Coming down from the trees: Is terrestrial activity in Bornean orangutans natural or disturbance driven?

The orangutan is the world's largest arboreal mammal, and images of the red ape moving through the tropical forest canopy symbolise its typical arboreal behaviour. Records of terrestrial behaviour are scarce and often associated with habitat disturbance. We conducted a large-scale species-level analysis of ground-based camera-trapping data to evaluate the extent to which Bornean orangutans Pongo pygmaeus come down from the trees to travel terrestrially, and whether they are indeed forced to the ground primarily by anthropogenic forest disturbances. Although the degree of forest disturbance and canopy gap size influenced terrestriality, orangutans were recorded on the ground as frequently in heavily degraded habitats as in primary forests. Furthermore, all age-sex classes were recorded on the ground (flanged males more often). This suggests that terrestrial locomotion is part of the Bornean orangutan's natural behavioural repertoire to a much greater extent than previously thought, and is only modified by habitat disturbance. The capacity of orangutans to come down from the trees may increase their ability to cope with at least smaller-scale forest fragmentation, and to cross moderately open spaces in mosaic landscapes, although the extent of this versatility remains to be investigated.

The orangutan is the world's largest arboreal mammal, and images of the red ape moving through the tropical forest canopy symbolise its typical arboreal behaviour. Records of terrestrial behaviour are scarce and often associated with habitat disturbance. We conducted a large-scale species-level analysis of ground-based camera-trapping data to evaluate the extent to which Bornean orangutans Pongo pygmaeus come down from the trees to travel terrestrially, and whether they are indeed forced to the ground primarily by anthropogenic forest disturbances. Although the degree of forest disturbance and canopy gap size influenced terrestriality, orangutans were recorded on the ground as frequently in heavily degraded habitats as in primary forests. Furthermore, all age-sex classes were recorded on the ground (flanged males more often). This suggests that terrestrial locomotion is part of the Bornean orangutan's natural behavioural repertoire to a much greater extent than previously thought, and is only modified by habitat disturbance. The capacity of orangutans to come down from the trees may increase their ability to cope with at least smaller-scale forest fragmentation, and to cross moderately open spaces in mosaic landscapes, although the extent of this versatility remains to be investigated.
T he Bornean orangutan Pongo pygmaeus is the largest arboreal species in the world and its survival is linked to forest habitat 1,2 . Despite the orangutan's iconic value and millions of dollars spent annually on its conservation 3 , the species is declining throughout its range. In Borneo, more than 70% of orangutans occur in fragmented multiple-use and human-modified forests that have lost many of their original ecological characteristics 4,5 . The consequences of these drastic habitat changes on orangutan survival, behaviour and ecology are only just starting to be documented 3,4,6,7 . Some authors have proposed that forest degradation may force the species to the ground more frequently 8,9 . We can suppose that increased terrestriality would increase predation risk, interactions with and persecution by humans, and exposure to novel pathogens. On the other hand, terrestrial behaviour could also facilitate movement and, therefore, dispersal, especially in degraded or fragmented landscapes as a result of natural or man-made processes. It could also create new opportunities to access different food sources 10 . Ultimately, a better understanding of the drivers of orangutan terrestriality and how this influences dispersal and movement is important for designing effective landscape management strategies for maintaining viable metapopulations of this species in Borneo 4 .
Most studies of orangutan locomotion have been based on direct observations 11,12 . However, orangutans may be reluctant to come to the ground in the presence of human observers, and remote camera traps present an opportunity to overcome this potential bias 8 . We conducted a large-scale species-level analysis of ground-based camera-trapping data to evaluate the extent to which Bornean orangutans travel terrestrially and to investigate possible drivers for this behaviour.

Results
Altogether we collected camera-trapping data from 16 study areas from Sabah (Malaysia) and East and Central Kalimantan (both Indonesia), for which reliable orangutan density estimates were available (Table 1). Pictures were collected between June 2006 and March 2013, and included data from all months of the year. The total dataset encompassed 159,152 trap days at 1,409 independent camera-trap stations.
Orangutans were recorded on the ground in all forest classes, indicating that terrestrial activity occurs regardless of habitat disturbance. Nevertheless, the regression model revealed that forest class, camera-trap placement and orangutan density influenced the photographic frequency and the probability of orangutans coming to the ground (Table 3, Figure 1 A-C). Photographic frequencies were significantly higher beneath large canopy gaps than under closed canopy ( Figure 1A & 1B, Table 3). In 428 out of 641 orangutan records the sex-age class could be reliably determined (see Methods). We observed females alone 27 times, females with clinging  babies or with walking young 63 and 25 times, respectively, unflanged males 48 times, and flanged males 265 times. We recorded flanged males significantly more often than expected based on their proportion in the population (see Methods) (x 2 5 32.050, df 5 1, p , 0.001), suggesting they are more terrestrial than unflanged males, and females. Only 15 orangutan records were obtained during night time (before 0600 h and after 1800 h).

Discussion
Overall, Bornean orangutan terrestrial activity appears more common than previous anecdotal observations suggested 8 , which indicates that the species exhibits a flexible and varied repertoire of locomotion. This is supported evolutionarily: the potential ancestors and fossil relative taxa of Pongo, are thought to have used more ground locomotion than the current Pongo 14,15 . This, in turn, has led to the suggestion that ancestral orangutans may have been able to cover larger distances on the ground 13,16 . It should be noted that the data used in the present analysis come predominantly from the subspecies P. pygmaeus morio in northeastern Borneo and there is the potential for regional differences in the species' response to forest characteristics driven by differences in ecological circumstances 13 . For example, an analysis of raw model residuals by study site (Supplementary Information) showed stronger negative outliers for photographic counts from the Sabangau peat swamp (i.e., subspecies P. p. wurmbi) than from other sites, suggesting that swamp habitat may reduce terrestriality in orangutans. Further, preliminary data indicate that the Sumatran orangutan Pongo abelii is much less terrestrial, possibly because of the presence of tigers Panthera tigris as potential predators, which are absent from Borneo 1 . Our analysis shows that the degree of terrestriality is modulated by forest structure. This suggests that anthropogenic canopy disruptions will increase terrestrial activity in orangutans, but habitat disturbance is not the only driver for this behaviour. Indeed, both photographic frequencies and probabilities of coming to the ground   were lowest in areas logged by reduced impact logging schemes between two and 20 years ago and not, as expected if terrestriality were determined by disturbance, in primary forests. This may be caused by a continuous lower canopy layer that develops after reduced impact logging and is relatively easy to travel through without coming to the ground to cross gaps. In contrast, the irregular canopy structure in primary forests and the large man-made gaps in forests recently logged using conventional logging methods may increase the energetic costs of arboreal locomotion 11 , offering an explanation for similar terrestriality in these forest classes. Although our data also support the perception that terrestriality in orangutans is most prevalent in larger and heavier flanged males 9 , our camera trap data showed that all age-sex classes travel on the ground irrespective of habitat type. The ability to cross forest gaps, especially for males, which are the dispersing sex [17][18][19] , suggests that gene flow can occur even in disturbed and possibly fragmented habitats 10 . In addition to occasional advantages in travelling from place to place, terrestriality could enhance the possibilities for foraging for terrestrial resources, such as succulent shoots, termites or mineral clay 10,20 . This may be particularly important during periods of habitat-wide fruit scarcity. Heightened orangutan terrestriality in human-modified landscapes and the resulting increase in contact with people creates new risks, such as increased susceptibility to hunting and exposure to new diseases, as great apes are vulnerable to many human diseases 21,22 .
Our findings reinforce the importance of incorporating degraded forests recovering from logging disturbance into orangutan conservation strategies 3,4,6 . The capacity of orangutans to come down from the trees may increase their ability to cope with at least smaller scale fragmentation 10 , and to cross moderately open spaces in mosaic landscapes, although the extent of this versatility remains to be investigated. In order to design conservation management strategies that will allow for the species to persist in anthropogenic landscapes, however, practitioners need to be mindful of the potential risks associated with terrestrial activities in orangutans. The ''man of the forest'' cannot be regarded as safely tucked away up in the trees.

Methods
Our analysis is based on a compilation of existing ground-based camera trapping data that had been collected at 16 sites in Borneo, mostly on the northern part of the island (Table 1).
Data base. To avoid inflated counts caused by repeated detections of the same individual, only one record per hour per camera site was included in the data analysis. We excluded all cameras that were placed at salt licks, and data from the orangutan rehabilitation site at Sepilok Forest Reserve, as we assumed that both factors could influence terrestrial behaviour (i.e., a particular resource for the former 20 and habituation to humans for the latter 23 ). To avoid spatial autocorrelation in the data, we only included camera-trap stations that were a minimum of 1 km apart from each other, assuming that orangutans do not move continuously along the ground over distances in the order of a kilometre 10 , so that events to come to the ground are independent at that scale. The independence of these observations would be compromised if orangutans routinely moved along the ground over 1 km. This assumption is backed by the observation that the overall (i.e. arboreal and terrestrial) maximum distance moved by orangutans within a day is in most cases below 1 km 24 . Considering these restrictions all analyses are based on 641 independent orangutan records taken at 1,409 stations during 159,152 trap days (see Table 1).
Analysis. All analyses were performed in R, version 2.15.1 25 .
Terrestrial activity as a function of forest disturbance and density. We hypothesised that regional forest structure and camera-trap site specific canopy structure (also referred to as camera-trap placement), as well as orangutan density, could influence how often orangutans are photographed on the ground. We therefore classified forests into six classes, based on their current and past management history: 1. primary forest (PRIM; not disturbed); 2. very old conventionally logged forest (VOL; last logging more than 20 years ago); 3. old slightly logged forest (OLD-RIL; exploitation using reduced impact logging (RIL) practices between 2-20 years ago); 4. old heavily logged forest (OLD-CL; conventional logging practices 2-20 years ago); 5. recent slightly logged forest (REC-RIL; RIL within the last 2 years); 6. recent heavily logged forest (REC-CL; conventional logging in the last 2 years). We further categorized camera trap placement according to Loken et al. 8 as under closed canopy (0-3 m gaps that orangutans are likely able to cross by tree swaying), small canopy gap (3-5 m gaps that orang-utans might be able to cross by tree swaying), or large canopy gap (.5 m gaps that orang-utans are unlikely to cross by tree swaying). Gap size was determined post hoc based on field notes and photographs. We recognize that there might be some inherent error in classifying gap size post hoc, but the coarse classification of gap size should buffer most of that error. Orangutan densities were obtained from the literature and from unpublished data of the authors for each site (Table 1). We acknowledge that density estimates were obtained with different methods and that some of these methods are controversial; however, the estimates used here represent the best currently available data for these sites and are widely used to assess the status of the species.
To quantify the influence of the above variables on orangutan terrestriality, we compiled the number of orangutan records taken at each camera location and analysed the data using a zero-inflated Poisson (ZIP) regression 26 . A ZIP model allows for overdispersion in counts in the form of excess zeros, which we observed in our data set. It is a Binomial-Poisson mixture that attempts to separate zero counts into structural zeros (sites where orangutans never come to the ground so that we can only observe a zero count) and sampling zeros (sites where orangutans do come to the ground but we happen to not record them there). The binomial component of the ZIP model estimates the probability of coming to the ground at a given camera trap station. The Poisson component of the ZIP model describes the number of records we expect to observe (referred to as the photographic frequency) at a camera trap, conditional on the species coming to the ground at all. Both parameters can be modelled as functions of covariates on the logit and log scale, respectively. We used station-specific survey effort (i.e., camera-trap days) as offset, and orangutan density, forest class, and camera-trap placement (characterised by station-specific canopy gap size) as model covariates.
We built a number of models, differing in the combination of explanatory covariates (forest class, canopy gap size and orangutan density), and used the Akaike Information Criterion (AIC) to select the most parsimonious ('best') model. Because we expect a direct relationship between the sampling effort at a camera-trap station and the number of pictures obtained, all models used the number of camera trap days each station was surveyed as an offset in the model for frequency.
To reduce the total number of candidate models, we first explored different models for the photographic frequency while holding the probability of coming to the ground constant (i.e. no explanatory variables of the probability of coming to the ground). Conditional on the best frequency model we then built candidate models for the probability of coming to the ground ( Table 2). The best model for photographic frequency contained all three covariates. Conditional on this frequency model, the best overall model additionally contained effects of forest class and density on the probability of coming to the ground. The second-best model, which also included camera placement as a covariate on the probability of coming to the ground, only had a delta AIC relative to the best model of 0.66 and was therefore essentially equally supported by the data ( Table 2). Since we were unable to determine a single best model, we employed model averaging, where parameter estimates are obtained as a weighted average over all candidate models 27 .
Age-sex classes on the ground. To investigate whether terrestrial behaviour is exhibited by all demographic classes (i.e. flanged males, unflanged males, females and females with offspring) we would have ideally run separate ZIP regressions for different classes, but identification of these classes was only possible with high certainty in about 50% of all camera trap records. We cannot assume that failure to identify the age-sex class occurs at random -it is much harder to distinguish a small male from a female, or ascertain that a female is with offspring, than it is to unambiguously identify a flanged male. The only analysis we conducted with respect to demographic class was, therefore, a comparison of the observed versus expected number of flanged male pictures using a Chi-square test. We expect a 50550 male-female ratio in the orangutan population 28 ; within the males, on Borneo there are typically 1.651 flanged to unflanged males 28,29 . Thus, if all demographic groups came to the ground relative to their occurrence in the population, we would expect flanged male pictures to comprise 31% of our total sample. We considered all pictures in which the demographic group could not be identified as ''not flanged male'' -this is unlikely to be true since pictures that only show small part of an orangutan could be of a flanged male. However, this procedure guaranteed a conservative approach towards the question of whether flanged males come to the ground more frequently than expected.