Understanding species distribution patterns has direct ramifications for the conservation of endangered species, such as the Asian elephant Elephas maximus. However, reliable assessment of elephant distribution is handicapped by factors such as the large spatial scales of field studies, survey expertise required, the paucity of analytical approaches that explicitly account for confounding observation processes such as imperfect and variable detectability, unequal sampling probability and spatial dependence among animal detections. We addressed these problems by carrying out ‘detection—non-detection’ surveys of elephant signs across a c. 38,000-km2 landscape in the Western Ghats of Karnataka, India. We analyzed the resulting sign encounter data using a recently developed modeling approach that explicitly addresses variable detectability across space and spatially dependent non-closure of occupancy, across sampling replicates. We estimated overall occupancy, a parameter useful to monitoring elephant populations, and examined key ecological and anthropogenic drivers of elephant presence. Our results showed elephants occupied 13,483 km2 (SE = 847 km2) corresponding to 64% of the available 21,167 km2 of elephant habitat in the study landscape, a useful baseline to monitor future changes. Replicate-level detection probability ranged between 0.56 and 0.88, and ignoring it would have underestimated elephant distribution by 2116 km2 or 16%. We found that anthropogenic factors predominated over natural habitat attributes in determining elephant occupancy, underscoring the conservation need to regulate them. Human disturbances affected elephant habitat occupancy as well as site-level detectability. Rainfall is not an important limiting factor in this relatively humid bioclimate. Finally, we discuss cost-effective monitoring of Asian elephant populations and the specific spatial scales at which different population parameters can be estimated. We emphasize the need to model the observation and sampling processes that often obscure the ecological process of interest, in this case relationship between elephants to their habitat.
Citation: Jathanna D, Karanth KU, Kumar NS, Karanth KK, Goswami VR (2015) Patterns and Determinants of Habitat Occupancy by the Asian Elephant in the Western Ghats of Karnataka, India. PLoS ONE 10(7): e0133233. doi:10.1371/journal.pone.0133233
Editor: Sadie Jane Ryan, University of Florida, UNITED STATES
Received: May 7, 2014; Accepted: June 4, 2015; Published: July 24, 2015
Copyright: © 2015 Jathanna et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: This work was supported by a grant to the Centre for Wildlife Studies, Bangalore from Vision Group on Science and Technology, Government of Karnataka vide G.O. No. Vi Ya Ei 170 Vi Tra Ma 2010 dated 19-4-2011, and from Department of Biotechnology, Government of India vide No. BT/PR13854/BCE/08/809/2010 dated 10-1-2012. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
The Asian elephant (Elephas maximus Linnaeus, 1758), despite being a species of immense cultural, ecological and conservation value, continues to be threatened by the loss, fragmentation and degradation of habitat, conflict with humans and poaching for ivory [1–3]. Even within India––a country that harbors nearly 60% of the current global wild population––much of the species’ former range has been lost to agricultural expansion, human settlements and developmental projects in the past 2–3 centuries [2,4].Consequently, elephants in India persist today in small, insular populations, restricted largely to protected areas [5,6] within fragmented landscapes. Therefore, understanding factors that underlie elephant spatial distribution patterns is critical to prioritize habitat conservation, identify threats that limit elephant presence and to inform management actions to secure these populations over time. Notwithstanding this need, factors that drive distribution patterns and the actual area occupied by elephants in India are poorly studied and largely unknown. In contrast, several studies have addressed the question of what determines the distribution of African elephants (e.g. [7–29]). The understanding thus gained on the ecology and conservation status of elephants in Africa has long informed their intensive management .Not only are such data scarce in the case of Asian elephants (but see[31–38]), what is known about their distribution patterns has been generated using varied field methods (dung encounter rates, direct observation, telemetry-based locations) and diverse analytical approaches (comparisons of summary statistics, correlations, multivariate analyses, presence-only modeling), none of which accounts for observation processes such as sample selection bias, imperfect and variable detectability[39–41]. Because detectability can vary with habitat, ignoring it could seriously bias inferred occupancy-habitat relationships[41–45].A recent study on Asian elephant occupancy that did account for detectability  found that estimated detectability ranged from 0.08 to 0.83, depending on covariate values and intensity of use by elephants, indicating that detection is both imperfect and highly variable. Moreover, most surveys of Asian elephant distribution were not conducted at appropriate spatial scales, given their wide-ranging behavior (see[47,48]for estimates of Asian elephant home ranges).
At the local scale, past efforts to monitor Asian elephant populations have focused on estimation of population size and density. However, regional, national and range-wide estimates are based on indefensible extrapolation of such local-scale (i.e. individual reserve) estimates that vary widely in their reliability. Blake and Hedges justifiably worry that beyond locations and relative abundances of some of these elephant populations, we know very little. Not only have unreliable field and analytical methods been applied to estimate elephant numbers, there is also a critical mismatch between the population parameter being estimated (i.e. abundance or density) and the spatial scale of such estimation effort. While abundance and density can be reliably estimated at smaller spatial scales (e.g. individual reserves) using appropriate methods[40,49],estimates of abundance or density at larger scales tend to be unreliable (see).
The appropriate metric for landscape level studies in wide-ranging species such as elephants is the estimation of habitat occupancy: the probability (Pr) that a species is present within a site(see[44,50]). Formal occupancy modeling explicitly accounts for imperfect detectability and other observation processes, which, if ignored, could lead to biased inferences of occupancy, habitat relationships and temporal changes in these. The approach also permits modeling of ecological and anthropogenic effects on detection probabilities as well as habitat occupancy (or both). This removes biases from heterogeneity among sites, in either detectability or occupancy, as well the confounding of true patterns in occupancy with detectability and sampling probability[41,44]. Consequently, rigorous occupancy modeling is rapidly emerging as a method of choice. It has been successfully used on other wide-ranging species, such as the tiger Panthera tigris .
In this study, we used habitat occupancy sampling of Asian elephants over a large and diverse landscape in the Western Ghats of Karnataka State in India. We employed a modeling approach recently developed by Hines et al.  that relies on spatial rather than temporal replication and uses surveys of animal signs, which has been successfully used to investigate the distribution of tigers  and dhole Cuon alpinus  based on the same survey effort. Our estimates provide a useful baseline to monitor future changes in spatial distribution of the largest wild population of Asian elephants. We also elucidate the role of key ecological and anthropogenic covariates as determinants of elephant habitat occupancy. We demonstrate how this approach can improve future management and monitoring of Asian elephants.
Materials and Methods
This study only involved non-invasive surveys of elephant signs, for which the Karnataka Forest Department provided field research permits.
We conducted our surveys across a c. 38,000-km2 landscape (Fig 1), with our sampling frame defined by the presence of natural vegetation types including forest (evergreen, moist deciduous, dry deciduous, thorn scrub), tree plantations (e.g. Tectona grandis, Acacia auriculiformis, Eucalyptus spp. Casuarina equisetifolia, among others), tree savanna, shrub savanna, grassland (in our landscape, primarily montane grasslands) and uncultivated revenue department or private lands. Thus, other than a few areas where small numbers of itinerant and conflict-prone elephants subsist in heavily human-dominated areas (e.g. in Tumkur and Hassan districts) and which have been identified for removal , our field surveys covered the entire distributional range of elephants in the state of Karnataka. The surveyed areas encompassed the full gradients in rainfall, land cover and intensity of human use.
Inset shows location of the landscape within India.
The altitude in the study area peaks at 1927m, with the terrain abruptly rising from the coastal plains and then descending from the ridge of the mountains eastwards, intergrading into the Deccan Plateau. The mean annual rainfall declines from well over 5000 mm in the coastal plains and crest of the Ghats to ~600 mm eastwards . Western slopes of the Ghats support wet evergreen rainforests and montane shola-grassland habitats; semi-evergreen and moist and dry deciduous forest types dominate areas eastward, where rainfall declines progressively. These blocks of natural forests are interspersed with or abut areas under varied forms of agriculture, horticulture and plantations as well as human settlements. We defined elephant habitat as comprising all the natural vegetation types described above, which together cover 21,167 km2 of the study landscape , and support some of the largest populations of elephants as well as other threatened large mammal species typical of the region [56,57].
The landscape includes 14 wildlife reserves that cover an area of 5,500 km2 and lie in a human-dominated matrix populated by >10.2 million people. The elephant habitats are subject to threats from illegal hunting, livestock grazing and forest biomass extraction as well as pressures from developmental projects and industrial growth . For detailed accounts of the study area see [51,56,58,59].
Occupancy survey design
Our goal was to estimate the proportion of the available potential habitat occupied by elephants. Thus defined, defined, ‘habitat occupancy’ corresponds to second-order habitat selection (selection of home ranges within the species’ distributional range) under the hierarchical framework proposed by Johnson . Our surveys were specifically designed for tigers  therefore the study design, field survey protocols and analytical approach used were planned considering tiger biology. We carefully reviewed these elements of our study in the context of surveying elephant occupancy to ensure that the modeling approach did indeed describe our observation process (variably imperfect detection of signs on successive segments along trails within specified grid cells) overlaid on the ecological process of interest (true elephant distribution, determined by the values of one or more covariates). We also carefully reviewed the literature on elephant home range size to confirm that the same survey design would be appropriate for Asian elephants. Sukumar  estimated home ranges of two elephant families at 105 km2 and 115 km2, based on repeated observations of recognizable individuals. Further, Fernando et al. reviewed a large number of Asian elephant home ranges estimates and the 18 estimates of elephant home ranges in south India in their Table 1 average to 275.8 km2 (minimum: 105 km2 ; maximum: 800 km2 ). When considering that a) these studies were conducted over periods ranging from nine months to nearly six years, and also b) what is of relevance to our survey design is ranging patterns over the dry season (October-May) each year, we are confident that our definition of sites as grid cells of size 188 km2each (Fig 1) is appropriate. Because our grid cells are larger than expected mean elephant home range sizes over the dry season, we are also confident that a) our estimates of ψ pertain to true pr(occupancy) rather than pr(use|occupancy) and that b) spatial autocorrelation in elephant occupancy among grid cells is minimized to the extent possible. The grid cells also coincided with graticules on the 1:50,000 topographic maps, facilitating planning of field surveys. Field surveys were not carried out in grid cells with natural habitat fragments totaling <10km2 since elephants and other large-ranging mammals were unlikely to occupy these cells. Of the 232 grid cells covering the overall landscape, we excluded 27 cells using this forest cover-based exclusion criterion, yielding 205 cells (sites) to be surveyed.
Because of low elephant densities, low sighting probabilities and logistical factors, rather than sightings, we chose to survey and detect elephant presence via signs, primarily fresh dung . Although direct sightings, tracks and other evidences were also recorded, we based our analyses on fresh dung, since other signs have very different times of persistence (direct sightings are instantaneous), affecting both detectability and occupancy. Because of logistical practicality (see) we chose spatial  over temporal replication, to estimate detectability of elephant signs. To address the potential problem of spatially autocorrelated non-closure of occupancy state along such spatial replicates, we used the recently developed occupancy model of Hines et al. , as more fully described below.
The survey effort (distance walked) invested within each grid cell was set proportional to the extent of elephant habitat within the cell. Elephant habitat was defined based on forest cover, classified using a rule-based eco-climatic vegetation classification [56,63]. The sampling effort was set at 40 km for cells with 100% forest cover (188 km2), and, decreased in proportion to extent of available habitat. Field surveys ensured adequate spatial coverage of cells but surveyors’ subjective route choice was constrained by dividing the grid cell into 16 sub-cells and ensuring that the field teams traversed one sub-cell randomly selected prior to the surveys .
Prior to the surveys, 6 permanent staff and about 12 long-term volunteers were trained by one of the authors (NSK). We conducted a pilot training survey to familiarize the teams with field protocols. Four core teams, each led by a regional coordinator, covered different parts of the landscape. Field research teams surveyed forest trails typically used by elephants during the dry seasons (October-May) of 2005–2006 and 2006–2007. Dry season surveys ensured that dung persistence was consistent, reducing heterogeneity in detection probability induced by rainfall variation [64,65]. In addition to detection or non-detection of elephant dung, habitat covariate information was also recorded. The data were collected for each trail segment of 100 m, but were subsequently aggregated to 1-km spatial replicates (see ). To limit the number of parameters to be estimated, covariates of detectability were estimated only at the site level. The data were used to construct elephant sign detection histories for each cell, ‘1’ representing detection and ‘0’ non-detection [44,50]. Thus detections along trails were used to assess elephant occupancy at the site (= grid cell) level, and not at the level of each trail or trail segment. The number of replicates surveyed per cell, which depended on the proportion of habitat in it, ranged from 4–42 (also see).
Modeling elephant occupancy
Accounting and testing for spatial dependence in our data.
An important assumption of the MacKenzie et al.  model is that occupancy status does not change among replicates within each site. Our surveys, using spatial replication may not always have met this assumption, because elephant presence in one part (or replicate) within a cell (site) did not necessarily mean that other parts (or replicates) within that cell were also occupied. As shown by Kendall  and MacKenzie et al. , if such within-site changes in occupancy were to occur completely randomly, then replicate-level detectability is the product of the Pr(species is present in site and replicate) and Pr(species is detected | species is present in site and replicate), while the interpretation of Pr(site is occupied) changes to Pr(site is used). However, if between-replicate changes are non-random (e.g. Markovian, as when replicates are sampled sequentially along a trail rather than randomly from within a cell), detectability cannot be decomposed as above, and estimates of both detectability and occupancy will be biased. MacKenzie et al.  and Kendall & White  recommend sampling of spatial replicates randomly with replacement to ensure changes in occupancy within a cell are completely random. However, such an approach is logistically impractical to implement in field surveys of elephants. For this reason our elephant sign survey data were generated from sequential sampling of successive spatial replicates along trails walked by teams. To fully account for this factor, we chose to model occupancy using a recent approach developed by Hines et al. . This model explicitly accounts for spatial dependence in elephant presence at the replicate level as a first-order spatial Markov process. Parameters estimated from this model are ψ = Pr(site is occupied by elephants), θ = Pr(elephants present on replicate | site is occupied and elephants absent on previous replicate), θ′ = Pr(elephants present on replicate | site is occupied and elephants present on previous replicate) and p = Pr(detection at a replicate | elephants present both in the site and in the replicate). Because the first replicate could be located anywhere within each cell, we calculated probability of replicate-level presence for the first replicate θ0 as the weighted (by ψ) average of θ and θ′.
This practically useful occupancy modeling approach was earlier developed specifically for surveys of tigers in our study area (see[51,52] for details). Because elephants do not use forest roads and trails as intensively as tigers do, we were unsure if there was substantial spatial dependence in elephant presence among successive replicates. Therefore, we initially tested for possible existence of spatial dependence in replicate-level elephant presence by comparing the fits of the standard MacKenzie et al.  model and the Hines et al.  model to our survey data based on AIC values , and also by comparing the estimated values of θ and θ′. We did not assess fit of the models to the data  as there is currently no way to assess fit under the Hines et al.  modeling approach.
Modeling covariate effects on detectability and occupancy.
Another major objective was to assess the role of ecological and anthropogenic factors that possibly determined elephant occupancy in our landscape. This involved realistically modeling variation in sign detectability also, so that patterns in detectability were not wrongly attributed to patterns in occupancy. The plausible covariates that influenced detectability and/or elephant occupancy included data collected in the field surveys as well as remotely sensed data. Modeling of such covariates is important to ensure there is no un-modeled variation in Pr[occupancy] or detectability among cells [44,50]. More importantly, comparing likely covariate models enables examination of methodological (what influenced detectability) as well as important ecological or conservation questions (what determines elephant spatial distribution/occupancy). Based on our empirical experience, and, previous studies of Asian and African elephants (Loxodonta cyclotis Matschie, 1900 and L. africana Blumenbach, 1797), we identified covariates that we expected to influence probability of habitat occupancy or detectability, or both, for elephants in our study area (Table 1). We note that some of these covariates were expected to potentially influence both ψ as well as p. Influences on detectability included both direct (e.g. rainfall, which determines dung persistence) as well as indirect effects (covariates that affect elephant abundance in the site, which then affects species-level detectability). We first assessed collinearity between the potential predictors using scatter plots and Pearson’s correlation coefficients. We did not include in the same model covariates that were substantially correlated (Pearson’s |r|> 0.7; ). To limit the number of models in our candidate set, we first assessed the role of four potential covariates of p (see Table 1) while holding the covariate structure for ψ constant, following Karanth et al.  (see  for a discussion of such two-step analyses). Because we effectively conditioned on the covariate structure for occupancy while assessing covariates for detectability, we chose a moderately parameterized form for modeling occupancy in this step (Table 2). Based on AIC values of the different models of detectability (Table 2), we then fixed the covariate structure for p and assessed the influence of our covariates on ψ, also based on AIC values (Table 3).
(2010) modeling approach. No. of sites = 205. Please see Table 1 for descriptions of covariates.
(2010) modeling approach. No. of sites = 205. Please see Table 1 for descriptions of covariates.
In addition to additive effects of covariates on occupancy, we also included interactions between some of the covariates where we expected the influence of one covariate on ψ to depend on the value of another covariate. Pettorelli et al. [71,72] show that NDVI is an excellent index of vegetation productivity and list several examples of the successful use of NDVI to understand patterns of distribution and abundance of animals. We included a quadratic term for vegetation productivity in some models because we expected probability of elephant occupancy to exhibit a non-linear, peaked response from low (thorn scrub-tree savannah-dry deciduous) through medium (moist deciduous) to highly productive (wet evergreen) habitats. All covariates were modeled using the logit link function[44,50]. Covariate values were scaled as described in Table 1. We used the software program PRESENCE v 4.1  to implement all these occupancy models and to estimate relevant parameters.
Estimation of overall elephant occupancy in our landscape.
Following Karanth et al., overall occupancy across the landscape was computed by weighting each site-specific occupancy estimate by the area of forest cover in that site as a proportion of the entire 21,167 km2 landscape, wi and summing over all 205 sites. Standard error of overall occupancy was computed as
The covariance between estimated site-specific occupancy estimates in the expression above was estimated using a parametric bootstrap  where the untransformed β parameter estimates and the associated variance-covariance matrix from the best model were used to generate 100 random deviates from a multivariate normal distribution using the R package MSBVAR [75,76]. These simulated β values were then used to compute 100 site-specific occupancy probabilities for each of the 205 sites using the inverse logit link function. For each pair of site-specific occupancy estimates, covariance was then computed as shown below (please see Appendix S2 in  for more details):
Description of the dataset
We surveyed n = 205 sites, covering 21,167 km2 of forests within our study landscape. In each site, we surveyed 4–42 1-km long spatial replicates (mean effort per site = 20.36km) depending on the amount of forest cover available in the site. A total walk effort of 4,172 km resulted in 2,712 detections of fresh elephant dung. In terms of frequencies, at least one elephant dung pile was detected in 1,230 of the 4,172 1-km replicates (0.295) and in 110 of the 205 sites (naïve occupancy = 0.537).
Spatial dependence in our data.
A comparison of the MacKenzie et al. and Hines et al.  models clearly shows that replicate-level elephant presence was spatially correlated (Δ AIC = 641.7 for the MacKenzie et al.  model). Additionally, the probability of replicate-level elephant presence estimated using the Hines et al. model without covariates, differed substantially depending on whether elephants were present or not present on the previous replicate ( = 0.953 [0.008] and = 0.084 [0.016], respectively). This finding indicates strong Markovian dependence in replicate-level presence. We therefore used the Hines et al. modeling approach for all subsequent occupancy analyses. As expected, detectability was negatively biased when the MacKenzie et al. model was used ( = 0.511 [0.010]) compared to the estimate from the Hines et al.  model ( = 0.737 [0.016]).
Effects of ecological and anthropogenic covariates on detectability and occupancy.
Initial exploratory analysis revealed substantial collinearity between mean normalized difference vegetation index (NDVI) values and mean annual rainfall (Pearson’s r = 0.81). Therefore, these covariates were not included together in any model. As described above, we first explored covariate effects on detectability while holding the covariate structure for occupancy constant––in this step, logit occupancy was modeled as an additive function of proportion of forest cover, NDVI and livestock presence. A model where logit detectability was an additive function of livestock presence and proportion of forest cover received the most support from the data (Table 2). There was evidence to suggest that logit detectability also varied as a function of (a) livestock presence, proportion of forest cover and rainfall (ΔAIC = 1.38) and (b) livestock presence, proportion of forest cover and NDVI (ΔAIC = 1.99). Since our primary interest was in modeling covariate effects on occupancy while accounting for heterogeneity in detectability, we used the covariate structure for detectability as indicated by the minimum AIC model for all subsequent analyses. We note that the covariate structure for occupancy used in this step includes the covariates determined to be most important in determining elephant occupancy (Table 3), with the exception of rainfall, which was highly correlated with NDVI.
Based on the literature on Asian and African elephants (Table 1), we constructed a set of 24 a priori candidate models (Table 3) where ψ was determined by one or more covariates. These models included additive effects of two or more covariates (on the logit scale), interactions between covariates and in the case of NDVI, a quadratic term for non-linear responses. The model with ψ as a function of the interaction between livestock presence and NDVI received the most support from the data, substantially more than the next best model (additive effects of livestock presence, proportion of forest cover and annual rainfall; ΔAIC = 3.03). We therefore did not use model averaging to derive our parameter estimates. Based on the top model, probability of replicate-level presence given (a) presence on the previous replicate and (b) absence on the previous replicate were = 0.962(0.007) and = 0.082(0.015), respectively. The probability of elephant presence on the first replicate was = 0.683(0.035). Site-wise estimates of replicate-level detectability ranged from 0.56 to 0.88, depending on covariate values, while site-wise probability of occupancy ranged from 0.19 to 0.98.
Akaike weights summed over all models containing each predictor () indicated that livestock presence was by far the most important predictor of elephant occupancy (summed Akaike weight ~ 1), mean NDVI (0.662) received less support, and mean annual rainfall (0.1922) and proportion of forest cover (0.1456) were far less important. Non-linear response of occupancy to NDVI and the coefficient of variation of NDVI (as a measure of habitat heterogeneity) received practically no support from the data (summed Akaike weight ~ 0).
Estimated slope parameters from the four models with ∆AIC < 10 (Table 4) showed that elephant occupancy was negatively associated with livestock presence and positively associated with proportion of forest cover, as expected. Contrary to our expectations, the main effects for mean NDVI and mean annual rainfall, as well as the interaction between livestock presence and NDVI were all negative, while the interaction term between livestock and rainfall was positive but small. The interaction term in the top model indicated that livestock presence––a collective surrogate for human disturbance in our study (see  for a discussion)––had the greatest negative effect on elephant occupancy in highly productive (i.e. rainforest) habitats.
Point estimates followed by standard error (SE) in parentheses. No. of sites = 205. Please see Table 1 for descriptions of covariates.
Overall elephant occupancy in our landscape.
Dividing the number of cells in which elephants were detected by the total number of cells (205) yielded a naive (i.e. assuming detectability = 1) occupancy estimate of 0.537. However, based on site-specific occupancy estimates from the minimum AIC model, overall occupancy was estimated as = 0.637 (0.04). Thus of the 21,167km2 of forest area within our study landscape, we estimated that 64% or 13,483 km2 (SE = 847 km2) was actually occupied by elephants in the Western Ghats of Karnataka. In comparison, the naïve occupancy estimate of 0.537, as derived using a traditional presence-absence analysis of our survey data, yields an estimated 11,367 km2 occupied by elephants, an underestimation of true occupancy by c. 16%.
Overall occupancy and covariate effects on detectability and occupancy
Our estimate of overall occupancy (the proportion of area occupied by elephants) of 64% (~13,500km2) is the first such estimate based on scientifically robust methodology for this landscape. It provides a baseline against which future changes in elephant distribution can be rigorously assessed. Our survey design and analytical approaches explicitly modeled the (a) observation processes through which data were collected and (b) effects of biologically reasonable covariates on both elephant habitat occupancy and detectability of their signs. We note that although our estimates of detectability are fairly high (ranging from 0.56 to 0.88 depending on covariate values), ignoring that they are less than perfect (i.e. <1.0) can seriously bias inferred occupancy-habitat (and occupancy-disturbance) relationships by (a) yielding estimated slope parameters with biased magnitude, sometimes with reversed direction and (b) changing the inferred relative importance of covariates [42–44].
Our results indicate that human intrusion and disturbance in these forests, as measured by frequency of livestock signs, is an overwhelmingly more important influence on elephant habitat occupancy compared to ecological attributes we examined. This may indicate that elephant occupancy is likely to be negatively influenced by competition for forage, particularly in the dry season. However, in this landscape and forest management system, livestock presence is also usually strongly correlated with other forms of human disturbance such as biomass extraction, habitat degradation, habitat fragmentation and hunting [51,77]. Clearly, it is not possible to tease apart such confounding effects in a purely non-experimental study such as ours. Nevertheless, our long-term empirical experience in this landscape suggests that the strong negative influence of livestock presence on elephant occupancy likely reflects the cumulative effects of several forms of human disturbances. This is not surprising given the past distribution and recent extirpation of this generalist species from apparently suitable habitats in India [6,78]. Sukumar et al.  and Rood et al.  observed a similar influence albeit using field and analytical methods that ignored imperfect detectability. Recent studies of Asian and African elephants that did account for imperfect and variable detectability also highlight the sensitivity of elephants to human disturbance [16,46]. Further, our finding that detectability was influenced by both human disturbance and proportion of forest cover in a cell suggest that in addition to influencing occupancy, anthropogenic factors may also influence elephant abundance (which can then determine detectability of the species). Although we did not use a modeling approach which explicitly models abundance effects on speciespfxpbe2tfsrtrzsv because of some strong assumptions and stringent design requirements that were unlikely to have been met in our surveys, we feel that this effect of disturbance on the species’ detectability likely arises from effects on abundance, as we had expected a priori when selecting putative predictors of detectability.
The less important effects of NDVI and rainfall were negative, contrary to results from studies in Africa. As pointed out by Borowik et al.  in the context of seasonal differences in forests and grasslands, the ability of NDVI to reflect ground vegetation biomass is mediated by leaf cover in shrub, lower and upper canopy layers. In evergreen forest areas, high NDVI values (due to high canopy closure) may not reflect the low forage availability to elephants at lower levels. Additionally, plants may invest in higher levels of secondary compounds  in the more permanent foliage of evergreen forests. Deciduous forest canopies are almost completely leafless at this time of the year (our image was from 18th February 2006), while the small proportion of montane grasslands in our landscape is covered by coarse, dry grasses with low nutrient and chlorophyll content and, therefore, low NDVI values. These results are consistent with the general observation that Asian elephants occur in lower densities in primary rainforests than in dry and more open forests .
Unlike the African savannah, where rainfall is a strong determinant of elephant occupancy (see Table 1), most areas in our landscape are well above the annual rainfall breakpoint below which rainfall alone determines woody vegetation cover . In the range of rainfall (and correlated NDVI) values found in the African savannah, elephant distribution as well as abundance has been shown to be strongly positively related to NDVI. In our landscape, the minimum long-term average annual rainfall for a grid cell (averaged across all pixels within that cell) was 630 mm, the mean was 2515 mm, and the maximum was 5883 mm. In such a wet landscape, we found that elephant distribution was negatively related to both rainfall and NDVI (strongly correlated to each other), and we believe that the preponderance of extremely wet (and dense, evergreen-canopied) areas in our landscape masks the peak in pr(occupancy) in deciduous forest areas with intermediate NDVI and rainfall values. This is supported by a recent study carried out in a purely deciduous part within our study landscape , which found that probability of habitat use by elephants was highest in areas which experienced the least reduction in NDVI from the end of the wet to the dry season (i.e. moist deciduous forests). However, we also believe that the strong, predictable correlation of elephant occupancy with NDVI and rainfall evident in the African savannah is more complicated in our landscape, potentially due to the overriding influence of anthropogenic disturbance.
Although our estimates of overall elephant occupancy are largely relevant and limited to monitoring and conserving elephants in our landscape (which, nevertheless, supports the largest population of Asian elephants globally), we suggest that our inferred elephant–habitat relationships are fairly generalizable across the Asian elephant’s geographical range. The evergreen and deciduous forests that comprised a majority of our sampled area are found throughout the the species’ range in south and southeast Asia. Similarly, anthropogenic disturbances abound across the elephant’s range in Asia . We therefore do not expect Asian elephants elsewhere to react very differently to forested habitats, rainfall regimes and, in particular, to anthropogenic pressures. While the specific types of disturbances may vary across the species’ range, causing some differences in how elephant distributions respond to levels of human activities in other landscapes compared to ours, we believe that these differences would be quantitative rather than qualitative.
Elephants, either in Asia or in Africa, are megaherbivores, with similar physiological, thermoregulatory, dietary, osmoregulatory and other requirements/constraints. This is particularly true for the forest elephant in Africa that occurs in relatively similar habitats as the Asian elephant. We therefore suggest that our findings likely have considerable relevance to patterns of habitat selection by African forest (but possibly not African savannah) elephants. At the very least, we feel that our results have the potential to inform the selection of putative predictors of African forest elephant distribution in future studies.
Management and conservation implications
Our findings clearly show that elephant distribution is currently limited more strongly by human disturbance compared to natural habitat characteristics we assessed. This result underscores the conservation value of limiting disturbances such as illegal hunting, cattle grazing, forest product and biomass extraction as well as habitat fragmentation arising from local and large scale economic development. Given that the landscape supports over 10.2 million people and is undergoing rapid economic growth, there is urgent need to provide more strictly protected enclaves for elephants across their distributional range in this landscape. Our study showed poor correspondence between elephant distribution and aridity of habitat types, even during the dry season. Given the presence of numerous natural and artificial sources of water in this habitat matrix, this is not surprising. Therefore, the utility of continually increasing surface water availability through engineering interventions, a standard management practice currently, is questionable.
Our occupancy analyses are also extremely useful for identifying priority sites for focused conservation of elephant populations and their habitats (Fig 2C). While this has been attempted previously (e.g. ), such previous analyses did not account for confounding factors such as spatially uneven sampling effort, imperfect and variable detectability and spatial dependence in detection of elephant signs, all of which can seriously affect inference and prediction. A recent study by Madhusudan et al.  provides a comprehensive and useful assessment of the species' distribution, relative abundance, conservation status and spatial patterns in human–elephant conflict in Karnataka, using a variety of information sources, and examines these with respect to protected area coverage, forest cover, human population density, type and extent of agriculture and extent of interface between forest and human-dominated areas. However, their assessment of elephant distribution does not account for imperfect detectability and, as also recognized by the authors, does not distinguish between occupancy and seasonal use (especially in human-dominated areas). Our sampling and modeling explicitly accounted for both these issues, and the resulting estimates and spatial predictions may therefore be considered extremely reliable. Furthermore, their assessment of relative abundance a) poorly accounts for detectability of dung piles (since they use average perpendicular distance rather than dataset-specific fitted detection functions (e.g. see Jathanna et al. ) to estimate dung pile density); b) is subject to biases from non-random sampling caused by the use of forest beats (area range given as 0.6–1802 km2) as sampling units (see Table 1, Jathanna et al. in press for a discussion) and c) is confounded with spatial variation in dung decay rates (Table 1, Jathanna et al. ), which we expect to be considerable, given the extremely wide range of annual precipitation across elephant habitats in Karnataka.
Estimated site-specific detectabilities and ψ are based on the minimum AIC model.
Monitoring of elephant populations in India
Our study demonstrates the practical utility of occupancy sampling to the monitoring of elephant populations at larger landscape and regional spatial scales. Our approach explicitly accounts for imperfect and variable detectability of elephant signs in surveys, variation in ψ and spatial dependence of elephant presence on trail segments that are surveyed. We submit that the use of a modeling approach that closely describes how the data could have been generated, designing and conducting field surveys so that they match the analytical framework used, and careful modeling of observation as well as state processes (using the best available information) are critical to making inferences that can reliably and substantively inform monitoring, conservation and management of elephants as well as other wide-ranging, terrestrial vertebrate species globally. Unfortunately, the country-wide Synchronised Elephant Census (SEC) currently carried out by the Indian government relies on very different methods that attempt to directly estimate elephant population size: waterhole counts, block counts and ‘indirect’ estimates of elephant densities derived from dung counts [86, 87]. These elephant ‘numbers’ / ‘densities’ obtained in specific areas are then used to produce state-, country- and range-wide estimates of abundance through extrapolation, despite the complete lack of randomization and inadequate replication. Blake and Hedges  argue that such “uncritical acceptance of poor-quality data…impedes planning for effective elephant conservation.” Elephant numbers obtained from the SEC lack scientific rigor, both because of the unreliability of the methods used and the mismatch between the population parameters and the spatial scale at which they are estimated.
As noted by the Elephant Task Force Report  the block and waterhole count methods are not rooted in estimation theory, are subject to a number of biases, and are likely to produce misleading elephant population numbers. Although the ‘indirect method’ can be used to estimate elephant dung densities, reliably deriving elephant densities by applying dung decay and defecation rates is challenged by factors that can introduce substantial biases .
Abundance and density are undoubtedly more useful parameters to inform conservation and management than habitat occupancy, if and when these can be reliably estimated. At the scale of individual sites or reserves (e.g. protected areas), approaches such as line transect surveys  based on visual detections of elephants [90–92] or capture–recapture sampling  of elephants from individual identifications based on photographs or DNA [49,94–96] have proven merit for abundance or density estimation.
However, at large spatial scales such as states, regions and countries, reliable estimation of abundance or density is not just difficult, but impossible at the current time, given constraints such as finite resources and human-power, sample size requirements for reliable estimation, variable detectability (which precludes the use of indices), among other issues. Therefore, we strongly recommend that the focus at such scales be moved away from estimation of abundance or density. Rather than producing biased and dangerously misleading ‘estimates’ of state- and country-wide elephant numbers (see  for a critique), the emphasis should be on producing population parameters that are reliable, and useful for informing conservation at these large spatial scales. As we show in this paper, the occupancy modeling approach is ideally suited for monitoring elephants at these scales. Over time, such data can help assess changes in available habitat as well as monitor rates of colonization and extinction of sub-populations. We believe such reliable and rigorous monitoring is critical for effective long-term conservation of the Asian elephant in the face of rapidly growing economies and accelerating land use changes.
S1 File. Data used to model Asian elephant occupancy in our landscape.
Separate worksheets contain a) detection—non-detection data and b) covariate data.
The analysis and manuscript preparation were carried out while DJ was visiting the Nicholas School of the Environment, Duke University, and we are grateful to S. Pimm, W. Chameides and P. Khasibatla for facilitating and supporting the visit. This work was partly supported by the Vision Group on Science and Technology, Government of Karnataka, and the Department of Biotechnology, Government of India. We thank the U.S. Geological Survey-Patuxent Wildlife Research Center for helpful comments, and the Karnataka Forest Department and Centre for Wildlife Studies for facilitating the field research. We are indebted to A. M. Gopalaswamy for analytical advice, M. Gonzalez Roglich for his help with downloading and summarizing remotely sensed data, K. Yadav for helping prepare the maps, and to V. Srinivas, Y. C. Krishna and the many volunteers and field assistants who helped collect the field data. We also thank the three anonymous reviewers and the two academic editors for their helpful comments, which greatly improved this manuscript.
Conceived and designed the experiments: KUK NSK. Performed the experiments: KUK NSK. Analyzed the data: DJ VRG. Contributed reagents/materials/analysis tools: KUK NSK DJ KKK. Wrote the paper: DJ KKK KUK VRG.
- 1. Leimgruber P, Gagnon JB, Wemmer C, Kelly DS, Songer MA, Selig ER. Fragmentation of Asia's remaining wildlands: implications for Asian elephant conservation. Anim Conserv. 2003; 6: 347–359. pmid:ISI:000186983400009.
- 2. Sukumar R. The living elephants: evolutionary ecology, behavior and conservation. New York: Oxford University Press; 2003.
- 3. Menon V, Sukumar R, Kumar A. A god in distress: threats of poaching and the ivory trade to the Asian elephant in India. Asian Elephant Conservation Centre: Indian Institute of Science; 1997.
- 4. Blake S, Hedges S. Sinking the flagship: the case of forest elephants in Asia and Africa. Conserv Biol. 2004; 18(5): 1191–1202. doi: 10.1111/j.1523-1739.2004.01860.x pmid:ISI:000224071800005.
- 5. Bist S. Elephant conservation in India-an overview. Gajah. 2006; 25: 27–37.
- 6. Karanth KK, Nichols JD, Karanth KU, Hines JE, Christensen NL Jr. The shrinking ark: patterns of large mammal extinctions in India. Proc R Soc B. 2010; 277: 1971–1979. doi: 10.1098/rspb.2010.0171.
- 7. Leuthold W, Sale JB. Movements and patterns of habitat utilization of elephants in Tsavo National Park, Kenya. Afr J Ecol. 1973; 11: 369–384.
- 8. Barnes RFW, Douglas-Hamilton I. The numbers and distribution patterns of large mammals in the Ruaha-Rungwa area of southern Tanzania. J Appl Ecol. 1982; 19: 411–425. doi: 10.2307/2403476
- 9. Ottichilo W. Population estimates and distribution patterns of elephants in the Tsavo ecosystem, Kenya, in 1980. Afr J Ecol. 1986; 24: 53–57. doi: 10.1111/j.1365-2028.1986.tb00342.x
- 10. Hoare RE, Du Toit JT. Coexistence between people and elephants in African savannas. Conserv Biol. 1999; 13: 633–639. pmid:ISI:000080720400020. doi: 10.1046/j.1523-1739.1999.98035.x
- 11. Blake S. The ecology of forest elephant distribution and its implications for conservation. Ph.D. Thesis, University of Edinburgh. 2003.
- 12. Redfern JV. Manipulating surface water availability to manage herbivore distributions in the Kruger National Park, South Africa. Ph.D. Thesis, University of California, Berkeley. 2002.
- 13. Grainger M, van Aarde RJ, Whyte I. Landscape heterogeneity and the use of space by elephants in the Kruger National Park, South Africa. Afr J Ecol. 2005; 43: 369–375. doi: 10.1111/j.1365-2028.2005.00592.x
- 14. Murwira A, Skidmore AK. The response of elephants to the spatial heterogeneity of vegetation in a Southern African agricultural landscape. Landsc Ecol. 2005; 20: 217–234. doi: 10.1007/s10980-004-3159-6
- 15. Ntumi CP, van Aarde RJ, Fairall N, De Boer WF. Use of space and habitat by elephants (Loxodonta africana) in the Maputo Elephant Reserve, Mozambique. S Afr J Wildl Res. 2005; 35: 139–146.
- 16. Buij R, McShea WJ, Campbell P, Lee ME, Dallmeier F, Guimondou S, et al. Patch-occupancy models indicate human activity as major determinant of forest elephant Loxodonta cyclotis seasonal distribution in an industrial corridor in Gabon. Biol Conserv. 2007; 135(2): 189–201. pmid:ISI:000245492600004. doi: 10.1016/j.biocon.2006.10.028
- 17. Chamaille-Jammes S, Valeix M, Fritz H. Managing heterogeneity in elephant distribution: interactions between elephant population density and surface-water availability. J Appl Ecol. 2007; 44: 625–633.
- 18. De Beer Y, van Aarde RJ. Do landscape heterogeneity and water distribution explain aspects of elephant home range in southern Africa's arid savannas? J Arid Environ. 2008; 72: 2017–2025. doi: 10.1016/j.jaridenv.2008.07.002
- 19. Harris GM, Russell GJ, van Aarde RI, Pimm SL. Rules of habitat use by elephants Loxodonta africana in southern Africa: insights for regional management. Oryx. 2008: 66–75.
- 20. Thomas B, Holland JD, Minot EO. Elephant (Loxodonta africana) home ranges in Sabi Sand Reserve and Kruger National Park: a five-year satellite tracking study. PLoS ONE. 2008: e3902. doi: 10.1371/journal.pone.0003902. pmid:19065264
- 21. Biru Y. Feeding preference and movement patterns of the African elephant (Loxodonta africana) and vegetation parameters in Babile Elephant Sanctuary, Ethiopia. M.Sc. Thesis, Adis Ababa University. 2009.
- 22. Loarie SR, van Aarde RJ, Pimm SL. Fences and artificial water affect African savannah elephant movement patterns. Biol Conserv. 2009; 142: 3086–3098. doi: 10.1016/j.biocon.2009.08.008
- 23. Loarie SR, van Aarde RJ, Pimm SL. Elephant seasonal vegetation preferences across dry and wet savannas. Biol Conserv. 2009; 142: 3099–3107. doi: 10.1016/j.biocon.2009.08.021
- 24. Shannon G, Matthews WS, Page BR, Parker GE, Smith RJ. The affects of artificial water availability on large herbivore ranging patterns in savanna habitats: a new approach based on modelling elephant path distributions. Divers Distrib. 2009; 15: 776–783.
- 25. Young KD, Ferreira SM, van Aarde RJ. Elephant spatial use in wet and dry savannas of southern Africa. J Zool. 2009; 278: 189–205. doi: 10.1111/j.1469-7998.2009.00568.x
- 26. Young KD, Ferreira SM, van Aarde RJ. The influence of increasing population size and vegetation productivity on elephant distribution in the Kruger National Park. Austral Ecol. 2009; 34: 329–342.
- 27. Hema EM, Barnes RFW, Guenda W. Distribution of savannah elephants (Loxodonta africana africana Blumenbach 1797) within Nazinga game ranch, Southern Burkina Faso. Afr J Ecol. 2011; 49: 141–149.
- 28. Martin J, Chamaillé-James S, Nichols JD, Fritz H, Hines JE, Fonnesbeck CJ, et al. Simultaneous modeling of habitat suitability, occupancy, and relative abundance: African elephants in Zimbabwe. Ecol Appl. 2010; 20(4): 1173–1182. pmid:20597299 doi: 10.1890/09-0276.1
- 29. Mukeka JM. Analyzing the distribution of the African elephant (Loxodonta africana) in Tsavo, Kenya. Miami University. 2010.
- 30. van Aarde RJ, Jackson TP. Megaparks for metapopulations: Addressing the causes of locally high elephant numbers in southern Africa. Biol Conserv. 2007; 134: 289–297. doi: 10.1016/j.biocon.2006.08.027
- 31. Mueller-Dombois D. Crown distortion and elephant distribution in the woody vegetations of Ruhuna National Park, Ceylon. Ecology. 1972; 53: 208–226. doi: 10.2307/1934074
- 32. Santiapillai C, Chambers M, Ishwaran N. Aspects of the ecology of the Asian elephant Elephas maximus L. in the Ruhuna National Park, Sri Lanka. Biol Conserv. 1984; 29: 47–61.
- 33. Sukumar R. The Asian elephant: ecology and management. Cambridge, UK: Cambridge University Press; 1989.
- 34. Diemer N. Environmental Suitability Analysis for Asian Elephants. M.Sc. Thesis, International Institute for Geo-information Science and Earth Observation, Enschede, Holland. 2003.
- 35. Varma S. Spatial distribution of Asian elephant (Elephas maximus) and its habitat usage pattern in Kalakad-Mundanthurai Tiger Reserve, Western Ghats, southern India. Current Science. 2008; 94: 501–506.
- 36. Gaucherel C, Balasubramanian M, Karunakaran P, Ramesh B, Muthusankar G, Hély C, et al. At which scales does landscape structure influence the spatial distribution of elephants in the Western Ghats (India)? J Zool 2010; 280: 185–194.
- 37. Kumar MA, Mudappa D, Raman TS. Asian elephant Elephas maximus habitat use and ranging in fragmented rainforest and plantations in the Anamalai Hills, India. Trop Conserv Sci. 2010; 3: 143–158.
- 38. Rood E, Ganie AA, Nijman V. Using presence-only modelling to predict Asian elephant habitat use in a tropical forest landscape: implications for conservation. Divers Distrib. 2010; 16: 975–984.
- 39. Williams BK, Nichols JD, Conroy MJ. Analysis and management of animal populations. San Diego: Academic Press; 2002.
- 40. Hedges S, editor.Monitoring elephant populations and assessing threats: a manual for researchers, managers and conservationists. Hyderabad: Universities Press; 2012.
- 41. Yackulic CB, Chandler R, Zipkin EF, Royle JA, Nichols JD, Grant EHC, et al. Presence-only modelling using MAXENT: when can we trust the inferences? Methods Ecol Evol. 2013; 4: 236–243.
- 42. Tyre AJ, Tenhumberg B, Field SA, Niejalke D, Parris K, Possingham HP. Improving precision and reducing bias in biological surveys: estimating false-negative error rates. Ecol Appl. 2003; 13: 1790–1801. doi: 10.1890/02-5078
- 43. Gu W, Swihart RK. Absent or undetected? Effects of non-detection of species occurrence on wildlife–habitat models. Biol Conserv. 2004; 116: 195–203.
- 44. MacKenzie DI, Nichols JD, Royle JA, Pollock KH, Bailey LL, Hines JE. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. San Diego: Academic Press; 2006.
- 45. Rota CT, Fletcher RJ, Evans JM, Hutto RL. Does accounting for imperfect detection improve species distribution models? Ecography. 2011; 34: 659–670.
- 46. Goswami VR, Sridhara S, Medhi K, Williams AC, Chellam R, Nichols JD, et al. Community-managed forests and wildlife-friendly agriculture play a subsidiary but not substitutive role to protected areas for the endangered Asian elephant. Biol Conserv. 2014; 177: 74–81. doi: 10.1016/j.biocon.2014.06.013
- 47. Fernando P, Wikramanayake ED, Janaka H, Jayasinghe L, Gunawardena M, Kotagama SW, et al. Ranging behavior of the Asian elephant in Sri Lanka. Mamm Biol. 2008; 73: 2–13. doi: 10.1016/j.mambio.2007.07.007
- 48. Alfred R, Ahmad AH, Payne J, Williams C, Ambu LN, Kotagama SW, et al. Home range and ranging behaviour of Bornean elephant (Elephas maximus borneensis) females. PLoS One. 2012; 7: e31400. doi:31410.31371/journal.pone.0031400. doi: 10.1371/journal.pone.0031400. pmid:22347469
- 49. Goswami VR, Madhusudan M, Karanth KU. Application of photographic capture–recapture modelling to estimate demographic parameters for male Asian elephants. Anim Conserv. 2007; 10: 391–399. doi: 10.1111/j.1469-1795.2007.00124.x
- 50. MacKenzie DI, Nichols JD, Lachman GB, Droege S, Andrew Royle J, Langtimm CA. Estimating site occupancy rates when detection probabilities are less than one. Ecology. 2002; 83: 2248–2255. doi: 10.2307/3072056
- 51. Karanth KU, Gopalaswamy AM, Kumar NS, Vaidyanathan S, Nichols JD, MacKenzie DI. Monitoring carnivore populations at the landscape scale: occupancy modelling of tigers from sign surveys. J Appl Ecol. 2011; 48: 1048–1056. doi: 10.1111/j.1365-2664.2011.02002.x
- 52. Hines J, Nichols J, Royle J, MacKenzie D, Gopalaswamy A, Kumar NS, et al. Tigers on trails: occupancy modeling for cluster sampling. Ecol Appl. 2010; 20: 1456–1466. pmid:20666261 doi: 10.1890/09-0321.1
- 53. Srivathsa A, Karanth KK, Jathanna D, Kumar NS, Karanth KU. On a dhole trail: Examining ecological and anthropogenic correlates of dhole habitat occupancy in the Western Ghats of India. PLoS One. 2014; 9: e98803. doi: 10.1371/journal.pone.0098803. pmid:24893166
- 54. Madhusudan MD, Sharma N, Raghunath R, Baskaran N, Bipin CM, Gubbi S, et al. Distribution, relative abundance, and conservation status of Asian elephants in Karnataka, southern India. Biol Conserv. 2015; 187: 34–40. doi: 10.1016/j.biocon.2015.04.003
- 55. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. Very high resolution interpolated climate surfaces for global land areas. Int J Climatol. 2005; 25: 1965–1978.
- 56. Das A, Krishnaswamy J, Bawa KS, Kiran M, Srinivas V, Kumar NS, et al. Prioritisation of conservation areas in the Western Ghats, India. Biol Conserv. 2006; 133: 16–31.
- 57. Vidya T, Fernando P, Melnick D, Sukumar R. Population differentiation within and among Asian elephant (Elephas maximus) populations in southern India. Heredity. 2004; 94: 71–80.
- 58. Daniels R. National Biodiversity Strategy and Action Plan: Western Ghats Eco-region. Report submitted to Ministry of Environment and Forests, Government of India; 2001.
- 59. Krishnaswamy J, Kiran MC, Davande S. Eco-climatic vegetation classification for the Western Ghats using multi-season NDVI. Bangalore: Ashoka Trust for Research in Ecology and the Environment; 2003.
- 60. Johnson DH.The comparison of usage and availability measurements for evaluating resource preference. Ecology. 1980; 61: 65–71. doi: 10.2307/1937156
- 61. Baskaran N, Balasubramanian M, Swaminathan S, Desai AJ (1993) Home range of elephants in the Nilgiri Biosphere Reserve, South India. In: Daniel JC, Datye H, editors. A week with elephants: proceedings of the international seminar on Asian elephants. Bombay: Oxford University Press; 1993. pp. 298–313.
- 62. Kendall WL, White GC. A cautionary note on substituting spatial subunits for repeated temporal sampling in studies of site occupancy. J Appl Ecol. 2009; 46: 1182–1188. doi: 10.1111/j.1365-2664.2009.01732.x
- 63. Krishnaswamy J, Kiran MC, Ganeshaiah KN. Tree model based eco-climatic vegetation classification and fuzzy mapping in diverse tropical deciduous ecosystems using multi-season NDVI. Int J Remote Sens. 2004; 25: 1185–1205.
- 64. Barnes R, Asamoah-Boateng B, Naada Majam J, Agyei-Ohemeng J. Rainfall and the population dynamics of elephant dung‐piles in the forests of southern Ghana. Afr J. Ecol. 1997; 35: 39–52.
- 65. Barnes R, Dunn A. Estimating forest elephant density in Sapo National Park (Liberia) with a rainfall model. Afr J Ecol. 2002; 40: 159–163. doi: 10.1046/j.1365-2028.2002.00362.x
- 66. Kendall WL. Robustness of closed capture-recapture methods to violations of the closure assumption. Ecology. 1999; 80: 2517–2525. doi: 10.1890/0012-9658(1999)080[2517:roccrm]2.0.co;2
- 67. Burnham KP, Anderson DR. Model selection and multimodel inference: a practical information-theoretic approach. New York: Springer; 2002.
- 68. MacKenzie DI, Bailey LL. Assessing the fit of site-occupancy models. J Agric Biol Environ Stat. 2004; 9: 300–318.
- 69. Dormann CF, Elith J, Bacher S, Buchmann C, Carl G, Carre G, et al. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. Ecography. 2013; 36: 027–046. doi: 10.1111/j.1600-0587.2012.07348.x
- 70. Doherty PF, White GC, Burnham KP. Comparison of model building and selection strategies. J Ornithol. 2012; 152: 317–323.
- 71. Pettorelli N, Vik JO, Mysterud A, Gaillard J-M, Tucker CJ, Stenseth NC. Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends Ecol Evol. 2005; 20: 503–510. pmid:16701427
- 72. Pettorelli N, Ryan SJ, Mueller T, Bunnefeld N, Jedrzejewsk B, Lima M, et al. The Normalized Difference Vegetation Index (NDVI): unforeseen successes in animal ecology. Clim Res. 2011; 46: 15–27. doi: 10.3354/cr00936
- 73. Hines JE. PRESENCE: Software to estimate patch occupancy and related parameters. Version 4.1. USGS-PWRC. Available: http://www.mbr-pwrc.usgs.gov/software⁄presence.html; 2006.
- 74. Efron B. The jacknife, the bootstrap and other resampling plans. Philadelphia, PA: Society for Industrial and Applied Mathematics; 1982).
- 75. R-Development-Core-Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2012.
- 76. Brandt PT, Appleby J. MSBVAR: Bayesian vector autoregression models, impulse responses and forecasting. R package version 03; 2007.
- 77. Kumar NS. Assessment of distribution and abundance of ungulate prey using spatial models in Nagarahole and Bandipur Tiger Reserves of India. PhD Thesis, Manipal University. 2011.
- 78. Karanth KK, Nichols JD, Hines JE, Karanth KU, Christensen NL. Patterns and determinants of mammal species occurrence in India. J Appl Ecol. 2009; 46: 1189–1200. doi: 10.1111/j.1365-2664.2009.01710.x
- 79. Royle JA, Nichols JD. Estimating abundance from repeated presence-absence data or point counts. Ecology. 2003; 84: 777–790. doi: 10.1890/0012-9658(2003)084[0777:eafrpa]2.0.co;2
- 80. Borowik T, Pettorelli N, Sönnichsen L, Jędrzejewska B. Normalized difference vegetation index (NDVI) as a predictor of forage availability for ungulates in forest and field habitats. Eur J Wildlife Res. 2013; 59: 675–682. doi: 10.1007/s10344-013-0720-0
- 81. Coley PD, Bryant JP, Chapin FS III. Resource availability and plant antiherbivore defense. Science. 1985; 230: 895–899. pmid:17739203 doi: 10.1126/science.230.4728.895
- 82. Sankaran M, Hanan NP, Scholes RJ, Ratnam J, Augustine DJ, Cade BS, et al. Determinants of woody cover in African savannas. Nature. 2005; 438: 846–849. pmid:16341012 doi: 10.1038/nature04070
- 83. Lakshminarayanan N. Assessing dry season habitat use of the Asian elephant Elephas maximus in Bandipur and Nagarahole National Parks, southwestern India, using occupancy modelling. MSc Thesis, National Centre for Biological Sciences, TIFR. pp. 37. 2014.
- 84. Sukumar R, Desai A, Lele S, Basappanavar CH, Bist SS, Kamath NR, et al. Report of the Karnataka Elephant Task Force submitted to the Karnataka High Court. 2012.
- 85. Jathanna D, Karanth KU, Kumar NS, Goswami VR, Vasudev D, Karanth KK. Reliable monitoring of elephant populations in the forests of India: analytical and practical considerations. Biol Conserv. 2015; 187: 212–220.
- 86. Bist S. An overview of the methods for enumeration of wild elephants in India. Gajah. 2003; 22: 67–70.
- 87. Rangarajan M, Desai A, Sukumar R, Easa PS, Menon V, Vincent S, et al. Gajah: securing the future for elephants in India. Report of the Elephant Task Force. New Delhi: Ministry of Environment & Forests, India. 2010.
- 88. Hedges S. Estimating absolute density from dung pile density. In: Hedges S, editor. Monitoring elephant populations and assessing threats. Hyderabad: Universities Press; 2012. pp. 61–111.
- 89. Buckland S, Anderson D, Burnham K, Laake J, Borchers D, Thomas L. Introduction to distance sampling: estimating abundance of wildlife populations. Oxford: Oxford University Press; 2001.
- 90. Karanth KU, Sunquist ME. Population structure, density and biomass of large herbivores in the tropical forests of Nagarahole, India. J Trop Ecol. 1992; 8: 21–35. doi: 10.1017/s0266467400006040
- 91. Varman KS, Sukumar R. The line transect method for estimating densities of large mammals in a tropical deciduous forest: An evaluation of models and field experiments. J Bioscience. 1995; 20: 273–287.
- 92. Baskaran N, Udhayan A, Desai AA. Status of the Asian elephant population in Mudumalai Wildlife Sanctuary, southern India. Gajah. 2010; 32: 6–13.
- 93. Amstrup SC, McDonald TL, Manly BF. Handbook of capture-recapture analysis. Princeton, NJ: Princeton University Press; 2005.
- 94. Morley R, Van Aarde RJ. Estimating abundance for a savanna elephant population using mark–resight methods: a case study for the Tembe Elephant Park, South Africa. J Zool. 2007; 271: 418–427. doi: 10.1111/j.1469-7998.2006.00225.x
- 95. de Silva S, Ranjeewa ADG, Weerakoon D. Demography of Asian elephants (Elephas maximus) at Uda Walawe National Park, Sri Lanka based on identified individuals. Biol Conserv. 2011; 144: 1742–1752. doi: 10.1016/j.biocon.2011.03.011
- 96. Hedges S, Johnson A, Ahlering M, Tyson M, Eggert LS.Accuracy, precision and cost-effectiveness of conventional dung density and fecal DNA based survey methods to estimate Asian elephant (Elephas maximus) population size and structure. Biol Conserv. 2013; 159: 101–108. doi: 10.1016/j.biocon.2012.12.010
- 97. Huete A, Didan K, Miura T, Rodriguez EP, Gao X, Ferreira . Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens Environ. 2002; 83: 195–213.