Assessing tiger–prey interactions in Sumatran rainforests


  • Editor: Andrew Kitchener

Matthew Linkie, Fauna & Flora International, Jupiter House, Station Road, Cambridge CB1 2JD, UK.


Little is known about interactions between the critically endangered Sumatran tiger Panthera tigris sumatrae and its prey because of the difficulties associated with detecting these species. In this study, we quantify temporal overlap between the Sumatran tiger and five of its presumed prey species from four study areas comprising disturbed lowland to primary submontane forest. Data from 126 camera traps over 8984 camera days were used to estimate species activity patterns and, in turn, their overlap through the coefficient Δ (ranging from 0 to 1, i.e. no overlap to complete overlap). A newly developed statistical technique was applied to determine confidence intervals associated with respective overlap, which is important, as such measures of precision are usually not estimated in these types of study. Strong temporal overlap was found between tiger and muntjac Muntiacus muntjac (Δ=0.80, 95%CI=0.71–0.84) and tiger and sambar Cervus unicolor (Δ=0.81, 0.55–0.85), with the latter illustrating the importance of measuring precision. According to the foraging theory, Sumatran tigers should focus on expending lower levels of energy searching for and then capturing larger bodied prey that present the least risk. Hence, surprisingly, there was little overlap between the crepuscular tiger and the largest-bodied prey species available, the nocturnal tapir Tapirus indicus (0.52, 0.44–0.60), suggesting that it is not a principal prey species. This study provides the first insights into Sumatran tiger–prey temporal interactions. The ability to estimate overlap statistics with measures of precision has obvious and wide benefits for other predator–prey and interspecific competition studies.


The ecology of most rainforest mammals in Indonesia is poorly known, because these species tend to be shy and secretive and therefore difficult to study. This situation is even more acute for medium-large bodied mammal taxa as they typically occur at naturally low population densities. It is therefore ironic that one of Indonesia's largest mammal species, the Sumatran tiger Panthera tigris sumatrae, which occurs at the some of the lowest population densities, due to its trophic status and poaching pressures, is one of the country's best studied.

Scientific research on the Sumatran tiger has been enabled by the introduction of camera-trap equipment and associated statistical sampling techniques (Karanth & Nichols, 1998). The overwhelming majority of these studies have estimated tiger population densities with varying levels of precision (O'Brien, Kinnaird & Wibisono 2003; Wibisono et al., 2009) and their spatial habitat use across landscapes with varying levels of disturbance (Kinnaird et al., 2003; Linkie et al., 2006, 2008a). Yet, basic information about interactions between Sumatran tiger and their prey, prey ecology or even which species represents principal tiger prey is still lacking. According to the foraging theory, prey such as tapir should be preferentially selected given its large body and presumed low risk posed, being predominantly solitary and lacking tusks, horns, antlers or other defence weapons that might injure a marauding tiger.

To investigate interactions between tiger and their prey, studies should focus on their spatial and temporal dimensions. Thus, strong overlap for both of these is expected for the principal prey species as this will increase tiger encounter rates with these species. In the only study to investigate such patterns, O'Brien et al. (2003) found a significant spatial relationship between Sumatran tiger and wild pig (Sus sp., P<0.05) and sambar Cervus unicolor (P<0.10). Camera traps also provide information about daily activity patterns of different species, but to date there has been no comparative study of activity patterns between the Sumatran tiger and its putative prey species.

Recently, Ridout & Linkie (2009) developed a statistical technique for estimating daily activity pattern overlap between sympatric felid species using camera-trap data that includes a measure of precision of the estimated overlap value. In this study, we apply the methodology of Ridout & Linkie (2009) on camera trap data from the Kerinci Seblat (KS) region to investigate the overlap of predator–prey activity patterns, focusing on the Sumatran tiger and its five presumed principal prey of sambar, red muntjac Muntiacus muntjac, wild pig Sus scrofa, pig-tailed macaque Macaca nemestrina and Malayan tapir Tapirus indicus.

Materials and methods

Study area

The 13 300 km2 Kerinci Seblat National Park (KSNP) in west-central Sumatra spans four provinces. The large blocks of contiguous forest in and around KSNP have been designated as a ‘Level 1 Tiger Conservation Landscape’, because they are considered to provide one of the best chances for the long-term survival of tigers (Dinerstein et al., 2007). Nevertheless, tigers living in the KS region are threatened, principally by loss of habitat, and then by poaching of tiger and prey. Deforestation (i.e. complete forest conversion to farmland) has fragmented KSNP into two parts and deforestation rates from 1995 to 2001 were 34.6 km2/year (0.28%/year) inside KSNP and 213.1 km2/year (0.96%/year) across the KS region (Linkie et al., 2008b).

Camera-trap data collection

Camera-trap field data for tiger and their presumed prey were collected from four study areas from 2004 to 2007: (1) Renah Kayu Embun – 26 camera traps set from 3 September to 30 November 2004 in 112 km2 ranging from 947 to 1941 m a.s.l. located in Jambi province; (2) Sipurak – 28 camera traps set from 3 January to 29 March 2005 in 88 km2 ranging from 694 to 1254 m a.s.l. located in Jambi province; (3) Bungo – 32 camera traps set from 16 April to 23 November 2006 in 237 km2 ranging from 363 to 1745 m a.s.l. located in Jambi province; (4) Ipuh – 40 camera traps set from 24 August 2006 to 2 May 2007 in 569 km2 ranging from 194 to 1064 m a.s.l. located in Bengkulu province. These four study areas were selected because of their presumed importance for tiger, for which the KSNP management authority had requested detailed information.

A combination of TrailMaster and Photoscout passive infrared camera traps, activated by a heat-motion sensor, was used. Cameras were set along ridge trails and medium-large bodied animal trails, as identified through the presence of tiger sign. Cameras were checked every 2 weeks and their films replaced.

Statistical analysis

To investigate the temporal tiger–prey activity patterns, photographs that were recorded within 30 min of a previous photograph of the same species and at the same camera placement were not used, because they were not considered to be independent. The remaining data were regarded as a random sample from the underlying distribution that describes the probability of a photograph being taken within any particular interval of the day. The probability density function of this distribution was then referred as the activity pattern, which presupposes that the animal is equally likely to be photographed at all times when it is active (Ridout & Linkie, 2009).

A two-step procedure for quantifying the extent of overlap between two activity patterns, based on a sample from each species, was performed. For the first step, each activity pattern was estimated separately, either non-parametrically, using kernel density estimation or by fitting a distribution from the flexible class of non-negative trigonometric sum distributions (Fernández-Durán, 2004). The kernel density estimates used a bandwidth parameter, which was selected following the procedure developed by Taylor (2008). The trigonometric sum distributions were determined from the number of components based on minimizing the Akaike information criterion (Fernández-Durán, 2004).

For the second step, a measure of overlap between the two estimated distributions was calculated. Ridout & Linkie (2009) reviewed several alternative measures of overlap between two probability distributions, favouring the coefficient of overlapping, Δ (Weitzman, 1970), which ranges from 0 (no overlap) to 1 (complete overlap). This is defined as the area under the curve that is formed by taking the minimum of the two density functions at each time point. One useful interpretation of the coefficient of overlapping is that for any time period during the day, the probability that a randomly selected camera trap photograph will have occurred during that period differs between the two distributions by <1–Δ.

Ridout & Linkie (2009) discussed three alternative ways of estimating Δ, given estimates of the two probability density functions. These were labelled inline image, inline image and inline image for consistency with an earlier work. Here, we use their estimators inline image and inline image, which were recommended for ‘small’ and ‘large’ sample sizes, respectively. Confidence intervals were obtained as percentile intervals from 500 bootstrap samples.


From 8984 trap nights, a high number of records were obtained for tiger, pig-tailed macaque, muntjac and tapir across all study areas combined (Table 1). Comparing between the study areas, the fewest records were obtained from Renah Kayu Embun and, except for wild pig, most were from Sipurak. The number of records for sambar and wild pig was low in each of the study areas.

Table 1.   Numbers of camera trap records of tiger and its presumed prey species within four study areas in the Kerinci Seblat region, Sumatra
SpeciesStudy areaTotal
Sumatran tiger Panthera tigris sumatrae15835251201
Muntjac Muntiacus muntjac11996129200
Pig-tailed macaque Macaca nemestrina231255966273
Sambar Cervus unicolor1145525
Wild pig Sus scrofa0761528

Kernel density and trigonometric sum estimates of activity patterns for tiger, pig-tailed macaque, muntjac and tapir showed little difference within each species (Fig. 1). For sambar and wild pig, sample sizes were considered too small to estimate the density reliably, as reflected in the larger differences between kernel density and trigonometric sum estimates for these species. Tigers were active throughout the 24 h period (54% of observations between 06:00 and 18:00), but exhibited peaks of activity around dawn and dusk, that is tending towards being crepuscular. Similarly, muntjac had peaks of activity around dawn and dusk, but its activity was more strongly diurnal (76% of observations between 06:00 and 18:00). The pig-tailed macaque was strongly diurnal (97% of observations between 06:00 and 18:00) and was more often photographed during the first half of the day. Tapir was strongly nocturnal (8% of observations between 06:00 and 18:00). Wild pig was predominantly diurnal (93% of observations between 06:00 and 18:00), while from the limited data available, sambar appeared to be cathemeral (52% of observations between 06:00 and 18:00).

Figure 1.

 Density estimates of the daily activity patterns of tiger and five presumed prey species in the Kerinci Seblat region of Sumatra. The solid lines are kernel-density estimates, whereas the dashed lines are trigonometric sum distributions. The short vertical lines above the x-axis indicate the times of individual photographs and the grey dashed vertical lines indicate the approximate time of sunrise and sunset.

The kernel density estimators inline image and inline image gave similar numerical values for these data and, for brevity, results are reported only for the estimator inline image. From this, muntjac and sambar had a high degree of overlap with tiger, as indicated by the estimated overlap coefficients ≥0.8 (Table 2; Fig. 1). However, for sambar, confidence intervals were much wider, due to the smaller sample size. Other species had notably lower degrees of overlap, tapir and pig-tailed macaque in particular (inline image). Estimates of overlap based on kernel-density estimates of the underlying activity pattern closely matched those based on trigonometric sum densities (Table 2).

Table 2.   Estimates of activity pattern overlap between tiger and its presumed prey species, with approximate 95% bootstrap confidence intervals in parentheses
Method of estimating activity patterns
Sumatran tiger
Panthera tigris sumatrae
Kernel densityTrigonometric
  1. The underlying activity densities were estimated either by kernel-density estimation or as trigonometric sum densities and the estimator of overlap being inline image

Muntjac Muntiacus muntjac0.80 (0.71–0.84)0.81 (0.73–0.87)
Tapir0.52 (0.44–0.60)0.53 (0.43–0.58)
Pig-tailed macaque Macaca nemestrina0.53 (0.46–0.58)0.54 (0.47–0.60)
Sambar Cervus unicolor0.81 (0.55–0.85)0.81 (0.52–0.83)
Wild pig Sus scrofa0.60 (0.44–0.71)0.67 (0.43–0.73)

Owing to the limited data recorded from study area 1 and for wild pig and sambar, variation between study areas (2, 3 and 4) was investigated for tiger with muntjac, tapir and pig-tailed macaque, respectively (Fig. 2). Muntjac, although predominantly diurnal in all three areas, showed some variability in the relative importance of morning and evening peaks. However, the overlap with tiger was similar in all three areas. Tapir had a higher overlap with tiger in the Sipurak area (inline image) than elsewhere (inline image), while for pig-tailed macaque, there was considerably higher overlap with tiger in the Ipuh area (inline image) than elsewhere (inline image).

Figure 2.

 Estimates of the daily activity patterns of tiger and three putative prey species in three study areas in the Kerinci Seblat region of Sumatra; Sipurak (row 1), Bungo (row 2) and Ipuh (row 3). The dashed lines are kernel-density estimates for tiger, the solid lines are kernel density estimates for the indicated prey species, based on individual photograph times that are indicated by the short vertical lines above the x-axis. The overlap coefficient is the area under the minimum of the two density estimates, as indicated by the shaded area in each plot. The estimate of overlap is indicated in each plot, with 95% bootstrap confidence interval in parentheses.


This study is the first to quantify the degree of overlap in activity patterns between the tiger and its putative prey species. These patterns revealed a close temporal overlap between tiger and both sambar and muntjac, which provide a complementary temporal perspective on tiger–prey spatial interactions to a previous study that found strong associations with tiger–sambar (O'Brien et al., 2003). Surprisingly, for the larger bodied and nocturnal tapir, which should not be too formidable a prey species, there was weak temporal overlap with the crepuscular tiger.

Statistical methodology

The results for the kernel-density estimation and the use of trigonometric series distributions were generally very similar in this study. However, the kernel-density estimation requires much less computing time, which is an important consideration when calculating bootstrap confidence intervals, where the difference in computing time can be a few minutes versus a few hours. Although the statistical methodology used within this study is quite complex, it was performed within the statistical package r (R Development Core Team 2009). The code and dataset used within this study have been made available online ( to support future work. Such work might, for example, focus on the development of formal statistical tests for investigating differences between overlap coefficients because our results revealed heterogeneity between study areas. When such heterogeneity exists, the estimate that results from pooling data across sites is always larger than the average of the separate estimates for each area (Ridout & Linkie, 2009). Further research into the differences between study sites would certainly be of interest in improving the understanding of how biophysical and anthropogenic landscape factors influenced temporal activity patterns.

Sumatran tiger–prey interaction

Large carnivore habitat selection should focus on the most profitable patches, where the lowest levels of energy are expended on searching for and then capturing the largest bodied prey with the least risk (Stephens & Krebs, 1987; Scognamillo et al., 2003; Carbone, Teacher & Rowcliffe, 2007). Thus, the Sumatran tiger would be predicted to select sambar (185–260 kg) and tapir (250–540 kg, Boonsong & McNeely, 1988). For sambar, which appeared to be less common in the KS study areas, there were insufficient data to determine activity patterns confidently. However, similarly designed camera-trap studies from southern Sumatra and Peninsular Malaysia found sambar to have predominantly crepuscular activity patterns (T. O'Brien, unpubl. data; Laidlaw & Shaharuddin, 1998). The high overlap between tiger and muntjac in our study was strong because both species exhibited peaks of activity around dawn and dusk.

To date, evidence of interactions between tiger and tapir is limited to photographic records of tiger attacks on tapir and speculation over the tapir's status as a prey species (Lynam, 1999; Holden, Yanuar & Martyr, 2003). Even though the tapir was frequently photographed and along trails used by tiger, our analysis found only a low level of temporal overlap; tapir was predominantly nocturnal. Thus, the lack of a tiger–tapir interaction may be because tapir is not a principal prey species, and this lack of relationship is suggested from Malaysia where overlap was low (Kawanishi & Sunquist, 2004). This is surprising because the Bengal tiger, which although larger than the Sumatran tiger (adult males of 180–258 and 100–140 kg, respectively, Nowell & Jackson, 1996), typically kills not only large prey (>176 kg) especially adult sambar but also occasionally adult male gaur Bos gaurus, which can attain an upper body mass of 1000 kg (Karanth & Sunquist, 1995; Andheria, Karanth & Kumar, 2007). Malayan tapir should not, therefore, be too large for a Sumatran tiger to kill.

An alternative explanation for the lack of positive tiger–tapir interaction may be the effect of predation risk on the prey. The ‘ecology of fear’ concept states that prey modify their behaviour by striking a balance for their need to forage against their need to avoid predators (Brown, Laundre & Gurung, 1999). Consequently, this trade-off may result in the avoidance of food-rich habitat patches, either spatially or temporally, which remain unoccupied by prey species if these patches also have significantly higher predation risks. Such risk has been shown to affect physiological and demographic patterns of elk, Cervus elaphus preyed on by grey wolves, Canis lupus (Creel et al., 2007) and spatial patterns of bighorn sheep, Ovis canadensis, avoiding open habitats that provide greater visibility for pumas, Puma concolor (Altendorf et al., 2001). As no studies exist of Malayan tapir temporal patterns in areas without tiger, we speculate that the tapir's strong nocturnal activity patterns is advantageous for avoiding its only predator in KSNP the tiger.

In comparison with the tapir, the pig-tailed macaque had an overall overlap coefficient almost identical to that of tapir, but in contrast was strongly diurnal, with significant activity occurring in the middle of the day, which is also the hottest time of the day. During this period, tiger activity was low, presumably because the species was resting. The limited data for wild pig suggest that it is also strongly diurnal (Laidlaw & Shaharuddin, 1998). Finally, there were several other putative prey species recorded in KSNP, argus pheasant Argusianus argus, mouse deer Tragulus spp., porcupine Hystrix brachyura and bearded pig Sus barbatus that may have influenced tiger temporal patterns. However, these were not included in the study as they were not considered to represent principal prey species because of their smaller body size (Karanth & Sunquist, 1995; O'Brien et al., 2003) or, in the case of the migratory bearded pig, an irregular food source.

Ideally, tiger scat samples would have been collected for a dietary analysis of prey species composition, but scats are notoriously difficult to collect in tropical forests, because of low tiger population densities and high scat decay rates, and none were encountered during our field surveys. However, in the absence of difficult-to-collect dietary data, it is also valuable to demonstrate the temporal relationships, as conducted in this study, to provide new and much needed insights into Sumatran tiger–prey interactions. The methodology used here has wide application, especially for future statistical studies of predator–prey interactions or interspecific species competition.


The authors thank the US Fish and Wildlife Service, 21st Century Tiger, Rufford Small Grants, and the Peoples Trust for Endangered Species for funding this work. The authors thank the Indonesian Department of Forestry and Nature Protection for assisting us in our work, Yoan Dinata, Agung Nugroho and Iding Achmad Haidir for their help with the data collection and Tim O'Brien, Phil Stephens, Patricia Medici and two anonymous reviewers for useful comments on an earlier draft of this paper.