Baluran National Park (BNP) is located on the northeastern tip of Java (7°50′0″S, 114°22′0″E) in Indonesia. The total area is 250km2 and includes primary and secondary forest, savannah, shrub forest, and mangroves (Fig. 1). Two species of primates are currently found in the national park, the Javan lutung and the long-tailed macaque, while the Javan slow loris (Nycticebus javanicus) may occur in parts yet to be surveyed . Potential predators of the primate species include the Javan leopard, dhole (Cuon alpinus), various birds of prey such as the changeable hawk-eagle (Nisaetus cirrhatus) and short-toed snake-eagle (Circaetus gallicus), and the reticulated python (Malayopython reticulatus).
Human activity within the park influences the biodiversity, especially mammal species richness in areas surrounding permanent settlements [39, 40]. These settlements were already established when the park was gazetted in 1980 , and their approximately 4000 domestic cows and goats use 22% of national park habitat for grazing, which has had a negative impact on native mammal wildlife occurrence in the area . Native wildlife in BNP is also threatened by invasive acacia (Acacia nilotica), that has invaded the native savannah, and in 2013 covered roughly 90% of it [41, 42]. Tourism is extensive in designated areas (86,000 visitors in 2017) [pers. comm. Arif Pratiwi], and a highly trafficked highway traverses the outer southern part of the park.
The Javan lutung is a diurnal colobine primate that occurs only in Indonesia; on the islands of Java, Bali and Lombok . It is classified as Vulnerable on the IUCN Red List with a decreasing population size . The Javan lutung diet is more versatile than other folivorous colobines, and includes young leaves, fruits and flowers. They are able to also feed on the leaves of teak trees (Tectona grandis) in plantations . They occur at different elevations ranging from 0-3500masl , and experience lower densities and group sizes at higher elevations . Information on home ranges is only available from the long-term field site in Pangandaran; here Javan lutung groups have well-defined home ranges of ~ 4-6 ha .
On 2 December 2019, we conducted a search in Google Scholar for articles written about Javan lutung to assess if there had been a study bias to one or more particular habitat types (Table 1). We used ‘Trachypithecus auratus’, ‘Presbytis auratus’ and ‘Presbytis aurata’ as search terms, and included only studies that were conducted on wild populations (thus excluding studies on captive lutungs) that were conducted at one location (thus excluding surveys). We included articles, reports, book chapters, and theses in English, German and Indonesian.
Follows and ad libitum observations
Research was conducted from February 2017 through June 2017 for preliminary observations, liaisons with stakeholders and reconnaissance, spending approximately 3 days per week inside BNP, and from July 2017 to May 2018 for more systematic data collection on the primates, spending 5 days per week inside BNP following focal groups or conducting line transect distance sampling (see below). Wet season in BNP occurs from November through April, and dry season from May through October with modifications each year. We conducted follows of groups at different time slots during the day, divided into morning (05.30–09.00 h), midday (09.00–15.00 h) and afternoon (15.00–17.30 h), and this was augmented by ad libitum observations . The lutungs were mostly located by vocalisations and sounds from shaking branches and leaves as the animals move through the canopy. None of the groups were habituated. We followed Javan lutung groups from the sleeping sites into other habitats as far as they and the landscape permitted in the morning, tried to locate them midday and follow them back to their sleeping sites in the afternoon, with detection certainty and following distance decreasing with habituation. We encountered six groups regularly. They were identified by the location of their sleeping sites, their group sizes and composition, especially the number of adult males (Table 2). Javan lutung groups have well-defined small home ranges in Pangandaran ~ 4-6 ha , and we expect this to be the case in BNP as well, although habitats differ. For observations, we maintained a distance of 20 m in areas such as sleeping sites, whereas in the dense scrub habitats in BNP this distance increased greatly until the groups were not visible anymore.
Line transect distance sampling
We created a systematic grid covering the entire area, excluding the higher parts of Mt Baluran (200masl – 1250masl), which was too treacherous for a systematic census. This did however; exclude a large part of the primary forest, which could have affected our results. Because Javan lutung are the most observed mammal species on the mountain [pers. comm. Arif Pratiwi], we decided to include the mountain in our estimates. We placed the grid randomly in respect to habitats and wildlife distribution [25, 45], but ensured all transects began at roads or trails (Fig. 1). This ensured that we did include all habitats, also anthropogenic ones. We did not place transects according to anecdotal knowledge of lutung distribution to ensure that we did not affect results through our transect grid.
All transects were 4.5 km long with at least 2.0 km between adjacent transects totalling 189.0 km for outbound and return trip. However, we were only able to walk 160.65 km due to inaccessibility (Fig. 1). Between October and December 2017, a team of three to four observers walked the transects at a speed of 1.25 km hr.− 1, finishing one transect trip (outbound or return) within 4 h with observations never exceeding 15 min [24, 25]. Outbound trips were conducted in the morning between 07.00 and 11.00 h, and return trips in the afternoon between 13.00 and 17.00 h. An observation was the detection of Javan lutung, repeated counts, and distance measurement. We aimed at detection of all Javan lutung at zero distance (on the line) [25, 46]. To keep our bearing and direction, find the perpendicular point on the transect (initial location of detection) and waypoint the position, and measure the perpendicular distance (PD) to sub-group centre (i.e. the midpoint of individuals that were within sight), we used a Garmin GPS Map 64 s, a compass, and a Nikon Aculon A11 Rangefinder [23, 38, 47]. We used repeated counts to count sub-group size . To increase our resolution of distance measurements, we focused on smallest visible clusters of individuals – “sub-groups”, where group centre is more accurately estimated [36, 48]. We always measured the PD from the line . We detected animals via sight and vocalisations, yet only counted them when visual . We recorded the pelage colour of Javan lutung, distinguishing between black and erythristic individuals.
Co-occurrence and polyspecific associations with long-tailed macaques
When searching for Javan lutung, we also systematically searched the surrounding landscape for long-tailed macaque groups, and registered all interactions. All encounters of long-tailed macaques were also registered during line transect distance sampling. We recorded encounters within 50 m and 5 min of a Javan lutung group as a spatial co-occurrence encounter .
Line transect distance sampling
We only included the outbound trip (41 sub-group sightings) in our analysis to avoid double counts. The analysis was conducted in Distance 7.1 . For calculating population size, we considered the entire BNP area, i.e. 250km2, acknowledging that lutung also range outside BNP, especially to the southwest towards Mt Ijen . We could not secure accurate PD for three sightings, and these were excluded from the analysis.
We right-truncated our data to increase robustness in estimating detection function, excluding all observations beyond 90 m . We investigated histograms without truncation to determine truncation distance [pers. comm. Eric Rextad and Tiago Marques]. This excluded four observations, approximately 10% of the data [24, 49]. We tested all detection function combinations: 1. Uniform key with cosine, simple polynomial, and hermite polynomial adjustments, 2. Half-normal key with cosine, simple polynomial, and hermite polynomial adjustments, 3. Hazard rate key with cosine, simple polynomial, and hermite polynomial adjustments. Due to lowest AICc and best goodness of fit (GOF) tests according to p < 0.05 and lowest χ2/df accumulated for χ2 tests half-normal key detection function with 2 cosine adjustments was chosen [22, 24, 46]. We used the lowest AICc due to the low observation size .
Species Distribution Modelling (SDM) environmental predictors
Shapefiles containing vector layers provided by BNP, updated using Google Earth enabled us to create a map of BNP in QGIS 2.18.19, in which vector layers from the LTDS census were inserted. Using the raster package in R , we generated 26 raster layers (15 × 15 m resolution) with potential environmental predictor variables for the habitat suitability analyses (described below). Topographic layers (n = 5) included elevation (m), slope (°), aspect (radians), hill-shade (radians) and terrain ruggedness (index), derived from a digital elevation model of the study area. We generated a raster layer using ESRI shapefiles, that included 13 major vegetation/habitat types found in the census region including: rice fields, livestock fields, teak plantations, dwarf forests, evergreen forest, shrub forest, primary forest, secondary forest, mangroves, acacia forest, savannah, restored savannah, and beach. Instead of using categorical classes for each habitat/vegetation type, and to capture potential edge effects in habitat suitability, we generated new rasters (15 × 15 m pixel size). We did this by calculating for each raster cell the Euclidean distance (km) to the nearest cell with a given vegetation type (n = 13). Euclidean distance to (km) paved roads, human settlements, trails, and rivers found in the census region were included in the eight last raster layers.
Once all raster layers were created, we excluded collinear raster layers by calculating the Variance Inflation Factor (VIF), then excluding the one with highest VIF and repeating this process until only layers with a VIF < 2 were remaining . With this procedure we ended up with 10 covariates for the final MaxEnt modelling; macaque suitability, distance to roads, distance to secondary forest, elevation, distance to rivers, distance to restored savannah, distance to savannah, distance to shrub forest, distance to trails and distance to agriculture and rice.
SDM evaluation and mapping
We used the SDM package in R for modelling and mapping of habitat suitability, and employed the MaxEnt algorithm . Presence locations (N = 62) comprised projected waypoints of observations from both the outbound and return trip of each transect from the LTDS population census. All non-correlated environmental raster layers were used to extract the values of environmental conditions for presence points.
Model accuracy of MaxEnt was assessed by calculating the area under the curve of the receiver operating characteristic (AUC ;). An AUC value of 1 indicates perfect performance, whereas an AUC value of 0.5 indicates that the model performs no better than a random model. AUC > 0.7 generally indicates good model accuracy . We furthermore calculated the True Skill Statistic (TSS). A TSS value below 0 indicates a no better than random model performance, and a value of 1 indicates perfect performance . Change in the AUC value (ΔAUC) with and without a specific environmental variable, but with all other variables included, was used to evaluate variable importance for the habitat suitability map.