Assessing site-safeguard effectiveness and habitat preferences of Bar-headed Geese (Anser indicus) at their stopover sites within the Qinghai-Tibet Plateau using GPS/GSM telemetry

Abstract

Background

The Bar-headed Goose (Anser indicus) breeds across the high plains and plateau of Central Asia and winters in the Qinghai-Tibet Plateau (QTP), the Yunnan-Guizhou Plateau and the Indian sub-continent. Of the two recognized discrete flyways of the Bar-headed Goose, the Eastern Tibetan Flyway (ETF) is the larger, comprising at least six migration routes. However, we remain ignorant about their migratory connectivity, habitat use and effectiveness of site-safeguard mechanisms set in place for the species.

Methods

We tracked 30 ETF Bar-headed Geese from Chinese and Mongolian breeding areas to their wintering grounds using GPS/GSM transmitters, to determine their migration routes and stopover staging patterns within the QTP, overlaying these upon GIS layers of protected area status and habitat type, to model their habitat selection.

Results

In total, 14 tagged Bar-headed Geese provided information on their entire autumn migration and 4 geese on their entire spring migration. Qinghai Lake marked birds overwintered in the QTP (n = 2), geese tagged in Mongolia wintered either in the QTP (n = 3) or in India/Bangladesh (n = 9), representing three of the migration routes within the ETF. In total, tagged birds staged at 79 different stopover sites within QTP in autumn and 23 in spring, of which 65% (autumn) and 59% (spring) of all fixes fell within the boundaries of either National Nature Reserves (NNRs) or Important Birds Areas (IBAs) in the QTP. Bar-headed Geese predominantly occurred on four land-cover types: grassland (mostly by day), water bodies (at night), wetlands and bare substrates (salt flats, dry lake/river substrates and plough) with little change in proportion. Generalized linear mixed models comparing presence with pseudo-absence data suggested geese strongly selected for wetlands as staging habitat, avoiding bare substrates in spring.

Conclusions

Based on our limited observations of these tagged geese, this study is the first to show that the current designated National Nature Reserves in place in the staging areas within the QTP appear adequate to protect this increasing population. In addition, Hala Lake in Qinghai Province and adjacent areas used as initial QTP staging during autumn migration (currently outside of designated as NNRs/IBAs) are recommended for protection, based on their use by tagged birds from this study. Habitat modelling confirmed the importance of natural wetlands as feeding areas and safe areas of open water as roosting places.

Background

The Bar-headed Goose (Anser indicus) has increased greatly in abundance since the 1990s, reaching an estimated world population of 97,000‒118,000, distributed across a fragmented breeding range throughout the high plains and plateaux of Central Asia (Liu et al. 2017; Fox and Leafloor 2018). Previous telemetry studies have demonstrated the existence of two major flyways of this species in Central Asia (Additional file 1: Fig. S1; Prosser et al. 2009, 2011; Bourouiba et al. 2010; Hao et al. 2010; Köppen et al. 2010; Zhang et al. 2011b; Newman et al. 2012). These can be summarized as follows. The first flyway is the Western Tibetan Flyway, breeding in SE Kyrgyzstan, Kazakhstan and the NW Xinjiang Uygur Autonomous Region in China. These birds winter in Pakistan and the Western Indian subcontinent and comprise more than 30,000 birds, based on wintering counts, including 1800‒5000 in Pakistan (Li et al. 2009; Van der Ven et al. 2010), and 19,000‒40,000 at Pong Dam (Takekawa et al. 2017), the major site for the species on the Indian subcontinent.

Birds from the second flyway, the Eastern Tibetan Flyway (hereafter ETF), breed from Western Mongolia, the SE Xinjiang Uygur Autonomous Region, China to the Qinghai-Tibet Plateau, China (hereafter QTP). These birds are thought to winter mainly in the southern QTP and the Yunnan-Guizhou Plateau in China, throughout the Central and Eastern Indian subcontinent, and in Myanmar. This population is estimated to number more than 80,000 birds based on wintering counts, including 67,000 along the Yarlung Zangbo River and adjacent area (Liu et al. 2017), 5300‒7200 on the Yunnan-Guizhou Plateau (Yang 2005; Yang and Zhang 2014) and 3500‒5800 in Nepal/Bangladesh/Myanmar (Li et al. 2009; Takekawa et al. 2009; Van der Ven et al. 2010). Unknown numbers also winter in East India.

Movements of the Bar-headed Goose were closely studied after an outbreak of highly pathogenic avian influenza (HPAI) H5N1, which killed more than 3000 geese in Qinghai Lake (Chen et al. 2006). These studies identified the importance of tracking their annual movements, because such an increasing and expanding species, showing diverse migration routes and which often uses agriculture areas during staging and in winter, has the potential to rapidly transmit HPAI along its migration routes (Liu et al. 2010).

The ETF includes more than 70% of the individuals in the global population, and most previous studies of this flyway have focused on their breeding and wintering grounds. For instance, breeding habitat use and selection of the Bar-headed Geese has been studied at Qinghai Lake in China (Liu et al. 2008; Cui et al. 2011; Zhang et al. 2011b) and Hangcuo Lake (Zhang et al. 2011a), west-central Mongolia (Batbayar et al. 2014) and Ladakh in India (Prins and van Wieren 2004). Similar studies have taken place on the wintering grounds, including at Caohai Lake (Yang et al. 2012, 2013), Lashihai Lake (Yan et al. 2014) and Yamdrok Lake (Zhang et al. 2016) in China, and Keoladeo National Park in India (Middleton 1992). These studies are important, because Bar-headed Geese distribute two-thirds of their year between the breeding and wintering sites (Takekawa et al. 2017). However, few studies have attempted to identify crucial stopover sites used during migration, especially in the ETF (Hao et al. 2010; Zhang et al. 2011b; Prosser et al. 2011). Although the migration period comprises only one-third of the annual cycle of the geese, energy acquisition to fuel migration can be critical and it is known that migration mortality can exceed 85% of overall annual deaths in some species (Ydenberg 2017). Hence, an understanding of stopover sites used by Bar-headed Geese is fundamental to our ability to effectively conserve the species and understand the potential for disease transmission.

The Bar-headed Goose is famous for its trans-Himalayan migration and studies have demonstrated its physiological adaptations for flying at altitudes over 6000 m where oxygen pressure is very low (Hawkes et al. 2011, 2013, 2017; Bishop et al. 2015). Due to the lower air density, Bar-headed Geese flying at such altitudes, suffer not only severe hypoxia, but increase wing beat rates compared to those at sea level to gain the same lift effect, doubling the power requirement (Bishop et al. 2015). Bearing this in mind, however, given the importance of their stopover sites within QTP to provide the necessary energy to cross the QTP, complete their migration and breed successfully, we know lamentably little about the precise Bar-headed Goose staging areas, habitat preferences and their conservation status there.

Here, we attempt to define Bar-headed Goose habitat selection (as distinct from simple habitat use; Jones 2001) to understand how the birds might be showing behavioral responses to changes in habitat conditions, which potentially affect the survival and adaptability of individuals (Block and Brennan 1993). Only by revealing habitat selection patterns, can we fully understand the spatial and temporal distribution of the Bar-headed Geese we follow. Based on results from a previous study, which used geographical information to model habitat selection in ETF breeding areas (Zheng et al. 2018), we identified six key parameters in modelling habitat selection, including land cover (each class is treated as a separate variable), elevation, slope, aspect, distance to river/lake and road.

In this investigation, we applied further telemetry devices to samples of birds on their key breeding areas in China and Mongolia, to extend earlier studies of this species, with the specific aim of identifying key staging areas throughout the migratory life cycle of these birds. Most importantly, we specifically compare the areas used by tagged Bar-headed Geese to assess the effectiveness of currently designated areas in the QTP to protect key staging areas during migration for the species and to determine their habitat use and selection at these areas.

Methods

Capture of individuals and transmitter attachment

In June 2016, three Bar-headed Geese (of unknown age and sex) rescued and rehabilitated by the Qinghai Lake National Nature Reserve Rescue Center were fitted with neck collar-mounted GPS/GSM loggers (weight 40 g, Hunan Global Messenger Technology Co., China) and released at Qinghai Lake (37.07 °N, 99.82 °E, China). The signal was lost from one of the birds during autumn migration, but the two other geese completed their autumn migrations in 2016.

In July 2018, 27 individuals (adult, unknown sex) were rounded up during the flightless moult period and captured at Terkhiin Tsagaan Lake (48.15 °N, 99.59 °E, Mongolia) and Bayan Lake (49.94 °N, 93.90 °E, Mongolia), and fitted with backpack-mounted GPS/GSM loggers (weight 27.5 g, Ornithology and Telemetry Applications, UAB). Of these marked birds from Mongolia, 13 geese provided complete data on the full autumn migration in 2018 (among them, two geese accompanied each other on migration), and four geese on the full spring migration in 2019. Although we cannot eliminate the possibility that the fitted devices affected the behaviour and ecology of these individuals, the loggers constituted much less than 3% to individual total body mass above which it is considered this may more likely occur (Millspaugh and Marzluff 2001; Bodey et al. 2018).

Devices attached in 2016 and 2018 were set to record the GPS locations at 2 h and 10 min intervals throughout the research project, respectively. Some individuals were tracked for more than one year and two individuals migrated together. To avoid pseudo-replication, only the first spring and autumn migration tracks were selected from each individual, and only one bird from the individuals that accompanied each other on migration were selected (Additional file 2: Table S1).

Identifying stopover sites

We applied the method of Wang et al. (2018) to segment movement tracks into “fly” and “non-fly”, then we plotted locations in Google Earth to visualize movements, pinpoint arrival and departure dates at sites, and evaluate where birds were wintering, summering or stopping (Page et al. 2014). We defined stopover sites as locations where a bird did not move > 30 km over a 48 h period (Kölzsch et al. 2016). The staging areas, which include at least two stopover sites with distance less than 30 km from each other, were identified and categorized by manual checking on maps and Google Earth according to duration and location (Chen et al. 2016). In order to exclude the moving positional fixes during stopovers, we only retained the GPS locations where the velocity between adjacent points was less than 1 km/h.

Conservation status of stopovers within the Qinghai-Tibet Plateau

We attempted to assess the effectiveness of the current extent of designated Important Bird and Biodiversity Areas (IBAs), and National Nature Reserves of China (NNRs) for the protection of stopover sites in the QTP used by these tagged Bar-headed Geese. We downloaded China’s IBA (BirdLife International 2020) and NNR boundary information with permissions (https://www.resdc.cn/data.aspx?DATAID=272). We then overlaid all stopover sites in the QTP used by tagged Bar-headed Geese in our study through ArcMap 10.6 to identify which GPS fixes were located within IBA and NNR boundaries.

Environmental variables

To characterize the landcover used by tracked geese, we used the “FROM-GLC 2017v1” land cover dataset (resolution 10 m × 10 m) created by the Department of Earth System Science, Tsinghua University (Gong et al. 2019). The dataset is based upon 10 land cover classification types, namely artificial surfaces, bare substrate, cropland, forest, grassland, permanent snow/ice, shrubland, tundra, wetland and water bodies. Wetland includes marshland and mudflats with natural and semi-natural aquatic or regularly flooded vegetation; water bodies include natural and artificial waterbodies, such as lake, river and reservoir/pond (for detailed definitions of land cover categories, see Gong et al. 2013). We assigned each GPS location within the stopover sites in the QTP to a specific land-use type defined by the land cover data using R 3.6.0 (R Core Team 2019). GPS fixes were assigned to day or night based on local sunrise and sunset times calculated by “solartime” package (v0.0.1; Wutzler 2018).

We used elevation data measured from SRTM3-DEM dataset (resolution 30 m × 30 m) created by NASA and NIMA, and calculated slope and aspect (defined as a parameter value running from 0 to 360, starting from the West and increasing clockwise) for each GPS fix using ArcMap 10.6. Finally, we calculated the shortest distance of each GPS fixes in stopovers to roads (downloaded from https://www.worldclim.org/) and rivers (downloaded from https://download.csdn.net/download/weixin_38779546/10613773), respectively.

Resource selection modeling

We used generalized linear mixed models (GLMMs) with a binomial error structure to evaluate stopover resource selection during autumn and spring migration respectively, with use/availability as response variable and environment variables as explanatory variables (Meng et al. 2020). We diluted GPS fixes to hourly intervals to reduce the potential for autocorrelation (Signer et al. 2019).

Availability data (i.e., pseudo-absence data) at each stopover site were generated by creating 100% minimum convex polygons (MCPs) based on each set of positions for tagged individuals. We extended these outwards by 11.3 km (the average maximum hourly displacement for all individuals at all stopover sites) in all directions around the MCPs to represent the area potentially available to each of the staging birds. We then randomly selected locations from the extended MCP for each stopover site as pseudo-absence data, generating 20 pseudo-absence points for each positional fix to gain stable and unbiased parameter estimates (Northrup et al. 2013).

Rare land cover types (< 5% of total land use by either use or availability data points) were excluded (namely artificial surfaces, cropland, forest, shrubland, tundra and snow/ice), to escape model convergence problems likely below such levels (Altman et al. 2004). We rescaled variables using the “scale” function in “base” package in R, following the method of Becker and Chambers (1984) to estimate the effect size of explanatory variables.

We used the “dredge” function in “MuMIn” package (v1.43.10; Bartoń 2019) in R to develop our resource selection model using model weights derived from AICc criteria. The cross-prediction accuracy of our resource selection model was tested by estimating the area under the receiver operating characteristic (ROC) curve (AUC). AUC values can range between 0 and 1, where 0.5 indicates predictions no better than chance, 1 indicates perfect discrimination, with values above 0.7 being generally accepted as indicating reasonable predictions (Hosmer et al. 2013). Finally, we applied odds ratios to evaluate effect size of the variables (Szumilas 2010).

Results

Tracking results

A total of 17 geese in autumn, and 8 geese in spring began migration with functioning transmitters, however, due to mortalities and transmitter malfunction, we were only able to obtain information on the full migration for 14 and 4 geese during autumn and spring migration, respectively. Based on the complete migration data from 2016 and 2018 combined, the tagged birds followed three different migration routes. Birds marked at Qinghai Lake overwintered in the Shigatse Prefecture of Tibet Autonomous Region (TAR) (n = 2), whereas geese marked in Mongolia either wintered in the Shigatse Prefecture (n = 3) or continued down into India (n = 7) or Bangladesh (n = 2) to winter (Fig. 1). In addition, tracked geese (another three geese in autumn and another four in spring that started but did not complete migration) staged at 79 different stopover sites in autumn and 23 in spring within the QTP (Fig. 1; Additional file 2: Tables S2, S3). The cumulative time that all birds spent at stopover sites within the QTP totaled 1445 h, which constituted 86.7% of the entire stopover duration (1667 h).

Fig. 1

Autumn and spring individual migration routes (black lines) and stopover sites (green circles) of Bar-headed Geese (Anser indicus) derived from GPS/GSM telemetry devices deployed on birds tracked in 2016 and 2018/2019. The shaded red region shows the Qinghai-Tibet Plateau. TtoT: Qinghai Lake − Shigatse Prefecture routes; MtoT, Mongolia − Shigatse Prefecture routes; MtoA, Mongolia-India/Bangladesh routes. Sample sizes are indicated by n (number of individuals finishing migration)

Tagged Bar-headed Geese that summered in Mongolia and wintered in India/Bangladesh, arrived in the QTP during autumn migration on average on September 8 (± 7 days standard deviation; n = 9 individuals) and left the QTP on November 19 (± 13 days). In spring, they arrived in the QTP from India or Bangladesh on average on March 15 (± 6 days; n = 6) and left on April 25 (± 6 days; n = 3; another three geese started but did not complete migration). Bar-headed Geese spent an average of 72.3 ± 17.3 days (n = 9) and 44.3 ± 7.1 days (n = 3) in the QTP during autumn and spring migration respectively.

Conservation status of stopover sites

Data from the goose-borne loggers generated 123,539 non-moving GPS fixes in autumn and 51,282 in spring at stopover sites within the QTP. Of these, 59% and 53% of GPS fixes in autumn and spring respectively fell within NNRs, while 27% and 23% of GPS fixes were from within IBAs. Overall, 65% and 59% of GPS fixes were from within NNRs/IBAs (Figs. 2, 3; Additional file 2: Table S4). In all, we found 12 important staging areas with at least two stopover sites (Table 1). Among these, there were three areas containing at least three stopovers, which fell outside existing NNRs/IBAs designation (Fig. 2). These were: (1) Shule River, Qinghai Province, N_s = 6 (N_s in each case represents the number of stopovers; No. 2 in Table 1; close to Hala Lake); (2) Dangqu River, TAR, N_s = 3 (No. 11; outside of the Se Lin Cuo NNR and near downtown Dangxiong county); (3) Duoqingcuo Lake, TAR, N_s = 11 (No. 12; a national wetland park and close to the highway).

Fig. 2

GPS fixes of stopover sites within the QTP for tagged Bar-headed Geese during autumn and spring migration in 2016 and 2018/2019. a stopovers during autumn migration; b stopovers during spring migration. NNRs: National Nature Reserves of China; IBAs: Important Bird and Biodiversity Areas; QTP, the Qinghai-Tibet Plateau. Sample sizes are indicated by n (number of instrumented individuals generating the data), N_s (number of stopovers within QTP) and N_g (total number of GPS fixes per category). The explanations apply also to Figs. 3, 4

Fig. 3

Conservation status of Bar-headed Geese stopovers within the Qinghai-Tibet Plateau during autumn (totally 63.3%) and spring (58.7%) migration in 2016 and 2018/2019

Table 1 Important staging areas used by tagged Bar-headed Geese (Anser indicus) migrating in the Eastern Tibetan Flyway in 2016 and 2018/2019

Habitat use and selection in stopover sites

Habitat types used by the Bar-headed Geese were predominantly natural ecosystems: 31% grassland, 29% bare substrate (including dry salt flats, bare herbaceous croplands and dry lake/river bottoms; Gong et al. 2013), 26% water bodies and 11% wetlands. Habitat types used by geese differed between day (the majority on grassland, 47% in autumn and 33% in spring) and night (when the majority were using waterbodies, 33% in autumn and 41% in spring), but were similar during autumn and spring migration (Fig. 4).

Fig. 4

Percentage habitat use of Bar-headed Geese during stopovers within the Qinghai-Tibet Plateau during autumn and spring migration in 2016 and 2018/2019 respectively, based on positional fixes from deployed GPS loggers overlaid on land-cover maps of the world (see Methods)

Comparing GLMMs results, the best fit models for predicting spring and autumn Bar-headed Goose stopover sites within the QTP were based on the same nine parameters. These included the four habitat types, as well as slope, aspect, elevation, distance to roads and rivers (autumn: weight = 1.000, AUC = 0.80; spring: weight = 1.000, AUC = 0.83; Table 2). All these parameters are significant (p < 0.001; Fig. 5). Among habitat types, Bar-headed Geese tended to strongly select wetlands (β_autumn = 1.94, β_spring = 2.03), slightly select water bodies (β_autumn = 0.62, β_spring = 0.80), slightly avoided to select grassland (β_autumn = ‒0.95, β_spring = ‒1.05), and were least likely to select bare substrates in spring (β_spring = ‒2.01). Among terrain variables, geese tended to select stopovers facing south (β_autumn = 0.30, β_spring = 0.32), with low elevation (β_autumn = ‒0.12, β_spring = ‒0.31) and slope (β_autumn = ‒2.00, β_spring = ‒1.17) both during autumn and spring migration. Among other variables, geese tended to select stopover sites, which were away from roads (β_autumn = 0.16, β_spring = 0.29) and close to rivers (β_autumn = ‒0.22, β_spring = ‒ 0.43) both during autumn and spring migration (see Additional file 2: Tables S5, S6 for full model details).

Table 2 Model selection results for the top five stopover habitat selection analysis models for the Bar-headed Geese during autumn and spring migration in 2016 and 2018/2019 respectively

Fig. 5

Odds ratios of fixed effects in highest-ranked generalized mixed-effect model for autumn and spring stopovers in 2016 and 2018/2019. Red symbols depict negative effects, and blue symbols depict positive effects. Significant results are marked with asterisks (*** indicates p-value < 0.001)

Discussion

The results of this tracking study clearly show that three NRRs: Yan Chi Wan (Gansu Province), San Jiang Yuan (Qinghai Province), and Se Li Cuo (TAR) are of critical importance to Bar-headed Geese in the ETF during both autumn and spring migration. In particular, the importance of rivers in Yan Chi Wan (No. 1 in Table 1) and San Jiang Yuan NNRs (No. 5, 6) were not previously known. Overall, 65% of autumn GPS fixes were from within NNRs/IBAs and 59% in spring. Our studies confirmed the importance of Hala Lake (close to Shule River, No. 2), San Jiang Yuan NNR (No. 3, 4) and Se Lin Cuo NNR (No. 7‒10) indentified by tagged Bar-headed Geese in previous studies (Hao et al. 2010; Zhang et al. 2011b; Prosser et al. 2011). Nevertheless, our tracking data found additional areas frequently used by tracked geese, which are not currently designated as NNRs/IBAs. Of these, the most important appear to be the Shule River, Qinghai Province, the Dangqu River, TAR and Duoqingcuo Lake, TAR, all of which are recommended for ground survey during the migration season based on their prolonged use (3‒26 days) by tagged birds from this study.

We would also recommend surveying the suitability of adding Hala Lake and adjacent areas (97.60 °N, 38.30 °E; Fig. 2) to the protected area network for this species, as part of the ecological redline for the region. This is part of the NE edge of the QTP which represents the first staging area encountered and used by geese during autumn migration, yet very few of our tracked birds used areas inside the current protected area boundaries, despite the importance of the position of this area in the overall migration network (Xu et al. 2020).

The ETF represents the larger of the two flyways of the Bar-headed Goose, supporting more than 80,000 individuals. Within this flyway, it is thought that there at least six migration routes (Additional file 1). These are: (1) Mongolia–Yarlung Zangbo River, China; (2) Mongolia–East Indian sub-continent; (3) Qinghai Lake–Yarlung Zangbo River, China; (4) Qinghai Lake–Yunnan-Guizhou Plateau, China (Zhang et al. 2011b); (5) Xinjiang–Yarlung Zangbo River, China (Liu et al. 2010); (6) Yarlung Zangbo River, China–Central Indian sub-continent (Newman et al. 2012). Our study only covered the first three of these migration routes, so there remains three other migration routes that are poorly studied, not to mention those used by geese from the Western Tibetan Flyway (Köppen et al. 2010).

After pooling the data from tracked Bar-headed Geese (both from this and other studies) which completed autumn migration and were captured in Mongolia (n = 24) and Qinghai Lake, China (n = 40), 38 of these individuals wintered in the QTP, 25 in Indian sub-continent, and only one on the Yunnan-Guizhou Plateau (Tian et al. 2015; Takekawa et al. 2017). This diversity of migration patterns underlines the need to apply telemetry studies to more Bar-headed Geese marked throughout their breeding distribution to enable us to better delineate the flyway structure of this species, which remains poorly known, and to ensure adequate site safeguard for geese of different breeding provenance. Such an understanding is essential if we are to be able to appraise the effectiveness of the cohesive site-safeguard network to protect all the elements of this complex population throughout its annual cycle. This is especially important because the species exploits arid and high altitude ecosystems at different times of its annual cycle, all of which are known to be particularly susceptible to the effects of current on-going climate change. For example, effects of climate change at one of the Bar-headed Goose’s major breeding sites in west Mongolia, which has experienced the most rapid rise in temperatures in the past decade outside of the Arctic regions (Batbayar et al. 2014), may have serious impacts on their breeding success. In the QTP, the extent of wetlands have increased in the eastern part and decreased in the western-central sectors (Xu et al. 2019), factors which may explain increases in breeding Bar-headed Geese in this area relative to numbers in Mongolia. On the other hand, in TAR, the species is considered to be more vulnerable to power line strikes (Li et al. 2011) and avian influenza (Liu et al. 2010).

The results from this analysis of habitat use by Bar-headed Geese reflect those of many northern hemisphere goose species, which typically feed out in wetlands and grassland by day but resort to open water bodies by night as protection from potential predators (Zhao 2017). Bar-headed Geese mainly used natural ecosystems during migration and on the summering areas, but cropland during winter. In summer, 35% of positional fixes from tagged geese captured at Qinghai Lake were from grassland and 54% from wetlands during the breeding and post-breeding period (Prosser et al. 2011). Zheng et al. (2018) reported 53% of positional fixes from their telemetry tracked Bar-headed Geese were from wetland, 21% forest and 18% bare substrate in Qinghai Lake (n = 8). We found very low forest cover when overlaying the home ranges from Zheng et al. (2018) on the ESA Global Cover 2009 maps and goose use of forest habitats seems extremely unlikely.

During migration, geese mainly used grassland, water bodies, wetlands and bare substrates at their stopover sites in our study. Prosser et al. (2011) reported 75% of positions of tagged Bar-headed Geese came from grassland and > 12% from wetlands during autumn and spring migration. Both studies therefore confirmed tagged geese mainly used natural ecosystems during migration. Small differences between our and Prosser et al.'s (2011) results may be the consequence of land-cover changes that have affected goose habitat availability in the last two decades and improvements in accuracy of the land cover dataset. Prosser et al. (2011) used the 2003 dataset based on 1 km resolution, whereas our study used the 2019 dataset with 10 m accuracy.

During winter, geese mainly used winter wheat fields, or fallow croplands, rivers, lakes and marshes along the Yarlung Zangbo River. Ground counts of feeding geese (n = 44,657 in January 2009) found 72.1% on fallow cropland (Liu et al. 2010). Prosser et al. (2011) also reported tagged geese used 39% cropland near Lhasa city. In addition, Bar-headed Geese also used grassland and cropland as their main diurnal feeding areas at Caohai, China (Yang et al. 2013), Lashihai Lakes, China (Yan et al. 2014) and Keoladeo National Park, India (Middleton and Van der Valk 1987). Hence, like many northern hemisphere goose species, the Bar-headed Goose has shown a plasticity that has enabled it to take advantage of the new farmland feeding opportunities on the wintering areas (Fox and Abraham 2017). We speculate that restricted use of croplands in summer and on migration simply reflects the general remoteness of the areas that they use at these stages of the life cycle, in regions where the area of cropland available for foraging is extremely restricted.

In our original GLMM modelling to compare the habitat types used by geese with pseudo-absence measures in the immediate vicinity, we also entered water recurrence (Pekel et al. 2016) as a potential explanatory parameter, but it was highly correlated with “water body”. Deletion of water recurrence from the model resulted in an AUC of more than 0.7, indicating reasonable predictions from this simplified model. In our final model, geese tended to select wetland and water bodies as habitat, because the seasonal growth of aquatic plants provides accessible digestible biomass which can be exploited to accumulate fat stores by day (Cong et al. 2012; Wang et al. 2013), and open water offers a night time roosting refuge to avoid predators. Geese also appeared to prefer south facing stopover sites, probably because these areas gain maximum solar insolation to stimulate the growth of food plants to support fat accumulation during the spring migration and shelter from the north wind during the autumn migration. Geese avoided bare substrates strongly in spring, probably because of the low surface temperatures and frost during spring migration in these areas. Summering Bar-headed Geese at Qinghai Lake preferred wetland, open land with sparse vegetation, sites close to rivers/lakes and away from roads, croplands and higher elevations (Zheng et al. 2018) and we contest that they also avoided forest. Wintering Bar-headed Geese at Caohai Lake showed a preference for sites with high vegetation cover, low vegetation height, far from human disturbance, close to the water, open habitats and lower elevation as their foraging sites (Yang et al. 2013). Our results closely followed theirs, confirming that all stages of the life cycle, Bar-headed Geese tend to select wetland areas, close to rivers/lakes, away from roads and at lower elevations than would be expected by chance.

The habitat selection patterns of Bar-headed Geese also reflect patterns shown by Greylag Geese in their breeding and wintering areas in East China (Li 2019). Both species tend to select water bodies and wetland as habitat, although Greylag Geese occur in sites nearer to roads and tend to show greater selection for cropland, suggesting they are more tolerant of human disturbance than Bar-headed Geese. Cropland was not used as a parameter in our model, because it contributed only 1% to habitat use among tagged Bar-headed Geese. In conclusion, these results, together with the results of previous studies of the species, continue to indicate that Bar-headed Geese rely predominantly on wetland and water bodies during migration staging in the QTP in the absence of major human impacts on the landscape in these areas.

In contrast, increasing numbers of Bar-headed Geese are shifting to croplands during the winter. This is especially evident along the Yarlung Zangbo River (where the area of cropland increased by 15.5% between 1990 and 2015; Wu et al. 2017). Numbers of Bar-headed Geese wintering along the Yarlung Zangbo River rose from 15,000 in the 1990s (Bishop et al. 1997) to 30,000 in 2007 (Bishop and Tsamchu 2007) to 67,000 in 2014 (Liu et al. 2017) as increasing numbers resorted to farmland to feed.

The Qinghai-Tibet Plateau is a vast territory with high biodiversity interest and a sparse human population, which has benefitted from the positive effects of nature conservation designation, particularly evident through the current large extent of protected areas. Relatively high levels of site protection for the Greylag Geese in East China have resulted in that species spending more than 65% of their stopover duration within IBA/protected areas during migration, and their population size is increasing (Li et al. 2020), suggesting that site protection can contribute to supporting increasing numbers of geese over time. It would therefore appear that Bar-headed Geese have benefitted from improved conservation action (especially protected areas designation) at migration stopovers en route to and from their breeding areas, as well as from the increasing use of energy rich agriculture areas as winter quarters. Nevertheless, as we identify above, with climate change imposing spatially explicit patterns of change in different parts of the annual range of the Bar-headed Goose, it remains essential we improve our understanding of their migration routes and flyways through extended telemetry studies and monitor their population dynamics and abundance to ensure the future of this unique Asian goose species.

Conclusions

This study is the first to identify the paramount importance of stopover sites within the QTP to Bar-headed Geese in their annual cycle, but also to confirm the satisfactory level of current site protection given the patterns revealed by our telemetry data, which is consistent with the upward trend in abundance of the ETF Bar-headed Goose. We recommend Hala Lake and adjacent areas for protection because of their disproportionate importance to the geese during the initial stages of their autumn migration, which are currently outside the present network of NNRs/IBAs. Bar-headed Geese mainly used natural ecosystems during migratory stopovers (the majority feeding on grasslands by daytime and roosting on water body at night). Habitat modelling confirmed geese tend to select wetland areas, close to rivers/lakes, away from roads and at lower elevations.