Remote Assessments of Site Damage: A New Ontology by Emma Cunliffe
Reprinted from
International Journal of
Heritage in the Digital Era volume 3 number 3 2014
Remote Assessments of Site Damage: A New Ontology Emma Cunliffe Department of Archaeology, Durham University, DH1 3LE
[email protected]
International Journal of Heritage in the Digital Era
volume 3 number 3 2014
453
Remote Assessments of Site Damage: A New Ontology Emma Cunliffe
Abstract: The archaeology of the Middle East is of immense significance to the history of mankind. However, due to modern development and the expansion of irrigation and agriculture, this priceless heritage is being damaged and lost at an unprecedented rate. It is neither practical, nor in many cases, possible, to visit sites to determine the extent of damage. Satellite imagery offers an unparalleled opportunity to assess and quantify the damage sites are experiencing. However, remote assessments present a new set of challenges in how to record aspects of damage which cannot be confirmed in the field, such as site visibility, and site depth. This paper presents a new ontology of damage, developed specifically for use in assessing sites remotely. In addition, it offers ways to incorporate additional factors such as site visibility on imagery. Whilst it was developed during an assessment of Syrian sites, it has far wider applicability. 1. INTRODUCTION Our heritage is facing a global crisis. The increasing population has a real need for shelter, food and a better quality of life. These needs can only be met by more intensive utilisation of the landscape, but unless this forms part of a major planned work, such as a dam or large irrigation project, in many countries no archaeological assessment or rescue work is usually carried out. Even then, limited definitions of the terms “sites” and “heritage” have meant that a great number of archaeological landscape features have been lost with no record made. Once rescue work has been carried out, sites are often consigned to their fate. Many sites, particularly in the Middle East, are located within increasingly intensivelyfarmed areas, supported by planned or unplanned irrigation schemes, or are in or near expanding developments built with the infrastructure they require – roads, electricity, gas, pipes and so on. No rescue work is conducted on these sites and the extent of damage is
454
largely unknown, although anecdotal evidence has suggested it is extensive [1-3]. Given the scale of the problem, it will never be possible to visit all the sites affected to gather data. Whilst many damage causes are broadly similar, the extent of damage sustained to each site is dependent on a number of localized factors, including geology, soil, and topography. Nonetheless, data is still required, so this AHRCfunded study was designed to attempt to quantitatively determine the effects of modern damage [4]. The study used data collected by Durham University’s Fragile Crescent Project to assess 161 sites in two case study areas of Syria, all of which had been surveyed as part of the Land of Carchemish Project [5-9] and the Tell Beydar Survey [1013]. Using ArcGIS and Google Earth, sites were located on CORONA satellite imagery from the 1960s, and their appearance was compared to imagery from the last decade. CORONA imagery is a vital tool for Middle Eastern archaeologists as it provides a record of the landscape at the advent of the implementation of intensive modernisation programs. Interpretation was aided by the survey notes (which provided control data), and by fieldwork conducted in 2010. The goal was to quantify change, without needing to visit the sites. However, there have been very few comprehensive studies of site damage, and even fewer using remote sensing. Damage is a widely used but poorly defined concept covering a wide variety of attritional effects. This led to difficulties with the existing terminology, which was inadequate for the methods used. As a result, a new methodology based on a new ontology of damage was developed to be of use to the heritage conservation community. At present, communities dedicated to sharing metadata are based in Europe and America. However, it is hoped that in the future, such projects will include the Middle East, offering a shared ontology that allows cross comparison of sites on a wider basis. The Europeana Data Model (EDM), for example, provides “a framework for collecting, connecting and enriching metadata” [14], available through the Europeana Professional website1, specifically designed to host and share standards such as this.
1. UNDERSTANDING DAMAGE The study of damage is hindered by a lack of definition. “Even the statement that a monument has been totally destroyed is only a relative statement based on the assumption that something once existed but now does not. … Thus survival is a point-in-time measure of the current state or condition of a monument relative to some former state, a 1 http://pro.europeana.eu/home [Accessed 29 August 2014]
Remote Assessments of Site Damage: A New Ontology 455
reflection of the cumulative effects of all the natural and man -induced processes that have operated on it” [15a]. A site which has been ‘damaged’ may have been completely flattened by a bulldozer, or cracked by weathering, or painted with graffiti. Damage is an all-encompassing term, which makes study of the problem particularly hard. To talk of the percentage of sites which have been damaged, for example, is meaningless unless the extent is also considered. Assessments must delve deeper into the concept, and refine it with new terminologies which can elucidate new insights into the nature of the problem. Site damage occurs as a result of a number of different sources, both natural and cultural [16, 17], with different extents and effects. Each threat also has specific characteristics which are visible on satellite imagery. Natural causes are part of the taphonomic processes which are part of the life-cycle of any site. Sites as we find them today, or would have found them a hundred years ago are the cumulative result of hundreds or thousands of years of abandonment processes. Cultural anthropogenic processes, on the other hand, have had far less time to affect sites, as they will only come into effect once an area is reoccupied. Not all damage is equally significant. There are several key factors which determine how significant damage to a site is, including the rarity of the damaged monument, whether it has greater importance due to incoporation in monument grouping, the amount of information available, and its condition before it was damaged [18]. This information is vital when determing preservation priorities, but it often requires comparison to the wider local, national and regional archaeological record, and such detailed information is not always available. As the goal here is only to asses damage – the first step in managing it - signficance has not been factored in, although it is hoped that this data could be used to target policies focussed on protecting the archaeological record. Finally, it must be acknowledged that whilst these threats are destructive, paradoxically some can also act to protect sites. At the Syrian site of Carchemish, orchards have partially destroyed the upper levels of a large part of the outer town, and yet also preserved the remaining sub-surface levels from the utter destruction caused by the expansion of the nearby town of Jerablus [19]. In their survey of the North Jazirah in Iraq, Wilkinson and Tucker note (p. 20) that several of the sites they surveyed were cut by canals and drains, the presence of which “enabled the context of the scatters to be examined” [20], which would otherwise have been impossible due to permit restrictions. However, the benefits can be overstated: eventually there comes a point in the life-cycle of a site where it is so damaged no
456
Emma Cunliffe
useful information can be extracted, and it is no longer of archaeological value.
2. PREVIOUS SURVEYS There have been no significant published studies of site damage in Syria, although some have discussed its effects on their field surveys [3]. Site survival, and measurements of it, is a point-in-time measurement to which different approaches have been taken, depending on data and time available. The main studies which have informed the development of this methodology are (briefly) summarised here.
2.1 Nationwide Studies The most notable study is the Monuments at Risk Survey [15], which looked at site damage across the whole of England. This survey, also known as the MARS survey, developed a framework to, amongst other things, “collate information on the condition of these monuments so that the resource requirements for future preservation, and the priorities for action, can be assessed ... on a systematic and nation-wide basis” 18. Data was collected through site visits on a national scale over an extensive period of time. The assessment was based on a combination of information sources, including the Sites and Monuments Record (SMR), which detailed the site type and materials, the known condition of the site and site visits to determine the actual extent as compared to the record. Various criteria were also developed to try and minimise observational bias by the different people conducting the study. Based on a consideration of the previous systems to quantify archaeological monuments and some trial and error15b, damage extent was calculated using the percentage area loss of sites over time. Acknowledging the conceptual and practical problems, as far as possible the Projected Archaeological Extent (PAE) (i.e. the largest the site could have been based on all known data), the Current Archaeological Extent (CAE) and the Current Maximum Height (CMH) were recorded for each site for all time periods where data was available between 1945 and 1995. From this the percentage area loss of each site was calculated.
2.2 Middle Eastern Studies One of the first damage assessment studies to use remote sensing was Parcak’s study of tells in Egypt [21], using a methodology which she then applied across the Middle East. Her method focused primarily on identification of the multispectral signature of sites on various imagery
Remote Assessments of Site Damage: A New Ontology 457
types, and discussed the different problems in the different areas of Egypt, comparing the Delta and Middle Egypt. Her primary imagery types were Landsat (30m resolution) SPOT (20m resolution), Quickbird (60cm) and CORONA (unspecified, usually 2-4m resolution). Parcak measured the visible area of sites on imagery and how they changed over time. In one area alone, 23% of ancient sites disappeared over a 30 year period, with 76% undergoing full to partial removal.
2.3 Topical Studies Several studies have been carried using remote sensing to assess the extent of looting on sites. Remote assessment of looting is particularly difficult, as a visible hole does not necessarily indicate disturbed archaeological layers. Looters may not have dug in the right place, or dug deep enough to reach archaeological material, and the site type and depth is often unknown, as is the amount of damaged and removed material. In the wake of the wars in Iraq in 1990 and 2003, and the ensuing loss of administrative control, reports indicated that looting was spiraling to previously unheard of proportions. The total area looted after the 2003 war was estimated to be greater than all the archaeological investigations ever conducted in southern Iraq [22]. Ground conditions prevented visits to all the sites to assess the extent of the problem, so in 2004, Van Ess et al. [23] successfully used a semiautomated detection method to analyse IKONOS imagery to assess the looting and site damage at Uruk. Manual change detection – carried out on purchased and free Google Earth imagery - was undertaken at Isin in Iraq [24], and in the Virú Valley in Peru [25]. Both studies assessed change in damaged areas. Another study, at Cahuachi in Peru, used spatial autocorrelation statistics to assess change on purchased high resolution imagery [26]: whilst the extent clearly increased, it was not quantified. Stone trialled a more detailed method to examine the scale of looting at over 1, 900 sites [22]. Data was collected to assess which were most likely to be in danger from looting: a limited number of those sites were examined on Digital Globe imagery acquired immediately before or shortly after the war. Attempting to account for differing conditions between images, nearly 10,000km [2] of imagery was examined. Criteria were developed for evaluating looting based on the density and distribution of the holes, and on visual traces, such as the sharpness of the edges and apparent size of the holes. The site circumference was estimated from the imagery, rather than site records, then digitised, and this was used to estimate the area damaged. Assessments were then made of the extent of the looted area, the targeted periods, and the intensity of looting as relating to size of site. Both demonstrated a clear increase in
458
Emma Cunliffe
looting holes on sites. A further advance was made using Google Earth imagery to assess cemetery looting in Jordan [27]. The study used free imagery and manual visual site assessment combined with information from the excavation reports to estimate how much of the looted area was actually archaeological, and to quantify the amount of looted material. Most recently (and post-dating this study), further work by Lasaponara et al. [26] measured the increase in the area of the looting, and for the first time used topographic correction to estimate the sub-surface depth of the illegal excavations. They were also able to examine the different sizes of looting holes, and related this to soil granulometry and therefore cavity stability.
2.4 Relevant Non-Damage Based Studies Some archaeologists have attempted to calculate site volume, as many sites in the Middle East are characteristically mounded. To do this, they calculated the size and height of sites from Digital Elevation Models (DEMs) using Shuttle Radar Topography Mission (SRTM) data, anthrosol data, and an estimate of the plain level the settlements were originally constructed on, and / or the level of shadows cast by tells [28, 29].
2.5 Problems and Issues However, these surveys were not without their problems. Firstly, many of them rely on field data to assess the current state of sites, which is often not possible, or expensive purchased imagery, which is not always feasible. Accurate comparisons can also rely on existing data, such as Sites and Monuments Records (SMRs). Middle Eastern countries often have no SMRs - in many cases the survey notes constitute the only available record. The second major problem is that methods used to determine site extent in the field are estimates only and there is often a significant variation between field measurements and those made from multiple satellite images [30], which can change from image to image. Studies that rely on area alone fail to account for the depth of a site, or its height. Measures such as the Percentage Area Loss used in the MARS survey did not take this into account as depth could not be determined in the field visits conducted. In order to be fully quantified, a percentage loss including the site depth requires the total volume of the site, which cannot be known without fully excavating (and thus destroying) the site. However, to only examine area can vastly underestimate the damage. Tell sites can be bulldozed down from their original height into low mounds: Mound D of Site 59 of the Tell Beydar Survey was bulldozed from its initial height of 6m to 3m during the years between two surveys [12, 31] but the area of the base (as would be mapped in the methods described
Remote Assessments of Site Damage: A New Ontology 459
above) remained unchanged, despite the loss of half the volume. Wells and looting holes in sites examined during the LCP Survey sometimes revealed several metres of subsurface deposits unaffected by changes to the surface topography. Even the volume models are unable to account for changes in ground level and subsurface remains, and no comparative data is available for the 1960s, when sites were in better condition. Nor is the data is updated regularly enough to monitor change now. Lasaponara et al’s [26] method offers new hope for accurate assessment of looting, as does the work by Contreras and Brodie [27], but these methods are site specific and damage threat specific. Neither are they necessarily viable over large areas if regional assessments of damage are needed. Finally, lower resolution multispectral imagery and automatic change detection methods, whilst less time-consuming, are often only capable of detecting changes to large sites. Parcak, for example, only studied tells, but most landscapes are covered with sites of all sizes representing the spectrum of human occupation. To focus only on large sites is to ignore the small sites which represent the substance of archaeological settlement density patterning and which are potentially more vulnerable and undergoing a higher rate of attrition. The true extent of a site requires a site visit, and even then without excavation the complete volume (from the point of the start of the excavation) cannot be known. Without this, exact volumetric loss cannot be estimated. However, a methodology is needed which is not reliant on expensive imagery, site visits or detailed excavation information, in order to be able to cover as many affected sites as possible.
3. METHODOLOGY AND TERMINOLOGY MARS was probably the first study to consider the area of a monument, the height / depth, and measurements relating to volume. These are vital characteristics of any damage assessment which seeks to understand patterns of damage, but all assessments are relative to the original site size, which can never be known. Satellite imagery has a long history of use in site prospection [28, 3238] but has rarely been used in damage assessment. Change detection on imagery is difficult, as the reflectance signature of sites can be altered by a variety of factors, including seasonality, moisture levels, and modification [33, 39, 40]. However, an initial trial assessment [41] determined that change detection was feasible using multiple sequential satellite images specifically comparing CORONA imagery, which is cheap, and the free imagery on Google Earth. Manual assessments of change in site condition were based on a thorough understanding of the effects of
460
Emma Cunliffe
each damage threat. The terminology trialled in that pilot study was too broad, and too vague, but the potential of the method was clear. Once corrected in ArcGIS, the CORONA imagery is of a fairly high resolution (between 2 - 4m), and sites can be located and clearly seen. Sites were then compared on a variety of more recent imagery – 2003 DigitalGlobe, 2004 SPOT2, and 2009 and 2010 Geoeye, all primarily available on Google Earth. Fieldwork was conducted in the survey areas to verify how sites and damage which were visible on imagery appeared on the ground, and to use this information to refine the methodology. It is not the intent here to discuss the nature or definition of sites, particularly in the Middle East3. However, it is worth noting that in some areas, such as the Tell Beydar survey area, a ‘site’ can consist of multiple site types, each of which will be affected differently by a given damage threat. In order to mitigate the chance that – if all sites types within a single surveyed site were examined together - the damage to large tell sites would hide the potentially greater damage to small mounded sites or flat sites, particularly when comparing tells to outer towns, sites were split into their component site types4. This increased the number of sites examined in the Beydar are from 83 to 108 site parts, and around Carchemish from 78 to 85 site types. In order to assess change, a series of relative categories was devised which would cover the damage visible on each sites. As each category is ordinal, they each reflect the increase in damage, but avoid inaccurate volumetric estimates. The first aspect which must be recorded is whether the site itself is visible, and then the extents of the horizontal and vertical damage. Damage extents were recorded for each threat, in order to calculate the greatest threats to sites. The number of threats affecting each site could also then be recorded. The goal was not to calculate the total volume of lost archaeological material, but to work out how it was being lost in order to prevent it. Finally it was important to assess the stability of each site: the various threats which affect sites are not static and can change over time, causing increasing (or more rarely decreasing) damage5. In the study described here, most the data is not discreet and sample sizes were too small to conduct normal statistical analysis. However, ordinal categories can be analysed with nonparametric statistical analyses, using the Wilcoxon Signed Rank Test, the MannWhitney-U test, and the Kruskal-Wallis test [42].
2
2004 is the date given by Google for the regional mosaic. According to the Astrium SPOT
website, the Syrian tile is most likely to be from 2006. 3
A full discussion can be found [4] - Chapter 2.
4
This is discussed more extensively in [4] and [38].
5
Examples of each category with images can be found in [4] - Appendix B.
Remote Assessments of Site Damage: A New Ontology 461
Given the caveats of this method, in all cases the least possible damage is assumed to have affected the site. This is termed the Principle of Least Damage. Whilst this could potentially under-estimate the damage, it will avoid over-stating it. Considering the extent of the anecdotal reports, it was expected that even if the damage was under-estimated, it would still be extensive.
3.1 Site Visibility Site visibility determines whether a site is visible on satellite imagery. In this study, sites were located using GPS points taken during the surveys. The early TBS points were only accurate to within 50m. In addition, the rectification process (the process of fixing the imagery to known points on a Cartesian map system) cannot always remove distortions in imagery of tens of metres. As a result, sites are not always exactly where the point places them. However, most sites were visible and could be located, so the following categories deal with finding sites (Table 1). Numbers in brackets indicate the variable used in statistical analyses. Once identified, sites were entered into a GIS by the Fragile Crescent Project Team, and boundary polygons were created in ArcGIS for each site6, 7[43]. These were then imported into Google Earth for comparison. (A distinction is made here between sites which are obscured, and sites which are not visible. Sites which are not visible may have been destroyed, whilst the condition of sites which are hidden from view is unknown).
Table 1 – Site
6
Visibility
Evidence
Visible (1)
A site is clearly visible
Partially visible (2)
The imagery resolution is poor so the site is only partially visible Part of the site is visible but part is not, either because the resolution of the imagery changes over the site, or some parts of the site are too small to see, such as water channels or rock cut tombs. Part of the site is obscured under a cemetery or modern village, or by clouds.
Barely visible (3)
Very little of the site is visible, usually because the resolution is too poor to make out more than the location of the site.
Obscured (4)
Something is preventing the site from being seen. It is not visible on imagery as it is under a cemetery, modern buildings, or underwater, for example, or the view is blocked by clouds or other weather conditions.
Not visible (5)
The site is not visible on the imagery at all, but nothing obvious (like buildings) is preventing it from being seen.
Thanks are due to Dan Lawrence and Andrea Ricci, who created many of the LCP
shapefiles, and Jason Ur, who created the TBS shapefiles) 7
For a full discussion of how sites were identified, see [43]. It included use of field maps,
textual descriptions, maps, and GPS points.
462
Emma Cunliffe
Visibility Categories
This is not to say that a site which cannot be seen cannot be assessed. For many sites, inferences about the damage can be made depending on the land cover. For example, a site under a town will almost certainly be damaged by the establishment of modern infrastructure and building foundations; the upper levels of a site in a field will be ploughed. Changes in visibility can also act as a proxy for damage. If visibility is considered as an ordinal variable, and counted numerically, then change in visibility can be considered. Visible = 1, Partially Visible = 2, Barely Visible = 3, Obscured = 4, Not Visible = 5 A site which was “Visible” in 1967 but becomes “Not Visible” by 2010 would move from 1 to 5 – a -4 rating. A site which becomes more visible would have a positive rating. These ratings can then be counted. As the resolution of imagery improves, sites should become more visible (and therefore gain a positive rating), but as sites become more degraded, they should become less visible (and therefore have negative ratings) (see Sample Results - Graphs 1 and 2). Although many factors can ‘hide’ sites on imagery, by considering the wider region area, more significant results can be achieved which are less influenced by the individual site constraints.
3.2 Horizontal Damage Horizontal damage is defined as the extent of each damage threat across the site. Table 2 shows the categories devised to monitor horizontal damage for each threat on each image. Numbers in brackets indicate the variable used in statistical analyses. Examples of these categories would include: a site which appears to be largely farmed would have the threat recorded as agriculture, and the extent as Majority. However, if a small road or track also crossed the site, this would be recorded as a second threat with the extent Fractional or Sectional depending on the size of the road / track relative to the site. The effect of each threat can vary for each site, and this must be factored into the assessment. A road which affects only a minute proportion of a large site can affect a significant area of a much smaller site. In such cases, it is not the size of the road which is important, but its effect on each site.
Remote Assessments of Site Damage: A New Ontology 463
Table 2 - Horizontal
Horizontal Damage Extent
Evidence
None / undamaged (0)
No damage is visible on the site. * Given the time span over which the sites have existed, and the natural decay processes which affect them, this was not actually applied to any sites
Unknown (1)
The field records for the site are missing, or the land cover on the site is not visible, so that the state of the site at the time of the visit is unknown. If no obvious damage is visible on a site, given the decay processes affecting them, damage is usually marked as Unknown, rather than none, as it is unlikely there is no damage. Natural taphonomic processes will always have affected a site, so No damage would be a misnomer. This therefore ranks more highly than None.
Peripheral (2)
The damage is around the edge of the site, and may be affecting the edges of the site, or may represent a threat which puts the site at risk, but is not currently affecting the site, such as an expanding quarry on the edge of the site. If damage extends on to the site, such as an orchard around and on the site, it is recorded as being on the site, rather than around it as well.
Intermittent / Fractional (3)
A very small part of the site is affected, or the amount of damage being done is very small, such as a hole dug for looting.
Sectional / Partial (4)
A large part of the site is damaged If the site can be split into clear sections, such as an upper and lower site, or areas of different date, that section (usually as defined by the field team) is affected.
Majority / Extensive (5)
Most of the site is affected by a single damage type except a small section.
Total / Wholesale (6)
The entirety of the site is affected, or the site is affected by wholesale damage.
3.3 Vertical Damage Detecting height on imagery is difficult, but not impossible. On prominent mounded sites, the height is clear from the distibution of light and shadow. On high resolution imagery height can be inferred from the pattern of plough lines. Disturbed plough lines can also often be used to indicate the presence of sub-surface remains [13]. Building on this, the vertical damage extent categories (Table 3) estimate the affected depth of different damage threats. This is estimated from analysing each damage threat combined with known site details. For example, based on accepted technology levels, ploughing depths can be reasonabley estimated. Site depths may be known from surveys, or can be inferred from the geographic region (soils are very thin in the limestone hills of the LCP area, and flat sites are rarely more than 1 or 2 metres deep). Therefore farming damages the upper levels of a site, unless the site is particularly shallow. Other vertical damage depths are estimated in a similar fashion. Numbers in brackets indicate the variable used in statistical analyses.
464
Emma Cunliffe
Damage Categories
Table 3 – Vertical Damage Categories
Vertical Damage Extent
Evidence
None / Undamaged (0)
As for horizontal damage extent
Unknown (1)
As for horizontal damage extent
Site Buried (2)
The site has been buried, perhaps by alluviation, colluviation, or the flood water and associated sediments behind a modern dam.
Pitted (3)
Pits have been dug in the site, perhaps for looting, or burials. The pits are of varying depth, and do not cover the entire site. This is considered to be a lesser degree of damage than upper levels damage . If the entire site is pitted, such as a cemetery covering the entire site, this is recorded as upper levels damaged .
Site Slightly Degraded (4)
This usually refers to a mound which is gradually degraded by ploughing, or a site which is being dispersed by erosion.
Upper Levels Damaged (5)
The upper levels of the site are being damaged, perhaps by ploughing or a small amount of levelling for a track.
Site Heavily Degraded (6)
The site is heavily eroded, or has been heavily dispersed / degraded by agriculture. This also refers to heavily looted rock cut tombs. Often when the tombs are looted, artefacts which are considered to be of little value, such as pottery, will be left strewn across the landscape outside the tomb. The tombs are heavily degraded by the looting.
Site Destroyed To Ground Level (7)
The site has been deliberately (rather than through natural processes) destroyed to the current ground level, but there are (or may be) subsurface remains.
Site Destroyed (8)
There is nothing left of the site, even below the ground.
3.4 Site Stability Not all damage threats are stable, affecting site condition over time. A gravel track may become a tarmac road, or scrubland may be converted to arable fields, with a corresponding increase in the effect on the site. Occasionally damage decreases – houses may be abandoned, for example, so damage associated with occupation ceases. These damage categories record whether the site is stable or if damage threats affecting it are changing. The different categories of stability are listed in Table 4, and were recorded for each threat on each image. Due to the complexities of recording the change, these categories are not ordinal. Damage increase is categorised by date, not by amount of increase, allowing an assessment of the rate of change. Whilst the terms given below are specific to the imagery used in this study, they can be replaced with any imagery dates, or can be left vague to determine increases or decreases without the finer precision afforded by dating imagery.
Remote Assessments of Site Damage: A New Ontology 465
Table 4 - Site Stability
Damage Stability
Categories
No increase visible
The damage threat is stable. Damage has occurred but it is not getting worse and no further damage is occurring or is expected to occur from that source: a road was built in the past, for example, or the site is under an orchard. (Although roots may grow, for example, this is not visible, and the damage it causes is dependent on site depth, which is usually unknown).
New
The threat is new: it was not recorded on earlier imagery.
Increasing
The damage which is occurring / has occurred from a given threat has increased since the previous image. For example a ploughed site has been converted to an orchard: the damage done by agriculture has increased (as the site has been subjected to ploughing and to the boreholes for trees). In these cases the date the damage increase was recorded is specified.
Increase since / between CORONA
The damage threat appears to have worsened since CORONA was acquired or even between CORONA image acquisitions.
Increase since DigitalGlobe 2003
The damage threat appears to have worsened between DigitalGlobe image acquisitions or since DigitalGlobe imagery (2003) was acquired.
Increase since / between field visits
The damage threat has worsened between field visits (recorded in field visit notes), or since the field visit.
Increase since SPOT 2004
The damage threat appears to have worsened since SPOT 2004 was acquired.
Damage Lessening
The damage has lessened (e.g. a track is no longer in use, or a field is no longer cultivated and thus erosion is lessened)
Not visible / Not applicable / Unknown
It is not possible to determine whether the threat has worsened or lessened.
4. SAMPLE RESULTS This methodology was tested in case study areas of Syria with excellent results. Over 600 satellite images of 161 sites (split by type into 193 parts) were analysed, and damage threats were recorded in a database using the above terminology. It was conclusively demonstrated that damage is under-recorded by archaeologists during field visits, and thus its impact on archaeological surveys cannot be estimated from existing records. The following represents a small sample of a much larger set of results achieved as part thesis research [4], intended to indicate the potential of the method and terminology. Sites were not always visible on the different images. Whilst most site visibilities did not change, many have decreased in visibility between the 1960s CORONA and the 20009 / 2010 Geoeye, despite a marked increase in imagery resolution. Around Beydar (Graph 1), approximately 1 in 5 (36 sites) became less visible, whilst only 15 became more visible. Around Carchemish (Graph 2), 20 sites became more visible, but 17 became less visible. This is due to the increasing damage sites suffered. In both areas, the difference is statistically significant: this is mostly attributable to the difference between
466
Emma Cunliffe
CORONA imagery and the later imagery. A Kruskal-Wallis test was conducted on all sets of imagery, where 1 = Visible and 5 = Not Visible: a higher mean rank indicates less visible sites. (TBS: X2 = 12.259, df 2, p=0.002. Mean ranks: CORONA = 138.15, SPOT = 178.65, Geoeye = 170.69. LCP: X2= 19.351, df 3, p
