Study site
The study was carried out at the Estación Biológica Quebrada Blanco (EBQB) between March 14 and June 27, 2011. The EBQB is located in north-eastern Peru (4°21′S, 73°09′W) on the right bank of Quebrada Blanco, a white-water affluent of the Rio Tahuayo, at an altitude of 110-140 m asl. The majority of the study area is primary terra firme forest with embedded swampy sections. The southern study area includes a secondary forest originating from former agricultural use (first for crop cultivation, later as buffalo pasture). Since 2001 the pasture has remained unused and regenerates, with typical pioneer plants like Cecropia sp. dominating the tree layer.
In June 2011 we surveyed the vegetation structure in nine randomly selected points in each of the two habitats. Vegetation cover — estimated visually within a radius of 10 m around the sampling point — was higher in primary than in secondary forest in all forest strata above 5 m height (Online Resource Fig. S1a). Estimated epiphyte coverage was 2.9 ± 2.9 % (mean ± SD) in primary and 0.9 ± 1.0 % in secondary forest, dominated by Araceae and Bromeliaceae (Wörner 2007). Other structure parameters were recorded with angle count-sampling (Bitterlich 1952, Kramer and Akça 2008) using Kramer’s dendrometer under a basal area factor of k = 2. We subtracted the influence of inclination and re-measured trunks at the threshold. Primary and secondary forest composition consisted of 86.8 and 92.1 % trees, 4.4 and 1.4 % palms, 0.7 and 0.0 % lianas, as well as 8.1 and 6.5 % deadwood, respectively (Online Resource Fig. S1b). While the secondary forest is dominated by trees with dbh < 15 cm (diameter at breast-height), tree size composition in primary forest showed a relatively even distribution (Online Resource Fig. S1c). In addition, we calculated mean heights of 24.5 ± 3.1 and 18.1 ± 2.7 m, total basal area of 29 m2/ha and 16.7 m2/ha, and stand density (for trees with dbh > 5 cm) of 2,495 trees/ha and 2,443 trees/ha for primary and secondary forest, respectively.
Mean annual rainfall in 2011 was about 2,300 mm (measured at Tamshiyacu, 40.4 km north of the EBQB, data provided by the Servicio Nacional de Meteorologia e Hidrologia del Peru). Our study period corresponded to the wet season, with maxima in April (269 mm) and May (368 mm), and the transition between late wet season and beginning of a drier season starting in June (181 mm).
Study groups
We observed two groups of S. nigrifrons in blocks of 6 subsequent days in a regular weekly change. Group 1 (3 adults, 1 subadult and 2 infants born in February) had access to secondary forest. Group 3 (2–3 adults, one of which died in April, 2 subadults and 2 infants born in May) lived directly north of group 1 solely in primary forest and served as a reference group to control for group-specific patterns when comparing foraging behaviour of group 1 in primary and secondary forest.
Observation usually started from 0600 hours, when the tamarins left a sleeping site, and continued until the afternoon at about 1600 hours, when they entered another sleeping site. In total, group 1 was observed for 202 h over 28 days and group 3 for 191 h over 29 days. Both groups formed stable mixed-species troops with moustached tamarins Saguinus mystax and were well habituated to the presence of human observers.
Observational methods
We used three different methods to collect behavioural data from the study groups, excluding infants. Instantaneous scan-sampling (Martin and Bateson 2007) at 15-min intervals focused on activity budget and diet composition. We recorded the type of activity (locomotion, resting, social interaction, prey searching, fruit feeding, exudate feeding, prey feeding, other) of each tamarin that became visible within 30 s after the scan sampling-point indicated by the beep of a timer. We obtained 745 activity records for group 1 (681 in primary, 64 in secondary forest), and 655 for group 3. At each scan-sample point, we recorded the group position using a Garmin GPSMap 76CSx. Between scan-sampling points, focal sampling with continuous recording (Martin and Bateson 2007) was employed to record prey searching strategies, capture techniques, prey characteristics and capture rates. The length of focal samples was 10 min, during which the focal individual had to be visible for at least 5 min. The selection of focal individuals followed a previously set rotational scheme. If we could not find the individual at the top of the scheme within 2 min, we selected the next individual. If this could not be found either, we started a new search again with the previous individual. By using this procedure we ensured an even distribution of the number of focal protocols over all studied individuals. We recorded 66.7 h of focal sampling for group 1 (61.0 h in primary and 5.7 h in secondary forest) and 55.7 h for group 3. During focal sampling the same activities as in scan sampling were registered. Additionally, we recorded support type and orientation as well as substrate type and height during prey search. We also categorized the technique of each successful prey capture: direct capture from open microhabitats, low intensity manipulation, e.g. opening epiphytes or unrolling dry leaves, and high intensity manipulation including breaking or biting open substrates (see also Nadjafzadeh and Heymann 2008). If captured directly, we registered whether this arose from prey flushing, meaning events where prey items fled from other tamarins. Also, we recorded colour and size of prey items. We defined prey capture success (S
i
) as the rate of captures per prey search time in focal samples. Outside of focal and scan sampling, other prey feeding events were recorded noting group, forest type, date, time and prey type, with as much detail as possible. We collected prey items discarded by the tamarins, e.g. orthopteran tegmina and hind wings, for later identification. To increase interobserver reliability, especially on height and in situ prey size estimation, we carried out a multi-week tutorial with the field assistants prior to data collection.
Prey abundance
We concentrated our survey of potential prey abundances in primary and secondary forest on nocturnal katydids since this is the dominant part of the tamarins’ prey (Nickle and Heymann 1996; Smith 2000; Nadjafzadeh and Heymann 2008). Katydids belong to the family Tettigoniidae, order Orthoptera, and produce distinct species-specific stridulation sounds. Therefore, we recorded prey abundance as the number of singing orthopteran individuals on three randomly placed 50-m transects in primary and secondary forest, respectively. All transects were walked with the same velocity within 10 min using a Petersson D200 heterodyne ultrasonic detector during rainless nights. To account for species-specific activity patterns (Belwood 1990; Nickle and Heymann 1996) we repeated the survey at three different times (1900, 2300 and 0300 hours) once a month from March to June.
Data analyses
We carried out statistical analyses with R 2.12.1 (R Development Core Team 2010) using the packages stats, vegan (Oksanen et al. 2011), and lme4 (Bates et al. 2011). For the analyses of activity budgets and diet composition, only complete observation days were used. We used Fisher’s exact test to compare activity budgets, diet composition, prey searching strategies and capture techniques between groups and habitat types. The function fisher.test in R uses a subroutine (FEXACT) to execute Fisher’s exact tests on contingency tables larger than 2 × 2 (Mehta and Patel 1986; Clarkson et al. 1993). All tests were two-tailed at a significance level of α < 0.05. For each test where the null hypothesis had to be rejected, we performed additional multiple testing to obtain specific information about diverging categories. We used the Bonferroni correction to correct the significance level in multiple tests.
Due to unequal numbers and variances in the comparative analysis of prey size, we employed Welch’s unequal variance t-test with previously ranked values instead of Mann–Whitney U-test (Ruxton 2006).
We analysed the diurnal distribution of secondary forest utilization of group 1 and compared it with diurnal patterns in fruit feeding and prey search.
We calculated the overlap of captured prey items between groups based on Morisita’s unmodified index of similarity without log-transforming data (Krebs 1999):
$$C = \frac{{2\sum {p_{ij} } p_{ik} }}{{\sum\nolimits_{{}}^{n} {p_{ij} \left[ {\left( {n_{ij} - 1} \right)/\left( {N_{j} - 1} \right)} \right] + \sum\nolimits_{{}}^{n} {p_{ik} \left[ {\left( {n_{ik} - 1} \right)/\left( {N_{k} - 1} \right)} \right]} } }},$$
where C is Morisita’s index, p
ij
and p
ik
are the proportions of prey item i in the total prey used by groups 1 and 3, respectively, n
ij
and n
ik
are the numbers of individuals that use prey item i in groups 1 and 3, respectively, and N
i
and N
k
are the total numbers of individuals in each group. This index produces a minimum bias of abundance and diversity in different data sets (Wolda 1981; Smith and Zaret 1982).
For taxonomic identification of collected prey items, we used literature (Beier 1962; Belwood 1990; Nickle and Castner 1995; Nickle and Heymann 1996; Bartlett and Bartlett 2003), the species online files for Orthoptera (http://orthoptera.speciesfile.org) and Phasmida (http://phasmida.speciesfile.org), a reference collection of tamarin prey provided by Andrew C. Smith (see also Smith 2000) as well as assistance by experts for Orthoptera, Holger Braun (División Entomología, Museo de La Plata, Argentina), and Phasmida, Sven Bradler (Johann-Friedrich-Blumenbach Institute of Zoology, Göttingen, Germany). Where exact taxonomic identification was not viable due to insufficient prey remains, the items were classified into morphotypes using colour and size as categories.
We assessed prey abundance using the generalized linear mixed model function lmer (Bates et al. 2011) with a significance level at p < 0.05. We set night-time as random, forest type and month as fixed effects.
GPS positions were processed in ESRI ArcGIS 9.3. We performed fixed kernel home-range estimation following Worton (1989) with the software extension ‘Home-Range-Tools’ (Rodgers et al. 2007). To analyse habitat utilization, we calculated intensities of primary and secondary forest habitat use indices (H
i
) for group 1, basically following Neu et al. (1974):
$$H_{i} = \log \frac{{\text{freq}_{\text{obs}} }}{{\text{freq}_{\exp } }}$$
This method compares the frequencies between observed and expected values for each habitat type. Observed frequencies were obtained from recorded GPS positions. A buffering of these points by 5 m represents the distribution of the tamarin group in the field and converts the point data to polygon. Krebs (1999) suggested using habitat type availability as expected frequencies. Thus, we calculated the proportions of forest types within the home range of group 1 based on the 100 % minimum convex polygon (MCP) (Mohr 1947), which we also buffered by 5 m. We log-transformed the term to get an index value between −1 (avoidance) and +1 (preference). Differences between expected and observed frequencies were tested using Fisher’s exact test.
All presented statistical information were referred to Fisher’s exact test, except where otherwise stated.