Spatial and temporal variation in the abundances of primary prey can have major demographic effects on predators (Dale et al. 1994, O'Donoghue et al. 1997, Ramakrishnan et al. 1999, Karanth et al. 2004). Large declines in prey abundances may reduce the viability of endangered predator populations (Karanth & Stith 1999), and predators on islands may be especially vulnerable to declines in prey abundances (Frankham 1998). A key feature of many predator-prey systems is the large prey base required to sustain the predator (Karanth et al. 2002a, 2004). Hence, apex predators typically occur at low densities, making robust inference about changes in the abundance of the predator difficult because of the impracticality of obtaining even modest sample sizes. In such systems, monitoring prey abundances may be particularly useful for providing an early warning of changes in predator populations (Karanth et al. 2002b, Lovari et al. 2009).

The Komodo dragon Varanus komodoensis is the world's largest lizard. Adult male dragons can reach 3 m in length and weigh up to 87 kg (Jessop et al. 2006). Currently, the Komodo dragon is listed in Appendix I of the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) and is classified by the International Union for the Conservation of Nature (IUCN) as ‘vulnerable’ due to its demographic decline and limited distribution (World Conservation Monitoring Centre 1996, CITES 2011). The Komodo dragon is endemic to five small islands in eastern Indonesia, with four island populations in Komodo National Park (KNP) and several fragmented populations on Flores (Ciofi & De Boer 2004). The Komodo dragon is an apex predator, with three ungulate species dominating the diet of adults: rusa deer Rusa timorensis, feral pig Sus scrofa and water buffalo Bubalus bubalis (Auffenberg 1981, Jessop et al. 2006; Fig. 1). Previous work has shown the distributions and abundances of these three species to be important determinants of the demography of Komodo dragon populations (Jessop et al. 2006; but see Laver et al. 2012).

Anthropogenic threats such as habitat loss (from illegal logging and agriculture) and illegal hunting could reduce the densities of ungulates (Groom 2006, Steinmetz et al. 2010), and hence Komodo dragon populations, inside and outside KNP (Ciofi et al. 2007). For example, a reduction in the density of deer caused by illegal hunting was considered the major cause of the extinction of the Komodo dragon on the island of Padar within KNP during the early 1980s (Ciofi & De Boer 2004). Regular patrols by rangers since 2000 are thought to have reduced illegal hunting of deer and feral pigs within KNP. However, outside KNP, especially on the island of Flores, illegal hunting of deer still occurs and may be affecting the local distribution and abundance of the Komodo dragon (Ciofi & De Boer 2004). Despite these concerns, no attempt has been made to implement monitoring of the distribution and abundance of the ungulate prey species of the Komodo dragon.

A wide variety of methods have been used to estimate the abundance of ungulates (reviews in Thompson et al. 1998, Mayle et al. 1999). Faecal counts are a commonly used indirect sampling method for estimating the relative abundance of ungulates, particularly in wooded habitats where animals are difficult to observe directly (Mayle et al. 1999). The faecal count method is a relatively inexpensive method that is easy for park rangers to use and can be implemented in a wide range of habitats (Forsyth et al. 2003, Månsson et al. 2011). However, as for any index of animal abundance, it is desirable that the relationship between faecal density and animal density be validated for the species and area of interest: the relationship should be positive and linear (e.g. Forsyth et al. 2007). Distance sampling is a direct sampling method used to estimate the abundance of animals (Buckland et al. 2001, Thomas et al. 2010). Distance sampling can provide robust estimates of abundance at less cost than methods based on mark-recapture (Buckland et al. 2001, Wingard et al. 2011). Density is estimated using the perpendicular distances to detected animals from line transects (Thomas et al. 2010).

Our study had two aims. First, to estimate the densities of rusa deer, feral pigs and water buffalo using faecal counts and distance sampling at 11 sites in and around KNP. Second, to evaluate the relationship between faecal densities and ungulate densities at the 11 sites.

Material and methods

Distance sampling

We conducted distance sampling surveys in the same months as the faecal counts. However, most distance surveys were conducted in the early morning (05:30-09:00) and late afternoon (16:00-18:30) when ungulates were most active, to increase the likelihood of sighting individuals of all three species (for deer, see Focardi et al. 2005; for feral pigs, see Choquenot et al. 1996 and for water buffalo, see Corbett 1995). Transects were located systematically in each of the 11 sites, with the distance between transects ≥ 500 m. The 111 transects were of variable length (0.5-6.15 km) and a total 163.65 km was surveyed.

The same observers (A. Ariefiandy and D. Purwandana) conducted all surveys. Observers first walked 13.5 km of transect together to standardise methodology. Thereafter, the two observers surveyed alternate transects at each site. Surveys were conducted at a slow walking speed of 2-3 km/hour. When one of the three ungulate species was observed, the radial distance from the observer to the animal(s) was estimated using a laser range finder (Bushnell Range Finder Elite 1500, Bushnell Corporation, Overland Park, Kansas, USA) and the bearing to the animal(s) determined with an electronic compass (Garmin Summit, Kansas, USA). The number of animals in the group (‘cluster size’) was also recorded. Perpendicular distances were calculated from the radial distances and sighting angles by trigonometry (Buckland et al. 2001).

We analysed data using the program DISTANCE 6.0 release 2 (Thomas et al. 2009; available at:  http://www.ruwpa.st-and.ac.uk/distance/). Buckland et al. (2001) recommend having at least 60 observations for robust estimation of density using distance sampling. Since there were far fewer than 60 observations of each ungulate species at most sites, we used multiple covariates distance sampling to estimate a global detection function and cluster size for each species: densities were then estimated for each site from the species-specific encounter rate (Marques et al. 2007, Thomas et al. 2009). Cluster size was estimated using the size-bias regression method (Buckland et al. 2001).

Our exploratory data analyses revealed that detection data for rusa deer and feral pigs had long tails, so the 5% of detections with the greatest distances were discarded for both species (Buckland et al. 2001, Thomas et al. 2009). There was also evidence of both species moving small distances away from the transect line prior to detection, so we left-truncated observations of rusa deer and feral pigs by 5 m (Buckland et al. 2001).

Following Thomas et al. (2010), we evaluated the following detection functions, g(y), for each species, where y is the perpendicular distance (m) of an observation from the transect. The half-normal key,

was evaluated with cosine adjustments, and the hazard rate,

with simple polynomial adjustments. Further details on these keys and adjustments are given in Buckland et al. (2001). Histograms, quantile-quantile plots and Cramér-von Mises tests were used in conjunction with our biological understanding of each species in our study area to assess the fit of the keys and adjustments. We sought a detection function with a good shoulder and used Akaike's Information Criterion corrected for small sample sizes (AIC_c) and Akaike weights (w_i) to assess the relative support for each model (Burnham & Anderson 2002). Following Buckland et al. (2001) and Thomas et al. (2010), site-specific estimates of ungulate density (UD) are presented with their 95% CI and coefficient of variation (CV).

Relationships between faecal and ungulate population densities

Following Forsyth et al. (2007), we evaluated the relationship between faecal and ungulate densities with three models. We first considered the linear model,

where a is the intercept and b is the slope of the relationship between the UFD and UD at site i. Because the relationship between ungulate faecal counts and density could be non-linear, we next considered the power model,

We also considered another non-linear model, i.e. the logistic model,

Model parameters were estimated by non-linear least squares regression using the NLS2 package in program R 2.11.1 (Grothendieck 2010). The data used in the models were the mean faecal count and distance sampling estimates obtained at each of the 11 sites. Since non-linear least squares regression assumes that the variation in X (i.e. distance sampling estimates) is substantially less than the variation in Y (i.e. faecal count estimates), we evaluated the variation in these estimates for each ungulate species. Since there was no evidence of overdispersion in the data (evaluated by comparing the degrees of freedom to the residual deviance for each model; Crawley 2007), we again used AIC_c and w_i to assess the relative support for each model.

Results

Discussion

The primary objective of our study was to evaluate the usefulness of faecal counts and distance sampling for monitoring the abundances of the ungulate prey of the Komodo dragon in and around KNP. The importance of estimating changes in the abundances of rusa deer, feral pig and water buffalo populations will depend on their relative importance as prey for Komodo dragons. Demographic and diet studies suggest that rusa deer are the most important ungulate prey species for Komodo dragons (Auffenberg 1981, Jessop et al. 2006, 2007). Our results indicate that distance sampling is a useful technique for estimating rusa deer density on large island sites, especially on Komodo Island where CVs were ≤ 21%. However, estimated rusa deer densities were much less precise at sites with few observations (i.e. CVs ≥ 42%).

There were two major limitations to the use of distance sampling in our study. First, the target species was not always observed during distance sampling despite being detected during faecal counts, resulting in underestimates of density at those sites (e.g. feral pig at Loh Buaya). Second, far fewer observations than the minimum 60 required for robust distance sampling estimation (Buckland et al. 2001) were recorded for the three ungulate species at some sites. Although using global detection functions and cluster sizes (i.e. multiple covariates distance sampling; Marques et al. 2007, Thomas et al. 2009) overcame that limitation for rusa deer and feral pig, there were still too few observations to robustly estimate the density of water buffalo at any site. Increasing the distance sampling effort (e.g. by walking more transects and/or by walking transects twice; Wingard et al. 2011) would require substantially more resources than were available for our study. However, our inability to estimate water buffalo densities using distance sampling is unlikely to be a major limitation for a Komodo dragon prey monitoring programme because this species is a less important prey species compared to rusa deer and feral pig (Auffenberg 1981).

We found positive relationships between estimated faecal and population densities for both rusa deer and feral pig at the 11 sites (see Fig. 5). Although there was a positive relationship between rusa deer pellet group and population densities, the form of the relationship was non-linear and contrasts with the linear relationship observed in 20 enclosures with known densities of deer in New Zealand (Forsyth et al. 2007). It is desirable to have a positively linear relationship between an index of abundance and true abundance (Thompson et al. 1998, Gibbs 2000, Månsson et al. 2011), and the non-linearity observed for rusa deer in our study is likely due to observers undercounting pellets at sites with very high deer densities. Further work is needed to determine if the linearity of the relationship between rusa deer population density and faecal counts can be improved by better training of observers and/or modifying the sampling protocol. One possibility is to use distance sampling to estimate the abundance of deer pellet groups along transects (Marques et al. 2001), although tall grass and thick leaf litter may obscure some pellet groups, potentially violating the assumption that all objects on the transect are observed. Estimating the rates at which faeces are deposited and decay would enable population densities to be estimated, as done for sika deer Cervus nippon in Scotland by Marques et al. (2001). However, estimating faecal deposition and decay rates is a major task that would require substantial resources (Forsyth et al. 2003). In contrast to rusa deer, the relationship between feral pig population density and faecal density was best explained by the linear model, although there was some evidence that the relationship may have been non-linear. An assumption of non-linear least squares regression is that X is measured without error, but ungulate densities estimated with distance sampling were often subject to substantial uncertainty. It would, therefore, be desirable to have more precise estimates of ungulate densities but, as noted above, this would require an impractically large sampling effort. It would also be desirable to include a larger number of sites in these analyses, particularly for feral pig which were present at only eight of the 11 sites.

The time required to conduct faecal counts was half that required for distance sampling. Compared to distance sampling, faecal counts were also technically easier for rangers to conduct (requiring cheaper equipment). Another advantage of faecal counts is that the standing crop of faeces accumulated during the dry season, when decomposition is limited (A. Ariefiandy, unpubl. data), is counted. Faecal counts, therefore, capture a broader temporal window of ungulate activity within any site and may be more representative of ungulate density than the direct observations needed to estimate density using distance sampling.

The maximum rusa deer and feral pig densities estimated using distance sampling in our study (165.5 and 25.2 animals/km², respectively; see Table 2) are similar to the highest densities documented elsewhere. Densities of 75-160 rusa deer/km² were estimated in New Caledonia using ‘nocturnal line-transect sight counts’ (de Garine-Wichatitsky et al. 2005), and densities of 27.0 and 46.0 feral pigs/km² were estimated using distance sampling in a lowland Malaysian forest by Ickes (2001). Understanding the drivers of ungulate abundances was beyond the scope of our study, but the low densities of rusa deer and feral pig estimated at WW were likely a consequence of illegal hunting. Preventing illegal hunting may increase the densities of rusa deer and feral pig at WW.

Additional parameters could also be monitored in the context of Komodo dragon conservation. In particular, changes in the body condition of rusa deer, feral pigs and water buffalo (e.g. kidney fat (Riney 1955) or bone marrow (Sinclair & Arcese 1995)) may provide an ‘early warning’ of changes in the quality of prey populations due to density-dependent processes (Bonenfant et al. 2009). Monitoring the hunting efficiency (e.g. kill rate) of adult Komodo dragons (sensu Jędrzejewski et al. 2002, Smith et al. 2004) would also be useful; declines in the kill rate would be a cause for concern, perhaps indicating reduced prey abundances/vulnerability. Developing protocols for estimating these additional parameters in and around KNP is a priority area for further work.

Management implications

Monitoring of endangered predators and their prey is conducted for three main reasons (e.g. Karanth et al. 2002a). First, to evaluate the success or failure of management actions. Second, to establish benchmark data as a basis for future management. Third, to increase general knowledge of the system and hence increase the capacity of managers to deal with new situations. Because spatial and temporal variation in ungulate density is likely to influence the population dynamics of the Komodo dragon, we recommend that annual monitoring of ungulates in and around KNP be undertaken using distance sampling and faecal counts. Continued annual monitoring of ungulate prey would provide valuable information for the management of Komodo dragon populations both inside and outside of KNP. Ungulates, particularly rusa deer, are the primary prey of large Komodo dragons (Auffenberg 1981), and Jessop et al. (2006, 2007) showed that the maximum size of Komodo dragons increased with increasing ungulate prey density. Subsequent work will further investigate the relationships between Komodo dragon population dynamics and ungulate prey density, including annual and seasonal changes in hunting efficiency and kill rates.

Although site-specific factors such as vegetation composition and microclimatic conditions, and also season, may influence the deposition and decay rates of ungulate faeces (Van Etten & Bennett 1965, Neff 1968, Putman 1984), the relationships reported here will be useful for managers establishing monitoring programmes for feral pig, rusa deer and water buffalo elsewhere in their native and exotic ranges. The positive relationships between the abundances of dung and densities of rusa deer and feral pig observed in our study enable managers elsewhere to use these techniques confident that they can identify substantial differences in abundance (i.e. the relationships are positive and approximately linear). The feral pig has a particularly large and expanding global distribution (Oliver 1993, Long 2003) and causes significant economic and environmental impacts (reviews in Choquenot et al. 1996, Lowe et al. 2000) but, despite dung counts being a commonly used index of feral pig abundance (e.g. Choquenot et al. 1996, Hone 2002), this is the first study to investigate the form of the dung-density relationship for this species. Our results for rusa deer support previous work in New Zealand showing that the abundance of faecal pellets from other deer species (primarily red deer Cervus elaphus scoticus) increases with deer abundance. However, we encourage managers and researchers, whenever possible, to validate their indices of abundance in their study areas.