Annual and spatial variation in composition and activity of terrestrial mammals on two replicate plots in lowland forest of eastern Ecuador

Study site

Research was conducted at Tiputini Biodiversity Station (TBS), Orellana Province, Ecuador (ca 0°37′S, 76°10′W, ∼190–270 m above sea level). Work at Tiputini Biodiversity Station was conducted in accordance with research permit No. 0005-FAU-MAE-DPOPNY, Ministerio del Ambiente, Ecuador. TBS is located adjacent to Yasuní National Park on a tract of largely undisturbed lowland forest within the biologically diverse Yasuní Biosphere Reserve (Bass et al., 2010). The station and nearby areas are dominated by terra firme forest but várzea forest, palm swamps, and various successional habitats also are present. Mean annual precipitation at Yasuní Research Station, approximately 30 km WSW of TBS, is about 3,100 mm. Two plots (Harpia, Puma; ∼1 km × 1 km each) were established in terra firme forest during 2001. Both plots are gridded (100 × 200-m grid lines) and marked (Fig. 1). Transects are approximately 0.5–1 m wide and are typically cleared during November-December of each year. The Harpia plot ranges from ∼201 to 233 m elevation and is characterized by more dissected upland forest. The Puma plot has less relief overall although elevation range is similar, from ∼209 to 235 m. Both areas experience partial, temporary inundation (approximately 5 to 10 ha, depending on height of the flood) when small streams back up as the Tiputini River rises; Puma has more areas that fill with persistent standing water during the rainy season (see Loiselle et al., 2007).

Camera traps

Cameras (Reconyx Hyperfire; Reconyx, Inc., Holmen, WI, USA) triggered by an infrared heat-and-motion detector were located ∼200 m apart in a grid of 16 cameras on each plot (Fig. 1); only one camera was placed at each location. Cameras were active from mid-January until mid-March, 2014–2017 (∼60 days/year). All cameras were set to record 5 images during each detection event, with a 1-sec delay between images. We set cameras with a minimum time of 5 min between sets of images. Cameras remained continuously activated (except when malfunctions occurred); date and time were automatically stamped on each image. Cameras were visited frequently to check batteries and camera placement (cameras were occasionally disturbed by Tayassu pecari). Assuming that individual animals within 100 m of the perimeter of the camera grid (i.e., half the distance between cameras) might be expected to pass in front of a camera at some point, then the 16 cameras would have covered an area of approximately 64 ha; the area sampled would be ∼100 ha if individuals within 200 m from the perimeter might be detected. Many individuals of large species (e.g., Panthera onca) certainly range well beyond the area covered by the camera grid; other, smaller species (e.g., Myoprocta pratti) may occupy areas sampled by only one camera. We calculated number of trap-days for each camera location as the time the camera was placed in operation until it was taken down or until the last image was taken (based on the date and time stamp on the images) if batteries had failed or other malfunctions had occurred. We combined records across months during each year.

Both plots were surveyed to record elevation (at 25 m intervals), locations of permanent markers, and stream courses. These data were used to produce GIS databases for permanent grid markers, streams, elevation, slope, and aspect (Loiselle et al., 2007). We used 50 and 100-m buffers around each camera to characterize environmental space around each camera location. Plots were first divided into 1 × 1-m grid cells and assigned unique numbers to each camera location. Thus, grid cells within either 50 or 100 m of each camera were identified by the same value. The environmental characteristics of each location were then determined using zonal statistics in GIS (SPATIAL ANALYST, ESRI, Redlands, CA, USA). Zonal statistics provide minimum, maximum, mean, and standard deviation of values for each environmental variable; camera identification values defined the zones. After examining correlations among variables, we retained mean and standard deviation of elevation (m), slope (degrees), and distance (Euclidean) from stream (m).

Analyses

We summarized images by species and date. We classified images as belonging to independent records if more than 30 min elapsed between consecutive images of the same species at a given location (O’Brien, Kinnaird & Wibisono, 2003; Datta, Anand & Naniwadekar, 2008; Blake et al., 2011). We indexed activity in terms of number of images/100 trap-days (hereafter referred to as capture rates; i.e., captures of images) irrespective of the number of individuals in each image (e.g., peccaries often were represented by multiple individuals). We do not assume that such rates are a true estimate of actual abundance (Burton et al., 2015) but simply assume that, for comparative purposes, such rates provide a reasonable estimate of activity (frequency with which animals pass in front of a camera) at a given camera location (Foster, Harmsen & Doncaster, 2010; Blake et al., 2011; Meyer et al., 2015; Blake et al., 2017). Capture rates may be affected by factors that do not relate directly to abundance so interpretation of results must keep these limitations in mind. The distance between camera locations in this study is considerably less than that found in most camera-trap studies (reviewed by Burton et al., 2015) and, as consequence, images from different cameras are not necessarily independent as individuals of a variety of species, particularly larger species (e.g., Panthera onca, Puma concolor, Leopardus pardalis, Tayassu peccari, Pecari tajacu, Tapirus terrestris) may be recorded at more than one location (Blake et al., 2014; Blake et al., 2016). Thus, rates of capture may reflect variation in spatial activity patterns as well as variation in numbers of individuals. Differences in yearly capture rates between plots were evaluated with a paired t-test. We used correlation (Spearman’s rank) tests to examine the level of similarity in capture rates of species between plots within a year and between years within a plot. We also calculated an interpolated jackknife estimate of number of species present during a given sample using program SPECRICH (Hines, 1996).

Spatial distribution of individual species (i.e., clumped, random, uniform; based on numbers of images at camera locations) was evaluated using Program PASSaGE Version 2.0.10.18 (Rosenberg & Anderson, 2011). PASSaGE computes means and variances from counts (e.g., number of images/camera) and calculates a series of indices to describe patterns of variation. We used the Index of Dispersion (ID; based on the variance-to-mean ratio) as an indication of whether distribution of images among camera locations was clumped (ID > 1.0), random (ID = 1.0), or uniform (ID < 1.0). Departure from random distribution was evaluated with a chi-square test (Rosenberg & Anderson, 2011). We also used PASSaGE to test for spatial autocorrelation in distribution of individual species (i.e., correlograms; Moran’s I) to examine if number of captures of a given species at one location was correlated with captures at cameras at different distances. Moran’s I (Moran, 1950) ranges from 1 to −1 with an expected value of ∼0 for large sample sizes and no spatial autocorrelation. Significance levels of correlations were examined with permutation tests. We used Mantel correlograms to determine if species composition (based on Jaccard similarity coefficients) at camera locations exhibited spatial autocorrelation depending on distances among cameras. We also used Mantel tests to examine whether the compositional similarity (all common species) between camera locations was related to distance between those locations (i.e., to determine if the composition was spatially autocorrelated). Significance levels were determined with permutation tests. Mantel correlograms were implemented in PASSaGE; Mantel tests were implemented in PC-ORD Version 6.12 (McCune & Mefford, 2011).

We used several approaches to analyze differences and similarities in community composition between plots and across years. First, we used a Non-metric Multi-dimensional Scaling ordination (NMS; Bray–Curtis distance measure) to graphically represent similarities (and differences) in species composition between plots and among samples, based on capture rates per species (Clarke & Warwick, 2001; McCune & Grace, 2002). Next, we used analysis-of-similarity (ANOSIM; Bray-Curtis similarity, standardized capture rate; see Clarke & Warwick, 2001) to compare the level of similarity in species composition among annual samples from a given plot (i.e., Puma or Harpia) to the level of similarity across all samples, to determine if plots differed in species composition more than expected by chance. The significance of the ANOSIM test statistic is determined by comparison with values obtained by a Monte Carlo randomization procedure. Ordination was performed with PC-ORD Version 6.12 (McCune & Mefford, 2011); ANOSIM was run with PRIMER Version 6.1.11 (PRIMER-E, 2008).

We used both indirect and direct gradient analyses to examine spatial variation in species composition based on camera location, with data from all years combined. We first used NMS to graphically portray variation in composition among the camera locations, without taking into account effects of environmental variables. Next, we ran detrended correspondence analysis (DCA) to detect gradients in species composition based on rates of capture in cameras. Lengths of the segments for each axis provide an index of the amount of species turnover among sample points (Clark, Palmer & Clark, 1999). Linear models are appropriate and advantageous when the standard deviation of ordination lengths is less than about 3; unimodal models are more appropriate when lengths are greater than 4 (see Clark, Palmer & Clark, 1999). In this case, gradients were all <2.0 so we used redundancy analysis (RDA) to evaluate the relationship between species composition and six environmental features at different camera locations (using 50 and 100-m buffers around each camera location. In RDA biplots, if species and environmental factors are represented by arrows, the cosine of the angle between the arrows approximates the correlation coefficient between the two variables. Arrows in the same direction indicate positive correlation and arrows in the opposite direction indicate negative correlations. Species located farther away from the origin provide more information regarding gradients among samples. DCA and RDA analyses were run on program CANOCO 4 (Ter Braak & Šmilauer, 1998).

We combined data across all years to compare the relative importance of typically terrestrial species (i.e., excluding primates and Coenodon prehensilis) by plotting capture rates from both plots combined (x-axis) vs capture rates from cameras located along trails at TBS (y-axis) following Pitman et al. (2001 see also Blake, 2007; Blake & Loiselle, 2009). Cameras along trails (Fig. 1; Blake et al., 2017) were located approximately 800 to 1,200 m apart. We calculated the slope of the line between plots and trails to test the null hypothesis that capture rates per species did not differ between the two sets of cameras. If true, the slope of the line should not differ from 1.0 (Pitman et al., 2001:2107). Points above a line with slope of 1.0 would indicate greater capture rate on trails; below the line, points would indicate higher rates on plots.

We took the same approach to compare capture rates of species on the two plots combined to capture rates of species at sites located farther from the plots. First, we compared results from cameras located at a site in the eastern part of Yasuní National Park, approximately 90 km ESE of TBS (Arguillin et al., 2010). That site was located in the ITT (Ishpingo, Tambococha, Tiputini) block, an area that has been the focus of questions regarding extraction of oil (Finer et al., 2008; Finer, Moncel & Jenkins, 2010). As with TBS, the area is dominated by terra firme forest and had not, at the time of the study, been subjected to substantial disturbance as a result of human activities, although roads used by hunters, tourists, and military personnel were present (Arguillin et al., 2010). Cameras (32 stations) were located 1–2.5 km apart and covered ∼58 km²; 603 independent events recorded 28 species in 2015 trap-days. Finally, to examine patterns at a larger scale, we used a similar approach to compare results from the two plots to those from a site in lowland forest in southeastern Peru, approximately 1480 km SSE of TBS, within and adjacent to Los Amigos Conservation Concession (Tobler et al., 2008); part of that study area lies within a logging concession. Cameras were operated in 2005 (24 locations, 2-km grid, ∼50 km²) and 2006 (39 locations, 1–2 km grid, ∼50 km²). Cameras accumulated 508 images of 21 mammals in 1,440 trap-days in 2005 and 814 images of 27 species in 2,340 trap days in 2006.

Prior to statistical tests, all variables were examined for assumptions of parametric tests and were transformed if necessary. Unless otherwise mentioned, analyses were run with STATISTIX Version 10 (Analytical Software, 2013).

General summary

We accumulated a total of 3,533 independent images (images separated by at least 30 min) of mammals, representing a total of 28 species, in 7,189 trap days (Table 1, Table S1). Numbers of images per 100 trap-days was considerably higher on Puma than on Harpia in 2014, 2015 and 2017 but approximately equal in 2016 (Table 1; paired t-test, t = 2.99, df = 3, P = 0.058). Despite the lower number of captures, more species were recorded in each year on Harpia than on Puma (paired t-test, t = 3.29, df = 3, P = 0.046) but the total across all years was similar on the two plots. Estimated species totals also were higher on Harpia except in 2017; estimated total numbers (all years combined) were similar on the two plots (Table 1).

Individual species

Capture rates across species were similar between plots within a year (2014: r_s = 0.75; 2015: r_s = 0.73; 2016: r_s = 0.85; 2017: r_s = 0.86; years combined: r_s = 0.84; P < 0.001 all cases). Capture rates also were similar within plots between years (Harpia: 2014–2015, r_s = 0.83; 2015–2016, r_s = 0.73; 2016–2017, r_s = 0.94; Puma: 2014–2015, r_s = 0.86; 2015–2016, r_s = 0.89; 2016–2017, r_s = 0.80; P < 0.001 all cases). Nine species on Harpia had capture rates of at least 1.0/100 trap-days (Table 2), with highest rates recorded for Pecari tajacu, Mazama americana, Dasyprocta fuliginosa, and Cuniculus paca. Ten species had capture rates ≥1.0/100 trap-days on Puma, with highest rates recorded for Mazama americana, Tayassu pecari, Dasyprocta fuliginosa, Myoprocta prattii, Pecari tajacu, and Tapirus terrestris (Table 2).

Some species exhibited large and consistent differences in capture rates between plots (Fig. 2). Tapirus terrestris, Tayassu pecari, and Myoprocta prattii were more common on Puma during each year whereas Pecari tajacu was the only common species with consistently higher capture rates on Harpia. Others, such as Leopardus pardalis, Mazama sp. and Dasyprocta fuliginosa, differed in capture rates between plots but direction and extent of the difference varied across years. Still other relatively common species, such as Cuniculus paca, Dasypus novemcinctus, Puma concolor and Panthera onca, showed little difference between plots in most years. Similarly, species differed in extent of variation in capture rates among cameras within a given year and plot. Capture rates of Myoprocta pratti, for example, were highly variable (large SEs) among cameras on Puma (Fig. 2).

Amount of variation in capture rates across years differed among species. Some, such as Myoprocta prattii, Tayassu pecari, Dasyprocta fuliginosa, Tapirus terrestris, and Cuniculus paca showed very little change from one year to the next, at least on one plot (e.g., Dasyprocta fuliginosa and Tapirus terrestris showed more variation across years on Puma than on Harpia). In contrast, other species, such as Puma concolor and Mazama americana showed substantial variation among years. In the case of P. concolor, the increase in capture rates from 2014–2015 to 2016 followed by a decrease to 2017 was especially notable, particularly as it occurred on both plots. In contrast, Leopardus pardalis showed a large decrease in capture rates on Puma from 2014 to 2015 with a large increase from 2016 to 2017.

Of 17 species represented by at least 10 images on at least one plot (all years combined; 16 species on Harpia, 13 on Puma), 11 demonstrated a clumped distribution pattern (Index of Dispersion, ID, significantly different from random) on each plot (Table 3). Panthera onca was the only species that did not show a clumped distribution pattern on either plot. Priodontes maximus and Leopardus pardalis did not show a clumped pattern on Harpia but did on Puma. Indices of dispersion were similar on both plots for some species (e.g., Puma concolor, Pecari tajacu, Tayassu pecari, Cuniculus paca) but differed strongly for others (e.g., Dasypus novemcinctus, Tapirus terrestris, Dasyprocta fuliginosa, Myoprocta prattii); all of the latter species had higher indices (stronger clumping pattern) on Puma than on Harpia (Table 3).

Despite the fact that a number of species gave evidence of a clumped distribution pattern, there was little evidence to suggest that numbers of images in particular cameras were correlated with numbers of images in other cameras—i.e., there was little to no spatial autocorrelation in numbers of images (based on Bonferroni corrected significance values; Table 3). A significant autocorrelation was noted for Mazama americana on Harpia but this reflected autocorrelation with distant camera locations (∼300–400 m away), suggesting that the autocorrelation was not ecologically meaningful (and not significant after correction for multiple comparisons). Thus, overall, there was little evidence to suggest that captures of a given species at one camera location were related to or influenced by captures at other cameras, even for species, such as most felids, that regularly travel along transects. Mantel correlograms showed no significant autocorrelations in species composition at any distance category. Similarly, results of Mantel tests indicated no significant correlation between composition of captures at camera locations and distance between cameras (Harpia—standardized Mantel statistic, r = 0.14, P = 0.22; Puma, r = 0.12, P = 0.32).

Community composition

A NMS ordination based on capture rates of 17 terrestrial species (species represented by at least 10 captures; cameras combined on each plot for each year) separated samples from the two plots (Fig. 3). From left to right, the first axis of the ordination reflected a gradient of decreasing importance of species such as Pecari tajacu, which was more important on Harpia, and increasing importance of others, such as Tayassu pecari, which was more important on Puma. This separation reflects the differences in capture rates between plots (Table 2). In contrast, the second axis primarily reflected differences in capture rates among years, rather than between plots. Puma concolor, for example, was more commonly encountered in 2016 on both plots than in other years. Results of an ANOSIM indicated that samples from each plot were, overall, more similar to each other than to samples from the other plot (global R = 0.9, P = 0.029).

Spatial variation among camera locations

A NMS based on capture rates (ln-transformed) from individual camera locations (all years combined) also indicated a separation between the two plots (Fig. 4). As with the comparison across years (Fig. 3; cameras combined within a year), the first axis of the ordination reflected differences in importance of peccaries, with Tayassu pecari more important on Puma and Pecari tajacu on Harpia. In addition, the gradient also reflected greater importance of Tapirus terrestris on Puma. The second axis largely reflected differences in importance of Dasyprocta fuliginosa and Myoprocta pratti and did not indicate a separation between plots, but rather differences among cameras irrespective of plot (Fig. 4); no species showed a strong negative correlation with the second axis.

Influence of environmental variation

Mean elevation did not differ among camera locations on the two plots at either the 50 or 100-m buffer scale (Table S2) but elevation was more variable on Harpia at both scales (i.e., significant differences in standard deviation values), reflecting the hillier terrain on Harpia. Similarly, slope and distance to streams were greater on Harpia, although significant only at the 100-m scale. These differences reflect the fact that the Puma plot tends to be flatter and with more permanent streams in comparison with Harpia.

Redundancy analysis based on environmental features within 50-m buffer around cameras explained a small but significant amount of variation in the distribution of species (Table 4). The first two axes accounted for almost 20% of variation in species distribution patterns, with slope, elevation, SD of slope, and SD of distance to stream all significant in a stepwise analysis. Thus, species such as Tayassu pecari and Tapirus terrestris were negatively associated with steeper slopes and greater variation in elevation (Fig. 5). Similarly, Leopardus pardalis was strongly negatively associated with slope and variation in slope and elevation. In contrast, Puma concolor and Atelocynus microtis were positively associated with elevation whereas Pecari tajacu tended to occur in areas farther from streams. In general, cameras on Puma separate from those on Harpia along the first axis; cameras on Puma tend to be in flatter areas and closer to streams, although this is not true for all camera locations on the plot (Fig. 6).

Environmental features explained slightly more variation in species distribution patterns when based on a 100-m buffer around camera locations (Table 4). Slope, SD of slope, and elevation all were significant in a stepwise analysis. Relationships between species and environmental features tended to be similar to those based on 50-m buffer (Fig. S1). Tayassu pecari and Tapirus terrestris were even more strongly negatively associated with variation in elevation, Pecari tajacu was more strongly associated with distances to streams, but Atelocynus microtis, for example, was not strongly associated with any variable. Cameras on Puma showed less variation in elevation than did cameras on Harpia and tended to be in areas with lower slopes (Fig. S1).

Comparison with other sites

Capture rates of species on the two plots combined (all years combined) were similar to capture rates from cameras located on trails within the boundaries of TBS (Fig. 1; r_s = 0.84; Fig. 7A). Capture rate of Tapirus terrestris was somewhat higher on the plots than on the trails whereas Sciurus igniventris, Leopardus pardalis, and Dasypus novemcinctus were somewhat more commonly encountered on trails. In contrast, capture rates were, overall, much higher on the two combined plots than at a different site within Yasuní National Park (ITT block, Varadero sector; Arguillin et al., 2010). Most species had higher capture rates on the two plots, as indicated by points lying below the line that indicates a 1:1 relationship (Fig. 7B). The lower slope was strongly influenced by species such as Puma concolor, Dasypus spp., Dasyprocta fuliginosa, and Mazama americana. Despite these differences, relative (rank) importances of species were similar in both sites (r_s = 0.81). Similarly, the slope of the relationship was much less than 1.0 when results from TBS were compared to a site in southeastern Peru (Tobler et al., 2008). For example, Pecari tajacu, Mazama americana, and Myoprocta pratti all were captured much more frequently on the TBS plots whereas Panthera onca and Didelphis marsupialis were more commonly captured at the Peru site (Fig. 7C). Capture rates were, nonetheless, highly correlated (r_s = 0.79) indicating that relative importance of species was similar in lowland forest of eastern Ecuador and in southeastern Peru.