Suitable habitat of wild Asian elephant in Western Terai of Nepal

Abstract Background There is currently very little available research on the habitat suitability, the influence of infrastructure on distribution, and the extent and connectivity of habitat available to the wild Asian elephant (Elephas maximus). Information related to the habitat is crucial for conservation of this species. Methods In this study, we identified suitable habitat for wild Asian elephants in the Western Terai region of Nepal using Maximum Entropy (MaxEnt) software. Results Of 9,207 km2, we identified 3194.82 km2 as suitable habitat for wild Asian elephants in the study area. Approximately 40% of identified habitat occurs in existing protected areas. Most of these habitat patches are smaller than previous estimations of the species home range, and this may reduce the probability of the species continued survival in the study area. Proximity to roads was identified as the most important factor defining habitat suitability, with elephants preferring habitats far from roads. Conclusions We conclude that further habitat fragmentation in the study area can be reduced by avoiding the construction of new roads and connectivity between areas of existing suitable habitat can be increased through the identification and management of wildlife corridors between habitat patches.

the wild Asian elephant are changes in the habitat and reduction in its suitable habitat, and these are caused by increased human activities (Zhang & Wang, 2003). Human expansion transforms natural habitats of wildlife into human settlements and agricultural lands (Cordingley, 2008;Hoare, 1999). Forests outside the protected areas have suffered extensive exploitation, due to the demands of human populations living along the fringe of the forest (Pradhan et al., 2011).
This study explored on how these threats are likely to impact current populations of elephants and the extent and connectivity of suitable habitat both inside and outside protected areas. Research into these factors is therefore crucial to ensuring the species continued survival within the country. The study identified the important habitat parameters and environmental variables within topographic, vegetation related, and anthropogenic category that determine suitable wild Asian elephant habitat in the Western Terai region of Nepal.

| Study area
The study was conducted in the Banke, Bardia, Kailali, and Kanchapur districts of western Nepal, with a total area of 9,207 km 2 (Figure 1).
Protected areas within the study site are Banke National Park and its Buffer Zone, Bardia National Park and its Buffer Zone, Shuklaphanta National Park and its Buffer Zone and Krishnasar Conservation Area (DNPWC, 2017). National parks belong to II and conservation area and buffer zone belong to VI according to Protected Area Categories System of International Union for Conservation of Nature (iucn.org).
Entry without permission of park authority is prohibited in national parks, but reasonable entry is accepted for local people for their daily activities in buffer zones and conservation area. The lowland Terai of Nepal is an area of high biodiversity and significant con-

| Elephant occurrence points
Occurrence points of wild Asian elephant were collected between September 2017 and March 2018. We first held discussions with officials responsible for protected areas in the region to identify F I G U R E 1 Study area in Nepal potential habitat of elephants and visited identified areas from these discussions to record evidence of elephant presence. Elephant presence was collected through direct observation of individuals, as well as indirect observation of tracks and droppings. We also used secondary sources of elephant occurrence records, previously recorded observations (GPS points) by park authorities, in each of the protected area site offices. We collected a total of 76 records (GPS points) of elephant presence during data collection.

Topographical variables
Digital elevation model (DEM) data of 30 m resolution were downloaded from the United States Geological Survey website (https:// earth explo rer.usgs.gov/), and the slope was computed from the DEM using ArcGIS software (ESRI, 2017). Shapefiles of water sources were downloaded from Geofabrik website (https://www.geofa brik. de/data/shape files.html) and converted to distance raster file using ArcGIS (ESRI, 2017). Elevation was used as a proxy of temperature due to the unavailability of high-resolution climatic variables.

Vegetation-related variables
Herbivores are depended on vegetation-related variables (Andersen et al., 2000). The elephant is a mega herbivore, so the inclusion of vegetation-related variables to predict suitable habitat for this species is a prerequisite for robust habitat modeling.
For the variable "forest cover," we used data prepared by Hansen et al. (2013) (Jönsson & Eklundh, 2004), to reduce cloud cover in Environment for Visualizing Images, a software of image analysis, and the EVI values were averaged over all the indices in order to obtain the final EVI index.

Anthropogenic variables
Human activities have been identified as a threat to wild Asian elephants and influence the species distribution (Choudhury et al., 2008;DNPWC, 2012). We, therefore, incorporated anthropogenic variables into our model. Anthropogenic variables were the distance to human paths (used by human and animal) and roads (used by vehicle), distance to settlements, and land use. Location of paths and roads was obtained from shapefiles available on the Geofabrik website (https://www.geofa brik.de/data/shape files. html). Settlement locations were obtained from the Department of Survey, Nepal. Distance raster files of paths, roads, and settlements were created using ArcGIS (ESRI, 2017). Land cover and land use (LULC) data were downloaded from the International Centre for Integrated Mountain Development website (ICIMOD; http://www. icimod.org) (Uddin et al., 2015) and incorporated into the model.

| Prediction of distribution of the wild Asian elephant
MaxEnt is a software package used to model species distributions using geo-referenced occurrence data and environmental variables to predict suitable habitat for a species (Phillips, Anderson, & Schapire, 2006). This software extracts a sample of background locations that it contrasts against the presence locations and estimate the density of presences across the landscape (Merow, Smith, & Silander, 2013;Phillips et al., 2006). We incorporated the variables listed in Table 1 into MaxEnt (version 3.4.1) along with our occurrence data to determine habitat suitability for wild Asian elephants within our study area. The MaxEnt program is widely used to map wildlife habitat and identify the influence of environmental variables on species occurrence in similar study areas Bista, Panthi, & Weiskopf, 2018;KC et al., 2019;Panthi, 2018;Panthi,  Multicollinearity between environmental variables described in Table 1 is acceptable (|r| < .70) (Dormann et al., 2013), so we used all variables in the model. We maintained at least 1 km distances between species presence points to lessen spatial autocorrelation. We Accuracies of the model were accessed by two methods: threshold independent and threshold dependent. In the threshold independent method, the value of accuracy was directly obtained from the model, but in the threshold dependent method, we provided the threshold to maximize the sum of specificity and sensitivity. We used the area under the receiver-operator curve (AUC), which is automatically calculated during the modeling without using any threshold. An AUC < 0.7 denotes poor model performance, 0.7-0.9 denotes moderately useful model performance, and >0.9 denotes excellent model performance (Pearce & Ferrier, 2000).

TA B L E 1 Environmental variables considered in the model
We chose true skill statistics (TSS) as the threshold dependent method.
The TSS = Sensitivity + Specificity − 1 and ranges from −1 to 1, where values less than 0 indicate a performance no better than random and 1 indicates a perfect fit (Allouche, Tsoar, & Kadmon, 2006). We calculated TSS for all 10 model outputs in R software (R Core Team, 2018), and the final TSS was averaged from all ten replications (Bista et al., 2018;Jiang et al., 2014;Panthi, 2018). For species distribution models, presence-only data threshold to maximize the TSS is recommended (Liu, White, & Newell, 2013); so, we used this threshold to convert the continuous habitat suitability map to a suitable/unsuitable binary map.

F I G U R E 3
Importance of variables to train the model. The regularized training gain explains how much better the model distribution fits the presence data relative to a uniform distribution. "With all variables" indicates the results of the model when all variables are run; "with only variable" denotes the results of the model when an only that variable is run; and "without variable" denotes the effect of removing that single variable from the model (Phillips, 2017). See Table 1 for full variable names and descriptions

| The suitable habitat of the wild Asian elephant
We identified a total of 3,194.82 km 2 as suitable habitat for wild Asian elephant in the study area ( Figure 2). About 39.11% (1,249.58 km 2 ) of this habitat occurs in existing protected areas ( Elephant habitat in the study area was highly fragmented, occurring as small, discrete patches. Connectivity between habitat patches was low in the southern and northern parts of the study area, but higher in the center (Figure 2).

| Important environmental variables
Of 10 variables used in the model, the distance to road, distance to water, elevation, and slope were found to be the most important variables determining habitat suitability. Distance to settlement, and mean EVI and LULC were identified as the least important variables (Figure 3).
In Figure 3, the regularized training gain of the model without distance to road was less than that of the model using without other single variables, so the distance to road is a more useful variable to the model.
Similarly, the regularized training gain of the models without distance to water, elevation, and slope is less, indicating that these variables are useful predictors of habitat suitability for the species.
The model, therefore, indicates that elephants prefer habitat far from roads, near to water sources, with low elevation and gentle slope (Figure 4).

| Model accuracy
Accuracies of the model are relatively good. We obtained 0.813 ± 0.072 AUC and 0.528 ± 0.031 TSS (Table 3). We obtained 0.214 threshold to maximize the sum of sensitivity and specificity.
We used this threshold to calculate the TSS and to covert the continuous habitat suability map to binary suitable/unsuitable map.  (Lamichhane et al., 2017;Neupane, Kwon, Risch, Williams, & Johnson, 2019;Pradhan & Wegge, 2007). Our result reveals most habitat (1,249.58 km 2 or 39.11% of the total study area), located inside the protected areas where natural vegetation cover exists. Our results agree with a previous study in India which identifies the importance of natural vegetation cover to provide suitable habitat for Asian elephant (Kumar et al., 2010).

| D ISCUSS I ON
The home range size of an elephant was estimate 105-320 km 2 in India (Sukumar, 1989). Three national parks within our study area (Banke National Park and its Buffer Zone, Bardia National Park and its Buffer Zone, and Shuklaphanta National Park and its Buffer Zone) contain a total area of suitable habitat larger than the Sukumar (1989) home range estimate ( Table 2). Habitat of Asian elephant is being fragmented in China (Zhang et al., 2015). Similarly, we also find this fragmentation in our study area. Although African elephants spend much of their time in less fragmented landscapes (Gara et al., 2016), Asian elephants have been shown to continue to occur in areas with fragmented habitat (Kumar et al., 2010). Therefore, the habitats identified in these three national parks may function as significant refuges for elephants despite the fact that they occur as fragmented patches. Kailali district contains more area of suitable habitat (942.55 km 2 ) outside protected areas although this district includes no protected area. Fragmented forests are more serious to human-wildlife conflict (Acharya, Paudel, Jnawali, Neupane, & Köhl, 2017). The connectivity of habitat patches in the central parts of the study area means that they have the potential to be managed as corridors to increase the likelihood of elephants moving between habitat patches and mitigate conflict between elephants and humans. According to our model, distance to roads was found to be the major component of habitat suitability of the wild Asian elephant.

| CON CLUS IONS
This study identified more than 3,000 km 2 of area as the suitable elephant habitat in the Western Terai region of Nepal. Around 40% of suitable habitat is covered by existing protected areas. Although there is large suitable habitat, the majority of suitable habitat occurs in small, discrete patches insufficient to accommodate the large resource requirements of the species. To increase connectivity between these patches, we recommend protecting existing habitat to provide corridors between Bardia National Park and Shuklaphanta National Park. The future road projects should consider the movement of wild Asian elephant and design accordingly.

ACK N OWLED G M ENTS
This study was possible due to research permission of Department of National Parks and Wildlife Conservation (DNPWC). We thank DNPWC, Banke National Park Office, Bardia National Park Office,

Krishnasar Conservation Area Office, and Shuklaphanta National Park
Office for their support. We thank Dr. Bhogendra Mishra (Science Hub, Kathmandu, Nepal) for his contribution in reviewing the manuscript.

CO N FLI C T O F I NTE R E S T
None declared.