Introduction

The persistence of an animal species in a human dominated landscape is often linked to the existence of protected refuge within the landscape as well as the human tolerance of the species using areas outside protected areas1,2. Protected areas are often essential, but for species that roam over large areas protected areas may not be sufficient to guarantee species persistence. Human tolerance of animal species is itself linked to local culture, economic conditions, and conservation efforts, especially efforts to mitigate human-wildlife conflict3. These underlying social factors all vary in space and can explain why species persist in human dominated landscapes in some portions of their geographic range and disappear from similar landscape in other portions of their geographic range4,5. These social factors can also change over time facilitating large scale range expansions or shrinkage6. The role of human tolerance in determining animal persistence is particularly pronounced in large mammal species, like Asian elephants (Elephas maximums)3,7.

Asian elephants are ecologically and culturally important to humans, however, they frequently come into conflict with humans. Elephants appear prominently in multiple Asian religions and elephants have played significant economic and military roles for millennia8. Asian elephants, allowed to roam in natural areas, act as ecological engineers and seed dispersers9. At the same time, elephants are frequently targeted by poachers and their habitat is increasingly being lost and fragmented by human development10. Asian elephants require large areas to fulfill their life-history requirements and large blocks of habitat required to meet their forage and water requirements have become rare in much of Asia11,12,13. Over time, human-elephant conflict (HEC) has increased throughout much of Asia, frequently leading to economic losses, loss of human life and retaliatory killings of elephants14,15,16. In many parts of the Asian elephant’s geographic range elephant populations are confined mainly to protected areas17,18. Given their cultural and economic significance and the intense pressure put on Asian elephants it is not surprising that they are listed as ‘Endangered’ in the IUCN Red List of threatened species19 and that there are ongoing conservation efforts to mitigate human-elephant conflict, especially in the lowlands of Nepal, an area known as the Chure Terai Madhesh Landscape (hereafter CTML)20,21.

The CTML has undergone significant forest loss and fragmentations over the last century but still supports a small Asian elephant population and contains many protected areas that have acted as refuges. Approximately 22% of elephant habitat in the CTML was lost between 1930 and 2020 due to deforestation and development into agricultural lands. At present, 19,000 km2 forest cover is available as elephant habitat in Nepal22 and there are approximately 400 elephants (~ 200 wild elephants and another ~ 200 captive elephants) across the lowlands of Nepal23. Studies elsewhere in the range of Asian elephants suggest a negative relationship between extent of human disturbance and the probability elephants used a portion of the landscape24 and an earlier study in the CTML suggested that Asian elephant used areas with fewer streams and greater protection (i.e., protected areas and national parks as opposed to community forests)25. While the positive relationship with protection was expected, the relationship with streams was not expected a priori. A posteriori, Lamichhane et al.25 hypothesized that increasing HEC in wetter areas might explain the avoidance of wetter areas by Asian elephants. Lamichhane et al.25 relied on a single season occupancy analysis of sign survey data collected throughout the forested portions of CMTL in 2012.

Here we build on Lamichhane et al.25 , by analyzing these data and subsequent survey data collected throughout forested portions of the CMTL in 2018 and 2020 within a dynamic occupancy modelling framework26,27. Between 2012 and 2020, conservation within the CMTL focused on managing within protected areas and decreasing HEC outside protected areas. The objectives or our study are to: (1) determine whether elephant landscape use changed over the study period, and (2) test hypotheses regarding potential drivers of change in landscape use. A dynamic occupancy modelling framework allows us to test hypotheses regarding initial landscape use (in 2012), colonization (defined here as use of an area in time t + 1 that was not used in time t), extinction (defined here as not using an area in time t + 1 that was used in time t) and detection probability (i.e., the probability of detecting Asian elephants given that they use the area). We only tested a limited set of hypotheses regarding initial use and detection that were directly informed by the prior analysis and instead focused our analysis primarily on hypotheses regarding colonization and extinction. Specifically, we hypothesized that: (1) Colonization (extinction) would be higher (lower) in protected areas, (2) colonization (extinction) probabilities would be lower during the second interval (because it was only 2 years as opposed to 6 years), (3) colonization (extinction) would be lower (higher) in wetter areas (i.e., areas with more streams that are used more by humans), (4) colonization (extinction) would be lower (higher) in areas with more roads (because humans use these areas more), and (5) colonization (extinction) was higher (lower) in areas with rougher terrain because humans use these areas less. We operationalized these hypotheses as a series of models.

Results

Our model selection strategy identified a best model with protected area effects on initial occupancy, colonization, and detection probability as well as temporal variation in colonization and detection probability (Table 1). The best model did not include any of the covariates representing hypothesized environmental drivers. The best model was > 8 AIC better than the next best model that did not include uninformative parameters suggesting strong support for the best model and the best model was responsible for the vast majority of the Akaike weight within the model set (w = 0.969). The two other models with Akaike weights greater than 0.01 included the same protected area effects and the temporal variation in detection probability but did not include temporal variation in colonization and in one case included temporal variation in extinction probability. Based on the high support for the best model, we focused additional inferences solely on this model.

Table 1 Models fit in this study (and associated model selection values) to test hypotheses regarding Asian Elephant use of the Chure Terai Madhesh Landscape (CTML) in Nepal between 2012 and 2020.
Full size table

The modelled elephant use of forested areas in the CTML increased from 0.42 (SE: 0.01) in 2012 to 0.68 (SE: 0.02) in 2018 and 0.70 (SE: 0.01) in 2020 (Fig. 1). These increases were consistent with naïve estimates of use that do not account for detection probability and increased from 0.40 (65 of 159 surveyed grid cells) in 2012, to 0.61 (97/159) in 2018, to 0.66 (105/197) in 2020. Initial use, colonization and detection probabilities were all higher in protected areas than in non-protected areas (Table 2). Colonization in the first interval was 0.94 (SE: 0.08) in protected areas and 0.51 (SE: 0.06) in non-protected areas as compared to 0.77 (SE: 0.26) and 0.18 (SE: 0.08) during the second, shorter interval. Extinction probability was estimated to be constant at a rate of 0.09 (SE: 0.03) during both intervals. Detection probabilities were lower in both the second and third surveys (Table 2). Whereas elephants were patchily distributed in the first year of surveys the distribution of grid cells increasingly consolidated into a western and eastern block in subsequent years, with the only gap in their distribution occurring between Chitwan and Bardiya National Parks (Fig. 2).

Figure 1
figure 1

Estimated use of protected and non-protected sites in the CMTL over the study period. Points represent means and whiskers represent the 95% confidence intervals.

Full size image
Table 2 Model coefficients (with standard errors in parentheses) from the best model in our model set estimating Asian Elephant use of the Chure Terai Madhesh Landscape (CTML) in Nepal between 2012 and 2020. Definition of symbols: β0 - intercept; P - protected area effect; β2 - difference between second interval and first interval or second sampling occasion and first sampling occasion; β3 - difference between third sampling occasion and first sampling occasion.
Full size table
Figure 2
figure 2

Elephant landscape use in Nepal; (a) 2012, (b) 2018, and (c) 2020 illustrating the increasing consolidation of use into a large eastern and western block. Map was generated by author NKY using ArcGIS 10.556.

Full size image

Discussion

Colonization was higher in protected areas and lower during the second, shorter interval, consistent with two of our a priori hypotheses. We did not find support for a priori hypotheses that included environmental conditions as drivers of colonization (e.g., stream length, road density, or terrain roughness) and no hypotheses regarding extinction were supported. The lack of support for any hypotheses regarding extinction may be related to the overall low rates of extinction during the study period. While our inferences were primarily focused on occupancy dynamics, we also tested a limited set of models describing initial use. Both protected areas and stream length were supported as correlates of use in a priori study25 that relied on the same survey data projected on a finer grid scale (i.e., 10 km × 10 km as opposed to 15 km × 15 km), however the analysis presented here found stronger overall support for an effect of protected areas on the distribution of use in 2012.

Our analysis of three surveys covering Nepal’s entire elephant range found that forest elephant landscape use increased substantially during a period of eight years (2012–2020) with increasing use of both protected areas and forested portions of the landscape outside protected areas (Figs. 1, 2). Increasing use of the overall landscape could be associated with an increasing elephant population, however trends in overall abundance over the course of the study are unknown. Future work addressing this uncertainty in interpretation of use trends could be useful. Increasing use of the overall landscape may also be linked to decreased conflict with, and increased tolerance of, elephants50. While colonization and initial use were both higher in protected areas, areas outside protected areas saw the largest increase in probability of use. This pattern occurred because as elephant use saturated in protected areas, the absolute difference in use between protected areas and outside protected areas declined, an observation that is consistent with theoretical and empirical expectations of how habitat associations are expected to change under range expansion48.

While detection probabilities are often thought of nuisance parameters, they can reflect differences in relative density or intensity of use49. We found that detection probabilities were higher in protected areas but declined over the course of the study. These differences might suggest higher densities of elephants in protected areas as well as declining overall densities within occupied grids in later years. A decline in densities does not suggest that the overall population has declined and could occur if the overall abundance of Asian elephants in the study area was relatively constant, but just spread out over a larger area. Alternatively, there may have been subtle differences in environmental conditions (e.g., more rain) in later survey years that led to quicker degradation of elephant signs and lower detection probabilities despite similar densities. Thus, future work focused specifically on abundance estimation could help to address uncertainties in detection probabilities as well as use.

Elephant use in 2012 was restricted to a series of patches, however in later years use was extended to two large areas. In 2018 and 2020, use was continuous from Central to Eastern Nepal indicating the habitat connectivity of elephant sub-populations of eastern and central Nepal, between Koshi Tappu WR and Chitwan-Parsa NPs51. Similarly, elephants continuously used the Western part of Nepal (Fig. 2). Population studies that occurred over a decade ago suggested four sub-populations of elephants in Nepal52,53, however our results indicate increased landscape connectivity leading to two larger subpopulations. One in the east located between Chitwan and Jhapa and another to the West of Kapilvastu. The first author has observed movement of elephant bulls from Koshi Tappu Wildlife Reserve in Eastern Nepal to Parsa National Park in central Nepal in 2022 and vice versa in 2023, further supporting the hypothesis of increased landscape connectivity. The increasing elephant population in core habitats23 and restoration of forest patches in Churia foothills54 may have contributed to the increased landscape use across the CTML and could be a subject of future studies. Increasing elephant use outside protected areas may increase HEC and management interventions that reduce HEC are likely to be important to maintaining human-elephant co-existence in both protected areas and outside protected areas55.

Methods and materials

Study area

The Chure Terai Madhesh Landscape (CTML) covers an area of 42,456 Km2 in southern Nepal (Fig. 3) and includes five National Parks (NP), one Wildlife Reserve (WR), one conservation area and the Terai Arc landscape (TAL) of Nepal28,29. The altitude in the CMTL ranges from 60 to 1500 m above sea level and major ecosystem include: (a) Himalayan subtropical broadleaved forest, (b) Gangetic plains and moist deciduous forest and (c) Terai-duar savannas20. Temperatures range from 14 to 40 °C30 and annual rainfall varies from 1138 to 2,680 mm with over 80% of the rain occurring during monsoon months31. The CTML is densely populated with an average human density of 390 persons/km2 29 and sixty percent of the people depend on subsistence agriculture20. In contrast, there are ~ 200 wild elephants in the CMTL23 and transborder movement of elephants are frequent along the Nepal–India border. Population trends in wild elephants over the study period are unknown.

Figure 3
figure 3

Study area map showing Nepal’s Chure Terai Madhesh Landscape (CTML), protected areas, and survey grids (15*15 km2). Map was generated by author NKY using ArcGIS 10.556.

Full size image

Survey design

We used a grid-based approach to design our survey of elephants across CTML30. We divided the CTML using grid cells that measured 15 km × 15 km which approximates the average home range of forest elephants in these environment conditions32,33,34. We identified 159 grid cells that were forested or adjacent to forested grids and surveyed all of these grids. Within each cell, 45 km (5 transects of 9 km length and ~ 5-m width) was surveyed following linear features (e.g., trails, stream beds, etc.). Each transect was treated as a spatial replicate 30. Use of spatial replicates is appropriate because elephants have a wide range and leave signs that persist for the survey period 35,36. We conducted two field surveys in 2017/2018 (herafter 2018 survey) and 2019/2020 (hereafter 2020 survey) during the cool-dry season (November–February) when elephant signs are likely to persist for many days. Elephant signs (dung, footprints, elephant trails, feeding signs, and evidence of tree peeling, and body rubbing marks) were recorded. If no signs were detected along a transect the transect was coded as a 0 in the detection history and if any signs were detected on a transect the transect was coded as a 1 (i.e., elephant use detected). We combined data from these surveys with an earlier survey conducted in November 2011–February 201237 (hereafter 2012 survey). The 2012 survey was conducted with similar protocol but at spatial scale of 10 × 10 km2. These survey data were transformed to 15 × 15 km2 by mapping the start and end points of the individual transects from the original finer scale (10 × 10 km2 grids) on the coarser scale (15 × 15 km2 grids). Those transects which entirely fell within a coarse scale grid were treated as full transects at the coarser scale. If a transect fell within two grids, the transect was split between the two grids and intersection of the transect and grid border was considered the start or end point for that transect. If more than 5 full transects fell within a coarse scale grid, we discarded additional partial transects. As data were collected at regular 1 km intervals along each transect it was possible to assign detections to transects split by the coarse spatial scale. All surveys were conducted by wildlife biologists and trained wildlife technicians with over five years of field experience in wildlife research and capable of identifying elephant signs from other taxa in the region. During each survey clear signs of elephant use were identified which usually consisted of footprints or dung.

Covariate calculation

We initially considered nine potential covariates for our analysis: (i) Management type—protected/non-protected areas, (ii) Seasonal water area, (iii) Permanent water area, (iv) stream length, (v) road length, (vi) human population density, (vii) terrain roughness index, (viii) Proximity to human settlements and (ix) Mean Perimeter Area Ratio (MPAR, a fragmentation matrix calculated as sum of all perimeters divided by the total area of forest patches). After multicollinearity test, we removed highly co-related (r > 0.7) variables and identified four covariates for the occupancy analysis i.e. (1) Management type – protected/non-protected areas (P); (2) stream length (S); (3) terrain ruggedness index (H); and (4) road length (R). Stream lengths (km) in each grid was derived from digital topographic maps of 1:25,000 scale produced by Survey Department of Nepal between 1995 and 1996 (https://dos.gov.np/). Similarly, road length (km) was calculated from the open-street maps (major roads) of 2020 (https://www.openstreetmap.org). The grids were overlaid with Protected Areas layer and codded 1 (PA) if the grid centroid fell within protected area and coded 0 (non-PA) if the centroid fell outside. The terrain ruggedness index value was calculated from a digital elevation model (SRTM 30 m resolution) and averaged for respective grid.

Dynamic occupancy modelling and model selection strategy

To test hypotheses regarding elephant’s landscape use dynamics, we fit multi-season (dynamic) occupancy models using the Presence software38,39,40,41. We used the parameterization based on initial use (ψ), colonization (γ), extinction (ε), and detection probability (p). We defined colonization probability as the probability that elephants use a site in time t + 1 given that it was not used in time t, and extinction probability as the probability that elephants do not use a site in time t + 1 given that it did use the site in time t 42,43,44.

We used prior research on single season (static) elephant habitat use in Nepal to inform the overall set of models we considered, and we used a secondary candidate set strategy to explore this overall set of models efficiently45,46. Specifically, we defined a base model based on the best model identified by Lamichhane et al. (2018) in which initial use was a function of stream length, detection was a function of whether the center of a cell fell within a protected area, and colonization and extinction were constant (Lamichhane et al. 2018). We then fit four primary groups of models in which the base structure was used for all parameters except the focal parameter and a series of hypotheses regarding the focal parameter were tested. For the ψ group, we tested whether stream length only, stream length and protected area status, protected area status only or a constant model best described use since these models were all competitive in Lamichhane et al. (2018). For the p group, we include protected area status in all models (based on strong support for this hypothesis in Lamichhane et al. (2018) and tested whether differences between intervals (t), variation in effort, interval and effort, or no additional variation were supported. For the colonization and extinction groups, we decided a priori that we would fit five univariate models first and then consider any multivariate models based on combinations of informative covariates identified from the univariate models (i.e., we used a step-forward process within these groups). The five univariate models were based on the five hypotheses (last paragraph of the introduction section). After fitting these primary model sets, we constructed our secondary model set as the combination of all competitive sub-models (i.e., by considering all models in each primary group that were within 5 AIC of the best model and included informative parameters46,47.

Research approval and ethics

We obtained research permissions from the Department of National Parks and Wildlife Conservation Nepal (Ref no: 3066/073/74; June 02, 2017). Our research was based on non-invasive survey of elephant signs and it didn't carry out any experiments with live animals. We only carry out stakeholder consultation, presence absence survey from the study area and collected their respective locations.