Introduction

Compared to many other mammals, primates are characterized by increased brain size and complex cognitive abilities that allow them to have highly flexible problem-solving skills (Dunbar 2000; Shettleworth 1998; Tomasello 2000; Tomasello and Call 1997). One critical question concerning the evolution of primate cognitive abilities is how individuals internally represent and integrate spatial and ecological information to develop a set of decision rules to locate and relocate productive feeding sites efficiently (Garber et al. 2009). Resources in a tropical rain forest vary in their spatial and temporal distribution, nutritional rewards, and ease of procurement. Thus, foragers face significant challenges in navigating between distant feeding sites and in deciding when to leave a food patch and which patch to visit next (Raubenheimer et al. 2009; Simpson et al. 2004; Ydenberg et al. 2007). Given these foraging challenges, adaptations that allow animals to retain, integrate, and update information on the location and phenological patterns of potential food sources should greatly increase their foraging success (Cheng 2010; DeLoache and Pickard 2010; Garber et al. 2009).

Two general models have been proposed to explain nonhuman primate spatial strategies in large-scale space, e.g., when moving between distant feeding and resting sites that are outside of their visual field: a route-based mental map (Di Fiore and Suarez 2007: Garber 2000) and a coordinate-based mental map (Normand and Boesch 2009; Poucet 1993). A topological or route-based mental map is associated with the ability of a forager to use and reuse a series of familiar travel routes (networks) and landmarks as orientation or switch points to navigate to and relocate feeding and resting sites. Using a route-based “map,” foragers are expected to maintain an overall understanding of the relative spatial relationships among several key locations or topological features in their home range, but using this representation they are unable take the most direct route to reach their target or goal. As suggested by Dolins and Mitchell (2010), within a route-based spatial framework foragers can generate novel routes of travel; however, these routes are constrained by their proximity to familiar landmarks (switch points) and pre-established intersecting travel routes.

Alternatively, animals may represent spatial information in the form of a coordinate-based mental map (a Euclidian map or a “view from above”) in which the precise locations of salient features of the environment are encoded and recalled as x and y coordinates. Using such a system, animals are expected to compute relatively accurate distances and directions from their current location to their goal and travel using direct routes and novel shortcuts. For example, Normand and Boesch (2009) argued that wild West African chimpanzees (Pan troglodytes verus) encode spatial information in a coordinate based representation, travel directly to feeding sites that lie far outside their field of view, rarely reuse the same travel paths, and do not redirect travel by orienting to a particular set of landmarks.

Primates also may use a combination of cognitive strategies to navigate in different spatial scales. In small-scale space, e.g., in a core area in which individuals can obtain views of the same targets from different directions and perspectives, animals may be able to form a coordinate-based spatial representation, whereas in large-scale space animals may use a route-based framework (Garber and Brown 2006; Garber and Dolins 2010; Poucet 1993; Urbani 2009). In the case of black-horned capuchins (Cebus nigritus: Presotto and Izar 2010), white-faced capuchins (Cebus capucinus: Urbani 2009), and Geoffroy’s spider monkeys (Ateles geoffroyi: Valero and Byrne 2007), authors have argued that individuals navigate using a combination of coordinate-based and route-based maps. An understanding of how different species represent spatial information and how this affects patterns of ranging, foraging strategies, diet, and patch choice is critical for developing models of cognitive ecology and decision making.

In this study, we examined foraging decisions and spatial memory in large-scale space in a group of wild Weddell’s saddleback tamarins (Saguinus fuscicollis weddelli). Saddleback tamarins are small-bodied monkeys (adult body mass ca. 350 g; Smith and Jungers 1997) found in forests throughout the Amazon basin (Rylands et al. 2008). Saddleback tamarins exploit a diet composed principally of insects, fruits, nectar, and exudates (Garber 1993; Porter 2001). They live in small cohesive social groups of 4–10 individuals, exploit and defend home ranges that vary in size from 20 to 150 ha, and have a daily path length of between 1400 and 2000 m (Digby et al. 2007; Garber 1988; Peres 1992; Porter 2004; Soini 1987).

Previous studies of tamarin foraging behavior indicate that individual feeding bouts are relatively short (5–8 min) and members of a group visit between 10 and 20 fruit, nectar, and exudate feeding trees in a single day (Garber 1993). Moreover, data indicate that the amount of food at a feeding site, as well as the spatial and temporal predictability of the food reward, play an important role in tamarin foraging decisions (Garber 1988, 2000; Garber et al. 2009).

To better understand foraging strategies and spatial memory in wild saddleback tamarins in large-scale space, we recorded the diet of the tamarins and located the positions of their feeding trees. We also assessed the phenology of these feeding trees to determine how productive the sites were over time. In this way, we were able to determine the distribution pattern of feeding trees of varying degrees of productivity in the monkeys’ home range. In addition, we tracked the tamarins’ movement patterns when moving between feeding sites. Finally, we analyzed these feeding paths to test several predictions based on three hypotheses.

Hypothesis 1

The monkeys have no cognitive maps; they move randomly through their home range. Prediction 1: We predict that the monkeys’ circuity indices (distance actually moved ÷ direct distance) when moving between feeding and resting sites are not significantly different than the circuity indices of randomly generated routes between these sites. Prediction 2: We predict that the monkeys move between feeding sites in a sequence that is no more efficient than if the sequence was chosen randomly.

Hypothesis 2

The monkeys have a coordinate-based spatial representation of the location of feeding and resting sites in their home range. Prediction 3: We expect travel to be characterized by a significant decrease in the CI index values as the monkeys revisit feeding and resting sites. Over time the monkeys should learn the precise location of those sites and take relatively straight-line paths and novel shortcuts to reach them.

Hypothesis 3

The monkeys have a route-based spatial representation of their home range. Prediction 4: We expect that in analyzing the monkey’s movement patterns we will be able to identify “paths” that are used repeatedly for travel. Prediction 5: In addition, we predict that the monkeys have a set of switch points in their home range that they repeatedly use to change direction during the course of travel.

Methods

We observed one group of Weddell’s saddleback tamarins (Saguinus fuscicollis weddelli) in northern Bolivia in the Department of the Pando, at a field camp Callimico (11°23′S, 69°06′ W, ca. 280 m a.s.l.). The forest is representative of sandy, clay forests of the south and southwestern Amazon Basin (Alverson et al. 2000) and experiences pronounced dry and rainy seasons.

Our focal group had been followed for 3 mo in 2008; thus individuals were fully habituated to the presence of observers at the start of this project in 2009. In 2008, we also had trapped and fitted beaded identification collars on all adult group members. In 2009, the study group was composed of five individuals: one adult female, three adult males, and one juvenile. The adult female and two adult males retained their collars from the previous year, and the juvenile was identifiable by its size; therefore we were able to distinguish all group members from each other.

We collected data on the study group during 32 d in June and July 2009. On 15 of these days we have data on the subjects’ ranging patterns from morning sleep site to evening sleep site, on 10 d we arrived soon after they had emerged from their sleep sites, on 2 d we ended observations before the tamarins entered their nighttime sleeping trees, on 2 days we had GPS data for only part of the day, and on 3 days the GPS unit was not used owing to technical problems. We collected data on diet, activity budget, and foraging behavior on a focal group member at 2-min intervals throughout the day. In total we obtained 7098 focal animal activity records accounting for 236 observation hours. We collected data on one individual for an entire day and then observed a different individual on each subsequent day of observation until all group members had been observed. We chose the order in which group members were observed randomly. In total we observed each group member on 7 individual days, except for one male, which we observed for only 6 d. At 10-min intervals throughout each observation day we recorded the spatial location of the focal animal using a Garmin e-Trex HCx GPS unit with a memory storage chip. Based on an analysis of 14 test points taken at the same location on different days we found the GPS unit to be accurate in our forest to 10.8 m.

We used the following definitions to record the behavior of the focal individual: feeding = the monkey was handling or ingesting food; foraging = the monkey was moving in the crown of a tree, or up and down a substrate in search of food; traveling = the monkey was moving but the movement was not directly food related; resting = the monkey was inactive; social = the monkey was grooming or engaging in other forms of affiliative or agonistic behavior; intergroup interaction = the monkey was approaching, vocalizing at, or engaging in threatening or aggressive behavior with conspecifics from a neighboring social group. We defined a feeding bout as the total number of 2-min intervals in which the focal individual fed or foraged on a particular food type or in a particular food patch, e.g., an individual tree bearing fruits, exudates, or floral nectar, or searching in a tree hole for insects. We assumed that the focal individual devoted the entire 2-min interval to exploiting that resource. A new feeding bout was recorded when the focal individual was observed to feed or forage at a new feeding site, e.g., different tree or microhabitat, regardless of whether or not it was the same plant species as the previous feeding site. We defined feeding sites by two different criteria. We identified trees of plant species that accounted for >1 % of total feeding time as major feeding tree species (MFTS). We also identified individual trees in which the monkeys fed on two or more days; we call these revisited feeding (RF) trees/sites.

Given that the amount of food present at a feeding site has been demonstrated to influence tamarin foraging decisions (Garber et al. 2009), we assessed the phenology of MFTS and RF trees each week. We estimated a tree’s productivity of fruits and flowers as follows: 1) we counted the number of food objects observed in three binocular fields while viewing the tree crown; 2) we counted the number of fields needed to view the entire canopy; and 3) we estimated the total number of fruits or flowers in the canopy by multiplying the mean number of fruits in 3 fields of view with the total number of fields needed to view the entire canopy. Finally, we gave each feeding site a productivity score based on these weekly estimates of flower and fruit productivity: 0 = no fruits/flowers observed; 1 = 1–50 fruits/flowers; 2 = 51–100 fruits/flowers; 3 = 101–500 fruits/flowers; 4 ≥ 501 fruits/flowers.

We plotted the distribution of low productivity sites (scores 1 and 2) and high productivity sites (scores 3 and 4) each week and used the nearest neighbor analyses in ArcView to determine if these sites were clustered, random, or dispersed in their distribution in the tamarins’ home range, as calculated by the minimum convex polygon (MCP) method. We compared the feeding bout lengths in high- (score of 3 and 4) and low- (score of 1 and 2) productivity trees using the Mann–Whitney U test using SPSS. We made no assessment of the nutritional quality of different species of fruits and flowers. As we only collected information on the location or productivity of trees that were visited by the tamarins, we cannot determine how many feeding sites they were unaware of, or if they chose to avoid some feeding sites because of their low productivity or location. Thus, we cannot assess whether the monkeys chose the most productive feeding sites among all that were available.

We entered all GPS location points into ArcView to map the distances, directions, travel routes, and spatial distribution of feeding sites. In a previous study, all group members were within 15 m of one another 95 % of the time (Lopez-Rebellon 2010); thus location data collected following one individual are representative of the location of the entire group. Using the Home Range Tools Extension (Rodgers et al. 2005), we calculated the group’s home range size using both the MCP method and the kernel density (KD) method. For the MCP method, a polygon was generated that encompassed all the location points in the data set. For the KD method, we used the least-squares cross-validation equation to calculate a smoothing parameter (or bandwidth) and generated volume contours to identify areas in which the group spent 95 % of its time (Rodgers et al. 2005).

We calculated the circuity index (CI) for paths in which the monkeys moved from one feeding site to another. The CI is calculated as the monkeys’ actual path length divided by the straight-line distance; thus a CI of 1.0 indicates straight-line travel whereas a CI of 1.30 indicates travel was 30 % greater than the straight-line distance. We eliminated any paths that included ≥10 min of insect foraging, resting, or missing data. We also excluded any paths that were interrupted by an intergroup encounter or the sighting of a predator from our analysis. From these data we calculated the tamarins’ CI values when traveling between sequential feeding sites. Given the accuracy of our GPS readings, we eliminated paths that were <20 m to avoid including potentially erroneously high CI values in our results.

Using the step model developed by Janson (1998) and Cunningham and Janson (2007) and adapted by David McGarry for general use, we tested whether the tamarins used random travel to encounter feeding sites. To use this model, we entered into a spreadsheet the distances the monkeys moved and the turning angles they took to move from one major feeding site to another. The model considers the distance traveled between each GPS data point and the turning angles from one point to the next to be a “step” in their daily movement. The model then randomly selected steps from these data to generate a path of a given target distance, e.g., to move the tamarins a distance of 100 m, 105 m, 110 m, etc. The model calculated target distances at 5-m intervals between 100 and 500 m and generated 200 random paths for each target distance. We then calculated a CI for each random path generated. We ranked these paths from highest to lowest in directness (CI ratios). Then, for each of our actual feeding paths, we identified the rank of the CI ratio that most closely matched the monkey’s actual directness ratio. Finally, we calculated the probability that the monkey’s actual directness ratio for any given path was better than the ratio predicted if the monkey had been moving randomly [P = (rank +1)/200] and used Fisher’s method of combining probabilities to determine if the difference was significant. We also used Excel to compare the CI values of first, last, and mean visits to repeatedly used feeding sites using a two-tailed Student’s t-test. In addition, we used StatPlus to run Pearson correlation tests to determine if CI values correlated with the distance traveled by the tamarins to reach MFTS and RF trees.

We also examined tamarin ranging patterns for evidence of travel networks or route segments that were reused on at least 2 observation days. We considered travel routes that ran parallel to one another (not deviating >30 m at any one point) for ≥75 m to be the same route. We identified routes visually from ArcView–generated maps. Although we evaluated path lengths of ≥100 m for the step model (a distance set by Janson 1998), we set our criterion for identifying routes as paths that ran parallel for ≥75 m, as several parallel paths in the 75–100 m range were apparent in our maps. If more than two paths ran parallel, we chose the two that ran parallel for the longest distance as the route. To calculate the percentage of daily travel that occurred on these routes, we converted all routes and daily paths into polygons by creating a 12.5-m buffer on each side of the route and path lines using ArcView. We then used ArcView to calculate the area of overlap between these polygons. Finally, we calculated the percentage of each daily path that was located on a route by dividing the area of overlap by the total area of the daily path.

To examine more closely the actual paths the monkeys took, and to determine whether these were more consistent with a route-based or coordinate-based spatial representation, we visually identified areas of the forest (based on a 25 m × 25 m grid superimposed on a map of the study area using ArcView) in which the tamarins made turns of ≥45°, indicating a change in direction, on three or more separate occasions. We call these locations switch points. We conducted a nearest neighbor analysis in ArcView to determine whether switch points were distributed randomly, clustered, or dispersed within the group’s home range (as calculated by the MCP method).

Assuming that tamarin foragers retain spatial information on the location of previously visited feeding sites, we used ArcView to calculate whether the tamarins reduced daily travel distance by visiting major feeding trees in an efficient sequence. We accomplished this by comparing the order in which five or more MFTS trees were visited by the monkeys on a given day (N = 18 d), with the most efficient order that would reduce overall travel distance. We repeated this same procedure using the RF trees (N = 16 d). If a feeding tree was visited twice on a given day, we still used this travel route for these analyses if the monkeys visited at least one other feeding site before the revisit occurred. We also generated four random routes, by taking the list of feeding sites used on that day and randomizing the list four times. Using the measure tool on ArcView we then calculated the travel distances required to visit the sites in the randomly generated order. We then compared the sum of these random path lengths to the sum of the path lengths taken by the monkeys to visit these feeding sites, based on the order of tree visitation.

In all statistical analyses, we set significance at P < 0.05. The methods for this study were approved by the IACUCs at Northern Illinois University and the University of Illinois Urbana-Champaign and adhered to the laws governing animal research in Bolivia.

Results

Saddleback Tamarin Feeding Behavior

Based on time spent feeding and foraging over the course of the 32 observation days, the tamarin diet was composed principally of ripe fruits (60.6 %), insects (21.8 %), exudates (11 %), and floral nectar (6.4 %). Fruits and insects were consumed on all 32 days, exudates were consumed on 31 days, and nectar was consumed on 21 days. Overall, we documented 715 individual feeding bouts. On each day the tamarins engaged in 16.8 (±4.9) plant feeding bouts and 6.8 (±2.1) insect feeding bouts (mean number of daily feeding bouts was 23.3 ± 6.0). Mean bout length was longest when feeding on fruits (8.2 min) and nectar (8.0 min) and shortest when feeding on insects (5.3 min) and exudates (5.4 min) (Table I). Across all food types, mean bout length was 7.0 min. Overall, trees from 13 plant species accounted for 77.4 % of tamarin plant feeding time. In total, we identified 85 MFTS and 69 RF trees.

Table I Tamarin feeding and foraging bout lengths on different food types
Full size table

During the study period, the tamarins occupied a home range of 25.5 ha as calculated using the kernel density method and 35.5 ha as calculated using the minimum convex polygon method. Mean daily path length was 1718 ± 333 m (N = 15 dawn to dusk day follows) as calculated from GPS points collected at 10-min intervals. In general, the tamarins exhibited a pattern of ranging in which several individual trees each from a small number of target tree species were the focus of daily feeding activities. For example, a mean of nine trees from three species accounted for 52.5 % (±18.4) of plant feeding bouts each day and 69.2 % (±19.2) of daily plant feeding time.

Based on a comparison of feeding behavior in MFTS sites of different productivity scores, we found that tamarins fed more frequently in trees of higher productivity [60 % of bouts (N = 71); productivity scores 3 and 4)] than in trees of lower productivity [(27 % of bouts (N = 30) productivity scores of 1 and 2]. The monkeys fed on trees with a productivity score of 0 on two occasions (2 % of bouts), and for 13 feeding bouts productivity scores were not collected (11 % of bouts). However, the total amount of time the tamarins fed per week in trees of higher and lower productivity trees did not differ (Mann–Whitney U test: N = 87, U = 763, P = 0.41). Highly productive MFTS sites (productivity scores of 3 and 4) did not exhibit a more clumped distribution; these sites were randomly distributed for all weeks except week 3, which was dispersed [nearest neighbor analysis: week 1, nearest neighbor ratio (NNR) = 1.07, z = 0.56, P = 0.57; week 2, NNR = 1.02, z = 0.27, P = 0.78; week 3, NNR = 1.27, z = 2.18, P = 0.05; week 4, NNR = 0.87, z = –1.22, P = 0.22; week 5, NNR = 1.18, z = 1.33, P = 0.18; week 6, NNR = 1.12, z = 1.07, P = 0.28)]. The less productive feeding sites were not clumped either [scores of 1 and 2; these sites were randomly dispersed for all weeks except week 5, which had a dispersed pattern (nearest neighbor analysis): week 1, NNR = 1.37, z = 1.41, P = 0.15; week 2, NNR = 1.22, z =1.13, P = 0.26; week 3, NNR = 1.18, z = 1.12, P = 0.26; week 4, NNR = 1.08, z = 0.80, P = 0.42; week 5, NNR = 1.44, z = 2.63, P = 0.01; week 6, NNR = 0.85, z = –0.80, P = 0.42] (Fig. 1). Therefore, the monkeys had to travel equally as far to encounter trees of greater or lesser productivity. Thus, the spatial relationship of nearby feeding sites rather than site productivity appeared to be a more important factor in tamarin foraging decisions.

Fig. 1
figure 1

Location of major feeding tree species sites with different productivity scores in weeks 1–6 of the study. The study weeks are listed below each map, which is delineated with the home range boundary (as generated by the maximum convex polygon method).

Full size image

Random vs. Directed Travel

A comparison of the actual paths (N = 99; for travel routes between 100 and 500 m) taken by the tamarins to those generated by the Janson’s (1998) and Cunningham and Janson’s (2007) step model (N = 200 for every 5-m increment between 100 and 500 m, total N = 16,000) indicated that tamarin travel was not random (Fisher’s method of combining probabilities: χ2 = 438.6, df = 198, P < 0.001). This result was consistent across all distances (100–500 m) examined.

Given evidence of nonrandom travel, we examined the sequential order in which major feeding sites and sleeping sites were visited each day to determine if tamarin travel was consistent with a distance reducing principle. The monkeys visited a mean of 6.5 ± 3.1 MFTS sites per day (N = 85). On 14 of 21 d for which we tracked the tamarins moving between ≥5 MFTS sites, the monkeys chose the most efficient route that minimized distance traveled. For each of the 7 suboptimal days, the monkeys performed better than the randomized routes 93 % of the time. Of the remaining cases, 3.5 % of tamarin routes were equal to the random routes and 3.5 % of the time the tamarins performed worse than the random routes. If we examine the RF trees, we find a similar pattern. The monkeys visited a mean of 7.5 ± 3.3 RF trees per day (N = 69). On 12 of 18 days for which we tracked the tamarins moving between ≥5 RF sites, the monkeys chose the most efficient route that minimized distance traveled. For each of the 6 suboptimal days, the monkeys performed better than the randomized routes 96 % of the time and performed worse than a random route on only 1 day (4 %).

Evidence for a Coordinate-Based Spatial Map

When moving between sequential MFTS sites (N = 109 travel segments; range, 31–499 m; mean distance traveled, 164.3 m ± 101.0; mean straight-line distance between sites, 120.3 ± 75.9 m), the tamarins’ CI value was 1.43, indicating that the monkeys traveled ca. 43 % farther than the straight-line distance. If we examine movement between RF sites the CI value is 1.34 (N = 83 travel segments; range, 31–479 m; mean distance traveled, 176.0 m ± 96.4; mean straight-line distance between sites, 128.9 m ± 69.8). Thus, the analyses using MFTS sites and RF trees lead to the same conclusion: Tamarins are not moving directly between feeding sites.

We found no direct correlations between CI and the distance traveled by the tamarins to reach MFTS sites (Pearson correlation: r 2 = 0.1, F = 1.33, df = 1, P > 0.25). For example, there is no consistent increase in the CI indices with path length or decrease in the percentage of routes with CI indices of <1.15 (Table II). In the case of RF feeding sites, we found that the straight-line distance between sites had no effect on the tamarins’ CI values (Pearson correlation: r 2 = 0.1, F = 0.93, df = 1, P > 0.35). When traveling distances of 50–100 m between RF sites the tamarin mean CI was 1.30, which was identical to the tamarin CI when traveling distances of >250 m (Table III). In addition, when traveling between RF sites in closer proximity to each other, e.g., <50 m, 50–100 m, >100–150 m, as compared to traveling to more distant feeding sites, e.g., 250–300 m, >300 m, there was no consistent difference in the percentage of direct or relatively straight-line routes taken by the tamarins (CI values of <1.15).

Table II Distance traveled to reach major feeding tree species sites and the corresponding circuity index values
Full size table
Table III Distance traveled to reach revisted feeding sites and the corresponding circuity index values
Full size table

We selected seven MFTS sites each visited a mean of 5.1 times to examine CI index values over repeated visits. These trees accounted for 29.1 % of tamarin plant feeding/foraging time. Although we cannot be certain whether we observed the monkeys’ first ever visit to a feeding site, we still would expect CI values to decline as the tree is revisited, unless the monkeys had already determined the shortest route to that tree before our observations began. Our results indicate no significant change in CI indices with repeated visits to these feeding sites (paired Student’s t-test: CI first documented visit vs. CI last visit, t = 0.45, df = 12, P = 0.66; CI first visit vs. CI mean visit, t = 0.48, df = 12, P = 0.48). The CI indices failed to show a pattern of decline over time for any individual tree and for five of the seven trees, the CI values for the last visit to the tree were higher than for the first documented visit to the tree (Table IV).

Table IV CI values, feeding bout lengths, and dietary percentage of major feeding tree species sites
Full size table

We repeated the analyses again using seven RF sites each visited a mean of 6.1 times. These RF trees accounted for 29.8 % of tamarin plant feeding/foraging time. Our results indicate no significant change in CI indices with repeated visits to these feeding sites (paired Student’s t-test: CI first documented visit vs. CI last visit, t = 1.04, df = 12, P = 0.31; CI first visit vs. CI mean visit, t = 1.91, df = 12, P = 0.33). As for the RF trees, the CI indices failed to show a pattern of decline over time for any individual tree (Table V).

Table V CI values, feeding bout lengths, and dietary percentage of revisited feeding sites
Full size table

Evidence for a Route-Based Spatial Map

During the course of our study, we identified 50 switch points each traveled through 3–9 times (Fig. 2). Although seven of these nodes were separated from other orientation points by ≥50 m, the remaining nodes formed tight clusters or perhaps arrays of landmarks (N = 9), which together were each traveled through from 9 to 18 times. These switch point clusters exhibited a dispersed distribution (nearest neighbor analysis: NNR = 1.25, z = 2.0, P = 0.05) across the tamarins’ home range. Overall, the mean number of MFTS sites and sleeping sites per individual node was 0.7 ± 0.8 (range 0–3), whereas the mean number of feeding and resting sites per node cluster was 2.8 ± 1.8 (range 1–6). If we examine the number of feeding and resting sites located in the 25 m × 25 m quadrats adjacent to nodes and clusters, we find there were 0.34 ± 0.84 per node and 0.89 ± 1.05 per node cluster.

Fig. 2
figure 2

Location of switch points in the tamarins’ home range.

Full size image

We next examined the travel route segments that were used by the tamarins on multiple occasions. Based on our analysis, we identified 29 habitually used route segments that ranged in length from 81 to 444 m (mean = 155 ± 84 m). The mean distance between paths (as calculated at 25-m intervals) that were included as the same route segment was 10.7 m (range of means for the individual routes was 2.7–21.5 m). The locations of these travel routes are indicated in Fig. 2. The mean degree of overlap between the monkeys’ daily paths (N = 28 days) and the routes (N = 29) was 50.0 % ± 10.6, range 32.4–74.2 %. Thus, on a daily basis, ca. 50 % of the distance the tamarins traveled was restricted to a small set of reused travel routes.

We also plotted the spatial relationship between travel routes, switch points, MFTS sites, and sleeping trees (Fig. 3). We found that a mean of 3.3 ± 1.2 different route segments intersected at a single switch point cluster. In addition, we examined 42 tamarin travel paths that were ≥100 m in distance to better determine route and node use. On these paths the monkeys visited 27 different sleeping and resting trees (5 trees were revisited on multiple occasions). To arrive at each feeding or resting site the monkeys traveled on a mean of 1.17 ± 0.7 different route segments (total 20 different routes) and passed through a mean of 1.90 ± 1.3 nodes (total 36 different nodes). Of the 27 individual feeding/resting sites visited 9 (33.3 %) were located in quadrats also defined as nodes or switch points.

Fig. 3
figure 3

Location of travel routes, major feeding tree species sites, and sleep sites in the tamarins’ home range.

Full size image

Discussion

Our findings are consistent with previous studies of tamarin foraging behavior in terms of dietary breadth and foraging bout length. Like Garber (1988, 1993), we found that individual feeding bouts were relatively short (5–8 min); members of a group visited 10–20 fruit, nectar, and exudate feeding trees in a single day; and several individual trees of two or three target tree species were the focus of daily ranging activities. However, although previous studies indicated that the amount of food at a feeding site was an important factor in determining tamarin foraging decisions (Garber 1988, 2000; Garber et al. 2009), our data indicate that on a weekly basis tamarins fed in trees of higher and lower productivity with the same frequency. It is possible that for the small-bodied tamarins, even trees with a limited number of fruits or flowers may represent an adequate feeding reward. In addition, it may be that the low productive trees at our site are of higher food quality (in terms of nutrient content) than those at other sites, leading the tamarins to revisit them more frequently. Future research projects on tamarin foraging decisions should assess the nutritional rewards of individual plant species in addition to their productivity.

Our analyses of the tamarins’ movement between important feeding trees indicate that the monkeys performed better than was predicted if their movement was random. In addition, our data show that in general, the monkeys visited feeding sites using the most efficient sequence. These results do not support predictions 1 or 2; thus, we reject hypothesis 1: The monkeys are not moving randomly through their home range.

The ability to reduce overall daily path length by selecting feeding sites in a relatively efficient progression (as opposed to straight-line travel between sequential feeding sites) also has been reported in other primate taxa including mixed species troops of mustached tamarins (Saguinus mystax) and saddleback tamarins (Saguinus fuscicollis nigrifrons: Garber 1989; Garber and Hannon 1993), vervets (Chlorocebus aethiops: Cramer and Gallistel 1997), Geoffroyi’s spider monkeys (Ateles geoffroyi: Valero and Byrne 2007), black-horned capuchins (Cebus nigritus: Janson 2007), white-faced capuchins (Cebus capucinus: Urbani 2009), chacma baboons (Papio hamadryas ursinus: Noser and Byrne 2007), and common chimpanzees (Pan troglodytes: Menzel 1973). The use of a distance-reducing principle suggests that the tamarins, like several other taxa of primates, are able to maintain some form of internal representation of either the specific or relative locations of these trees to one another and visit them in a spatially efficient manner.

Our results indicate that the monkeys’ ability to revisit trees in an efficient order is not due to a coordinate-based mental representation. We found that the monkeys’ CI indices were considerably greater than 1.0, indicating that they do not travel directly between feeding sites. Further, we found no evidence that the CI indices improved over time to sites that were revisited frequently by the monkeys. Thus, there is no support for prediction 3, and we reject hypothesis 2.

Our data support predictions 4 and 5, indicating that tamarin ranging patterns were most consistent with a route-based spatial representation in large-scale space (hypothesis 3). The tamarins reused a limited set of route segments (N = 29) that intersected at a small number of switch points or landmark clusters (N = 9) to navigate between fruit, nectar, and exudate feeding sites (Figs. 2 and 3). Although the monkeys did not take direct routes or novel shortcuts to reach sequential feeding sites, the order in which trees were visited across an entire day provides evidence for an internal spatial representation in which the relative positions or “place” of salient points in the environment were evaluated to generate a foraging itinerary along their travel network. In addition, given that many switch points contained or were adjacent to feeding sites, it appears that over time, the tamarins may build or revise their internal spatial representation to incorporate the spatial position of frequently visited feeding sites as landmarks in navigating to and locating other nearby feeding sites.

In addition to tamarins, many species of primates are reported to travel across their range in a manner consistent with a route-based spatial representation. This includes several species of lemurs, platyrrhines, and catarrhines (Di Fiore and Suarez 2007; Erhart and Overdorff 2008; Garber and Jelinek 2006; Janson 1998; MacKinnon 1974; Milton 2000; Noser and Byrne 2007; Presotto and Izar 2010; Sigg and Stolba 1981; Urbani 2009). It is likely, however, that many if not all primate species use a combination of cognitive strategies to navigate across their home range including dead-reckoning or path integration, visual sightings of familiar objects, and detailed spatial knowledge (Dolins and Mitchell 2010). Moreover, researchers have made an important distinction between navigational strategies used across different spatial scales (Poucet 1993). Additional research is necessary to investigate the degree to which travel routes, switch points, novel routes of travel, and resource use covary over time and across different spatial scales to understand better species differences in spatial memory and mental map formation in primates.

We argue that the route-based internal spatial representation constructed by wild saddleback tamarins offers both flexibility and reliability in effectively locating productive feeding sites. Within such a framework, tamarins appear to label and group salient places in the forest into a hierarchy of localized routes, switch points, possibly more and less productive feeding sites, and local landmarks (Garber 1988). This is expected to involve the integration of large-scale spatial relationships that are relatively stable over time and a comparison of past, current, and possibly expected future characteristics of local places such as the phenology and productivity of individual feeding sites, which can change over the course of hours or days (Cunningham and Janson 2007; Garber et al. 2009; Presotto and Izar 2010; Thinus-Blanc et al. 2010). Experimental studies in captivity (Saguinus oedipus: Deipolyi et al. 2001; Dolins 2009; Garber and Dolins 2010) and in the wild (Saguinus mystax, S. fuscicollis, and S. imperator: Garber 2000; Garber and Dolins 1996; Garber et al. 2009) support the contention that many species of tamarins are able to integrate disparate types of spatial and nonspatial information in foraging decisions.

A few additional factors, however, should be considered when making conclusions from our results. Field observations of several species of wild primates provide evidence that, in many instances, individuals do not take the most direct or straight-line route when traveling between sequential feeding sites. Although this generally has been discussed in terms of route-based vs. coordinate-based spatial representations, there are several other reasons why a forager might not take the most direct travel route. These reasons include the value of taking a detour in order to sample environmental information and monitor nearby resource availability, the availability of suitable or safe travel routes in some but not all areas of a group’s home range (Laundre et al. 2010; Willems and Hill 2009), and the need to monitor range borders for the presence of conspecific groups (Cunningham and Janson 2007; Di Fiore and Suarez 2007; Erhart and Overdorff 2008; Noser and Byrne 2007). Thus, animals may move indirectly between feeding and resting sites, leading researchers to conclude they lack coordinate-based spatial representations, when in fact they may choose these indirect routes to avoid areas with elevated predation risk, limited appropriate substrates for travel, or few food resources.

In addition, it is important to consider that data on the energetics of travel in both arboreal and terrestrial primates indicate that the costs in time and energy of increased travel are relatively small. For example, based on data from exercise physiology research, Steudel (2000) estimated that the cost of locomotion for a 1-kg mammal accounts for only 1.2 % of its daily total energy expenditure. Even for primates that travel large distances per day, such as hamadryas baboons (Papio hamadryas), which have a mean daily path length of 9500 m, the ecological cost of transport represents <10 % of their total daily energy expenditure. In the case of small-bodied arboreal primates, such as squirrel monkeys (Saimiri sciureus), a daily path length of 3500 m accounts for ca. 4.5 % of total daily energy expenditure (Steudel 2000). Thus for many primates, the time and energy required to travel an additional 25–200 m (most primates can travel 200 m in the span of a few minutes) when moving between a given set of feeding sites may represent a relatively low cost investment. Further, the costs that movement incurs would be offset if additional travel is used to update information on the presence, availability, and productivity of new and previously visited feeding sites. As a result, tamarins may choose to take an indirect route because the potential benefits gained by monitoring resources are greater than the travel costs incurred by moving a longer distance.

Finally, it has been argued that foragers that exploit widely scattered and ephemeral feeding sites face significantly greater cognitive challenges in locating resources than foragers that feed on more concentrated and less seasonal foods (Milton 1988). In the case of primates, when controlled for body mass and phylogeny, species that exploit larger home ranges and species that include a greater proportion of ripe fruit in their diet are generally characterized by increased brain size relative to more folivorous species (Milton 1988, 2000; cf. Reader and Leland 2002). In this study, ripe fruits and floral nectar accounted for 65 % of tamarin feeding time, and the tamarins visited ca. 17 different plant feeding sites per day; thus the tamarins must keep track and retain spatial information regarding the location of many different food resources. Our data suggest that it is the tamarins’ ability to integrate several different types of ecological information, e.g., location and productivity of current feeding sites, phenological state of future feeding sites, along with a route-based mental map that allows them to increase their foraging efficiency in large-scale space.