Population and distribution of wild Asian elephants (Elephas maximus) in Phu Khieo Wildlife Sanctuary, Thailand

Study area

Our study was conducted from 2016 to 2019 in the Phu Khieo Wildlife Sanctuary (PKWS) which spans over 1,560 km² in the Chaiyaphum Province of Thailand (latitude 16°5′ and 16°35′N and longitude 101°20′ and 101°55′E) (Fig. 1). Eight connecting protected areas cover more than 4,594 km² in the Western Issan Forest Complex. The sanctuary is near three other protected areas: the Nam Nao National Park to the north, Tat Mok National Park to the west, and Ta Bao-Huai Yai Wildlife Sanctuary to the southwest. The annual rain fall is 1,368 mm per year, and the average temperature is approximately 18 °C to 27 °C. The average elevation is 900 m (min. 200 m to max. 1,300 m) above the average sea level (ASL). Dry evergreen forests cover approximately 68% of the area. Of this total area, 27% are mixed deciduous forests dominating the low lands, 4% are dry dipterocarp forests, 0.6% are pine forests, and 0.4% area secondary forest (Faculty of Forestry, 2010).

Camera trap survey

Global Positioning System (GPS) coordinate information was collected from camera trap placements in the field. The environmental conditions were the same ones recorded by Pearce & Boyce (2006). A set of environmental factors that likely model a species’ environmental requirements was obtained from a set of occurrence localities, influencing the suitability of the environment for the species (Phillips, 2017). Seven environmental factors were combined with the five predicted factors to generate the model. These were: human activity (road, stream, wildlife guard station, and villages), forest types, salt licks, elevation, and slope. As previously reported in Chaiyarat et al. (2019), environmental parameters generated from available Geographic Information System (GIS) layers and habitat composition was analyzed using the land-use layers of the Phu Khieo Wildlife Sanctuary digital map. Topography data was obtained from a digital elevation model (DEM) generated by the Faculty of Forestry (2010) from 1:50,000 topographic maps. A DEM was used to generate the slope and then images were resampled to a 30-m pixel resolution (Fig. 2). These parameters were used to estimate the distance between parameters in each pixel and each elephant observation point. A model built using 49 camera sites captured wild Asian elephants in 4,319 photographs. The other 100 camera trap stations were considered pseudo-absent from the total 149 camera trap stations, which captured 38,834 photographs. Images were captured using MaxEnt set to a 25 random test percentage with 15 replicates (more than the sample size) and 5,000 maximum iterations.

We used the HCO SG565 flash camera-traps (HCO Outdoor Products, Norcross, Georgia, USA) to obtain photographs of individual elephants. Fifteen camera trap stations systematically were set up in 3  × 3 km² grid cells within 3 months of sampling blocks. Two camera traps were used per grid with 10 m apart from each other. The camera traps operated continually for 24 h a day, recording the date and time of each photograph. The batteries were changed in the camera traps monthly, and the SD card data was transferred to HD data storage. Camera traps were attached to trees approximately 0.75 m from the ground (Rowcliffe et al., 2008) with a view range of at least five m to six meters to capture a single wild elephant up to 20 m away to allow for a complete view of a wild elephant herd (Varma, Pittet & Jamadagni, 2006).

Relative frequency and relative abundance index

The correlation structure of the set of environmental factors (salt lick, wildlife sanctuary guard station, stream, village, road, elevation, and slope) using a matrix of Kendall’s rank correlation coefficients, τ (Kendall, 1938; Halvorsen et al., 2016) presented by Chaiyarat et al. (2019) were used to test the autocorrelation among each environmental factor. Relative frequency (RF) was used to estimate the distribution of the wild Asian elephants, and the relative abundance index (RAI) was used to estimate the abundance of the wild Asian elephants.

Population survey

Photographs were used to identify and record the location, date, and time of wild elephant sightings. A score of 0 indicated bad image quality and the photograph was discarded. A score of 5–10 indicated that the picture quality was sufficient to identify individual herd characteristics. Photographs’ ratings were based on their quality, clarity, and the position of the elephants in the frame (Varma, Pittet & Jamadagni, 2006; Varma, Baskaran & Sukumar, 2012). A rating of five or above would allow us to determine whether elephants were individual adult males, females, sub-adult males, females, juveniles, or calves (Arivazhagan & Sukumar, 2008). We recorded elephants’ clear morphological distinguishing features and basic body measurements to help identify individuals (Goswami et al., 2012; Vidya, Prasad & Ghosh, 2014) (Supplementary S1). Photos were used to identify individual elephants and unique herds. We conducted a census of individual adult males (AM) and adult females (AF), sub-adult males (SM), sub-adult females (SF), juveniles (JU), and calves (CA) for each herd. Herd density was calculated using crude density.

Herd distribution and distribution model

We used camera traps (Aebischer, Robertson & Kenward, 1993) to obtain statistically independent wild Asian elephant distribution data. Photographs taken by our traps were used to identify each herd. The location of each herd was used to predict distribution patterns and population habitat with MaxEnt. We only used MaxEnt with species records when individuals were either present, pseudo–absent, or truly absent at any given point on the landscape within a given time frame (Pearce & Boyce, 2006).

We used the jackknife procedure and percentage factor contributions to estimate the relative influence of different predictive factors in MaxEnt. Model performance was evaluated using the area under the receiver-operating characteristic (ROC) curve (AUC) (Fielding & Bell, 1997) as previously reported in Chaiyarat et al. (2019).

Population home range

Wild Asian elephants are a herd animal; therefore, we grouped their home ranges for our study. In Chaiyarat, Youngpoy & Prempree (2015) and Chaiyarat et al. (2019), we used the kernel density-estimate (KDE) bounds on the innermost 95% of the 49 presence data points to estimate habitat areas (Seaman et al., 1999). The model derived from this equation was used to create a habitat use map in ArcGIS 9.3 (Environmental Systems Research Institute; ESRI, 2007).

Statistical analysis

The RF was calculated for all camera trap locations as previously described in Chaiyarat, Youngpoy & Prempree (2015):

RF = No. camera locations that captured photographs ×100/total camera locations

The RAI calculated for all camera trap locations as:

RAI = No. of detections for wild Asian elephants ×100/total number of camera trap nights

Wild Asian elephant detection was considered to be independent if the time between consecutive photographs of the same individual was more than 0.5 h apart. This definition follows (O’Brien, Kinnaird & Wibisono, 2003).

The crude density (D) was calculated as:

D = Total number of wild Asian elephants/total number of Phu Khieo Wildlife Sanctuary

The habitat suitability area ranged from 0 (completely unsuitable habitat) to 1 (optimal habitat). Suitable habitats were classified into three categories:

Most suitable = most optimal habitat for wild Asian elephants >0.66 to 1

Moderate suitable = moderate optimal habitat for wild Asian elephants >0.33 to 0.66

Lowest suitable = low optimal habitat for wild Asian elephants >0 to 0.33

In Chaiyarat et al. (2019) we used MaxEnt to calculate the AUC value in a slightly different manner (Phillips, 2017) an AUC value of 0.5 indicated that the model did not perform better than a random model, whereas a value of 1 indicated perfect discrimination (Swets, 1988).

A one-way ANOVA was used to compare the population structure in different areas. The correlation coefficient was used to analyze the relationship between population structure and water sources, natural licks, elevation, slope, and forest types. We used SPSS as previously described in Chaiyarat, Youngpoy & Prempree (2015). Environmental factors affecting population structure were considered to be significant at p < 0.05. This work was conducted under an appropriate animal ethics approval (COA. No. MU-IACUC 2016/17) with permission from the Department of National Parks, Wildlife, and Plant Conservation (NRCT No. 0402/3908).

Autocorrelation test

The correlation structure for seven environmental factors (station and stream, station and village, station and elevation, village and elevation, salt lick and road, and road and slope) had a very strong positive correlation. In contrast, there was a negative correlation between two factors (stream and village; stream and elevation; t = 0.192) that was very statistically significant (p < 0.001) (Table 1). These environmental factors helped determine the distribution models.

Relative frequency and relative abundance index

The RF was 32.9% for all camera trap stations and the RAI of wild Asian elephants in the PKWS was 62.6 captures per 100 trap nights (Table 2).

Population survey

A total of 114 wild Asian elephants were identified from 4,319 photographs taken at 49 camera trap stations in the PKWS (Table 2). The crude density of wild Asian elephants in the sanctuary was 0.07 individuals per km². The population consisted of 30 adult males, 43 adult females, 14 sub-adult males, nine sub-adult females, 11 juveniles, and seven calves, with the population ratio was 0.7:0.1:0.3:0.2:0.3:0.2. The ratio of reproductive ability between adult females (including sub-adult females), juveniles, and calves was 1.0:0.2:0.1 (F = 1.072, df = 5, p = 0.382) (Table 3).

Herd distribution

Wild Asian elephants were separated into seven herds based on individual classifications (Fig. 3). Herds 1–5, roamed in the northwestern part of the sanctuary, while the other two preferred the eastern (herd 6), or western regions (herd 7).

Distribution model

We used MaxEnt to calculate the habitat model of the wild Asian elephants as previously described in Chaiyarat et al. (2019). Our results revealed that all 15 models generated training or testing models when compared with a random model where the average AUC was 0.61 ± 0.13. The test AUC values were lower than the training AUC values (Figs. 4A, 4B). The average training AUC values were 0.689, while the test AUC values ranged from 0.393 to 0.819. The model was run 15 times (Table 4). The contribution of the environmental factors and the results of the jackknife test analysis are presented in Fig. 4C. Analysis of the environmental factors independently indicated that distance from salt licks, elevations, land covers, and forest types were more important than slopes. The distance from salt licks was the most important predictor (36.4%) of habitat suitability. The second and third most important factors were elevation (23.5%) and human activities (20.2%), respectively. The contribution of environmental factors (Table 4) and response curves (Fig. 5) showed that the main environmental factors affecting habitat suitability were elevation (52%), forest types (23.2%), and distance from the salt licks (14.9%). Using the MaxEnt habitat model, we determined that species’ suitable area was 1,288.9 km²; 276.9 km² was found to be unsuitable (Table 5). Highly and moderately suitable areas were 672.3 and 616.6 km², respectively. Most elephant herds were situated between 700-900 m above average sea level (ASL) and found in dry evergreen forests. However, most solitary males and a herd of males were found 400-700 m above the PKWS. The MaxEnt habitat model was similar to the minimum convex polygon (MCP) and 95% KDE. The model showed that wild Asian elephants used a wide range of habitats (Table 5 and Fig. 6). The total area of 95% of the MCP was 1,098 km². The whole area inside the PKWS using 95% of the KDE was 1,554.9 km².

Population home range

The population of herd 3 was the largest of the seven female herds at 72 km² (Table 5 and Fig. 3D). The second largest was herd 2 at 67.5 km² (Fig. 3C). The smallest was herd 7 at only 4.5 km² (Fig. 3H).