Study Site

Fieldwork was conducted within the territory of the indigenous Kichwa community of San José de Payamino (Payamino hereafter) in the Amazonian province of Orellana, Ecuador, 300 m elevation (Fig. 1). Their extensive territory consists of ca. 17,000 hectares of lowland Amazonian rainforest and is bordered to the north by, and partially ranging within, the Gran Sumaco Biosphere Reserve. Much of their area is only used for hunting grounds, as most community members live near the forest bordering the Payamino River. The habitat is characterized by large tracts of primary terra firme forest, disturbed secondary forest primarily around the community square and near the Payamino River, and riparian forest and oxbow lakes along the Payamino River. Rainfall in the region is higher than that of the rest of the Amazon—averaging 3,711 mm over a 10-year period at a nearby site—and is generally considered aseasonal with irregular annual patterns [22].

Fig. 1.

Location of the study site, San José de Payamino (indicated by star), in Orellana, Ecuador, 300 m elevation.

Less than 10 years ago, Payamino was relatively isolated, and transportation to the nearest towns of Loreto and Coca was accomplished via river. This has now changed, as there has been a substantial amount of primitive road creation in the area within the past decade. One such road established in 2010 was poorly executed, as a river crossing with no bridge makes it inaccessible to most vehicles, and in multiple areas erosion has made the road impassable for vehicles since its construction (Paul Bamford, pers. comm.). Oriented north-south, this narrow stretch of dirt road (5-10 m wide) is about 4 km long and runs 1–2.5 km west of, and roughly parallel to, the Payamino River (Fig. 2; Appendix 1). Despite its size, the road leaves a noticeable scar in the landscape, creating a gap in the canopy (Appendix 1). Daily low-volume foot traffic occurs along the clearing, which, along with machete use, continues to maintain it as a road-sized disturbance.

Fig. 2.

Study design depicting sites of 20 m diameter plots and 10 m long drift fences. Plot center-points were situated at three perpendicular distances from the dirt road: 10 m (forest edge), 50 m (intermediate), and 100 m (interior forest). All sampling sites were restricted to secondary forest. Plot and drift fence locations are not to scale.

Study Design

The project study area was restricted to secondary forest 110 m from each side of the dirt road (0°28'58“S, 77°17'5”W). Sampling greater distances from the road was not possible, as secondary forest occurred only as far as ∼125 m from both sides of the road (i.e., primary forest began roughly 125 m interior from the forest edge in areas west of the road, but no primary forest was to the east side of the road). Therefore, sampling different habitat types from opposite sides of the road was avoided. We established eighteen alternating 100 m transects running perpendicular to the road, with the first running west of the road, the second running east, and so on (i.e., nine transects per side; Fig. 2). A Garmin 62S model GPS unit was used to navigate the forest perpendicular to the road. Each transect was separated by 30–50 m from the nearest opposing transect. On each transect, three 20-m diameter plots were established at different distances from the road edge, with plot center-points marked at: 10 m (edge plot), 50 m (intermediate plot), and 100 m (interior plot). Plots did not extend onto roadside ditches or narrow grassy verges (i.e., edge plots reached only to the forest edge) to avoid sampling individuals in standing water along the road periphery. Also, no plots overlapped small streams within the forest, although having some plots closer to streams than others was unavoidable.

Sampling Methods

All data were collected from June – August 2013. A nocturnal visual encounter survey (VES) was chosen as the primary sampling technique [30–31]. We performed two 30-minute VES sample efforts at least one week apart for each plot. All VES samples were conducted by the same two biologists (RJM and NCA). Each VES was performed between 1900 - 0030 h with the use of high-powered LED headlamps and a stopwatch to keep time. Prior to each night of surveys, the order at which plots were to be sampled along transects was selected at random to avoid bias in the timeframe sampling distances were surveyed.

Plots were 20 m in diameter with plot center points marked with flagging tape [32]. During a VES sample, a plot was surveyed as thoroughly as possible by both observers for 30 min. When an individual amphibian or reptile was observed, the stopwatch was paused to allow data collection for the individual. Data collected on all animals were: in situ photographs, species identification, substrate or perch type, perch height (cm), perch diameter (cm), and a visual estimate of snout-to-vent length (mm). Although most individuals were not captured, all records from a plot were crosschecked via photographs to assure no individuals were counted twice during a survey.

As a secondary sampling method, we constructed drift fences with both funnel and pitfall traps [33–34]. Combining this method with VES enabled a broader description of the herpetofauna assemblage along the sampled edge-interior forest gradient, since drift fences with traps are effective at capturing species easily overlooked by VES, such as those which are primarily fossorial or leaf litter inhabitants. Drift fence sites were chosen in a similar fashion to that of VES plots, however drift fences were constructed only at the forest edge (just inside the tree line) and interior forest (100 m), with 10 fences at each distance (Fig. 2). Drift fence sites did not overlap with plot sites and were distributed as evenly as possible from one another within the study area. Drift fences were 10 m in length, unidirectional, and oriented parallel to the road. Each fence had two mesh nylon funnel traps (25 × 46 × 25 cm; dual 6 cm entrances) placed at both ends of a fence on the side facing away from the road. Additionally, a large plastic bucket (40.5 I) was sunk flush to the ground at the center of each drift fence. All drift fences were open for a total of ten days and were checked at least once every 24 hours when open. Captured individuals were released after identification > 15 m away from the drift fence site to help avoid recaptures, corroborated by photos of captured individuals.

Six habitat variables were measured to determine structural differences along the forest edge-interior habitat gradient. At each plot site, we recorded leaf litter depth, understory density, percent canopy cover, vine abundance, abundance of mature trees (defined as number with DBH >12 cm), and number of downed trees (DBH ≥20 cm). Leaf litter depth was measured by placing a graduated ruler into the leaf litter until it reached the soil. Understory density was calculated by vertically placing a 3.5-cm diameter and 2-m tall PVC pole on the ground and counting the number of contact points made on the pole by the vegetation (branches, stumps, leaves; [28]). Leaf litter and understory density were measured five meters from the center point of a plot in each of the four cardinal directions, and then averaged. Percent canopy cover was measured using a spherical densiometer held 0.5 m above the ground at the center of a plot. Vine abundance (woody and herbaceous) of plots was independently scored on a scale from 0 to 4 by both observers and then averaged, as outlined by Schlaepfer and Gavin [35]. Additionally, relative humidity and ambient temperature (°C) were recorded prior to each VES sample using a digital thermo-hygrometer.

Data Analysis

Analyses pertain to data from VES samples and are exclusive of drift fence data unless specified. Drift fence data are primarily used to compare sampling methods as well as to detect species that were unique to this method. A complete list of all individuals recorded by either method is reported (Appendix 2).

Abundance of amphibians and reptiles for each plot was calculated as the average of the total abundance values from each of the two VES samples of a plot. We calculated amphibian and reptile species richness in the same fashion. Diversity was also calculated for each plot using the Shannon-Weaver index, but only for amphibians, as sample sizes were generally too low for reptiles. We used Pearson correlation coefficients to identify correlated measures of the habitat. Pairwise permutation tests [36] were used to examine differences in habitat variables and herpetofauna abundance and diversity among habitat types. Each permutation test was performed for 9,999 iterations, using the PopTools 3.2 plugin for Microsoft Excel. Pairwise differences among sampling distances were calculated by taking the absolute difference between the mean values for the variable being tested. The original plot values were then randomly re-assigned (without replacement) to treatments, allowing an additional absolute difference to be calculated using the resampled data. Next, the distribution of absolute differences from resampled data was compared to the absolute difference from the actual observed values to determine the likelihood that the pattern found in the original data could be found by chance [36]. The P-value for the test represents the proportion of the 9,999 iterations that were greater than, or equal to, the observed difference in treatment means. Observed values were considered significant if the P-value was < 0.05.

We compared abundance patterns and species evenness between sampling distances using rank-abundance curves [37]. For each sampling distance, we plotted relative abundance of each species on a logarithmic scale against the rank order for the species from most to least abundant [28]. Relative abundance of a species was calculated as the total number of individuals of i species divided by the total number of individuals recorded at that sampling distance (n_i/N). Rank-abundance curves were calculated for amphibians and reptiles separately. All records were included for analysis from both surveys of each plot.

To better understand which measured habitat variables best predict animal abundance and diversity, we created models using stepwise multiple regressions in program JMP (v.10; SAS Institute). After consideration of co-linearity, a standard least squares regression was used to obtain the parameter estimates, followed by forward stepwise multiple regressions. Habitat variables were included in models using the selection threshold of P ≤ 0.10. We used the lowest Akaike's Information Criterion (AIC_c) score to select the model that best explains the dependent variable using the fewest habitat variables. Additional models that had a difference in AIC_c score (ΔAIC) within 2.0 of the best model (i.e., that with the minimum AIC_c score) were also noted for consideration. Models were only created for amphibians due to the relatively small sample size for reptiles.

Ethics Statement

Field work was carried out in strict accordance with the recommendations in the guidelines for use of live amphibians and reptiles in field research compiled by the American Society of Ichthyologists and Herpetologists (ASIH). Research was conducted with permission of the Ministerio del Ambiente, Ecuador, (permit #: 001–11 IC-FAU-DNB/MA).

Drift Fences

Drift fence captures accounted for 84 records (13% of total records) consisting of 14 amphibian species from four families and 10 reptile species from six families. More amphibians were captured (n = 54) than reptiles (n = 30). We experienced an overall capture rate of 0.42 animals per drift fence/day. In contrast to VES plot samples, there were more individuals and species of herpetofauna captured at drift fences within the forest edge (abundance = 58; richness = 18) than in interior forest habitat (abundance = 26; richness = 12); these differences were not significant (P = 0.0880; P = 0.3186; respectively).

Habitat Structure and Climate

There were various differences found in habitat structure across the forest edge-interior gradient (Table 1). Understory density and vine abundance decreased with increasing distance from the road; both variables were significantly greater at the forest edge habitat than in intermediate and interior forest habitats (all comparisons: P 0.001; Table 1). Downed tree abundance and average temperature demonstrated a general decline with distance from the road, although differences between habitats were not significant. Mature tree abundance and relative humidity increased with distance from the road; mature tree abundance was significantly different between habitats (edge vs. intermediate: P 0.001; edge vs. interior: P 0.001). Canopy cover was greater at the intermediate habitat than at forest edge (P = 0.004). No trend in leaf litter depth was found relative to distance from the road (edge vs. intermediate: P = 0.6639; edge vs. interior: P = 0.3570; intermediate vs. interior: P = 0.1719).

Table 1.

Habitat and environmental variables measured at each plot along the forest edge-interior gradient adjacent to a dirt road at San Jose de Payamino, Orellana, Ecuador. Values in a row with same letters did not differ (P ≥ 0.05) after pairwise tests comparing the absolute difference between observed values and permuted values. LLD = leaf litter depth; USD = understory density; CC = canopy cover; VA = vine abundance; MT = mature tree abundance; DT = downed tree abundance; RH = relative humidity; AT = ambient temperature.

Various habitat and climatic features were correlated (Table 2). Vine abundance was most strongly correlated with understory density, mature tree abundance, and distance from the road (P < 0.001). Mature tree abundance also had notable correlations with understory density and canopy cover. Correlations between distance from the road and both understory density and mature tree abundance were also highly correlated.

Table 2.

Pearson correlation coefficients among nine environmental variables measured within the secondary rainforest bordering the primitive dirt road.

Abundance and Diversity Models

Stepwise regressions suggest that vine abundance is the single most important predictor variable for amphibian abundance, richness, and species diversity within the study area (Table 3). Models with the lowest AIC_c score for each of the dependent variables indicate a negative relationship with vine abundance. Models within two ΔAIC of the best model for amphibian abundance also include downed tree abundance and relative humidity. For amphibian richness and diversity, the model with the lowest AIC_c score includes vine abundance and mature tree abundance (Table 3). As vine abundance decreases and mature tree abundance increases, richness and diversity are predicted to increase.

Table 3.

Best-supported habitat predictor model (using lowest AIC) for amphibian abundance, richness, and Shannon-Weaver (SW) diversity.

Amphibian Distribution Patterns

Corroborating similar studies, species-specific responses to differences in habitat strongly influenced amphibian abundance and diversity trends [28, 35, 39]. For example, both the highly disturbed forest-edge habitat and moderately disturbed intermediate habitat were dominated in relative abundance by a disturbance specialist (Pristimantis variabilis), particularly at the forest edge (Appendix 2). This same species was uncommon in the interior forest habitat and was observed only in the few plots that showed clear signs of disturbance. Such species are often indicative of forest disturbance levels, as their abundance along a forest gradient generally reflects the degree of disturbance at any given point. A similar trend has been found with the numerically dominant Cachabi Robber Frog (P. achatinus) at various sites in western Ecuador, where its abundance also declined in habitats with decreasing levels of disturbance [27; Paul Hamilton, unpubl. data].

Contrarily, many rainforest amphibians have likely evolved life history traits that are specific to interior forest conditions. Lower amphibian abundance, richness, and diversity at the forest edge is perhaps due to an inability of many interior forest specialists to persist there, due to lack of necessary resources or proper environmental conditions. In this study, 16 amphibian species were absent from the forest edge, seven of which were restricted to the interior forest habitat. Species that were found at the forest edge that are otherwise generally associated with interior forest habitat were often found in close proximity to, or directly on, old growth trees still standing adjacent to the road (e.g., Brown-eyed Treefrog [Nyctimantis rugiceps]; Osteocephalus deridens; Pristimantis paululus). Whether or not mature trees at the forest edge remain healthy and persist over time could strongly influence how permeable the forest nearest the road is to such species.

In some cases, greater or equal animal abundance and diversity at the forest edge compared to interior forest are attributed in part to a source effect from the matrix in fragmented landscapes [404142–43]. However, the little-disturbed interior forest at Payamino still harbored more individuals and species of amphibians than the highly-disturbed forest edge habitat, adding to the relatively few studies reporting negative effects on amphibian abundance and diversity from forest clearing or edge effects in Neotropical forests [29, 35, 44–45]. The source effect of matrix habitat is potentially reduced in situations where only a narrow linear disturbance bisects otherwise mostly continuous tropical rainforest, such as that at Payamino. The road in this study is comprised of compact soil and lacks the vegetative substrate characteristic of open matrix conditions such as pastures, which might reduce or preclude certain matrix specialists. Interestingly, species known to use the open dirt road at Payamino, such as numerous hylid frogs that exploit water-filled roadside depressions as breeding sites, mostly avoided overlap within the forest edge. Of at least eight hylid species known to make use of such habitat in relatively high numbers at Payamino, only the Upper Amazon Treefrog (Dendropsophus bifurcus; n = 1), Basin Treefrog (Hypsiboas lanciformis; n = 9), and Red-snouted Treefrog (Scinax ruber; n = 4) were recorded in edge plots [Appendix 2; Ross Maynard, unpubl. data]. None of these species were recorded beyond the forest edge habitat. A similar pattern has been reported from forest fragments in central Amazonia, where nearly 20% of anurans were recorded exclusively in matrix habitat [46].

The increase in amphibian abundance along the forest edge-interior gradient was driven not only by greater assemblage diversity at the interior forest habitat, but this is also where five of the overall six most abundant species were most numerous. Although some studies have found either positive or neutral effects of edge on amphibians, the results generally reflect a particular species or guild [28, 45]. Nevertheless, terrestrial-breeding frogs in the genus Pristimantis played a large role in the patterns of distribution of amphibians reported herein due to their prominence in the dataset. Pristimantis was by far the most abundant and species rich of all the genera observed, with a majority of Pristimantis spp. demonstrating greater abundance with increasing distance from the forest edge. Three Pristimantis spp. were completely absent from the forest edge habitat.

Other studies have also found a negative effect of forest clearing or edge effects on abundance and diversity of Neotropical terrestrial-breeding frogs [29, 38, 45], perhaps because many species are highly sensitive to edge changes in moisture (e.g., humidity in important microhabitats), microhabitat availability, and food availability. Such associations are presumably specific to interior forest conditions for many species. However, teasing apart which habitat components are most closely tied to patterns of abundance and diversity is difficult. McCracken and Forstner [29] found no relationships between several habitat variables and high-canopy anuran abundance, and Ernst and Rödel [47–48] found that habitat factors were not significant predictors of species incidence in tropical frog assemblages in Africa and South America. When models suggest significant habitat associations with patterns of animal distribution, they are generally drawn from broad-scale environmental measures rather than specific measures of quality and quantity of important microhabitats [e.g., 28], which are unknown for many tropical taxa. In our models, the suggested importance of vine abundance is noteworthy, as it is a conspicuous element of disturbed secondary forest structure and likely serves as a proxy variable for other habitat factors that are more difficult to measure. Both vine abundance and understory density, which we found to be correlated, have had a negative impact on amphibians elsewhere in Neotropical forests [35, 38]. Rising vine abundance in the Neotropics has known negative effects on tree diversity, recruitment, growth, and survival [49], and its conspicuous structural influence at the forest edge presumably translates to differences in resource availability from that of interior forest conditions. Such differences likely have a profound impact on amphibian distribution patterns. For example, increased vine abundance at the forest edge could alter the quality and quantity of perch and oviposition sites, change the acoustics of the habitat and thereby influence the ability of frogs to effectively transmit or receive auditory signals, or alter inter-specific dynamics (i.e., predator-prey relationships).

Regarding other habitat variables, the lack of difference in leaf litter depth among sampling distances was likely due to the high abundance of large, multi-lobed leaves shed from the many Cecropia at the forest edge, which formed a leaf litter layer comparable in depth to that of interior forest. Instead of depth, perhaps a difference is more likely to be found in leaf litter quality and composition. Since the large Cecropia leaves become gnarled and concave in shape after abscission, the leaf litter within the forest edge habitat is less compact and likely retains less moisture than leaf litter at the forest interior. Additionally, leaf litter at the forest edge is presumably lower in compositional diversity, which could alter micro-fauna and flora diversity. Such differences, should they exist, could influence clutch success for terrestrial-breeding herpetofauna as well as the availability of essential microhabitats such as refuges during particularly hot temperatures or dry spells. Future studies could address these issues by measuring dried leaf litter mass in addition to leaf litter depth.

Reptile Distribution Patterns

The greater reptile abundance at the interior forest than at the forest edge habitat was somewhat unexpected, as tropical reptile species that thrive in open or disturbed habitat generally persist in greater abundances than forest interior species [50]. Also, light gaps are particularly important for heliothermic lizards [51], which are usually characteristic of disturbed areas of forest, such as that at the edge. Interestingly, reptile trapping with drift fences was more successful at the forest edge than at the interior forest and recorded species otherwise not detected by VES samples (Appendix 2). The conflicting results from the two survey methods underscores the importance of using multiple survey methods for reptiles, of surveying by both day and night, and of the difficulty of acquiring adequate samples of Neotropical reptile assemblages to properly assess their distribution patterns.

Lastly, even though the overall species accumulation curve for the herpetofauna suggests a nearly complete survey of the species in the study area, this is certainly not the case, as numerous amphibian and reptile species either known from outside the immediate study area or not yet recorded at Payamino but known from nearby sites, are still likely to be present [22, 39; Ross Maynard & Paul Hamilton, unpubl. data]. Even survey efforts spanning nearly 30 years have yet to completely sample the entire amphibian and reptile assemblage at a site close to Payamino, as additional species continue to be recorded there [22; Ross Maynard, Paul Hamilton, and Greg Vigle, unpubl. data].

Future Research

We suggest several avenues for future research: 1) with road creation increasing in the Amazon Basin [12] coupled with the paucity of information on road impacts on biota in the region, projects examining roads differing in width, position (e.g., roads on gradients vs. upland vs. lowland), traffic intensity, and road type (dirt vs. gravel vs. pavement) are all necessary; 2) research needs to encompass extended timeframes to better survey the diverse biotic assemblages that characterize the region; 3) careful observations are needed for many taxa to determine their natural history. Such information is necessary to quantify vital habitat features for particular species or animal groups that are potentially altered by anthropogenic disturbances; and 4) additional distances from the edge as well as a road plot should be sampled to gain a clearer perspective of changes occurring throughout the ecotone.