When Survey Goes East: Field Survey Methodologies and Analytical Frameworks in a Central Asian Context
- First Online:
- Cite this article as:
- Markofsky, S. J Archaeol Method Theory (2014) 21: 697. doi:10.1007/s10816-013-9172-9
- 410 Downloads
This paper investigates the applicability and transferability of conventional frameworks of archaeological survey in the context of marginal alluvial environments, particularly the unique inland deltas of Central Asia. These dynamic and visually obstructed landscapes pose unique challenges not only to survey methodologies but also to theory and interpretation. Here, an exploratory approach to data analysis is used that applies three distinct yet integrated methodologies: visibility analysis, multi-scalar spatial analysis and directional (anisotropic) statistics. This approach thereby moves beyond many of the existing conceptual constraints about how we understand surface distributions in arid alluvial landscapes and ultimately identifies both transferable analytical methods and new fieldwork agendas that are relevant to a wide range of survey projects.
KeywordsSurvey Central Asia Spatial analysis Murghab
One particularly challenging domain for the field survey of archaeological landscapes is the point at which theory, method and interpretation meet. Because archaeologists seek to infer broad-scale human processes from highly vestigial material, the ways in which research questions are framed and investigations conducted can have as profound an effect on our understanding of archaeological phenomena as does the evidence itself. The degree to which researchers, faced with palpably new challenges, can fall back to the familiarity of pre-defined research frameworks, accepted methodologies and well-trodden theoretical discourses is particularly evident in archaeological survey, not least because surface data are by nature nebulous and there is often limited recourse to core, invasive datasets traditionally associated with archaeology (i.e. excavated materials).
Because surveys deliberately try to cover a larger spatial area than excavations, and only a fraction of subsurface data is usually available, research can sometimes be constrained by familiar and established dichotomies such as onsite versus offsite or extensive versus intensive methods. To be sure, researchers are well aware of the irregular and dynamic processes that influence surface distributions, and survey approaches increasingly seek to understand artefact variability (Bevan and Conolly 2004; Bintliff et al. 2007), the behaviour of pottery within dynamic or obstructed landscapes (Ammerman 1995; Tartaron et al. 2006), or the role of integrated analytical scales in the interpretation of survey data (Bevan and Conolly 2006; Burger and Todd 2006). However, despite these welcome developments, certain unhelpful conceptual boundaries remain.
One type of survey environment in which such issues are particularly salient is in marginal alluvial deltas; those transitional regions that straddle the line between habitable, sustainable regions and adverse, often infertile zones (Barker and Gilbertson 2000; Wilkinson et al. 2004; Cremaschi and Zerboni 2010). Such regions were often characterized in antiquity by geomorphologically nebulous boundaries, complex hydrologies and social marginality, and these features persist in many cases into the present. In particular, the considerable impact of these complex and dynamic environments on the present-day distribution of relict material culture on the landscape surface can hardly be overstated. While no archaeological environment is static, the constant flux of the alluvial margin necessitates approaches to survey that can not only infer occupational activities and post-depositional processes from surface distributions but also deconstruct interpretative biases that may vary from one marginal environment to the next.
This article explores the applicability and transferability of marginal-zone survey in the context of the Murghab delta, an inland alluvial fan in southeastern Turkmenistan. While this region bears similarities to other arid-zone alluvial fans in terms of its complex hydrology, dune cover and transitional features, this ‘inland oasis’ also presents some unique considerations for archaeological survey (Salvatori 1998; Cattani and Salvatori 2008; Markofsky and Bevan 2012). Winds and infrequent but occasionally heavy rains can contribute to the development of landforms in which erosional and deflationary processes constantly redistribute surface material (see discussion of ‘badlands’ in Imumorin and Azam 2011). In such regions, the fragile dynamism of the delta–desert boundary perhaps makes it even more important to develop a sophisticated understanding of local surface distributions.
This data presented below is drawn from the Northern Murghab Delta Survey (NMDS), an intensive survey of part of the northernmost fringe of this alluvial margin, conducted from 2007 to 2009. The research framework described in this article may be considered exploratory as it seeks to discern patterns in a discrete set of survey data while eschewing prior assumptions of how sites and distributions ought to be manifest (as opposed to strict hypothesis testing—see Tukey 1977). Three aspects are here considered: surface visibility, multi-scalar spatial clustering of artefacts and directional (or anisotropic) patterns. Chronological data, while central to any archaeological survey and carefully investigated during the course of research, is not considered a primary focus of this paper, which addresses general distributional characteristics in the context of survey method and practice. My motivation here is twofold: first, to gain a clearer sense of the characteristics of surface distributions in a particular marginal alluvial environment that can set some reasonable expectations for other surveys facing similar challenges, and second, to provide a broader assessment of the transferability of theory and method in archaeological survey.
Binary conceptual frameworks can be traced throughout the history of survey and bear significantly on the way in which research has been conducted. One well-known example is the ongoing debate between ‘site-based’ archaeological surveys, often extensive or regional in scope, and ‘off-site’ or ‘siteless’ approaches (Ebert 1992; Wilkinson 2003b; Caraher et al. 2006). Sites, of course, are no longer viewed as the monolithic entities that once defined archaeological survey, and decades of research have brought to light the extreme complexity inherent in the site concept. Substantial research has sought to quantify, for example, statistical deviations of occupational areas from background scatter (Bintliff et al. 2007), or the variability of activity areas within occupational loci (Laurenza et al. 2005). Even the term ‘site’ is a fluid concept—as increased concentrations of surface material may represent domestic areas, outdoor activity areas, regions of intensive agricultural activity or manuring (Wilkinson 1982), secondary depositional areas (Allison 1999), farmsteads or other manifestations of human activity over a concentrated area (Bintliff et al. 2007). Complementing—and sometimes opposing—such investigations is a growing body of what may be thought of as ‘offsite’ research. These latter projects comprise a broad spectrum of work and may examine factors as diverse as agricultural practices, post-depositional processes (Millett et al. 2000; Taylor et al. 2000) or recovery bias (Van Leusen 2002). Even in offsite studies, the concept of ‘site’ is often retained as a familiar reference point against which these studies can be understood. While useful both conceptually and administratively (Dunnell and Dancey 1983; Bevan and Conolly 2004), the sometimes opposing concepts of offsite vs. onsite can influence archaeological interpretation (Sullivan et al. 2007).
To a degree, modern research agendas have attempted to bridge such conceptual dichotomies in hopes of representing archaeological landscapes more equitably. For instance, the large-site dominance that characterized many early surveys (Wilkinson 2003a) has been substantially tempered by the recognition that occupational evidence on any scale can contribute greatly to archaeological understanding. Small sites or scatters, variously described in terms such as seasonal, rural or temporary (see Foxhall 2000), have increasingly become topics of study—whether to evaluate the effectiveness of previous surveys or to investigate activity that may not be prominent in the surface record (e.g. sedentary–nomadic relationships or rural–urban interactions). Other investigations have focused on the fuzziness of the site concept, recognising that neither settlements nor their present-day manifestation in the surface record need conform to defined boundaries, but that they instead reflect complex patterns of occupation, settlement-derived activities or post-depositional processes (Gallant 1986; Banning 2002).
Both the identification and interpretation of such patterns, however, relies on a keen understanding of how, where and when perceptual biases may come into play. Central to this issue is archaeological visibility, not only in terms of what can be recovered but also how we recover the surface data. Archaeological evidence may be masked by agricultural or urban development (Wilkinson and Tucker 1995), geomorphological or hydrological processes (Brown 1997: 42), even fragmentation of artefacts themselves, resulting in a greatly transformed and often largely invisible record (Taylor et al. 2000). Although visibility bias is a well-recognized problem (Van Leusen 2002), formal analysis and standardization can be problematic. Some progress has been made in this area, however. Quantitative approaches such as regression analysis (Shennan 1985) have been used to identify specific biases that affect recovery. More recently, the impact of variables such as artefact obtrusiveness, clustering and fragmentation have been tested against known or seeded artefact distributions, providing an effective control against which such variation can be measured (Wandsnider and Camilli 1992; Schon 2002). Even survey methods themselves have come under recent scrutiny in the context of visibility, leading to the development of quantitative functions to measure detectability in various survey landscapes (Banning et al. 2006, 2011). Despite these investigations however, correction and standardization of visibility bias have been less successful. In many surveys, an estimated visibility level is assigned to individual survey units (Bintliff and Snodgrass 1988; Bevan and Conolly 2004; Stark and Garraty 2008), and attempts have been made to re-weight artefact totals based on this estimated visibility (Given and Knapp 2003: 54; Bintliff et al. 2007: 21; Bevan and Conolly 2009). Such corrective models, however, are subject to inherent landscape complexities and varying recovery methodologies (Mattingly et al. 2000; Bevan and Conolly 2004). While progress has been made in the development of quantified models to correct for visibility (e.g. Schon 2002; Banning et al. 2011; also see approaches in Bintliff et al. 2007), more work is required to assess the degree to which such models can effectively standardize our assessment of differential visibility.
While perhaps apparent to archaeologists working today, the recognition of the complexity of surface distributions represents a major theoretical advance over earlier, largely site-based interpretations of archaeological landscapes, and attempts to deconstruct this complexity continue to drive research methodologies. Nevertheless, integrated studies of settlement landscapes remain problematic. Surveys that can merge equitably both site and non-site data remain more theoretical than practical, and while many researchers recognise the need for such approaches, these are not always achievable. Archaeological landscapes represent continuous rather than discrete datasets (Lloyd and Atkinson 2004), the result of myriad depositional and post-depositional processes that are further subject to intrinsic biases in the landscape. Methodological or analytical delineations, therefore, are largely arbitrary relative to the unknown and nebulous realities of the archaeological landscape under study. Fortunately, research designs that address this problem more formally have increased in recent years. On the methodological side, distributional or siteless approaches to survey, pioneered by researchers such as Dunnell and Dancey (1983) and Ebert (1992), have spurred new ways of considering interpretative lenses. Additionally, multi-scalar statistical approaches (Premo 2004; Bevan and Conolly 2006; Burger and Todd 2006), geostatistical and directional (anisotropic) analyses (Bevan and Conolly 2009; Markofsky and Bevan 2012) have helped researchers re-think pre-existing structural confines. The consideration of such spatially varied processes, operating over multiple analytical scales, offers the potential for more robust and integrated approaches to survey methodology.
These methodological developments come amidst changing theoretical approaches that have helped reshape the perception of space and region (e.g. Kantner 2008; Ryzewski 2012), promising new frameworks that may further our understanding of integrated entities and spaces. Recent work in northern Mongolia, for example, has incorporated nested survey methodologies to investigate different levels and scale of influence within nomadic confederacies (Honeychurch et al. 2007). Scale integration has also been a pre-dominant factor in re-evaluating survey data from northern Mesopotamian sites under the auspices of the Fragile Crescent Project (Galiatsatos et al. 2009). However, such frameworks are still developing, and more work needs to be done to fully integrate these approaches into the conventional discourse on survey.
The emphasis on regional aspects of delta occupation and large sites has to some extent hindered a clearer understanding of local and multi-scalar aspects of the surface distribution. Such investigations, however, offer the potential to elucidate fine-scale variability in surface distributions across marginal alluvial environments more generally. While micro-dynamics of survey data have been the focus of several recent investigations (Cleuziou 1998; Cattani and Salvatori 2008; Cattani et al. 2008; Cerasetti et al. (2013) Walking in the Murghab Alluvial Fan (Southern Turkmenistan): an Integrated Approach between Old and New Provides New interpretations about the Interaction between Settled and Nomadic People, unpublished), these represent punctuated areas of more intensive analysis, resulting in a kind of interpolated view of settlement dynamics in which limited areas of intensive study have been draped against a complex, largely obscured and still poorly understood regional backdrop.
Although the implications of this recent work have been far-reaching, research agendas have relied in part on a set of conventional assumptions that may not be fully applicable to the landscape. Factors such as deflation or dune cover can strongly affect estimations of site dimensions, and even large sites are often low or obscure. As a result, survey methodologies that may be effective elsewhere, such as in the tell-dotted landscapes of the Near East, may be less effective. Additionally, intensive methodologies that have relied primarily on sherd counts or densities have limited utility in the Murghab due to complex post-depositional processes and visual obstruction in the region, although work continues in this area (B. Cerasetti and L. Rouse, personal communication). Given these inherent difficulties, new methodological and analytical lenses may help to address non-uniformity and variability more specifically, and some prospects for such lenses are introduced below.
The methodology for the NMDS survey was designed to address the pressing need for a better understanding of sub-regional and multi-scalar characteristics of settlement patterns, particularly in marginal alluvial regions such as the Murghab, where visual obstruction and spatial heterogeneity of surface distributions can confound site identification and complicate interpretation. In such regions, in which both landscape and surface distribution are distinctly non-uniform, even rigorous sampling strategies may be heavily skewed. A methodology is here developed that seeks to quantify some of these complicating factors within the archaeological landscape, and to develop transferable methods for effective interpretation that can be applied in other marginal regions as well.
Analytical units in the NMDS survey area
Analytical unit No.
Characteristics of land cover and material distribution
Primary area of occupation—comprises extent of high-density surface scatters. Landscape is marked by fluvial signatures and takyrs
Area to the east of the high density scatters in Area 1. Apparently intensive palaeohydrographic activity but very sparse surface distributions. Some dune cover, but generally exposed landscape
Intermediate (and largest) region consisting largely of variable sand dunes, over-sanded/eroded takyrs. Represents the region between the primary surface scatters in Area 1 and the secondary scatters in Area 4
Small area of heavy dune cover and little exposed surface. Low sherd counts
Large takyr surface (0.7 km2)
Region with variable dune cover and moderate palaeohydrographic activity. Moderate scatters. Dune cover is clearly distinguishable from Area 6 based on multi-spectral imagery
Region with stable dune cover but moderately dense surface material. Clearly distinguishable from the variable dunes in Area 5 based on multi-spectral ASTER imagery
Small area with apparently intensive palaeohydrographic activity and substantial surface material
Large region with heavy sand cover but extremely sparse surface material
The first aspect of the exploratory approach laid out in this paper examines visibility—the degree to which we can detect archaeological evidence by eye, and the extent to which visual obstruction can be treated as uniform or variable across archaeological landscapes. In the Murghab itself, the visibility problem has not been formally quantified and is often treated as binary: sand cover is (sensibly) assumed to obscure material that would otherwise be visible on exposed surfaces.
The methodology employed here uses supervised classification algorithms to develop an interpretable visibility model for the delta landscape. Two sets of remote sensing data were used for the analysis: multi-spectral ASTER imagery and high-resolution Quickbird imagery. These datasets were spatially integrated via an image-stacking technique in which each set of imagery was re-sampled to a common resolution; this allowed each pixel in the high-resolution imagery to be associated with more robust multi-spectral data that could then be assigned a ‘land type’ according to a classification algorithm based on training sites chosen during the course of in-field investigation. This derived visibility model could then be evaluated in conjunction with sherd count data from the survey.
Sherd distributions, by virtue of the processes that influence their ultimate deposition, are spatially heterogeneous and non-random (Bevan and Conolly 2009; also see discussions on assuming spatial randomness in Banning 2002: 51 and Nance and Ball 1986: 40), so any methodology that investigates visibility must seek a way to deconstruct visibility bias from actual archaeological processes. To this end, the visibility analysis is presented as a foundation for the following two statistical methods that are orientated towards the evaluation of such distributions. Furthermore, several lines of test pits were conducted during the course of the survey in order to investigate the subsurface character across varying topography and land cover types. Although these results are beyond the scope of this paper, they contribute to the qualitative aspect of landscape visibility discussed below.
The second analysis is designed to investigate sherd clusters at multiple analytical scales. The tendency for surface material to aggregate is pivotal to many surveys not only in terms of site definition but also to assessing the likelihood that material may actually be recovered. Furthermore, such investigations can offer clues to geomorphological, agricultural and other post-depositional processes that may influence the surface distribution. However, most such approaches only consider one or very few analytical scales and therefore miss an opportunity to consider integrated processes that may be manifested at varying spatial levels.
The approach considered here employs K functions (Ripley 1977), statistical methods that evaluate clustering or dispersal over a range of distances. The K function is more robust than traditional methods of cluster analysis in that it investigates multiple analytical scales simultaneously, and while not often used in archaeology (see Orton 2005; Bevan and Conolly 2006; Sayer and Wienhold 2012), it can help detect patterns that may not be visually apparent.
To conduct the analysis, the survey data (sherd counts) were converted to a point pattern, in which each sherd is represented as an individual point. Within each 20 × 20 m collection unit, the location of these sherds was randomized, so that for any unit, the maximum distance between the actual location of the sherd (not recorded) and the randomized location could not exceed 28 m (the hypotenuse of each unit). The distances under consideration ranged from 50 m (exceeding the randomization error) to 500 m in increments of 50 m. The analysis used Monte Carlo simulations (n = 99, p < 0.01) to establish confidence intervals. Clustering is indicated when the graph of the K function exceeds the high confidence interval; conversely, dispersal is indicated when the graph is below the low confidence interval.
Because of the non-stationary nature of the surface distribution, the analysis was conducted both globally and for each of the nine interpretative units defined above. While these units are not statistically derived (and distributional processes and land cover may vary to some extent within these units), they provide a means to assess general multi-scalar variability in sherd clustering that can then be further explored via local variants of the K function as well as other geostatistical methods.
The third method focuses on directional or anisotropic characteristics of the surface distribution, and employs an aspect of geostatistics called variography to measure whether sherd counts exhibit continuity in certain directions more than others (Bevan and Conolly 2009). Variography provides a measure of spatial dependence, the similarity between two distinct values as a function of distance (Fortin and Dale 2005), and is an effective way to assess underlying continuity from sparsely sampled and non-uniform data such as sherd counts. This is effectively the first step in the method known as kriging which is often used for predictive modelling (e.g. Journel 1974); however, this article focuses on the directional aspect of this continuity. By way of illustration, consider the landscape of dune ridges already described in the study area. Measured sherd counts may be expected to exhibit continuity along the course of the north–south oriented ridges: if one were to travel north along a ridge, sherd counts would likely be consistently low. In the inter-dune valleys, however, sherd counts would—if bias were the only factor—be consistently higher. By choosing the direction of investigation, such continuity can be measured.
For the purposes of this article, only a brief discussion of the methodology is necessary (for a full discussion of geostatistical methods in the NMDS survey, see Markofsky and Bevan 2012). To conduct the analysis, the central point (centroid) of each of the collection units was assigned the value of the sherd total within that unit. This provided a valued grid of sherd counts. While actual surface distributions may be expected to be organic and certainly not cardinally oriented, both the exactness and the high spatial resolution of the point pattern (27,500 valued points) facilitate the identification of directional trends that may run counter to the primary orientation of the dataset.
For this article, directional influences are presented using a fairly simple yet robust geostatistical plot called a variogram map that plots spatial dependence as a function of distance. The primary direction of continuity is indicated by the major axis of the observed ellipse in the variogram maps.
The first objective of the analysis was to investigate, describe and, if possible, quantify potential visibility biases inherent in the landscape that could potentially affect the recovery of surface material. The analysis employed a supervised classification algorithm to identify five land-types ranging from presumably low-visibility stable sand dunes to high-visibility takyr surfaces. Land cover types were determined by first selecting training areas then applying a maximum likelihood classification to assign each pixel to its closest (statistically speaking) land type. The categories labelled ‘Variable Dune’, ‘Moist Takyr’ and ‘Sanded Takyr’ are in-field descriptions of the training sites used in the classifications. Since these are statistically determined based on a set number of training sites, these categories represent a thematic model of land cover, which is in actuality far more complex. However, these classifications are very useful to broadly characterize the land cover found in the NMDS survey region in the context of general recovery potential.
The following results employ a chi-squared statistical test which evaluates the observed surface sherds against the number that would be expected to occur if proportionally distributed according to land-cover percentage. However, any such analysis must consider the potential for circular reasoning: absence of artefacts in a certain region does not necessarily mean they are simply obstructed; actual depositional processes may account for the lack of material in one region as opposed to another. While certain methods such as artefact seeding offer the potential for more direct control to measure deviations in artefact counts (Ebert 1992; Wandsnider and Camilli 1992; Schon 2002; Banning et al. 2006, 2011), such an approach to evaluate variability over an entire archaeological landscape can be resource intensive, particularly in regions that exhibit a high degree of local heterogeneity. The following analysis, therefore, may be seen as a general approach to visual non-uniformity that is further evaluated in the context of field observation and the parallel spatial and statistical analysis in the following sections.
Effects of land cover on sherd counts (analytical unit 7 not included because of its extremely small area)
Land Cvr, %
Land Cvr, %
Land Cvr, %
Land Cvr, %
Full NMDS area (sherd totals)
NMDS (units with over 1 sherd)
This global result, however, assumes rather simplistically that visibility bias is a uniform phenomenon. In reality, complex taphonomic processes and variable land cover suggest that visibility itself is dynamic, and that such biases may not be manifest evenly across a survey landscape (Ammerman 1995). To investigate the local variability of visibility bias, the nine analytical units described above were considered individually. Although these units are not fully homogeneous in terms of land cover or the character of the surface distribution, each unit is qualitatively distinctive and thus provides a means to compare and contrast the variable nature of visibility bias across a full survey landscape. The results of this analysis revealed a much more complex situation. In Area 1, an area of widely scattered material and variable dune cover, sherd totals in sandy areas actually exceeded expected values. This seemingly paradoxical result is likely explained by the fact that surface material here was too dense and too widespread to be significantly restricted by land types, and field observations confirmed significant spillover into potentially more visually restrictive areas. Erosion and deflation, as well as shifts in active dunes that comprised a portion of the landscape, may further complicate the trend and lead to increased exposure of material even in areas classified as dune covered.
Inter-dune areas, primarily takyrs in the Murghab context but with geomorphological parallels in other arid alluvial environments, yielded some unexpected results. Open takyr surfaces were often largely devoid of material, with the exception of certain high-density regions in Areas 1 and 4. Because test pits conducted across takyrs in each of these regions revealed only natural stratigraphy, much of this material was likely spillover from adjacent regions of varied land cover. This illustrates an important disconnect between visibility as it pertains to archaeological material and archaeological process: since these regions represent wide-open, unvegetated spaces in which artefacts would likely to be detected if present, the lower-than-expected counts in such areas attest to underlying archaeological processes rather than bias. One striking example may be seen in a comparison between Areas 1E and 8; in each case, sherd counts were almost equally low; however, in Area 8, the landscape was largely covered by stable dunes, as opposed to the largely open fluvial landscape in Area 1E. The presence of substantial surface material to the west, in Area 1, but not in 1E in which the landscape was markedly similar, may suggest preferential occupational choices, although such a possibility is speculative and a full discussion is beyond the scope of this paper. In Area 8, however, the presence of at least one moderately dense sherd cluster in an isolated exposed location suggests that visual obstruction has indeed hindered recovery.
Relationship between sand cover index (SCI) and sherd totals (by number of sherds per collection unit)
More than 20 sherds
Chi-sq (df = 4, p < 0.0001)
These results show that, although surface distributions and land cover are spatially heterogeneous, remote sensing data can be used effectively alongside qualitative information to assess variable visibility. The high resolution offered by the integration of multi-spectral and high-resolution imagery allows non-uniform obstruction to be assessed and measured, and thus offers excellent potential for not only arid alluvial environments but also other regions in which multiple geomorphological and post-depositional processes are in play. The effectiveness of the straightforward classification used here bodes well for future investigation in other environments, both in terms of differential visibility assessment as well as in conjunction with systematic recovery-oriented investigations such as artefact seeding.
Clustering and Spatial Aggregation
The formalisation of potential visibility bias provides a clearer picture of local variability in the context of recovery potential. At the same time, it raises an important question: if bias is itself spatially variable even at small scales, how reliable are traditional parameters of archaeological survey if the interpretative frameworks are so dynamic? In the Murghab, these issues can be seen in large discrepancies in site size estimates, which in some cases can exceed dozens of hectares (Kohl 1984: 143), but these questions may be applicable in any situation in which erosion, deflation or other processes confound conventionally accepted site metrics. This is particularly problematic for survey-based interpretation: if basic parameters vary widely, how can researchers define data points suitable for analysis and interpretation? The following analysis therefore eschews familiar metrics often associated with the discrete site (e.g. size and extent) in favour of statistical descriptions that seek to describe the clustering behaviour of the surface material.
The statistical method described in the methodology section, the K function, offers a way to investigate patterns of sherd clustering at widening spatial scales and can therefore help to identify patterns in the surface pottery that transcend fixed analytical boundaries and may not be detectable by eye. The K function was first applied to the full survey area to gain a general sense of global sherd clustering and shows significant clustering at all scales. This is a general indicator of spatial heterogeneity and often indicates multiple processes at different spatial scales, signalling the need for more targeted investigations (Orton 2005).
A local variant of the K function can be used to ‘drill down’ to examine even more localized behaviours (Getis 1984). This local K function measures clustering within a certain radius of a given point against what would be expected in a random distribution of points (sherds) (Walker et al. 2007). This localized variant can be used in conjunction with the global K function to assess specific clustered and/or dispersed areas that may be contributing to the overall pottery distribution. Furthermore, it offers a systematic way to suggest meaningful sub-regional units of analysis, ultimately offering a more statistically sound means to divide up the continuous survey data given the fact that the distribution is clearly neither random nor produced by a single process (cf. Bintliff et al. 2007; Bevan and Conolly 2009).
The K function thus highlights some key patterns in the survey data. Firstly, spatial heterogeneity was present even within seemingly discrete distributions, a finding that offers strong statistical support for complex local processes and post-depositional activities, and in some cases, may indicate multiple activity loci. Secondly, off-site scatters are themselves highly variable, and likely reflect various post-depositional processes acting over several spatial scales, although the presence of scattered diagnostics may attest to small-scale occupational activity as well. This non-uniformity of off-site material provides statistical evidence that varied processes may be expected to affect surface distributions in different ways (see also Alcock et al. 1994; Taylor et al. 2000). Thirdly, statistical interpretations can offer a way of mitigating interpretative problems caused by visibility issues (i.e. settlement areas that may be directly obstructed may still leave signatures in the observable background scatter that may be more apt to be detected statistically).
So far, two frameworks have been put forth through to which interpret surface aggregation within obstructed landscapes: visibility and clustering. To further characterize the dynamic behaviour of pottery in a landscape, a third can be applied: directionality. In the alluvial fan of the Murghab, two prevailing directional patterns may be expected to influence both the perceived and actual distributions of material: (a) the predominant north–south topography of sand ridges, and (b) the delta geomorphology, in which both the underlying geology and some relict watercourses trend towards the northwest (Marcolongo and Mozzi 1998). While directional influences may be observed qualitatively in many alluvial systems, as with quasi-linear distributions of settlements along channels, influence on surface distributions has not been sufficiently investigated.
To gain a clearer sense of behaviour in ‘off-site’ regions, the areas of highest density (Areas 1 and 4) were eliminated from consideration, revealing subtle trends that became increasingly clear (Fig. 8c). The variogram map suggests a slight offset in the NNW–SSE direction at medium separation distances of 200–500 m (also detected in a directional variogram, a geostatistical plot not shown here). This tendency, most evident in the western portion of the survey area, suggests that fluvial geomorphology contributes to the directional patterning, distinct from the north–south anisotropy of the dune ridges. This directional variation may suggest underlying distributional processes attributable to fluvial erosion or occupation along watercourses, and some support for this possibility may be seen in the parallel distribution of diagnostic pottery in these regions. However, dune formation may in some cases be influenced by underlying fluvial morphology, so it is important to resist interpretations that are too directionally deterministic and to consider the anisotropic data in context.
To this end, the directional evidence may be evaluated in tandem with the statistical evidence discussed above. Both directional and aggregational influence are most prominent in the 300–400-m range; at shorter distances, the resolution becomes too low to penetrate the noise caused by patchy and variable sherd distributions. Beyond 500 m, the prevailing topography of the current landscape exerts undue influence and is more indicative of visibility bias than of actual distributional processes (e.g. the north–south banding mentioned above). One possibility is that this spatial restriction, acting over distances of only a few hundred metres and with particular directional orientation, reflects the clustering behaviour of settlement-derived processes along watercourses. This pattern recurs throughout the area and may reflect distributional behaviours representing quasi-linear settlement chains. However, the ascription of current distributional patterns to actual settlement processes is tenuous, and it is probable that post-depositional behaviours, particularly along palaeochannels, strengthen the directional signatures in the direction of these watercourses.
The distinct yet integrated frameworks of visibility, clustering and directionality reveal spatially varied distributional patterns that may be difficult to encapsulate using conventional approaches to archaeological survey. The organic nature of the surface distribution, as well as the fact that much of the archaeological landscape is deflated or obstructed, suggests that conventional off-site or on-site frameworks may not adequately describe surface patterning. In the specific context of the Murghab delta, spatially heterogeneous sherd scatters distributed across varied local topography may reflect uneven and shifting occupational patterns, influenced in part by a dynamic and perhaps unstable fluvial system at the fragile desert/delta frontier. Stream channels on the alluvial margin may be expected to be smaller and shallower than trunk channels (Marinangeli et al. 2004), and in some ways more sensitive to processes of desertification, aridisation and water loss (see also Cremaschi and Zerboni 2010). It is likely that social structures and adaptation within such an environment (e.g. settlement foundations, agricultural activities such as irrigation or manuring (Wilkinson 1982; Bintliff et al. 2007), pastoral activities, etc.) contributed to complex settlement-derived depositional patterns, which were subsequently modified by geomorphological and hydrological processes. Furthermore, the continuity of material in certain directions as opposed to others may attest to the diachronic role of the fluvial system in shaping distributional processes: first offering fertile and accessible quasi-linear routes along which settlements initially were founded (cf. Adams 1981: 21), and subsequently providing a transport mechanism for redeposition of surface material. Even along the relict channels that now characterize the palaeodelta, such processes may continue, aided by heavy precipitation, and such a possibility is worth considering in the context of other arid alluvial environments.
The complexity of the surface distribution presented here therefore suggests a conflation of settlement-derived and post-depositional processes whose manifestations vary with changing analytical frameworks, and operate over various scales of analysis. Such heterogeneity highlights the inherent difficulties in theory, method and analysis that may arise when survey-based approaches are tested in new and unfamiliar environments (Glatz 2012), particularly in cases in which the depositional character is obscured or significantly transformed, or where sites are not prominent, topographically or otherwise. The problem of extracting reliable data from obstructed landscapes is well known, and new adaptive strategies continue to inform recent projects (e.g. Düring and Glatz 2010; Kaptijn 2009; Van Leusen 2002). Such strategies are of particular importance in landscapes such as those of the type described in this paper and other marginal or otherwise transitional regions, in which the highly dynamic interplay of archaeological, taphonomic and other factors further confound metrics such as site dimension or size. Non-site or artefact-based methods that rely on sherd counts or densities may be similarly compromised in distinctly non-uniform archaeological landscapes.
In cases such as these, survey implementation and interpretation can benefit from frameworks that are better suited to disentangling overlapping archaeological processes, post-depositional processes and inherent biases. Such issues continue to be explored in the context of ‘hidden landscapes’, for example, in which small and often inconspicuous prehistoric assemblages have been shown to represent dynamic but often undetected archaeological landscapes (Bintliff and Howard 1999; Bintliff 2011). The effective interpretation of such landscapes necessitates a deep understanding of local taphonomic, geomorphological and archaeological processes. As a result, more reliable corrective factors continue to be developed that can provide better models for understanding sparse and heterogeneous surface scatters.
In light of such developments, the NMDS results presented in this paper offer some complementary approaches to the consideration of non-uniformity and local heterogeneity. The multi-scalar investigation of anisotropy, for example, offers a plausible way to dissociate spatially restricted patterns from inherent directional biases. Although the discrepancy was subtle in this particular survey area due to the slight deviation between underlying geology and later dune morphology, this methodology may be further applicable in other dune environments, ridged landscapes or landscapes in which modern activity (e.g. roads and pathways) may contribute to directional bias. From the perspective of survey methodology, the recognition of anisotropy bears on both surface and subsurface sampling methodology as well: recovery along directional lines may yield different results than those that run counter to them. Sampling methodologies, then, should take into account directional influences (e.g. topography) that may influence perceptions of archaeological remains (also see discussions in Nance 1979: 44; Banning 2002: 101). Further, what we think of as sherd ‘clusters’ are clearly subject to perception: particularly in visually obstructed environments where continuity may not actually be observable in the field. Statistical evaluation of multiple analytical windows offers a way of transcending these visible obstructions. In the NMDS survey, for example, many cases exist in which apparently isolated sherd clusters could later be associated more definitively with other processes, not observable in the field due to visual obstruction, and similar barriers to identifying the full extent of sherd aggregations exist in other environments (e.g. Ur et al. 2011: 7). By employing multi-scalar statistical approaches, broader patterns could not only be identified but also some measure of their influence in the surrounding landscape could be determined. For example, if a statistically significant sherd aggregation is much larger than an observable concentration of diagnostic material, this discrepancy may be partially attributable to post-depositional processes that can themselves be measured. Effective assessment of such patterning, however, relies on integrated knowledge of landscape, bias and distribution, all of which can benefit from the spatial-continuum approach described in this article.
More broadly, these analyses further underscore the importance of context-sensitivity in survey. While aspects of the NMDS surface distribution and the delta landscape have parallels in other alluvial environments, each archaeological landscape is unique, and methods of standardisation or re-scaling across different survey projects, or even within a single project, can be problematic. In this sense, statistical approaches that include local, directional and multi-scalar analysis can offer a greater degree of transferability than do metrics that rely on absolute or relative values (such as sherd counts, densities or site sizes). Characteristics that may at first seem to be abstract such as statistical significance or directional character can offer a great deal of information that can be of great comparative value in conjunction with other more qualitative factors. Visibility analysis may similarly benefit from statistical and algorithmic approaches, and such trajectories may be even more promising in light of novel, integrated approaches to survey (e.g. Campana and Francovitch 2007; Campana 2011) that incorporate remote sensing, geoarchaeology and other techniques that can shed light on artefact behaviour and recovery potential in complex landscapes.
There is one further issue to consider, the implications of such investigations on archaeological discourse itself. Because conceptual frameworks can be pervasive, different sets of assumptions can result in interpretative disconnects among researchers. Research projects are all-too-often criticized on the grounds that they failed to address questions that they may not have even been designed to answer. Such tension has surfaced in criticism of intensive surveys on the grounds that they are too narrowly focused, too resource intensive or inadequate to address broader-scale political or socio-economic questions (Fentress et al. 2000; Blanton 2001; Kowalewski 2008, contra Caraher et al. 2006). Such criticism, while sometimes valid, can overlook the potential contributions of these projects to the integrated study of archaeological landscapes. Indeed, if the goal is simply to locate sites, or to offer a regional assessment of settlement character, then such scepticism may be warranted; even in the NMDS study area, cursory coverage was often sufficient to locate the largest scatters and thus gain a general idea of settlement character at broader analytical scales. What is becoming increasingly evident, however, is that observable archaeological phenomena on the surface—sites, scatters, even offsite material—present only part of the overall picture. Indeed, that ether in which such phenomena occur—whether we speak of background scatter, local morphology, surface obstructions or other factors—is just as relevant and may provide answers to questions previously unconsidered.
When surveys are conducted in regions so heavily defined by unknowns, trusted frameworks can be constraining. The exploratory approaches discussed here provide new ways of describing and characterising data that can be repeated and applied both in other alluvial environments presenting similar challenges, and indeed in a wider array of survey contexts. The close study of patterning both in landscape and surface material in this particular inland alluvial delta therefore provides a baseline for understanding how surface distributions may be manifested in dynamic, arid environments. At the same time, this research has used multi-scalar and directional analysis to offer some points of departure from some of the more familiar concepts in archaeological survey, as concepts of site, background scatter and even visibility often resist discrete characterisation and may vary locally. As the exploratory approaches presented here have shown, survey data act very differently from one case to another, and require methods and conceptual frameworks that respect this contextual variation.
I wish to express my sincere gratitude to Tim Williams and Andrew Bevan for their encouragement and support of this research. Thanks also go to Gaigysyz Joraev and to Maurizio Tosi, Barbara Cerasetti and the University of Bologna researchers associated with the Archaeological Map of the Murghab Delta, without whose collaboration, this work would not have been possible. Further thanks to the anonymous reviewers for their helpful suggestions.