Wild Tibetan Macaques Use a Route-Based Mental Map to Navigate in Large-Scale Space
Shi Cheng and Bo-Wen Li have contributed equally to this work.
ABSTRACT
Many animals face significant challenges in locating and acquiring resources that are unevenly distributed in space and time. In the case of nonhuman primates, it remains unclear how individuals remember goal locations and whether they navigate using a route-based or a coordinate-based mental representation when moving between out-of-sight feeding and resting sites (i.e., large-scale space). Here, we examine spatial memory and mental map formation in wild Tibetan macaques (Macaca thibetana) inhabiting a mountainous, forested ecosystem characterized by steep terrain that limits direct vision to 25 meters. We used an instantaneous scan sampling technique at 10-min intervals to record the behavior and location of macaques on Mt. Huangshan, Anhui Province, China, from September 2020 to August 2023. Over 214 days, we obtained 7180 GPS points of the macaques' locations. Our study revealed that the macaques reused 1264 route segments (average length 204.26 m) at least four times each. The number of feeding and resting sites around the habitual route segment, terrain roughness, and dense vegetation areas significantly influenced the use of route segments by our study group. In addition, we found evidence that the monkeys reused 48 nodes to reorient their travel path. We found that monkeys approached a revisited foraging or resting site from the same limited set of directions, which is inconsistent with a coordinate-based spatial representation. In addition, the direction in which the macaques left a feeding or resting site was significantly different from the straight-line direction required to reach their next feeding or resting site, suggesting that the macaques frequently reoriented their direction of travel to reach their goal. Finally, on average, macaques traveled 24% (CI = 1.24) farther than the straight-line distance to reach revisited feeding and resting sites. From our robust data set, we conclude that Tibetan macaques navigate large spaces using a route-based mental representation that appears to help them locate food resources in dense, rugged montane forests and heterogeneous habitats.
Summary
-
The daily travel patterns of wild Tibetan macaques were nonrandom and goal-oriented.
-
Tibetan macaques reused the same route segments to reach previously visited feeding and resting sites and would consistently change travel direction at a small set of choice points (landmarks).
-
Tibetan macaques are most consistent with a route-based spatial representation in large-scale space.
1 Introduction
Wild animals face significant challenges in orienting and navigating through forested landscapes where their field of view or line of sight is restricted by dense vegetation (Abrahms et al. 2021; Garber 1988, 1989, 2024; Porter and Garber 2013). Furthermore, given the temporal and spatial variability in resource availability and distribution, the ability of foragers to encode, store, and recall information about the location of previously encountered feeding sites is critical for enhancing foraging success (Fagan et al. 2013; Milton 1981; Ranc et al. 2021; Robira et al. 2021; Rosati 2017; Garber 2024). Several theories have been proposed to explain how foragers relocate productive feeding sites. These include random foraging (Bartumeus et al. 2005), systematic foraging (Baum 1987; Ohashi, Thomson, and D'souza 2007), a Lévy walk model (Shaffer 2014; Codling, Plank, and Benhamou 2008), the use of olfactory cues (Garber and Hannon 1993), or the use of mental maps (O'Keefe and Nadel 1978; Poucet 1993; Tolman 1948; Garber 2000, 2024). These alternative ways of relocating feeding sites differ in how spatial information is encoded and in the types of sensory information used in decision-making (Garber 2000).
Given that many plant species are characterized by temporal synchrony in the timing of fruiting, flowering, and leafing, this offers a degree of predictability in resource acquisition for those foragers that can recall and integrate spatial and temporal information. For example, Javan gibbons (Hylobates moloch) have been reported to retain knowledge of the intraspecific synchronous fruiting patterns of several tree species in their home range and to use this information to visit several trees of the same species on a given day (Jang et al. 2021). A similar behavioral pattern has been found in gray-cheeked mangabeys (Lophocebus albigena), Japanese macaques (Macaca fuscata), mustached tamarins (Saguinus mystax), saddleback tamarins (Leontocebus nigrifrons and L. weddelli), and chimpanzees (Pan troglodytes) (Garber 1988, 1989; Janmaat et al. 2012, 2013a; Menzel 1991; Porter and Garber 2013; Porter et al. 2021). In contrast, other tree species may flower and fruit asynchronously (Milton 1981; Hanya, Tsuji, and Grueter 2013). This presents a different set of challenges, as foragers may need to periodically monitor individual trees of a given species across their range to determine when those trees contain food resources (Janmaat et al. 2016). In addition, given that virtually all primates are social foragers, individuals may locate feeding sites by following “knowledgeable” group members to a feeding site or by taking advantage of conspecific food calls (Bicca-Marques and Garber 2005; Sacramento and Bicca-Marques 2022).
Previous studies of spatial memory in primates indicate that foragers often travel in a relatively straight line to reach distant feeding or resting sites that are out of their field of view. This is referred to as goal-directed travel, as it is assumed that the forager has encoded a mental representation (map) of its current location and the general or specific location of its goal (Janson 1998; Valero and Byrne 2007; Noser and Byrne 2007a). While it is theoretically proposed that primates use “mental maps” to encode locations and spatial relationships between resources, the specific form of the mental map(s) used by primate species remains less clear (Janson and Byrne 2007; Garber and Dolins 2014; Garber 2024). Debates persist about how primates represent spatial information in large-scale space, where the forager cannot see its final destination from its current location (Garber 2024). Specifically, it has been argued that primates represent spatial information as either a route-based mental map (also known as a topological map) or a coordinate-based mental map (also known as a Euclidean map; Poucet 1993; Garber 2000, 2024).
In the context of large-scale space, these maps serve as a framework for navigating vast landscapes (Byrne 2000; Garber 2024). Despite the challenges associated with empirical testing, researchers have proposed examining primate travel routes for indicative patterns that support both types of cognitive maps. A forager relying on a route-based mental map is expected to reach the goal by reusing a set of familiar routes that intersect with salient landmarks or CPs that serve to reorient travel (Porter and Garber 2013). Typically, landmarks or CPs are located within the daily travel routes of animals and have salient features (such as cliffs, emergent trees, or mountain ridges) that serve as decision points that guide them to their next destination (Bufalo et al. 2024; Byrne et al. 2009; Gregory, Mullett, and Norconk 2014; Presotto et al. 2018). Together, these routes and CPs form a route network (Byrne 2000). Foragers using a route-based spatial representation are not expected to take the most direct (shortest distance) path when traveling to an out-of-sight goal and typically have a circuit index (defined as the actual distance traveled divided by the straight-line distance from the starting location to the goal location) greater than 1.2 (i.e., traveling at least 20% farther than the straight-line distance between the current location and the goal; Garber 2024). In contrast, using a coordinate-based mental map, a forager is expected to encode spatial information as X and Y coordinates, allowing it to calculate the actual direction and distance between its current location and its goal (O'Keefe and Nadel 1978). A coordinate-based spatial representation allows individuals to navigate using a least-distance principle, to have a CI approaching 1.0, and to calculate novel shortcuts to reach out-of-sight locations (Gallistel 1990).
In a recent review, Garber (2024) reported that in large-scale space, the movements of virtually all primate foragers are most consistent with a route-based map. The only primate species that are reported to rely on a coordinate-based spatial representation are humans (Homo sapiens) (Epstein et al. 2017; Howard et al. 2014; Lee, Sovrano, and Spelke 2012, but see Warren 2019, for an alternative argument) and possibly chimpanzees (P. troglodytes) (Normand and Boesch 2009, but see Janmaat, Ban, and Boesch 2013b, for an alternative explanation), and Western gorillas (Gorilla gorilla) (Salmi et al. 2020). The sophistication of these cognitive maps across species has been likened to their respective cognitive abilities (Poucet 1993; Tolman 1948; Warren 2019). Furthermore, the evolution of enhanced cognitive abilities in primates has been attributed to selection pressures associated with foraging requirements (Clutton-Brock and Harvey 1980; Milton 1981). The cognitive mechanisms underlying primate movement have received particular attention due to their comparatively large brains, the spatiotemporal dispersion of their food resources, and their phylogenetic proximity to humans (Clutton-Brock and Harvey 1980; Janmaat et al. 2016; Trapanese, Meunier, and Masi 2018; Garber 2024). Although the navigating strategy of primates is likely influenced by constraints and affordances resulting from cognitive mechanisms (e.g., spatial memory), they are also partially based on current local ecological factors, including resource distribution, topographic features, and forest structure (Abreu et al. 2021; de Guinea et al. 2019; de Raad and Hill 2019; Trapanese, Meunier, and Masi 2018). For instance, Abreu et al. (2021) found that wild common marmosets (Callithrix jacchus) living in a semiarid environment use the habitual route network characterized by substantial reuse of route segments located near feeding sites and within sparsely vegetated tree areas. Furthermore, various geographic and landscape features, such as elevation, slope, aspect, or substrate play a pivotal role in determining the location of frequently traversed routes (Abreu et al. 2021; de Guinea et al. 2019; Di Fiore and Suarez 2007). Additionally, visual access to the landmarks has been considered critical for directional decision-making and for travel route adjustments in animals that navigate using a route-based map. Landmarks appear to be more prominent in areas with steeper terrain and proximity to important resources, thereby facilitating easier visual access to and acquisition of these resources (Bufalo et al. 2024; Presotto et al. 2018). This visual access not only aids in making informed directional choices but also facilitates the reorientation of travel routes, demonstrating its importance in primate navigation.
In addition to understanding the use of mental maps, it is also critical to investigate how ecological factors interact with cognitive processes to gain a better understanding of primate navigation strategies and spatial cognition. In this study, we examine navigation and spatial memory in Tibetan macaques (Macaca thibetana), a near-threatened primate species that is endemic to China. Tibetan macaques inhabit dense, rugged montane forests and cliff ledges at altitudes from 300 to 2100 m a.s.l. (Li and Kappeler 2020), and form multimale multifemale groups that average 35.3 individuals (range from 21 to 51 individuals; Li 1999; Li and Kappeler 2020). A study by Li et al. (2022a) found that their diet is composed principally of 52.5% fruits, 23.3% leaves, 14.4% bamboo shoots, 4.5% flowers, and 3.2% young buds and stems. Tibetan macaques have an average daily travel distance of 1,735.6 ± 288.3 m and a home range area of up to 13.6 km2 (Li et al. 2022b). Given that Tibetan macaques spend 61.3% of their time on the ground and travel through a densely forested landscape (Li et al. 2022c), we examined spatial strategies used by these primates when navigating to encounter distant feeding and resting sites. We hypothesized (H1) that travel in Tibetan macaques is not random and that patterns of movement are consistent with the ability to encode and recall spatial information regarding the location of previously visited feeding sites across their home range. If H1 is confirmed, and the movement patterns used by Tibetan macaques are most consistent with a route-based spatial representation in large-scale space, then we expect their circuity index to exceed 1.2 and increase with increased distance to their goal (H1P1), travel routes would be frequently reused (H1P2), the monkeys would reuse the same route segments to reach previously visited feeding and resting sites (H1P3), and the macaques would consistently change travel direction at a small set of choice points (landmarks) or clusters of landmarks (nodes) (H1P4). In contrast, if the macaques were navigating in a large-scale space using a coordinate-based spatial representation (H2), we expected their circuity index to be less than 1.2, the macaques would take direct paths and novel shortcuts to reach their destination (H2P1), and the monkeys would reach the same feeding or resting sites from any number of different paths and directions (H2P2). We also hypothesized (H3) that the reuse of route segments and the distribution of nodes would correlate with the location of frequently used resources and the structure of the forest. Specifically, we expect higher reuse of route segments in the vicinity of feeding/resting sites and located in areas with less terrain roughness (H3P1); nodes would occur in association with important resources (feeding/resting sites) and high elevations or steep slopes with better visual accessibility (H3P2).
2 Methods
2.1 Study Area and Subjects
The study was conducted at the Niejiashan Field Station (30°9′53″ N, 118°15′21″ E) located in the periphery of Mt. Huangshan and the Tianhu Nature Reserve, Anhui Province, China (Figure 1). This region falls within the north subtropical monsoon climate zone and experiences four distinct seasons. The average yearly temperature is 15.5°C. The highest temperature reaches 38.1°C in July, while the lowest temperatures drop to −13.1°C in February (Li et al. 2022a). Total annual rainfall is 2639.4 mm, and precipitation is concentrated in early spring and summer (from March to July). The macaques at Mt. Huangshan typically forage at an altitude of up to 1500 m a.s.l. across a landscape characterized by steep slopes and rugged terrain (Xiong 1984). The area is characterized by subtropical vegetation with a strong vertical distribution with elevation. Pinus massoniana forest is present below 400 m a.s.l. At elevations of 400–600 m a.s.l., the area is dominated by evergreen broad-leaved forest, evergreen broad-leaved and deciduous broad-leaved mixed forest, while at 600–1200 m a.s.l., evergreen broad-leaved and deciduous broad-leaved mixed forest dominate. Deciduous broad-leaved forest is present at an elevation of 1200–1500 m a.s.l. (Li et al. 2022c; Xiong 1984). Approximately 82.5% of the group's home range is forested.
We followed and observed a wild, non-provisioned group of Tibetan macaques that had been tracked for three consecutive years before our study (Li et al. 2021). The group was habituated to the presence of observers at a distance of 10 m (Figure 2). All adult male and female group members were unambiguously identifiable by their size, facial appearance, pelage characteristics, and surface wounds. The group traveled as a relatively cohesive social unit. The home range of our study group was approximately 13.6 km2 and ranged in elevation from 225 to 1100 m a.s.l. (Li et al. 2022b). During our study, group composition varied from 21 to 29 individuals, due to the emigration of males upon reaching adulthood. At the end of data collection, the study group consisted of 22 individuals: 4 adult males, 7 adult females, 2 sub-adults, 6 juveniles, and 3 infants. To estimate the maximum sighting distance for the macaques group members, we randomly selected 50 locations within the group's home range. At each location, we measured visibility or the distance we could see in each of the four cardinal directions, using a range finder at the height of 1.8 m (Kell et al. 2023). The mean sighting distance was 22.45 ± 9.48 m. Therefore, we assumed that the macaques' direct sighting distance was limited to 25 m on average.
2.2 Data Collection
We followed the group from its morning sleep site to its evening sleep site, whenever possible, for 4 to 18 consecutive days each month. Factors that caused us to lose or not locate the group included bad weather (rainy or snowy), difficult terrain (e.g., ravines or cliffs that the group could easily cross), and periods when the monkeys were moving too fast for us to follow. In total, we observed the group for 214 full days (7.6 ± SD 1.4 h/day, range: 6.0–12.5 h/day) from September 2020 to August 2023. During each full-day follow, we studied the behavior of the monkeys (e.g., feeding, resting, traveling, socializing, and other) using an instantaneous scan sampling method at 10-min intervals (Altmann 1974). During each interval, we recorded the behavior of each visible adult group member, as well as the location of the center of the group every 10 min at 1-min intervals (automatically recorded by a Garmin Etrex 329x GPS; mean GPS error = 3.5 ± SD 1.8 m). We recorded the GPS center of the group by having a researcher walk to the estimated center of the group. Given that we estimated the group spread to be generally between 5 and 15 m, our GPS point for group location was approximately 7.5 m from the most distant group member. To collect dietary data, we recorded the items consumed (e.g., fruits and seeds, young leaves, mature leaves, flowers, stems, bamboo shoots, fungi, insects, other), the plant or fungal species consumed, and the number of adult individuals feeding simultaneously at 2-min intervals. We were able to identify 88.8% of the 112 plant species consumed in the field. For those species we could not identify in the field, we collected specimens for identification. These samples were identified by Doctor Si-Yu Zhang from Anhui Normal University.
2.3 Data Analyses
We loaded all GPS data points into a Geographic Information System (ArcGIS 10.8.1, ESRI 2020). We transformed the points from the WGS 1984 geographic coordinate system to the WGS 1984 UTM Zone 50 N projected coordinate system (Fei et al. 2022). To make the location more realistic and accurately represent travel routes, we removed clearly erroneous GPS points or large numbers of GPS points taken at the same location (Bebko 2021; Asensio et al. 2011; Noser and Byrne 2010). After finalizing the GPS points, we created a daily route taken by the macaques.
We defined the daily travel route as the set of GPS points collected from the time group members left their sleeping site in the morning until the time they entered their sleeping site in the late afternoon or early evening (Figure 3a). To ensure that the analyses focused on feeding sites that were important to the group and not just to an individual, only feeding sites where at least two adults had a cumulative feeding time of more than 30 min (three scans) were included in the analyses. We assume that the location of these more productive feeding sites is likely to be encoded in the spatial memory of the group. Our metric of 30 min is based on a study of spatial cognition in mantled howler monkeys (Alouatta palliata; Hopkins 2016). We defined a resting site as a tree or area on the ground where one or more adult group members were found to rest for 20 min (two scans). We defined a sleeping site as a location where the entire group spent the night. The trajectory of the monkeys between two goal sites was described as a travel bout consisting of at least four steps (start point, step 1, step 2, endpoint, or step 3; Figure 3b). During the travel bout, the repeated path is identified and defined as a route segment (Presotto et al. 2019). For all behavioral activities, if the same behaviour was performed within 30 m of the previous GPS point, it was considered the same location. In addition, we defined a direction change as the waypoint at which the macaques changed their direction of travel from their current direction based on the CPT (see Byrne et al. 2009 and below). We calculated the home range of the study group using the Minimum Convex Polygon method (MCP method: Hayne 1949) in ArcGIS.
2.3.1 Random or Directed Travel (H1)
We used the Step Model originally developed by Janson (1998) and later adapted for general use by David McGarry to test whether the likelihood that Tibetan macaques traveled to feeding and resting sites located at least 100 m from the starting location differed from random movement (Cunningham and Janson 2007; Janson 1998; Porter and Garber 2013). The model treats the distance from a starting GPS point to the next point as a step and the goal site may be reached in many steps (Figure 3b). It operates by randomly selecting a start point from the observed data set with the given target distance range of 100 to 500 m and generates a path to the actual goal site in 5-m increments and 5-degree increments. It proceeds by randomly selecting steps, discarding any that would result in a cumulative distance exceeding the target range, which is defined as being within 10% of the target distance. Once the target range is reached, the model calculates the linear distance between the start and end points to derive the circuity index. This process is iterated 200 times for each target distance, with the resulting circuity indices sorted from highest to lowest and ranked from 1 to 200. The circuity index for the actual paths observed during the study period is calculated and similarly ranked. The observed circuity indices are then compared with the closest expected circuity index generated by the model. The model then estimates the probability that the actual circuity index for any given path is better than what would be expected from random movement, using the formula . Finally, Fisher's method is applied to combine these probabilities and assess whether the observed differences are statistically significant.
2.3.2 Circuity Index (H1P1/H2P1)
We calculated the circuity index (CI) for the travel paths taken by the study group as they moved from one goal site to another (Figure 3b). To ensure that Tibetan macaques could not see the goal from their starting location, we only included travel bouts containing at least 4 GPS points, with the final step covering a distance of greater than 30 m, which was beyond the average sighting distance (25 m). In addition, we used Spearman rank correlations to test whether the CI value varied with the distance traveled by the group to the target.
2.3.3 Habitual Route Segments and Route Networks (H1P2 and H1P3)
To investigate whether macaques repeatedly reused the same route segments throughout the study period, we applied the Habitual Route Analysis Method (HRAM) proposed by Presotto et al. (2019). The HRAM tool loads daily routes in line shapefile format and creates a buffer distance around the routes. The buffer created around the route is based on the visual field of the study species in their environment. Given that the field of view of the macaques in our study group was 25 m, we chose 30 m as the buffer size in our analysis. HRAM then overlaid the buffers on all daily travel routes, one at a time, to determine which route segments were previously used. We then calculated the number of times a route segment was reused. We defined a habitual route segment as a distance of at least 75 m that the macaques reused at least 4 times during the study period based on de Guinea et al.'s (2019) study of black howler monkeys (A. pigra).
2.3.4 Change Points and Nodes (H1P4)
To identify change points (CPs) or areas in the group's range where the macaques consistently changed direction, we applied the Change Point Test (CPT) developed by Byrne et al. (2009). This test is a statistical procedure that assesses the degree of collinearity between the vectors representing the path after a given point is aligned with those used previously. If the vectors are not aligned, the test assumes that a change in direction has occurred. Since CPT is sensitive to the number of vectors selected (q), we used q values between 2 and 10 on all daily routes and found that the best predictor for our data set was q = 4. Therefore, we set the q at 4 and the alpha level at p < 0.05 (Byrne et al. 2009; Noser and Byrne 2014). To determine whether Tibetan macaques consistently used a small set of nodes as reorientation or decision-making points, we tallied the number of times each change point was used and conducted a nearest-neighbor analysis in ArcGIS to determine whether they were randomly distributed, clustered, or dispersed within the group's home range (MCP method: Figure 1). We considered all locations where CPs concurred with intersections as nodes (Presotto et al. 2018; Watkins et al. 2022).
2.3.5 Leaving Directions (H2P1)
We recorded the initial direction the Tibetan macaques took when traveling toward a goal and compared it to the straight-line direction leading to that goal (for more details, see de Raad and Hill 2019). We then determined whether the macaques consistently used the straight-line direction to reach the goal from different starting points. The resulting angle would be close to zero. Such a pattern is consistent with a coordinate-based spatial representation. We measured the angular deviation and converted it to a scale from 0° to 180°. The null hypothesis of Moore's paired test is that there is no difference between the two samples. A probability less than the chosen significance level (usually 0.05) indicates a difference between the sample pairs and is consistent with a route-based rather than a coordinate-based spatial representation.
2.3.6 Approach Directions (H2P2)
To determine whether Tibetan macaques reached the same travel goal from a single, multiple, or random direction, we calculated the angle of the last step taken each time that goal was visited. This was treated as a compass direction that deviated from true north by an angle between 0° to 360° (Abreu et al. 2021; de Raad and Hill 2019; Lührs et al. 2009; Normand and Boesch 2009). This calculation was performed using ArcGIS, based on the GPS point at the start of the final step and the coordinates of the GPS point at the final step (de Raad and Hill 2019). For this analysis, we only included travel goals with more than 10 visits. This was done because a smaller sample size has the potential to skew a normal distribution (Landler, Ruxton, and Malkemper 2019). We performed a parametric Rao's spacing test for each travel goal in the software program Oriana 4.0 (Kovach Computing Services 2013). Rao's spacing test takes as its null hypothesis that the data are uniformly distributed. This was tested by determining whether the spacing between adjacent points roughly equaled a 0° to 360° distribution, using the equations from Batschelet (1981). For a uniform distribution, the spacing between points should be roughly 360°/n. If the actual spacing is significantly different from this value, then the distribution is nonuniform, indicating that the angles are clustered in a specific direction.
2.3.7 Environmental Factors on the Route Segment Repetition (H3P1)
To investigate if the reuse of route segments is associated with some environmental factors, we extracted the correlation factors of the habitual route segments, including the mean slope, the roughness of the habitual route segments, the vegetation coverage in which the habitual route segments are located, and the number of feeding, resting, and sleeping sites located within 30 m of the buffer zone of the habitual route segment. We created 30 m buffers around the habitual route segments (Presotto et al. 2018) and used the spatial joining method to count the number of feeding, resting, and sleeping sites in the buffer zone of all the route segments in ArcGIS (de Guinea et al. 2019). To quantify the mean slope, and the roughness of the habitual route segments, route segments were transformed into strings of points at 10 m intervals, and we extracted values using the Spatial Analyst Tool from the Digital Elevation Model (obtained from NASA-STRM-Global Digital Elevation Map: http://srtm.csi.cgiar.org/srtmdata/) with a resolution of 30 m. We generated slope from DEM data, extracted slope values by the strings of points at 10 m intervals, and calculated the mean value of each segment of the travel route (Gregory, Mullett, and Norconk 2014; de Guinea et al. 2019). To determine the roughness of the habitual route segments, we extracted elevation values by the strings of points of 10 m intervals and calculated the rate of change in elevation of each route segment. We obtained the Normalized Difference Vegetation Index (NDVI) via Google Earth Engine (GEE) from the Geospatial Data Cloud (http://www.gscloud.cn/home) to match the study area for each month with a resolution of 30 m. NDVI is a measure that reflects changes in vegetation coverage, vegetative biomass, and ecosystem parameters (Sun et al. 2020; Cai et al. 2022). We generated the Fractional Vegetation Cover (FVC) value, which is a measure of vegetation coverage. FVC values greater than 0.6 represent dense tree vegetation, and FVC values less than or equal to 0.6 as sparse tree vegetation, according to Cai et al. (2022). Based on NDVI, we extracted FVC values for all habitual route segments. We also use Landsat 8 remote sensing satellite imagery, which has a resolution of 30 meters, obtained from the USGS Earth Explorer (https://earthexplorer.usgs.gov/), to confirm and identify the land cover of all habitual route segments.
We performed a Generalized Linear Mixed Model (GLMM) with a Poisson distribution and a logit link function using the function “glmer” from the R package “lme4” within R 3.6.3 (R Core Team 2020). The response variable was the number of times that each habitual route segment was used, the predictor variables were the mean slope, the roughness of the habitual route segments, the vegetation coverage in which the habitual route segments are located (land cover), and the number of feeding, resting, and sleeping sites located within 30 m of the buffer zone. As random effects, we included the ID of each route segment to avoid issues related to pseudoreplication. We used the vif function from the car package to measure the multicollinearity among predictor variables. The mean slope, the roughness, the number of times that each habitual route segment was used and the number of feeding, resting, and sleeping sites were continuous variables. Land cover was a categorical variable.
2.3.8 Environmental Factors on the Distribution Nodes (H3P2)
To test whether Tibetan macaques reused specific nodes based on their proximity to areas of high resource availability or different topographic features (e.g., higher elevations or steeper slopes with better visual accessibility, dense or sparse forested vegetation, assuming that sparse forested vegetation helps to improve visual field), we extracted the following environmental variables, elevation, slope, aspect, dense or sparse forest cover for each change point (Presotto et al. 2018; Abreu et al. 2021). In addition, we calculated the number of feeding, resting, and sleeping sites located within a 30 m radius of each change point. We extracted elevation values for all CPs from DEM data using the Spatial Analyst Tool. We generated slope and aspect from DEM data and extracted values of slope and aspect for each change point (de Guinea et al. 2019; Newmark and Rickart 2012; Valderrama-Zafra et al. 2024). We extracted FVC values for all CPs using the Spatial Analyst Tool in ArcGIS.
To determine whether the distribution and reuse of nodes were affected by environmental features, including the number of feeding, resting, and sleeping sites within the macaques' sighting distance of each node. Within the R environment, we analyzed FVC (dense or sparse forest vegetation), and terrain characteristics, including elevation, aspect, and slope. For this analysis, we used the “glmer” function from the “lme4” R package, which uses a binomial error structure and a logit link function. We used the vif function from the car package to measure multicollinearity among the predictor variables. In this model, we selected the node (yes/no) in CPs as the response variable, and the predictor variables were the number of feeding, resting, and sleeping sites, elevation, slope, aspect, and FVC. The ID of each change point was included as a random effect to avoid pseudoreplication problems. Number of feeding, resting, and sleeping sites, elevation, and slope were continuous variables. Aspect and land cover were categorical variables. This approach was chosen to examine the effect of these environmental factors on the nodes used repeatedly by the macaques. P was set at 0.05 for all analyses. For all models, we compared the full model (including all predictor variables) with the null model (including only control predictors and random effects) using likelihood ratio tests. Significant differences led us to assess individual predictor effects and interactions using the ‘drop1’ function in both the full and a reduced model (excluding interactions). Model stability and normality of residuals were checked by visual inspection and comparison of estimates. All models showed stability and no multicollinearity problems, as indicated by VIF values below three.
3 Results
3.1 Range and Feeding Behavior of Tibetan Macaques
Members of our study group traveled an average distance of 1620.35 ± 514.64 m per day and their home range size was 21.87 km2 (Figure 1). Throughout our 36-month study, the monkeys used 893 feeding sites, visited 749 resting sites, and slept in 178 sleeping sites (Figure 4). On average, the macaques spent 169.52 ± 123.33 min at a feeding site. The macaques consumed fruits and seeds from 482 trees of 53 species. They consumed leaves from 190 trees of 34 species, bamboo shoots from 150 sites of three species, stems from 37 trees of 6 species, and flowers from 34 trees of 16 species. In total, the macaques consumed food from a total of 112 plant species.
3.2 Random or Directed Travel
We scored 1642 travel bouts engaged in by the monkeys (Figure 4). A comparison of the actual routes traveled by the macaques with the routes (N = 162, mean = 295.06 ± 115.05 m, range from 100 to 500 m) generated by the Step Model indicated a significant difference (χ2 = 625.32, df = 324, p < 0.001). The routes taken by the monkeys were not random.
3.3 H1P1/H2P1: Circuity Index
The distance traveled by the macaques to reach a feeding or resting site was 165.59 ± 130.54 m (range of 30.65 to 1424.81 m). Additionally, the mean straight-line distance between feeding and resting sites was 138.45 ± 107.88 m. Overall, the macaques had a CI value of 1.24, indicating that, on average, they traveled 24% further than the straight-line distance (Table 1). An analysis of CI values indicated that when the macaques traveled longer distances to reach their goal, their CI significantly increased (Spearman rank correlation: rs = 0.268, n = 1,056, p = 0.000) (Table 1).
| Distance categories (m) | Circuity index | Actual distance (m) | Sample size | Proportion (%) |
|---|---|---|---|---|
| 30 ≤ D < 50 | 1.06 ± 0.08 | 35.58 ± 2.47 | 9 | 0.8 |
| 50 ≤ D < 100 | 1.15 ± 0.37 | 78.83 ± 14.1 | 214 | 20 |
| 100 ≤ D < 200 | 1.21 ± 0.37 | 144.94 ± 28.79 | 451 | 43 |
| 200 ≤ D < 300 | 1.26 ± 0.45 | 244.81 ± 30.3 | 200 | 19 |
| 300 ≤ D < 400 | 1.33 ± 0.54 | 342.43 ± 26.96 | 92 | 8.7 |
| ≥ 400 | 1.45 ± 0.69 | 524.67 ± 147.04 | 90 | 8.5 |
| Total | 1.24 ± 0.45 | 165.59 ± 130.54 | 1056 | 100 |
3.4 H1P2 and H1P3: Habitual Route Segments and Route Networks
Over the 214 observation days, we identified 1264 route segments (mean length = 204.26 ± 151.68 m), which were used a total of 8823 times (mean reuse = 6.98 ± 3.95, range 4–26 times; Figure 5; Figure S1). The top 100 route segments were each reused an average of 14.11 ± 3.44 times.
3.5 H1P4: Change Points and Nodes
Using the method of CPT, we detected 400 CPs, 48 nodes at the intersections of the route segments that showed a clumped distribution (nearest neighbor analysis: NNR = 0.43, z = –11.84, p = 0.00). The average number of times a node was reused was 10.15 ± 4.93 (range 5–23), and the average number of times a change point was used was 8.27 ± 8.13 times.
3.6 H2P1: Leaving Directions
We next examined whether the initial direction taken by Tibetan macaques when traveling toward a goal was consistent with straight-line or direct travel. We found that the initial direction traveled deviated significantly from the most direct angle of travel (Moore's paired test: R = 9.054, p < 0.001, N = 1449).
3.7 H2P2: Approach Directions
We expected that if the macaques were using a coordinate-based spatial representation, they would approach their target from any number of different directions and, from their current position, travel to their target using novel shortcuts and straight-line travel. An analysis of travel directions taken by the macaques to reach feeding trees that were visited more than 10 times (Table 2) showed that the distribution of their approach angles was significantly clumped, with the monkeys frequently arriving at a given feeding site from the same previous direction (Rao's Spacing test, p < 0.01; Figure 6).
| ID | N | μ | r | U value | p |
|---|---|---|---|---|---|
| 3 | 15 | 272.77 | 0.19 | 226.45 | < 0.001 |
| 12 | 15 | 338.73 | 0.28 | 217.09 | < 0.001 |
| 16 | 16 | 48.69 | 0.39 | 195.16 | < 0.001 |
| 21 | 13 | 70.67 | 0.30 | 191.33 | < 0.001 |
3.8 H3P1: Environmental Factors Influencing Route Segment Repetition
Around the habitual route network, we found that the greater the number of feeding and resting sites around the habitual route segment, the more often the same route segment was used. The number of sleeping sites around a route segment was negatively significant when reusing the habitual route segment. Roughness and vegetation density were significantly correlated with the frequent use of the same habitual route segment, which indicates that the route network is usually distributed in lower roughness and densely vegetated areas. The slope of the terrain did not affect route segment repetition (Table 3).
| Estimate | Std. error | Adjusted SE | z value | Pr (> |z|) | |
|---|---|---|---|---|---|
| (Intercept) | 0.77077 | 0.03464 | 0.03466 | 22.236 | 0.000 |
| No. of feeding sites | 0.22985 | 0.01813 | 0.01814 | 12.668 | 0.000 |
| No. of resting sites | 0.36055 | 0.02976 | 0.02979 | 12.105 | 0.000 |
| No. of sleeping sites | −0.06534 | 0.02735 | 0.02737 | 2.387 | 0.016 |
| Roughness | 0.07678 | 0.01999 | 0.02000 | 3.838 | 0.000 |
| Mean slope | −0.02454 | 0.02186 | 0.02188 | 1.122 | 0.262 |
| Land cover (sparse) | −0.09222 | 0.04520 | 0.04523 | 2.039 | 0.041 |
3.9 H3P2: Environmental Factors Influencing the Distribution or Reuse of Nodes
We also found that the macaques reused a given node to reorient travel more frequently when that node was associated with an increased number of nearby feeding sites (mean distance: 47.78 ± 73.33 m). In addition, when traveling to resting sites, the macaques used these trees to reorient travel (mean distance: 50.25 ± 74.09 m, Table 4). This was not the case, however, for sleeping sites (mean distance: 166.17 ± 136.65 m). Moreover, elevation, aspect, slope, and land cover did not affect the distribution or reuse of a certain node.
| Estimate | Std. error | Adjusted SE | z value | Pr (> |z|) | |
|---|---|---|---|---|---|
| (Intercept) | −1.40402 | 0.11397 | 0.11430 | 12.283 | 0.000 |
| No. of feeding sites | 0.40465 | 0.07022 | 0.07044 | 5.745 | 0.000 |
| No. of resting sites | 0.43813 | 0.10284 | 0.10315 | 4.247 | 0.000 |
| No. of sleeping sites | −0.28264 | 0.11251 | 0.11286 | 2.504 | 0.012 |
| Elevation | −0.03782 | 0.10222 | 0.10253 | 0.369 | 0.712 |
| Slope | −0.02414 | 0.08948 | 0.08975 | 0.269 | 0.787 |
| Aspect | −0.13329 | 0.09129 | 0.09157 | 1.456 | 0.145 |
| Land cover (sparse) | 0.32131 | 0.22924 | 0.22995 | 1.397 | 0.162 |
4 Discussion
4.1 Nonrandom Travel Pattern
During a 3-year field study, we investigated the navigation strategies and spatial cognition of a wild group of Tibetan macaques. Our analyses indicated that the daily travel patterns of wild Tibetan macaques were nonrandom and goal-directed. Travel to previously visited and out-of-sight feeding, sleeping, and resting sites was consistent with expectations that the macaques encode and recall spatial information about the location of a large number (n = 645) of previously visited locations in their home range.
4.2 Evidence for a Route-Based Mental Map
The macaques were found to navigate by consistently reusing a familiar set of travel routes, and travel routes were found to intersect at nodes, which the macaques used to reorient travel to reach out-of-sight goals in large-scale space. Notably, certain route segments were reused up to 26 times, with the top 100 most frequently used segments being reused an average of 14 times each. On average, each node facilitated access to distant and out-of-sight goals in large-scale space on 10 occasions. A recent study of spatial memory in wild common marmosets (C. jacchus) showed a similar pattern, suggesting that individuals consistently reused the same set of travel routes (43 route segments were reused 20–43 times) and reoriented travel in locations that contained clusters of CPs (i.e., nodes; Abreu et al. 2021). Presotto et al. (2018) reported that bearded capuchin monkeys (S. libidinosus) reused 170 route segments between four and twenty times and reoriented travel at CPs (on average, each change point was visited 9.6 times), which were located at route intersections. Our findings demonstrated that wild Tibetan macaques travel over habitual routes; however, their efficient navigation suggests that they may possess some form of mental maps.
Similar results were found in other primate species, such as Ateles belzebuth and Lagothrix poeppigii: Di Fiore and Suarez 2007; Eulemur fulvus and Propithecus edwardsi: Erhart and Overdorff 2008; C. jacchus: Abreu et al. 2021; Pongo pygmeaus: MacKinnon 1974, Bebko 2021; Papio ursinus: Noser and Byrne 2007b, de Raad and Hill 2019; P. hamadryas: Schreier and Grove 2014; A. palliata: Hopkins 2011; A. pigra: de Guinea et al. 2019; S. fuscicollis weddelli: Porter and Garber 2013; S. libidinosus: Presotto et al. 2018; Leontopithecus chrysopygus: Bufalo et al. 2024. Consistent with the existing literature, wild animals possess the ability to maintain internal representations of environmental information by recognizing a sequence of landmarks and forming intersections along well-established routes, enabling them to detect their destination and navigate through the route network system (Blattodea: Perna and Latty 2014; Hymenoptera: Collett 2010; Columbiformes: Freeman et al. 2011; Cetacea: Horton et al. 2017; Rodentia: McNaughton et al. 2006; Primates: Trapanese, Meunier, and Masi 2018; Garber 2000, 2024). Our data set aligns with the findings reported in Garber (2024), which suggest that numerous species, including lemurs, New World monkeys, Old World monkeys, and apes, utilize a route-based mental representation to encode spatial information in large-scale space. This reinforcement underscores the likelihood that a route-based spatial representation strategy is prevalent among most primates for navigation, potentially prompting further research to test if other primate species might similarly rely on such route-based spatial representations in large-scale spaces.
Furthermore, we found that the distance traveled by Tibetan macaques to revisit feeding sites was 24% (CI = 1.24) greater than the linear distance between their location and goal. Primates navigating using a route-based spatial representation have been reported to exhibit a CI value between 1.2 and 1.6 (Garber 2024). Although it can be argued that taking the most direct or shortest route offers benefits associated with reduced energy expenditure, there are several reasons why taking the direct route to reach a feeding or resting site may not be optimal. First, taking alternative routes may facilitate resource monitoring, which can be critical for discovering new feeding sites and asynchronously producing trees of a particular species that may produce edible food items in the coming days or weeks (Porter et al. 2021; de Guinea et al. 2019; Di Fiore and Suarez 2007; Hopkins 2011, 2016). Alternatively, foragers may avoid taking the same route as an antipredator strategy (Willems and Hill 2009; Makin et al. 2012; Hodges, Cunningham, and Mills 2014). Moreover, primate foragers may choose more distant routes that coincide with areas of high visibility, such as the uppermost canopy, ridge tops, stream margins, and human-made trails (MacKinnon 1974; de Raad and Hill 2019; Gregory, Mullett, and Norconk 2014; Noser and Byrne 2014; Green, Boruff, and Grueter 2020). For example, chimpanzees (P. troglodytes) living in the tropical montane forests of southwestern Rwanda often travel using human-made trails. Although these trails were not consistent with shortest distance travel, they were consistent with reduced travel costs associated with navigating steep slopes and dense vegetation (Green, Boruff, and Grueter 2020).
When revisiting the same travel goal, the Tibetan macaques were found to approach the goal from a small number of directions. Based on the movement patterns of the macaques, their CI value, and the fact that they rarely took the most direct route to reach their goal, H2 was not supported. We found no evidence that in large-scale space, the macaques traveled using a coordinate-based mental map. de Raad and Hill (2019) reported a similar pattern in chacma baboons (Papio ursinus) and concluded that these primates rely on a route-based map in large-scale space. This contrasts with a study of Western chimpanzees, in which Normand and Boesch (2009) argued that these African apes navigate in large-scale space using a coordinate-based spatial representation. They reported that the chimpanzees traveled directly to reach their target (CI = 1.04) and revisited the same feeding site from a large number of different directions. Moreover, the initial direction taken by the chimpanzees to reach a food source corresponded to the general direction needed to reach that resource. In contrast, several studies on spatial memory and foraging decisions in Western chimpanzees by Janmaat and colleagues argued that chimpanzees rely on long-term spatial memory and botanical knowledge of the fruiting patterns of large fruiting trees in their home range but do not navigate using a coordinate-based spatial representation (Janmaat et al. Ban, and Boesch Ban, and Boesch 2013a, 2013b, 2014; Ban et al. Boesch, and Janmaat 2014, 2016).
The initial direction taken by Tibetan macaques when traveling toward a goal deviate from the straight-line direction, suggesting that they may not have knowledge of the exact direction at the start and end points of each traveling event. This aligns with the assumptions of a route-based spatial representation: a significant proportion of direction changes are expected to result from a reorientation process (de Raad and Hill 2019; Di Fiore and Suarez 2007; Presotto et al. 2018). In contrast, for chimpanzees and gorillas, both of which rely on a coordinate-based spatial representation, the direction of the initial and the actual direction at the start between goals were nearly identical; they may have had the knowledge of the exact direction as coordinate-based spatial representation assumes (Normand and Boesch 2009; Salmi et al. 2020).
4.3 Environmental Factors Influence on the Reuse Route Segment and Distribution of Nodes
We also found that environmental factors influenced both the reuse of route segments and the distribution of nodes. First, our results indicated that the study group exhibited greater reuse of route segments in the vicinity of feeding and resting sites, as well as in areas typically characterized by less terrain roughness. The same pattern has been observed: the repeated use of routes may serve to connect sites with higher densities of feeding trees (MacKinnon 1974; de Guinea et al. 2019; Abreu et al. 2021) and aid in monitoring potential future feeding sites (Di Fiore and Suarez 2007; Porter and Garber 2013; Bebko 2021). Additionally, areas with less terrain roughness enhanced the efficiency of their movement and probably reduced energy expenditure. Second, the study group demonstrated a greater reuse of route segments in areas with dense forested vegetation compared to areas with sparse vegetation. That result was consistent with the results of previous studies (de Guinea et al. 2019; Hopkins 2011) but contrasted with findings in common marmosets (Abreu et al. 2021). Contrary to common belief, Tibetan macaques, as a primarily terrestrial, large-bodied primate species, do not travel more in sparse vegetation, despite the potential for these regions to reduce their travel time and energy expenditure. It is important to note that the open areas were generally located at lower elevations and contained a small number of abandoned tea plantations overgrown with shrubs and multiple roads traversing them (Li et al. 2022c; Xiong 1984). These conditions pose additional energetic challenges and hazards, and increase the risk of exposure to aerial predators such as hawks.
In addition, the macaques reused nodes that were spatially associated with many current feeding and resting sites, reinforcing both the importance of these sites for navigation decisions and same as previous observations on other primates (Presotto et al. 2018; Asensio et al. 2011; Valero and Byrne 2007). However, this was not the case for the sleep sites. Differences in the macaques navigation when traveling to sleep sites compared to resting sites are best explained by the fact that the macaques often sleep near their last feeding site of the day (cf. Li et al. 2022b). Contrary to our predictions, elevation, and slope did not exert influence on the distribution or reuse of a particular node. This may be due to the fact that, although the study area has steep terrain, the variation in elevation gradient (225 to 1100 m a.s.l.) is relatively small, allowing the study group to access and monitor food resources by following the route network. The study group uses the route network, considering both the distribution and the topographic relief of these pathways, to facilitate navigation and food acquisition within the mountain forest.
In conclusion, based on our 3-year study, Tibetan macaques were found to navigate in large-scale space using a route-based mental map characterized by the reuse of travel routes that intersected as nodes (landmarks) that the macaques used to reorient travel. On average, the monkeys traveled 24% farther than the straight-line distance. The Tibetan macaques' large-scale spatial navigation is consistent with that reported for most primate species. In the future, we plan to study Tibetan macaques' spatial memory in small-scale space to better understand their foraging and cognitive abilities.
Author Contributions
Shi Cheng: conceptualization (lead), data curation (lead), formal analysis (lead), investigation (lead), methodology (lead), writing-original draft (lead), writing-review and editing (lead). Bo-Wen Li: conceptualization (equal), investigation (equal), writing-original draft (equal); writing-review and editing (equal). Paul A. Garber: conceptualization (equal), methodology (equal), writing-original draft (equal), writing-review and editing (equal). Dong-Po Xia: conceptualization (equal), funding acquisition (equal), methodology (equal), supervision (equal), writing-review and editing (equal). Jin-Hua Li: conceptualization (equal), funding acquisition (lead), data curation (equal), methodology (equal), supervision (lead), writing-review and editing (equal).
Acknowledgments
The authors are grateful to the Huangshan National Park for permission to conduct research and to all the villagers of Niejiashan for facilitating and supporting our research. We acknowledge our field assistants, Zhong-Bao Yang, and Chun-Sheng Wang, for their assistance in collecting data. We want to thank Si-Yu Zhang for his help in the identification of plant specimens. The authors thank Qi-Xin Zhang and Shen-Qi Liu for their advice on the statistical analysis. This study was funded by grants from the National Natural Science Foundation of China (No. 32370535, 31971404, 32370533, 32070455) and the Anhui Provincial Natural Science Foundation (No. 2108085Y12). Paul A. Garber acknowledges the support provided by Chrissie, Sara, Jenni, and Dax during the writing of this manuscript. We thank the International Collaborative Research Center members of the Huangshan Biodiversity and Tibetan Macaque Behavioral Ecology group, who provided more advice with data collection and paper improvement, especially Bing-Hua Sun, Xi Wang, Wen-Bo Li, Ya-Dong Li, Peng-Hui Li, Jia-Kai Lu, and others.
Ethics Statement
All research protocols reported in this manuscript were approved by the Chinese Wildlife Management Authority. The study was purely observational and did not involve invasive experiments on wild primates. Thus, no review from an institutional ethics committee in China was required. This research was conducted in compliance with the Wildlife Protection Law of the People's Republic of China. All research activities reported in this study followed the regulatory requirements of the Huangshan Garden Forest Bureau in China.
Conflicts of Interest
The authors declare no conflict of interest.
Open Research
Data Availability Statement
The datasets analyzed in this study are available from the corresponding author upon request.
