Differential Impacts of Climatic and Land Use Changes on Habitat Suitability and Protected Area Adequacy across the Asian Elephant’s Range

: Climate change and human activities have caused dramatic impacts on biodiversity. Although a number of international agreements or initiatives have been launched to mitigate the biodiversity loss, the erosion of terrestrial biome habitats is inevitable. Consequently, the identiﬁcation of potential suitable habitats under climate change and human disturbance has become an urgent task of biodiversity conservation. In this study, we used the maximum entropy model (MaxEnt) to identify the current and potential future habitats of Asian elephants in South and Southeast Asia. We performed analyses for future projections with 17 scenarios using the present results as baseline. To optimize the modelling results, we delineated the core habitats by using the Core Mapper Tool and compared them with existing protected areas (PAs) through gap analysis. The results showed that the current total area of core habitats is 491,455 km 2 in size and will be reduced to 332,544 km 2 by 2090 under SSP585 (the shared socioeconomic pathway). The projection analysis under differential scenarios suggested that most of the core habitats in the current protected areas would remain stable and suitable for elephants in the future. However, the remaining 75.17% of the core habitats lay outside the current PAs, and ﬁnally we mapped approximately 219,545 km 2 of suitable habitats as priority protected areas in the future. Although our model did not perform well in some regions, our analyses and ﬁndings still could provide useful references to the planning of protected areas and conservation of Asian elephant.


Introduction
The majority of terrestrial habitats have been modified by climate change and human pressures. The erosion of terrestrial biodiversity has been inevitable. A number of international plans have been initiated to establish protected areas to mitigate biodiversity loss. The selections of protected areas or conservation sites usually rely on indicators related to species richness and the matrix of flagship species distribution [1][2][3]. Species distribution models (SDMs) are important tools in species habitat projection because they combine specific species with niche factors [4][5][6]. Of all the SDMs, MaxEnt is the among the mostly commonly used models with regard to habitat simulation and the relationship between species distribution and determinant variables [7][8][9][10][11], including rare and endangered species [12,13], invasive species [14], etc. The advantage of the MaxEnt model is that it can predict the suitable distribution of species when the sample size is small [15]. Further, to 8000 m above sea level in the Himalayas. The climate is also spatially different, varying from extremely arid regions to extremely humid regions, from extremely hot to extremely cold, and from drought-prone areas to flood-prone areas [51]. A large proportion of annual rainwater in the region comes from summer monsoons [52]. The entire region is strongly affected by southwest and northeast monsoons [53]. The land cover types in the study area are complex, with forest and cropland being the most predominant.

Elephant Occurrence Data
The recorded Asian elephant occurrence events were compiled from the Global Biodiversity Information Facility (GBIF) [54] and the published literature [45]. We only chose the events recorded after 1990 whose uncertainty distance was less than 1 km. To minimize the sampling bias that could result in model overfitting, we only selected locations which were a minimum of 1 km apart from each other, and we also performed a manua check to ensure that there was only one point in each pixel [16]. The final records were 178 occurrence points.

Topography
We used SRTM DEM to calculate topographic slope, land surface roughness, the terrain ruggedness index, the topographic position index, and the topographic wetness index [55]. We also used hydrographic data generated from Shuttle Elevation Derivatives at multiple scales (HydroSHEDS, [56]) to calculate the distance from a certain point to the adjacent surface water. This area has substantial topographic relief, with the elevation ranging from sea level to 8000 m above sea level in the Himalayas. The climate is also spatially different, varying from extremely arid regions to extremely humid regions, from extremely hot to extremely cold, and from drought-prone areas to flood-prone areas [51]. A large proportion of annual rainwater in the region comes from summer monsoons [52]. The entire region is strongly affected by southwest and northeast monsoons [53]. The land cover types in the study area are complex, with forest and cropland being the most predominant.

Elephant Occurrence Data
The recorded Asian elephant occurrence events were compiled from the Global Biodiversity Information Facility (GBIF) [54] and the published literature [45]. We only chose the events recorded after 1990 whose uncertainty distance was less than 1 km. To minimize the sampling bias that could result in model overfitting, we only selected locations which were a minimum of 1 km apart from each other, and we also performed a manual check to ensure that there was only one point in each pixel [16]. The final records were 178 occurrence points.

. Anthropogenic Variables
Land use is a direct transformation of nature by humans [62,63] and has a substantial impact on elephant habitats [27]. We focused on urban area and cropland because these land use types have significant pressure on the environment. The expansion of cropland and urban areas has led to a certain degree of loss, degradation, and fragmentation in the natural and semi-natural habitats of elephants and has also increased human-elephant conflict [20,[64][65][66]. The urban and cropland covers were extracted from the ESA CCI land cover product (Table 1). Urban expansion has affected different species due to land use change and its consequent effects (e.g., urban heat island, night-time lights, and impervious surface construction, etc.) [67][68][69]. Here, we used the distance to urban areas as the measurement to determine urban effects on elephant habitats. We also used the WorldPop database to calculate the population density [70]. Roads and railways can break habitat connectivity and prevent elephant migration within their habitat range. We calculated the distance from Asian elephants' habitats to roads and railways based on the OpenStreetMap [71] and Global Roads Inventory Project (GRIP) [72] to map road interruption.

Climatic Variables
We collected climate data in the current period from WorldClim with 30 arc second resolution [73]. We used 19 bioclimatic variables of the WorldClim dataset in our study, and we also derived 15 other bioclimatic variables calculated from WorldClim data by ENVIREM [74]. The annual AET was obtained from CGIAR-CSI [75]. All the climatic variables are shown in Table 2.

Variable Selection
In total, we obtained 52 input variables, including 7 topographic, 5 vegetation, 5 anthropogenic, and 35 climatic variables. In order to avoid overfitting, highly correlated variables were removed. We calculated Pearson's correlation coefficient between all variables and removed variables with |r| > 0.8 ( Figure 2) [16,76]. The remaining 22 variables were input into the MaxEnt model as predictor variables; however, the variables with contributions less than 1% were removed [77]. Finally, 15 independent variables were used to model the suitable habitat area for Asian elephants ( Figure 2).

Species Distribution Modeling and Evaluation
We used 178 elephant occurrence events as input of the MaxEnt model [78], and all environment variables were resampled to 1 km. In this study, we performed four-fold cross-validation, with all analyses based on the average [79] and all other parameters as default. A jackknife test was used to determine the contribution of each variable to elephant distribution. The performance of the MaxEnt model was evaluated by using the AUC (Area Under the ROC-receiver operating curve) [80]. An AUC value of 0.5-0.7 indicates poor performance, 0.7-0.9 indicates moderate performance, and a value greater than 0.9 indicates high performance [81]. The determination of the potential habitat was based on the 10 percentiles of the training presence, which is a common threshold used in species distribution modeling [16].

Projecting Future Potential Habitat
In order to assess the impact of future climate change and human disturbance, we used the MaxEnt model to project potential habitats of elephants during four future periods (i.e., 2030, 2050, 2070, and 2090) under the four most representative scenarios: (1) SSP126 (SSP1) is a sustainable pathway and uses green roads; (2) SSP245 (SSP2) is a middle pathway Sustainability 2022, 14, 4933 6 of 22 between SSP126 and SSP370; (3) SSP370 (SSP3) is a regional rivalry pathway contrary to global cooperation; (4) SSP585 (SSP5) is a fossil-fueled development pathway in which the global economy grows rapidly, but people face severe mitigation challenges [82]. We obtained future climatic variables from Centre National de Recherches Météorologiques Coupled Global Climate Model, version 6 (CNRM-CM6) [83], which is under the latest Coupled Model Inter-Comparison Project (CMIP6) framework [84] (www.worldclim.org). We used global projections data regarding future urban land expansion [82] to simulate the change of human disturbance. These data projected the future (2020 to 2100) urban land expansion at 1 km resolution under SSPs (SSP1, SSP2, SSP3, SSP4, and SSP5 based on ESA CCI land cover in 2015. Using this data, we calculated the distance of the given place to urban areas and the land use type (only the category of urban) dynamic. The remaining variables remained static because they had no projection data.

Species Distribution Modeling and Evaluation
We used 178 elephant occurrence events as input of the MaxEnt model [78], and all environment variables were resampled to 1 km. In this study, we performed four-fold cross-validation, with all analyses based on the average [79] and all other parameters as default. A jackknife test was used to determine the contribution of each variable to elephant distribution. The performance of the MaxEnt model was evaluated by using the AUC (Area Under the ROC-receiver operating curve) [80]. An AUC value of 0.5-0.7 in-  being contiguous, highly suitable habitat areas. We used the Core Mapper tool from the Gnarly Landscape Utilities to identify core habitats throughout the study area [85]. We identified highly suitable habitat patches in the study area by using a moving window with a 9.4 km radius [40]. In addition, cells within a certain effective distance were included in the targeted core habitats. First, we defined a resistance surface (i.e., resistance = 1/habitat suitability). Then, the resistance value was 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 [86] were included in the core habitat area [40]. This could connect small nearby patches within a reachable distance, resulting in fewer but larger core habitats.

Human Footprint in Core Habitats
To evaluate the human pressure on the core habitat, we calculated the degree of human footprint proposed by Sanderson and Venter, which included 8 types of human activity stressors [87,88]. In addition, we used reservoirs as another input layer. These pressures were weighted following Venter, Sanderson, and Theobald [87][88][89]. All of this information can be found in Table 3. We generated an annual HF dataset from 2000 to 2018 by calculating all stressor layers in our study area at 1 km 2 . We used five stressors as statics layers in the temporal dataset because they only had data from a fixed period, and most of these stressors existed before 2000. In addition, we validated our human footprint results by using the same validation set as Venter's, and the statistical parameters for model validation (RMSE, R 2 , and Kappa) showed that our products were improved.

Gap Analysis
Gap analysis is a popular method for use in evaluating the effectiveness of PAs [16]. In order to assess the effectiveness of existing PAs, we compared the core habitats with existing PAs [90]. Then, we extracted core habitats inside and outside the PAs and calculated their area and proportion to evaluate the conservation efficiency of the existing PAs.
The working procedure of all our analyses can be found in Figure 3. We applied Maxent version 3.4.3 to perform habitat suitability modeling [91]. We finished all the analyses by using the packages usdm, seaborn, and matplotlib in R and Python. Spatial processing was completed in ArcMap 10.2.
In order to assess the effectiveness of existing PAs, we compared the core habitats with existing PAs [90]. Then, we extracted core habitats inside and outside the PAs and calculated their area and proportion to evaluate the conservation efficiency of the existing PAs.
The working procedure of all our analyses can be found in Figure 3. We applied Maxent version 3.4.3 to perform habitat suitability modeling [91]. We finished all the analyses by using the packages usdm, seaborn, and matplotlib in R and Python. Spatial processing was completed in ArcMap 10.2.

Model Performance
The modeling accuracy for the current and future distribution of Asian elephants in our study area was reliable. The AUCtrain and AUCtest were found to be 0.971 and 0.946, respectively, indicating high performance (AUC score > 0.9). This showed that the MaxEnt model performed well in predicting the suitable habitat of elephants based on currently available data.
Among the 15 environmental variables in the model simulation, the 5 variables with the highest contributions included precipitation of coldest quarter, population, aridity index, LULC, and distance to urban area, which accounted for 71.6% of the modeling results ( Table 4). The response curves for these five variables showed that elephants prefer forest and mixed areas with trees and shrubs, and herbaceous, relatively dry environments with high PET, free from high human impact ( Figure 4).

Model Performance
The modeling accuracy for the current and future distribution of Asian elephants in our study area was reliable. The AUC train and AUC test were found to be 0.971 and 0.946, respectively, indicating high performance (AUC score > 0.9). This showed that the MaxEnt model performed well in predicting the suitable habitat of elephants based on currently available data.
Among the 15 environmental variables in the model simulation, the 5 variables with the highest contributions included precipitation of coldest quarter, population, aridity index, LULC, and distance to urban area, which accounted for 71.6% of the modeling results ( Table 4). The response curves for these five variables showed that elephants prefer forest and mixed areas with trees and shrubs, and herbaceous, relatively dry environments with high PET, free from high human impact ( Figure 4).

Current and Future Potential Habitats of Asian Elephants
The area of potential habitats for elephants in the study area was about 516,753 km 2 in total, mainly located in southern and northeastern India, southern Nepal, most areas of Sri Lanka, and southern Thailand ( Figure 5). The protected areas defined by IUCN had a total area of 627,367 km 2 , which was 21.4% larger than the modeled potential suitable habitats. The intersecting area between these two kinds of habitats was only 159,726 km 2 . This difference probably lay in the fact that the modeled potential habitats were derived from elephant occurrence data, while the IUCN-protected areas were artificially delimited regions. The modeled potential habitats could be reserved to be used as protected areas for Asian elephants.

Current and Future Potential Habitats of Asian Elephants
The area of potential habitats for elephants in the study area was about 516,753 km 2 in total, mainly located in southern and northeastern India, southern Nepal, most areas of Sri Lanka, and southern Thailand ( Figure 5). The protected areas defined by IUCN had a total area of 627,367 km 2 , which was 21.4% larger than the modeled potential suitable habitats. The intersecting area between these two kinds of habitats was only 159,726 km 2 . This difference probably lay in the fact that the modeled potential habitats were derived from elephant occurrence data, while the IUCN-protected areas were artificially delimited regions. The modeled potential habitats could be reserved to be used as protected areas for Asian elephants. The current potential distribution of elephants was superposed with the potential distribution in future scenarios to obtain the spatial change of habitats ( Figure 6). The results indicated that the spatial changes in the potential distribution of elephants would display the same trend under four scenarios. The suitable area increased initially and then decreased at the 2050 or 2070 mark (Figure 7). The modeled suitable areas were mainly The current potential distribution of elephants was superposed with the potential distribution in future scenarios to obtain the spatial change of habitats ( Figure 6). The results indicated that the spatial changes in the potential distribution of elephants would display the same trend under four scenarios. The suitable area increased initially and then decreased at the 2050 or 2070 mark (Figure 7). The modeled suitable areas were mainly located in India and the border of Thailand, Myanmar, and Laos, expanding northward, while the decrease mainly occurred in Thailand.

Identification of the Core Habitats
Core habitats were considered to be large, contiguous key areas with a high degree of suitability. We used the Core Mapper tool to delineate the core habitats of elephants under all 17 scenarios (1 for current and 16 for the next four periods under four different scenarios). We calculated the mean suitability in each core habitat patch (Figure 8). The modeled current core habitats were around 491,455 km 2 . Several core habitats with high suitability and large areas were distributed in India, Sri Lanka, and Nepal. The core habitats located in Southeast Asia had low suitability. The future projection under different scenarios showed that the changes in the core habitats displayed the same trend as the changes in the potential habitats (Figure 7). We compared future core habitats and the current core habitats and found that the core habitats remained relatively stable under the SSP1 scenario, while the core habitats experienced significant loss under the SSP5 scenario ( Figure 9).

Identification of the Core Habitats
Core habitats were considered to be large, contiguous key areas with a high degree of suitability. We used the Core Mapper tool to delineate the core habitats of elephants under all 17 scenarios (1 for current and 16 for the next four periods under four different scenarios). We calculated the mean suitability in each core habitat patch (Figure 8). The modeled current core habitats were around 491,455 km 2 . Several core habitats with high suitability and large areas were distributed in India, Sri Lanka, and Nepal. The core habitats located in Southeast Asia had low suitability. The future projection under different scenarios showed that the changes in the core habitats displayed the same trend as the changes in the potential habitats ( Figure 7). We compared future core habitats and the current core habitats and found that the core habitats remained relatively stable under the SSP1 scenario, while the core habitats experienced significant loss under the SSP5 scenario ( Figure 9).

Human Footprint Mapping
We mapped the HF in South and Southeast Asia during 2000-2018 ( Figure 10). In core habitats for Asian elephants, the human footprint continued to increase over the 19 years studied, but the changing rate slowed down in recent years ( Figure 11). The core habitats located in India had high HF values, and core habitats in other regions had low HF values (Figures 10 and 12a). We calculated the changing rate of HF using the least-squares regression method (Figure 12b). The HF's major regions that changed substantially were the core habitats with low HF values, such as Vietnam and Cambodia. The core habitats with higher HF values in southern India and Sri Lanka changed slightly.

Human Footprint Mapping
We mapped the HF in South and Southeast Asia during 2000-2018 ( Figure 10). In core habitats for Asian elephants, the human footprint continued to increase over the 19 years studied, but the changing rate slowed down in recent years ( Figure 11). The core habitats located in India had high HF values, and core habitats in other regions had low HF values (Figures 10 and 12a). We calculated the changing rate of HF using the leastsquares regression method (Figure 12b). The HF's major regions that changed substantially were the core habitats with low HF values, such as Vietnam and Cambodia. The core habitats with higher HF values in southern India and Sri Lanka changed slightly.

Gap Analysis
The area of current core habitats intersected by the existing PAs was 122,008 km 2 , while the area of core habitats outside the PAs was 369,447 km 2 . The area of protected core habitats only accounted for 24.83% of the protected areas, and 75.17% of the core habitats were not considered as protected areas (Figure 13).

Gap Analysis
The area of current core habitats intersected by the existing PAs was 122,008 km 2 , while the area of core habitats outside the PAs was 369,447 km 2 . The area of protected core habitats only accounted for 24.83% of the protected areas, and 75.17% of the core habitats were not considered as protected areas (Figure 13).
We also evaluated the protection efficiency for existing protected areas in protecting the core habitats ( Figure 14). Overall, the protection efficiency of the existing protected areas increased by the 2030s under all scenarios, with the highest value of 28.34% under the SSP1 scenario. Even under the SSP5 scenario by the 2090s, the conservation efficiency was still shown to be 23.73%, with a lower rate of decrease. We combined all the core habitats under all 17 scenarios and identified approximately 219,545 km 2 of land as areas that should be protected as a priority (Figure 15), which could provide a basis for the establishment of future PAs.

Gap Analysis
The area of current core habitats intersected by the existing PAs was 122,008 km 2 , while the area of core habitats outside the PAs was 369,447 km 2 . The area of protected core habitats only accounted for 24.83% of the protected areas, and 75.17% of the core habitats were not considered as protected areas (Figure 13).  We also evaluated the protection efficiency for existing protected areas in protecting the core habitats ( Figure 14). Overall, the protection efficiency of the existing protected areas increased by the 2030s under all scenarios, with the highest value of 28.34% under the SSP1 scenario. Even under the SSP5 scenario by the 2090s, the conservation efficiency was still shown to be 23.73%, with a lower rate of decrease. We combined all the core habitats under all 17 scenarios and identified approximately 219,545 km 2 of land as areas that should be protected as a priority (Figure 15), which could provide a basis for the establishment of future PAs.

Habitats Shift Caused by Climate Change
Among 15 variables, 9 climatic variables accounted for 55.1% contribution, 3 human activity variables accounted for 33.2%, 2 vegetation variables accounted for 9.1%, and 2 topographic variables accounted for 8%. Based on the currently available data and model configuration, the hydrometeorological variables were identified as the most important factors in the spatial distribution of elephant habitats, which agrees with the results of a previous study [40]. Historical documents also demonstrated that climate change has led to a significant shift in the distribution of elephant habitats over the last few centuries [18]. Such habitat shifts will continue to occur in the future and will pose great challenges to the conservation of elephants.
Forage and water availability are critical for elephant survival [30]. In this study, the precipitation of coldest quarter was determined as the key bioclimatic variable. The importance of the bioclimatic conditions with regard to controlling Asian elephant distribution lies with the fact that bioclimate affects vegetation's spatial pattern and its productivity [92]. The precipitation of the coldest quarter was not considered in a previous study [40], probably because the previous investigations only focused on some specific regions, where the spatial variation of the precipitation of coldest quarter was not significant in such areas. However, when we expanded the study area to the whole of South and Southeast Asia, the spatial differences of this variable were highlighted. In a previous study, the AET was commonly deemed as the most important bioclimatic variable, possibly due to its local importance, especially in relatively dry regions [40]. However, in our study, most of the study area (especially potential and core habitat areas) is located in hot-wet regions, and therefore, it has a weak contribution and was removed in model-based variable selection. In fact, the contribution of the AET was reflected in the aridity index, which con- Figure 15. Distribution of areas that should be protected as a priority for Asian elephants.

Habitats Shift Caused by Climate Change
Among 15 variables, 9 climatic variables accounted for 55.1% contribution, 3 human activity variables accounted for 33.2%, 2 vegetation variables accounted for 9.1%, and 2 topographic variables accounted for 8%. Based on the currently available data and model configuration, the hydrometeorological variables were identified as the most important factors in the spatial distribution of elephant habitats, which agrees with the results of a previous study [40]. Historical documents also demonstrated that climate change has led to a significant shift in the distribution of elephant habitats over the last few centuries [18]. Such habitat shifts will continue to occur in the future and will pose great challenges to the conservation of elephants.
Forage and water availability are critical for elephant survival [30]. In this study, the precipitation of coldest quarter was determined as the key bioclimatic variable. The importance of the bioclimatic conditions with regard to controlling Asian elephant distribution lies with the fact that bioclimate affects vegetation's spatial pattern and its productivity [92]. The precipitation of the coldest quarter was not considered in a previous study [40], probably because the previous investigations only focused on some specific regions, where the spatial variation of the precipitation of coldest quarter was not significant in such areas. However, when we expanded the study area to the whole of South and Southeast Asia, the spatial differences of this variable were highlighted. In a previous study, the AET was commonly deemed as the most important bioclimatic variable, possibly due to its local importance, especially in relatively dry regions [40]. However, in our study, most of the study area (especially potential and core habitat areas) is located in hot-wet regions, and therefore, it has a weak contribution and was removed in model-based variable selection. In fact, the contribution of the AET was reflected in the aridity index, which contains information pertaining to the AET [40]. Overall, our results demonstrate the importance of bioclimatic conditions with regard to the habitat of Asian elephants, and the response curves showed that elephants prefer places with relatively low precipitation as well as higher PET.
The response of LULC and the 30% threshold value of VCF for elephant habitats suggested that elephants prefer sites with low forest density, such as mixed forest, shrub, and grassland, which elephants prefer to feed on [20,93,94].
Under future climate change scenarios, the spatial transformation of elephants' habitats showed a trend of northern migration overall ( Figure 6). Meanwhile, habitats in the southern part showed degradation, especially under the SSP585 scenario in 2090, with an overall potential habitat reduction of about 25%. Long-term meteorological data projected that the temperature would continue to rise in the future, which would strengthen the Indian monsoon [95], causing an increase in evaporative demand and water availability. This would cause elephants to move northward and to higher altitudes, probably resulting in the shrinkage or complete loss of elephant habitats in central India ( Figure 6). Moreover, we projected a similar shift to higher latitudes and higher altitudes in Southeast-Asian countries ( Figure 6). Although suitability decreases as elevation increases [31], Asian elephants have to move higher to keep pace with climate change. However, the majority of the Asian elephant habitats are located within the range of lowlands and foothills ( Figure 1).

Habitats Compression Driven by Human Disturbance
Human activities are typically the main threats to endangered species [96]. The human modification of the natural environment significantly affects the population of different species. According to our modeling results, the human population was found to be the most important component of human activities in affecting the suitability of habitats for Asian elephants. Within its habitat range, the Asian elephant faces threats caused by increased proximity to humans, including poaching and conflict arising from human-elephant interactions [64,[97][98][99][100][101]. In Karnataka, Kerala, and Tamil Nadu, where the world's largest Asian elephant populations are distributed, the main threat to the longterm conservation of elephants comes from human disturbance [20]. Elephants' responses to human populations and the distance to urban areas showed that elephant should be prevented from experiencing human disturbance [45]. Urban expansion, especially under the SSP585 scenario in the 2090s, showed that the elephant's habitats would become more fragmented ( Figure 6).
Our results showed that the human footprint value increased rapidly in Vietnam and Cambodia due to the massive local deforestation and rapid expansion of cultivated land. Although the human footprint value growth in India and Sri Lanka was not significant, it reached a relatively high level. Native forest loss and land degradation, which affect about 18% of India's territory [102,103], are major threats to its biodiversity and will decrease the availability of forage for wild herbivores [104,105]. Higher levels of importance should be attached to human pressure in future endeavors to protect the habitats of Asian elephants.

More PAs Need to Be Established
Our gap analysis showed that the existing PAs were inadequate in protecting elephants, and there were many conservation gaps. The efficiency of existing PAs under different scenarios indicated that most of the existing protected areas intersected with the modeled core habitats are stable priority habitats for elephants. However, the total protection efficiency was relatively low, because the remaining 75.17% of the core habitats lie outside the current PAs.
As we mentioned in the analyses, the core habitat was composed of large, contiguous key areas with high suitability. Thus, the modeled priority habitat areas from the Max-Ent could be reliable habitats for protected areas for Asian elephants in future planning ( Figure 16). The modeled core habitats were the most suitable areas for elephant habitats when taking into account various natural and anthropogenic conditions. We suggested that the core habitat areas could be considered as a priority in the future [106,107]. could be reliable habitats for protected areas for Asian elephants in future planning (Figure 16). The modeled core habitats were the most suitable areas for elephant habitats when taking into account various natural and anthropogenic conditions. We suggested that the core habitat areas could be considered as a priority in the future [106,107].

Limitations of the Current Study
The MaxEnt model is one of the most widely used species distribution models. Many studies have confirmed that MaxEnt is reliable in modeling the potential distribution of species [12,107]. However, the MaxEnt model is greatly dependent on the field observation data. In areas without observed occurrence events (e.g., Yunnan and Southeast Asia), the modeled potential and core habitat did not match well with the actual distribution. More field investigations are required, especially in Southeast Asia, where there was less occurrence data. The modelled distribution range did not match well with the extent provided by the IUCN in some places because our results represented the potential areas which are suitable for the elephants to live. The Asian elephants may move out of protected areas and encroach on agricultural land due to the increase in population [31,32]. The MaxEnt could not capture this process based on the current collection of filed investigations. Individual-based models are desirable to simulate the range shift of Asian elephant in the future [108][109][110][111].
In addition, the accuracy of the model's projections for the future were highly dependent on the precision and resolution of the future projection data. For example, VCF, population, and LULC variables were set to be static when making projections for the future because these datasets were generally unavailable at present. As the results showed, human disturbance is quite important. The best way to determine the future effects of human disturbance and climate change on the habitats of Asian elephants is to place one type of variable as constant in the model and then compare the rate of habitat

Limitations of the Current Study
The MaxEnt model is one of the most widely used species distribution models. Many studies have confirmed that MaxEnt is reliable in modeling the potential distribution of species [12,107]. However, the MaxEnt model is greatly dependent on the field observation data. In areas without observed occurrence events (e.g., Yunnan and Southeast Asia), the modeled potential and core habitat did not match well with the actual distribution. More field investigations are required, especially in Southeast Asia, where there was less occurrence data. The modelled distribution range did not match well with the extent provided by the IUCN in some places because our results represented the potential areas which are suitable for the elephants to live. The Asian elephants may move out of protected areas and encroach on agricultural land due to the increase in population [31,32]. The MaxEnt could not capture this process based on the current collection of filed investigations. Individual-based models are desirable to simulate the range shift of Asian elephant in the future [108][109][110][111].
In addition, the accuracy of the model's projections for the future were highly dependent on the precision and resolution of the future projection data. For example, VCF, population, and LULC variables were set to be static when making projections for the future because these datasets were generally unavailable at present. As the results showed, human disturbance is quite important. The best way to determine the future effects of human disturbance and climate change on the habitats of Asian elephants is to place one type of variable as constant in the model and then compare the rate of habitat change. In any case, such experiments could not be performed because high-quality human disturbance projection datasets are not available. Although the absence of projection data inevitably affected the accuracy of the projections, in this study, the modeling results were demonstrated as sensible in the current stage [40].
The establishment of PAs often requires a great deal of justification and sophisticated design, while our analysis only focused on analyzing the potential hotspots for elephant habitats without considering the social and economic costs and governance. Future studies are expected to involve more variables and practical scenarios to enhance the rationality of these findings.

Conclusions
This study identified the potential habitat area for Asian elephants using the MaxEnt model. The Core Mapper Tool was used to identify the core habitats of Asian elephants. The results showed that the habitat of Asian elephant will experience a trend of expansion and then shrink due to the backdrop of global change, with a northward shift overall. We constructed the annual human footprint in the study area from 2000 to 2018, which showed a flat change in areas where the human footprint reached a high level and a rapid growth in areas with a low human footprint in Southeast Asia. Protection efficiency was assessed using gap analysis, indicating that the existing protected areas are suitable, but measures are still needed to expand these protected areas. A total of 219,545 km 2 of land was identified as areas that should be protected as a priority; this was achieved by overlaying all of the core habitats under all of the scenarios. Our work is of great significance to the conservation of Asian elephants.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.