Anim Cogn (2016) 19:387–403 DOI 10.1007/s10071-015-0941-6

ORIGINAL PAPER

Mantled howler monkey spatial foraging decisions reflect spatial and temporal knowledge of resource distributions Mariah E. Hopkins1

Received: 23 June 2015 / Revised: 9 November 2015 / Accepted: 11 November 2015 / Published online: 23 November 2015 Ó Springer-Verlag Berlin Heidelberg 2015

Abstract An animal’s ability to find and relocate food items is directly related to its survival and reproductive success. This study evaluates how mantled howler monkeys make spatial foraging decisions in the wild. Specifically, discrete choice models and agent-based simulations are used to test whether mantled howler monkeys on Barro Colorado Island, Panama, integrate spatial information in order to maximize new leaf flush and fruit gain while minimizing distance traveled. Several heuristic models of decision making are also tested as possible alternative strategies (movement to core home range areas instead of individual trees, travel along a sensory gradient, movement along arboreal pathway networks without a predetermined destination, straight-line travel in a randomly chosen direction, and random walks). Results indicate that although leaves are the single most abundant item in the mantled howler monkey diet, long-distance travel bouts target the areas with the highest concentrations of mature fruits. Observed travel patterns yielded larger estimated quantities of fruit in shorter distances traveled than all alternative foraging strategies. Thus, this study both provides novel information regarding how primates select travel paths and suggests that a highly folivorous primate integrates knowledge of spatiotemporal resource distributions in highly efficient foraging strategies.

Electronic supplementary material The online version of this article (doi:10.1007/s10071-015-0941-6) contains supplementary material, which is available to authorized users. & Mariah E. Hopkins [email protected] 1

Department of Anthropology, University of Texas at Austin, Mail Code C3200, Austin, TX 78712, USA

Keywords Spatial memory  Primate cognition  Foraging strategies  Foraging cognition hypothesis  Alouatta

Introduction The cognitive demands of foraging may provide an important selective pressure influencing the evolution of spatial cognition, as an animal’s ability to exploit resources that are unevenly distributed in space and time directly impacts its survival and reproductive success (Eisenberg and Wilson 1978; Clutton-Brock and Harvey 1980; Milton 1981, 1993; Platt et al. 1996; Shettleworth 1998; Smulders et al. 2010). Since spatial foraging decisions are often made in nature at a scale that is difficult to replicate in captive environments, studies of naturalistic foraging patterns can provide important insights into a species’ spatial memory form and capacity (Fagan et al. 2013). However, naturalistic foraging patterns are determined both by cognitive ability and foraging strategy (Janson and Byrne 2007). Thus, conclusions regarding cognitive ability from naturalistic foraging patterns will be strengthened by studies that can determine how animals integrate spatial and temporal information into foraging strategies. A general assumption derived from optimal foraging theory is that if an animal has complete spatiotemporal knowledge of available resources and represents this information in a mental map that is able to encode the distances and directions between food sites (i.e., a vector mental map; Tolman 1948; Poucet 1993), it should travel to the closest and most productive resources in its habitat in an energy-minimizing fashion (often interpreted as a straight line; Garber and Hannon 1993; Normand and Boesch 2009b; Porter and Garber 2013). Failure to observe

123

388

this type of travel pattern may indicate incomplete spatial knowledge or the use of a mental map in which exact distances and directions between resources cannot be calculated (e.g., a route-based network map; Milton 1980, 2000; Byrne 1982; Di Fiore and Suarez 2007). However, an equally valid explanation for observed nonlinear travel is the complexity of the decision-making process, in which animals weigh a variety of social, ecological, and/or spatial variables against one another when selecting a foraging strategy (Janson 1998; Hopkins 2011, 2013). In assessing this decision-making process, foraging strategies ranging from random walks to energy maximization have been proposed for a wide range of taxonomic groups, with varying levels of support. Random walks remain the null model for studies of animal travel (Codling et al. 2008; Fronhofer et al. 2013). However, the majority of animal travel patterns exhibit non-random elements inconsistent with true random walks, in which direction and distance traveled between directional changes are chosen at random (Shettleworth 1998; Rodrigo 2002; Schreier and Grove 2010). Correlated random walks that allow for directional fidelity and step models, in which the direction of travel and the distance traveled before the next directional change are randomly chosen from a distribution of observed angles and distances, better approximate observed travel patterns (Conradt et al. 2003; Bartumeus and Levin 2008; Barton et al. 2009). Yet, even these modified random walks are unable to explain much of the directionality and complexity in animal travel patterns, causing researchers to turn to other explanations (Benhamou 2007; Barton et al. 2009; Boyer and Walsh 2010). Maximizing energetic and nutritional gains while minimizing energy expenditure may yield the highest fitness payoff and would require extensive use of spatial and temporal memory of resource distributions (Pyke 1984). Indications that animal travel is more efficient than expected by random search strategies has been found in the behavior of a wide variety of organisms ranging from insects to mammals (Stephens and Krebs 1987; Fagan et al. 2013). Some of the most convincing evidence that animals draw upon spatial memory to maximize energetic and nutritional gain comes from studies of primates, as primates offer unique opportunities for detailed and continuous observation in nature (reviewed in Janson and Byrne 2007). Primates reach food trees in shorter distances traveled than expected from random travel (Garber 1989; Garber and Hannon 1993; Janson and Byrne 2007; Erhart and Overdorff 2008; Luehrs et al. 2009; Normand and Boesch 2009a, b; Asensio et al. 2011; Suarez et al. 2014). A few primate species even appear to incorporate previous visitation patterns when deciding which resources to visit at a given point in time, and may have the capacity to plan

123

Anim Cogn (2016) 19:387–403

long-distance multi-destination routes in an energy-minimizing fashion (Menzel 1973; Altmann 1974; Gallistel and Cramer 1996; Janmaat et al. 2004, 2006; Janson 2007; Suarez 2014). Nevertheless, while many studies have revealed that animals are more efficient than expected from random travel patterns, far fewer assess whether observed statistical patterns could result from simpler ‘non-optimal’ search strategies that rely to lesser degrees on spatial and temporal knowledge. Numerous methods for condensing quantities of spatial information for ease of encoding and/or selection of a travel route have been documented in humans, and several of these methods have been the subject of limited investigation in animals (Wiener et al. 2004). One method to simplify the search process is to condense spatial information into ‘regions’ or ‘neighborhoods,’ rather than remember the exact location of individual sites (Wilton 1979; Chown et al. 1995; Smulders et al. 2010). Humans often condense and average spatial information, making decisions based upon regional values rather than individual metrics, though the degree to which we do so may depend on spatial scale (Wiener and Mallot 2003; Wiener et al. 2004). Unfortunately, identifying how and when animals condense spatial information can be problematic since determining the influence of nearby sites on a particular site’s likelihood of selection (i.e., the ‘neighborhood effect’) depends on the ability of the researcher to match the scale of study to that at which the animal condenses information (Saracco et al. 2005; Jackson and Fahrig 2012). The scale at which the landscape is measured impacts the strength of the measured ecological response, and the most appropriate scale differs across species (Holland et al. 2005; Jackson and Fahrig 2012). Given these documented scale effects, analyzing foraging decisions across a number of spatial scales becomes critical to isolating the cognitive process involved in foraging decisions. For example, among primates, the appropriate feeding patch size is often assumed to be the individual tree within which a primate group feeds (Leighton and Leighton 1982; Chapman 1988; Gentry and Terborgh 1990; Wallace 2008). As a result, studies of the efficiency of primate foraging are conducted at the scale of individual trees, leaving a number of questions regarding the scale at which primates make foraging decisions unanswered. Specifically: Do primates make travel decisions on the basis of individual trees, collecting and remembering information on the size and phenological patterns of each tree? Or can primates approximate optimal energy yields and observed travel patterns simply by targeting areas of their home range that have groups of trees (i.e., larger patches) belonging to preferred species, without integrating any knowledge as to the phenological patterns of individual trees at all? The latter strategy still relies upon spatial

Anim Cogn (2016) 19:387–403

memory, but the quantity and resolution of spatial information remembered could be reduced (e.g., Milton 2000; Hopkins 2008; Smulders et al. 2010). The use of habitual routes in navigation is another method that may simplify foraging problems for humans and animals alike. Encoding spatial information as a series of routes between known landmarks (a ‘network’ map) instead of a coordinate-based system in which the exact distance and direction between locations of interest is known (‘a vector map’) may serve to reduce cognitive load when the number of routes required to navigate is comparatively low (Milton 2000). Indeed, the network map is the more common form of spatial representation used in navigation by humans, with vector maps less commonly used at larger scales, among women, and among those \7 years old (Dabbs et al. 1998; Sandstrom et al. 1998; Istomin and Dwyer 2009). Repeated use of the same travel paths is also common among animals ranging from insects to mammals (Shettleworth 1998; Ohashi et al. 2007), potentially offering similar support for the use of network maps by animals at larger spatial scales (Poucet 1993). Primates, for example, repeatedly use travel routes that follow landscape features such as streams and hilltops (MacKinnon 1974; Milton 1980; Sigg and Stolba 1981; Boonratana 2000; Di Fiore and Suarez 2007; Noser and Byrne 2007; Presotto and Izar 2010). In addition to ease of navigation, such a travel pattern could reduce the energy expenditure required from frequent altitudinal changes in areas with great topographical variation, facilitate resource monitoring, and increase the frequency with which animals come in contact with desired food items (Milton 2000; Di Fiore and Suarez 2007). Nonetheless, it is unclear whether use of a mental map (either vector or network) at all is required to achieve observed energetic gains. Seemingly complex travel patterns can be approximated with relatively simple search strategies that require little use of spatial memory or remembered knowledge of the surrounding environment (Wiener et al. 2009; Petit and Bon 2010). For example, impressive feats of navigation by pigeons could be the result of following a gradient of sensory information, in which the animal orients its direction to the strongest sensory cue at each point in time (Wallraff 2001). Thus far, however, the few simple heuristics tested have provided imperfect explanations for food search by primates. Computer simulations of travel along an olfactory gradient could not explain the directness of tamarin travel between food sources (Garber and Hannon 1993). Other simple heuristics such as picking a direction at random and traveling in a straight line until a food source has been reached (Janson’s ‘geometric model’) or traveling along a habitual route network without a predetermined resource in mind have failed to approximate food search by primates such as gibbons, capuchin, spider, and saki monkeys (Janson 1998; Cunningham and Janson 2007a, 2013; Asensio et al.

389

2011; Suarez et al. 2014). However, these studies have largely focused on primates belonging to one ecological guild (arboreal frugivores) and have yielded results that are both more efficient and less efficient than observed travel patterns, suggesting that further study in this area is warranted. In this study, I employ spatially explicit discrete choice and individual-based movement models to quantitatively assess how mantled howler monkeys (Alouatta palliata) make spatial foraging decisions, using many of the search strategies described above. Specifically, I target three questions: (a) Is there evidence that mantled howler foraging strategies yield greater quantities of food in shorter distances traveled than would be expected from random search strategies? (b) Are mantled howler monkeys making decisions based upon individual trees or neighborhoods? and (c) Can simple heuristics (Janson’s geometric model, traveling along a sensory gradient or habitual path) approximate the efficiency of observed travel patterns? There are no mantled howler monkeys kept at any of the 911 zoos and aquaria across 85 countries evaluated by the International Species Information System (2015), and published studies of howler monkey cognition in captivity are nonexistent. Thus, studies of the naturalistic foraging patterns of mantled howler monkeys are currently our only window into howler monkey physical cognition. Furthermore, the selection of mantled howler monkeys as a model species has important implications for those interested in the evolution of cognition. Early models hypothesizing a dietary basis for cognitive differences drew heavily upon observations of mantled howler monkey travel for support (Milton 1981, 1988, 2000). Mantled howler monkeys belong to the most folivorous New World primate genus; over 50 % of their annual diet and up to 100 % of their daily diet is comprised of leaves, with lesser portions of their diet comprised of fruits and flowers (Milton 1980). The travel patterns of howler monkeys can be highly circuitous, unlike the travel patterns of more frugivorous primates living in the same habitat (Milton 1993). While the nutrient and chemical composition of leaves varies greatly across space and time in tropical forests, preferred leaves are not as patchily distributed as ripe fruits (Glander 1978; Nagy and Milton 1979a, b; Ganzhorn 1992, 1995; Chamberlain et al. 1993; Ganzhorn and Wright 1994; Milton 1998; Chapman et al. 2003). Early studies relied on these observations to support the hypothesis that patchier distributions of fruits generated greater selective pressure for frugivores to develop detailed spatiotemporal representations of resources, and that this cognitive difference could be reflected in the more linear travel patterns of frugivorous primates (i.e., the foraging cognition hypothesis; Milton 1998, 2000; Cunningham and Janson 2007b). Nonetheless, despite this early focus on mantled howler monkey travel patterns, few studies have subsequently

123

390

assessed the efficiency of mantled howler monkey travel using quantitative methods (but see Milton 2000; Garber and Jelinek 2006). Among primates, studies of foraging efficiency have focused on the species thought to have the greatest capacities for spatial memory—the frugivores and insectivores—with much lesser attention on those with perceived lesser capacities, including folivorous primates such as mantled howler monkeys (reviewed in Janson and Byrne 2007). While highly frugivorous primates likely integrate knowledge of resource size, distance to the resource, and previous visitation patterns when deciding which resources to visit at a given point in time (Janson and Byrne 2007; Erhart and Overdorff 2008; Luehrs et al. 2009; Normand and Boesch 2009a, b; Suarez 2014), the same has not been concluded for highly folivorous primates. Previous studies of mantled howler monkeys, specifically, have demonstrated that members of this species travel in a non-random fashion. Groups repeatedly use habitual pathway networks that do not appear to be tied to topography or landscape features such as streams (Milton 1980; Hopkins 2011), and exhibit seemingly ‘goal-directed’ travel, in which an individual or group travels large distances from one feeding tree to the next via relatively straight-line direct routes (Milton 1980, 2000; Garber and Jelinek 2006). They also appear to encounter trees of particular species at a much higher rate than would be expected if they chose food trees at random, and given the foraging sites visited in a single day, mantled howlers have been observed to take a distance-limiting path between them (Milton 2000; Garber and Jelinek 2006). However, while these observations demonstrate non-random habitat use, they cannot be used to make conclusions regarding whether the foraging choice made was the most efficient out of the options available; nor can they accept or reject the use of particular foraging strategies. Thus, results from a study such as this one that quantitatively evaluates the foraging decisions of the most folivorous New World primate better illuminate the cognitive abilities of a species that has been central to the development of theory explaining the evolution of primate cognition.

Anim Cogn (2016) 19:387–403

Methods

study site is 18 individuals (Milton et al. 2008). Groups travel cohesively, often in single file (Carpenter 1934; Milton 1980). This study focuses on the spatial foraging decisions of one mantled howler monkey group with a home range size of 44 ha (95 % MCP; Hopkins 2008; Figure S1). Monthly censuses of this group indicated that it averaged 23.4 individuals (SD = 1.2), with 11.3 adult females (SD = 1.2) and 4.2 adult males (SD = 0.3; Hopkins 2008). The spatial and temporal distribution of available resources were estimated using data from the BCI 50-ha forest dynamics plot (Condit 1998; Hubbell et al. 1999, 2005), which overlapped the study group’s home range. The Center for Tropical Forest Science (CTFS) maps and measures every stem on the 50-ha plot greater than 1 cm diameter at breast height (dbh) every 5 years. Estimated error of locations in CTFS forest plots is\1 m (Condit et al. 2014). The 2005 50-ha plot census took place during this study, providing up-to-date information on the distribution of howler monkey food items. In addition, CTFS conducted a phenology-monitoring program on the 50-ha plot for the duration of this study, in which 200 0.5 m 9 0.5 m fruit traps were checked weekly for fruit fall. Locations of travel paths and feeding data were obtained through behavioral follows by M. Hopkins and three assistants. The study group was followed for 7–10 consecutive days per month from dawn to dusk, with interruptions to data collection only occurring for severe weather patterns (N = 84 days; total observation time = 849 h; mean daily follow length ± SD = 10.1 h ± 1.8 h). During group follows, instantaneous group scan samples were conducted at 10-min intervals. As part of each scan, the locations of all visible individuals were recorded from plot tree tags, and each individual’s behavior was noted. Continuous sampling methods were applied specifically to foraging behaviors, with every tree foraged upon by an individual noted and the item foraged upon identified (i.e., mature fruit, immature fruit, over-ripe fruit, new leaf flush, immature leaves, mature leaves, flowers, petioles). Differentiation between ripe, unripe, and over-ripe fruit was accomplished through qualitative visual inspection. Inter-observer reliability was assessed with a Cohen’s Kappa test (Cohen 1960), with an initial value [0.95 required for fruit differentiation prior to data collection.

Study site and focal animals

Analyses

This study took place between August 2005–June 2006 on Barro Colorado Island (BCI), Panama, a 1500-ha protected reserve in the Panama Canal with an estimated howler monkey population of approximately 1300 individuals (Figure S1). While mantled howler monkeys live in multimale multi-female groups that range greatly in size (2–40? individuals; di Fiore et al. 2010), the mean group size at the

On BCI, mantled howler monkey movements have been characterized as long-distance movements between important feeding trees, followed by a series of scatter feeding events on less important resources within close proximity to these important feeding trees (Milton 1980). Here, all analyses focus only upon long-distance movements between food trees, as movements exceeding the

123

Anim Cogn (2016) 19:387–403

visual range of primates are the most likely to utilize spatial memory. I defined a ‘long-distance foraging movement’ a priori as a travel bout longer than 50 m that ended at a food tree. By traveling 50 m, mantled howler monkeys would exceed previously estimated visual detection distances of primates, thereby ensuring that they were unlikely to have been able to monitor the target of their travel bout from their starting point (Janson and DiBitetti 1997; Milton 2000). To ensure that analyses focused on feeding sites that were important to the group as a whole and not just to an individual, analyses include travel to those trees in which at least two adults foraged for more than one scan. This metric used both the number of individuals and time spent foraging to designate the importance of foraging events to the group as a whole as has been done in previous studies (White and Wrangham 1988; Janmaat et al. 2006; Hopkins 2008). Discrete choice models: Evidence for energy maximization and use of neighborhoods in search strategies? Discrete choice models determined the likelihood of the howler monkeys selecting a tree or group of trees as the target of a long-distance travel bout, given the distance to the target and the quantity of fruit or new leaf flush available. Discrete choice models took the form of a conditional logit model stratified by each long-distance foraging movement (McFadden 1974; Manly et al. 2002; Stata SE10, StataCorp 2007). In other words, at the start of every long-distance movement bout, the howler monkey group was able to choose from all sites on a virtual landscape outside of a 50-m radius from its current location. The discrete choice model then compares the target actually chosen by the howler monkeys with available trees to determine whether the howlers’ selection was the most efficient choice available to them at that given point in time. The virtual landscape was programmed in MATLAB (Mathworks 2008) with field data to be as close as possible to the actual forest and was parameterized in three ways to test the following hypotheses: Hypothesis 1 Long-distance movement bouts target the individual trees that yield the most new leaf flush and fruits at a specific point in time, while minimizing distance traveled: In the first set of analyses, all trees selected by the howlers were compared to a random selection of 20 trees [25 cm dbh belonging to the same tree species. A minimum size requirement was instated as the targets of longdistance travel bouts are known to be large trees, and given the relative abundance of saplings, a truly random selection would be biased towards trees of small size (Milton 1980). All trees were monitored for both new leaf flush and fruit crop size (Chapman et al. 1992). New leaf flush was

391

estimated by multiplying the crown area by a visually estimated percentage of new leaf flush present. Estimates of fruit availability in all trees were completed by taking visual counts of fruits in five randomly selected m2 areas, and then multiplying the mean of these calculations by the measured crown area. Randomly selected trees had to be monitored within 1 day of the travel bout to be included in analyses. As mantled howler monkeys at this site show little evidence of consecutively targeting the same species of fruiting trees (Milton 1980; Hopkins 2008), models had to allow for the inclusion of multiple tree species in a single discrete choice model. In order to do this, fruit crop size was normalized by dividing each tree’s crop size by the maximum crop size measured for that species on that day. This ensured that a bias did not result from the different sized fruits of different tree species. Hypothesis 2 Long-distance movement bouts target the groups of trees (‘neighborhoods’) that yield the most resources at a specific point in time, while minimizing distance traveled: To identify the neighborhoods with the highest concentrations of food items, the 1 km 9 0.5 km 50-ha CTFS plot was discretized into 25 m 9 25 m analysis cells, and discrete choice models stratified by foraging movement were used to determine whether howler monkeys selected the most productive cells, after accounting for distance traveled. Twenty-five meter cells were selected as the unit for analysis, since the mean observed group spread of mantled howler monkeys at this site during foraging events is 24.6 m (Hopkins 2008). During each time period in which a long-distance movement occurred, the primate group was able to choose from all cells within its home range that overlapped the 50-ha plot (N * 390). Those cells within 50 m of the group’s current location (*24 cells) were assumed to be part of the group’s current visual food search and were excluded as target cells. The value of a cell at the particular time of the travel bout was calculated by estimating the relative fruit availability probabilistically. Each tree (i) within a cell (k) was weighted by its size (dbh-meters) and relative production (Rlq). Relative production was calculated as the percentage of the total fruit wet weight (grams) present in the 200 CTFS fruit traps that week (q) that belonged to the tree’s species (l): Estimated fruit productionkq ¼

p n X X 

 dbhil  Rlq ;

i¼1 l¼1

ð1Þ where fruit equivalentslq  wet weightl Rlq ¼ Pp ; l¼1 fruit equivalentslq  wet weightl

123

392

Anim Cogn (2016) 19:387–403

Only tree species responsible for [1 % of the annual diet of mantled howler monkey groups at this site during the study year were considered, and only trees[5 cm dbh were included in analyses, as this was the smallest tree foraged upon by at least two mantled howler monkeys. Weighting each tree by its species’ relative production (Rlq) identifies those cells with the largest concentrations of the tree species responsible for the most fruit production in the given week. Values were then normalized by dividing each cell’s estimated fruit production by the maximum cell value for that week prior to any analysis. Hypothesis 3 Long-distance movement bouts target the core areas that yield the most resources on average across an annual cycle, not at a specific point in time: This model was generated as for Hypothesis 2, but cells were parameterized with a slightly different metric of food availability. The value of a cell across an annual cycle was quantified for each cell (k) by summing the diameter at breast heights in meters of all trees[5 cm dbh (i) belonging to species (j) that were responsible for [1 % of the annual diet of mantled howler monkey groups at this site during the study year (Nspecies = 60; Hopkins 2008). dbh sumk ¼

p n X X

dbhij ;

ð2Þ

i¼1 j¼1

Adjusting for previous visitation patterns To allow for the possibility of resource monitoring, a binary variable indicating whether the howler monkey group had visited the tree in the last 3 days was also included in all discrete choice models. Three days were selected as the analysis time frame to be consistent with previous studies on more frugivorous primates as well as a study of howler revisitation patterns to trees at the study site (Cunningham and Janson 2007a; Hopkins 2008; Suarez 2014). Decision tree models: Can simple heuristics approximate the efficiency of observed howler monkey travel patterns? Individual-based movement models were generated to evaluate whether observed movements led to areas with greater levels of estimated fruits in shorter distances traveled than foraging strategies generated by simple heuristics. The landscape was discretized into 25 m 9 25 m cells as above. Simulations began by placing the group at the starting location of an observed long-distance movement and allowing the group to move across the landscape according to a specified movement hypothesis (Fig. 1):

123

Hypothesis 4 Howler monkey travel follows a random walk. Two types of random walks were considered: correlated random walks with fixed step lengths and a step model using observed angles and distances traveled. In correlated random walks, the group repeatedly traveled for fixed movement steps of 25 m (Pyke 1978; Garber and Hannon 1993; Janson 1998). Angle deviations between steps were selected randomly from a uniform distribution, but could not exceed 60 degrees in order to maintain the directionality common to primate movements (Garber and Hannon 1993). In the step model (Cunningham and Janson 2007a), the group selected the step length and the angle deviation between movement steps from a frequency distribution of observed distances and angles (Fig. 2). One hundred simulations of each type were completed for each observed long-distance foraging movement, and the mean value for each observed movement was used as the unit for analysis. Hypothesis 5 Howler monkeys follow arboreal paths with the highest concentration of food trees. Simulations of primate movement along arboreal pathway networks are inherently difficult as they require independent estimates of where arboreal pathway networks are likely to occur. However, Hopkins (2011) found that the locations of arboreal pathways repeatedly used by mantled howler monkeys on BCI were highly predictable. Moreover, the most significant predictor of the locations of arboreal pathway networks used repeatedly by mantled howler monkeys at this site was total food tree availability as calculated here in Eq. (2). Thus, simulations of movement that follow gradients of food trees likely approximate movement along arboreal pathway networks, even in the absence of complete knowledge of all pathway locations. Simulations began at the locations where long-distance movements were observed to begin. The simulated group was then allowed to examine all neighboring cells (8 cells possible according to first-order ‘queen’ contiguity) and move to the cell with the highest concentration of food trees (Eq. 2) that fell within the group’s home range and that had not been visited previously during the same movement bout. To maintain directional fidelity, angle deviation between subsequent movements followed previous simulations of primate movement and was not permitted to exceed 60 degrees (Garber and Hannon 1993). Hypothesis 6 Howler monkeys follow a visual sensory gradient. This simulation approximates movement through the canopy in which the howler monkeys repeatedly travel to the largest fruit source within their sensory range. Food availability was estimated as per Eq. (1). Howler monkeys traveled to the most productive resource within a 75-m search window diameter (centered on the centroid of the

Anim Cogn (2016) 19:387–403

393

Fig. 1 Description of the heuristic individual-based model employed to approximate mantled howler monkey movement. An artificial landscape was created by partitioning the group’s home range into 25 m 9 25 m cells. Simulations began in the cell where an observed movement bout was initiated, and proceeded by following random walks (Hypothesis 4), gradients of all food trees (Eq.2; Hypothesis 5), or gradients of estimated fruit availability (Eq. 1; Hypothesis 6)

current cell) repeating the process until they traveled the distance that the observed howler monkey group traveled. Once howler monkeys visited a cell, that cell was assumed to be depleted. As in the previous simulation, angle deviation between steps was not permitted to exceed 60 degrees. Hypothesis 7 Howler monkeys select a direction at random and travel in that direction until arriving at a large food source (the ‘Geometric Model’; Janson 1998). In this simulation, the group travels in a straight line in a random direction and searches for fruit within a specified detection distance of the simulated straight line path (Janson 1998). Once a quantity of fruit as large as that actually selected by the mantled howler monkeys is detected, the group travels directly to that cell. Two detection distances were employed: 25 m and 50 m (i.e., ‘search windows’ of 50 m or 100 m). A search window of 50 m is roughly consistent with Milton’s (2000) previous estimate of search windows for mantled howler monkeys at this site derived from mean tree crown size (*38 m). One hundred simulations at each detection distance were completed for each observed longdistance movement, and the mean value of the 100 simulations for each observed movement was used as the unit for analysis.

Restrictions Several restrictions were placed upon all simulations to make them more consistent with primate movement. Once a cell was visited, it was assumed to be depleted and would not renew for the remainder of the travel bout. As a result, backtracking was minimal in the simulations. Movement beyond home range boundaries was not allowed. If the group reached the edge of its home range, it had to turn around and was permitted to backtrack to a previous cell it had visited if needed for one movement step. In simulations of Hypothesis 7, if the group did not reach a comparable cell prior to exiting the group’s home range, another angle was selected at random to replace the initial angle. Simulations were never permitted to exceed the maximum distance traveled between two food trees by the focal group. This restriction ensures that mean simulation values were not artificially inflated by paths with unrealistic travel distances. Comparison between simulated paths and observed movement patterns To compare model foraging efficiency, the distances a howler monkey group would need to travel along the

123

394

Fig. 2 Histogram of mantled howler monkey group path characteristics when traveling between two food trees [50 m apart: a Distance traveled between every 10-min scan (i.e., on a single step), grouped in 25-m bins (N = 2713); b Angle deviation between consecutive step trajectories, grouped in 10 degree bins (N = 2713). Howler monkey group paths were calculated from the median of all individual monkey locations

simulated movement path until it reached a cell with the same or greater level of estimated fruit availability as the cell actually chosen by the focal group were compared to the distances of observed movement paths with both paired t tests and measurements of mean squared error (MSE). Metrics are compared both for all long-distance foraging movements and for only long-distance movements ending at fruiting trees, as it is difficult to rule out the possibility that mantled howler monkeys were searching for leaves when travel did not end at fruiting trees.

Results Foraging comprised 11.6 ± 4.57 % (mean ± SD) of the group’s daily activity budget, with at least one member of the group foraging in 41 ± 1.4 % of daily scans. The focal group’s annual diet was comprised mostly of leaves (15 %

123

Anim Cogn (2016) 19:387–403

new leaf flush, 39 % immature and mature leaves), with lesser periods of time devoted to fruit (40 %), flowers (5 %), and petioles (1 %). The majority (63.6 %) of howler monkey food trees were reached by short distance movements (\50 m) from other feeding sites (N = 451). A total of 164 feeding sites were reached during the study period with long-distance travel bouts [50 m. Movement step length during long-distance travel bouts (defined as the difference in the median group location between two consecutive 10-min scans) averaged 23.32 m ± 47.6 m (mean ± SD), with a mean angle deviation between steps of 60.03 ± 64.4 degrees (Fig. 2). In 33 % of all scans, angle deviation from the previous scan was less than 10 degrees (median angle deviation: 53.7). The mean straightline distance between the starting point and the ending point of long-distance travel bouts to food trees was 118.8 m (SD = 67.6.4 m, N = 164). In long-distance travel bouts between two feeding sites, the focal group covered between 80 and 956 m (mean ± SD: 196.9 ± 137.4 m). Mean circuity of long-distance paths taken between two feeding sites averaged 1.6 ± 1.1 indicating that observed paths were 60 % longer than a straight line between the starting and ending point (paired t test: t = 6.99, df = 163, p \ 0.001). Overall group mean velocity during long-distance travel bouts between two food trees was 0.15 km/h (SD = 0.34 km/h, range = 0.023.5 km/h; calculated from the median of all monkey locations). Long-distance travel bouts ended in large trees (mean dbh ± SD = 93.5 ± 54.5 cm), belonging to 60 different species. Out of the 164 total long-distance movements, 52 ended at fruiting trees and 21 ended at trees with new leaf flush, with the remaining movements to trees with older leaves and/or flowers. Discrete choice models: Evidence for energy maximization and use of neighborhoods in search strategies? Discrete choice models comparing individual tree characteristics support the hypothesis that howler monkeys minimize distance traveled and maximize food availability at the level of individual trees (Hypothesis 1 supported), but only for crops of mature fruits. Size of new leaf flush crop was not significantly related to likelihood of selection, though the group exhibited a trend to select trees with greater quantities of new leaf flush (Table 1a). The presence of fruit in general was also not a significant predictor of tree selection (z = 0.01, p [ 0.90), but when fruiting stages (immature, mature, over-ripe) were considered separately, presence of mature fruits was a highly significant explanatory variable (Table 1b). After accounting for resource yield, monkeys were more likely to target their long-distance movements to closer trees containing larger

Anim Cogn (2016) 19:387–403

crops of mature fruits that had been visited in the previous 3 days (Table 1c). Tree size (dbh) was not a significant predictor of tree selection, given that methods preselected for only large trees greater than 25 cm dbh (Table 1b). Support was also found for the hypothesis that howler monkeys target neighborhoods in their home range with the highest estimates of total fruit availability at a given point in time (Hypothesis 2 supported). Total fruit availability in an area, estimated probabilistically according to Eq. (1), was a highly significant predictor of foraging site selection after long-distance travel bouts (p \ 0.001; Table 2a; Fig. 3; Figure S2). The hypothesis that howler monkeys target neighborhoods in their home range with high concentrations of food trees, without knowledge of phenological patterns of different tree species, was not supported

395

(Hypothesis 3 not supported). Relationships between site selection and the concentration of preferred tree species did not approach Bonferonni-corrected levels (Table 2b). Decision tree models: Can simple heuristics approximate the efficiency of observed howler monkey travel patterns? Observed travel was significantly more efficient than simulated travel strategies in all but one case (the geometric model with a 50-m detection radius). Simulations following a correlated random walk required up to 80 % longer distances to reach the same quantity of food as obtained by the howler monkeys on BCI, while step models required up to 75 % longer distances (Hypothesis 4 not supported;

Table 1 Discrete choice models testing whether mantled howler monkeys were more likely to select trees with greater relative fruit and leaf crop sizes as the targets of their long-distance travel bouts b

z score

p value

(a) Movements ending at trees with new leaf flush Distance (m) Estimated crop of new leaf flush (m2) Visit previous 3 days (0 or 1)

-0.024

-2.75

0.003 2.17

1.56 1.92

0.006* 0.118 0.051

Likelihood ratio: 108.58 (p \ 0.001, df = 3) McFadden’s R2 = 0.565 (b) Movements ending at trees with fruits Distance (m)

\0.001*

-0.02

-3.89

dbh (cm)

0.01

0.42

Immature fruits (0 or 1)

0.28

0.35

0.730

Mature fruits (0 or 1)

3.26

3.98

\0.001*

-1.62

-1.19

2.21

2.71

Over-ripe fruits (0 or 1) Visit previous 3 days (0 or 1)

0.671

0.235 0.007*

Likelihood ratio: 108.76 (p \ 0.001, df = 6) McFadden’s R2 = 0.71 AIC: 57.2 (c) Movements ending at trees with fruits Distance Estimated crop of fruits (# fruits, normalized by maximum value for each species on the same day) Visit previous 3 days (0 or 1)

-0.018 2.35 1.88

-4.77

\0.001*

3.60 3.10

\0.001* 0.002*

Likelihood ratio: 96.08 (p \ 0.001, df = 3) McFadden’s R2 = 0.62 AIC: 63.8 Trees chosen by the howler monkeys were compared to a random selection of trees [25 cm dbh belonging to the same species. Discrete choice models took the form of a conditional logit model stratified by long-distance movement. Model 1a evaluates the impact of the size of new leaf flush crop on foraging decisions, using only those long-distance movements that ended at trees with new leaf flush (N = 152 trees stratified according to 10 long-distance movements). Model 1b identifies the fruit stage most associated with foraging movements ending at fruiting trees (N = 431 trees stratified according to 30 long-distance movements). Model 1c evaluates the impact of the size of fruit crop on foraging decisions, using only those long-distance movements that ended at trees with fruits (N = 431 trees stratified according to 30 long-distance movements). All models include a binary variable indicating whether the tree had been visited in the past 3 days * Significant to Bonferonni-corrected value p = 0.01

123

396

Anim Cogn (2016) 19:387–403

Table 2 Discrete choice models examining the likelihood of site selection given the food availability present b

z score

p value

-0.02

-12.36

\0.001*

1.25

3.09

0.002*

-0.012

-12.50

\0.001*

0.001

2.40

0.002*

(a) Relationship between feeding site selection and the distribution of trees likely to be fruiting Distance (m) Weighted dbh-sum of fruiting tree species (normalized by the maximum weekly value) Likelihood ratio: 245.36 (p \ 0.001, df = 2) McFadden’s R2 = 0.13 AIC: 1630.3 (b) Relationship between feeding site selection and distribution of all food trees Distance (m) dbh-sum of top 60 feeding tree species Likelihood ratio: 242.18 (p \ 0.001, df = 2) McFadden’s R2: 0.12 AIC: 1633.5 Discrete choice models took the form of a conditional logit model stratified by long-distance movement and were conducted on a grid with a 25 m 9 25 m cell resolution. Model 2a tests whether selected sites have significantly higher quantities of trees likely to be fruiting at the time of the movement, calculated per Eq. (1). Model 2b tests whether selected sites have higher quantities of food trees, calculated per Eq. (2). N = 50,778 stratified according to 164 movement steps * Significant to Bonferonni-corrected value p = 0.01

625900 626000 626100 626200 626300 626400 626500 626600 626700 626800 1012100

1012100

1012000

1012000

1011900

1011800

Observed Path Start (11/19/05 07:10) End (11/19/05 08:00)

Normalized Fruit Availaibility

1011900

1011800

0-0.25 1011700

0.26-0.50

1011700

0.51-0.75 1011600

0.76-1.0

1011600

625900 626000 626100 626200 626300 626400 626500 626600 626700 626800

Fig. 3 Illustrative long-distance foraging movement overlaid on its weekly estimate of fruit availability. Fruit availability in each cell was estimated according to Eq. (1) and then normalized by the maximum cell value for that week. A value of 50 % indicates that this cell has

approximately 1/2 of the estimated fruit available in the most productive cell during that week. Coordinates (easting and northing) are provided in the Universal Transverse Mercator (UTM) system, with units in meters

Table 3). Travel along arboreal pathways through high concentrations of food trees without a predetermined food tree in mind required 36–65 % the distance monkeys were observed to travel, a significant difference (Hypotheses 5 not supported). Models using this decision method also consistently had the highest MSEs, indicating the poorest fit to observed data (Table 3). Searching along a sensory gradient with a detection distance of 25 m (75-m search window) also received little support (Hypothesis 6 not supported). In order to reach the same levels of fruit as

observed monkeys, virtual monkeys following a visual search pattern had to travel 37 % farther than observed travel paths (a significant difference; Table 3a). Models in which virtual monkeys selected a direction at random and traveled in a straight line until a comparable food source was reached (Janson’s ‘Geometric Model’) had the best fits (lowest MSEs; Table 3). However, the relative efficiency of the model depended on the assumed detection width (Hypothesis 7 mixed support). When a detection width of 25 m from the current cell was modeled (i.e., a search

123

Anim Cogn (2016) 19:387–403

397

Table 3 Distances required by simulated foraging strategies to reach cells with equal or higher levels of fruit availability as the cells chosen by the mantled howler monkey group Strategy

Distance traveled (m)

SE

SD

Paired t test

p value

MSE

(a) Movements to all feeding sites [50 m in length (N = 164) Correlated random walk

299.84

18.28

234.15

5.26

\0.001*

73.0

Step model

267.53

15.69

201.01

4.05

\0.001*

54.4

Geometric model (25 ma)

259.32

5.64

72.23

5.50

\0.001*

24.8

Geometric model (50 ma)

124.74

12.79

163.82

-4.27

Gradient model (all feeding trees)

325.17

28.35

363.08

4.44

\0.001*

152.3

Gradient model (estimated fruits) Observed movements

268.59 196.93

11.25 10.73

144.09 137.37

4.16 –

\0.001* –

53.5 –

\0.001***

40.9

(b) Movements to fruiting sites [ 50 m in length (N = 52) Correlated random walk

377.62

34.58

249.43

4.87

\0.001*

96.0

Step model

369.83

38.74

279.35

4.27

\0.001*

87.6

Geometric model (25 ma)

264.12

10.47

75.52

2.66

\0.005*

23.2

Geometric model (50 ma)

198.14

33.53

241.62

-0.40

Gradient model (all feeding trees)

343.12

51.91

374.32

Gradient model (estimated fruits)

291.34

19.11

Observed movements

210.91

21.26

[0.688

51.5

2.68

\0.008*

141.2

137.81

2.83

\0.006*

47.5

153.35

–

–

–

Cell fruit availability was estimated according to Eq. (1) using a 25 m 9 25 m cell resolution. Mean distances required by simulations were compared to those traveled by the mantled howler monkeys during long-distance foraging movements using paired t tests (df = 163). p values are two-tailed with significance comparisons made to a Bonferonni-corrected p value (p = 0.008) * Significantly greater than observed *** Significantly less than observed a

Modeled detection distance

window of 50 m), simulated monkeys had to travel 25–32 % farther than observed distances (Table 3). In contrast, models using a 50-m detection distance (i.e., a search window of 100 m) reached similar levels of estimated fruits in 5–36 % shorter distances than observed. This difference was significant when considering travel bouts ending at all trees (including those that did not contain fruits), but was not significant when considering only those travel bouts ending at fruiting trees. However, MSE values were consistently greater for 50-m detection distances than for 25-m distances, suggesting that while the overall mean values when using 50-m detection distances better approximated observed values, model variance as measured by residuals was greater.

Discussion This study furthers our knowledge of primate cognition by comparing the observed foraging behaviors of mantled howler monkeys to a series of hypothesized foraging strategies that differ both in the type of information used (e.g., spatial, temporal) and in the underlying decisionmaking process. Mantled howler monkeys were more likely to end their long-distance travel bouts at the

individual trees with the largest mature fruit crop sizes, and to target neighborhoods within their home range with the largest fruit yields at a specific point in time. Observed travel patterns were consistently more efficient than simulated foraging strategies that did not rely on memory of the spatial distribution and phenological patterns of individual fruiting trees. These results provide quantitative evidence that a highly folivorous primate likely integrates knowledge of spatial and temporal resource distributions in a foraging strategy that maximizes resource gain while minimizing distance traveled, and demonstrates that primate foraging patterns cannot be explained well by five commonly suggested search heuristics. Similar to many frugivorous primates (Janmaat et al. 2004, 2006; Cunningham and Janson 2007a; Asensio et al. 2011; Suarez 2014), the mantled howler monkeys studied here likely encode information regarding both the location and fruiting states of individual food trees. Observed preferences for larger fruit crops were not simply a reflection of preferences for larger trees, as dbh did not differ significantly between selected and non-selected trees. Furthermore, howler monkeys in this study directed their long-distance movements only to those trees bearing fruits in the most palatable state, while showing no preference for trees bearing unripe or over-ripe fruits, and were more

123

398

likely to revisit trees that they had visited in the previous 3 days. In contrast, the howler monkeys in this study did not select trees with significantly more new leaf flush than other trees present within their home range. This may indicate that the patchier fruit distributions influence longdistance travel decisions as a whole more than leaf distributions. However, since nutrient content and quality vary so greatly between individual leaves, even among those on the same tree (Ganzhorn 1992; Chapman et al. 2003), it is difficult to make strong conclusions from this finding without additional tests of the spatial distribution of leaf quality. Howler monkey travel patterns were also consistent with neighborhood-based selection methods (Wiener and Mallot 2003) but only with those that incorporated spatiotemporal knowledge of resource distributions. Howler monkey movements were not consistent with a strategy that targeted only core areas of their home range containing high concentrations of food trees important across an annual cycle. Instead, they appeared to target only those areas with high concentrations of food trees belonging to species that were fruiting in that specific week. The presence of neighborhood effects is consistent with previous studies on foraging patterns by frugivorous birds (Saracco et al. 2005; Smith and McWilliams 2014), and indicates that further investigation is warranted to determine the scale at which primate foraging decisions are made. None of the simple heuristics tested could approximate observed movement patterns as measured by both mean distance traveled and MSE of model residuals, suggesting howler monkeys incorporate memory of both the spatial and temporal distributions of resources into their travel decisions. Both correlated random walk and step models consistently overestimated distances required to obtain observed fruit amounts. In fact, random models required almost twice the distance traveled to reach the same quantities of fruit as those obtained by the BCI howlers. This observation that howler monkey travel patterns were significantly more efficient than simulations of random patterns is consistent with earlier studies identifying nonrandom properties in mantled howler monkey travel (Milton 2000; Garber and Jelinek 2006). Traveling along arboreal pathways without a predetermined destination in mind was the least efficient heuristic tested. As this is the first attempt to simulate primate travel along arboreal pathway networks using independent continuous information on both spatial and temporal resource distributions, these results have particular relevance for those interested in primate cognition. The single previous quantitative evaluation of the efficiency of route-based travel among primates indicated that spider monkeys (one of the most frugivorous New World primates) reached food trees in shorter distances than if following an arboreal

123

Anim Cogn (2016) 19:387–403

pathway network without a predetermined destination, suggesting use of spatial memory (Suarez et al. 2014). The study reported here builds upon this important work and advances our knowledge of network travel among primates in three ways. First, it demonstrates that like more frugivorous primates, the most folivorous New World primate also uses arboreal pathway networks to reach predetermined destinations rather than relying upon the concentrations of resources along the paths themselves to meet nutrient requirements. Second, this study adds a temporal memory component by indicating that howler monkeys not only use habitual pathways to target food trees, but that they also use these paths to reach the most productive food trees in their habitat at a given time (i.e., measured within 1 day for individual tree-based analyses and within 1 week for cell-based analyses). Third, differences between this study and Suarez et al. (2014) suggest that further research is needed to identify the conditions and scales at which route-based travel offers selective advantages. Unlike Suarez et al. (2014), this study indicates that random walks were more efficient than undirected use of habitual pathways. This may be an artifact of both the distribution of howler monkey foods compared to spider monkey foods and the properties of their habitual pathway networks. Suarez et al. (2014) found that both spider monkey travel paths and used fruit trees were heavily constrained to ridgetops, with large gaps in between routes (mean density of food trees from which spider monkeys foraged across the study area: 5 trees per ha). Alternatively, mantled howler monkey travel paths are not heavily constrained by topography as there is very little altitudinal change within a single howler monkey group’s home range at this site (Hopkins 2011). Furthermore, the howler monkeys studied here consume many more species of trees ([180) that are distributed across the study area with few large gaps (mean density of mapped potential large food trees [25 dbh = 115.16 trees per ha; mean density of trees from which howler monkeys foraged across the study period = 40.4 trees/ha; Hopkins 2008). These differences indicate that different null movement models may be appropriate for primates with different movement and foraging ecologies. In addition, results support the idea that different foraging strategies and associated spatial memory forms yield unequal advantages across primates with different diets. Given the predominance of route-based travel among animals, further research examining the relative efficiencies of route-based travel can help determine which movement strategies give the greatest benefits under which conditions and for which species. While focusing on species falling into broad dietary categories (i.e., folivore vs. frugivore vs. insectivore; herbivore vs. carnivore; food storing vs. non-food storing) can reveal general trends, this

Anim Cogn (2016) 19:387–403

categorization scheme leaves out important variation when evaluating species for spatial cognition (Platt et al. 1996; Long and Platt 2005). For example, while howler monkeys are the most folivorous New World primate, they also eat considerable amounts of fruit. In addition, many Old World folivores have more extensive morphological and physiological adaptations for leaf eating and may have experienced different selective pressures for spatial cognition (Garber 1987; Milton 1998; Wasserman and Chapman 2003). Models intended to approximate following a sensory gradient, in which simulated monkeys were programmed to repeatedly look to all adjacent cells (a 75-m search window) and travel directly to the most productive cell, performed better than both random models and models simulating habitual pathway use without a predetermined destination. However, sensory gradient models were still significantly less efficient than observed travel. Since the visual detection distance of howler monkeys at this site was not measured during the study period, I cannot exclude the possibility that howler groups are able to search a window greater than 75 m. However, primate detection distances have been estimated with a variety of methods, including using human visual detection distances as a proxy, measuring ecological characteristics such as mean crown size of canopy trees, and testing detection distances experimentally (Milton 1980; Janson and DiBitetti 1997; Milton 2000; Janmaat et al. 2006; Normand and Boesch 2009a; Noser and Byrne 2010). In these studies, estimated search windows for primate groups are less than or consistent with a 75-m window (range 10–82 m). One would expect the search window of a mantled howler monkey group to be towards the lower end of the estimated distribution of search windows, given the propensity of mantled howler monkeys to travel through the canopy in single file (Milton 1980; Hopkins 2011). Geometric models in which the howlers traveled in a straight line in a randomly selected direction performed the best out of all heuristics tested. Geometric models consistently had the lowest MSEs, as well as the lowest differences between mean distance values. However, support for these models was inconsistent. Similar to more frugivorous primates (Janson 1998; Asensio et al. 2011), traveling in a straight line in a randomly selected direction yielded similar levels of fruit in the same distances as observed only when the howler monkeys’ detection abilities reached seemingly unrealistic levels (i.e., a 100-m search window), and when considering only those observed travel bouts that concluded at fruiting trees. Furthermore, the MSEs produced by this model are comparatively high. In contrast, the geometric model with a seemingly more biologically appropriate search window of 50 m could not approximate the efficiency of observed searches as measured by mean

399

distance traveled, but had the lowest MSEs out of any model. Unfortunately, statistical model comparison using metrics such as AIC is inappropriate for nonparametric models relying upon rule-based decision-making processes such as those tested here, making it difficult to determine whether differences between model fits are significant. While these comparisons of models to observed travel paths provide important information regarding the strategies and types of information used by mantled howler monkeys in their foraging decisions, they must be interpreted within the framework of the current study. Although hypotheses were considered here as separate alternatives, foraging strategies are not necessarily mutually exclusive. Animals may employ different foraging strategies under different environmental conditions, at different spatial scales, and when searching for different food items (Senft et al. 1987; Wallace et al. 1995; Fortin 2003; Garber and Dolins 2010; Cunningham and Janson 2013; Amato and Garber 2014). This study cannot tease out the conditions under which each foraging strategy might be most likely. In addition, by making a number of simplifying assumptions, models cannot address the full complexity involved in primate travel decisions. For example, models rely on general trends observed for howler monkey travel and foraging behaviors (e.g., limits to backtracking and circuity of allowed paths, travel in single file, depletion of a resource after a same-day visit to the tree) and assume howler monkey long-distance movements target only one food item at a given time, even though howler monkeys will consume a variety of items such as leaves, fruits, petioles, and flowers within the same day (Milton 1980). Models also treat one study group as the unit of analysis. Thus, they cannot account for the sociality inherent in group foraging decisions, in which multiple individuals must reconcile different needs as well as different sets of knowledge regarding the spatial distribution of resources when deciding on travel paths (Boinski 2000; Garber 2000; di Bitetti and Janson 2001; BiccaMarques and Garber 2005; Couzin 2005, 2009; Garber et al. 2009). Furthermore, while the results presented here reflect use of both temporal and spatial knowledge of resource distributions by the most folivorous New World primate and confirm that there may be important differences between howler monkey movement ecology and that of more frugivorous primates, further research is required to make conclusions regarding the form of howler monkey memory (i.e., network or vector cognitive map). Data support the hypothesis that mantled howler monkeys remember the characteristics of individual productive trees or small groups of trees, not just a series of routes. Nonetheless, it is possible that in navigating to these trees, howler monkeys rely on their knowledge of arboreal networks and associated landmarks instead of knowledge of the angles and

123

400

distances between starting locations and target trees. Mantled howler monkeys in this study traveled along paths that were more circuitous than the paths of capuchin monkeys, chimpanzees, and even mantled howler monkeys inhabiting much smaller home ranges at different study sites (Janson 1998; Garber and Jelinek 2006; Normand and Boesch 2009b), which would seem consistent with a network map representation. Yet, nonlinearity of travel paths is not conclusive evidence of use of a network map, especially given that howler monkeys at this site have been shown to deviate from straight-line paths for a number of reasons including adherence to paths with high canopy connectivity, potential resource monitoring, and avoidance of neighboring groups (Milton 1980; Hopkins 2011, 2013). In conclusion, studies of the foraging strategies of animals such as the one presented here, which both increase our current sample size and broaden our taxonomic perspective, are critical to our understanding of animal cognition. In observational studies, field biologists attempt to infer cognitive processes from behavior, while making certain assumptions regarding the most ‘optimal’ foraging strategy (Janson 2007; Janson and Byrne 2007; Normand and Boesch 2009b). However, the most ‘optimal’ foraging strategy may differ between species, depending upon variance in morphology, physiology, social structure, and ecology, as well as between shifting contexts such as seasonal changes to resource distributions (Garber 1987; Cunningham and Janson 2013; Fagan et al. 2013). Thus, when attempting to use the natural foraging behaviors of animals to make conclusions regarding spatial memory form or capacity, we must first determine what these animals are trying to accomplish. This study provides a necessary step in this direction. Specifically, results indicate that the long-distance travel bouts of mantled howler monkeys target both the most productive trees and the most productive neighborhoods within their home range while minimizing distance traveled. This relationship cannot be approximated by search heuristics that do not rely on knowledge of the location and phenological patterns of important food sources. Thus, while the specific decisionmaking process remains unknown, the most likely explanation for these results is that howler monkeys integrate detailed memory of spatial and temporal resource distributions into a highly efficient and nonrandom foraging strategy that works towards maximizing resources gained while minimizing distances traveled. Acknowledgments I would like to thank the following individuals for helpful comments and/or assistance with research: Katharine Milton, E. Lacey, J. Brashares, F. Rodriquez, V. Jaramillo, A. Wendt, M. Senf, P. Thompson, K. Ellis, B. Gaard, L. Rich, R. Kays, M. Wikelski, D. Obando, J. Wright, S. Hubbell, R. Condit, and the staff of STRI and the Center for Tropical Forest Science. I would also like to thank three anonymous reviewers for their important and helpful

123

Anim Cogn (2016) 19:387–403 comments. Funding was provided to M. Hopkins by the following organizations: the National Science Foundation (#0622611), The Wenner-Gren Foundation, the American Association of University Women, The Leakey Foundation, the Smithsonian Tropical Research Institute, and the University of California at Berkeley. CTFS provided the 50-ha plot data, which were made possible by NSF Grants to S. Hubbell, support from CTFS, STRI, the MacArthur Foundation, the Mellon Foundation, the Celera Foundation, and numerous private individuals, and through the hard work of over 100 people. Compliance with ethical standards Conflict of interest interest.

The author declares there were no conflicts of

Ethical standard This study received approval from the Institutional Animal Care and Use Committee at the University of California, Berkeley, and complies with all national and international regulations regarding the care and use of animals in research, including the guidelines for research with animals as outlined by the Association for the Study of Animal Behaviour.

References Altmann J (1974) Baboons, space, time, and energy. Am Zool 14:221–248 Amato KR, Garber PA (2014) Nutrition and foraging strategies of the black howler monkey (Alouatta pigra) in Palenque National Park, Mexico. Am J Primatol 76:774–787 Asensio N, Brockelman WY, Malaivijitnond S, Reichard UH (2011) Gibbon travel paths are goal oriented. Anim Cogn 14(3):395–405 Barton KA, Phillips BL, Morales JM, Travis JM (2009) The evolution of an ‘intelligent’ dispersal strategy: biased, correlated random walks in patchy landscapes. Oikos 118(2):309–319 Bartumeus F, Levin SA (2008) Fractal reorientation clocks: linking animal behavior to statistical patterns of search. Proc Natl Acad Sci USA 105:19072–19077 Benhamou S (2007) How many animals really do the Le´vy walk? Ecology 88:1962–1969 BiccaMarques JC, Garber PA (2005) Use of social and ecological information in tamarin foraging decisions. Int J Primatol 26(6):1321–1342 Boinski S (2000) Social manipulation within and between troops mediates primate group movement. In: Garber PA, Boinski S (eds) On the move: how and why animals travel in groups. University of Chicago Press, Chicago, pp 421–469 Boonratana R (2000) Ranging behavior of proboscis monkey (Nasalis larvatus) in the lower Kinabatanjan, Northern Borneo. Int J Primatol 21:497–518 Boyer D, Walsh PD (2010) Modelling the mobility of living organisms in heterogeneous landscapes: does memory improve foraging success? Phil Trans R Soc B 368:5645–5659 Byrne RW (1982) Geographical knowledge and orientation. In: Ellis A (ed) Normality and pathology in cognitive functions. Academic, London, pp 239–264 Carpenter CR (1934) A field study of the behavior and social relations of howling monkeys. Comp Psychol Monogr 10(48):1–168 Chamberlain J, Nelson G, Milton K (1993) Fatty-acid profiles of major food sources of howler monkeys (Alouatta palliata) in the Neotropics. Experientia 49(9):820–824 Chapman CA (1988) Patch use and patch depletion by the spider and howling monkeys of Santa Rosa National Park, Costa Rica. Behaviour 105:99–116

Anim Cogn (2016) 19:387–403 Chapman CA, Chapman LJ, Wrangham R, Hunt KD, Gebo DL, Garnder L (1992) Estimators of fruit abundance of tropical trees. Biotropica 24(4):527–531 Chapman CA, Chapman LJ, Rode KD, Hauck EM, McDowell LR (2003) Variation in the nutritional value of primate foods: among trees, time periods, and areas. Int J Primatol 24:317–333 Chown E, Kaplan S, Kortenkamp D (1995) Prototypes, location and associative networks (plan): towards a unified theory of cognitive mapping. Cogn Sci 19:11–51 Clutton-Brock TH, Harvey PH (1980) Primates, brains, and ecology. J Zool 190:309–323 Codling EA, Plank MJ, Benhamou S (2008) Random walk models in biology. J R Soc Interface 5(25):813–834 Cohen R (1960) A coefficient of agreement for nominal scales. Educ Psychol Meas 20:37–46 Condit R (1998) Tropical forest census plots. Springer and R. G. Landes Company, Berlin and Georgetown Condit R, Lao S, Singh A, Esufali S, Dolins S (2014) Data and database standards for permanent forest plots in a global network. For Ecol Manag 316:21–31 Conradt L, Zollner PA, Roper TJ, Frank K, Thomas CD (2003) Foray search: an effective systematic dispersal strategy in fragmented landscapes. Am Nat 161(6):905–915 Couzin I (2005) Effective leadership and decision-making in animal groups on the move. Nature 433(7025):513–516 Couzin I (2009) Collective cognition in animal groups. Trends Cogn Sci 13(1):36–43 Cunningham E, Janson C (2007a) Integrating information about location and value of resources by white-faced saki monkeys (Pithecia pithecia). Anim Cogn 10(3):293–304 Cunningham E, Janson C (2007b) A socioecological perspective on primate cognition, past and present. Anim Cogn 10(3):273–281 Cunningham E, Janson C (2013) Effect of fruit scarcity on use of spatial memory in a seed predator, white-faced saki (Pithecia pithecia). Int J Primatol 34(4):808–822 Dabbs JMJ, Chang EL, Strong RA, Milun R (1998) Spatial ability, navigation strategy, and geographic knowledge among men and women. Evol Hum Behav 19:89–98 di Bitetti MS, Janson C (2001) Social foraging and the finder’s share in capuchin monkeys (Cebus apella). Anim Behav 62:47–56 di Fiore A, Suarez SA (2007) Route-based travel and shared routes in sympatric spider and woolly monkeys: cognitive and evolutionary implications. Anim Cogn 10(3):317–329 di Fiore A, Link A, Campbell CJ (2010) The Atelines: behavioral and socioecological diversity in a new world monkey radiation. In: Campbell CJ, Fuentes A, Mackinnon KC, Bearder SK, Stumpf RM (eds) Primates in perspective. Oxford University Press, Oxford, pp 155–188 Eisenberg JF, Wilson DE (1978) Relative brain size and feeding strategies in the Chiroptera. Evolution 32(4):740–751 Erhart EM, Overdorff DJ (2008) Spatial memory during foraging in prosimian primates: Propithecus edwardsi and Eulemur fulvus rufus. Folia Primatol 79(4):185–196 Fagan WF, Lewis MA, Auger-Me´the´ M, Avgar T, Benhamou S, Breed G et al (2013) Spatial memory and animal movement. Ecol Lett 16:1316–1329 Fortin D (2003) Foraging ecology of bison at the landscape and plant community levels: the applicability of energy maximization principles. Oecologia 134:219–227 Fronhofer EA, Hovestadt T, Poethke HJ (2013) From random walks to informed movement. Oikos 122:857–866 Gallistel CR, Cramer AE (1996) Computations on metric maps in mammals: getting oriented and choosing a multi-destination route. J Exp Biol 199:211–217 Ganzhorn JU (1992) Leaf chemistry and the biomass of folivorous primates in tropical forests. Oecologia 91(4):540–547

401 Ganzhorn JU (1995) Low-level forest disturbance effects on primary production, leaf chemistry, and lemur populations. Ecology 76(7):2084–2096 Ganzhorn JU, Wright PC (1994) Temporal patterns in primate leaf eating: the possible role of leaf chemistry. Folia Primatol 63(4):203–208 Garber PA (1987) Foraging strategies among living primates. Annu Rev Anthropol 16:339–364 Garber PA (1989) Role of spatial memory in primate foraging patterns—Saguinus mystax and Saguinus fuscicollis. Am J Primatol 19(4):203–216 Garber PA (2000) Evidence for the use of spatial, temporal, and social information by primate foragers. In: Garber PA, Boinski S (eds) On the move: How and why animals travel in groups. University of Chicago Press, Chicago, pp 261–298 Garber PA, Dolins FL (2010) Examining spatial cognitive strategies in small-scale and large-scale space in tamarin monkeys. In: Dolins FL, Mitchell RW (eds) Spatial cognition, spatial perception: mapping the self and space. Cambridge University Press, London, pp 181–195 Garber PA, Hannon B (1993) Modeling monkeys–a comparison of computer-generated and naturally-occurring foraging patterns in 2 species of neotropical primates. Int J Primatol 14(6):827–852 Garber PA, Jelinek PE (2006) Travel patterns and spatial mapping in Nicaraguan mantled howler monkeys (Alouatta palliata). In: Estrada A, Garber PA, Pavelka M, Luecke L (eds) New perspectives in the study of mesoamerican primates: distribution, ecology, behavior and conservation. Developments in primatology: progress and prospects. Kluwer, New York, pp 287–309 Garber PA, BiccaMarques JC, Acevedo-Lopez MADO (2009) Primate cognition: integrating social and ecological information in decision making. In: Garber PA, Estrada A, BiccaMarques JC, Heymann EW, Strier K (eds) South American primates: comparative perspectives in the study of behavior, ecology, and conservation. Springer Press, New York Gentry AH, Terborgh J (1990) Composition and dynamics of the Cocha Cashu ‘mature’ floodplain forest. In: Gentry AH (ed) Four neotropical rainforests. Yale University Press, New Haven, pp 542–564 Glander KE (1978) Howling monkey feeding behavior and plant secondary compounds: a study of strategies. In: Montgomery GG (ed) The ecology of arboreal folivores. Smithsonian Institution Press, Washington DC, pp 561–574 Holland JD, Fahrig L, Cappuccino N (2005) Body size affects the spatial scale of habitat-beetle interactions. Oikos 110:101–108 Hopkins ME (2008) Spatial foraging patterns and ranging behavior of mantled howler monkeys (Alouatta palliata), Barro Colorado Island, Panama. PhD Dissertation, University of California, Berkeley, Berkeley Hopkins ME (2011) Mantled howler (Alouatta palliata) arboreal pathway networks: relative impacts of resource availability and forest structure. Int J Primatol 32:238–258 Hopkins ME (2013) Relative dominance and resource availability mediate mantled howler monkey (Alouatta palliata) spatial responses to neighbors’ loud calls. Int J Primatol 5:1032–1054 Hubbell SP, Foster RB, O’Brien ST, Harms KE, Condit R, Wechsler B, Wright SJ, de Lao SL (1999) Light-gap disturbances, recruitment limitation, and tree diversity in a Neotropical forest. Science 22(283):554–557 Hubbell SP, Condit R, and Foster RB (2005) Barro Colorado forest census plot data. http://ctfs.si/edu/datasets/bci International Species Information System (2015). https://zims.isis. org. Reported 13 Jan 2015 Istomin KV, Dwyer MJ (2009) Finding the way a critical discussion of anthropological theories of human spatial orientation with

123

402 reference to reindeer herders of Northeastern Europe and Western Siberia. Curr Anthropol 50(1):29–49 Jackson HB, Fahrig L (2012) What size is a biologically relevant landscape? Landscape Ecol 27:929–941 Janmaat KRL, Zuberbuhler K, Byrne RW (2004) Do mangabeys remember fruiting states? A study on visiting frequencies of and travel speed towards rain forest trees with different fruiting states. Folia Primatol 75:169 Janmaat KRL, Byrne RW, Zuberbuhler K (2006) Evidence for a spatial memory of fruiting states of rainforest trees in wild mangabeys. Anim Behav 72:797–807 Janson CH (1998) Experimental evidence for spatial memory in foraging wild capuchin monkeys, Cebus apella. Anim Behav 55:1229–1243 Janson CH (2007) Experimental evidence for route integration and strategic planning in wild capuchin monkeys. Anim Cogn 10(3):341–356 Janson CH, Byrne R (2007) What wild primates know about resources: opening up the black box. Anim Cogn 10(3):357–367 Janson CH, DiBitetti MS (1997) Experimental analysis of food detection in capuchin monkeys: effects of distance, travel speed, and resource size. Behav Ecol Sociobiol 41(1):17–24 Leighton M, Leighton DR (1982) The relationship of size of feeding aggregate to size of food patch: howler monkeys (Alouatta palliata) feeding in Trichilia cipo fruit trees on Barro Colorado Island. Biotropica 14(2):81–90 Long A, Platt M (2005) Decision making: the virtue of patience in primates. Curr Biol 15(21):R874–R876 Luehrs M-L, Dammhahn M, Kappeler PM, Fichtel C (2009) Spatial memory in the grey mouse lemur (Microcebus murinus). Anim Cogn 12(4):599–609 MacKinnon J (1974) Behavior and ecology of wild orangutans (Pongo pygmaeus). Anim Behav 22:3–74 Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Erickson WP (2002) Resource selection by animals: statistical design and analysis for field studies. Kluwer, London Mathworks (2008) MATLAB and Statistics Toolbox. Natick, Massachusetts, USA McFadden D (1974) Conditional logit anlaysis of qualitative choice behavior. In: Zarembka P (ed) Frontiers of econometrics. Academic Press, New York Menzel EW (1973) Chimpanzee spatial memory organization. Science 182:943–945 Milton K (1980) The foraging strategy of howler monkeys: a study in primate economics. Columbia University Press, New York Milton K (1981) Distribution patterns of tropical plant foods as a stimulus to primate mental development. Am Anthropol 83:534–538 Milton K (1988) Foraging behaviour and the evolution of primate intelligence. In: Byrne RW, Whiten A (eds) Machiavellian intelligence: social expertise and the evolution of intellect in monkeys, apes and humans. Clarendon Press, Oxford, pp 285–305 Milton K (1993) Diet and primate evolution. Sci Am 269(2):86–93 Milton K (1998) Physiological ecology of howlers (Alouatta): energetic and digestive considerations and comparison with the Colobinae. Int J Primatol 19(3):513–548 Milton K (2000) Quo vadis? Tactics of food search and group movement in primates and other animals. In: Boinski S, Garber PA (eds) On the move: How and why animals travel in groups. University of Chicago Press, Chicago, pp 375–418 Milton K, Lozier J, Lacey E (2008) Genetic structure of an isolated population of mantled howler monkeys (Alouatta palliata) on Barro Colorado Island, Panama. Conserv Genet 10(2):347–358 Nagy KA, Milton K (1979a) Aspects of dietary quality, nutrient assimilation and water balance in wild howler monkeys (Alouatta palliata). Oecologia 39(3):249–258

123

Anim Cogn (2016) 19:387–403 Nagy KA, Milton K (1979b) Energy metabolism and food consumption by wild howler monkeys (Alouatta palliata). Ecology 60(3):475–480 Normand E, Boesch C (2009a) Forest chimpanzees (Pan troglodytes verus) remember the location of numerous fruit trees. Anim Cogn 12(6):797–807 Normand E, Boesch C (2009b) Sophisticated Euclidean maps in forest chimpanzees. Anim Behav 77(5):1195–1201 Noser R, Byrne RW (2007) Mental maps in chacma baboons (Papio ursinus): using inter-group encounters as a natural experiment. Anim Cogn 10:331–340 Noser R, Byrne RW (2010) How do wild baboons (Papio ursinus) plan their routes? Travel among multiple high-quality food sources with inter-group competition. Anim Cogn 13(1):145–155 Ohashi K, Thomson JD, D’Souza D (2007) Trapline foraging by bumble bees: IV. Optimization of route geometry in the absence of competition. Behav Ecol 18:1–11 Petit O, Bon R (2010) Decision-making processes: the case of collective movements. Behav Process 84:635–647 Platt ML, Brannon EM, Briese TL, French JA (1996) Differences in feeding ecology predict differences in performance between golden lion tamarins (Leontopithecus rosalia) and Wied’s marmosets (Callithrix kuhli) on spatial and visual memory tasks. Anim Learn Behav 24(4):384–393 Porter LM, Garber PA (2013) Foraging and spatial memory in wild weddell’s saddleback tamarins (Saguinus fuscicollis weddelli) when moving between distant and out-of-sight goals. Int J Primatol 34:30–48 Poucet B (1993) Spatial cognitive maps in animals: new hypotheses on their structure and neural mechanisms. Psychol Rev 100:163–182 Presotto A, Izar P (2010) Spatial reference of black capuchin monkeys in Brazilian Atlantic Forest: egocentric or allocentric? Anim Behav 80:125–132 Pyke GH (1978) Optimal foraging: movement patterns of bumblebees between inflorescences. Theor Popul Biol 13:72–97 Pyke GH (1984) Optimal foraging theory: a critical review. Ann Rev Ecol Syst 15:523–575 Rodrigo T (2002) Navigational strategies and models. Psicologica 23:3–32 Sandstrom N, Kaufman J, Huettel S (1998) Males and females use different distal cues in a virtual environment navigation task. Cogn Brain Res 6:351–360 Saracco JF, Collazo JA, Groom MJ, Carlo TA (2005) Crop size and fruit neighborhood effects on bird visitation to fruiting Schefflera morototoni trees in Puerto Rico1. Biotropica 37(1):81–87 Schreier A, Grove M (2010) Ranging patterns of hamadryas baboons: random walk analyses. Anim Behav 80:75–87 Senft RL, Coughenour MB, Bailey DW, Rittenhouse LR, Sala OE (1987) Large herbivore foraging and ecological hierarchies. Bioscience 37:798–799 Shettleworth SJ (1998) Cognition, evolution, and behavior. Oxford University Press, Oxford Sigg H, Stolba A (1981) Home range and daily March in a hamadryas baboon troop. Folia Primatol 36(1–2):40–75 Smith AD, McWilliams SR (2014) Fruit removal rate depends on neighborhood fruit density, frugivore abundance, and spatial context. Oecologia 174(3):931–942 Smulders TV, Gould KL, Leaver LA (2010) Using ecology to guide the study of cognitive and neural mechanisms of different aspects of spatial memory in food hoarding animals. Philos Trans R Soc B 365:883–900 StataCorp (2007) Stata Statistical Software: Release 10. TX: StataCorp LP, College Station

Anim Cogn (2016) 19:387–403 Stephens DW, Krebs JR (1987) Foraging theory. Princeton University Press, Princeton Suarez S (2014) Ecological factors predictive of wild spider monkey (Ateles belzebuth) foraging decisions in Yasuni, Ecuador. Am J Primatol 76:1185–1195 Suarez SA, Karro J, Kiper J, Farler D, McElroy B, Rogers BC, Young T (2014) A comparison of computer-generated and naturally occurring foraging patterns in route-network-constrained spider monkeys. Am J Primatol 76(5):460–471 Tolman EC (1948) Cognitive maps in rats and men. Psychol Rev 55:189–208 Wallace RB (2008) The influence of feeding patch size and relative fruit density on the foraging behavior of the Black spider monkey Ateles chamek. Biotropica 40:501–506 Wallace LL, Turner MG, Romme WH, O’Neill RV, Wu Y (1995) Scale of heterogeneity of forage production and winter foraging by elk and bison. Landsc Ecol 10:75–83 Wallraff HG (2001) Navigation by homing pigeons: updated perspective. Ethol Ecol Evol 13:1–48

403 Wasserman MD, Chapman CA (2003) Determinants of colobine monkey abundance: the importance of food energy, protein and fibre content. J Anim Ecol 72(4):650–659 White FJ, Wrangham RW (1988) Feeding competition and patch size in the chimpanzee species Pan paniscus and Pan troglodytes. Behaviour 105:148–164 Wiener JM, Mallot HA (2003) ‘Fine-to-Coarse’ route planning and navigation in regionalized environments. Spat Cogn Comput 3:331–358 Wiener JM, Schnee A, Mallot HA (2004) Use and interaction of navigation strategies in regionalized environments. J Environ Psychol 24:475–493 Wiener JM, Ehbauer NN, Mallot HA (2009) Planning paths to multiple targets: memory involvement and planning heuristics in spatial problem solving. Psychol Res 73:644–658 Wilton RN (1979) Knowledge of spatial relations: the specification of the information used in making inferences. Q J Exp Psychol 31:133–146

123