The fundamental hominin niche in late Pleistocene Central Asia: a preliminary refugium model

Authors


Abstract

Aim

We examine hominin presence in Central Asia during the late Pleistocene in order to identify the abiotic characteristics that best predict site distribution during interglacial and glacial periods. Our goal is to build a preliminary framework for climate-mediated hominin dispersals in this understudied part of the Old World.

Location

Tajikistan, Uzbekistan.

Methods

We developed an ecological niche model using presence-only data to explain the spatial relationship of abiotic variables and hominin locations (= 15) during glacial–interglacial transitions. The model was evaluated using the Cramér–von Mises goodness-of-fit statistic and empirical K-function.

Results

Hominin locations were spatially aggregated during both glacial and interglacial periods. Of the abiotic variables analysed on a small scale (30-m resolution), only distance to water differed significantly between glacial and interglacial periods, although most locations were within 5 km of water. At a coarse scale (5-km resolution), hominin locations appear to have been constrained by low temperatures during glacial periods, but not during interglacials.

Main conclusions

Hominin groups did not abandon Central Asia during colder periods. This suggests one of three possibly complementary scenarios: (1) late Pleistocene hominin groups had a more flexible behavioural repertoire than previously anticipated and were able to buffer climatic instability culturally; (2) our study area was not as hostile an environment as traditionally considered; and (3) the area examined here represents a refugium during late Pleistocene glaciations.

Introduction

The archaeological record suggests that hominins first appeared in Central Asia roughly 800,000 years ago and intermittently inhabited the region during the middle and late Pleistocene. With respect to the late Pleistocene, explanatory models of hominin dispersals within Central Asia during periods of climatic instability are mostly anecdotal and rely on single lines of evidence such as the location and technotypological character of lithic assemblages and how these track against coarse-scale biome reconstructions (e.g. Nat, 1971; Davis, 1990; Schäfer et al., 1998; Davis & Ranov, 1999; Mangerud et al., 2004).

The aim of this study is to examine the degree to which hominin site locations are aggregated on the landscape, and identify the abiotic variables that best predict the observed pattern during both glacial and interglacial conditions. Our ultimate goal is to provide direction for future fieldwork so that biogeographical and archaeological data sets from the region match the quantity and quality of those from other, better-explored parts of the Old World. The significance of this endeavour is underscored by recent genetic and archaeological evidence that has identified three metapopulations of late Pleistocene hominins (Neanderthals, Denisovans and modern humans) within Central Asian landscapes (Krause et al., 2007, 2010; Reich et al., 2010). Once robust biogeographical models are framed and tested, it will be possible to better characterize hominin population structure in Central Asia and perhaps use these insights to examine the context of the disappearance of the Neanderthals not only in Europe but across their range.

It has been assumed that Central Asian environments were harsh and probably inhospitable to hominins during much of the last two glacial periods, oxygen isotope stages (OIS) 4 and 2 (e.g. Vishnyatsky, 1999; Glantz, 2010). Existing biogeographical models emphasize either dispersals out of greater Central Asia (Ranov & Davis, 1979; Davis & Ranov, 1999) or to preferred biomes – possibly refugia – within the region during the coldest and driest phases of the last 100,000 years (Nat, 1971; Mangerud et al., 2004). Whether the mountainous zones, foothills or the inner deserts/plains were differentially inhabited during cold periods and/or transitional phases requires further testing. However, recently reported radiometric dates from three hominin sites in Uzbekistan clearly indicate that hominins did not disperse out of the region during climatic downturns (Glantz et al., 2008; Mallol et al., 2009; Kolobova et al., 2011).

The first step in model-building is to falsify the null hypothesis, which states that hominin sites are randomly distributed across the landscape. If overturned, an exploration of which abiotic variables best predict the distribution of interglacial versus glacial sites is warranted. To describe the abiotic requirements of the hominin fundamental niche, we use concepts from ecological niche modelling (ENM) to construct a testable framework of hominin dispersal and distribution within Central Asia. The framework is based on the synthesis of multiple lines of evidence and their analysis within a geographical information system (GIS).

The late Pleistocene hominins of Central Asia: whose distribution is being modelled?

Genetic and archaeological evidence suggests the roughly synchronous presence of three hominin metapopulations in greater Central Asia: Neanderthals at Okladnikov Cave; Denisovans, Neanderthals and modern humans at Denisova Cave in the Siberian Altai; and Neanderthals at Teshik-Tash Cave in Uzbekistan (Krause et al., 2007, 2010; Derevianko et al., 2008; Reich et al., 2010; Mednikova, 2011; Meyer et al., 2012). These groups might not, however, have been on the landscape at the same time. Here, we treat Central Asian hominins as metapopulations, because genetic evidence supports some degree of intermixture between them, and they seem to be spatially structured (Krause et al., 2010; Reich et al., 2010). The re-analysis and interpretation of the Teshik-Tash child and the description of the Obi-Rakhmat Grotto hominin as reflecting mosaic morphologies that depart from the western European Neanderthal pattern (Glantz et al., 2008, 2009) also offer morphological support for the genetic evidence of intermixture. However, it is presently impossible to identify these three groups on the landscape by bony morphological and/or archaeological criteria alone.

With regard to the archaeological record, some scholars have suggested that Middle and Upper Palaeolithic industries represent an in situ technological transition. These industries, however, cannot be attributed exclusively to any hominin group from the region (Rybin, 2005; Shunkov, 2005; Derevianko et al., 2008). It appears that archaeological models used to explain the local extinction of Neanderthal populations in western Europe (d'Errico & Sánchez Goñi, 2003; Banks et al., 2008) cannot be used in Central Asia. Here, we take the approach that hominin population structure should be modelled according to the resilience of source and sink populations of a single interbreeding group during the late Pleistocene (e.g. Hublin & Roebroeks, 2009; Dennell et al., 2011). We therefore make no assumptions regarding either the potential affinity of Central Asian hominins or the authorship of the lithic assemblages that define the hominin locations of our interglacial and glacial samples.

Existing evidence for a Central Asian refugium during the late Pleistocene

The genetic and putative morphological and archaeological diversity reflected in the Middle-to-Upper Palaeolithic record of Central Asia may be interpreted through the lens of non-human mammalian biogeographical research from the same region. Studies of Coelodonta antiquitatis (woolly rhinoceros) (Kahlke, 1999; Kahlke & Lacombat, 2008), Capreolus pygargus (Siberian roe deer) (Vorobieva et al., 2011) and Microtus oeconomus (tundra vole) (Galbreath & Cook, 2003) support the notion of endemism and extensive intraspecific variability during the late Pleistocene. For instance, several taxa (i.e. Saiga, Coelodonta) of the greater MammuthusCoelodonta faunal group originated on the steppe of continental Central Asia and underwent a series of small-scale dispersal and divergence events within the region before dispersing across Eurasia (Kahlke, 1999; Lister et al., 2005). Additionally, multiple replacements of Capreolus pygargus occurred in the Altai throughout the last glacial–interglacial transition, and some of the ancestral haplotypes point to an origin from populations in the Tien Shan Mountains (Vorobieva et al., 2011). Furthermore, studies of Microtus oeconomus suggest that the extant Beringian clade was colonized by a small founder population in Central Asia that dispersed following glacial retreat (Galbreath & Cook, 2003). These lines of evidence generally support the expectations described by Stewart & Stringer (2012) regarding the impact of refugia on the pattern of diversification and coalescence in species.

Defining Central Asia and its contemporary climate

Central Asia is delineated from often contradictory political, cultural, historical and physio-geographical criteria. The region is commonly understood to consist of the former Soviet republics of Kazakhstan, Uzbekistan, Turkmenistan, Tajikistan and Kyrgyzstan; this is the definition used herein (Fig. 1).

Figure 1.

Map showing the distribution of stratified glacial and interglacial hominin locations in the study area. Hominin locations (= 15) are distributed within the current political boundaries of Uzbekistan and Tajikistan. Maximum glacial extent is derived from Kuhle (2004).

Central Asia is ecologically diverse, with juxtaposed steppe, high mountain ranges and arid plains. The contemporary climate of the region is the result of the permanent high pressure ridge systems created by westward expansion of the Siberian Highs in the winter and the Azores Highs in the summer, and the impact of the topography of the region on these systems (Mestdagh et al., 1999; Koppes et al., 2008; Feng et al., 2011). The study area examined here is under the influence of westerlies and is not affected by the summer monsoon, due to an orographic divide created by the Tien Shan and Pamir mountains (Koppes et al., 2008; Feng et al., 2011). Annual temperature and precipitation varies considerably throughout the study area. For instance, high precipitation (> 1000 mm yr−1) characterizes the northern and western portions of the Tien Shan, with decreasing precipitation following a southward gradient (Zech, 2012). Annual precipitation in the Kyzyl Kum desert and surrounding plains is roughly 200 mm, with extensive daily and seasonal temperature fluctuations (Mestdagh et al., 1999). Precipitation mostly falls during the winter and spring (Dushanbe: January, 72.2 mm, July, 2.4 mm; Samarkand: January, 43.9 mm, July, 4.3 mm), with broadly fluctuating annual temperatures (Dushanbe: January −2.4/8.1 °C, July 18.3/35.1 °C; Samarkand: January −3.6/6.2 °C, July 17.8/33.9 °C).

Climate regimes of the last glacial period: is Central Asia unique?

A full examination of the effects of climatic change and tectonic activity on the Central Asian Palaeolithic is beyond our remit (see Glantz, 2010). Instead, we describe current research on mountain glaciation that provides a more precise account of the timing and intensity of maximum glacial extents and the equilibrium line altitude (ELA) in the Tien Shan, Pamir and Quin Lin mountains (Fig. 1). This work has a direct bearing on climate reconstructions and water availability in the region during the late Pleistocene.

Substantial efforts have been devoted to the reconstruction and interpretation of the environmental history of the Central Asian mountain systems (e.g. Grosswald et al., 1994; Ding et al., 2002; Koppes et al., 2008) and the degree to which climate change there has mirrored that of Europe and East Asia. Two major processes contributed to local climate change throughout the Pleistocene in Central Asia: (1) irreversible aridification; and (2) significant orogeny, resulting in favourable conditions for glacier expansion (Merzlyakova, 2003). A curve reflecting short-period palaeoclimatic events recorded in the Central Asian loess-soil sections correspond to those from the loess sections of China (Liu, 1985; Ding et al., 2002), the glacial–interglacial sequence of Europe (Kukla, 1978; Bronger, 2003; Molodkov & Bolikhovskaya, 2006), and the oxygen isotope curve. However, the timing and intensity of palaeoclimatic fluctuations in Central Asia differ from those experienced in Europe and East Asia (Owen et al., 2005; Wei et al., 2006). First, the maximum glacial extent probably occurred during the First Glacial Maximum (FGM) (Merzlyakova, 2003). Second, recent regional surveys have suggested that the local Last Glacial Maximum in Central Asia (LGML) was asynchronous with the global LGM (LGMG) in terms of initial onset (Finkel et al., 2003; Ehlers & Gibbard, 2004; Owen et al., 2005; Wei et al., 2006; Koppes et al., 2008).

The LGML in the Kyrgyz Tien Shan appears to have occurred c. 37 ka or earlier (Koppes et al., 2008), placing it within OIS 3. A similar pattern is reported from the Tibetan Plateau, as well as the Hunza and Chitral regions in Pakistan (Benn & Owen, 1998; Finkel et al., 2003). An early LGML also agrees with 10Be and 26Al ages for moraines south of the Tarim Basin in the Karakax Valley of the Kunlun Mountains (Clark et al., 2001). It is therefore likely that an early onset of the LGML influenced the hominin distribution examined in the present study, and we rely on local chronologies of glacial advance to identify our sites as glacial, interglacial or both.

Materials and methods

The study area and hominin location information

The hominin locations sampled here (= 15 Palaeolithic sites) are from a relatively well-circumscribed area within Central Asia, in Uzbekistan and Tajikistan (Fig. 1). This represents a small part (240,000 km2) of greater Central Asia, but captures all of the ecosystem types identified as potential hominin niches in previous studies (e.g. Davis, 1990; Mangerud et al., 2004). Site locations are dispersed within the Tajik depression, several small mountain ranges associated with the greater Tien Shan and Pamir, and along the plains and foothill zones between the Kyzyl Kum desert to the west and the mountain systems to the east (Fig. 1).

The sites examined herein (Table 1) represent a subset of those reported in the literature and were chosen because they are associated with a chronological marker and are stratified. Although fewer than half of the sites (= 7) have been dated using radiometric techniques, the remaining sites are associated with coarse-scale chronological markers, such as those derived from biostratigraphy and/or fauna (e.g. Gromova, 1940; Bibikova, 1958), pollen (e.g. Ranov et al., 1973; Ranov & Nesmeyanov, 1973), or geomorphological attributes (e.g. Tetyukhin et al., 1978). These data provide enough resolution to place each site in either a glacial or an interglacial period. Some sites were attributed to a specific climatic period based on the technotypological similarity of their lithic assemblages to sites that had a more robust chronological marker. For example, Karak-Kumy has a very similar lithic industry to Khudji, and Khudji has a radiometric date (Vishnyatsky, 1999; Trinkaus et al., 2000). Additionally, Shugnou's chronological attribution is due to its technotypological similarities to Dodekatym-2 (Kolobova et al., 2011; Ranov et al., 2012). Although this method is certainly fraught with potential error, it is standard practice in the archaeological literature in the absence of radiometric dating. Descriptions and interpretations of lithic assemblages from Central Asia are rife with relative ‘cultural’ dating (Kolobova et al., 2011; Ranov et al., 2012). Our goal, however, was not to create a specific ‘date’ for the site, but instead to place it within a glacial or interglacial palaeoclimatic window.

Table 1. List of stratified hominin locations in the study area of Uzbekistan and Tajikistan (= 15) with relevant information concerning chronological attribution and archaeological deposits. Hominin locations that are dated using radiometric techniques are indicated with an asterisk (*), and dates that do not require calibration are in bold type. Latitude/longitude coordinates are in WGS 1984 geographical coordinate system
Hominin locationLatitudeLongitudeTechnotypologyDate (yr bp)Date (cal. yr bp)Dating techniqueaClimate periodBiomeReference
  1. a

    AMS, accelerator mass spectrometry; ESR, electron spin resonance; OSL, optically stimulated luminescence; U-series, uranium-series; UT, uranium–thorium; 14C, carbon-14.

Ak-Dzhar37.996268.8451Levallois/Mousterian; Soanski Type  TechnotypologyGlacial/InterglacialPlainsKostenko et al. (1961), Ranov (1965), Ranov & Nesmeyanov (1973)
Aman Kutan39.306766.9621Levallois/Mousterian 50,000  BiostratigraphyInterglacialMountainsLev (1949), Bibikova (1958), Vishnyatsky (1999)
Anghilak Cave*39.285066.6870Typical Mousterian

Stratum III: 27,310 ± 270

Stratum IV: 43,900 ± 2,000

Stratum IV: 38,100 ± 2,100

31,663–31,255

48,933–45,769

44,734–41,162

AMSGlacial/InterglacialFoothillsGlantz et al. (2003, 2008)
Dodekatym-2*41.572370.1636UP assemblage; small blades produced by subprismatic and narrow-face flaking

Layer 4: 23,800 ± 190

23,600 ± 330

21,850 ± 180

28,909–28,295

28,732–27,981

26,604–25,950

AMSGlacialMountainsKolobova et al. (2011)
Dzhar-Kutan39.790368.2971Typical Mousterian; Nanai and Tashkent Complex  Geomorphology, biostratigraphyInterglacialFoothills 
Kairak-Kumy40.098169.4306Acheulian/Mousterian or early Mousterian  TechnotypologyGlacial/InterglacialPlainsRanov & Nesmeyanov (1973)
Kara-Bura37.778568.9502Mousterian; Soanski Type  TechnotypologyGlacial/InterglacialFoothillsRanov (1965), Ranov & Nesmeyanov (1973), Vishnyatsky (1999)
Khodjamazgil39.835366.7732Similar to Upper Palaeolithic of Samarkandskaya  TechnotypologyGlacialPlainsTashkenbaev & Suleĭmanov (1980), Vishnyatsky (1999)
Khudji*38.616768.2167Later Mousterian, Levallois38,900 ± 70043,786–42,64614CGlacial/InterglacialFoothillsTrinkaus et al. (2000)
Kulbulak*41.008670.0061Mousterian blade/Upper Palaeolithic microblade assemblages 50,000  OSLInterglacialFoothillsKasymov & Grechkina (1994), Flas et al. (2010)
Obi-Rakhmat Grotto*41.009470.0022Transitional from MP to UP with Levallois-Mouterian blade based variant

Layer 7: 36,170 ± 810

Layer 14: 48,800 ± 2400

Layers 15–21: 90,000–50,000 (at least 17 cultural horizons)

Obi-Rakhmat-1 hominin: 87,000–77,000

41,976–40,563

52,055–46,490

14C; ESR; U-seriesGlacial/InterglacialMountainsKrivoshapkin et al.(2006, 2010), Glantz et al. (2008), Blackwell et al. (2006), Skinner et al. (2007), Mallol et al. (2009)
Ogzi-Kichik*38.731870.1663Typical Mousterian

15,700 ± 900

30,000

38,000

21,273–17,089

34,766–34,607

42,676–42,350

14CGlacial/InterglacialMountainsRanov (1965), Ranov et al. (1973), Vishnyatsky (1999)
Sel-Ungur*40.721171.0746“Meridonal Acheulian” handaxe/cleaver/chopper 126,000 ± 5000  UTInterglacialMountainsVishnyatsky (1999)
Shugnou38.484171.0971Mousterian, resembles upper layers of Obi-Rakhmat

Layer 2: 20,000–25,000

Layer 3 and 4: 30,000–35,000

Palynological, technotypologyGlacialMountainsRanov & Nesmeyanov (1973), Vishnyatsky (1999), Ranov et al. (2012)
Teshik-Tash38.290067.0500Mousterian 70,000–50,000  BiostratigraphyGlacial/InterglacialMountainsMovius (1953), Vishnyatsky (1999), Dalen et al. (2012)

According to the attributes listed above, each site was designated as either a glacial or interglacial occupation (Table 1). The sites that had chronological associations with archaeological materials from both glacial and interglacial periods, such as Obi-Rakhmat Grotto, or sites associated with a single date with an error range that spans a glacial–interglacial transition, such as Khudji, were used in the spatial analyses for both periods (Trinkaus et al., 2000; Bailey et al., 2008; Glantz et al., 2008). Interglacial sites are those that were occupied during the period spanning the Last Interglaciation (130–75 ka) and the initial stable warm phase of the OIS 3 interstadial (59–44 ka). Sites corresponding to glacial periods were occupied during the FGM (74–60 ka), the transitional and terminal phases of OIS 3 (LGML; 43–28 ka), and the LGMG (27–16 ka) (Meese et al., 1997).

Test for randomness of site distribution

We first tested whether the distribution of Palaeolithic sites is random. A Monte Carlo test based on the Cramér–von Mises statistic (Cressie, 1991) was applied to the xy coordinates of each observation to test the null hypothesis of complete spatial randomness (CSR). These neighbourhood analyses are designed to test hypotheses of randomness by simulating a homogeneous Poisson process and calculating the empirical K-function for each simulation (n = 100) (R.M.R., pers. obs.). If the empirical K-function lies within the bounds of the upper and lower 99% confidence interval from simulation, the sites can be considered spatially independent. A higher empirical value indicates an aggregated distribution, while a lower value suggests regularity (Reich et al., 2000). The null hypothesis of CSR is rejected if the goodness-of-fit P-value is below 0.05. Glacial and interglacial sites were evaluated independently. Although we could not model hominin distribution using only securely dated sites, because of statistical limitations on sampling size and model fitting, we report the results of the total sample as well as the undated and dated subsets when possible.

Late Pleistocene hominin ecological niche model

Environmental variables

We modelled the abiotic requirements representing part of the fundamental ecological niche (Hutchinson, 1957) of hominins in this circumscribed region of Central Asia. The analyses were carried out at two spatial scales, one fine (30 m resolution) and the other coarse (5 km resolution). All of the procedures explained below were carried out using the R statistical package and ArcGIS 10 (ESRI, Redlands, CA, USA).

At the fine scale, we assessed the spatial relationship between site location and: (1) elevation (LP DAAC: http://lpdaac.usgs.gov/get_data); (2) slope; (3) proximity to streams (see Appendix S1 in Supporting Information); and (4) distance to maximum glacial ice extent (Kuhle, 2004). An ANOVA test was used to examine if the means for each variable were statistically different between glacial and interglacial sites.

To explain the spatial variability of hominin locations at a coarse resolution, we used bioclimatic variables derived from monthly temperature and precipitation data (http://www.worldclim.org/; Hijmans et al., 2005) thought to represent biologically meaningful aspects of local climate (Waltari et al., 2007). Ocean–atmosphere simulations for past climate are available through the Paleoclimate Modelling Intercomparison Project (Braconnot et al., 2007). The present study used the Community Climate System Model (CCSM3; Otto-Bliesner et al., 2006a) to represent environmental layers for the LGM, and a global climate simulation for the Last Interglaciation (Otto-Bliesner et al., 2006b). Predictor variables with high correlation coefficients (> 0.8) were excluded when correlated with more than one variable, and, for paired variables, we excluded monthly averages in favour of quarterly averages.

The present study used LGM bioclimatic layers for glacial periods as well as the terminal phases of OIS 3, while sites occupied during the Last Interglaciation and the stable warm phase of OIS 3 were evaluated using the interglacial data set. Although this is a coarse proxy for the broadly fluctuating climatic conditions of the last glacial period (LGP), glacial/interglacial conditions can be universally characterized as the oscillation between cold, dry conditions and warm, wet conditions, regardless of their chronology (Dodonov & Baiguzina, 1995; Swann et al., 2005; Dodonov et al., 2006; Feng et al., 2011; Murakami et al., 2012).

Model development

The ENM presented here is derived from habitat association models that require presence-only data as the response variable (Clark et al., 1993; Knick & Dyer, 1997; Dettmers & Bart, 1999; Reich et al., 2000). Based on minimum threshold theory (Neter et al., 1985), Reich et al. (2000) developed this approach to fit a logistic regression model to the spatial distribution of bird presence as a function of canopy opening. The ENM was carried out using spat99.tar (Reich et al., 2000), a software package created for spatial analysis in the R statistical package, and ArcGIS 10. The following section provides a general description of the statistical theory used to develop and evaluate our model.

The aim of our ENM is to explain the spatial relationship between a suite of bioclimatic variables and the spatial distribution of hominin locations. To develop the model, we assumed that each abiotic variable possessed a minimum threshold, such that an increase beyond a threshold would render that geographical space uninhabitable. A further assumption is that the response is logistic and that hominin presence can be modelled with a logistic function following a curvilinear relationship with asymptotes of 0 and 1, such that 0 indicates areas of unsuitable habitat and 1 indicates suitable habitat (Neter et al., 1985). The logistic response function can then be transformed to allow for parameter estimation using least-squares methods (Neter et al., 1985; Reich et al., 2000). From this assumption, we were able to derive probability density functions (PDFs) that denote the likelihood of observing a hominin location as a function of individual abiotic variables.

The method presented here was carried out in two steps. First, we defined an optimal range of values for predictor variables based on the concept of maximizing the difference between cumulative distribution functions of predictor variables where a species is present and the value of those variables across the landscape (Dettmers & Bart, 1999). We overlaid the PDFs describing the range of variables at hominin locations and a random background sample (= 100,000) to detect potential preferences or selection for specific ranges of variables, and to provide the first step in understanding constraints on hominin distribution. Second, we used ArcGIS to fit probability surfaces for each significant variable. Probability surfaces were created with values ranging from 0 to 1, using equal-interval classification. Joint probability surface models were created by multiplying individual probability surfaces.

Model evaluation

To test the hypothesis that our ENM explains the spatial variability in hominin distribution, we developed a spatial point process simulation for both glacial and interglacial periods (Reich et al., 2000). The null hypothesis is that the distribution of hominin locations follows a non-homogeneous Poisson process, with probabilities proportional to the ENM probability surfaces. The simulations are evaluated using the Cramér–von Mises goodness-of-fit statistic (Cressie, 1991) and the transformed K-function. An empirical K-function that remains within the 99% confidence interval simulation envelope provides support for the null hypothesis, indicating that the chosen variables adequately describe the spatial variability in hominin distribution. Rejection of the null hypothesis suggests that another variable not used in the model is driving the distribution.

Results

Are hominin sites randomly distributed across the landscape?

The distribution of both interglacial and glacial locations are spatially dependent (P < 0.05) and therefore we can falsify our null hypothesis of CSR (Fig. 2). Because the empirical K-function extends above the 99% simulation envelopes, interglacial and glacial locations are considered to be aggregated on the landscape. The Cramér–von Mises test was not possible because of the small sample sizes when examining the distribution of radiometrically dated interglacial and glacial sites. However, in both cases, the empirical K-function extended beyond the 99% simulation envelopes and supports the interpretation of those sites as also being aggregated (Fig. 2). Although this is not direct validation of a ‘correct’ assessment of those sites as glacial versus interglacial, these results underscore the observation that, on the whole, hominin locations in the study region are not spatially random. During the LGP, hominins were either tracking resources within their local environments or their distribution is systematically biased in other meaningful ways (i.e. biases related to site preservation, taphonomy and/or archaeological survey logistics).

Figure 2.

Plot of transformed K-function against distance [L(h)] to test for spatial randomness (CSR) of hominin locations in the study area of Uzbekistan and Tajikistan. The dark black stair-step line represents the empirical K-function calculated from the data and the light, dashed lines represent the upper 99%, average (middle line) and lower 99% simulation envelopes (= 100) for our point process model for the following subsets: (a) all glacial; (b) dated glacial; (c) all interglacial; and (d) dated interglacial.

Explaining the spatial variability in site location using ENM

Fine-scale analysis

ANOVA results indicate that only proximity to water differs significantly between the glacial and interglacial sites when either the entire sample (= 8.09, P = 0.01) or the dated subset is considered (= 13.05, = 0.004) (Table 2). The undated subset, in contrast, shows no significant difference in the mean distance from water between interglacial and glacial locations (P > 0.05). All sites are on average less than 5 km from water. Only three sites modelled in the present study (Ogzi-Kichik, Shugnou and Kara Bura) are further than 5 km from a potential water source. Kara Bura appears to be an outlier in the sample as it is roughly 12 km from water. Kara Bura is treated as both an interglacial and a glacial hominin location because of its chronological range. The other two sites were occupied during glacial periods and are near the glacial terminus.

Table 2. Summary statistics of abiotic variables evaluated in the fine-scale ecological niche model (ENM) analysis of stratified hominin locations in Uzbekistan and Tajikistan. Values are reported [mean (standard deviation) min–max] for the total number of glacial and interglacial sites and also for the dated and undated subsets
SubsetClimate periodDistance to streams (km)Elevation (m)Slope (°)Distance to glacier terminus (km)
AllGlacial4.24 (3.44) 0.30–12.381086 (532) 406–200012.85 (13.31) 1.69–43.5136.58 (36.89) 0–96.32
Interglacial1.27 (1.10) 0.05–3.961099 (474) 406–20007.74 (8.56) 1.69–29.1836.08 (27.48) 0–96.32
DatedGlacial4.20 (2.05) 0.97–6.281324 (400) 796–200018.04 (15.31) 3.44–43.514.51 (20.24) 0–52.19
Interglacial1.06 (0.59) 0.28–1.641248 (404) 796–20009.61 (10.02) 3.04–29.1824.31 (21.43) 0–52.19
UndatedGlacial4.63 (4.48) 0.30–12.381000 (705) 406–200012.78 (16.60) 1.69–43.5152.56 (41.09) 0–96.32
Interglacial1.49 (1.48) 0.05–3.96949 (527) 406–18005.88 (7.24) 1.69–20.2647.85 (29.49) 19.61–96.32

No hominin locations are over 2000 m in elevation, and glacial and interglacial sites are within similar elevation ranges. Although no significant difference in the distance from glacial terminus exists between glacial and interglacial sites, our dated glacial-period locations are closer to glacial termini when compared to the entire sample and the undated locations, with dated glacial sites being on average within 15 km of the terminus. Finally, glacial and interglacial sites are present in the three major biomes sampled in our study area (mountains, foothills and desert/plains), although dated sites are disproportionately located in the foothills and low-elevation mountains (Table 1), perhaps indicating a bias in site preservation and/or sampling.

PDFs indicate no preference or selection of specific abiotic ranges when compared to a background sample. Hominin sites are distributed in areas that are roughly proportional to the overall availability of those variables on the landscape: most sites are within 5 km of water, as are most points on the landscape. These results indicate that model-building at a fine scale using distance to water, elevation, slope and distance to LGM glacial termini, would be no more robust than one generated randomly. Although the ANOVA test indicates a significant difference between distance to water when glacial and interglacial sites are compared, this difference was not captured as significant in the PDF test.

Coarse-scale analysis

Spatial variability in hominin site locations is not strongly predicted by any variable examined in our fine-scale analysis. However, a few robust results were produced when examining the relationship between hominin location and coarse-scale (c. 5 km) bioclimatic indicators. We compared the PDFs of random background points to those of the glacial sites, and identified two variables that affect the distribution of hominin sites: temperature seasonality and the mean temperature of the coldest quarter (Fig. 3a,b). To evaluate the degree to which the variables adequately describe the spatial variability in site location, we tested the null hypothesis that hominin sites follow a non-homogeneous Poisson process with probabilities proportional to the probability surface. The Cramér–von Mises goodness-of-fit statistic indicates that the null hypothesis cannot be rejected for temperature seasonality (= 0.14) or temperature of the coldest quarter (= 0.09); both variables therefore adequately explain the spatial distribution of sites. The empirical K-function for temperature seasonality is fully contained within the simulation envelope, while the K-function for temperature of the coldest quarter falls slightly outside the envelope, which suggests that temperature seasonality may represent a stronger predictor of hominin distribution (Fig. 3c,d).

Figure 3.

(a,b) Overlaid probability density function (PDF) plots for variables used in our coarse-scale ecological niche model (ENM) of stratified hominin locations during glacial periods in the study area of Uzbekistan and Tajikistan. Dotted lines represent the range of (a) temperature seasonality and (b) temperature of the coldest quarter observed at each hominin location; the solid black line represents a random sample of background points (= 100,000). (c,d) Plot of the transformed K-function against distance [L(h)] testing the degree to which (c) temperature seasonality and (d) temperature of the coldest quarter adequately describe the spatial variability in site location. The dark black stair-step line represents the empirical K-function calculated from the data and the light, dashed lines represent the upper 99%, average and lower 99% simulation envelopes (= 100) of the ENM.

The combination of these two variables (joint probability surface) provides a better fit when describing hominin distribution than the individual surfaces alone (= 0.34), a result that is supported by the empirical K-function (Fig. 4a). Those areas within the study region that have the least seasonal temperature variation are also the warmest during the coldest periods of time; hominin locations are disproportionately aggregated in these areas during glacial periods (Fig. 4b). Moreover, it appears that cold temperatures, rather than variation in precipitation, affect the distribution of hominin locations during glacial periods (contra Jennings et al., 2011). Table 3 provides the ranges for bioclimatic variables at each hominin location, thus providing the framework to define the abiotic niche requirements of late Pleistocene Central Asian hominin groups.

Table 3. Summary statistics for 19 bioclimatic variables (BioClim) evaluated in the coarse-scale ecological niche model (ENM) analysis of stratified hominin locations in the study area of Uzbekistan and Tajikistan. Values are reported for glacial and interglacial periods [mean (standard deviation) min/max]. Predictor variables used in the ENM are bold type. Temperature is in °C × 10 and precipitation in mm
BioClim variableGlacial hominin locationsInterglacial hominin locations
Annual mean temperature 60.09 (47.71) −33/123 98.75 (30.91) 35/147
Mean diurnal range 193.27 (15.65) 156/214136.92 (6.63) 128/150
Isothermality42.45 (3.24) 35/4625.17 (1.27) 23/27
Temperature seasonality 8840.73 (460.98) 8221/9482 13947.42 (953.21) 12697/15335
Maximum temperature of the warmest month 298.64 (52.04) 188/359384.42 (32.44) 323/433
Minimum temperature of the coldest month−151.45 (51.55) −81/−249−148.75 (46.92) −228/−80
Temperature annual range 450.09 (14.12) 435/472533.17 (34.52) 493/581
Mean temperature of the wettest quarter 33.18 (37.77) −39/9783.17 (57.32) −53/146
Mean temperature of the driest quarter162 (47.37) 76/232230.67 (28.99) 177/274
Mean temperature of the warmest quarter172.36 (46.47) 81−232276.58 (30.86) 216/320
Mean temperature of the coldest quarter −56.73 (51.17) −157/9 −75.25 (34.28) −143/−26
Annual precipitation 562.55 (232.83) 355/982 476.92 (166.13) 253/908
Precipitation of the wettest month 100.55 (33.85) 59/17083.42 (29.79) 37/57
Precipitation of the driest month 4.09 (2.77) 1/101.67 (2.06) 0/5
Precipitation seasonality 69.64 (8.48) 55/7869.75 (9.30) 57/83
Precipitation of the wettest quarter 263.36 (103.02) 156/477222.25 (84.20) 99/443
Precipitation of the driest quarter 20 (13.86) 9/4710.92 (9.11) 1/25
Precipitation of the warmest quarter 30 (24.14) 11/6718.08 (15.79) 3/49
Precipitation of the coldest quarter 187.36 (63.74) 122/310162.75 (48.79) 86/283
Figure 4.

(a) Transformed K-function against distance [L(h)], testing the degree to which our joint surface ecological niche model (ENM) (temperature seasonality × temperature of the coldest quarter) adequately explains the spatial distribution of stratified hominin locations during glacial periods in the study area of Uzbekistan and Tajikistan. The dark black stair-step line represents the empirical K-function calculated from the data and the light, dashed lines represent the upper 99%, average and lower 99% simulation envelopes (= 100) of our joint ENM. (b) The joint probability surface depicted here is the product of the individual surfaces, temperature seasonality and temperature of the coldest quarter. The probability surface values range from 0 to 1, and is classified into 10 equal intervals. Sites are indicated by black triangles. Areas in white represent the highest probability of observing a hominin location, and darker regions represent lower probability.

In a comparison of the PDF of random background points and those of the interglacial hominin locations, annual precipitation and annual temperatures appear to most strongly influence hominin distribution (Fig 5a,b). However, these variables alone do not adequately describe the spatial variability in hominin locations (Fig. 5c,d; < 0.05). Moreover, the Cramér–von Mises goodness-of-fit test on the joint probability surface (= 0) supports a rejection of the null hypothesis. Although hominin distribution tends towards regions that have the highest annual temperatures and highest annual precipitation, these variables do not adequately capture the spatial variability in distribution. Some other variable or combination of variables not analysed in the present study, or confounding biases must drive the distribution instead. These results suggest that the range of abiotic variables representing suitable areas for hominins were present in our study area throughout the glacial–interglacial transitions of the late Pleistocene (Table 3).

Figure 5.

(a,b) Overlaid probability density function (PDF) plots for the variables used in our coarse-scale ecological niche model (ENM) of stratified hominin locations during interglacial periods in the study area of Uzbekistan and Tajikistan. Dotted lines represent the range of (a) annual temperature and (b) annual precipitation observed at each hominin location; the solid black line represents a random sample of background points (= 100,000). (c,d) Plot of the transformed K-function against distance [L(h)], testing the degree to which (c) annual temperature and (d) annual precipitation adequately describe the spatial variability in site location. The dark black stair-step line represents the empirical K-function calculated from the data and the light, dashed lines represent the upper 99%, average and lower 99% simulation envelopes (= 100) of the ENM.

Discussion

The present study represents an initial step towards characterizing the abiotic requirements of the hominin fundamental niche in Central Asia during periods of climatic instability. Three significant aspects of hominin distribution in the study area can now be addressed. The first point requires significant elaboration because of its ability to speak directly to previously existing biogeographical models of the region.

First, new radiometric dates associated with anthropogenic material from three Palaeolithic sites in Uzbekistan contradict the suggestion that glacial climates in Central Asia, specifically those of OIS 2, were too harsh to support suitable niches for hominins (Ranov & Davis, 1979; Davis & Ranov, 1999). Dates from Obi-Rakhmat Grotto, Anghilak Cave and the open-air Upper Palaeolithic site of Dodekatym-2 imply that hominins found suitable niches not only during the LGMG, but also in OIS 4, as well as the coldest phases of OIS 3 (LGML), respectively (Table 1).

The radiocarbon dates from Dodekatym-2 and Anghilak Cave are from anthropogenic materials, i.e. charcoal from hearths and faunal elements that were clearly processed by local hominins. The chronological framework from Obi-Rakhmat Grotto relies on multiple dating techniques, such as optically stimulated luminescence (OSL), electron spin resonance (ESR, linear uptake model), uranium-series dating and traditional radiocarbon methods. ESR dates derived from ungulate tooth enamel create upper and lower temporal boundaries for the hominin remains (OR-1) from stratum 16. Strata 12–14 are dated to 55–77 ka, while the basal aspect of stratum 21.2 is dated to 87 ka. Consequently, the hominin remains can be placed within an age range of 77–87 ka, suggesting that the population from which the OR-1 fossil is derived was present at the site during the OIS 5 interglacial (Mallol et al., 2009). However, Obi-Rakhmat Grotto consists of roughly 10 m of continuous archaeological deposits; these dated anthropogenic assemblages are synchronous with the coldest phases of OIS 4, the initial stable warm phase of OIS 3 and the LGML.

Although hominins were present in the study area during the coldest periods, existing biogeographical models hypothesized that it was the combined effects of the cold and increasingly severe aridity that prompted hominin dispersals out of Central Asia (Davis & Ranov, 1999). Our results suggest that the extent to which glacial periods were also arid (or more arid than during interglacial periods) is open for some debate. Recently dated lacustrine sediments in the Tarim Basin, an area within Taklamakan Desert, indicate that some localities were well watered during the LGMG (Zhao & Xing, 1984; Olsen, 1992; Keates, 2004) and that, on the whole, Central Asia was continuously wetter than East Asia during the LGMG (Feng et al., 2011).

The notion that hominins persisted in this region throughout the glacial–interglacial transitions from OIS 5 to OIS 2 suggests that the study area may represent a glacial refugium, and possibly served as a zone of hybridization (see Hewitt, 1996; Arnold, 1997; Stewart & Stringer, 2012). Concepts of refugia and their relationship to subspecies variability and the potential for subsequent divergence are multifaceted (see Bennett & Provan, 2008; Stewart et al., 2010 for review). It is generally agreed that refugia exhibit high levels of endemism and intra-specific variability. Although the interpretation of our study area as a glacial refugium requires further testing, both the high degree of variability in the hominin fossil and genetic record of the region (Krause et al., 2007; Glantz et al., 2009; Reich et al., 2010), coupled with findings from the non-human mammalian literature (Kahlke, 1999; Galbreath & Cook, 2003; Kahlke & Lacombat, 2008; Vorobieva et al., 2011), offer preliminary support for this working hypothesis.

In fact, the location of our study area – 2000 km to the south and west of the Siberian Altai – fits a southern dispersal model for hominins during climate downturns, and is in agreement with Stewart & Stringer's (2012) suggestion that the Siberian Altai represents an unlikely late Pleistocene refugium. This view might also support the idea that our study area represents a hybridization zone, although this conclusion is confounded by the fact that it is only in the Altai, and specifically at the sites of Okladnikov and Denisova Caves that hominin genetic diversity (i.e. multiple hominin metapopulations) is observed. It is important, in this regard, to investigate how the Altai ecosystems compare to those of our study area, and if hominin locations there persist across glacial–interglacial transitions.

Second, the results presented here indicate the distribution of hominin locations during glacial and interglacial periods is non-random and that, third, a small subset of the abiotic variables analysed strongly predict the pattern of that aggregation. While the empirical K-function distribution and the goodness-of-fit statistic demonstrate that locations are aggregated on the landscape (Fig. 2), most of the tested variables do not have an impact on hominin distribution. Generally, the landscape contains plenty of water resources. For example, glacial sites located more than 5 km from water are also positioned closest to the glacial terminus (within 3 km). In addition, hominin locations are present in each of the major biomes that characterize the study area, regardless of climatic period (Table 1). The lack of contraction of hominin locations within a particular biome may be related to the ubiquity of water on the landscape and underscores the notion that the study area was ecologically productive throughout the LGP.

In the coarse-scale analysis, our ENM indicates that the distribution of hominins was constrained during glacial periods by temperature, while no observable abiotic or bioclimatic variable influenced hominin distribution during interglacial periods. Although interglacial sites are spatially aggregated, none of the analysed variables robustly predicted hominin distribution during milder climatic phases (Fig. 5). It is likely that other variables not modelled here – such as the distribution of specific flora and fauna and/or culturally significant factors, such as distance from raw material resources – have an impact on hominin locations during interglacial periods. These factors may also affect the distribution of glacial sites, as flora and fauna may track those locations that are warmer during the coldest phases of the LGP.

The major uncertainty involved in our analyses, however, is the influence of other forms of bias on how hominin niches are modelled in the context of analyses that cannot take into account archaeological and taphonomic biases in a systematic manner. For example, archaeologists have long nursed the perspective that high-elevation ecosystems were not commonly exploited by prehistoric humans (Steward, 1938; Bettinger, 1991; Aldenderfer, 2006). The fact that the hominin locations in our study are all below 2000 m a.s.l. and average at least 15 km from glacier termini may be the consequence of an archaeological bias against surveying for hominin localities at higher elevations. It is presently impossible to discriminate between an archaeological bias that influences decisions concerning field protocol and survey goals from geological phenomena that may also affect the ability to identify hominin presence on the landscape; continuous processes that have shaped topography over the LGP, such as orogeny, glacial dynamics and erosion, may have destroyed evidence of hominin presence in high-elevation areas. In addition, radiometrically dated hominin locations also are disproportionately located in the foothills and low-elevation mountain belts, regardless of climate phase. This may be the result of site preservation bias, in that only sites located in this terrain possess the conditions necessary for the derivation of radiometric dates (i.e. long occupational histories, highly stratified, well-preserved archaeological material, etc.). However, in the archaeological literature, the propensity of sites to aggregate in the foothills is interpreted as a preference of hominins to inhabit the piedmont zones along mid-elevation foothill/mountainous regions, with easy access to the plains and resources therein. While there is little doubt that this zone represented an important part of the hominin niche, we should be cautious of interpreting these data without considering sources of bias and how they confound our efforts to describe the hominin fundamental niche or characterize the limits of the hominin Palaeolithic adaptation.

Conclusions

In comparison to the richness and refinement of the data sets used to model hominin niche characteristics and dispersals within Europe during the late Pleistocene (e.g. van Andel & Davies, 2003), the sources of evidence used to define the fundamental hominin niche in Central Asia are rather limited. This is due to the patchy distribution of late Pleistocene hominin locations, highly localized climatic regimes, and a relatively poor chronological framework for both phenomena. However, improved geospatial methodologies allow for an examination of existing data sets from Central Asia in new and fruitful ways.

The present study represents an initial step in the examination of competing hypotheses concerning hominin distribution in Central Asia during periods of climate instability. In particular, our results provide limited support for the contraction of the hominin niche to specific locations on the landscape where temperatures were less extreme during glacial periods. On the whole, the abiotic and bioclimatic variables analysed do not seem to have limited hominin distribution during interglacial periods. It appears that this region of Central Asia provided suitable habitats for hominins throughout the LGP. To what extent the study area represents a refugium during glacial periods is currently unknown. Although this analysis offers weak support for the foothills and lower-elevation mountain ranges in southern Central Asia as defining a refugium, more dated sites and a larger region need to be analysed before we can confidently draw that conclusion. Contrary to popular opinion, late Pleistocene Central Asia – even during the coldest phases of the glacial periods – may have been better watered and more suitable for hominins than previously thought.

The identification and subsequent characterization of refugia within Central Asia is important to our interpretation of the demise of the Neanderthals. From a broader geographical perspective, Central Asia represents another area where Neanderthal populations may have flourished during times when they were unusually stressed in north-western Europe, and it seems that Neanderthals were not the only hominin populations in Central Asia, thereby setting the geographical stage, as it were, for the intermixing of hominin groups during the late Pleistocene that genetic research has recently revealed (Reich et al., 2010).

Our model offers a means to directly compare the pattern of Central Asian hominin distribution to other, similar Old World data sets. GIS and allied spatial statistical tools allow researchers to identify potential weaknesses in data sets and foster the development of hypotheses that can be addressed directly in the field. The validity of the geospatial model presented here ultimately relies on the robustness of the spatial models used and the temporal associations of the hominin locations examined. Our results, therefore, should be considered preliminary and contingent upon validation in the field.

Acknowledgements

We would like to thank Lucy Burris and the Geospatial Laboratory at Colorado State University for technical help throughout the model-building process. The ASTER GDEM data was provided by the NASA Land Processes Distributed Active Archive Centre (LP DAAC), USGS/Earth Resources Observation and Science (EROS) Centre, Sioux Falls, South Dakota (https://lpdaac.usgs.gov/get_data).

Biosketch

The research presented here was supported by the Human Origins Laboratory (HOL), Colorado State University. Ongoing projects are positioned at the interface between palaeoanthropology and Quaternary sciences. HOL is ultimately interested in understanding the effects of global environmental change on Neanderthal and modern human evolution.

Author contributions: T.A.B. and A.K.T. devised the study; T.A.B., A.K.T., S.S.T. and M.M.G. collected the data; T.A.B. and R.M.R. analysed the data; and T.A.B. and M.M.G. led the writing.

Ancillary