Surface Survey in Wadi Ziqlab, 1981

Who studious in our art
Shall count a little labour unrepaid?
I have shed sweat enough, left flesh and bone
On many a flinty furlong of this land.
- Robert Browning, An Epistle.

The Recovery of Evidence: Survey in Wadi Ziqlab
E. B. Banning

The Wadi Ziqlab Project began in 1981 with a survey conducted from 21 November to 19 December of that year (Banning and Fawcett 1983). The goal of that survey was to associate site locations with environmental variables in such a way as to reveal changes in pastoral and agricultural land use (Banning 1982; 1985). The ability of survey data, alone, to provide this type of information is limited by several problems.

One of the inevitable problems if the survey is not accompanied by test excavations is that it is frequently difficult or impossible to determine the functions of sites which the survey discovers (Banning 1985: 75-95). In order to meet the goals of the Wadi Ziqlab Project it is crucial to be able to distinguish pastoral camps from various types of agricultural and special-purpose sites, and these distinctions are difficult at best when the only evidence comes from material lying on the modern surface.

If subsurface data can be made available, by selected test excavations, which strengthen or replace the assumptions of surface survey, then regional survey has the potential to tell us much more about economic development in ancient Jordan than could any number of excavations on large tells.

Identifying the course of agricultural and pastoral development in antiquity is a problem which is inherently rural. Excavated samples from large towns, like most of the tell sites in the southern Levant, may not be reliable indicators of the structure of subsistence systems. This is because central places may draw on an agricultural and pastoral hinterland over 700 square kilometres in area, and import trade goods over large distances.
Only regional data from groups of small sites which, like campsites and farm outbuildings, reflect the spatial organization of rural activities, will inform us in detail about the economic basis of ancient societies. A combination of regional survey and small-scale excavations is one way to obtain such data.

Another potential problem associated with using the data of a regional survey as evidence for changes in land use is that we cannot be certain that the physical environments of sites have not changed since antiquity. Provisional reconstruction of palaeoenvironments may be fraught with difficulties when it is based only on regional climatic and vegetation trends, on modern physiographic, topographic and edaphic features, on imperfect palynological evidence, and on infrequent references in historical literature. Combining archaeological research with a programme of palaeoethnobotanical study and botanical survey would provide more complete data on which to base a more reliable picture of humans' exploitation of the environment in antiquity, and this is what we attempted in Wadi Ziqlab.

Methods of the 1981 Survey

In retrospect it is easy to identify ways in which the procedures of the 1981 survey might better have served the purposes of the research there. Time constraints, a somewhat hopeless optimism, and a failure to appreciate, until too late, some of the major obstacles to archaeological survey in this rugged, forested and well peopled valley conspired to prevent us from collecting the representative sample that was our goal. Substantial mantles of colluvium, vegetation or leaf litter, in particularly uncooperative sample areas, also made it next to impossible to identify the remains of small rural settlements and camps that were a major focus of the research. If nothing else, the problems that arose out of the 1981 survey, and some of the solutions that arose in subsequent field work, have provided raw material for some hard thinking that resulted in suggestions for others facing the prospect of tackling a regional archaeological project in Levantine highland areas (Banning 1996).

The following pages document in detail the methods of the survey, with all its pitfalls, on which much of the interpretation of Wadi Ziqlab's cultural history is based. This will permit other researchers to evaluate the survey results realistically and make it easier to recognize any biases we have not already identified. Reference to Appendices A and B will provide the most specific details on survey transects, the circumstances of sites and samples.
The description of survey procedures in the following sections uses terminology that is now fairly widespread in archaeological publications (e.g., Mueller 1975; Schiffer and Gumerman 1977; Plog, Plog and Wait 1978; Schiffer, Sullivan and Klinger 1978; Banning 1986; 1988; 1996).

The Survey Universe and Sampling Frame

The drainage basin of Wadi Ziqlab was the area of interest for this survey but, to provide a straightforward sampling frame for this area, we adjusted its borders to fit the 1-km intervals of the Universal Transverse Mercator Grid Zone 36 (Series K737 maps at 1:50,000). The resulting universe (fig. 1) contained 115 squares of one square kilometre each, and this grid provided the basic sampling frame. Sites outside this grid received only purposive visits in rare instances (e.g., Site WZ 113) in which large or obtrusive sites were known to occur very close to the universe's borders.

From this sampling frame of 115 quadrats a 20% spatially random selection with replacement targeted 20 quadrats for sub-sampling (three squares were selected twice). Each of the target quadrats, highlighted in fig. 2, was the subject of inspection by pedestrian transects. The transect records and overlays for aerial photographs in Appendix B show the routes of our transects in detail; where possible the crew of three walked four parallel transects across each target quadrat. In exceptionally difficult circumstances we walked only three transects across a quadrat, and in some cases cliffs or military restrictions prevented us from completing portions of transects. The resulting distribution of inspected landscape lacks consistency, but is well documented and easy to relate to the natural features of interest.

In addition a purposive survey of some sites known to exist in or adjacent to the wadi catchment and of unsampled areas which we suspected might harbour important information supplemented the transect sub-sample. The sites from purposive survey, of course, are unsuitable for inclusion in statistical analyses of site distributions in Wadi Ziqlab.

The sites from the transects constitute a cluster sample and their statistics suffer from spatial autocorrelation. Treating the quadrats or transects as units of analysis, and the number of sites or site areas within these units as a statistic of the unit, sidesteps most of the problems that would result from treating the sites themselves as units of analysis (cluster sampling), but the interpretation of site distributions in any case is not simple.

Intensity and Coverage

While the placement of transects in the sub-sample is not as regular as might have been desirable, the distribution of transects none the less permits us to make quite precise measurements of survey intensity in given quadrats, or to estimate coverage for sites of given size or obtrusiveness in the universe as a whole (Banning 1985: 64-66; 1988).

Coverage, as the term is used here, is the proportion of the survey universe in which we assume that the survey would locate or record every site, of a given definition, if visibility were perfect. It serves, in essence, as a crude estimate of the probability of site discovery.
We estimate coverage for the survey by reference to individual sites' "discovery radius" (Banning 1985: 66; 1988: xx; n.d.1). For a site of relatively low obtrusiveness, such as a sherd scatter, but with good visibility the discovery radius is simply one-half the minimum width of the site. This is more conservative - more pessimistic if you like - than basing the measure on the average width of the site. Basically the assumption is that as long as a crew member walks over a portion of the site, he or she will recognize it, but is much less likely to recognize the site from a distance. For more obtrusive sites, such as buildings with standing walls, the discovery radius will be equal to the distance from the centre of the site to the point where the site first becomes visible to a crew member on foot. In practice it is much less obvious where this point should be than in the example of a sherd scatter, but we can make reasonable, conservative estimates for sites such as mills, large khurub, dolmens, or large stone cairns on hilltops.

As the graph in fig. 3 indicates, the estimated coverage of the 1981 Wadi Ziqlab Survey for sites 250 m or more in width is at least 0.24, while for sites 100 m in width minimum coverage is 0.11 and for sites only 25 m in diameter it is 0.047. As such, and particularly since one of the goals of the survey was to investigate small rural sites, the coverage of our small survey project is not very impressive. On the positive side, the implication is that many more sites await discovery in the valley.

Visibility and Obtrusiveness

Survey of the region during November and December provided optimum site visibility in that most agricultural fields and orchards had been plowed only a short time earlier. As a result, sherds and lithics that plowing had turned up from just below the surface were highly visible, especially after the light rains we received early in December, while there was no obscuring vegetation in the plowed fields.

Not all areas subject to survey, however, were agricultural, and variation in visibility was substantial. In some portions of the survey universe visibility is poor throughout the year. The oak forests, in particular, were a great obstacle to the visibility of ancient cultural materials (cf. Alexander 1983); apart from the trees themselves, which reduce horizontal visibility to as little as three metres, the dead leaves which have accumulated beneath the trees have covered all but the most obtrusive sites and have contributed to soil development over them. As a result we would expect the probability of finding scatters of sherds or lithics in forested portions of the transects to be near zero when visual inspection is the only detection technique, as was the case in 1981. In valley bottoms, similarly, we would expect large numbers of sites to be hidden by an overlying mantle of colluvium redeposited from nearby hillsides. Again, the probability of finding most kinds of archaeological sites in this context by visual inspection alone is near zero, although checking the eroded sections which streams have cut through wadi terraces can reveal some archaeological materials.
Obtrusiveness of sites during the 1981 survey varied from a very low discovery probability (low-density scatters from possible camps) through medium (high-density scatters, cisterns, wine presses) to very high (khirab or "ruins" with visible traces of architecture on the surface), often occupying mounds within valleys where they could be seen without difficulty at distances of 2 km.
The Site Inventory in Appendix A records the vegetation cover, site height, estimated probabilities of erosion and colluviation and sampling intensity for each site. These factors are summarized by an estimate, ranging between 0 and 1, of the discovery probability for any given site. Because of the magnitude of accumulated errors in this figure, it should only be used as a rough guide. Furthermore, in comparing the discovery probabilities of different classes of sites, I would recommend converting the data to an ordinal scale.

Data Collection at Sites

Once a crew member identified a probable site, he or she usually called adjacent crew members to see if they were also noticing features or increases in artifact density and, if the locality met the criteria for a site, recorded descriptions of the site on a form and, where possible, took a sample of artifacts.

Site description forms recorded each site's location on maps (aerial photo references were added later, as the photos were not yet available during the survey), the date of visit and initials of discoverer, and whether the site was discovered on a transect or by purposive visit. Descriptions of the site included a general site type (e.g., cistern or sherd scatter), the site's dimensions and probable depth of deposit, an estimate of artifact density on the surface, and a description of that surface, including vegetation or erosion on it. The recorder also noted any visible features, such as architecture, providing a sketch if appropriate, assessed the likelihood of preserved stratification on the site, and listed the bag numbers and types of samples. Most of the samples were purposive, "grab bag" ones that simply included as many diagnostic artifacts, or often simply as many artifacts, as we could find. Only in the rare cases where artifact density was fairly high did we attempt "random" samples. These were not strictly speaking random, but consisted of all of the artifacts we could collect on the surface of a circle the centre of which we determined arbitrarily by tossing a rock onto the site. In two cases we delineated several such circles. For each sample bag, a form recorded the type and location of the sample and would later be used to record the characteristics and probable dates of the artifacts it contained. In an extremely large number of cases, however, sites with architecture or other features produced no samples of artifacts at all in spite of close inspection of the surface.

Site Definition and Chronology

A number of alternative criteria served for the definition of sites during the survey. The most common criterion was a marked increase in the density of artifacts encountered as we walked survey transects. This contrast between the density of on-site sherds and lithics, and the extremely low density of the "background" scatter encountered almost throughout the survey area (xxx) we evaluated subjectively, but our decision to delineate site borders is analagous to removing the tails from a normal distribution (fig. xx). This was more satisfactory than an arbitrary threshold of artifact density, especially since some fairly obvious loci of former human activity had little material evidence on the surface. In other cases, we defined sites by the presence of architectural features such as wall lines or artificial rock cuttings. During the field work crew members made no attempt to discriminate between primary and secondary deposits, but did record any information which they thought might be useful for making this distinction at a later date, such as evidence for erosion or indications that recent farmers had fertilized fields with sabakh (soil dug from archaeological sites).
Crew members attempted to record all sites they encountered which dated from Lower Palaeolithic to Ottoman times. In practice, however, it was often very difficult to distinguish ancient from modern remains, while some of the sites which are probably ancient lack evidence for more precise dates.

For most of the sites the author has inferred dates from the presence of distinctive ceramic or lithic types. Given the relative paucity of good "diagnostic" sherds from many of the sites, the longevity of most lithic and some pottery types, and the lack of representative assemblages at most sites, particularly from low-density scatters, fairly broad periods are the only chronological units we can use with any confidence, and further aggregation of periods is frequently necessary to provide sample sizes large enough for statistical analysis. Unfortunately this crude chronology makes it more difficult to assess the validity of apparent "gaps" in settlement history.

As noted above, many sites provided few or no diagnostic sherds or other artifacts. Stone fences (cf. Evenari et al 1958: 235-36), wine presses (cf. Hirschfeld 1983), cisterns and tombs were rarely associated with surface pottery. In some of these cases we are simply unable to infer dates of construction or use, and we must exclude them from analysis. In others where morphology or presumed function may be distinctive we may suggest probable dates. These dates must in most cases involve even broader periods than those for pottery, and we must in every case treat them with considerable caution. The design of some structures, such as mills, may have changed very little for centuries after their first appearance in the survey area. One of the goals of 1986 and 1987 excavations in Wadi Ziqlab was to test some of these distinctive structures (a circular tower and a mill) to find datable artifacts in use and disposal contexts.

Survey Data as Evidence for Agricultural Economy and Settlement Systems

Many skeptics among Near Eastern archaeologists continue to doubt the value of the data from regional surveys, and especially from ones that employ sampling theory, in addressing general archaeological problems. G. E. Wright (1970) claimed that excavations at tells are the legitimate concern of Near Eastern archaeologists; certainly they have been the focus of most of them. Kay Prag (1984: 66) suggests that only botanical remains from excavated sites, principally tells, are of any use in documenting agricultural developments. Survey data, Prag claims, are of doubtful validity in such applications. Furthermore, Bar-Yosef and Goren (1980: 15), addressing a different set of problems associated with regional surveys, "warn against the use of sampling within a given region."

While many of the skeptics' concerns about existing regional data bases are legitimate, to respond with a complete dismissal of data from regional surveys, as Prag seems to do, or to eschew sampling, as do Bar-Yosef and Goren, betrays a misconception of the goals of sampling and a failure to recognize the unique potential of archaeology's spatial dimensions.

Existing survey data do suffer from biases and unevenness in survey coverage, from the attrition of sites through erosion, construction, burial and other processes, and from lack of adequate publication. Ideally survey publication should be sufficiently thorough to inform us to the degree to which survey coverage and site attrition affect the site inventory. As Bar-Yosef and Goren (1980) point out, geomorphological processes are important contributors to the shape of archaeological distributions and we must assess how they have shaped those distributions if we are to interpret correctly how sites are related to their environments and each other. But these processes are just as damaging to so-called "total" surveys as to spatial samples and just as subject to assessment no matter what type of survey we carry out. As I point out elsewhere (Banning n.d. 1), the real limitation of most spatial samples is that they are not appropriate for studying spatial structure - Christalleran settlement lattices or road systems, for example - but only for estimating population parameters. Most of the research of the Wadi Ziqlab Project, as it happens, focusses on such parameters as settlement size, settlement frequency, and the proportion of settlement associated with particular soil types, making representative samples appropriate.

Furthermore, regional survey data are not merely "poor cousins" of data from excavations. Regio nal survey is uniquely capable of providing some categories of data which quite routine archaeological research often requires. One simple illustration is the case of the archaeologist who concludes that the construction of fortified villages in a region was a response to military threat.
No number of visits to fortified tells will help us test this hypothesis. The only way to refute such a hypothesis is to survey the countryside between the tells. If the survey reveals large numbers of small, unfortified habitation sites between the tells, we would have to abandon or modify the original hypothesis. Our assumption that it was unsafe for populations to scatter over the countryside because of military unrest would have been proven wrong (Banning 1982: 4).

Similarly, problems concerning agriculture and pastoralism in antiquity are inherently rural. It is doubtful that plant remains or animal bones from large towns, like most of the tell sites in the southern Levant which have been excavated to date, will be a reliable indicator of the structure of subsistence systems. Pre-industrial central places could have drawn on an agricultural and pastoral hinterland more than 700 km in area,x potentially populated by small sites that are not typically the target of excavation.

Only regional data from groups of small sites which, like campsites and farm outbuildings, reflect the spatial organization of rural activities will inform us in detail about the economic basis of ancient societies. The spatial distributions of rural sites reflect cumulative decisions about agricultural and pastoral production, storage, and transport. The distributions of carbonized seeds from excavations in tell sites, meanwhile, reflect a complex combination of decisions about food selection, processing, storage, use, discard and post-depositional disturbance; they are just as subject, if not more so, to problems of attrition and bias, although recent researchers attempt to assess and deal with these problems. It is unwarranted to assume that isolated botanical remains from tell excavations, even from primary storage contexts, are automatically superior sources of evidence for subsistence systems than are patterns in the spatial distributions of agricultural, pastoral and special-purpose sites. Of course excavation data from as many sites in a region as possible are welcome contributors of information on exactly what plants and animals may have been exploited at different places and times, but financial and other limitations usually preclude excavated samples from very many sites in a region. Consequently survey data must usually provide the bulk of our evidence, supplemented by the relatively meagre record of published plant and animal remains from excavations.

The Rationale for Sampling

To point out that drawing conclusions from the entire population of archaeological sites in a region is better than generalizing from only 10 or 20 per cent of them is to belabour the obvious. As it happens, such complete data bases are usually quite impractical, and often impossible, to obtain without reducing the scale of survey to the point that it misses spatial variation. Except in cases where investigating some spatial structure, such as canal or road systems, is the goal, controlled sampling is the obvious alternative. In non-archaeological fields, such as political polling, probability sampling has proven itself reasonably accurate, precise and cost-effective in estimating population parameters. Stratified random samples of only 1000 voters typically provide accurate predictions of election outcomes with a standard error of only 5% (Canadian Institute of Public Opinion, p.c.). While natural attrition of sites is often a serious archaeological problem (Bar-Yosef and Goren 1980: 7) from which political pollsters are immune, a good sampling strategy would simply account for environmental factors which are associated with probability of site attrition (Banning 1985; 1986). Like archaeologists, political pollsters also have problems of inaccessibility: some people will refuse to be interviewed. Controlled sampling permits us to make corrections for such problems; haphazard or uncontrolled sampling, misconstrued as "non-sampling," provides data with uncontrollable biases.