Remote Sensing and Archaeological Prospection in Apulia, Italy (original) (raw)

Maney Publishing

Trustees of Boston University

Remote Sensing and Archaeological Prospection in Apulia, Italy
Author(s): Shawn A. Ross, Adela Sobotkova and Gert-Jan Burgers
Source: Journal of Field Archaeology, Vol. 34, No. 4 (Winter, 2009), pp. 423-437
Published by: Maney Publishing
Stable URL: http://www.jstor.org/stable/25608604
Accessed: 20-11-2015 01:10 UTC

REFERENCES

Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/25608604?seq=1&cid=pdf-reference#references\_tab\_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Remote Sensing and Archaeological Prospection in Apulia, Italy

Shawn A. Ross
The University of New South Wales
Sydney, New South Wales, Australia
Adela Sobotkova
The University of Michigan
Ann Arbor, Michigan

Gert-Jan Burgers
The Royal Netherlands Institute in Rome
Rome, Italy

When deployed in combination with ground control, archaeological surface survey, and environmental research, remote sensing based upon high-resolution multispectral satellite imagery allows large areas to be evaluated efficiently by a small team of researchers and contributes to a better understanding of an archaeological landscape. In 2007 and 2008, we analyzed ca. 100 sq km of imagery centered on L’Amastuola, Italy. Combining the evaluation of high-resolution multispectral imagery with concurrent ground control led to the discovery of 29 sites and significant off-site scatters during about four weeks of fieldwork. Our analysis indicates that most of the detected features reflect geological conditions amenable to past buman habitation rather than subsurface archaeological remains. Earlier fieldwork by the Murge Tableland Survey (MTS) provided independent definitions for various types of sites and a large sample of sites and off-site scatters in the study area. Comparison of our remote-sensing guided efforts with the results of that survey suggests that our success rate is too high to be explained by random association and also illuminates the strengths and weaknesses of the respective methods, underscoring the need to integrate satellite image analysis with ground control and surface survey.

Introduction

Until recently, high-resolution multispectral imagery, such as QuickBird, has been used to analyze only relatively small areas (Lasaponara and Masini 2007; Masini and Lasaponara 2006: 536-537). Larger areas have been investigated using multispectral but low-resolution imagery such as Landsat or medium- to high-resolution but panchromatic imagery like CORONA (Fowler 1996; Harrower, McCorriston, and Oches 2002; Philip et al. 2002; Wilkinson, Ur, and Casana 2004; Casana and Cothren 2008). Lower-resolution imagery is useful for producing base maps and studying large-scale environmental and geological phenomena, while higher-resolution panchromatic imagery can detect prominent archaeological sites such as tells. Only high-resolution multispectral imagery reveals the relatively small soil marks, crop marks, and shadow
marks often associated with subsurface archaeological remains. Use of high-resolution multispectral imagery allows for the detection of smaller sites, and for the efficient investigation and management of large, archaeologically rich landscapes (Madry 2007).

The use of high-resolution multispectral imagery as a primary means of archaeological prospection is methodologically underdeveloped, and an assessment of the utility of all types of imagery remains a pressing need (Kantner 2008). Few projects have combined satellite image analysis with field survey to evaluate the numbers of sites discovered with each method, a task necessary in order to determine the utility of satellite imagery for landscape archaeology. Madry’s paper (2007) on the use of QuickBird imagery exemplifies this trend, where identified sites are never confirmed through ground control or compared with the results of field survey. Rates of recovery also need

to be related to sensor type and resolution, accounting for different environments and various types of archaeological remains.

Our project evaluated 100 sq km of high-resolution multispectral imagery using established methods of image analysis to discover features associated with past human activity. Image analysis was supplemented and extended by ground control and geological investigation to improve the accuracy and efficiency of site detection, as well as to determine the nature of the relationship between the features visible in the image and archaeological evidence found in the field. The process, built around iterative image analysis and ground control, led to the discovery of previously unknown sites, shedding new light on settlement patterns in the environs of L’Amastuola. Results were compared with existing data from the Murge Tableland Survey (MTS), a systematic archaeological surface survey that was conducted in a transect across the Salento Isthmus between 2003 and 2007 and included 10 sq km within our study area. Variables such as rates of site recovery, time and labor costs, and the overall character of the results were compared. Here, we assess the relative utility of both approaches in an extensive regional investigation, indicating how surface survey and satellite image analysis profitably complement one another.

The L’Amastuola Archaeological Project and the Murge Tableland Survey

The area investigated in the present project corresponds to the target area of the L’Amastuola Archaeological Project, designed by Jan Paul Crielaard and Gert-Jan Burgers of Vrije Universiteit Amsterdam and begun in 2003 with the aim of investigating the material culture, settlement patterns, and landscape archaeology of the site of L’Amastuola (Figs. 1, 2). The MTS, a sister project conducted between 2003 and 2007, involved systematic survey of a transect extending from the coastal plain of Taranto into the karstic Murge uplands. In 2007, members of this survey also assisted with ground control and systematic investigation of selected features located through remote sensing. The MTS provides a comparative dataset for the present remote sensing project.

Both the L’Amastuola Archaeological Project and the MTS fit into a research program started by Vrije Universiteit in 1981 and encompassing a much wider area-from the Salento Isthmus between Taranto on the Ionian Sea and Brindisi on the Adriatic, which connects the Salento Peninsula to the rest of Italy. From the beginning, the Vrije Universiteit fieldwork in Salento combined excavation, field survey, environmental research, and remote sensing to investigate settlement and landscape evolution in the study

Figure 1. Map of southern Italy showing the 100 sq km research area centered on L’Amastuola.
region. Initially, Vrije Universiteit fieldwork focused on the effects of Romanization on regional societies (3rd-1st centuries B.C.), but gradually the scope was widened to include the impact of early Greek colonization (8th-7th centuries B.C.) and subsequent urbanization-issues that have also motivated research around L’Amastuola.

L’Amastuola is considered a key site for the study of early Greek colonization, a much-debated phenomenon among modern classical archaeologists and ancient historians. The debate informs the interpretation of the archaeological evidence from excavations in the 1980s by the Soprintendenza dei Beni Archeologici della Puglia and from 2003 onwards by Vrije Universiteit. While some consider the site to be a colonial Greek stronghold in Greek-controlled territory, others believe that it was an autonomous settlement with a mixed Greek-indigenous population (Dunbabin 1948; Herring 1991; Boardman 1999; D’Andria 2002). The first scenario is thought to demonstrate the aggressive, expansionist nature of early Greek colonization as well as Greek superiority over the indigenous population, while the latter supports a model of coexistence and integration.

Setting aside the historical debate, the site occupies one of the most prominent hill peaks in the entire Taranto region. Apart from its dominant location, the settlement seems to have been selected with a view to exploiting a range of resource zones; its catchment area offers both fer-

Figure 2. The environs of L’Amastuola. Photograph taken from the Murge looking towards the Gulf of Taranto.
tile inland soils for cereal cultivation and a coastal lagoon zone amenable to animal husbandry.

It comes as no surprise that early archaeological prospection culminating with the intensive MTS has established that the area was densely inhabited throughout antiquity. Indeed, the region was much more settled in antiquity than during later periods, for which very few habitation sites have been identified. Early prospection discovered dozens of small, ancient, rural sites that are spread evenly over the landscape, but clustering in some places in village-like settlements. The large amount of ancient offsite material recorded by the MTS also suggests that the area was exploited intensively, in particular during the late Classical and early Hellenistic periods (late 5th through 3rd centuries в.с.). This existing body of settlement information acquired through survey offered an ideal test case
for remote sensing. The results of the present investigation contribute to a fuller understanding of settlement patterns and the environmental conditions underpinning them.

Satellite Image Data

This project used a QuickBird satellite image as the basis for analysis. At the time the project was initiated, QuickBird was the highest-resolution satellite imagery commercially available, with optimal panchromatic resolution of 0.61 m and multispectral resolution of 2.44 m . QuickBird’s multispectral information includes separate red, green, blue, and near-infrared (NIR) bands. Our image was archival rather than newly tasked, collected on 18 March 2004 (the cost savings offered by archival imagery was justified because there were only modest changes in the landscape over the three-year interval between image

acquisition and the beginning of our project). The early spring date of the image captured vigorous plant growth, increasing the contrast between healthy and stressed vegetation that can reveal archaeological remains. At the same time, plant growth had not proceeded so far by mid-March as to entirely obscure the ground, allowing for the detection of soil marks. A combination of factors including clear sky, excellent environmental quality, and a low off-nadir angle combined to produce an unobstructed image with relatively little distortion. A full description of QuickBird imagery can be found on the Digital Globe website (http:// www.digitalglobe.com/product/product\_docs.shtml).

Methods

Analysis of the imagery began with georeferencing and projection, followed by image overlay and enhancement. Ground control was carried out concurrently with feature identification because of time constraints and to improve the accuracy of feature identification through immediate feedback from the field. Information from ground control was used to refine ongoing feature identification in the image, as spectral responses or patterns consistently denoting modern or natural features were eliminated from consideration and characteristics of features frequently associated with ancient surface material became clearer. Ground control also provided the location and extent of sites and off-site scatters as defined by the MTS.

Image interpretation was performed blind, without knowledge of the location of sites previously identified by the MTS. Only after image analysis and ground control were complete did we compare the sites and off-site scatters newly discovered through remote sensing with previously known sites. This approach was employed in order to compare the results of remote sensing with surface survey and to avoid introducing a bias in favor of areas with known sites.

Georeferencing, Projection, and Image Processing

Before georeferencing using ground control points, the image had a root mean square error (RMSE) of 14 m , meaning that each pixel in the image had a 63%63 \% probability of being referenced to within 14 m of its actual location on the earth’s surface. After georeferencing by Samsung Lim, its accuracy was improved to an RMSE of approximately 3 m , an excellent result facilitated by the low offnadir angle of the image. The investigators determined that no further correction (such as orthorectification) would be worthwhile since the size and variability of an artifact scatter (the most common archaeological phenomenon encountered) does not require more precise mapping. After georeferencing, the image was projected onto a local coor-
dinate system (WGS 84, UTM 33N), and the two components of the image were combined so that the higher spatial resolution of the panchromatic layer was enriched by information from the multispectral layer.

Image Analysis

Archaeological analysis of satellite images relies on the assumption that certain spatial and spectral patterns or characteristics of vegetation or topsoil can be correlated with buried archaeological remains (Lasaponara and Masini 2006b; Lillesand and Kiefer 1994: 20-21). For many decades conventional photographs taken from a bird’s eye perspective have been used to identify patterns such as soil marks, crop marks, and shadow marks that may indicate past human activity (Crawford 1929; Parrington 1983; Riley 1987). Some of these patterns, however, only become visible through manipulation of the various color bands that constitute multispectral satellite imagery (typically blue, green, red, and NIR). Indicators of buried archaeological remains amenable to spectral analysis include vegetation vigor and soil moisture. To enhance patterns invisible in standard panchromatic or color photographs, the bands in multispectral satellite imagery may be manipulated manually or through the use of indices (automated mathematical operations on combinations of bands) such as the Normalized Difference Vegetation Index (NDVI) (Lasaponara and Masini 2006a).

Our analysis began with band combinations prioritizing red and NIR, as they best reveal differences in vegetation growth sometimes associated with subsurface archaeological remains. The ratio of NIR to red light reflected from plants indicates the health of vegetation, in turn revealing the quality of the soil and substrate. The chlorophyll in healthy vegetation reflects NIR and absorbs red. Such “positive” crop or weed marks may identify aerated, moist, or fertile soils, sometimes revealing filled ditches, graves, or other “cuts” containing disturbed soil that retain water and nutrients. Conversely, stressed plants are associated with low NIR and high red values. These “negative” crop or weed marks may reveal packed, dry, or infertile soils, sometimes indicating underground masonry that is depriving plants of water and nutrients (FIGS. 3B-C, 4A-D) (Parrington 1983: 183; Masini and Lasaponara 2006).

Contrasts between high NIR values (healthy vegetation) and high red values (stressed vegetation) were sought primarily through manual band recombination displaying, for example, NIR as purple and red as green while excluding all other color information (a 4-1-4 band combination). The results of manual band recombination were then supplemented with transformations such as NDVI and Principal Component Analysis, although the latter did not

Figure 3. Features detected in the image. Left column is panchromatic; right column is multispectral (4-1-2 band combination). A-B) Rectilinear feature F1002 proving to be a false positive (protruding bedrock formation); C-D) Soil mark F1014 prominent in the 4-1-2 band combination (confirmed by ground control as a prehistoric site); E-F) Linear feature (crop mark) F1007 is a false positive (modern pipeline).

prove useful in our study (Lillesand and Kiefer 1994: 536-537). In their comprehensive discussions of the use of multispectral satellite imagery to detect spatially limited archaeological remains, Masini and Lasaponara relied heavily on the use of NDVI, an index which further increases the contrast between vigorous and stressed vegetation (Lasaponara and Masini 2006a, 2006b, 2007; Masini and Lasaponara 2006). Since there is such wide variation in NIR versus red reflectance produced by variable ground cover across our large study area, we used NDVI primarily to quickly distinguish between bare and vegetated areas and to confirm discoveries made through manual band recombination rather than as the principal means for detecting crop marks.

In addition to detecting crop and weed marks, remote sensing can also reveal variations in soil characteristics that may be directly associated with past human activity. Different soil characteristics (e.g., texture, chemistry, moisture, etc.) produce distinct spectral responses visible as “soil marks” in aerial photography or satellite imagery (Riley 1983: 9). The visibility of soil marks depends largely on the difference between the reflectance of anthropogenic residues and surrounding surface material. Some soil marks, for example, may appear as lighter spots against the darker background for most of the year (e.g., F1025; see FIGs. 4A-B). The higher reflectance in this case can probably be attributed to differences in soil moisture resulting from the better drainage of disturbed soils and/or the presence of subsurface masonry (Ur 2003: 105). In other cases, such as ditches or middens, the buried organic contents may leave a darker imprint on the surface. The level of contrast will depend on particular soil types and season; thin, xeric, and calcareous soils such as those in the Murge are fairly sensitive to fluctuations in moisture and other disturbances, facilitating the detectability of soil marks.

Soil marks, which are caused by variations in texture and chemical composition, tend to appear across all bands of a multispectral image (Lillesand and Kiefer 1994: 18-19; Ur 2003) As a result, soil marks should be more easily traceable than crop marks, a phenomenon dependant upon subtle differences in the vegetation health visible only in particular band combinations. In our image, some soil marks, particularly where bedrock had been plowed into surface soils, were readily visible in the panchromatic image (FIG. 3A). Elsewhere, soil marks appeared more clearly in a particular band combination, as was the case with Feature F1014, a soil mark produced by differences in soil composition and moisture retention (Vincenzo Simeone, personal communication 2008) (fIGs. 3C-D and compare Feature F1017, FIGs. 4E-F). Determining the nature of soil marks and distinguishing soil marks from crop or weed marks re-
quires ground control, especially when considering a large image. A greater degree of certainty about the origin of soil marks can be achieved if ground control takes place soon after image capture (one of the disadvantages of using archival imagery).

Idiosyncrasies in our image were selected for ground control based on whether or not they displayed distinctive patterns that had no immediately obvious natural or modern explanation. Both rectilinear and circular patterns were scrutinized under the working assumption that Greek and Roman sites would have a rectilinear form, while prehistoric sites might be circular. Particular attention was paid to features that did not align with the modern field system, roads, or structures such as field division walls.

In short, over the course of analyzing this image we found that, given the constraints of time and resources, the most effective way to quickly evaluate a large image with wide variations in vegetation cover, topography, and other parameters involved the use of a limited range of manual band combinations (4-2-1; 4-1-2; 4-1-4), paying special attention to idiosyncratic features that did not align with the orientation of modern structures and field divisions. Features were identified visually; in most cases they appeared as spatial patterns in the intensity of reflectance, usually in band combinations emphasizing the contrast between red and NIR. Sometimes we could determine whether these patterns were crop or soil marks through careful comparison of the panchromatic and multispectral images and automated transformations, but in most cases doing so required ground control. The immediate feedback provided by simultaneous ground control improved image interpretation; spectral responses associated with false positives identified early in the process could be excluded as the analysis proceeded, while those associated with ancient surface material could be sought out.

Ground Control

As noted above, ground control was conducted simultaneously with image analysis and informed feature identification in the image. Features were visited to identify any ancient surface material associated with them. Ultimately, they were placed into one of five categories: sites, off-site scatters, ambiguous (significant image anomaly but little or no surface material), false positives, and unassessed.

A team consisting of two or three people visited each feature, walked its perimeter and then walked several paths across it. Modern or natural features were noted as such, while features that were not obviously modern or natural were fully documented. The density of ancient surface material (if present) was systematically recorded. We employed the same site definition criteria as the MTS (a

Figure 4. Features detected in the satellite image. Left column is panchromatic. B) and F) are multispectral (4-1-2 band combination); D) is a Normalized Difference Vegetation Index (NDVI); A-B) Rectilinear feature F1025 (confirmed by ground control as a Hellenistic/Roman site); C-D) Grid pattern F1023, only visible in the 4-1-2 band combination and the NDVI reproduced here (confirmed by ground control as a Hellenistic/Roman site); E-F) Rectilinear feature F1017, a promising anomaly perhaps indicating a buried structure (ground cover prevented confirmation).

threshold of five sherds per sq m for historical sites and two sherds per sq m for prehistoric sites), and, like the MTS, we corrected for low surface visibility (Burgers, Attema, and van Leusen 1998: 3-4). Correction for surface visibility was particularly important because, unlike a typical surface survey, we could not choose fields based primarily on agricultural condition and visibility. Again, following the procedures of the MTS, off-site scatters that did not meet the site threshold (even after correction) were also recorded. Wherever ancient material was present, a grab sample was collected. The data collected through ground control allowed us to ascertain whether or not the features identified in the satellite image were associated with ancient material, and provided some indication of each site’s period of habitation and function. When no material was present, ground control often explained the origin of these false positives.

Several types of false positives were identified in the first days of ground control. Outcroppings of bedrock, modern agricultural improvements or soil conditioning, and underground pipelines were the most common. As ground control proceeded, image patterns associated with these features became readily identifiable, and were eliminated during the subsequent image analysis. Conversely, features identified in the image that proved to be associated with ancient surface material were scrutinized, and a careful search was conducted for similar features thereafter. Nine features identified in the satellite image could not be subjected to ground control because of inaccessibility or destruction between the date the image was acquired and the time of investigation. These features were excluded from consideration.

Results

Over the course of approximately three weeks of fieldwork in July 2007, this iterative process of image analysis, ground control, image review, and subsequent ground control was performed across the entire northern half of the image. The southern half of the image was completed during an additional 10 days in June and July 2008. In total, over 70 sq km were assessed (FIG. 5). One hundred and twenty-three features of interest were identified in the image and inventoried. Ground control evaluated 1.45 sq km , including 114 features. Urban areas were omitted, as was an area in the extreme southwestern part of the image that was deemed very unlikely to yield any ancient remains since it was a wetland before drainage in the 20th century and is now subject to intensive use.

Ground control determined that 14 image features corresponded to ancient surface scatters that met the MTS’s definition of a site (after modest correction for surface vis-
ibility). Significant off-site scatters were associated with an additional 15 image features. Another 13 image features displayed such distinctive and unusual reflectance patterns that they remain ambiguous; they had no obvious explanation and often low or no surface visibility (e.g., F1017 see FIGs. 4E-F).

As is the case with any archaeological survey, ground control associated with satellite image analysis is not immune to the problem of site definition. Moreover, it adds complications to site definition, which traditionally depends upon the quality, density, and boundedness of surface material. The issue of past human activity associated with chemical residues rather than surface material is of particular interest in remote sensing. Herding enclosures and pastoral camps that contain little surface material are in this category. While in theory such “sites” should be detectable through remote sensing under propitious circumstances (repeated use, accumulated deposits, and environmental conditions amenable to preservation), the impossibility of their confirmation and dating through non-invasive techniques prevents conclusive identification. We believe that at least some of our “ambiguous” sites fall into this category.

The characteristics of features associated with ancient surface material varied widely across the image. Most striking were unusual patterns likely caused by subsurface structures or cuts oriented against the pattern of modern agricultural divisions. Among confirmed sites were rectilinear features (F1024, F1025), semicircular and circular features (F1107, F1108, F1036, F1054-1062), and grids (F1018, F1023; for F1023 see FIGs. 4C-D). Another group of features associated with surface material was distinguished by areas of high NIR reflectance. Occasionally, a particular spectral response would go hand in hand with a rectilinear pattern visible in other bands. Features F1018 and F1023, visible in the panchromatic image, were accompanied by an intense NIR response. One general association involved historical sites; such sites were often accompanied by contrasts in brightness cutting across all bands creating rectilinear patterns interpreted as negative crop marks that probably indicate the presence of subsurface structures.

Three major sites, two prehistoric and one historical, were recognized by their unusual, intense, and spatially bounded NIR reflectance (features F1014, F1018, and F1054; for F1014 see FIGs. 3C-D). This characteristic reflectance may be caused by soil chemistry, but its exact origin has yet to be determined. Feature F1009, a discrete rounded area in the image, produced a high NIR reflectance similar to that of prehistoric site F1014 but yielded no material during ground control. Herding and stabling may have been economically important, especially in

Figure 5. Satellite image of the L’Amastuola research area analyzed in 2007-2008 with image features indicated.
marginal agricultural areas such as the Murge, but they leave few durable artifacts (Cribb 1991). Could the NIR reflectance of these features, one prehistoric, one historical, and one unidentified, represent a chemical or mechanical residue characterizing pastoral enclosures, where soil
might have been disturbed, phosphate-rich, and moistureretentive as a result of herding?

Overall, 14 out of 114 features ( 12.3%12.3 \% ) identified in the satellite image and assessed in the field yielded surface finds that met the density criterion for a “site” employed by the

project. Another 15 features ( 13.1%13.1 \% ) yielded some ancient material below the site threshold, while 13 ( 11.4%11.4 \% ) remain features of interest despite the fact that they have not yet yielded any ancient material. Thus, some 25.4%25.4 \% of features yielded significant surface material, while another 11.4%11.4 \% could neither be confirmed nor eliminated from consideration.

False Positives

Still, some 72 of 114 (63.1%) features were eliminated from consideration after ground control. Such false positives further illustrate the strengths and weaknesses of archaeological prospection using satellite image analysis. Many were the result of modern agriculture or other activity ( 27 or 23.6%23.6 \% ) or natural phenomena ( 33 or 28.9%28.9 \% ). In one case (F1007), the feature proved to be an underground pipeline (Figs. 3E-F). Another feature (F1001) consisted of filled ditches once dug for the irrigation of olive trees, but never used. Other features represented the ruins of masserie, the abandoned early modern farm complexes of southern Italy (F1105 and F1106); since ancient sites-including L’Amastuola itself-sometimes occur near masserie, these features were not rejected outright, but were thoroughly investigated. Common agricultural practices that produce suspicious image features include importation of topsoil, installation of irrigation systems, and a land amelioration process that involves excavating ditches in the bedrock, pulverizing the resulting debris, and returning the crushed rock to the ditches. Indeed, many of the fields in the southern half of the image (on the coastal plain, where intensive, large-scale, industrial agriculture is the norm) had undergone disruptive soil remediation, rendering archaeological prospection impossible.

In short, most of the false positives were similar in nature to the phenomena sought by archaeological prospection: either (modern) crop or weed marks accompanied by (modern) surface material, or soil marks arising from variations in the characteristics or composition of the soil. Other features resulted from the presence of modern material on bare ground (total of nine features: piles of modern pot sherds, brick fragments, or other rubble).

Comparison with the Muyge Tableland Survey

Even considering the number of false positives, the number of sites associated with features identified in the satellite image proved higher than would be expected from a random sample. The MTS, which explored a representative transect of our study area, yielded an average of 6.3 sites and off-site scatters per sq km ( 63 sites in total). At this rate, an area the size of that analyzed during remote
sensing ground control ( 1.45 sq km ) should have produced a total of about nine ancient sites and off-site scatters. The discovery of 29 sites and off-site scatters exceeds the number expected from a randomly chosen area of equal size by more than three times.
“False negatives,” sites or off-site scatters previously discovered by the MTS but not detected during our image analysis (FIG. 6), also reveal the value and limitations of remote sensing. Our project encountered 51 such false negatives. Most remarkably, the large ( 2.91 ha ) necropolis located south of the settlement of L’Amastuola could not be located; the exposed tombs cut directly into the limestone bedrock and partially covered by pines and macchia (Mediterranean scrub) were invisible in the satellite image. The small size of individual tombs ( 1.5×0.5 m1.5 \times 0.5 \mathrm{~m}, equivalent to one or two pixels in the image) combined with the bright reflectance of the exposed bedrock, uneven topography, and patchy surface vegetation, rendered the necropolis indistinguishable from naturally eroded bedrock outcroppings in the vicinity (even now, knowing its location and having visited it several times, none of the investigators can distinguish the tombs of this necropolis from unaltered bedrock in the image). Similarly, neither the shallow soil of the hilltop where L’Amastuola itself is situated, nor the vegetation it supported, revealed any traces of the rectangular structures recovered through excavation; only the fortification wall around the site was visible as shadow and weed marks produced by macchia protected from the farmers’ plow by collapsed masonry (Burgers and Crielaard 2007). Most of the sites and off-site scatters missed during satellite image analysis, however, consisted of numerous small scatters, a tendency reflected in the difference between the median size of scatters discovered through remote sensing ( 0.65 ha ) versus surface survey ( 0.1 ha ) (table I). The smallest tier of sites proved difficult to detect through image analysis, even using high-resolution imagery.

Twelve image features corresponded to surface concentrations previously defined as sites or off-site scatters by the MTS (out of a total of 63). Curiously, only eight of these 12 features yielded site or off-site sherd densities during ground control. The other four produced little if any ancient surface material (listed as “ground control failure” in Table 1). Since variations in surface visibility are probably not responsible (most of the sites in question are located in fields characterized by well-established perennial agriculture), this discrepancy highlights a familiar problem in archaeological surface survey: later fieldwork may fail to reproduce initial results when sites are resurveyed (Barker 1984; Terrenato and Ammerman 1996: 94). The incon-

Figure 6. Comparison of satellite image features and MTS sites (2007-2008) in the northeastern part of the study area. Image features are solid black. The MTS sites that overlap or fall within 25 m of the remotely sensed features are solid white; false negatives (MTS sites more than 25 m beyond the image features) are crosshatched. Solid light gray background represents the transect units of the MTS.

Table 1. Surface survey and remote sensing comparison (all figures reflect only the 100 sq km study area investigated by both the present project and the MTS).

gruity may also be partly attributable to the size mismatch between (large) image features and (small) average MTS site sizes. MTS scatters often comprise only a tiny fraction of image features, and thus may have been overlooked by ground control, which did not duplicate the intensity of the MTS (perhaps revealing a tendency toward underestimating the presence of small sites in image analysis and ground control).

Conversely, image analysis led to the discovery of one previously unknown site and two new off-site scatters within the MTS transect. Although survey was more likely to find sites missed by remote sensing than vice versa, both techniques contributed towards comprehensive prospection. Remote sensing, furthermore, revealed seven sites and 11 off-site scatters that lay beyond the MTS transect, demonstrating how image analysis can extend the reach of traditional survey and lead to a fuller understanding of the study area.

Comparison of MTS and remote sensing results demonstrates both the efficiency and idiosyncrasies of satellite image analysis in this region. Image analysis and ground control involved a team averaging three people working for a total of four weeks (ca. 70 person-days). Considering only work conducted in the 100 sq km study area, the MTS involved a team of five working two to four weeks per year for five years (ca. 350 person-days). Satellite image analysis discovered 29 site and off-site scatters (plus 14 ambiguous features requiring further investigation), while surface survey discovered 63 scatters. Satellite image analysis was particularly effective at recovering (comparatively) large sites, so much so that the total area of scatter inventoried through remote sensing exceeds that recovered by surface survey ( 28.96 ha versus 18.04 ha ) (table i). In part, this
difference is explained by the fact that the MTS treated discrete scatters in close proximity to one another as separate entities, while we considered an image feature containing multiple concentrations of material as a single scatter. Moreover, once ground control demonstrated the presence of ancient surface material, we considered the entire feature to be a site or off-site scatter (in the case of off-site scatter F1034, for example, the area covered by visible scatter was considerably smaller than the image feature).

Overall, however, we found that image features corresponded reasonably well with surface scatters. In 2007, seven image features associated with site-density surface material were fully and systematically surveyed by the MTS team using the same methodology that they employed elsewhere. This survey found that site-density scatters in and immediately around the image features totaled approximately 85%85 \% of the area of the features. Even if the mean, median, and total scatter area discovered through remote sensing are adjusted downwards to reflect these results, satellite image analysis still tended to locate comparatively larger sites. As a result, we believe that high-resolution imagery is best suited for finding sites somewhat larger than those recovered through systematic surface survey-and that it is a very efficient means of accomplishing that task. In the L’Amastuola region, these larger sites were often associated with crop or soil marks reflecting particular geological phenomena rather than with subsurface archaeological remains.

Environment, Geology, and Image Feature Interpretation

Geological and pedological expertise are essential for conducting archaeological remote sensing focused on

buried remains; they are the keys to understanding the processes that mediate between the surface, which produces the reflection patterns visible in the image, and subsurface strata potentially containing archaeological remains. In geological terms, our study area comprises a section of the western coastal plain of Apulia, extending inland through the transitional zone to the Murge Tableland. The Murge belongs to the Apulian karst and is marked by rolling hills and ridges with an average altitude of 420 masl (Burgers, Attema, and van Leusen 1998: 2, 6). It was formed by tectonic uplift that separated it from the coastal plain and in the process created a network of gravine, impressive canyon-like valleys (Vincenzo Simeone, personal communication 2008). Nowadays, the plateau is only marginally exploited, as it is neither suited for olive nor cereal cultivation. Viticulture and orchards prevail in the accessible areas, while the rest is covered by macchia and pine groves.

Quaternary sediments in the study area display a profile resembling a sandwich: two permeable layers (calcarenite sandstone and limestone) bracket an impermeable clay layer. The top sandstone layer (calcare di Castiglione) is soft and fragile, while the lower limestone (calcare di Gravina) is hard. Water that falls on the surface percolates through the upper sandstone layer and is blocked by the clay (argile di Bradano). Wherever this layer of clay approaches the surface (as a result of uplift and erosion), it provides low-volume but reliable near-surface water sources (Vincenzo Simeone, personal communication 2008; van Joolen 2003: 5-7). Such phenomena are abundant within our study area, especially to the ne of Taranto, and proved to be important factors in interpreting the satellite imagery for archaeological purposes.

The geology and associated water cycle of the Murge region affects the distribution of archaeological remains. Climate in the Salentine region is meso-Mediterranean and the soil regime is xeric, indicating water deficiency for more than 90 days a year. The Brindisi region mean annual precipitation is 548 mm , but evaporation exceeds precipitation during hot summer months (van Joolen 2003: 4−5)4-5). Given these data, it is likely that during antiquity access to water was the principle factor limiting human habitation in the region.

Although some features discovered through image analysis and associated with ancient surface material likely indicate subsurface remains (e.g., F1023, F1025; figs. 4A-D), the majority do not directly reveal traces of past human activity. Instead, they correspond to near-surface water sources or areas of high soil moisture. In most cases, these features reveal well-watered areas in zones where the interface between layers of permeable sandstone and im-
permeable clay brings water near the surface (e.g., features F1036, F1018, and F1024). In other cases, they reflect depressions filled with water-retaining clayey soils (e.g., features F1009, F1017, F1023, and F1107).

Thus, our image analysis mostly revealed locations amenable to human settlement rather than buried archaeological remains. The majority of sites or off-site scatters discovered through remote sensing were detected due to their association with easily accessible sources of water, the limiting resource in the region. Access to water determines the productivity, or even the possibility, of most agriculture in Apulia. At the same time, near-surface water sources affect vegetation growth and soil moisture, and as a result are readily apparent in multispectral satellite imagery. Surface material recovered during ground control generally reflected habitation rather than burials. In some cases, sites did not lie within features visible in the image, but instead were clustered nearby, a pattern particularly true for smaller scatters (under 1 ha ). Reliable but low-volume near-surface sources produced by the geology of the region likely supported small settlements or seasonal camps.

Compared to surface survey, image analysis was most successful at finding sites in well-watered areas of a broadly xeric region, further supporting an environmental origin of features and associated archaeological sites. In relatively moist regions, the majority of sites discovered through field survey were also located through remote sensing (along with some “new” sites). Areas lacking near-surface water sources or marked by low soil moisture showed a much lower correlation between our results and those of the MTS, with more false positives and false negatives.

In short, through a combination of the nature of remote sensing, the propitious date of image capture, the fact that water is the limiting resource in the region, and the particular geological formations that produce near-surface water sources in the study area, image features associated with ancient surface material generally represent environmental conditions conducive to human habitation rather than subsurface archaeological remains. Image analysis produced less successful and more erratic results in areas lacking such water sources.

Conclusions

Our project used satellite image analysis based on highresolution multispectral imagery to assess a large, archaeologically rich study area quickly and efficiently, extending and complementing the results of surface survey. It produced positive associations of features visible in the satellite image and artifact scatters on the ground at a rate over three times higher than would be expected by random chance. Although some of the features identified in the

image were the product of subsurface archaeological remains, most represent environments conducive to settlement, particularly zones of near-surface groundwater or moisture-retaining soils. Image analysis was more successful in places containing such water sources than in uniformly dry areas. Judging from surface finds, habitation sites were more amenable to detection than funerary sites. Some features, areas with particularly high NIR reflectivity, have yet to be explained and would especially benefit from pedological analysis.

The differential ability of satellite image analysis to locate various types of sites in different environments must be considered when assessing its capacity and limitations for archaeological reconnaissance. Image analysis allows efficient assessment of large areas, but its inability to locate certain types of sites in certain environments (such as rockcut tombs or low-impact habitation in uniformly dry areas) means that it works best in combination with other methods of prospection, particularly archaeological surface survey. Image analysis reflects a multitude of factors, including the nature of cultural residues present, the environmental and geological characteristics of the study area, the propensity of the land cover to reveal subsurface structures, and other phenomena that vary by culture and region. The season and time of day in which the image is taken may affect the visibility of subsurface features through variations in vegetation growth, soil moisture, and surface reflectance. In short, remote sensing has its limitations; differential recovery of archaeological sites argues for remote sensing and systematic surface survey as complementary methods of reconnaissance.

Although satellite image analysis can produce results (in terms of the discovery of sites, even some missed by conventional surface survey) it still lacks a mature, rigorous, and systematic methodology. Image analysis needs to be deployed on a larger scale and comprehensively assessed to determine rates of site recovery and their variations across different archaeological cultures and natural environments. Until then (and perhaps even after) remote sensing is best used to complement other means of prospection, such as surface survey. Despite these limitations, our project has demonstrated that remote sensing allows the rapid and efficient identification of some subsurface archaeological remains and, especially, of particular environmental conditions amenable to ancient habitation. These results suggest that one of the most useful applications of archaeological remote sensing may be to predict areas of human activity near places where a critical resource such as water exists in an otherwise deficient environment. An approach which combines surface survey, geological and environmental analysis, site location modeling, and remote sensing will
produce a powerful tool for regional archaeological prospection.

Acknowledgments

The authors would like to thank Samsung Lim, School of Surveying and Spatial Analysis, University of New South Wales, Vincenzo Simeone, Department of Environmental Engineering and Sustainable Development at Polytechnic University of Bari, and students from the Archaeological Center of the Free University of Amsterdam, whose assistance greatly facilitated this research. We would also like to thank three anonymous reviewers whose comments greatly improved this paper.

[1]

1. Shawn Ross (Ph.D. 2001, University of Washington) is currently a Lecturer in Ancient Mediterranean and World History in the School of History and Philosophy at the University of New South Wales, Sydney, Australia. His research interests include pre-Classical Greece, early Imperial Rome, the bistory and archaeology of trade, colonization, and imperialism, and the application of information technology to the humanities. Mailing address: American Research Center in Sofia, V. Petleshko 75, Sofia 1500, Bulgaria. E-mail: shawn.ross@unsw.edu.au
   Adela Sobotkova (M.A. 2005, Masaryk University, Brno, Czech Republic) is a Doctoral Candidate in the Interdepartmental Program in Classical Art and Archaeology at the University of Michigan, Ann Arbor. Her research interests include Black Sea archaeology, especially the rise of complex societies, empires and frontier environments, and archaeological applications of remote sensing and GIS. Mailing address: American Research Center in Sofia, V. Petleshko 75, Sofia 1500, Bulgaria. E-mail: adelas@umich.edu
   Gert-Jan Burgers (Ph.D. 1998, Vrije Universiteit Amsterdam, Netherlands) is the Head of Archaeology and an Assistant General Director at the Royal Netherlands Institute in Rome (KNIR) and directs all archaeological projects in Apulia operating under the auspices of KNIR. His research interests include landscape archaeology, archaeological method and theory, Greek-indigenous interaction, Greek colonization, Roman imperialism, and questions of acculturation. Mailing address: Koninklijk Nederlands Instituut Rome, Via Omero 10/12, 00197 Roma, Italy. E-mail: archeo@knir.it
   Barker, Graeme
   1984 “The Montarrenti Survey, 1982-3,” Archaeologica Medievale 11: 278−289278-289.
   Boardman, John
   1999 The Greeks Overseas: Their Early Colonies and Trade. New York: Thames and Hudson.
   Burgers, Gert-Jan, and Jean-Paul Crielaard
   2007 "Greek Colonists and Indigenous Populations at L’Amas- ↩︎

tuola, Southern Italy," Bulletin Antieke Beschaeving 82: 77−11477-114.
Burgers, Gert-Jan, Peter Attema, and Martijn van Leusen
1998 “Walking the Murge. Interim Report of the Ostuni Field Survey (Apulia, Southern Italy),” Studi di Antichità 11: 257-282.
Casana, Jesse, and Jackson Cothren
2008 “Stereo Analysis, DEM Extraction and Orthorectification of CORONA Satellite Imagery: Archaeological Applications from the Near East,” Antiquity 82: 732-749.
Crawford, Osbert Guy Stanhope
1929 Air Photography for Archaeologists. London: H. M. Stationery Office.
Cribb, Roger
1991 Nomads in Archaeology. Cambridge: Cambridge University Press.
D’Andria, Francesco
2002 “Greek Colonization and Romanization from a Native Perspective,” in P. A. J. Attema, Gert-Jan Burgers, Esther van Joolen, Martijn van Leusen, and B. Mater, eds., New Developments in Italian Landscape Archaeology: Theory and Methodology of Field Survey, Land Evaluation and Landscape Perception, Pottery Production and Distribution. Oxford: Archaeopress, 152-159.
Digital Globe, Inc.
2009 “Digital Globe,” (http://www.digitalglobe.com/product /product_docs.shtml)
Dunbabin, Thomas James
1948 The Western Greeks, The History of Sicily and South Italy from the Foundation of the Greek Colonies to 480 B.C. Oxford: Clarendon Press.
Fowler, Martin J. F.
1996 “High-Resolution Satellite Imagery in Archaeological Application: A Russian Satellite Photograph of the Stonehenge Region,” Antiquity 269: 667-671.
Harrower, M., J. McCorriston, and E. A. Oches
2002 “Mapping the Roots of Agriculture in Southern Arabia: The Application of Satellite Remote Sensing, Global Positioning System and Geographic Information System Technologies,” Archaeological Proisection 9: 35-42.
Herring, Edward
1991 “Socio-Political Change in the South Italian Iron Age and Classical Periods: An Application of the Peer Polity Interaction Model,” Accordia Research Papers 2: 33-54.
Kantner, John
2008 “The Archaeology of Regions: From Discrete Analytical Toolkit to Ubiquitous Spatial Perspective,” Journal of Archaeological Research 16: 37-81.
Lasaponara, Rosa, and Nicola Masini
2006a “Identification of Archaeological Buried Remains Based on the Normalized Difference Vegetation Index (NDVI) from QuickBird Satellite Data,” Institute of Electrical and Electronics Engineers Transactions on Geoscience and Remote Sensing Letters 3: 325-328.
2006b “On the Potential of Quickbird Data for Archaeological Prospection,” International Journal of Remote Sensing 27: 3607−36143607-3614.
2007 “Detection of Archaeological Crop Marks by Using Satellite QuickBird Multispectral Imagery,” Journal of Archaeological Science 34: 214-221.

Lillesand, Thomas M., and Ralph W. Kiefer
1994 Remote Sensing and Image Interpretation. New York: Wiley.
Madry, Scott L. H.
2007 “An Evaluation of Google Earth for Archaeological Aerial Prospection and Site Survey,” in Jeffrey T. Clark and Emily Hagemeister, eds., Digital Discovery: Exploring New Frontiers in Human Heritage. CAA 2006: Computer Applications and Quantitative Methods in Archaeology. Proceedings of the 34th Conference, Fargo, United States, April 2006. Budapest: Archaeolingua, 303-311.
Masini, Nicola, and Rosa Lasaponara
2006 “Satellite-Based Recognition of Landscape Archaeological Features Related to Ancient Human Transformation,” Journal of Geophysical Engineering 3: 230-235.
Parrington, Michael
1983 “Remote Sensing,” Annual Review of Anthropology 12: 105−124105-124.
Philip, G., D. N. M. Donoghue, A. R. Beck, and N. Galiatsatos
2002 “CORONA Satellite Photography: An Archaeological Application From the Middle East,” Antiquity 76: 109-118.
Riley, Derrick N.
1983 “The Frequency of Occurrence of Cropmarks in Relation to Soils,” in G. S. Maxwell, ed., The Impact of Aerial Reconnaissance on Archaeology. London: Council for British Archaeology, 59-73.
1987 Air Photography and Archaeology. London: Duckworth.
Terrenato, Nicola, and Albert J. Ammerman.
1996 “Visibility and Site Recovery in the Cecina Valley Survey, Italy,” Journal of Field Archaeology 23: 91-109.
Ur, Jason
2003 “CORONA Satellite Photography and Ancient Road Networks: A Northern Mesopotamian Case Study,” Antiquity 77: 102−115102-115.
van Joolen, Esther
2003 “Archaeological Land Evaluation. A Reconstruction of the Suitability of Ancient Landscapes for Various Land Uses in Italy Focused on the First Millenium B.C.,” unpublished Ph.D. dissertation, Rijksuniversiteit Groningen, Groningen.
Wilkinson, Tony J., Jason Ur, and Jesse Casana
2004 “From Nucleation to Dispersal: Trends in Settlement Pattern in the Northern Fertile Crescent,” in Susan E. Alcock and John F. Cherry, eds., Side-by-Side Survey: Comparative Regional Studies in the Mediterranean World. Oxford: Oxbow Books, 189-205.