1Historically, the overlap zones of wild equids were small in Africa but extensive for Przewalski's horses Equus ferus przewalskii and Asiatic wild asses Equus hemionus in Asia. Currently, the Great Gobi B Strictly Protected Area in south-western Mongolia is the only place where sympatric, free-ranging populations of these equids occur. This provides a unique opportunity to test the hypothesis that Przewalski's horses are primarily adapted to mesic steppes and Asiatic wild asses to arid desert steppes and semi-deserts. Understanding the spatial needs and habitat requirements of these little-studied species is a pre-requisite for setting aside and managing protected areas and planning future re-introductions.
2From 2001 to 2005, we followed nine Przewalski's horses and seven Asiatic wild asses using satellite telemetry and direct observations to assess differences in their resource selection strategies and social organization.
3Przewalski's horses had non-exclusive home ranges of 152–826 km2, selected for the most productive plant communities and formed stable harems groups.
4Asiatic wild asses had non-exclusive home ranges of 4449–6835 km2, showed little preferences for any plant community and seemed to live in fission–fusion groups.
5Synthesis and applications. Our results provide evidence for different resource selection strategies in two sympatric equid species. Our findings indicate that the Gobi areas provide an edge, rather than an optimal habitat for Przewalski's horses. Consequently, only small and isolated pockets of suitable habitat remain for future re-introductions. Asiatic wild asses, on the other hand, need access to large tracts of land to cope with the unpredictable resource distribution of the Gobi. Managers should be aware that protecting habitat where Asiatic wild asses occur does not necessarily benefit Przewalski's horse restoration, whereas setting aside habitat for the conservation of Przewalski's horses will only locally benefit Asiatic wild asses.
There are seven species of wild equids, of which four occur in Africa and three in Asia (Moehlman 2002). All species are similar in size and body shape and have a polygynous mating strategy with monomorphic sexes. They inhabit open, grass- or shrub-dominated habitats and are predominantly grazers (Bauer, McMorrow & Yalden 1994; Rubenstein 1989; Moehlman 2002). Equids are highly efficient hind-gut fermentors, adapted to compensate for low-quality food by consuming large quantities (Janis 1976). Different species seem to have very similar ecological requirements and interspecific competition can be expected to be high where species share the same range (Hutchinson 1957).
Overlap zones of the African equids were small and occurred only between two pairs of species. Competitive exclusion, specific adaptations, and historic colonization patterns are believed to be the driving forces behind the observed succession of species along the environmental gradient from desert, over steppe to savanna habitat (Bauer et al. 1994). Bauer et al. speculated that the degree of similarity in the social system and resource selection pattern does not allow for a sympatric occurrence of African wild ass Equus asinus and Grévy's zebra Equus grevyi, whereas Grévy's and plains zebra Equus quagga are sufficiently different to co-exist in areas with mixed habitats and high environmental fluctuations.
Equids have a polygynous mating system (Emlen & Oring 1977) and appear to follow two major strategies based on climate as a proxy for resource availability. Under mesic conditions, where resource distribution tends to be predictable in time and space, females form year-round stable groups. Female cohesion allows males to monopolize mating access by defending a harem group (Emlen & Oring 1977). Under arid conditions, resource distribution tends to be patchy and unpredictable. During droughts, food plants are scare and highly dispersed, discouraging aggregations of animals due to high interspecific competition (Jarman 1974, Moehlman 1998). After rainfall events, biomass becomes temporarily overabundant and attracts many conspecifics, making it difficult to monopolize access to females. Therefore, equids in arid environments tend to live in fission–fusion groups, characterized by loose associations of female–offspring pairs and single individuals (Emlen & Oring 1977; Rubenstein 1989; Moehlman 1998). Empirical evidence from the small overlap zone of plains and Grévy's zebras in Kenya seems to confirm that differences in resource selection strategy are linked with the social system. Plains zebras do not venture far from open water sources and form stable harem groups, whereas Grévy's zebras, which do not have to drink daily, forage farther away from water and live in fission–fusion groups (Rubenstein 1989).
From central Asia, there is evidence for a large overlap in the historical distribution range of the Przewalski's horse Equus ferus przewalskii (Wakefield et al. 2002) and the Asiatic wild ass Equus hemionus (Feh et al. 2002). This strongly suggests differences in resource selection strategies. Empirical data is lacking because the Przewalski's horse became extinct in the wild and the Asiatic wild ass disappeared from most of its historic range. Anecdotal evidence and data from closely related species indicates Przewalski's horses are primarily adapted to mesic steppe habitats and Asiatic wild asses to arid desert-steppes and semi-deserts (van Dierendonck & Wallis de Vries 1996; Bahloul et al. 2001; Feh et al. 2002; Wakefield et al. 2002). The social organization of wild asses is still poorly understood and may vary depending on environmental conditions (Feh, Boldsukh & Tourenq 1994; Feh et al. 2002). Przewalski's horses are known to form stable harem groups, at least under mesic conditions (Bahloul et al. 2001; King 2002; Zimmermann 2005). Today, Mongolia is the most important stronghold of the Asiatic wild ass (Reading et al. 2001; Lkhagvasuren 2007) and the only country where Przewalski's horses have been successfully released into the wild (Zimmermann 2005). One of the two release sites is the Great Gobi B Strictly Protected Area. Although the last wild horses were seen in this area in the 1960s, it is not known whether the Gobi area represents a mere refuge or was once typical Przewalski's horse habitat (van Dierendonck & Wallis de Vries 1996).
In this study, we aim to identify differences in resource selection patterns of the two equid species derived from satellite telemetry and observational data collected from 2001–2005. Based on the literature, we expected to see the following differences:
1Przewalski's horses are primarily adapted to mesic steppe habitats. They stay close to open water sources and show a stronger preference for plant communities with a high productivity and a low inter-annual variance in biomass. Given the fairly reliable and predictable nature of their selected resources, Przewalski's horses are organized in stable harem groups.
2Asiatic wild asses are primarily adapted to arid desert steppes and semi-deserts. They can venture farther away from water and are better able to exploit resources that vary in space and time. Given the unreliable nature of their resource distribution, Asiatic wild asses live in fission–fusion groups.
Besides the theoretical aspects of exploring differences in social organization and resource selection in two sympatric species, a better understanding of habitat requirements of these little-studied species is important to design and adapt conservation strategies (Clark et al. 2006).
Materials and methods
The Great Gobi B Strictly Protected Area was established in 1975 and encompasses some 9000 km2 of desert steppes and semi-deserts (Zhirnov & Ilyinsky 1986). Herder camps are allowed outside of the 1800 km2 core zone at pre-established locations (Fig. 1). These are used by ~100 families with ~60 000 head of livestock (sheep and goats, horses, cattle and camels), predominantly in winter and during the spring and autumn migration (Kaczensky et al. 2007). In summer, human presence in the park is almost negligible. No paved roads exist and dirt tracks are not maintained. In winter, access and mobility within the park is often limited by snow cover. Poaching occurs but, based on the small number of wild ass carcasses found during this study, seems to be of minor importance compared to other Gobi areas (Kaczensky et al. 2006).
The climate of the Great Gobi B Strictly Protected Area is continental with long cold winters and short, hot summers. Monthly temperatures average 14 to 19 °C in summer (May–September) and 4 to −20 °C in winter (October–April; Baitag weather station 1970–2007). Average annual rainfall is 96 mm with a peak during summer. Average snow cover lasts 97 days. Rain and snowfall can be highly variable from year-to-year in space and time (Zhirnov & Ilyinsky 1986). The area is generally considered to follow a non-equilibrium dynamics as biomass production, and as a consequence, ungulate population fluctuations are driven by the amount and timing of rainfall events (Fernandez-Gimenez & Allen-Diaz 1999; von Wehrden & Wesche 2007).
The landscape of the Great Gobi B Strictly Protected Area is dominated by plains in the east and rolling hills in the west. The Altai Mountains flank the park to the north, and the Takhin Shar Naruu Mountains form the southern border with China. Elevations range from 1000 to 2840 m above sea level. Although the international border is fenced in the more accessible areas, the rest of the park is not surrounded or dissected by fences. Open water (rivers and springs) is unevenly distributed with almost no water in the central or western part of the park (Fig. 1). In locations where several springs occur together, they are surrounded by intermittent swamps and form oases. Desert areas are widely dominated by Chenopodiaceae, such as saxaul Haloxylon ammodendron and Anabasis brevifolia. The steppe areas are dominated by Asteraceae, such as Artemisia and Ajania, and Poaceae like Stipa and Ptilagrostis (Hilbig 1995; von Wehrden, Tungalag & Wesche 2006a). High-productivity riparian vegetation and Nitraria sibirica communities are rare and restricted to larger oases and intermittent river valleys (Figs 1 and 2).
The ungulate community of the steppe areas consists of goitered gazelle Gazella subgutturosa, Asiatic wild ass, and Przewalski's horse. Common mammalian predators are the grey wolf Canis lupus and red fox Vulpes vulpes (Zhirnov & Ilyinsky 1986).
group size and composition
Park rangers check individual Przewalski's horse groups 1–4 times each week for group composition, independent of radiotelemetry (Kaczensky et al. 2007). Between October 2002 and December 2005, rangers located different Przewalski's horse groups on 478 days for a total of 1739 group observations. All rangers were experienced domestic horse breeders and were able to identify individual Przewalski's horses based on shape, coat colour, size, scars and/or freeze brands. The Przewalski's horse population increased from 59 in 2003 to 95 in 2005.
From April 2003 until December 2005, we documented wild ass group sizes during 29 surveys, attempted on a monthly basis. Nineteen surveys covered the eastern part of the park (track length per survey: 350–370 km, search area: 3500 km2) and 10 surveys covered the entire Great Gobi B Strictly Protected Area (track length per survey: 766–803 km, search area: 9000 km2). We always travelled with a minimum of four people at a maximum speed of 40 km h−1. Flight distances of wild asses were in the range of 0·5–2 km, which only rarely allowed for a reliable determination of sex or age classes. Previous estimates of the wild ass population in the Great Gobi B Strictly Protected Area have been around 2000 individuals (Reading et al. 2001; Lkhagvasuren 2007).
Between November 2001 and May 2004, we captured and monitored nine Przewalski's horses and seven wild asses (see Supporting Information Appendix S1). For darting, we approached animals by jeep or hid at water points (for details see Walzer et al. 2007). In any given year, only one collared Przewalski's horse was present per group. Although Przewalski's horses move in social groups, we treated eight out of the nine individual horses as independent units because group composition changed among years and can be assumed to have affected group leadership and thus group movements (see Supporting Information Appendix S2). We treated the two horses where group composition did not change as one individual for all statistical analysis. Because we were unable to observe if collared Asiatic wild asses travelled together, we used the average distance between wild ass pairs on the same day and not separated by more than 1 h as an indirect measure to detect possible associations. The average distance of 533 locations of wild ass pairs was 37 km. Only 2 pairs (0·4%) were within 500 m and only 39 pairs (7·3%) within 5 km of each other. We thus concluded that none of the collared animals moved in close association with another collared conspecific. Of the 16 animals collared, we equipped four with ARGOS collars (2-D cell Doppler PTT; NorthStar, Baltimore, Maryland USA) and 12 with GPS/ARGOS collars (2-D cell PTTs, NorthStar, Baltimore, USA and TGW-3580, Telonics, Mesa, Arizona, USA). For animal welfare reasons and to allow collar retrieval, we equipped all collars with pre-programmed drop-off devices (CR-2a, Telonics). Precisions of the GPS locations were in the range of ±15–100 m (P. Kaczensky, unpublished data). For ARGOS locations, we only used the three most precise location classes, where the expected error is ±150–1000 m (Hays et al. 2001).
For all statistical analysis, we used the individual animal and not the telemetry location as sample unit, except for the mixed models, where individuals were set as a random factor. To avoid problems of autocorrelation in bivariate comparisons, we first used the average value of all locations per day, than calculated mean values per individual and subsequently used those to compare species and seasons. We tested for a possible influence of season, by splitting the data into summer as defined from May to September and winter as defined from October to April. Summer has a low probability of temperatures below freezing and a higher probability of rain, thus allowing for plant growth and biomass production. Winter has a high probability of temperatures below freezing and snow cover provides an alternative water source away from open water. We calculated the average daily distances travelled by calculating the straight line distance between locations that were 21–27 h apart and standardized the average distance for 24 h by assuming a linear relationship.
habitat use analysis
von Wehrden et al. (2006a) described 12 plant communities for the Great Gobi B Strictly Protected Area based on supervised Landsat imagery (also see von Wehrden et al. 2006b). Of those, two montane plant communities did not occur within the wild equid range and five others were merged into two new categories (Fig. 2). We expected equid use of different plant communities to match productivity. Because we lacked plant community specific biomass production data, we based productivity estimates and expected equid use on the association of plant communities with different moisture regimes (H. von Wehrden, unpublished data). We expected to see the following ranking: riparian vegetation > Nitraria > Stipa > Nanophyton > Caragana > Reaumuria > Haloxylon (see Supporting Information Appendix S3).
For an overall estimate of biomass production, we used the global layer of biomass production expressed in gram carbon per square metre and year (gC m−2 year−1) for 1981–2000 (Prince & Small 2003). This open-source GIS data set is available on an 8 × 8 km raster basis under http://glcf.umiacs.umd.edu/data/glopem/ with data processing described in Prince & Goward (1995). For our analysis, we used the mean biomass production over all 20 years. We digitized rivers, springs and elevation from Russian 1:100 000 topographic maps. For data analysis we used ArcView 3·1 and ArcMap 9·1 (ESRI, Environmental Systems Research Institute, Inc., Redlands, California, USA) with the Spatial Analyst and Animal Movement (Hooge & Eichenlaub 1997) extensions.
We tested for habitat preferences comparing availability (random points) and use (animal locations). For availability, we created 1000 random points within the buffered minimum convex polygon (MCP) of each individual. Based on the average daily distance covered within 24 h, buffers were 8·3 km for Asiatic wild asses and 3·5 km for Przewalski's horses. From the animal locations, we used only GPS and ARGOS locations separated by ≥ 1 h. For each location and each random point, we derived plant community, slope, average biomass production, elevation, and distance to nearest water source.
To correct for temporal autocorrelation, we checked for the relationship of distance covered vs. time (since the last location, the before last, the two before last and so on; for a similar approach see Swihart & Slade 1985) for all possible time intervals up to 72 h apart. For both species, the relationship was best fitted by a power function with exponent 0·6792 (r2 = 0·141, d.f. = 7,824, P < 0·001) for Asiatic wild asses and exponent 0·3793 (r2 = 0·044, d.f. = 27 694, P < 0·001) for Przewalski's horses. All locations separated by ≤ 72 h were subsequently given a weight less than 1 based on the following species-specific equation:
(hours since the last locationspecies exponent)/(72 hspecies exponent)
Weighting of locations reduced effective sample size to N = 2400 in Przewalski's horses and N = 1181 in Asiatic wild asses. Weighting locations over longer time periods seemed unnecessarily conservative as animals are highly mobile, and within 72 h, all parts of the individual home range may be accessed, especially because there are no movement barriers. We recalculated the model with a less conservative weighting approach, only down-weighting locations separated by less than 24 h. However, this had little effect on the main results (see Supporting Information Appendix S5A). To check for a possible influence of the different precision of ARGOS and GPS locations on model performance, we recalculated the regression model using only those ARGOS locations with an expected error of ±150 m (see Supporting Information Appendix S4B). Differences in model output were small as compared to the model including all locations, and did not alter the main results; hence, we used the latter in the final results.
All variables were fed into a species-specific mixed-effect logistic regression model. To account for the unbalanced sampling design, we treated individuals as random factors (Gillies et al. 2006) and habitat variables as fixed effects. We selected habitat variables stepwise in a forward fashion, dropping those that failed to be significant or those with elevated Akaike Information Criterion values. For the factor variable ‘plant community’, the effect size of the different communities were first tested relative to the most productive riparian vegetation. With a subsequent Tukey post hoc test, we tested for significant differences in the effect size of each plant community relative to all others, thus creating a plant community ranking scheme. The logistic regression models were performed in the software r (r Development Core Team 2005) with the lme4, MASS, and multcomp packages. All other statistical analyses were done in spss 14·0 (Statistical Package for the Social Sciences; SPSS Inc., Chicago, Illinois, USA). Mean values between species were compared using non-parametric U-tests when the data was not normally distributed, else parametric T-tests. For multiple comparisons of count data, we used Kruskal–Wallis H-tests and subsequent T-tests with Bonferroni corrections. To test for the simultaneous influence of collar type, monitoring period and sex on MCP size, we used a linear regression model. We selected variables stepwise in a backwards fashion, removing those that failed to be significant. For all tests the significance level was set to P < 0·05.
home range size and overlap
Total home range sizes, measured as the 100% MCP, of individual Przewalski's horses and wild asses were significantly different and averaged 471 km2 (range: 152–826 km) for horses and 5860 km2 for asses (range: 4449–7186; T = −13·05, P < 0·001; see Supporting Information Appendix S1). MCP size was independent of the length of the monitoring period, sex and sensor type (ARGOS vs. GPS) in both species (linear regression model, all variables P > 0·05) and it seems reasonable to lump data by species. MCPs started to level out after 5–6 months for both species (see Supporting Information Appendix S5).
Range use was not exclusive and reciprocal home range overlap averaged 60% among individual horses (range: 13–100%) and 84% among individual asses (range: 61–100%; Fig. 2). Wild ass home ranges were almost identical with the boundaries of the Great Gobi B Strictly Protected Area, but excluded the high mountains in the south. Przewalski's horse ranges were confined to the north-eastern corner of the Great Gobi B Strictly Protected Area and also did not include any steep mountain ranges (Fig. 1 and Fig. 2).
selection for different plant communities
Plant community was the strongest predictor for resource selection in both species. Slope, biomass, distance to the nearest water source, and elevation had no or only very limited additional predictive value and the effect was negligible (Table 1). Przewalski's horses showed the expected use of plant communities, suggesting a preference from most to least productive community. Except for the two most productive plant communities, riparian vegetation and Nitraria, all differences were highly significant. Asiatic wild asses, on the other hand, did not use plant communities as expected. Although riparian vegetation was ranked highest, differences to Nitraria and Caragana were not significant and Stipa grassland was ranked second to the last (Table 1, see Supporting Information Appendix S6). In addition, effects were generally small, suggesting little preference for any particular plant community.
Table 1. Results of the species-specific mixed-effect logistic regression models
Ranks of plant community based on post hoc tests***
distance to water, average biomass production and daily travel distance
Distance to the nearest water source was significantly shorter for Przewalski's horses (9·0 ± 2·9 km) than for Asiatic wild asses (13·5 ± 0·9 km; T = 4·12, d.f. = 8·43, P = 0·003). The mean distance to the nearest water source was significantly longer in winter as compared to summer in both species (Przewalski's horses: 10·4 ± 2·7 km in winter, 6·9 ± 1·9 km in summer; T = 2·6, d.f. = 12, P = 0·023; Asiatic wild asses: 15·8 ± 0·9 km in winter, 10·3 ± 2·9 km in summer; T = 4·9, d.f. = 12, P < 0·001). Average biomass production in 8 × 8 km habitat squares used by Przewalski's horses was significantly higher (209·2 ± 22·4 gC m−2 year−1) than in those used by Asiatic wild asses (149·2 ± 4·9 gC m−2 year−1; T = −7·4, d.f. = 7·8, P < 0·001). Mean daily straight line distance between consecutive days was significantly longer for wild asses (8·3 ± 0·7 km) than for Przewalski's horses (3·5 ± 0·9 km; U-test, U < 0·01, P < 0·001).
group sizes and stability
Przewalski's horses were organized in 3–5 female–offspring groups with one dominant stallion and 1–3 bachelor groups of variable composition. Group stability of female–offspring groups was high for adult mares (≥ 3 years) and the dominant stallion, and confirms social organization in stable harem groups. Most changes of adult females occurred when a new harem group formed in 2005 (see Supporting Information Appendix S7). Averaged over all years and groups, a harem group consisted of 1 harem stallion, 5·6 adult mares (range: 3–9), and their offspring (mean total: 11·1, range: 4–23). Single Przewalski's horses were seen in 3%, groups of 2–3 in 5·3%, groups of 4–23 in 87·8%, and groups ≥ 24 (close associations of 1–5 groups) in 3·9% of all group observations.
Between April 2003 and December 2005, we counted 1036 wild ass groups. Mean group size was 28·4 animals (range: 1–1000, median: 5) and was almost identical from year to year (28·1 in 2003, 28·3 in 2004, and 28·8 in 2005; χ2 = 5·73, P = 0·06). Although groups tended to be bigger in winter (mean: 38·8, range: 1–634, median: 10) than in summer (mean: 24·2, range: 1–1000, median: 4; U = 77 185·00, P < 0·001), large groups of several hundred animals were encountered in both seasons. Single individuals accounted for 20%, groups of 2–3 for 20·8%, groups of 4–23 for 38·5%, and groups ≥ 24 for 20·7% of all groups encountered. A statistically sound comparison of group size distributions between species is hindered by the small population size of the Przewalski's horse population and the resulting pseudo-replication in the horse group data.
space and habitat use
The two equid species used the landscape at totally different scales, with the ranges of Asiatic wild asses being 10 times larger than those of Przewalski's horses. Differences are unlikely an artefact of releasing zoo-born animals into the wild because home range sizes of Przewalski's horses in the Great Gobi B Strictly Protected Area are more than 10 times larger than those of horses in the mountain steppe ecosystem of Hustain Nuruu (King & Gurnell 2005; N. Bandi, unpublished data). This suggests that zoo-born Przewalski's horses are able to adapt their spatial use to differences in the local habitat conditions.
Przewalski's horses seem to drink daily (Scheibe et al. 1998). For wild asses, reliable data are lacking, but it is often assumed that they can ‘regularly do without water’ (e.g. Bahloul et al. 2001: p. 320). However, range contraction around water sources (Kaczensky et al. 2007) and shortest distance to the nearest water source during the summer months show that availability of water is an important factor determining space and habitat use for both species. Although average distances to the nearest water source were significantly longer for individual Asiatic wild asses, the high variability among individual Przewalski's horses shows that distance to water is unlikely to be the key factor explaining the spatial difference in range use.
Striking differences also occurred with respect to the use of different plant communities. Przewalski's horses selected plant communities relative to their estimated productivity. Selection was strongest for the two most productive habitat types of the Gobi ecosystem, the riparian vegetation and Nitraria sibirica communities associated with oases, a strategy generally expected when dealing with resource selection by animals in stable environments (Manly et al. 2002). In contrast to Przewalski's horses and to earlier observations made by Feh et al. (2001), Asiatic wild asses do not seem to strongly target the most productive habitat types, but rather use plant communities almost relative to their availability. Avoidance of Przewalski's horses is not a likely explanation for these differences, as both species often graze in close proximity (O. Ganbaatar, unpublished observation).
Then why do Przewalski's horses prefer high-productivity habitats and Asiatic wild asses do not? Apparently, Asiatic wild asses can thrive on lower-quality pastures. They are about 20% smaller than Przewalski's horses (Clark et al. 2006) and need less food to cover their daily energy demand (Nagy 2001). Furthermore, anecdotal observations (Bannikov 1958) and preliminary results of scat composition analysis indicate that Asiatic wild asses make more use of coarse and woody plants than do Przewalski's horses (J. Lengger, unpublished data). Furthermore, the Gobi regions are considered non-equilibrium landscapes, where rainfall patterns and subsequent pasture production are highly stochastic in time and space (Fernandez-Gimenez & Allen-Diaz 1999). Based on annual or multi-annual means, habitat productivity varies among different plant communities. However, a patch of habitat which does not receive rain has little biomass to offer, regardless of its mean productivity. Exceptions are riparian and oases vegetation which profit from the surrounding high mountains providing them with a year-round stable supply of water. In the remaining habitats, productivity changes across different habitat types according to local rainfall patters, resulting in a mosaic of ‘green-up’ patches. We believe that Asiatic wild asses are able to track these green-up patches as has been documented for several other large ungulates in semi-arid and arid ecosystems in Africa (Wolanski et al. 1999; Musiega & Kazadi 2004) and Asia (Mueller et al. 2008). Tracking green-up patches that occur decoupled from a particular plant community necessitate large-scale movements and can be expected to blur selectivity.
As expected, Przewalski's horses in the Great Gobi B Strictly Protected Area live in stable harem groups. Thus, social organization in Przewaski's horse is remarkably stable over a wide range of ecological conditions (Bahloul et al. 2001; King 2002; Wakefield et al. 2002; Zimmermann 2005). These findings are somewhat in contradiction to Moehlman (1998) who postulated that social organization in equids may be primarily habitat-specific. One could, on the other hand, argue that even under the very arid conditions of the Great Gobi B Strictly Protected Area, Przewalski's horses are able to find pockets of suitable habitats that are predictable and stable enough to allow them to maintain a harem social organization.
While we were unable to determine the nature of social organization in Asiatic wild asses, we observed an average group size that was double that of Przewalski's horses, and seasonally large aggregations throughout the year including the mating season. In contrast, close associations of several Przewalski's horse harem groups were only observed during winter, outside of the mating season (O. Ganbaatar, unpublished data). However, differences in the group sizes of the two equids may also be attributed to a factor 20 difference in the overall population size.
Previous studies claimed that Asiatic wild asses lived in family groups because stallions were repeatedly observed herding several mares and their foals (Feh et al. 1994; Feh et al. 2001). However, data from other ungulates show that although one might observe a family-group-type structure, the composition of these ‘family groups’ may well change as females and juveniles move freely among groups (e.g. Sarno et al. 2006). The huge annual ranges, the long flight distances, the relative large size of the population and the uniform coat coloration make it difficult to determine group composition and stability in Asiatic wild asses. We therefore suggest caution in overstressing the results of Feh et al. (1994) as facts, and would rather label them as a working hypothesis.
The habitat-use analysis suggests that Asiatic wild asses are better adapted to cope with unpredictable resource distribution than Przewalski's horses. Under such conditions, a harem-type social organization might not allow sufficient flexibility to quickly locate and fully exploit temporary overabundant green-up patches. We suggest an additional working hypothesis that Asiatic wild asses in the Gobi are more likely to be organized in fission–fusion groups, as observed in the onager Equus hemionus khur (Sundaresan et al. 2007).
consequences for conservation
Given their different resource selection strategies, managers should be aware that protecting habitat where Asiatic wild asses occur does not necessarily benefit Przewalski's horse restoration, whereas setting aside habitat for the conservation of Przewalski's horses will only locally benefit Asiatic wild asses.
The habitat types preferred by Przewalski's horses – riparian and oasis vegetation – constitute only 1·5% of the Great Gobi B Strictly Protected Area. During severe droughts or in areas where water sources regularly dry up or riparian vegetation is minimal (e.g. in the Great Gobi A or the Small Gobi Strictly Protected Areas), we predict that Przewalski's horses will fail to thrive. Thus, we support the view of van Dierendonck & Wallis de Vries (1996) that the Great Gobi B Strictly Protected Area probably represents an edge, rather than representative habitat for Przewalski's horses. The dilemma is that most productive steppe habitats with sufficient water supply are intensively used for livestock grazing (Kaczensky et al. 2006). In these areas, several of the original causes for the species extinction – competition with livestock and interbreeding with domestic horses – are still in place. This leaves only small and isolated pockets of suitable habitat for future re-introductions.
Asiatic wild asses, on the other hand, can make use of a wider variety of habitats in the Gobi. This was most probably the reason why Przewalski's horses became extinct, whereas Asiatic wild asses are still numerous. However, in order to thrive, the species needs access to large tracts of land. Given the stable water resources, the Great Gobi B Strictly Protected Area seems large enough to support and protect a relatively large population. However, the majority of the wild asses live in the south-eastern Gobi, and preliminary data suggest that animals may need access to much larger tracts of land when water availability is more variable (Kaczensky et al. 2006). To guarantee the survival and connectivity of the wild ass population in the entire Gobi, conservation management should aim for a landscape-level approach.
This research was conducted within the framework of the Przewalski's horse re-introduction project of the International Takhi Group (ITG), in cooperation with the Mongolian Ministry of Nature and Environment and the National University of Mongolia. Funding was provided by the Austrian Science Foundation (FWF project P11529 & P14992) and the Austrian National Bank (Jubileums Fonds). We thank R. Samjaa, N. Enkhsaikhaan, D. Lkhagvasuren, J. Lengger, R. Tungalag, ITG staff and the local rangers and their families for their much needed support. J. Hanspach, T. Ruf, F. Knauer and R. Reading provided valuable input and comments on earlier drafts of this manuscript. This work is dedicated to Z. Suchebaatar (1963–2007) a true pioneer of Przewalski's horse re-introductions.