SEARCH

SEARCH BY CITATION

Keywords:

  • climate change;
  • conservation planning;
  • GAMM;
  • HSM;
  • long distance migration;
  • NDVI;
  • poaching;
  • protected areas;
  • Saiga tatarica tatarica;
  • ungulates

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

1. Conservation of declining migratory species is a challenging task, as the factors that may have determined their past distribution may not determine their current and future distribution. Saiga antelope Saiga tatarica populations have massively declined due to poaching. The species is now beginning to recover in Kazakhstan and protected areas are being implemented. Using 25 years of aerial monitoring data, we identified changes in the spring distribution and predicted densities of saiga to prioritize areas for protection under scenarios of climate change together with changes in disturbance and population size. Conserving the spring distribution is critical as spring calving aggregations are of particular importance to population viability.

2. The current distribution is strongly influenced by disturbance, whereas climate had a stronger influence in the past. The area of highly suitable habitat has halved and become fragmented in the last decade. The existing and proposed protected areas are relatively complementary and perform well under most scenarios of future change. However there is a need to widen the geographical scope of protected area planning if potential future high suitability areas are to be effectively protected.

3. Climate change interacts with other factors to determine the distribution of suitable habitat within and outside protected areas. Scenarios in which conservation has increased saiga population size and density tend to show limited impacts of climate change, while scenarios in which the saiga population fails to recover and disturbance continues show, worsening patchiness and reduced suitable habitat.

4.Synthesis and applications. We provide evidence for changing distribution and density of a migratory species over a large spatio-temporal scale, and suggest that future distribution may be more constrained and spatially heterogeneous. These results have important implications for designing future conservation measures for migratory species, such that areas that robustly show high suitability under a range of potential scenarios of change can be included in protected area expansion plans. Protected area placement based only on current, rather than projected distribution risks wasting opportunities for proactive conservation, particularly for a highly disturbed, recovering species likely to be affected by climate change.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Many ungulate migrations worldwide have been disrupted in the last two centuries due to human impacts (Bolger et al. 2008; Harris et al. 2009). Several studies have raised concern over the plight of migratory species and urged the need for proper monitoring and establishment of protected areas (PAs) (Bolger et al. 2008; Harris et al. 2009). However, protection of such species is challenging because the animals occur in large numbers, are frequently on the move and are distributed over vast areas (Mueller et al. 2008). Often, migratory routes vary by season and year, which makes establishment of static PAs challenging. Many migratory ungulates are also subjected to poaching and habitat modification, which cause changes in habitat use (Shuter et al. 2010). The prioritization of survey areas and PA planning can become difficult in such situations, especially for species whose populations have declined substantially and so are potentially not currently occupying typical habitat.

The issue of climate change has received significant attention recently, as it is expected to induce major changes in migratory systems and hence raise further challenges to the management of migratory species (Lundberg & Moberg 2003; Hodgson et al. 2009). Especially in the case of migratory ungulates for which the timing of spring green up is tightly coupled with the onset of the spring migration and dense calving aggregations, climatic changes in temperature and productivity are likely to induce major range shifts during spring (Lundberg & Moberg 2003; Wilcove & Wikelski 2008; Singh et al. 2010a; b). This implies that an understanding of the drivers of the spring distribution of these species is particularly important for conservation planning. Pressey et al. (2007) suggest that effective conservation planning must recognise both the inherent dynamism of ecological processes and the effects of anthropogenic alterations on habitat use and availability. Following on from this, there is a need to integrate uncertainty into spatial predictions of species presence and to make precautionary decisions that avoid underestimating threats. Decisions also need to be revisited regularly in the light of novel and potential threats (Wilson et al. 2007). It is also important to evaluate the effects of conservation on future species distributions, in the light of ongoing environmental change. When planning responses to climate change scenarios, Wiens & Bachelet (2010) urge conservationists to align the scale of conservation action to the scales of climate change projections.

An additional challenge is the uncertainty in identification of important habitats for threatened or declining species. Freckleton, Noble & Webb (2006) explored the likely relationship between population density and occupancy and found a generally positive relationship, but departures from this relationship were potentially informative as to population status. Especially for endangered or declining species, there may not always be a linear relationship between occurrence and abundance. Predicting the spatial relationship between density and occupancy is hence potentially important in evaluating the proportion of the population likely to benefit from particular PAs. Effective stratification of population monitoring also depends upon an understanding of the relationship between the density and distribution of a population.

An excellent example of a species facing these issues is the migratory saiga antelope Saiga tatarica (Linnaeus 1766) of Central Asia, which has experienced a 95% reduction in population size over the last two decades (Milner-Gulland et al. 2001, 2003). Simultaneously, group sizes have decreased (McConville et al. 2009), density has decreased and sex ratio and conception rates (Milner-Gulland et al. 2003) and the location and density of calving aggregations have all been affected by human disturbance (Fry 2004; Singh et al. 2010a). For the last four decades, annual aerial surveys have been conducted over the spring range of saiga in Kazakhstan to estimate population size, density and identify aggregations. Considerable conservation efforts are now being undertaken for saiga on a landscape-scale, mostly within the spring range. For example the Altyn Dala Conservation Initiative (ADCI) is working in Central Kazakhstan, carrying out activities such as research, community engagement, anti-poaching patrols, improvement of monitoring and the designation of new PAs (Klebelsberg 2008). As a result of investment in initiatives such as these by the Government of Kazakhstan and international NGOs, saigas are currently recovering rapidly in this region, and new PAs are being planned and implemented (CMS 2010). Under these circumstances, it is essential that the designation of PAs, which consumes substantial time, resources, and legislative activity, is based on best estimates of the future needs of the species. The regional predictions for Kazakhstan from global climate change models predict precipitation changes and an increase in spring temperature in the region (IPCC 2007). Recent studies based on analysis of time series satellite data also predict a decline in productivity across the saiga range (de Beurs, Wright & Henebry 2009; Zhao & Running 2010). Kazakhstan is currently undergoing rapid development, including new infrastructure, urban growth and human population increases, which are likely to change patterns of human disturbance, and together with increasing wealth, may affect poaching pressure (Kühl et al. 2009). These changes will combine to affect saiga densities and distribution in future, which may render PAs based upon current saiga presence less well targeted than they could be.

Long term aerial survey data provide a unique opportunity to identify changes in spring saiga distribution patterns and to assess the adequacy of different PA plans (Singh et al. 2010a; b). We used the last 25 years of aerial survey data to identify the changing drivers of saiga distribution in Central Kazakhstan. We analysed trends in distribution patterns, explored factors affecting the spring distribution, and then identified areas of high potential saiga density, where monitoring and protection could be focused. We compared density and occupancy predictions to identify areas of mismatch between the two. Since future changes in climate and human activities may cause changes in saiga distribution, we projected the current distribution under a range of scenarios, based on IPCC projections of spring temperature rise and vegetation productivity changes in the region (IPCC – Intergovernmental Panel for Climate Change 2007), in combination with assumptions about the success or failure of current conservation interventions in reducing human disturbance. We then evaluated the location of current and planned PAs with respect to potential future spring saiga distributions. This approach provides a robust basis for the continuing conservation planning process in Central Kazakhstan, as well as providing guidance for similar studies on other migratory species. It fills an important gap in the existing literature on modelling the effects of climate and anthropogenic factors to understand their synergistic impact on species density and distribution.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study area

We consider the Betpak Dala saiga population, located in central Kazakhstan (Fig. S1, Supporting information). This was historically both the largest and the widest ranging of the four populations of sub-species S.t. tatarica, covering an estimated area of about 1·08 million km2. The area is generally flat, covered by treeless steppe, semi-desert and desert vegetation. Vegetation zones occur in a latitudinal gradient. The northernmost steppe zone is followed by semi-desert and desert zones with annual precipitation of around 300 mm, 200–250 mm and <200 mm, respectively. Saigas inhabit all three zones on a seasonal basis (Fadeev & Sludskii 1982; Bekenov, Grachev & Milner-Gulland 1998). Their annual migrations, 600–1200 km in length, are driven by the need for new green pastures in the spring (March–May) and by the presence of deep snow in the steppe zone starting in autumn (September–October). The area in which spring calving was historically observed in this population is 400 km in length north–south and 700 km east–west (Bekenov, Grachev & Milner-Gulland 1998). The period, route, distance and speed of migration differ from year to year depending on climatic conditions, the condition of pastures, the number of watering places and the degree of disturbance experienced by the animals (Bekenov, Grachev & Milner-Gulland 1998; Singh et al. 2010a).

This population was severely and rapidly affected by poaching due to its relative accessibility from the capital Almaty and from the main saiga horn markets in China. The population declined from an estimated 375 000 animals in the 1980s to an estimated 1800 at its nadir in 2003 (although this is likely to be a substantial underestimate; McConville et al. 2009). The 2010 population estimate is 53 400, around 14% of the 1980s size, but substantially larger than only a few years previously (CMS 2010; Fig. S2, Supporting information). This population has been the recipient of significant conservation action, which is likely to be one reason for its rapid recovery.

Aerial survey data

We acquired aerial survey reports from the Institute of Zoology of Kazakhstan for the period 1983–2008, with some years missing (1984, 1990, 1998 and 2000). These reports contain information on the dates and timing of aerial surveys, the areas surveyed, locations of the groups with ∼500 m resolution, group sizes observed and the population estimates made (see Fig. S3, Supporting information for an example of the maps generated). These observations were transferred into a Geographic Information System. The survey technique was developed during Soviet times and remained unchanged throughout the study period. The aerial survey team flew transects over the areas of highest saiga density, determined through local and expert knowledge. Transects were, where possible, 10 km apart and the assumed strip width was 1 km each side of the aircraft, such that 20% of the area was covered. The estimated size and approximate location of each group were marked on a topographic map of scale 1:1 000 000. The team then extrapolated the population estimate by dividing the number of saiga seen by the coverage (usually 0·2) and multiplying by the size of the area of saiga concentration, calculated by drawing a line around the observations and counting grid squares within that line. There was no measure of error.

The population crash towards the end of the 1990s led to a change in the group size distribution, which affected the population estimates. Formerly, saiga aggregated in large numbers during the spring migration, with approximately 40% of the groups containing over 500 individuals (Fadeev & Sludskii 1982). This has changed considerably in recent years, with maximum herd sizes in the low hundreds, and the majority of herds numbering less than 50 (Institute of Zoology, unpublished data). The extent to which these changes have affected the bias in the population estimates is unclear but potentially significant (McConville et al. 2009). Significant issues with the sampling procedures have also recently been identified (Norton-Griffiths & McConville 2007), which are now being addressed (Zuther 2009).

Variables

The spring distribution of saiga is determined by temperature, availability of water and green forage (Bekenov, Grachev & Milner-Gulland 1998; Singh et al. 2010b). We selected mean diurnal range of temperature during the survey period (March–May), vegetation productivity and distance to water and settlements as predictor variables. Singh et al. (2010b) showed for the whole of Kazakhstan that all these variables were important determinants of saiga presence/absence in spring. We used temperature range as it made sense biologically and was a better fit to the data than other temperature metrics. In early spring the maximum temperature may be influential in the onset of spring migration as a result of snow melt, but later in the period, during calving, survival of new born saiga calves is limited by minimum temperature (Milner-Gulland 1994; Bekenov, Grachev & Milner-Gulland 1998). Mean diurnal range of temperature, henceforth ‘temperature’ was estimated from monthly maximum and minimum temperature data for the survey period, obtained from the National Climate Data Center (ftp://ftp.ncdc.noaa.gov/pub/data/ghcn/v2). These data were obtained from various ground stations and were gridded to be used as raster layers with 1 km2 resolution (http://www.esrl.noaa.gov/psd/data/gridded/). We rescaled all the variables to a pixel size of 25 km2, since the scale of the aerial survey is large and groups are spread over vast areas.

We used vegetation productivity instead of precipitation as these two variables are confounded (Singh et al. 2010a). We used the Integrated Normalized Difference Vegetation Index (INDVI) as an index of vegetation productivity (Pettorelli et al. 2006). We acquired NDVI data from 15-day GIMMS (Global Inventory Modelling and Mapping Studies) dataset (8 km2 spatial resolution –ftp://ftp.glcf.umiacs.umd.edu/glcf/GIMMS/Geographic/; Tucker et al. 2005) and from the bimonthly MODIS dataset (for years 2007–08, 1 km resolution; Global Land Cover Facility, http://wist.echo.nasa.-gov/wist-bin/api/ims.cgi?mode=MAINSRCH&JS=1) resampled to the scale of GIMMS data as GIMMS was only available up to 2006. We estimated INDVI as the sum of NDVI from March to mid-May, because the aerial surveys were conducted towards the end of April and first half of May. The straight line distance to the nearest natural water and settlements were estimated for each calving location based on maps acquired from ‘Biogeomencer project’ (http://www.biogeomancer.org/) for water and ‘Global database on Administrative areas’ (http://biogeo.berkeley.edu/gadm/) for settlements. Both temporary and permanent water sources were included, because temporary water sources are full in spring. The straight-line distance to the nearest settlement in the database was regarded as an indicator of the level of disturbance due to poaching pressure. Settlement location has remained consistent in the region over the period, despite changes in settlement size, and poaching pressure is not likely to be straightforwardly frequency dependent (Kühl et al. 2009), hence the size of the settlement was not included.

Saiga distribution modelling

Habitat suitability models (HSMs) apply ecological niche principles, using environmental variables to predict the presence/absence or abundance of a species throughout a study area, with the primary aim of identifying key variables that determine the niche (Hirzel & Le Lay 2008). However, the strength of the distribution-niche linkage depends on the ecology of the species, local constraints and historical events (Hirzel & Le Lay 2008). We hence tested five different modelling approaches, consisting of both presence-pseudo-absence and presence only methods, to select the best approach for modelling annual saiga distributions based on climatic and anthropogenic variables (Table S1, Supporting information; Thuiller et al. 2009). A presence-pseudo-absence approach using logistic Generalized Additive Models (GAM) with a binary response variable and continuous explanatory variables consistently performed better than other methods based on the area under the relative operating characteristic curve (AUC) values (Hastie & Tibshirani 1990; Table S1). We therefore used this approach instead of a presence-only approach such as Ecological Niche Factor Analysis (ENFA) or Maxent as they over-predicted suitable habitat, as has been observed in other studies (Chefaoui & Lobo 2008; Matthiopoulos & Aarts 2010). We generated 2000 pseudo-absences randomly within the annual range area of the Betpak Dala population, of which 1890 were left after removing the points from water bodies.

After estimating the yearly models, we tested whether pooling the data into three periods would be more representative of the changes over the study period. This was done to overcome the high yearly variability in saiga locations, probably a result of the confounded effects of weather and observer bias. The yearly observations were pooled into three periods – Period I: 1986–1996, II: 1997–2002 and III: 2003–2008, representing three different saiga population states: I – high and relatively stable, II – declining and III – low/recovering (Fig. S1). We ran these models using generalized additive mixed models with a cubic spline smoother value of 3 (GAMMS; Wood 2004), year as the random effect and other explanatory variables as fixed effects (Table S2, Supporting information). We accounted for spatial autocorrelation by incorporating ‘X’ and ‘Y’ coordinates of the centroids of the saiga group locations as fixed effects. Spatial predictions were made for all three periods and models were validated using k-folds cross validation (Boyce et al. 2002). We used <0·2, 0·2–0·5 and >0·5 suitability thresholds to classify the predictions into low, medium and high suitability areas. To estimate variable contributions we first estimated the correlation score between the standard predictions vs. the predictions of the model from a randomly permutated variable (Thuiller et al. 2009). We then estimated 1 minus the mean correlation for each variable where a high score meant a high importance. The results were then converted to give a relative importance (summing to one; Thuiller et al. 2009).

Change scenarios

The best model for period III as the current baseline was used to test the effects of potential future changes on saiga distribution. Two dimensions of change were explored; climate change leading to a rise in spring temperature and/or reduced vegetation productivity, and anthropogenic change leading to increased or reduced disturbance and/or increases in population and group size. We manipulated the dataset and environmental variables keeping the model structure constant, to get a general idea of the consequent distributional changes. The main scenarios tested were (Table 1): T+: Temperature + 2·5 °C, modelled by creating a future temperature layer using the predicted rise in maximum spring temperature from the 2007 IPCC report. N−: NDVI −13·5% due to increased droughts, based on predictions for the area in de Beurs, Wright & Henebry (2009) and Zhao & Running (2010), modelled similarly to the T+ scenario. D+: Increased disturbance; in this scenario saigas may move further away from towns, hence each saiga observation was moved 4 km further away from the nearest town than its observed location. The value of 4 km was chosen based on the observed shift of saiga groups away from settlements between periods I and III (Table S2, Supporting information). D−: Disturbance decreases; we decreased the distance of observations from the nearest settlement to Period I levels (i.e. by 4 km). P+: Population recovers to 150 000 (based upon the Government of Kazakhstan’s goal to increase saiga numbers in Kazakhstan to 200 000 in the next 10 years, and because the Betpak Dala population is the only one that is currently increasing rapidly). We increased the frequency of all group sizes in the database equivalently, to mimic an increased population size but the same group size distribution as currently. G+: Population recovers and also becomes less fragmented (based upon the fact that group sizes were very much larger in previous periods than in period III): We increased the population size to 150 000 by increasing the number of saigas in the top 10% of group sizes. Finally, we combined some of the scenarios to give composite scenarios that included the most likely potential combinations of effects. These composites included one that mimicked conservation success, in which disturbance was reduced and the population increased, with a concomitant increase in mean group size, but in the context of climate change (T+P−D−G+), and one mimicking conservation failure, in which disturbance increased and the population stayed at the current level (T+P−D+; Table 1).

Table 1.   Percentage area under three habitat suitability categories of predicted saiga presence for nine different change scenarios
 Period IIIT+N−D+D−P+G+T+ N−T+N−D+T+N−D−G+
  1. Period III is the current situation. T+: Temperature increases by 2·5 °C; N−: NDVI decreases by 13·5%; D+: Disturbance increases by 4 km (i.e. more people or more hunting); D−: Disturbance to 1980s levels (conservation programmes succeed); P+: Population increases to 150 000; G+: Population increases and groups become less fragmented. Combined scenarios include climate change only (T+N−), climate change with increased disturbance (‘conservation failure’; T+N−D+), and climate change with decreased disturbance and a larger, less fragmented population (‘conservation success’, T+N−D−G+).

High16·846·9613·1513·5030·8242·4351·1210·0914·9127·26
Medium42·726·7356·2839·2620·1440·6733·2721·6035·2136·34
Low40·4386·3130·5847·2449·0416·9015·6168·3149·8836·40

Future PA evaluation

To assess the relevance of existing and proposed PAs to potential future distributions, we mapped the two existing and three proposed PAs in the Betpak Dala population’s spring range, based on information from the Association for the Conservation of Biodiversity in Kazakhstan (ACBK; http://www.acbk.kz; Fig. S3) and the United Nations Development Programme’s Steppe Conservation Project (http://www.undp.kz). We overlaid this map over our model predictions for period I (representing healthy saiga populations in the past), period III (representing the current situation) and for the future change scenarios, then identified the suitable habitat falling under each PA type. We considered five PAs of which two already exist, and three are proposed (Fig. S4, Supporting information). The total area covered by these PAs is 20 250 km2, made up of 14 681 km2 of existing PAs and 5568 km2 of proposed PAs, compared to the identified current spring saiga range of about 650 000 km2. We then estimated the proportion of the PA with high (>0·5), medium (0·2–0·5) and low (<0·2) habitat suitability. We also estimated the proportion of each of these suitability types within the PAs vs. in the entire spring range.

Predicting future density

We predicted the spatial distribution of saiga density, to compare the distribution of density and occupancy, understand model limitations and identify which model is a better metric for future scenarios. Predicted relationships for period III and the future scenarios could also be used to guide the allocation of monitoring and protection efforts to areas of high saiga density. The aerial survey data included information about group size and extent from which density could be inferred. However, observed group size distribution from Kazakhstan’s aerial surveys are likely to be severely biased by observation error, artificially reducing the number of small groups in the dataset (McConville et al. 2009). Hence we reconstructed putative original group size distributions using the bias correction algorithm developed by McConville et al. (2009), obtaining an estimate and range of corrected group sizes for each year. We assumed the total population size and then back-calculated the actual group size distribution using log-normal curves. This distribution was then corrected using the detectibility function used by McConville et al. (2009) and was simulated 100 times for each year to be used in the models of density distribution.

Once the bias-corrected group size distribution had been obtained from the original survey data, we fitted a kernel density estimator to our observed point pattern with a smoothing parameter estimated using cross validation (Baddeley 2008; Hengl et al. 2009). We fitted the relative density as a ratio between the local density and maximum density from all locations (Hengl et al. 2009; Hengl 2009). The advantage of using the relative density is that the values are in the range [0, 1], regardless of the bandwidth and sample size. We then used the logistic regression kriging technique from Hengl et al. (2009) to predict saiga group density by: (i) converting the group sizes to logits (if the input values are equal to 0 or 1, replacing them with the second smallest/highest value); (ii) fitting a regression model (GAMM in our case) with temperature, NDVI, distance to towns and water as explanatory variables, based on the variables found important in previous modelling exercises (Singh et al. 2010a,b and the distribution model in this paper); (iii) fitting a variogram for the residuals (logits); (iv) producing predictions by first predicting the regression model part, then interpolating the residuals using ordinary kriging, and finally adding the predicted trend and residuals together; (v) back-transforming the interpolated logits to the original [0, 1] scale. We performed these analyses for period III. The density predictions were then stratified into three strata based on the predicted density of the groups (<0·02, 0·02–0·05, >0·05 animals km-2).

To identify the relationship between predicted density and occupancy of saiga, we used mean density and mean habitat suitability for each pixel and estimated the product–moment correlations (Freckleton, Noble & Webb 2006).

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Saiga distribution modelling

The spring saiga distribution was determined by an intermediate range of temperature, intermediate productivity, areas at intermediate distance from water and away from settlements. The yearly models demonstrate a high variability in the percentage contribution of these variables, with temperature consistently explaining most of the variability, while other variables varied in time (Fig. S5, Supporting information). The periodic models captured the same consistency (Table S2), but were more robust to interannual variation, and so were used for the scenario analyses in preference to the yearly models. In period I, temperature explained most variation in group location (52·8%) followed by NDVI (15·8%; Fig. 1, Table S2). For period II, the trends were similar, with temperature and NDVI still explaining most of the variability (Temperature – 51·3% NDVI – 11·0%), but the contribution of distance to settlements slightly increased (from 2·35 to 6·7%). However during period III, the contribution of temperature and NDVI declined (29·5 and 13·9%, respectively) while distance to settlements increased substantially (22·6%, Fig. 1). Percentage variation explained by spatial autocorrelation of groups was 9·9% and by year, about 15·1%. There was no significant spatial autocorrelation observed in spatial variograms of the residuals. The high Spearman rank correlation coefficients from cross validations for each model showed that the models had good predictive ability (Table S3 Supporting information).

image

Figure 1.  Percentage variable contribution of explanatory variables in each periodic Generalized Additive Mixed Model for the distribution of saiga antelopes in Betapak Dala. Temperature–Mean diurnal range of temperature during the survey period; INDVI – Integrated Normalized Difference Vegetation Index; Dist2water – Distance to nearest water source; Dist2town – Distance to nearest settlement; X,Y – Geographic coordinates of the groups; Year – Year of survey. Period I – 1986–1996; II – 1997–2002, III – 2003–2008.

Download figure to PowerPoint

There was a 50% decline in the amount of high suitability area in period III compared to period I and a slight decline compared to period II. The extent of medium suitability areas increased by about 25% in period III compared to period I (Fig. 2).

image

Figure 2.  Spring distribution of the Betpak Dala population of Kazakhstan in three different time-periods, based on the GAMM models. (a) Period I 1986–1996; (b) II 1997–2002 and (c) III 2003–2008. Black cells represent areas with high suitability (>0·5), dark grey cells – areas with medium suitability (0·2–0·5), and white with low suitability (<0·2). Light grey polygons represent water bodies.

Download figure to PowerPoint

Change scenarios

The predictions from the different scenarios were highly variable both in terms of the quantity and spatial distribution of suitable habitat. A 60% loss in high and 80% loss in medium suitability areas was predicted with rising temperatures, with patchier spatial occurrence of high suitability areas (Table 1, Fig. S6, Supporting information). A decline in productivity had less effect on the total amount of each habitat type, but increased the patchiness of the distribution (Table 1, Fig. S6). Increasing disturbance had a particularly strong effect in the northwest, whereas decreased disturbance lead to consolidation of the distribution, with both high and low probability habitat gaining at the expense of the medium probability habitat. Similar, and more extreme, consolidation within high probability areas was seen in the population increase scenarios, with a reduction in group fragmentation in particular leading to more than half of the occupied saiga range having a high suitability. The climate-only combined scenario, T+N− led to a shift in the distribution to the south and east, with large areas of the northern part of the spring range becoming low probability. Including the anthropogenic factors into either a ‘best’ or ‘worst’ conservation scenario illustrated that population fragmentation and the location of the areas of high probability of presence were strongly dependent on the outcome of conservation, although this effect was modulated by the predicted changes in climate. ‘Conservation success’ led to a homogenous distribution similar to period I, whereas ‘conservation failure’ led to a widely distributed, patchy population. In all three combined scenarios, there was more high probability habitat in the southeast part of the range than currently, and less in the north-west.

Future PA evaluation

The current PA system was more effective at protecting current, rather than past, high suitability areas, with the proportion of current high suitability habitat within the PA system being 23%, compared to 8% under the period I saiga distribution (Table 2, Fig. 3a,b). However it was not substantially better than random placement within the spring range, as 17% of the current spring range was rated as high suitability (Table 2). If the proposed PAs were added, the proportion of current high suitability habitat in PAs would be much improved, to 43%. The current PA system performed much better than random under increased disturbance (62% high suitability habitat), whereas the proposed PAs performed best under reduced disturbance and increased population size. The proposed PAs performed very differently to the existing PAs in some scenarios, which suggested a degree of complementarity which may enhance robustness to uncertainty. For example under reduced productivity, the proposed PAs performed very badly (11%) and existing PAs improved their performance (to 27%), while if the population became larger and less fragmented, proposed PAs remained effective (75%) while existing ones performed very badly (8%). This was due to changing distributions under these different scenarios (Figs. 3c, 3d and S6, Supporting information). The PA system performed relatively poorly under all three of the combined scenarios; in all cases the combined performance dropped below current levels, substantially so in the case of conservation failure (43% high probability habitat currently, 24% projected). The actual coverage of the existing and proposed PA system was low, which was expected given the very large areas involved; the percentage of the total high quality habitat currently covered by the existing PAs was only 1%, with an additional 2% in the proposed PAs (Table S4, Supporting information).

Table 2.   Percentage of high, medium and low probability habitat occurring within current PAs, proposed PAs, and current and proposed PAs combined (‘Total’), compared to the Predicted Total Area of each habitat type in the saiga spring range for Periods I and III and nine future scenarios (see Table 1 for scenario explanations)
 CurrentProposedTotalPredicted total area
HighMediumLowHighMediumLowHighMediumLowHighMediumLow
  1. The values are rounded to the nearest whole number.

Period I84646751313333333293635
Period III23542375250434314174340
T+8092131375105867786
N−384615114444274527133156
D+62231538063521433133947
D−38154688130571429161569
P+23621588130484310174142
G+8385475250333333172756
T+N−38313138063381943163351
T+N−D+312346131375241957102268
T+N−D−G+8464675250333829172954
image

Figure 3.  Maps showing the ADCI Boundary and PAs overlaid on modelled spring distribution of saiga for the (a) past and (b) current situation as well as the (c) conservation success and (d) failure scenarios (see Table 1 for details on scenarios and Fig. S6 for all other scenarios). Black cells show the predicted high suitability, grey – medium and white cells the low suitability areas for saiga.

Download figure to PowerPoint

Predicting future density

The density models from periods I and III produced similar results to the occupancy models, where distance to towns explained most of the variation in the model in Period III. The regression-kriging model explained 91% of the original variation and predicted densities well (Table S5 Supporting information). The maximum observed densities were 0·08 animals km-2 for Period I, with the groups spread in a relatively homogeneous line at a constant latitude (Fig. 4a). The stratified map of current (period III) predicted highest density showed at least three major areas of high density for saiga in the spring range, near the towns of Turgay, Irgiz and Zhezkazgan (Fig. 4b). The high density areas were shifted north and west compared to Period I. For the conservation failure scenario (T+N–D+; Fig. 4c), the predicted density was much lower on average, at 0·02 animals km-2 and the high density areas were more centrally located. In the conservation success scenario (T+N–D–G+; Fig. 4d), with a density of 0·09 animals km-2, the population was similar in distribution to Period III, but with high density areas reappearing in the east. However the results for the conservation success scenario suffer from a degree of circularity, given the assumption of population increase being concentrated in the largest herd sizes.

image

Figure 4.  Predictions of saiga spring density in Betpak Dala based on (a) Period I model (healthy past population); (b) Period III (current situation); (c) a scenario representing conservation success, in which the population recovers and group sizes get larger, with reduced disturbance but the climatic changes still occur (T+N−D−G+); and (d) a scenario representing conservation failure, in which disturbance increases and the population does not grow (T+N−D+). Refer to Table 1 for details on scenarios.

Download figure to PowerPoint

Density–distribution relationships

The distribution of habitat suitability was bimodal for period I, whereas for period III and the two conservation scenarios it was left skewed (Fig. S7, Supporting information). There was a significant positive relationship between mean density and occupancy for all the models (Period I –rs = 0·458, n = 1998, < 0·001; Period III –rs = 0·876, n = 1998, < 0·001; Conservation success –rs = 0·785, n = 1998, < 0·001; Conservation failure –rs = 0. 434, n = 1998, < 0·001). In period III and under the conservation success scenario, the strong correlations were matched by a strong spatial overlap between high density areas and areas with a high probability of occupancy (Fig. S8, Supporting information). However in period I, when the population was large and healthy, there was a substantial swathe of the south of the range which was highly likely to be occupied, but where density was not predicted to be high. In the conservation failure scenario, with more disturbed and fragmented populations, areas of predicted high occupancy were also spatially mismatched with areas of predicted high density, except in core areas in the centre of the range (Fig. S8).

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Saigas show a similar pattern to other long distance migratory ungulates, where the spring migration is driven by a rise in temperatures and an increase in productivity (Bolger et al. 2008; Singh et al. 2010a). Suitable habitat in the spring range is characterized by intermediate temperatures and productivity, is an intermediate distance from water and away from settlements. Temperature is a limiting factor for the survival of new born calves in spring, which may drive the role of temperature in defining suitable habitat (Bekenov, Grachev & Milner-Gulland 1998). With suitable habitat characterized by intermediate productivity, saiga may trade-off forage quantity for quality, as observed in several other migratory ungulates (Hebblewhite, Merrill & McDermid 2008; Mueller et al. 2008; Holdo, Holt & Fryxell 2009). Sites at an intermediate distance from water may indicate avoidance of disturbance from other animals, predators and humans, which also use water bodies. Moreover, availability of water may not be limiting in spring, when precipitation is relatively high. Finally, sites away from settlements and an increase in the role of this variable with time clearly indicates avoidance of areas with high disturbance (Singh et al. 2010b), although it is not possible to determine whether this is because at low population sizes there is no need to approach settlements in order to access high quality habitat, or whether increased disturbance in recent years has driven saigas away from otherwise high quality areas.

As expected, the scenarios of future change lead to significant predicted changes in saiga spring distribution. Climate change was predicted to reduce the extent of highly suitable habitat, and produce a more heterogeneous pattern of habitat. It was also predicted to lead to shifts in highly suitable habitat with increases in some areas (especially the south east) and decreases in others (the northwest). This may have implications for migratory routes, foraging behaviour and selection of favourable areas for calving. An indirect effect on population productivity might be expected. These likely effects of climate change are similar to those predicted for other migratory species (Lundberg & Moberg 2003; Shuter et al. 2010), but in our study we have been able to characterise potential changes spatially and in more detail than is commonly possible. It has also been instructive to consider how conservation outcomes may interact with climate change; successful conservation tends to mitigate the effects of climate change, while conservation failure exacerbates the patchiness and loss of suitable habitat. For the saiga, as is likely for many similar species (Harris et al. 2009), disturbance currently overrides the contribution of other variables. This suggests that a conservation priority should be to reduce the effects of human disturbance in otherwise highly suitable areas, for example by reducing poaching levels, limiting human use of natural watering places or restricting entry to areas of high saiga density such as calving aggregations.

Currently, only a few PAs exist in the vast spring range of the Betpak Dala saiga population, covering only 1·05% of the current high suitability areas (Fig. S3). This is a typical picture for migratory species which cover large areas during their long-range movements (Harris et al. 2009). A particular concern for saigas is that the current and planned PAs are concentrated in the north-west of the spring range, where saiga is currently at high density. However, the eastern parts of the range were important in the 1980s and could be again, as populations increase. The eastern areas were particularly prominent in our more realistic combined scenarios. It is important to aim for geographic representativeness when planning PAs (Hodgson et al. 2009), and our analyses demonstrate that if making PAs robust to plausible future scenarios is also an aim (Pressey et al. 2007; Hannah 2008), a priority should be to improve PA coverage in areas of past and predicted high density in the east.

Our study only considers the spring range, prior to calving, because this is when the available data were collected. The winter range of the saiga is under-represented in the PA system at present with only one PA, and no new PAs proposed for this area. However, the winter period is also critical for the saiga, as the winter range is more limited in extent than the other seasonal ranges and is shared with livestock, with the potential for food limitation as well as direct mortality from harsh weather (Sludskii 1963; ; Bekenov, Grachev & Milner-Gulland 1998; Coulson, Milner-Gulland & Clutton-Brock 2000). There is an urgent need for similar analyses to be conducted within the winter range, but this requires data on saiga distribution at this time of year, which are currently lacking. GPS collars could provide an effective way to gather this data, given the difficulties of aerial and ground surveys in the winter months.

The recent and potential increases in patchiness of saiga distributions have important implications for monitoring, since surveying patchy distributions entails increased cost and lower reliability. In such a situation, our predicted density maps may be useful in guiding monitoring efforts towards high density areas. Our results showed a significantly positive relationship between predicted density and occupancy for a migratory species in a variety of circumstances ranging from a healthy and abundant population phase to a critically endangered low density phase. However, the spatial mismatches observed in the healthy population phase and conservation failure scenario showed that such relationships may not follow a consistent pattern at large spatial and temporal scales, and as population status changes. Freckleton, Noble & Webb (2006) showed that when habitat suitability was left-skewed, density–occupancy relationships were positive and linear, whereas for bimodal distributions, the relationships were weak; this finding was echoed in our study. Models of habitat suitability based on data collected from disturbed populations may be misleading if density and occupancy are assumed to be proxies for each other. Overall, however, the positive relationships between density and occupancy which we found suggested that either may be useful when designing PAs or monitoring strategies.

Management implications

Work of the type presented here is crucial in providing baselines for landscape-scale conservation initiatives. An approach such as ours may ensure that monitoring remains cost effective and that PA networks are robust to future changes in distribution and densities of key target species. Here we have used the available data to model the likelihood of saiga presence and density in a particular season in a spatially explicit manner. This is a fundamental requirement for embarking on a systematic conservation planning process. We have also made some extrapolations to the future, based upon realistic scenarios. These are crude, and assume that the structure of the models governing current distribution remains the same as conditions change and as we extrapolate away from the data. However they are the best we have currently available and can give general indications of the effectiveness of current and proposed PAs for saiga conservation at the landscape and decadal scales.

This approach also has broad resonance for the conservation of similar long distance migratory species. A dynamic predictive approach that both assesses past and current drivers of distribution and predicts future distribution under plausible scenarios is more robust than relying on current status. It can provide insights into the contrasting drivers of current, potentially highly disrupted and fragmented distributions, and of healthier populations in the past. One potential application of the scenario-based approach is evaluation of conservation interventions against a spatially explicit set of objectives. For example, a PA network might be planned to ensure that it contains more than a threshold level of predicted high quality habitat under a range of likely scenarios of environmental and anthropogenic change.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We would like to thank the Institute of Zoology, Almaty, Kazakhstan for providing access to their long term datasets and especially Dr Iu. A. Grachev and Dr A.B. Bekenov for hosting NJS in Almaty. This work was supported by the Leverhulme Trust and by a Royal Society Wolfson Research Merit Award to EJMG. Thanks to Joaquin Hortal, Miguel Á Olalla-Tárraga, Steffen Zuther and Nils Bunnefeld for helpful comments. The editors and one anonymous reviewer also provided valuable comments that greatly improved the manuscript.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Baddeley, A. (2008) Analysing Spatial Point Patterns in R. CSIRO, Canberra, Australia.
  • Bekenov, A.B., Grachev, I.A. & Milner-Gulland, E.J. (1998) The ecology and management of the saiga antelope in Kazakhstan. Mammal Review, 28, 152.
  • de Beurs, K.M., Wright, C.K. & Henebry, G.M. (2009) Dual scale trend analysis for evaluating climatic and anthropogenic effects on the vegetated land surface in Russia and Kazakhstan. Environmental Research Letters, 4, 045012. doi:10.1088/1748-9326/4/4/045012.
  • Bolger, D.T., Newmark, W.D., Morrison, T.A. & Doak, D.F. (2008) The need for integrative approaches to understand and conserve migratory ungulates. Ecology Letters, 11, 6377.
  • Boyce, M., Vernier, P., Nielsen, S. & Schmiegelow, F. (2002) Evaluating resource selection functions. Ecological Modelling, 157, 281300.
  • Chefaoui, R.M. & Lobo, J.M. (2008) Assessing the effects of pseudo-absences on predictive distribution model performance. Ecological Modelling, 210, 478486.
  • CMS (2010) Overview Report for the Second Meeting of the Signatories to the CMS MOU on Saiga Conservation. Convention on Migratory Species, http://www.cms.int. Bonn, Germany.
  • Coulson, T.N., Milner-Gulland, E.J. & Clutton-Brock, T. (2000) A comparison of the relative roles of density and climatic variation on fecundity rates in three contrasting ungulate species. Proceeding of the Royal Society of London, B, 267, 17711779.
  • Fadeev, V.A. & Sludskii, A.A. (1982) Saigak v Kazakhstan, Alma-Ata, Kazakhstan: Academy of Sciences, USSR, pp. 160. [The Saiga in Kazakhstan].
  • Freckleton, R.P., Noble, D. & Webb, T.J. (2006) Distribution of habitat suitability and the abundance-occupancy relationship. The American Naturalist, 167, 260275.
  • Fry, M. (2004) The status of the Saiga antelope in the Ustiurt region of western Kazakhstan. MSc Thesis, Imperial College, London. Available online at http://www.iccs.org.uk.
  • Hannah, L. (2008) Protected areas and climate change. Annals of the New York Academy of Sciences, 1134, 201212.
  • Harris, G., Thirgood, S., Hopcraft, G., Cromsigt, J.P.G.M. & Berger, J. (2009) Global decline in aggregated migrations of large terrestrial mammals. Endangered Species Research, 7, 5576.
  • Hastie, T.J. & Tibshirani, R. (1990) Generalized Additive Models. Chapman and Hall, London.
  • Hebblewhite, M., Merrill, E.H. & McDermid, G. (2008) A mutli-scale test of the forage maturation hypothesis for a partially migratory montane elk population. Ecological Monographs, 78, 141166.
  • Hengl, T. (2009) A Practical Guide to Geostatistical Mapping, 2nd edn. University of Amsterdam, Amsterdam. pp. 291, http://www.lulu.com .
  • Hengl, T., Sierdsema, H., Radovic, A. & Dilo, A. (2009) Spatial prediction of species’ distribution from occurrence-only records: combining point pattern analysis, ENFA and regression-kriging. Ecological Modelling, 220, 34993511.
  • Hirzel, A. & Le Lay, G. (2008) Habitat suitability modelling and niche theory. Journal of Applied Ecology, 45, 13721381.
  • Hodgson, J.A., Thomas, C.D., Wintle, B.A. & Moilanen, A. (2009) Climate change, connectivity and conservation decision making: back to basics. Journal of Applied Ecology, 46, 964969.
  • Holdo, R.M., Holt, R.D. & Fryxell, J.M. (2009) Opposing rainfall and nutrients gradients best explain the wildebeest migration in the Serengeti. The American Naturalist, 173, 431445.
  • IPCC – Intergovernmental Panel for Climate Change (2007) Fourth Assessment Report on Climate Change. http://www.ipcc.ch/ipccreports/assessments-reports.htm .
  • Klebelsberg, E. (2008) The Altyn Dala Conservation Initiative: a long term commitment to save the steppe and its saigas, an endangered couple. Saiga News, 8, 12.
  • Kühl, A., Mysterud, A., Grachev, I.A., Bekenov, A.B., Ubushaev, B.S., Lushchekina, A.A. & Milner-Gulland, E.J. (2009) Monitoring population productivity in the saiga antelope. Animal Conservation, 12, 355363.
  • Lundberg, J. & Moberg, F. (2003) Mobile link organism and ecosystem functioning – implications for ecosystem resilience and management. Ecosystems, 6, 8798.
  • Matthiopoulos, J. & Aarts, G. (2010) The spatial analysis of marine mammal abundance. Marine Mammal Ecology and Conservation: A Handbook of Techniques (eds I.L. Boyd, D. Bowen & S. Iverson), pp. 6897. Oxford University Press, Oxford.
  • McConville, A.J., Grachev, Iu.A., Keane, A., Coulson, T., Bekenov, A.B. & Milner-Gulland, E.J. (2009) Reconstructing the observation process to correct for changing detection probability of a critically endangered species. Endangered Species Research, 6, 231237.
  • Milner-Gulland, E.J. (1994) A population-model for the management of the saiga antelope. Journal of Applied Ecology, 31, 2539.
  • Milner-Gulland, E.J., Kholodova, M.V., Bekenov, A.B., Bukreeva, O.M., Grachev, Iu.A., Amgalan, L. & Lushchekina, A.A. (2001) Dramatic declines in saiga antelope populations. Oryx, 35, 340345.
  • Milner-Gulland, E.J., Bukreevea, O.M., Coulson, T., Lushchekina, A.A., Kholodova, M.V., Bekenov, A.B. & Grachev, I.A. (2003) Reproductive collapse in saiga antelope harems. Nature, 422, 135.
  • Mueller, T., Olson, K.A., Fuller, T.K., Schaller, G.B., Murray, M.G. & Leimgruber, P. (2008) In search of forage: predicting dynamic habitats of Mongolian gazelles using satellite based estimates of vegetation productivity. Journal of Applied Ecology, 45, 649658.
  • Norton-Griffiths, M. & McConville, A.J. (2007) Technical evaluation of the aerial survey programme for saiga antelope (Saiga tatarica) with recommendations for improvements. Final Report – ZGF-Project No. 1293/02 Saiga Schutz Kasachstan/Evaluation der Tierzahlungen. Frankfurt Zoological Society, Frankfurt, Germany.
  • Pettorelli, N., Gaillard, J.M., Mysterud, A., Duncan, P., Stenseth, N.C., Delorme, D., van Laere, G., Toigo, C. & Klein, F. (2006) Using a proxy of plant productivity (NDVI) to find key periods for animal performance. Oikos, 112, 565572.
  • Pressey, R.L., Cabeza, M., Watts, M.E., Cowling, R.M. & Wilson, K.E. (2007) Conservation planning in a changing world. Trends in Ecology & Evolution, 22, 583592.
  • Shuter, J.L., Broderick, A.C., Agnew, D.J., Jonzen, N., Godley, B.J., Milner-Gulland, E.J. & Thirgood, S. (2010) Conservation and management of migratory species. Animal Migration: A Synthesis (eds E.J. Milner-Gulland, J. Fryxell & A.R.E. Sinclair), pp. 172206. Oxford University Press, Oxford.
  • Singh, N.J., Grachev, I.A., Bekenov, A.B. & Milner-Gulland, E.J. (2010a) Tracking greenery in central Asia – the migration of the saiga antelope. Diversity and Distributions, 16, 663675.
  • Singh, N.J., Grachev, I.A., Bekenov, A.B. & Milner-Gulland, E.J. (2010b) Saiga antelope calving site selection is increasingly driven by human disturbance. Biological Conservation, 143, 17701779.
  • Sludskii, A.A. (1963) Dzhuts in Eurasian steppes and deserts. Trudi Instituta Zoologii AN KazSSR , [Proceedings of the Zoological Institute of the Kazakhstan Academy of Sciences] , 20, 588.
  • Thuiller, W. (2003) BIOMOD: Optimising predictions of species distribution and projecting potential future shifts under global change. Global Change Biology, 9, 13531362.
  • Thuiller, W., Lafourcade, B., Engler, R. & Araújo, M.B. (2009) BIOMOD – a platform for ensemble forecasting of species distribution. Ecography, 32, 369373.
  • Tucker, C.J., Pinzon, J.E., Brown, M.E., Slayback, D., Pak, E.W., Mahoney, R., Vermote, E. & ElSaleous, N. (2005) An extended AVHRR 8-km NDVI data set compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing, 26, 44855598.
  • Wiens, J.A. & Bachelet, D. (2010) Matching the multiple scales of conservation with the multiple scales of climate change. Conservation Biology, 24, 5162.
  • Wilcove, D.S. & Wikelski, M. (2008) Going, going, gone: is animal migration disappearing? PLoS Biology, 6, 13611364.
  • Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer, K.R., Murdoch, W.W., Reyers, B., Wardell-Johnson, G., Marquet, P.A., Rundell, P.W., McBride, M.F., Pressey, R.L., Bode, M., Hoekstra, J.M., Andelman, S., Looker, M., Rondinni, C., Kareiva, P., Shaw, M.R. & Possingham, H. (2007) Conserving biodiversity efficiently: what to do, where, and when. PLoS Biology, 5, 18601861.
  • Wood, S.N. (2004) Stable and efficient multiple smoothing parameter estimation for generalized additive models. Journal of the American Statistical Association, 99, 673686.
  • Zhao, M. & Running, S.W. (2010) Drought-induced reduction in global terrestrial net primary production from 2000 through 2009. Science, 329, 940943.
  • Zuther, S. (2009) Aerial census of the Betpak Dala population of saiga antelope. Technical Report, Altyn Dala Conservation Initiative. Association of Conservation of Biodiversity of Kazakhstan, Kazakhstan.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Figure S1. Map of Kazakhstan showing the annual ranges of three Saiga populations: Betpak Dala (71° 24’ E, 48° 7’ N), Ustiurt and Ural.

Figure S2. Estimates of the size of the Betpak Dala population of Kazakhstan from the spring aerial surveys for the three periods I, II and III. Y axis represents saiga numbers. Data from the Institute of Zoology, Kazakhstan.

Figure S3. Past aerial survey maps of saiga population surveys from Institute of Zoology, Almaty, Kazakhstan, showing flight paths and the location and size of saiga aggregations. The years shown here are a) 1995 and b) 2001.

Figure S4. Map showing the key protected areas (PAs) of different categories and status in the study area. Map adapted from the UNDP project on steppe conservation (2010). The Green areas show the existing PAs and the blue Proposed PAs. The thick black line is the boundary of the Altyn Dala Conservation Initiative. The three blue areas are proposed to be combined into the ‘Altyn Dala’ State Nature Reservat.

Figure S5. Percent contribution of explanatory variables in each yearly GAM for the distribution of saiga antelopes in the Betapak Dala population. NDVI- Normalized Difference Vegetation Index; Dist2water- Distance to nearest water source; Dist2towns- Distance to nearest settlement; Temperature – Mean diurnal range of temperature during the survey period.

Figure S6. Maps showing the Altyn Dala Conservation Initiative Boundary and protected areas (red lines) overlaid on modelled spring distribution of saigas under 9 different scenarios. Black cells show the predicted high suitability, grey the medium and white cells the low suitability areas for saigas. See Table 2 for the details on the scenarios.

Figure S7. Distribution of habitat suitability for saiga antelope under four different scenarios, where habitat suitability is defined as the probability of occupancy. Period I, III and Conservation Success and Failure scenarios.

Figure S8. Spatial overlap between occupancy (grey; probability >0.5) and predicted density (black; >0.05 animals/km2) of saiga antelopes in Kazakhstan.

Table S1. Cross validation score for the best model for each survey year from BIOMOD (Thuiller 2003, Thuiller et al. 2009).

Table S2. Estimates and standard errors for the variables used in GAMM models of saiga antelope distribution in Betpak Dala, Kazakhstan for each period. Temperature = Mean diurnal range of temperature. Period I- 1986-1996; II- 1997-2002, III – 2003-2008.

Table S3. Cross-validated Spearman rank correlations (rs) between final GAMM model bin ranks and area-adjusted frequencies for individual and average model sets for each decadal model.

Table S4. Cross validation score for the period III regression kriging model

Table S5 Proportion of the total amount of High, Medium and Low suitability habitat in the spring saiga range that occurs within the ‘Current’ and ‘Proposed’ protected areas, and in the PA system as a whole (‘Total’), for different scenarios of change (see Table 1 for scenario details and Table 2 for the percentage of each habitat type within the protected areas).

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

FilenameFormatSizeDescription
JPE_1905_sm_FigS1-8-TableS1-5.doc3689KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.