2.1 Chronology of the elephant movements

Figure 1 and Supporting Information: Figure S2 and Appendix S1 provide a detailed chronology of the movements of the Kunming herd between March 2020 and October 2021. On 2nd June 2021, when the elephants entered the outskirts of Kunming, showing no signs of changing direction, a decision was made to influence their movements (by means of food lures and path blocks) to guide them back toward the south. From this moment, the authorities had a strong influence on the direction and path of the elephants' movements. By mid-September 2021, the Kunming herd was back within recent elephant range and in October 2021 they had returned to their estimated former range in Mengyang. The other herd (known as XTBG herd) returned to Mengyang in August 2021, also directed through ‘lures and blocks' by the authorities.

2.2 Animal movement data

We compiled a total of 519 elephant location points along the elephant trajectory from three different source types. Most of the data (94.6% of the points; Supporting Information: Figure S3) were collected by the Asian Elephant Research Center of National Forestry and Grassland Administration during their drone monitoring of the herd between 23rd September 2020 and 23rd July 2021. Drone monitoring was generally conducted using the model DJI-M300 (Shenzhen DJI Sciences and Technologies Ltd.). The drones were flown at 300–500 m above the elephants and locations were recorded with the drones' “point positioning” function, with an estimated error of generally less than 5 m. This monitoring and data collection was conducted for management rather than research purposes, hence, the number of daily locations recorded was not systematic and it depended on the elephant movements (e.g., if elephants stayed in the same place for many hours, only one point was recorded). This resulted in 1–6 GPS location per day, with some data gap periods (e.g., 4th–15th May 2021; Supporting Information: Figure S3).

We complemented the drone monitoring data with information (2.3% of the points; range: 27th July 2020 to 1st June 2021) from reliable news sources (e.g., photos and videos where we could confidently verify the location). In this case, we used landmarks from the photos or videos and assigned them the geographical coordinates as found in the online mapping service Baidu maps (https://map.baidu.com; Baidu, Inc.). The remaining data (3.1% of the points; range: January 2009 to 5th September 2020) were obtained from “other sources,” including photos and videos of the elephants in Mengyang and personal accounts from rangers. Data from “other sources” were geographically less accurate, and we only used them for the broad-scale analysis of movement type, but not for the finer-scale analysis of habitat selection.

2.3 Movement type analysis

To identify the general type of movements of the elephant herd and evaluate whether these movements should be considered as migratory or something else we used a two-step methodology similar to Purdon et al. (2018), analyzing (1) the overlap of seasonal ranges; and (2) net squared displacement (NSD). Our criteria to define the elephant movements as migratory was that both the overlap and the NSD methods classified them as such (Purdon et al., 2018). See Supporting Information: Appendix S2 for details of both analyses.

In the analysis of seasonal range overlap, we considered that the first and second seasonal ranges did not overlap at Bhattacharyya's affinity (BA) index values < 0.15, and high overlap between the first and third seasonal ranges at BA > 0.5 (Purdon et al., 2018). We used the NSD approach to classify the trajectory followed by the elephants within one of the four broad movement types: migratory, dispersal, sedentary, and nomadic (Bunnefeld et al., 2011). This method relies on the straight-line distance between the initial point and subsequent locations of a movement path and allows comparisons between the shape of the actual animal trajectory and the expected (theoretical) shape for different movement types. We chose the NSD model that best represented the elephant movements by means of model selection with AIC and AIC weights (Burnham & Anderson, 2002).

Climatically, Xishuangbanna has two marked seasons: a rainy season that goes from May to October, and a dry season that goes from November to April. The dry season can be further divided into a dry-cool (November to February) and a dry-hot (March and April) period (Cao et al., 2006). Due to data availability constraints, for the analyses of type of movement, both seasonal range overlap and NSD, we used 1 year of movements starting from 30th May 2020 (the first date for which we had data in the rainy season) to 29th May 2021.

2.4 Movement characteristics: Habitat use and selection

To analyze the characteristics of the elephant movements and the influence of landscape features on their habitat use and selection, we conducted two types of analyses representing landscape-scale (displacement) and patch-scale (step selection function) spatial scales. For these analyses we only used the high-resolution location data, that is, those from 27th July 2020 to 2nd June 2021, obtained by drone monitoring.

First, we compiled a geospatial data set representing the landscape characteristics of the study area (Table 1 and Supporting Information: Table S1). This data set included variables associated with land use and terrain, the Normalized Difference Vegetation Index (NDVI) and the Tasseled Cap Wetness Index (hereafter “wetness”), nightlight intensity, distance to main roads, and distance to towns and villages. We used covariates from 30 m to 1 km of resolution.

To study broader-scale landscape changes along the elephant path, we analyzed the relationship between landscape covariates and displacement (straight-line distance) from the original home range in Mengyang. Specifically, we used a location known as Wild Elephant Valley, where this elephant herd had frequently been seen since 2009 (Shen et al., 2021) as the starting point for the analysis. We built several models using different combinations of habitat covariates and used the Akaike Information Criterion (AIC) to identify the best model (Burnham & Anderson, 2002). We then used step selection function models (SSF; Thurfjell et al., 2014) to evaluate the elephants patch-scale movement decisions. We selected the best-fitting models using AIC with ΔAIC < 2, we calculated model average for the best candidate models (Burnham & Anderson, 2002), and we estimated the importance of predictor variables by the Sum of Weights (SW; Galipaud et al., 2014).

2.5 Body condition assessment

The body condition of wild Asian elephants can oscillate widely, so visual estimates of individuals' body condition scores can be used to assess the health of individuals or the population (Fernando et al., 2009). We used a simple visual system to assess Asian elephant's body condition developed by Wijeyamohan et al. (2015) using both wild and captive Asian elephants. The authors report that they have never found scores above 8 among wild individuals, and that a score of 10 represents obese captive elephants. Two independent and experienced scorers assigned a body condition score to 14 elephants from the Kunming herd using photographs taken in June 2021 (Supporting Information: Appendix S3 and Figure S4). The scores were then averaged for the whole herd. We were unable to obtain images of the elephants from early 2020 for comparative purposes.

3.1 Type of movement

The Kunming herd moved a straight-line distance (net displacement) of ~300 km from March 2020 to early July 2021, including nearly 200 km between March and June 2021. According to the seasonal range overlap method, the movement of the elephants was classified as nonmigratory, since there was a relatively high degree of overlap (BA = 0.32) between the first and second successive ranges, and a lower degree of overlap (BA = 0.25; Supporting Information: Figure S5) between the first and third seasonal ranges, indicating the herd did not return to the initial seasonal range. The NSD analysis classified the movement of the herd as nomadic behavior (AIC weight = 1; Figure 2a). The migratory model did not converge, and the dispersal and sedentary (home range) models had very low AIC weight (0) compared with the nomadic model. These results indicate that the trajectory followed by the elephants to Kunming cannot be classified as a migratory behavior.

3.2 Habitat use and preference

At landscape scale, as the elephants moved away from Mengyang, they walked into areas with higher elevation, higher intensity of nightlights, farther away from main roads, and with less forest cover (Supporting Information: Tables S2 and S3).

The SSF analysis showed the complex effects of natural and anthropogenic factors on the elephants' patch-scale movements (Supporting Information: Figure 2b–f; Supporting Information: Tables S4–S6). In their movements, elephants had a complex relationship with forest because, although they avoided areas with greater forest cover (negative effect of forest cover, SW = 1), they preferred areas closer to forest (negative effect of distance to forest, SW = 1). They chose areas of intermediate values of wetness (effect of wetness and wetness²), which indicates preference for areas with intermediate values of habitat structure (in between very thick primary forest and open environments); wetness and wetness², however, had lower support in the models than other variables (SW = 0.58 and 0.41, respectively). Finally, the elephants showed a clear preference for areas with high anthropogenic disturbance (negative effect of distance to towns and villages, SW = 1).

3.3 Body condition

The overall mean (±SD) body condition score (BCS) of the elephants was 7.0 ± 0 (n = 14 elephants; Figure 3), which represents good health (BCS of 7–8 in this scale represent the best conditions for wild individuals).

4.1 Movement type: irruptive nomadism or dispersal, but not migration

The movements of the Kunming herd have often been described as a migration, both by media and experts (e.g., H. Wang et al., 2021). Our results, however, are clear that this behavior should not be considered as migration. Migration is defined by four general features (Glossary; Roshier & Reid, 2003) of which the Kunming herd movements fail to meet three (nonoverlapping ranges, short migratory movements relative to time at seasonal ranges, and temporal predictability). They partially fit the fourth requirement (having symmetrical paths), but this was strongly influenced by the Chinese authorities directing their movements since June 2021. Migratory behavior is expected in highly seasonal environments, but the largely evergreen tropical rainforests of Xishuangbanna are not strongly seasonal.

But, given that the Kunming herd returned to Mengyang, as the XTBG herd had previously done, how should we characterize their behavior? In our NSD analysis, the movements of the Kunming herd were best described as nomadic, and we argue that the behavior of both elephant herds from March 2020 to ~September 2021 can be described as a case of partial irruptive nomadic movements (sensu Teitelbaum & Mueller, 2019; Glossary). Partial because only part of the population (the Kunming and XTBG herds) adopted these movements, while the other herds remained in their home ranges; and irruptive because regular sedentary behavior was punctuated by unpredictable, long-distance movements that were neither seasonal nor dependent on life-stage. Irruptive nomadism can manifest as an escape from abnormally poor conditions, such as the severe drought that hit Southwest China in 2019 and 2020 (H. Wang et al., 2021). Recent progress in animal tracking suggests that irruptive nomadic movements are more common than previously thought (e.g., Dean et al., 2009; Kaczensky et al., 2011).

Until March 2020, the Kunming herd seemed to have a stable home range in Mengyang (we have evidence of their presence there since 2009; Figure 2a). If the herd had established a new home range outside Mengyang, their behavior could have been described as a case of female dispersal, leading to a range shift, like previous events that led to herds from Mengyang establishing in the neighboring areas of Simao, Jiangcheng, and Menghai (black arrows in Figure 1). Dispersal and home range shifts are uncommon for Asian elephant herds, but several cases have also been reported in India (Sivasubramanian & Ramakrishnan, 2021; Sukumar, 2003; Anwaruddin Choudhury pers. comm.) and Indonesia (Gaius Wilson pers. comm.) since the 1990s, usually linked to drastic habitat losses or drought events.

Long-distance female Asian elephant movements, in the hundreds of km, away from their home range are exceptional and we only found evidence of two previous cases, both associated with severe droughts in India (Sukumar, 2003). One took place in 1983, when several herds shifted their home ranges from Tamilnadu and Karnataka into Andhra Pradesh, 200–300 km away, where they had been absent for over two centuries. The other was in 1987, when elephants moved from Bihar to West Bengal, but in this case the elephants did not shift their home range, they just expanded their former one (Sukumar, 2003). In Africa, there are numerous accounts of elephants shifting their home ranges and abandoning protected areas during droughts (e.g., Abraham et al., 2019; Kaszta et al., 2021).

Overall, the movements of the Kunming and XTBG herds are testimony to elephants' ecological and behavioral plasticity, being able to modify their behavior in response to unexpected environmental change. While sedentarism with male natal dispersal is their most common pattern of behavior, elephants' range of potential movement types includes female dispersal, irruptive nomadism, migration (at least in Africa), and nonconforming movement types (e.g., circular movement patterns by elephants in Mali; Wall et al., 2013).

4.2 Moving close to people and staying in good body condition

As the elephants moved north, the characteristics of the landscapes they used changed considerably from their familiar landscapes in Xishuangbanna. Specifically, the elephants moved into landscapes at higher altitude and higher human presence (nightlight intensity). At finer scale, the step selection function identified somewhat unusual habitat preferences for Asian elephants. In their steps, the Kunming herd actively chose to move close to towns and villages, and avoided areas with high forest cover, while wetness had a moderate influence (Supporting Information: Tables S5 and S6). In Malaysia, where similar step selection function analyses have been conducted (de la Torre et al., 2019, 2021; Wadey et al., 2018), wetness had a stronger influence on their movements, and elephants avoided human presence (negative effect of nightlight intensity). In both environments, elephants avoided moving far away from forests, showing Asian elephants' preference for forest edges (Campos-Arceiz, 2013; de la Torre et al. 2020). We attribute the Kunming herd's preference of moving close to human presence to their loss of fear of people, reliance on crops as food, and the disturbances produced by people along their path, particularly since May 2021.

The high values of the Kunming herd body condition scores are indicative of good health, even a long time after initiating their trip out of Mengyang. The Kunming herd elephants (BCS = 7.0 ± 0, n = 14) had body condition scores considerably higher and more consistent than elephants in Nangunhe National Nature Reserve (BCS = 5.75 ± 2.4, n = 12; Sun et al., 2021), the only other wild elephant population previously assessed in China. The fact that elephants in unfamiliar habitats and moving rapidly over long distances had better body condition than elephants living inside a protected area is testimony to the foraging benefits of feeding on crops as opposed to feeding on wild vegetation in the forest. Additionally, the Kunming herd delivered two babies successfully during their trip. Severe droughts can cause high mortality among elephants (Dudley et al., 2001; Wato et al., 2016). Altering regular movement patterns and moving out of traditional home ranges has been found to mitigate drought-related mortality, at least among calves (Foley et al., 2008). In the case of the Kunming herd, their high body condition scores and successful baby deliveries indicate good health and suggest that the decision of setting off from Mengyang paid off for them.

Our analyses provide a quantitative and solid interpretation of the Kunming herd movements and health condition during their critical time far away from Mengyang and advance our understanding of Asian elephant movement ecology in response to environmental and anthropogenic changes. Our movement data came largely from drone-based monitoring at a resolution of just one to six data points per day. Although not as comprehensive as GPS-collar telemetry, our approach sufficed to address the objectives of this study. To our knowledge, this is also the first quantitative analyses of elephant movements in China. Given the ongoing geographic expansion of the elephant population in China (Bai et al., 2022; Zhang et al., 2015) and the intense fragmentation of human-dominated landscapes in their range (Liu et al., 2017), we encourage the use of GPS-collar telemetry to better understand the spatial dynamics of the population. Studying movements was recently identified as one of the research priorities for elephant conservation in China (S. Chen et al., 2021). For future drone-based wildlife monitoring, we encourage the systematic collection of locations at regular periods of time, which is advisable for the analyses of animal movement attributes.

4.3 Integrating unusual movements into conservation planning

There are important conservation lessons to learn from the movements of the Kunming herd. First, to date, there is no convincing evidence of elephant migration in China (or anywhere else in Southeast Asia). Accordingly, their conservation should be primarily based on area-based and site-specific measures (as corresponds to sedentary animals; Figure 4) such as the new National Park that China is currently planning within elephant range (e.g., S. Chen et al., 2021; Li et al., in press; N. Yang et al., 2021). Since late-succession forest is not an attractive habitat for Asian elephants (e.g., de la Torre et al., 2021; Evans et al., 2018) we recommend introducing small-scale disturbances, such as maintaining a mosaic of early-succession vegetation patches within the forest matrix, which would enhance habitat suitability and elephant carrying capacity within the new park. Such forest gaps provide additional benefits, such as opportunities to monitor the elephant population and to promote conservation-friendly tourism activities. On the other hand, this kind of intervention can lead to a conflict of conservation priorities (i.e., forest vs. elephants) that needs to be carefully considered.

Second, elephants have large home ranges, and they also occupy mixed-use landscapes outside protected areas (de la Torre et al., 2022). Understanding the population distribution and home ranges outside the National Park is important to plan and implement site-specific interventions. As part of the planning process, we recommend identifying “no-go-areas,” where elephant presence is considered unacceptable (based on multi-stakeholder decision-making). In areas where elephants are not welcome, elephant access and presence can be prevented using barriers, reducing habitat connectivity, and establishing response units that can drive elephants out of the no-go-areas, if necessary. In areas where elephants are welcome, we recommend two key management objectives (a) promoting connectivity and (b) mitigating human-elephant conflicts (HEC). Connectivity can be promoted by protecting and restoring forest corridors, even in the form of “many small patches”, and preventing the development of urban or infrastructure barriers within them. HEC mitigation in China can be approached with a combination of intense elephant monitoring outside protected areas, early-warning systems, safe behavior awareness campaigns, spatial concentration of crops, fences, and financial compensation (see S. Chen et al., 2013; Y. Chen et al., 2016; S. Chen et al., 2021 and Campos-Arceiz et al., 2021 for more details).

Third, since the elephant population in China is expanding, we should expect elephants to continue dispersing into new, currently unoccupied, areas (Bai et al., 2022). It is important to anticipate these expansions and, again, define “no-go-areas,” where elephant recolonization is not acceptable. Where elephants are welcome, we recommend (a) protecting existing suitable habitats and (b) being prepared to mitigate future HEC, for example, by running awareness campaigns, training local officers, and establishing a “mobile elephant response unit” that can be mobilized to assist local authorities in case of elephant arrival. The movements of the Kunming herd, have shown that the scale of elephant movements can be measured in the hundreds of km. Hence, preparedness should be at the same scale.

Fourth, the Kunming herd chose to move close to towns, villages, and nightlight intensity, rather than avoiding them. We interpret this as an indication that the elephants felt safe to move near people and to feed on crops. The strategy of “getting people out of the elephants' way”, currently practiced in China (Campos-Arceiz et al., 2021), creates a risk of elephants losing fear of people and becoming reckless, a dangerous situation that may increase the risk of human-elephant encounters and elephant attacks to people. We recommend the use of nonconfrontational methods (e.g., electric fences, permanent barriers) to prevent elephants becoming habituated to crop-based diets and discourage the regular use of so-called “elephant canteens,” whereby crops are planted to subsidize elephant diet. Food subsidies should be used only in exceptional circumstances, for example, during drought.

Fifth, the Kunming and the XTBG herds have shown that, in case of severe environmental change, elephants can respond by altering their behavior and movement patterns. This requires temporal preparedness to respond in case of events that might affect conditions in their home range. Examples of such events include droughts, floods, forest fires, and other forms of large-scale disturbance that might affect their habitats and that are likely to increase their frequency with climate change (IPCC, 2021).

Sixth, these lessons, although specific for elephant conservation in China, will become increasingly relevant to other parts of the Asian elephant range, and for the conservation of other large and conflict-prone species. The ongoing expansion of China's small Asian elephant population is associated with climate change, strict protection, and broad-scale socioeconomic changes (Bai et al., 2022). This expansion is taking place in highly developed and fragmented landscapes where close contact and intense conflict with people are inevitable. This scenario, similar to what is happening with large carnivores in parts of China, Europe, and North America (e.g., Chapron et al., 2014; Gompper et al., 2015; T. Wang et al., 2016), will become increasingly common throughout tropical Asia.