Predicting range shifts of Asian elephants under global change

Climate change alters the water cycle, potentially affecting the distribution of species. Using an ensemble of species distribution models (SDMs), we predicted changes in distribution of the Asian elephant in South Asia due to increasing climatic variability under warming climate and human pressures.


| INTRODUC TI ON
Presently, the Asian elephant (Elephas maximus), an endangered species, occupies only a small fraction of its historical range (Choudhury et al., 2008;Sukumar, 2006). Between 26,390 and 30,770 elephants are reported in India, and between 100 and 125 are found in Nepal (Choudhury et al., 2008). The two countries harbour the bulk (more than 60%) of the total population of wild Asian elephants (Choudhury et al., 2008;DNPWC, 2008). The species survives in human-dominated habitats with human densities varying from approximately 149 to 292 people per km 2 (Worldometers, 2017), and is under threat from land use changes due to the continuous conversion of its habitat to agriculture, urbanization, transportation and industry.
Even though climate change is projected to generally increase rainfall in South Asia in the near future (IPCC, 2013;Jayasankar, Surendran, & Rajendran, 2015), the intensity of changes will vary spatially and depend upon location (Rajendran, Sajani, Jayasankar, & Kitoh, 2013). Climate change is expected to increase the occurrence of monsoon break periods, delay monsoon onset and reduce summer precipitation by the end of 21st century (Ashfaq et al., 2009;Schewe & Levermann, 2012). The increase in the frequency, intensity and duration of weak spells or breaks can lead to enhanced drought (Rajeevan, Gadgil, & Bhate, 2010 and their Figure 1).
Warming will lead to shifts in the distributions of species poleward or to higher altitudes, mostly due to changes in temperature (Hickling, Roy, Hill, & Thomas, 2005;Moritz et al., 2008). Plants' distributions are influenced by actual evapotranspiration and water deficit (Stephenson, 1998;Stephenson & Das, 2011). Ultimately, climate-driven geographical redistribution of plant and animal species affects ecosystem functioning, human well-being and the dynamics of the climate system (Lenoir & Svenning, 2015;Pecl et al., 2017).
Land use change, one of the direct threats to the elephant (Choudhury et al., 2008), will also amplify the effects of climate Editor: Wilfried Thuiller of 1 km 2 , and an extensive database on current distribution of elephants. For variable selection, we first developed 14 candidate models based on different hypotheses on elephant habitat selection. For each candidate model, a series of 240 individual models were evaluated using several metrics. Using three climatic and one land use change datasets for two greenhouse gas scenarios, ensemble SDMs were used to predict future projections.
Results: Nine predictor variables were selected for ensemble SDMs. Elephant distribution is driven predominantly by changes in climatic water balance (>60%), followed by changes in temperature and human-induced disturbance. The results suggest that around 41.8% of the 256,518 km 2 of habitat available at present will be lost by the end of this century due to combined effects of climate change and human pressure.
Projected habitat loss will be higher in human-dominated sites at lower elevations due to intensifying droughts, leading elephants to seek refuge at higher elevations along valleys with greater water availability in the Himalayan Mountains.
Main conclusions: Changes in climatic water balance could play a crucial role in driving species distributions in regions with monsoonal climates. In response, species would shift their range upwards along gradients of water availability and seasonal droughts. Conservation and management of elephant populations under global change should include design of movement corridors to enable dispersal of the elephant and other associated species to more conducive environments.

K E Y W O R D S
Asian elephant, climate change, habitat loss, range shift, seasonality, South Asia, species distribution modelling, water balance change indirectly due to alterations in surface energy budgets and thus temperatures at the local scale (Lenoir et al., 2010;Vanderwal et al., 2013). Changes in monsoon dynamics with increased variability in surface temperature and rainfall will affect climate water balance (Singh, Tsiang, Rajaratnam, & Di, 2014), thereby disrupting plant community structure and productivity (Condit, Engelbrecht, Pino, Perez, & Turner, 2013). Availability of forage and water influenced by seasonality is known to drive the habitat utilization of elephants (Bohrer, Beck, Ngene, Skidmore, & Douglas-Hamilton, 2014).
Consequently, the interaction between climate change and land use will compound existing threats to the elephant.
Ensemble modelling approaches, which combine a series of species distribution models (SDMs), have become one of the widely used tools to forecast the anthropogenic climate change impact on species distributions (Araújo & New, 2007;Coetzee, Robertson, Erasmus, Van Rensburg, & Thuiller, 2009), as they produce consensus projections that usually outperform single SDMs (Marmion, Parviainen, Luoto, Heikkinen, & Thuiller, 2009). However, when a large set of predictor variables are considered for analysis, mainly due to the multidimensional abundance of environmental components, the model complexity increases and may lead to the development of heavily parameterized and overfitted models (Merow et al., 2014). Moreover, different SDMs and their underlying modelling decisions about inputs and settings further add to model complexity (Merow et al., 2014;Radosavljevic & Anderson, 2014). One way to reduce model complexity and avoid overfitting is to use an intermediate step of predictor variable selection before embarking into ensemble modelling. This step consists of developing candidate models of differing complexity by grouping the predictor variables based on their environmental types (Carroll, Dunk, & Moilanen, 2010) and on different hypotheses on the mechanisms governing species' habitat selection. These models can be fitted to the observed (presence-only) data, using widely used modelling technique such as MaxEnt (Phillips, Anderson, & Schapire, 2006). MaxEnt is known to facilitate species-specific tuning of modelling decisions for improved performance and the calibration and evaluation of models via spatially independent training and test datasets to reduce model overfitting (Muscarella et al., 2014). The performance of the fitted models can be assessed through the Akaike information criterion (Burnham & Anderson, 2004) and other adequate evaluation metrics (Muscarella et al., 2014). The variables identified by the most parsimonious model can then be used in the ensemble modelling approach.
In this study, we compiled a large database on environmental variables (115) covering India and Nepal with presence observations of Asian elephants (4,262), and subsequently used ensemble modelling and consensus projections to assess the current geographical distribution of the elephant and impacts due to future climate and land use change. We expect that seasonal variation in temperature and precipitation under the monsoonal regime and human disturbance plays a dominant role in driving the distribution of elephants. Our specific objectives were to (a) assess the relative contributions of climatic factors and their seasonal variation, human-induced disturbance, vegetation and topography on F I G U R E 1 Methods for ensemble species distribution modelling for Asian elephant in India and Nepal (a-f); current consensus prediction (g); projection of models into future climate and land use change scenarios (h); assessment of core habitat and connectivity (i). See section species distribution modelling and evaluation for details of the SDMs used in the ensemble approach for the consensus projections

| Study area
Our study area includes India and Nepal (8º-37ºN, 68º-97ºE), encompassing areas of 32,872,631 km 2 and 147,181 km 2 , respectively (Balasubramanian, 2017;Karki, Hasson, Schickhoff, Scholten, & Böhner, 2017). The physiographical distribution is classified into Our study area has a complex topography ranging from sea level in the south of India to elevation higher than 8,000 m in Himalayas in the north (Supporting information Figure S1). Two major weather systems dominate the general climate: the summer and winter monsoon circulations in India and the former in Nepal. The average annual rainfall in India and Nepal is about 1,190 mm and 1,530 mm, respectively, with heterogeneous spatial and temporal distribution (Li & Deng, 2017;Purohit & Kaur, 2017). The study area experi-  In order to reduce spatial autocorrelation in our dataset (i.e., pseudoreplication) that may bias parameter estimates and increase type I error rates (Dormann et al., 2007), we selected only locations which were a minimum of 5.6 km apart from each other. This spatial thinning resulted in 631 locations. We selected this distance because it represents the radius of the minimum circular home range size of elephants, which is approximately 100 km 2 in the study area (Desai, 1991;Sukumar, 1989a).

| Environmental data
We focused on the environmental variables that were important in affecting the distribution of Asian elephants (see Supporting information Tables S1 and S2). As most of our field data were from the years 2000-2017, whenever possible, we compiled our environmental variables to cover this period. We assembled data on land use composition and human disturbances from various sources, as both habitat and human influence are likely to affect the distribution of Asian elephants. We obtained land cover variables and a time series of various vegetation indices from the MODIS database (https://lpdaac.usgs.gov). We then calculated for each variable several metrics across time (i.e., minimum, maximum, mean and SD) to be used as input into our models. We provided a detailed explanation of the collection and preparation of our environmental data in the digital Supporting information (see Appendix S2 and Table S2).
We also included various climatic factors and their seasonal variation (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005;Title & Bemmels, 2017; Supporting information Table S2). We obtained a total of 115 predictor variables: 60 climatic, 16 human disturbance, 29 forest and vegetation and 10 topographic variables. All analyses were done at a grid resolution of 1 km. We obtained our predictor variables at the resolution that was available at source. Variables with a finer resolution were combined by taking the average of grid cell at a particular neighbourhood. Variables with a coarser resolution were converted to fine resolution using bilinear interpolation.

| Candidate model construction and predictor variable selection
To make projections of elephant habitat suitability in space and time, we proceed in two steps: first (this section), we reduced the set of predictor variables by removing correlated variables and testing several models that are based on different hypotheses on elephant habitat selection, and second (next section), we used the variables of TA B L E 1 Evaluation metrics of the most parsimonious model (in bold) in comparison with the other candidate models of differing complexity (fitted in MAXENT) ordered based on increasing AICc score Notes. Models were constructed by grouping the predictor variables based on their environmental types and the available knowledge on elephant habitat selection. For each model type, 240 individual models were fitted in order to improve performance and reduce overfitting, resulting in total of 3,360 individual models. Models used 10,000 random background locations. The model with the lowest evaluation metric scores (i.e., that showed less severe overfitting; bold) and its predictor variables were used to fit the 11 SDMs in the ensemble modelling approach for the consensus projection the most parsimonious of these models in an ensemble of SDMs to project habitat suitability in space and time.
First, we grouped the environmental variables into three groups that correspond to basic hypotheses on elephant habitat selection: vegetation and topography (VT), human disturbance (D) and climate (C) (Supporting information Table S2). The candidate models were then assembled by the combinations of these three groups, thus ranging from specific models reflecting only one hypothesis (e.g., D) to general models containing variables of all three groups (i.e., VT + D + C). Whenever possible, we also defined "minimal" models, denoted by subscript m in italic (i.e., D m and C m ) that contained only those variables expected a priori to be the most important ones.
In total, we developed in this way 14 candidate models of differing complexity (see Supporting information Table S3 for details). We then performed a variable reduction procedure in each candidate model based on Pearson's correlation coefficient r, excluding variables with |r| > 0.6 (see Tables 1 and Supporting information Table   S3).
To fit the candidate models to the data, we used MAXENT, one of the most commonly used presence-only methods (Phillips et al., 2006). MAXENT models the statistical relationships between explanatory variables at observed locations of species, and "back- AUC DIFF , the difference between training and testing AUC, with high values for overfit models (Warren & Seifert, 2011)

| Species distribution modelling and evaluation
An ensemble of species distribution models (SDMs) (Araújo & New, 2007) was generated using the variables identified in the best performing MAXENT model of the previous step ( Figure 1). We con- The occurrence and background locations were split into 80% being a random sample to train the models, and the remaining 20% of the data were used to evaluate the fitted models using true skill statistic (TSS) as measure of accuracy with TSS = sensitivity + specificity -1 (Allouche, Tsoar, & Kadmon, 2006) and the area under the curve of the receiver operating characteristic (AUC, Fielding & Bell, 1997). Models were run five times, each time the training samples were taken randomly to be evaluated against a random test partition. Models with the TSS scores >0.8 were kept in the consensus analysis. The final consensus prediction was based on the mean weighted probability of occurrence of elephants in each 1-km grid cell and a proportional decay, with more weight given to the models that had better TSS scores. The continuous ensemble prediction values were converted into binary ones (suitable/unsuitable) based on the optimal threshold identified by the TSS.

| Future projections under climate and land use scenarios
The

| Assessing core habitat and connectivity
To identify core habitat areas throughout the study area, we used the Core Mapper tool within the Gnarly Landscape Utilities package (Shirk & McRae, 2013). Using a moving window with a 9.4 km radius (i.e., average home range 278 km 2 ; Fernando et al., 2008), this tool identified highly suitable (>0.64) habitat patches in the study area (Supporting information Appendix S4). To assess the potential connectivity among cores, we defined a resistance surface (i.e., resistance = 1/habitat suitability). The resistance value is then multiplied with the minimum distance of this cell from the core area to obtain an effective distance. The cells within an effective distance of 60 km (Sukumar, 1989b) are included into the extended core habitat area. This establishes connectivity between nearby cores that are within a reachable distance, resulting in fewer but larger cores.
All analyses were conducted in R (R Core Team, 2016) using the package ENMeval (Muscarella et al., 2014) for MAXENT fits for evaluating candidate models and BIOMOD2 (Thuiller et al., 2016) for ensemble approach. For the latter, we used the default options, except for MAXENT (Phillips) for which the RM and FC combinations were selected in consistence with the best performing MAXENT model (see section Candidate models construction and predictor variables selection).

| Candidate model evaluation and predictor variable selection
Overall, candidate models generally exhibited good performance (mean AUC TEST > 0.8), especially models that included climatic variables, and the final model was selected from including combination of the climate and the disturbance groups comprising nine predictor variables ( Table 1). The final model (C m + D m ) was selected based on its low AICc scores and evaluation metric scores for overfitting (Table 1).

| Species distribution models
The selected nine predictor variables were chosen to model suit-   (Table 2). This was followed by presence of land cover class "Croplands and Cropland/natural vegetation mosaic" (15.7%), isothermality (13.3%), potential evapotranspiration (PET) in the driest quarter (7%) and the variation of monthly AET over the year (5.6%). The contributions of remaining four variables were <5%.
The response curves for the predictor variables that contributed substantially (% contribution > 5; see Table 2) to the consensus model showed that elephants prefer habitat that experience higher AET (Figure 3, Supporting information Figure

| Future projections
The and Supporting information Figure S4a). Under the climate-only change scenarios, the loss of potential habitat is more moderate (Supporting information Figure S3), but still substantial with a 17.1% loss under RCP8.5 in 2070 (Supporting information Figure   S4b). Gain in potential habitat areas is indicated in the northern (N) and north-eastern (NE) habitats particularly along the valleys towards north avoiding high mountains (Figure 5a,b), with a maximum gain of 42.2% under climate-only change scenario of RCP8.5 in 2070 (Supporting information Figure S4b). Further, a possible northward shift along the valleys in the northern populations' distribution (N) is expected, while losing the relatively flat habitat areas in the foothills (Figure 5a,b).

| Cores and habitat connectivity
The Core Mapper tool identified several core habitat areas ("cores") in the study area (Figure 6a-d and Supporting information Figure S5).
In particular, two large cores with high habitat quality (mean habitat suitability value around 0.91) were identified in the southern (S) part of the study area (S; Figure 6a and Supporting information Figure   S5). Although the total area of cores of current elephant habitat was lower for the northern (N) and north-eastern (NE) regions compared to southern (S) region of the study area, the number of cores is comparable and increased under cost-weighted dispersal, especially in north-east under RCP2.6 in 2070, due to area gains of some patches smaller than 600 km 2 (Figure 6d), revealing the fragmented nature of these habitat areas in terms of high suitability values (Figure 6c,d and Supporting information Figure S5).

| D ISCUSS I ON
The historical range of the Asian elephant has shrunk due to anthropogenic land use change (Choudhury et al., 2008). Currently, we show that seasonal rainfall patterns and human disturbance are strongly associated with the distribution of extant Asian elephant populations over the majority of its range in the Indian subcontinent. Our model projections suggest that future changes in the

% contribution
Climate Actual evapotranspiration (AET mean monthly minimum) 58.4 Isothermality (mean monthly temperature diurnal range/annual range between maximum temperature of warmest month and minimum temperature of the coldest month)

13.3
Potential evapotranspiration (PET) of driest quarter 7 Actual evapotranspiration (SD) 5.6 Aridity index-Thornthwaite 1.5  Table S2 for definition of variables distribution of the elephants in India and Nepal would be driven predominantly by changes in climatic water balance, followed by changes in temperature and other ongoing human-induced disturbance. We anticipate that elephant range would likely shift towards higher elevations in the Himalayas and along a gradient of water availability, that is, low elevation valleys in the mountains, instead of a simple unidirectional range shift towards higher elevations and latitudes typically expected when temperature is the principal factor (Hickling et al., 2005;Lawler et al., 2009;Moritz et al., 2008). Our study also outlines complex local-scale interactions among precipitation and temperature, complicated by seasonal monsoon, and land use changes in the distribution of elephants in South Asia.

Disturbance
F I G U R E 3 Response curves for the environmental variables that contributed substantially (% contribution > 5; see Table 2) to the consensus model for Asian elephant in India and Nepal

| Effect due to climatic factors
Forage and water availability is critical for elephants (Sukumar, 2006). Actual evapotranspiration (AET) is a strong predictor of aboveground net primary productivity for terrestrial ecosystems as it simultaneously measures water availability (rain plus other water sources) and energy (heat and solar radiation), the key factors that enhance photosynthetic rates (Rosenzweig, 1968;Stephenson & Das, 2011). For the past several decades, both India and Nepal have been experiencing significant changes in spatial-temporal patterns of monsoon rainfall, with breaks in wet spell and the prolongation of the dry spell leading to seasonal droughts (Karki et al., 2017;Vinnarasi & Dhanya, 2016). Prolonged dry seasons have been observed to cause drought-related mortality in African elephant   Table 2 Continual rise in greenhouse gases and, thereby, temperatures is predicted to increase the intensity of the summer monsoon in India (scenario RCP8.5;IPCC, 2013;Jayasankar et al., 2015) causing increase in evaporative demand and water availability. This can potentially increase AET and lead to shifts in species' ranges towards higher elevations and latitudes (Hickling et al., 2005;Lawler et al., 2009). Although increase in water availability alone can lead to range expansion within an elevation belt (Stephenson & Das, 2011), additional increase in evaporative demand may modify this pattern in unexpected ways depending on its magnitude relative to water availability. Our results show that this can cause range shifts towards higher elevations but along the water availability gradient F I G U R E 5 the onset of the summer monsoon could be delayed and lead to a rapid decrease in rainfall by the turn of the 22nd century (Ashfaq et al., 2009;Schewe & Levermann, 2012). This will additionally increase the magnitude of evaporative demand in relation to available water and would shift suitable elephant habitat further northwards in the Himalayan region and accelerate habitat loss at lower altitudes.
The adverse effect of increases in global temperature and anomalies in seasonal monsoon rainfall on elephant distribution is also reflected by other climatic factors in our model. Our results suggest that elephants were fewer in areas that experience high minimum surface temperature in the warmest month, smaller day-to-night temperature oscillation relative to the annual oscillations (isothermality) and low PET during the driest quarter of the year in the model (Table 2). These results are in agreement with a study of Mumby et al. (2013) that found that an average monthly temperature increase of ~1ºC over 35 years (between 1965 and 2000) increased mortality in calves and young elephants.

| Effect of human disturbance
Human disturbance and land use change have historically been the main drivers of elephant population decline (Choudhury et al., 2008). Several studies have highlighted the sensitivity of Asian and African elephants to human disturbances (Buij et al., 2007;Jathanna, Karanth, Kumar, Karanth, & Goswami, 2015;Srinivasaiah, Anand, Vaidyanathan, & Sinha, 2012). Our results indicate here that human disturbances represented by the proportion of croplands in a 1.5 × 1.5 km area exerted negative influence on elephant occurrence patterns (Figure 3). The loss of habitat is projected to occur mostly in the human-dominated regions, that is, the Central-Eastern (Eastern Ghats) and the Southern Western Ghats of India and in the relatively flat areas of the Himalayan region (Figure 5a,b). Loss of native forests (Puyravaud, Davidar, & Laurance, 2010) and land degradation, which affects about 18% of its territory (Bai, Dent, Olsson, & Schaepman, 2008), are major threats to India's biodiversity that will decrease the availability of forage for wild herbivores (Jathanna et al., 2015;Madhusudan, 2004). Climate change will further amplify these threats leading to species declines and eventual extinction (Thomas et al., 2004).

| Importance of habitat connectivity
Our habitat model predicts only the suitability for a given 1-km 2 grid cell without accounting for the movement capacity of elephants and their home range sizes. We therefore added an analysis of core areas, being contiguous high suitability areas larger than the size of two average home ranges, and an analysis of connectivity of core areas that considers the maximal displacement capacity of elephants. The core area analysis identified the major habitat patches, such as two large core areas in the southern part of India that had the highest mean habitat suitability compared to the patches of the other three regions (Supporting information Figure S5). Indeed, field estimates document that the southern region holds by far the largest population of the elephants in the world (Madhusudan et al., 2015). The connectivity analysis suggests that the fragmented core areas that are located along the foothills forests and floodplains of the Himalaya in the northern part of India and Nepal could be connected by a mixture of poor-and high-quality habitat that should form specific targets for management. These analyses provide a first assessment of areas that could provide connectivity among core areas, and future studies using more detailed least-cost path analysis (Cushman, Lewis, & Landguth, 2014) or individual-based movement models (Kanagaraj, Wiegand, Kramer-Schadt, & Goyal, 2013) can test specific management actions in their effectivity to provide connectivity.

| Modelling constraints
Although our modelling approach makes a number of simplifications, for example, that the vegetation shifts are in equilibrium with climate shifting immediately after climate changes or that some variables are held constant (e.g., population density, fire probability), it is the best we can do at the moment to assess the sensitivity of elephant habitats to potential changes in climate and land use.
Though such limitations inevitably affect the accuracy and generality of projections, there are independent tests demonstrating that correlative models, especially when placed in the context of ensemble forecasting, provide inferences that are useful (Araújo, Whittaker, Ladle, & Erhard, 2005) and as powerful as more complex mechanistic models (Fordham et al., 2017). We acknowledge the inaccuracy associated with combining two IPCC assessments, but our approach of combining low emissions with balanced resources scenario and high emissions with adverse impact on the environment scenario provides a balanced approach to make the best use of available data.

| Conservation implications
Loss of local species is considered to be a major driver altering ecosystem structure and function in any region (Hooper et al., 2012). Our study area covers several biodiversity hotspots (Myers, Mittermeier, Mittermeier, da Fonseca, & Kent, 2000) and has a high proportion of threatened species; for example, India supports approximately 4.9% of the world's total threatened plant and faunal species (MEFI, 2009;Squires, 2014), and its extinction risk has increased in the recent past, primarily due to habitat loss and illegal hunting (Secretariat of the Convention on Biological Diversity, 2010). Given the limitations in human and financial resources and political will in conserving all desired species and their habitats in South Asia (Persha, Fischer, Chhatre, Agrawal, & Benson, 2010), prioritizing biodiversity conservation, strengthened by management shortcuts such as umbrella-species concept, should remain high priority in this region. The umbrella-species concept has shown promise in forest restoration and conserving biodiversity (e.g., Bell, 2015;Yamaura, Higa, Senzaki, & Koizumi, 2018), and can be used as an effective strategy for conservation in the South Asian region. Further, the centrepiece of biodiversity conservation always remains to be conservation of remaining habitat from any further loss. By studying a charismatic megafauna and a flagship of conservation, we have not only quantified and mapped the current suitable habitat and projections for this species under a warming climate and land use change scenarios, but also provided conservation opportunities for a large number of other species that co-occur under elephants' umbrella. For example, our suitability maps and future projections can be effectively used to identify critical habitat areas that require immediate conservation attention in order to minimize biodiversity loss through habitat degradation and loss. Further our fine-scale maps at 1-km 2 resolution can be used by park managers and conservationists to identify and prioritize the conservation requirements of those critical habitat areas, for example, preservation of critical remnant patches and connective habitats (e.g., corridors) that are projected to be lost in the future.
Adding to our difficulties in conservation of biodiversity in this region is the changing climate, which is predicted to affect the seasonal monsoon system, leading to variation in local climates. Our comprehensive treatment of relevant climatic factors at the spatial resolution of 1 km 2 enabled us to capture the variation in local climates over a major land part of the South Asian monsoon domain and assess its influence on the current and future distribution of elephants in this region. The wide array of scenarios tested here revealed clear tendencies in the expected habitat changes and provides an understanding of how climate and human-induced threats can influence the elephant distribution, critical prerequisite for the management of endangered species that will help mitigate potential threats. Habitat loss/degradation and range shift of the elephants in human-dominated landscape will also cause increase in human-elephant conflict (Lamichhane et al., 2018). Our study also provides a first assessment on the effect of climate change on the distribution of the Asian elephant in its major habitats in India and Nepal, which could help other assessments over its entire range across South and South-East Asia, and be useful for developing management plans for wildlife conservation under the aegis of climate change.

ACK N OWLED G EM ENTS
The authors thank the State Forest Departments of Arunachal Pradesh, Assam, Karnataka, Kerala, Meghalaya, Odisha, Tamil Nadu, Uttarakhand and Uttar Pradesh for allowing permission and facilities to collect samples and local forest staffs for their support.