, these studies suggest that mammal responses to human activity in this landscape may be expressed not only through spatial avoidance, but also through changes in detection, activity, and community composition.

Large carnivores play central ecological roles by structuring prey populations, influencing mesopredator dynamics, and contributing to trophic cascades and “landscapes of fear” (Estes et al., 2011; Gil-Sánchez et al., 2021; Ritchie and Johnson, 2009; Ritchie et al., 2012; Ripple et al., 2014; Terborgh et al., 2001). Pumas, specifically, have been found to play a role consistent with the mesopredator release hypothesis by suppressing the densities of smaller felids, such as ocelots, margays and jaguarundis, in the Ecuadorian Amazon (Gil-Sánchez et al., 2021). Through these pathways, predators can influence ecosystem resilience, disease dynamics, biological invasions, and broader ecosystem function (Estes et al., 2011; Ritchie et al., 2012; Ripple et al., 2014). Robust estimates of large carnivore density and space use therefore provide important baselines for conservation planning, especially in biodiversity-rich tropical forests where monitoring is logistically difficult and ecological data remain limited.

Puma density has been studied most extensively in North America, where estimates often remain below 1 individual/100 km² (Negrões et al., 2010; Nielsen et al., 2015). In contrast, estimates from Central and South America are fewer and geographically uneven, but they suggest substantial variation among ecosystems. Reported puma densities include estimates from Argentina, Bolivia, Belize, Ecuador, Mexico, and Brazil (Gil-Sánchez et al., 2021; Kelly et al., 2008; Paviolo et al., 2009; Quiroga et al., 2016; Rich et al., 2014; Soria-Díaz et al., 2010; Vega-Flores and Farías-González, 2021; Zanón Martínez et al., 2023). However, direct comparisons among these estimates are difficult because many studies used non-spatial methods or effective sampling areas based on MMDM or ½MMDM, which can overestimate density relative to spatially explicit approaches. For example, Kelly et al. (2008) reported a puma density of 3.42 individuals/100 km² in Belize using a non-spatial approach, whereas spatial reanalysis of the same dataset produced a substantially lower estimate of approximately 0.60 individuals/100 km² (Rich et al., 2014). Similarly, Negrões et al. (2010) estimated puma density at 3.40 individuals/100 km² in central Brazil using non-spatial models, while Janson and Emmons (1990) estimated 2.40 individuals/100 km² in Manu National Park based on ranging data. More recently, Gil-Sánchez et al. (2021) estimated puma densities ranging from 0.26 - 1.92 individuals/100 km² across sites with varying levels of human disturbance in the Ecuadorian Amazon using spatial capture-recapture models. However, comparable density estimates remain limited, and few studies have produced rigorous spatially explicit density estimates for pumas in lowland Amazonian forests.

Puma space use may be shaped by multiple bottom-up, top-down, environmental, and anthropogenic drivers. Pumas often respond to prey availability, with greater use of areas where prey are more abundant or more detectable (Ávila-Nájera et al., 2020; Guerisoli et al., 2019; Negrões et al., 2010). Habitat structure, water availability, forest cover, and productivity may also influence puma occurrence or movement by affecting prey distribution, cover, and travel routes (Ávila-Nájera et al., 2020; Coon et al., 2020; Dias et al., 2019; Nanni et al., 2023; Sollmann et al., 2012). Pumas also display variable responses to human disturbance. In some landscapes, they persist in moderately modified areas, while in others they avoid areas of high human activity or persecution risk (Gil-Sánchez et al., 2021; Guerisoli et al., 2019; Nanni et al., 2023; Quiroga et al., 2016). These context-dependent responses suggest that local landscape structure, prey communities, and human-access patterns may jointly influence puma space use.

In the Neotropics, pumas co-occur with jaguars (Panthera onca) across much of their range (Ávila-Nájera et al., 2020). Because both species are large solitary felids with broadly overlapping energetic requirements, previous studies have examined whether their coexistence is facilitated by spatial, temporal, or dietary partitioning (Ávila-Nájera et al., 2016, Ávila-Nájera et al., 2020; Ayala et al., 2021; de Oliveira and Medellín, 2002). However, the strength and direction of puma-jaguar associations vary among landscapes. Although several studies have documented high temporal overlap between the two species (Ávila-Nájera et al., 2016, Ávila-Nájera et al., 2020; Ayala et al., 2021; Foster et al., 2013), temporal separation has also been reported in some dry-forest systems (Romero-Muñoz et al., 2010). Similarly, spatial co-occurrence patterns appear variable (Ávila-Nájera et al., 2020; Negrões et al., 2010; Sollmann et al., 2012). In some systems, pumas and jaguars may co-occur in prey-rich or structurally favorable areas, while partitioning resources through prey size, fine-scale habitat use, or localized movement behavior. At a broader population scale, evidence from the Ecuadorian Amazon suggests that pumas reach higher densities where jaguars are scarce (Gil-Sánchez et al., 2021). Nevertheless, such interactions are likely context-dependent and may be mediated by landscape heterogeneity (Curveira-Santos et al., 2022).

Terrain ruggedness may be particularly important for understanding puma space use and coexistence with jaguars. Although ruggedness effects are often considered in mountainous systems, even subtle topographic variation in lowland Amazonian forests can influence drainage, vegetation structure, trail formation, prey movement, and predator travel routes (Detto et al., 2013; Riley et al., 1999). Structurally complex terrain may provide alternative movement pathways or microhabitats that reduce direct overlap with dominant competitors while still allowing access to prey-rich areas. Thus, terrain may not simply increase or decrease puma space use directly; it may modify the strength of puma associations with prey and jaguars.

Temporal activity provides another important axis of coexistence. Predators may adjust diel activity to maximize prey encounters while reducing risk or competition (Luttbeg and Sih, 2004). Pumas often overlap temporally with prey, though the strength of this relationship varies among systems (Ávila-Nájera et al., 2016; Ayala et al., 2021; Azevedo et al., 2018; Foster et al., 2013). If jaguars strongly suppress puma activity, pumas may reduce temporal overlap with jaguars. Alternatively, high temporal overlap may indicate that coexistence is mediated more strongly by prey selection, fine-scale spatial segregation, or habitat structure than by broad diel partitioning.

Building on a broader camera-trap monitoring effort, we provide an integrated assessment of puma density, space use, and diel activity patterns in a human-impacted Amazonian landscape of southeastern Peru. Specifically, we asked: (1) What is the density of pumas in the study area based on spatially explicit capture-recapture models? (2) Which detection covariates influence puma detection in SECR and occupancy models? (3) Is puma space use better explained by anthropogenic disturbance, habitat and environmental structure, prey availability, jaguar activity, or landscape-mediated biotic interactions? and (4) Do pumas show temporal overlap with prey and jaguars? We predicted that puma space use would increase with prey activity and decrease with anthropogenic disturbance. For jaguars, we considered two alternative predictions: pumas may avoid areas of high jaguar activity if competitive avoidance dominates, or they may be positively associated with jaguar activity if both felids use shared high-quality areas. We further predicted that terrain ruggedness would modify puma associations with jaguars or prey by increasing fine-scale spatial heterogeneity. Finally, we predicted high temporal overlap between pumas and prey, and either reduced or high puma-jaguar overlap depending on whether temporal partitioning is a primary coexistence mechanism in this landscape.

Methods

Study area

The study was conducted along the Las Piedras River in Madre de Dios, southeastern Peru, centered around the Hoja Nueva research station near Puerto Maldonado (UTM 19L 442212E, 8667724N; Figure 1). Madre de Dios is characterized by humid lowland Amazonian forest, broadly divided into two main habitat types: Floodplain forests are seasonally inundated and often have dense understory vegetation, whereas terra firme forests occur on higher, non-flooded ground and are generally associated with more leached soils, more open understory structure, and taller woody vegetation (Bredin et al., 2020).

The study area includes a mosaic of land-use types. Privately protected areas and forest concessions occur alongside areas used for Brazil nut extraction, ecotourism, small-scale agriculture, logging, and subsistence hunting. Human land use is more intensive in the southeastern portion of the study area near the community of Lucerna, where agriculture, logging, road access, and settlement expansion are more pronounced. Within the spatial extent shown in Figure 1, anthropogenic disturbance - including settlements, agriculture, and deforestation - increased substantially during the study period. This combination of intact forest, moderate human disturbance, riverine habitat, and expanding access routes makes the Las Piedras landscape an important system for evaluating puma ecology and large-carnivore conservation in a changing Amazonian forest.

Camera trapping

Camera-trap data were collected across multiple survey grids and sampling periods between 2019 and 2026. The full dataset used for occupancy and activity analyses included camera stations distributed across approximately 275 km² of lowland forest and human-modified habitat (Figure 1). Camera stations were deployed in multiple survey grids, with survey-specific details provided in Table 1. For occupancy modeling and activity analyses, the final dataset included 263 camera stations and 25,730 active trap nights. This study forms part of a long-term camera-trap monitoring program in the Las Piedras landscape that has previously been used to evaluate felid activity and occupancy patterns, including ocelots and margays, as well as mammal responses to moderate anthropogenic disturbance (Zwicker, 2023; Zwicker and Gardner, 2024; Zwicker et al., 2024, Zwicker et al., 2025). Here, we extended this monitoring framework to pumas by integrating individually identified puma detections for SECR with a broader multi-year camera-trap dataset for space-use and activity analyses.

Table 1. Summary of survey grids: Grid, dates sampled, the number of stations, the number of trap nights, and the average camera spacing in m.

Camera-trap locations were selected based on predefined coordinates and installed facing mammal trails. Cameras were mounted on trees approximately 40 cm above ground and configured to record 10-second videos with a one-second trigger delay. Paired camera traps were used where possible, although single-camera stations occurred in some survey grids due to equipment limitations. Cameras were placed both on and off trails to sample a range of movement pathways and habitat conditions. Trail placement was recorded as a covariate because large felids commonly use trails and linear features as movement routes, which can strongly influence detection probability (Harmsen et al., 2010).

All camera-trap records were identified to species level by the authors, and videos that did not allow reliable species-level identification were excluded. For activity analyses and prey/jaguar indices, independent detection events were defined as records of the same species at the same camera station separated by at least one hour, unless individuals could be distinguished based on unique physical characteristics. This one-hour threshold reduced temporal clustering from repeated videos of the same event while retaining biologically meaningful activity information.

Covariates

Spatial covariate processing and map preparation were conducted in QGIS (QGIS Development Team, 2022), using remote-sensing and spatial data accessed through Copernicus and NASA Earthdata resources where applicable (European Commission, 2026; NASA, 2026). Statistical analyses were conducted in R (R Core Team, 2021) using RStudio (RStudio Team, 2024).

We evaluated site-level covariates representing factors that may influence detection probability, anthropogenic disturbance and access, habitat and environmental structure, biotic associations, and landscape-mediated biotic interactions (Table 2). Covariates were derived from camera metadata, field classification (e.g., trail placement), spatial layers, and camera-trap detection rates.

Table 2. Site-level covariates evaluated in SECR and occupancy analyses of puma detection and space use.

Table legend: Means and ranges are reported using raw covariate values prior to standardization. Continuous covariates were standardized prior to model fitting; categorical covariates are summarized by their observed levels. Jaguar and prey indices were calculated as camera-station detection rates per active trap night. Individual sex and survey session were evaluated as SECR candidate covariates but are described in the Methods rather than included as site-level covariates.

Detection covariates included survey grid, trap nights, and trail placement. Survey grid accounted for differences in camera layout, sampling year, deployment area, and survey conditions. Trap nights represented total active camera effort at each station. Trail placement indicated whether the camera was placed on or off a trail and was evaluated because pumas may preferentially use trails as movement corridors, increasing detection probability.

Anthropogenic covariates included disturbed cover, distance to nearest mapped disturbance, distance to nearest road or access route, and distance to the Lucerna community. Anthropogenic disturbance layers were generated based on Sentinel-2 Level-2A true color imagery from July of the corresponding year, downloaded from the European Commission’s (2026) Copernicus Browser. Disturbance was defined as any human-induced alteration to the canopy, identified using Sentinel-2 Level-2A true color imagery from July of the corresponding year, downloaded from the European Commission’s (2026) Copernicus Browser. Included were visible settlements, agricultural clearings, and deforested areas (Figure 1). Distance-based covariates were calculated as Euclidean distances from each camera station to the nearest mapped feature. These variables were used to evaluate whether pumas responded to local disturbance intensity, proximity to human-modified habitat, road access, or settlement proximity.

Habitat and environmental covariates included habitat type, season, elevation, distance to river, Enhanced Vegetation Index (EVI), terrain ruggedness, and elevation variability. Habitat type was classified as floodplain or terra firme. EVI was used as a remotely sensed index of vegetation productivity because dense tropical forest canopy can limit the variation and ecological usefulness of simpler forest-cover metrics; vegetation indices are widely used to evaluate ecological responses to environmental variation, and EVI is less prone to saturation than NDVI in high-biomass tropical vegetation (Huete et al., 2002; Pettorelli et al., 2005). Elevation was extracted from a digital elevation model (DEM) which had been obtained in 2015 from the local forestry department in Puerto Maldonado (SERFOR, 2015). Terrain ruggedness was derived from the Terrain Ruggedness Index (TRI), which measures the change in elevation between a grid cell and its neighboring cells (Riley et al., 1999) and was used to represent local topographic heterogeneity. Elevation variability was calculated as the standard deviation of elevation around each camera station and used as an additional measure of local topographic heterogeneity. Distance to river was calculated as the Euclidean distance from each camera station to the nearest mapped river or watercourse.

Biotic covariates included a prey index and a jaguar index. The prey index was calculated as the number of independent medium-sized prey detections divided by active trap nights at each camera station. Medium-sized prey species (1–15 kg) were included because they are known or likely components of puma diets in Neotropical forests (de Oliveira and Medellín, 2002; Emmons, 1987; Karandikar et al., 2022), and pumas are broadly opportunistic predators whose diets often vary with local prey availability (Karandikar et al., 2022). The jaguar index was calculated as independent jaguar detections divided by active trap nights at each camera station. These indices were treated as relative activity or detection-rate indices rather than estimates of abundance, because they do not explicitly account for imperfect detection.

All continuous covariates were standardized to mean zero and unit variance prior to model fitting. Categorical covariates were included as factors. We assessed collinearity among continuous covariates using Pearson correlation coefficients. Covariate pairs with |r| ≥ 0.70 were not included in the same occupancy model. Excluded pairings included elevation with distance to river, elevation with EVI, distance to Lucerna with distance to disturbance, distance to Lucerna with distance to road, distance to disturbance with distance to road, and terrain ruggedness with elevation variability.

Spatially explicit capture-recapture density estimation

For SECR analyses, we used data from two seasonal camera-trap sessions (2023 and 2023W; 7,679 trap nights) at the same 48 camera stations. Ten stations from grid 2023W were excluded because of operational failures. These stations had an average spacing of 1,364.7 m and were originally arranged in a grid design suitable for large felids (Tobler et al., 2013). Because pumas and jaguars have broadly comparable movement scales in many Neotropical systems, camera arrays designed for jaguars can also be effective for puma SECR studies when individual pumas can be reliably identified (Kelly et al., 2008).

We estimated puma density using spatially explicit capture-recapture models implemented in the secr package in R (Efford, 2023). Individual pumas were identified from natural markings and physical characteristics following methods previously used for pumas in camera-trap studies (Kelly et al., 2008; Negrões et al., 2010; Rich et al., 2014; Alexander and Gese, 2018). We used both time-stable features, such as undercoat spot patterns, kinked tail shape, tail morphology, and tail-tip coloration, and time-variable features, such as scars and botfly marks (Figure 2). To minimize the risk of false individual assignment, only detections for which two independent observers agreed on individual identity were included in the SECR analysis. Ambiguous detections were excluded.

Capture histories were constructed using five-day sampling occasions. Site-specific effort was incorporated into the capture histories to account for variation in camera availability and functionality. The trap detector type was treated as a proximity detector, allowing multiple detections of identified individuals across occasions. Individual sex was included as an individual covariate in candidate models.

We evaluated alternative detection functions and selected a half-normal detection function for final density modeling because it produced biologically reasonable and stable estimates across buffer widths. We used a mask spacing of 0.5 km, consistent with previous recommendations that mask spacing between 0.5 and 2 km is sufficient for puma SECR applications (Noss et al., 2012). Rather than using ½ mean maximum distance moved (½ MMDM) to define the effective sampling area, we evaluated density stabilization using effective sampling area plots. Because ½ MMDM can overestimate density when used to define effective area in camera-trap studies (Gerber et al., 2012; Noss et al., 2012), we selected a buffer of approximately 14.3 km, corresponding to eight times the initial estimate of sigma. This value was similar to the 15 km buffer found to be adequate for pumas by Noss et al. (2012), and density estimates stabilized at this buffer width.

We fitted a candidate set of SECR models evaluating the effects of trail placement and session on baseline detection probability (g0), and sex and session on the spatial scale parameter (σ). Density was modeled as constant because the SECR dataset included a small number of identified individuals and because density covariate models were not supported during exploratory testing. We also evaluated whether adding sex to detection probability improved the top detection model. Models were compared using corrected Akaike Information Criterion (AICc; Burnham and Anderson, 2002), and the model with the lowest AICc was used for density inference. We report density as individuals/100 km², and σ in kilometers. We also derived an approximate 95% home-range area from the estimated σ using the circular home-range radius from the half-normal detection function.

Occupancy modeling/puma space use

We modeled puma space use using single-species occupancy models implemented in the unmarked package in R (Chandler et al., 2023). Because puma home ranges are larger than the spacing among camera stations and because detections were pooled across multiple survey grids and years, we interpreted occupancy estimates as probability of site use rather than strict demographic occupancy (MacKenzie and Nichols, 2004).

Detection/non-detection histories were constructed using camtrapR with five-day sampling occasions. Camera operation matrices were generated from camera-specific deployment and retrieval dates and camera-functioning records, so inactive periods were coded as unavailable rather than as true non-detections. We first evaluated detection probability using candidate models that included survey grid, trail placement, season, and trap nights. Based on model support, we retained a fixed detection structure including survey grid and trail placement (p ~ grid + trail) for subsequent occupancy model selection. This structure accounted for spatial and temporal variation among survey grids and the strong influence of trail placement on puma detection probability.

We constructed a hypothesis-driven candidate set of occupancy models representing five ecological themes: (1) anthropogenic disturbance and access, (2) habitat and environmental structure, (3) biotic associations, (4) integrated biotic-landscape models, and (5) landscape-mediated biotic associations. Anthropogenic models tested disturbed cover, distance to disturbance, distance to roads, and distance to Lucerna. Habitat/environmental models tested EVI, habitat type, distance to river, elevation, terrain ruggedness, and elevation variability. Biotic models tested prey index, jaguar index, and their additive effects. Integrated models tested whether prey and jaguar effects persisted when combined with environmental or disturbance covariates. Interaction models tested whether terrain ruggedness, elevation variability, river proximity, elevation, or EVI modified puma associations with jaguar or prey indices.

Models were ranked using Akaike’s Information Criterion (AIC; Akaike, 1973). Models within ΔAIC < 2 were considered competitive. Because multiple models received similar support, we used conditional model averaging across the competitive model set to estimate covariate effects. Conditional model averaging was performed using the MuMIn package in R (Barton, 2024). Conditional averaging considered each parameter only among models in which it appeared, thereby reflecting parameter-specific support while accounting for model-selection uncertainty. Mean predicted space-use probability and detection probability were extracted from the top-ranked occupancy model.

Activity patterns and temporal overlap

We assessed diel activity patterns of pumas and evaluated temporal overlap between pumas, medium-sized prey species, and jaguars using independent camera-trap records from the full camera-trap dataset. Time stamps were extracted from camera metadata and converted to radians for circular kernel density estimation.

We first described puma activity using activity density, histogram, and radial plots generated in camtrapR (Niedballa et al., 2016). We then quantified temporal overlap between pumas and two biologically relevant groups: medium-sized prey species included in the prey index, and jaguars. Temporal overlap was estimated using the coefficient of overlap, Δ, implemented in the R package overlap (Meredith et al., 2024) following Ridout and Linkie (2009). Because all pairwise comparisons included more than 75 detections, we used the Δ4 estimator, which is recommended for larger sample sizes. We calculated 95% confidence intervals using 999 smoothed bootstrap replicates. Overlap curves were plotted on a 24-hour scale, and the shared area under the two activity density curves was shaded to visualize temporal overlap.

Results

Puma detections and SECR density

Across the SECR dataset, we recorded 85 individual puma captures. Four detections were excluded because of uncertainty in individual identity, leaving 81 strict-ID detections from seven adult pumas. Identified individuals included four males and three females.

The AICc-selected SECR model included trail placement as a covariate on detection probability, with density and σ held constant: D(.) g0(trail) σ(.). This model had the lowest AICc among the candidate set (AICc = 370.557), followed by a model including trail and sex.

on detection probability (AICc = 371.453). Models including session effects or sex/session effects on σ received substantially less support (see Table 3).

Table 3. Comparison of fitted SECR candidate models estimating puma density.

Table legend: Models are ranked by corrected Akaike Information Criterion (AICc). Density was modeled as constant across models, while trail placement, session, and individual sex were evaluated as covariates on detection probability or the spatial scale parameter.

The top model estimated puma density at 1.012 individuals/100 km² (SE = 0.440, 95% CI = 0.448 - 2.286). Detection probability was higher for cameras placed on trails than for off-trail cameras. The spatial scale parameter σ was estimated at 3.393 km (SE = 0.605, 95% CI = 2.399 - 4.799). Based on this estimate, the derived 95% home-range area was 182.203 km². The observed sex ratio in the identified sample was slightly male biased, with males representing 54.5% and females 45.5% of individuals.

Occupancy model selection and puma space use

The occupancy detection-history matrix included 263 camera stations, with an average of 19.27 five-day sampling occasions per station and a maximum of 47 occasions per station. Puma detections occurred at 114 stations. Across all five-day occasions, there were 334 puma detections, 4,714 non-detections, and 7,266 unavailable observations due to camera inactivity or non-overlapping survey periods.

Detection probability was strongly influenced by trail placement and survey grid. In the final model-averaged estimates, trail placement had a strong positive effect on detection probability (β = 2.011 ± 0.223, p < 0.001), indicating substantially greater detection probability at trail cameras than off-trail cameras. Detection probability also varied among survey grids, with lower detection in the 2021M grid (β = -0.694 ± 0.181, p < 0.001) and 2025LG grid (β = -0.604 ± 0.287, p = 0.036) relative to the reference grid. Mean predicted detection probability from the top occupancy model was 0.121 ± 0.017.

The best-supported occupancy models consistently included jaguar index, prey index, terrain ruggedness, and interactions between terrain ruggedness and biotic covariates. The top-ranked model included terrain ruggedness, jaguar index, prey index, and a terrain ruggedness × jaguar index interaction (AIC = 2070.11, ΔAIC = 0.00, AIC weight = 0.22). Two additional models were within ΔAIC < 2: a model including terrain interactions with both jaguar and prey indices (ΔAIC = 0.31, AIC weight = 0.19), and a model including a terrain ruggedness × prey index interaction with jaguar index as an additive effect (ΔAIC = 1.19, AIC weight = 0.12). The additive model containing terrain ruggedness, jaguar index, and prey index was near-competitive but fell just outside ΔAIC < 2 (ΔAIC = 2.17). Anthropogenic-only models and habitat/productivity-only models received little support relative to biotic and terrain-mediated models. Figure 3 demonstrates predicted puma space-use probability as a function of jaguar detection index and terrain ruggedness.

Conditional model averaging across the three competitive models indicated that puma space use increased with both jaguar and prey indices (see Table 4). Jaguar index had a positive effect on puma space use (β = 1.090 ± 0.419, z = 2.604, p = 0.009), and prey index also had a positive effect (β = 0.541 ± 0.243, z = 2.228, p = 0.026). Terrain ruggedness had a weak positive main effect, but this effect was not statistically supported (β = 0.247 ± 0.188, z = 1.315, p = 0.188). The interaction between terrain ruggedness and jaguar index was negative (β = -0.610 ± 0.265, z = 2.304, p = 0.021), indicating that the positive association between puma space use and jaguar activity weakened as terrain ruggedness increased. The terrain ruggedness × prey index interaction was also negative but not statistically supported (β = -0.313 ± 0.213, z = 1.471, p = 0.141). Mean predicted puma space-use probability from the top-ranked model was 0.573 ± 0.069.

Table 4. Conditional model-averaged coefficient estimates from competitive puma occupancy models.

Occupancy parameters describe puma space-use probability, and detection parameters describe detection probability. Continuous covariates were standardized prior to analysis. Coefficients are shown on the logit scale. Significance codes: *p < 0.001, p < 0.01, *p < 0.05,. p < 0.10.

Activity patterns and temporal overlap

The activity dataset included 406 independent puma detections, 5,532 independent prey detections, and 220 independent jaguar detections. Puma activity was broadly crepuscular to nocturnal, with activity concentrated from late afternoon through early morning and reduced activity during midday. The radial activity plot showed puma activity distributed across much of the 24-hour cycle, but with strongest activity during nocturnal and low-light periods.

Pumas exhibited high temporal overlap with prey species (see Figure 4A). The coefficient of overlap for pumas and prey was Δ4 = 0.888, with a 95% bootstrap confidence interval of 0.844 - 0.922. Overlap was strongest during crepuscular and nocturnal periods, particularly around dawn and from early evening through midnight.

Temporal overlap between pumas and jaguars was also high (see Figure 4B). The coefficient of overlap for pumas and jaguars was Δ4 = 0.834, with a 95% bootstrap confidence interval of 0.782 - 0.902. Both felids showed substantial nocturnal and crepuscular activity, although density curves indicated some differences in the magnitude and timing of activity peaks. These results suggest that pumas did not strongly avoid jaguars at the broad diel scale.

Discussion

This study provides an integrated assessment of puma density, space use, and temporal activity in a human-impacted lowland Amazonian landscape. We found that pumas occurred at an expected density, were detected more frequently on trails, used areas with greater prey and jaguar activity, and showed high temporal overlap with both prey and jaguars. The clearest ecological result was that terrain ruggedness weakened the positive relationship between puma space use and jaguar activity as terrain ruggedness increased. Together, these findings suggest that pumas in the Las Piedras landscape are not broadly avoiding jaguars at the spatial or temporal scales sampled here. Instead, puma space use appears to be structured by biologically active areas, with fine-scale landscape structure influencing how pumas and jaguars share space. While the anthropogenic covariates incorporated here were not strongly associated with puma space use, this should not be interpreted as evidence that human disturbance is unimportant. A growing body of camera-trap research from the Las Piedras landscape suggests that Amazonian carnivores and disturbance-sensitive mammals may respond to human impacts through multiple ecological pathways, including changes in temporal activity, detection, and community composition, rather than through simple spatial avoidance alone (Zwicker and Gardner, 2024; Zwicker et al., 2024, Zwicker et al., 2025).

SECR density and puma detection

Our SECR estimate of 1.012 individuals/100 km² provides one of the few spatially explicit density estimates for pumas in lowland Amazonian forest and establishes a conservative baseline for southeastern Peru. Comparisons with previously published Neotropical puma density estimates should be made cautiously because many studies used non-spatial approaches or effective sampling areas based on MMDM or ½MMDM, which can inflate density estimates relative to spatially explicit methods (Gerber et al., 2012; Noss et al., 2012). Although our estimate is lower than several published estimates from Neotropical systems, including the Bolivian Chaco (Rich et al., 2014), Tehuacán-Cuicatlán Biosphere Reserve in Mexico (Vega-Flores and Farías-González, 2021), central Brazil (Negrões et al., 2010), and Manu National Park (Janson and Emmons, 1990), those differences may reflect methodological variation as much as biological variation. For example, Kelly et al. (2008) reported a higher non-spatial estimate for pumas in Belize, but spatial reanalysis of the same dataset produced a much lower estimate of 0.60 individuals/100 km² (Rich et al., 2014). Comparable estimates also exist from the Ecuadorian Amazon ranging from 0.26 - 1.92 individuals/100 km² (Gil-Sánchez et al., 2021). Within this context, our estimate should not be interpreted simply as unusually low, but rather as a conservative spatial estimate within the expected range for tropical forest pumas. Its main value is as a rigorous baseline for future monitoring in a landscape undergoing increasing road access, forest conversion, and human activity.

Several factors may contribute to the lower-to-middle density estimate. Dense Amazonian forest may support lower puma densities than more open or seasonal systems where prey are more concentrated, or easier to detect and catch, especially if pumas benefit from thick understory that provides stalking cover while still allowing greater visibility than typical rainforest understory (Guerisoli et al., 2019; Kelly et al., 2008). Coexistence with jaguars may also influence puma density or spatial organization through exploitative competition (Gil-Sánchez et al., 2021), prey partitioning, or localized avoidance, although our occupancy and activity results do not support broad-scale exclusion. Human access, hunting, and disturbance may further affect prey communities or carnivore behavior in ways not fully captured by the SECR sampling grid. Finally, even though we only excluded four out of 85 detections, our strict individual-identification criteria likely produced a conservative dataset. By excluding ambiguous detections, we reduced the risk of false individual assignment but also reduced sample size and precision.

The wide confidence interval around the density estimate reflects the small number of identified individuals and the difficulty of estimating density for low-density carnivores in dense tropical forests. Nevertheless, the estimate is valuable because it is based on strict individual identification, an explicit spatial model, site-specific effort, and an effective sampling area selected from density stabilization rather than ½ MMDM. Future studies that expand the sampling area, increase the number of paired camera stations to obtain images from both flanks and improve individual identification, or apply spatial mark-resight or partial-identity models could improve precision and allow stronger inference about spatial variation in density. For conservation, the immediate value of this estimate is as a repeatable baseline: future surveys can test whether puma density changes as roads, settlement, hunting access, and forest conversion increase across the Las Piedras landscape.

Trail placement strongly influenced puma detection in both SECR and occupancy analyses. In the SECR model, the AICc-selected model included trail effects on detection probability, and in occupancy models, trail placement also had a strong positive effect on puma detection. This is consistent with Harmsen et al. (2010), who showed that forest mammals use trails differentially and that camera placement on trails can substantially influence camera-trap detection rates. Cameras placed on trails are more likely to detect animals that use linear routes, like large felids, whereas off-trail cameras may better represent broader habitat use. In our study, explicitly modeling trail placement helped separate variation in detection from variation in space use, but the strong trail effect also underscores that raw detection rates should not be interpreted as habitat preference without accounting for camera placement.

This issue is also relevant to interpreting the jaguar and prey indices. Although puma detection was modeled as a function of trail placement and survey grid, the prey and jaguar indices were detection-rate covariates rather than abundance estimates and may reflect a combination of true activity, abundance, movement behavior, and detectability. Thus, the positive association between puma space use and jaguar index could partly reflect shared use of movement corridors or shared detectability rather than only ecological attraction to jaguar-active areas. We interpret the indices as a relative activity indices, not as true density. Multi-species occupancy models, spatial co-occurrence models, or independent density estimates for jaguars, and prey would help clarify the mechanisms behind the observed associations. Even with this caveat, the consistency of jaguar, prey, and terrain effects across the top occupancy models suggests that biotic activity and landscape structure are important components of puma space use in this system.

Biotic interactions

The occupancy results indicate that puma space use was more strongly associated with biotic covariates than with the measured anthropogenic or broad habitat/productivity covariates. Puma space use increased with prey index, supporting a bottom-up interpretation in which pumas concentrate activity in areas with greater prey activity or availability. This pattern is consistent with puma ecology across diverse systems, where prey availability, prey vulnerability, or prey encounter rates influence movement, habitat selection, and space use (Ávila-Nájera et al., 2020; Coon et al., 2020; Guerisoli et al., 2019; Negrões et al., 2010). In our study, this interpretation is strengthened by high temporal overlap between pumas and prey. Similar moderate to high temporal overlap has been reported in other Neotropical systems (Ávila-Nájera et al., 2016; Ayala et al., 2021; Azevedo et al., 2018; Foster et al., 2013). This suggests that pumas in the Las Piedras landscape are using both space and time in ways that likely increase encounter rates with medium-sized prey.

Puma space use also increased with jaguar index. We do not interpret this as direct attraction to jaguars. A more plausible explanation is that pumas and jaguars are both responding to high-quality areas that combine prey activity, cover, movement routes, or habitat conditions not fully captured by individual environmental covariates. Positive or neutral spatial associations between pumas and jaguars have been reported in other Neotropical systems, including the Brazilian Amazon (Negrões et al., 2010). However, puma-jaguar coexistence appears to be highly context-dependent (Ávila-Nájera et al., 2020; Curveira-Santos et al., 2022; Sollmann et al., 2012). Carvalho et al. (2024) showed that environmental and anthropogenic variables can influence large felids in degraded Amazonian landscapes, while the presence of primary forest may support spatial co-occurrence between pumas and jaguars. At the same time, Gil-Sánchez et al. (2021) found that in better-conserved areas of the Ecuadorian Amazon, higher jaguar densities were associated with lower puma densities. In Las Piedras, remaining connected forest may therefore allow both species to use the same productive areas without necessarily indicating direct facilitation.

This spatial pattern was complemented by the activity results. Pumas showed high temporal overlap with jaguars, consistent with previous studies documenting substantial diel overlap between jaguars and pumas (Ávila-Nájera et al., 2016; Ávila-Nájera et al., 2020; Ayala et al., 2021; Foster et al., 2013). High temporal overlap suggests that pumas do not avoid jaguars primarily through broad diel shifts in activity. However, this does not rule out finer-scale temporal avoidance, such as short-term changes in trail use after recent jaguar passage, or spatial avoidance at scales smaller than our camera-station analysis. The activity results therefore support the interpretation that broad temporal partitioning is not the dominant coexistence mechanism in this system; dietary partitioning, microhabitat use, prey selection, and terrain-mediated spatial structure are more plausible mechanisms.

The positive puma-jaguar association also makes sense when considered through the lens of competition theory. Interference competition and intraguild killing are most likely when the dominant predator is 2 - 5.4 times larger than the subordinate predator (Donadio and Buskirk, 2006). Because jaguars and pumas fall below this size‐ratio threshold, the risk of intraguild killing of pumas is minimal. In such systems, exploitative competition may be more important than direct interference. Comparative work on Neotropical felids shows that coexistence among sympatric felids can be facilitated by partitioning across diet, habitat, and activity dimensions (Di Bitetti et al., 2010; de Oliveira and Medellín, 2002).

Dietary partitioning is a particularly important alternative explanation for why pumas and jaguars can overlap strongly in both space and time. Classic work by Emmons (1987) at Cocha Cashu in Manu National Park showed that sympatric felids in Amazonian rainforest partition prey resources, with larger felids differing in prey composition and size classes. More recent work in the Amazon piedmont of Bolivia also found that jaguars and pumas differed in food habits, with jaguars relying more on large- and medium-bodied prey and pumas using more medium- and small-bodied prey, producing a relatively low trophic niche overlap of 0.46 (Flores-Turdera et al., 2021). Within this context, both felids may use the same prey-rich forest areas while reducing competition through prey-size selection and fine-scale hunting behavior.

Environmental and anthropogenic variables

The negative interaction between terrain ruggedness and jaguar index provides the strongest evidence that landscape structure modified puma space use. Puma space use increased with jaguar index, but this positive relationship weakened as terrain ruggedness increased. In less rugged areas, pumas and jaguars may be more likely to use the same trails, ridgelines, low-resistance paths, or prey-rich movement corridors, producing a strong positive association. In more rugged areas, pumas may have access to a greater variety of alternative travel routes, stalking cover, drainage features, or microhabitats, allowing them to use space less tightly linked to jaguar activity. This does not imply that rugged terrain causes pumas to avoid jaguars. Rather, ruggedness appears to reduce the degree to which puma space use tracks jaguar activity.

This distinction is important. Terrain ruggedness alone did not have a strong independent effect on puma space use. Its importance emerged primarily through its interaction with jaguar activity. In other words, terrain may not simply make sites more or less suitable for pumas; it may alter the context in which pumas respond to dominant competitors. In lowland Amazonian forest, even subtle topographic variation can influence drainage, vegetation structure, animal trails, and prey movement (Detto et al., 2013). Rugged or locally heterogeneous terrain may therefore increase the number of usable movement pathways and reduce forced overlap along shared travel corridors. This interpretation aligns with the broader idea that carnivore co-occurrence is context-dependent and that habitat structure can mediate whether species segregate, overlap, or respond independently across space (Curveira-Santos et al., 2022).

Contrary to our initial prediction, anthropogenic covariates were not strongly supported in the final occupancy models. Models containing disturbed cover, distance to disturbance, distance to road, or distance to Lucerna ranked poorly relative to biotic and terrain-mediated models. This suggests that, within the spatial extent and disturbance gradient sampled here, puma space use was more strongly associated with prey activity, jaguar activity, and terrain structure than with our measured disturbance variables. This result should not be interpreted as evidence that human disturbance is unimportant. Rather, it indicates that our specific disturbance metrics did not explain puma space use as well as biotic and terrain covariates at the camera-station scale.

This interpretation is consistent with recent local evidence that mammal responses to moderate disturbance in the Peruvian Amazon may be subtle and multidimensional. Ocelots in the Las Piedras landscape, for example, appear to respond to moderate anthropogenic impacts through changes in temporal niche structure rather than through strong shifts in spatial distribution (Zwicker and Gardner, 2024). Work on persecuted mammals in the Peruvian Amazon also suggests that hunted or disturbance-sensitive species can function as indicators of moderate human disturbance, emphasizing that anthropogenic effects may be expressed through changes in community composition, activity, or detection patterns rather than simple presence–absence responses (Zwicker et al., 2025). Our puma results fit this broader pattern: pumas did not show strong spatial avoidance of mapped disturbance variables, but this does not rule out disturbance effects operating through prey depletion, hunting pressure, road access, or altered activity patterns.

Several mechanisms could explain the weak support for anthropogenic covariates. First, pumas are behaviorally flexible and can persist in human-modified landscapes where forest cover, prey, and refuge habitat remain (Gil-Sánchez et al., 2021; Guerisoli et al., 2019; Nanni et al., 2023; Quiroga et al., 2016). Second, the Las Piedras landscape still contains extensive connected forest, which may buffer pumas from moderate, patchy disturbance. Third, our disturbance covariates were coarse spatial proxies. Disturbed cover did not differentiate among settlements, agriculture, logging, roads, hunting access, or other land-use types, and distance metrics may not capture actual human presence, hunting intensity, dog activity, traffic, or persecution risk. These unmeasured processes may influence pumas and prey even where mapped disturbance is not strongly associated with space use. Finally, disturbance effects may occur at broader spatial scales than the camera-station covariates captured, especially for wide-ranging carnivores.

Habitat and productivity covariates also received less support than biotic and terrain-mediated models. EVI and habitat type may not have captured the fine-scale structural variation relevant to puma movement in a mostly closed-canopy forest landscape. The narrow range of EVI values across dense forest may also have limited its explanatory power. This result does not mean that habitat quality is unimportant; rather, habitat effects may be expressed indirectly through prey activity, jaguar activity, and the movement pathways created by terrain and forest structure.

Lastly, it is important to point out that occupancy models were interpreted as space-use models rather than true occupancy models because puma home ranges exceed camera-station spacing and because data were collected across multiple grids and years. This interpretation is appropriate for wide-ranging carnivores, but ψ should be understood as probability of site use during the survey rather than demographic occupancy.

Conclusion

Our results highlight the importance of maintaining intact, connected, prey-rich forests for pumas and other large carnivores in southeastern Peru. Puma space use was most strongly associated with prey and jaguar activity, indicating that pumas are embedded within broader predator-prey and carnivore assemblage dynamics. Conservation strategies focused only on pumas may therefore be less effective than strategies that protect prey communities, reduce hunting pressure, and maintain habitat connectivity for the full carnivore guild. The conservative density estimate makes this especially important: populations of wide-ranging carnivores can appear stable in presence-only monitoring while still occurring at low densities and remaining sensitive to prey depletion or increasing human access.

The terrain-mediated puma-jaguar association suggests that structural heterogeneity may contribute to carnivore coexistence. In lowland Amazonian landscapes, subtle variation in topography, drainage, and forest structure may provide alternative movement routes and microhabitats that allow large carnivores to share space. Conserving contiguous forest mosaics that include floodplain and terra firme habitats, river corridors, and topographically heterogeneous areas may therefore support coexistence among large felids. Together with recent camera-trap studies of ocelots, margays, and persecuted mammals in the Peruvian Amazon, our results suggest that conservation monitoring in this landscape should consider multiple response pathways, including density, site use, detection, temporal activity, and community composition (Zwicker and Gardner, 2024; Zwicker et al., 2024, Zwicker et al., 2025).

Although measured anthropogenic covariates were not strongly supported in our occupancy models, continued expansion of roads, agriculture, settlements, and hunting access could erode prey populations, increase conflict risk, and reduce habitat connectivity. The relevance of persecuted mammals as indicators of moderate disturbance further underscores the need to monitor not only large carnivores, but also the prey and hunted mammal communities that sustain them (Zwicker et al., 2025). Long-term camera-trap monitoring across the Las Piedras landscape is therefore essential for detecting whether puma density, space use, and temporal behavior shift as disturbance intensifies.

Overall, this study provides a baseline for puma conservation in southeastern Peru and contributes to a growing understanding of Neotropical carnivore coexistence. Pumas occurred at a density consistent with expectations for tropical forest populations estimated using conservative spatial methods, used prey- and jaguar-active areas, overlapped temporally with both prey and jaguars, and showed evidence that terrain structure modified their association with jaguar activity. These findings suggest that conserving Amazonian pumas will require protecting not only forest cover, but also the prey communities, large-carnivore assemblages, and fine-scale landscape heterogeneity that sustain predator coexistence in human-impacted tropical forests.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During preparation of this manuscript, the authors used ChatGPT for language editing and organizational support. All scientific content, analyses, interpretations, citations, and conclusions were reviewed, verified, and approved by the authors, who take full responsibility for the final manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Mingsheng Hong, China West Normal University, China

Reviewed by: Patrick Ross, Duke University, United States

Rebecca Zug, Universidad San Francisco de Quito, Ecuador