Study area
The fieldwork was carried out in Tabin Wildlife Reserve (Tabin or TWR) (5°15'–5°10'N, 118°30'–118°45'E), a 1,205 km2 protected area (gazetted as a Wildlife Reserve in 1984) in the eastern part of the Malaysian State of Sabah in north-eastern Borneo (Fig. 5). Sabah lies close to the equator, possessing a relatively constant tropical climate with an annual rainfall in the study area of about 3000 mm. Tabin is Sabah's largest and oldest wildlife reserve [44, 45], currently managed by the Sabah Wildlife Department and the Sabah Forestry Department.
Excluding the so-called core area and seven smaller Virgin Jungle Reserves, all other areas of Tabin (more than 80% of the reserve) have been selectively logged between 1969–1989 [45, 46]. No legal logging has taken place after 1989 [47]. Thus, TWR is a mosaic of forest types in different succession stages. A gravel road running north to south along the western boundary separates the forest reserve from the adjacent palm oil (Elaeis guineensis) plantations.
The reserve plays an important role as a dedicated ground for the conservation of protected mammals in Sabah. Tabin is home to some endangered flagship mammals, such as the Borneo pygmy elephant (Elephas maximus borneoensis), the Sumatran rhinoceros (Dicerorhinus sumatrensis harrissoni), the banteng (Bos javanicus lowi) and the orang utan (Pongo pygmaeus pygmaeus).
The study site was located adjacent to the Tabin field station on the western boundary of the reserve comprising 6 km along the North-South road, and another 6 km east and west along an old logging road.
Determining the size of the area surveyed
An existing road, trail and stream system was used for all tracking operations. This method promised to be more successful than a square-based area approach with a straight transect grid, because large cats are likely to travel on existing paths [21, 48–50]. A buffer was created around each transect to estimate the size of the surveyed area as accurately as possible. To calculate the buffer width, ecological factors of the target species were required [51]. Recent studies used the distance moved by tigers between two photo-recaptures to calculate this parameter [33, 39]. However, Soisalo and Cavalcanti [35] recently pointed out that, due to an underestimating of the distance moved by the animals, the calculations might overestimate the true densities. In contrast, other studies used functions of home range size, density and trap spacing to calculate the buffer width [51]. To overcome the uncertainties we considered both approaches and designed the following equation to determine the buffer width W in our study:
where C is the core area of home range sizes an (M) is the average daily movement. Values for C (C = 6 km2) and M (M = 1.932 km) were obtained from Grassman et al. [9] in Phu Khieo Wildlife Sanctury, Thailand, since there were no data available on these parameters from Borneo. We preferred to use the core area instead of the total home ranges to calculate the size of the area surveyed because long distances travelled by large cats may increase the total home range size significantly.
The North-South road forms the boundary separating Tabin Wildlife Reserve from the adjacent palm oil plantations. A buffer calculated by equation 1 would have overestimated the surveyed area, because it would have included the nearby plantations which do not constitute suitable habitat for clouded leopards. Although clouded leopards were observed entering plantations in Borneo (Sabah Wildlife Department pers. comm.; pers. obs.), presumably following their prey, they were never seen deeper than 300 m inside the palm oil plantation (Sabah Wildlife Department pers. comm.; pers. obs.). Thus it was assumed that a smaller buffer width of 300 m to the west of this road transect would be adequate to describe the survey area.
Data collection
During March and August 2005, eight transects crossing different habitats were established and each transect was surveyed 20 times. In addition to two transects along the gravel road and one along the old logging road towards the reserve's centre, two transects followed existing jungle trails and three transects followed streams. The total length of all transects was approximately 35 km. Every 250 m a GPS coordinate was taken and a digital map showing all transects was produced using the program ArcGIS 9.1 (ESRI Inc.)
Our sampling unit was a track set (TS), defined as one or more contiguous pugmarks from any paw made by the same clouded leopard. Tracks were photographed with a digital camera (4 mega pixel) fixed on a monopod perpendicular to the track. An umbrella was used to adjust to the light conditions. A scale was placed on two sides of the track to standardize measurements. Only those tracks in good condition with clear edges and in flat terrain were included in the analysis to ensure accurate measuring. GPS coordinates were taken of each track set and later digitized in ArcGIS 9.1 (ESRI Inc.).
Track measurement
To discriminate individual animals, 14 linear and five area measurements were taken from each track (Fig. 6). Measurement technique were adopted from a variety previous studies [11–13, 52] with the intent to increase the level of discrimination. The units of linear and area measurements were millimetres with a 1 mm and 1 mm2 level of precision respectively. Angle measurements, which proved to have a high level of discrimination in previous studies [12, 52], varied greatly among tracks of a given individual thus making this tool inadequate for discriminating individuals in our study. All digital track photographs were measured using Adobe Acrobat 7.0 Professional™ (Adobe Systems, Inc.).
Statistical and analytical analysis
For the analysis, it was presumed that we could differentiate between pugmarks made by front and rear feet as well as by left or right feet. Confusion with pugmarks from other cat species could be excluded as a possibility, because no other large cats are present in Borneo. Confusion with tracks of bay cats (Catopuma badia), which might have tracks sizes similar to small clouded leopards, can be ruled out because no confirmed observations of bay cats have been made in Tabin. In order to determine if left and right tracks could be combined for the analysis to enlarge the data set, we used a paired t-test to compare the means of the total width from left and right tracks of each TS. This was done independently for front and rear tracks. We tested the other linear and area measurements as well to determine any differences between the variables. The t-test could be applied because it could be assumed that the means have a normal distribution.
To achieve an optimal separation of each TS, a standardized principal component analysis (PCA) was applied. Principal component (PC) 1 against PC 2 separated individuals better in a scatter plot than two of the original variables did [12]. We excluded the width of the heel pad and of each toe in our analysis, because the information in these variables is correlated highly with the length and area of the heel pads and toes, respectively. The remaining 14 variables were treated as being equally important, having the advantage of coping with linear and area measurements. We favoured the PCA over a discriminant analysis, which has been applied in similar studies [11, 13, 52], because this method does not require that the number of clouded leopards is known prior to the analysis. The PCA does not classify data into fixed groups of clouded leopards, because the number of groups was unknown. Rather it associates each track with another, even if they derive from the same TS. Tracks from the same TS should cluster together in space, as will tracks from different TSs made by the same individual. All data were analyzed using STATISTICA 6 (StatSoft, Inc. 2001).
After matching the tracks to individual clouded leopards the capture histories for each animal were developed, in a manner utilized by camera-trapping studies [33, 36, 53–55]. The capture history data were analyzed using the software CAPTURE [29, 32, 56] developed to implement closed population capture-recapture models. This program uses a number of different models to generate abundance estimates for a sampled area, based on the number of individual animals captured and the frequency of recaptures. The models differ in assumed sources of variation in capture probability, including individual heterogeneity, behavioural response, variation over time and various combinations of these. CAPTURE uses a discriminant function model selection algorithm to provide an objective criterion for selecting the best approximating model. In addition, CAPTURE statistically tests the closure assumption.
The abundance estimates were then used to estimate the clouded leopard densities, defined as D = N/A, where N is animal abundance and A is the effective surveyed area sampled.
Application of the results on the landscape level
Digital maps of all protected areas within Sabah and the results of their last faunal survey (2000–2001) provided by the Sabah Wildlife Department were used for the large scale analysis. To estimate future prospects of clouded leopards in various protected areas different variables were taken into account. Most important for the evaluation were the presence of clouded leopards, the reserve size, connectivity and classification of the protected areas. We classified the reserves as a) totally protected reserves and b) commercial forest reserves, where the commercial forest reserves are consistent with class 2 of the classification by Sabah Forestry Department. We pooled the classifications of class 5 (mangrove forests) and class 7 (wildlife reserves) within the designation of class 1 (totally protected areas), because in all of these classes hunting and selective logging are prohibited, making them subject to protective conditions. Only areas which were big enough to hold a minimum population of 50 individuals [57, 58] were included in the analysis since smaller populations of large cats, such as the Florida panther (Puma concolor coryi), experienced reduced viability and fecundity caused by inbreeding [59]. Furthermore, smaller populations are more susceptible to environmental and demographic stochasticity. Therefore a minimum reserve size was calculated based on our density estimation in TWR. Due to a lack of detailed data, we assumed densities to be similar in all protected areas and calculated a rough number of clouded leopards within each reserve, based on the reserve sizes and the density obtained from our results in the TWR.