1. Introduction
Landscape archaeology is fundamentally concerned with recording of the distribution of ancient settlement and the environments within which they were situated, and makes use of a range of different sources of data to aid the multi-faceted and multi-phased process of landscape prospection [
1]. Remote assessment is important for directing effective survey on the ground, but it also provides insight into the distribution of archaeological sites in areas and/or regions that are otherwise inaccessible (e.g., areas of Afghanistan [
2,
3,
4,
5]; Syria [
6]; Thessaly [
7]). Satellite imagery and aerial photography now play an indispensable role in the landscape archaeology of many regions, but their usefulness is curtailed by their relatively restricted time depth. In contrast, for some parts of the world there are historical map series that document the landscape long before major intensive agricultural practices and development programmes removed heretofore preserved vestiges of past landscapes. These historic maps provide an important additional source of data for landscape archaeology that is complementary to remote sensing imagery and aerial photography, and has a similar but relatively under-utilised potential to generate major new insights into past landscapes. This paper argues that the digitisation and analysis of historic maps has unique potential for the landscape archaeology of South Asia, where there are a range of historical map resources and modern development has profoundly obscured complex archaeological landscapes.
The Survey of India 1” to 1-mile (1:63,360 scale) map series originated in a very specific imperial context [
8], and were based on surveys carried out from the mid–late nineteenth century onwards. These surveys were started in the wake of the conquest of Punjab and the then North-West Frontier Province and the Indian rebellion of 1857, which is also known as the First War of Independence. The resulting maps were published from the early twentieth century onward as part of series that were issued progressively and updated incrementally, and they were primarily designed to present information that was relevant to the military, such as the location of villages, roads, irrigation canals and the nature of land-use [
9]. The surveys and the maps they produced also incidentally documented the locations and, to some extent, the height and area, of thousands of elevated mounds that were visible in the landscape at the time that the surveys were carried out. It appears that in most instances the surveyors did not recognise these mounds as anything unusual. A significant proportion of these mounds were actually the remains of ancient settlements, some of which built up during the processes of the formation and abandonment of ancient settlements millennia ago.
While archaeologists have long been aware of the potential of the Survey of India 1” to 1-mile maps, and made use of them as early as the late 1940s/early 1950s [
10], their use has been limited in terms of the areal extent. J.R. Knox carried out a systematic assessment of a significant number of map sheets as part of graduate research in the 1970s, but these maps have not been used to guide more recent archaeological surveys or as a data source in combination with remote sensing data. As such, the significance of these maps as a data source for large-scale systematic mapping of archaeological sites has not been explored. This paper considers the historical context within which the Survey of India 1” to 1-mile map series was created, its incidental documentation of archaeological sites, and the degree to which surveyors knew what they were recording. To highlight the utility of this rich data source, this paper also outlines a systematic method for (a) georeferencing these maps and (b) identifies symbols that represent features and/or archaeological site locations.
2. Reconstructing Archaeological Landscapes Remotely
The declassification of spy satellite photographs (e.g., CORONA, GAMBIT, HEXAGON keyhole series) and the availability of Open Access remote sensing data (e.g., Landsat, Aster, Copernicus; [
1,
11,
12]) has meant that a wide range of imaging system data sources are now used (both extensively and intensively) for prospection. Furthermore, a substantial (and growing) body of literature now highlights the importance of historic satellite and remote sensing imagery for documenting ancient landforms and settlements, particularly those at risk of being modified, obscured and/or obliterated in the process of modern urban and rural development [
1,
3,
4,
5,
13,
14,
15,
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37].
Such high-resolution imagery has had a profound impact upon archaeological knowledge, but it does have limitations. Declassified spy satellite imagery only post-dates the 1950s and though it saw markedly improved resolution in the 1970s [
1], it has issues with distortion [
38]. Furthermore, not all parts of the world have coverage, let alone high-quality coverage, as much spy satellite imagery acquisition focused on geopolitical “hot spots”. While China, the former USSR, and parts of the Middle East are covered extensively in US imagery, and Soviet imagery covered parts of Europe, imagery for other areas is either poor or non-existent. Coverage for much readily available remote sensing imagery is global, but is often limited in terms of its resolution and chronological range, which means that many features are not visible because the imagery was acquired after instances of modern disturbance [
1,
15].
The use of aerial photography for archaeology has a long history, beginning with the use of balloons in the nineteenth century and being advanced through the methods of reconnaissance and documentation using aeroplanes developed during the First World War [
1,
39,
40]. Aerial archaeology in many parts of the world is now well developed (e.g., UK, USA, France, Italy, Jordan), and important historic imagery is available via substantial archives (e.g., National Collection of Aerial Photography, which contains images from the UK and much of the world). Historic aerial photos provide an invaluable record of many landscapes before they were disturbed by modern activities, and our understanding of the archaeology of some regions has been completely transformed through their use (e.g., Jordan; [
1,
7,
40,
41,
42,
43,
44]). In relative terms, however, aerial photo coverage is again limited and/or unsystematic. Vertical and oblique images also have specific advantages and disadvantages related to distortion and visibility of features, with the latter being affected the timing of the photographs (time of day and time of year etc. [
45,
46,
47,
48]).
Maps have a longer history, and thus considerable potential for documenting lost or disappearing landscapes. While early maps contain limited topographic information and lack accuracy, the information documented by mapping projects in various nation states and former imperial dominions from the eighteenth and nineteenth century AD onwards was of a different order of accuracy due to advances in methods and the scale of the endeavour. Starting from the Cassini maps of France and the maps created of the UK prior to the establishment of the Ordnance Survey, surveyors set out to document landscapes and topography systematically. As these maps were created at particular points in time, it is also possible to monitor change in landscape over time by comparing different maps, particularly where individual maps are updated and reissued [
49]. The maps produced by these systematic surveying projects have considerable potential for documenting archaeological sites, and it is notable that in the UK, there was clear interaction between the Ordnance Survey and the various Royal Commissions for Historic Monuments, such that easily visible archaeological sites are clearly marked on the maps and their legends [
49,
50].
There is growing interest in the history and execution of the mapping projects carried out in various imperial dominions, particularly those of the UK [
8,
51,
52,
53,
54], but also the former USSR. The inclusion of archaeological sites on Soviet era maps of parts of the former USSR and Afghanistan produced by the Soviet Military Topographic Service has been noted [
5,
28,
55], but these maps have been used in relatively limited areas and/or on a relatively small scale. It is notable that the types of sites that are easy to distinguish on these maps are clearly ancient sites with distinctive morphologies.
4. Geo-Referencing Survey of India 1” to 1-Mile Maps
In order to make use of the Survey of India 1” to 1-mile maps as a prospecting resource we have established methods for georeferenciation using both ArcMap and QGIS georeferenciation tools (a plugin using GDAL in the case of QGIS). The maps themselves, and the processes of reproduction, were comparably accurate with the Ordnance Survey maps of the same period, and likely used the Everest 1830 spheroid. They were, nonetheless, printed on paper and have also subsequently been held in library collections where they are often folded, which creates the potential for distortion, particularly in digitisation processes that involve photography or scanning, and both of these can also introduce distortion. All of these factors affect the process of georeferenciation.
Using WGS84 as the geodetic datum, the first step in the georeferenciation process consisted of the selection of two points from opposite corners at the graticule of meridians and parallels available at the frame of the map (the corner of the map frames are always coincidental with every 15 minutes on a lat/long grid, so each map covers ¼ of a degree). The coordinate values of these points transformed to decimal degrees were then inserted manually into the georeferenciation tool. These steps produced a rough initial georeferenciation that made it possible to scale, orientate and situate the map in its approximate geographical location. It is notable that the use of coordinates from the map graticule (even when employing four or six points geometrically distributed) were not accurate enough when compared with independently geo-referenced imagery, and produced large deviations across the map. After this first step, a georeferenciation method using target data from an already georeferenced source was adopted. Typically, ground control points (GCPs) were obtained in ArcGIS and QGIS using their world imagery map services. Both feature high-resolution aerial imagery that can be employed to identify features that were extant at the time that the initial surveying for the maps was carried out.
A minimum of 20 GCPs distributed evenly across each map was considered necessary for a reliable georeferenciation. GCPs were obtained from clearly delimited structures and features visible on each map sheet and modern remote sensing imagery, particularly canals, old structures and intersections in villages. Given the degree of modifications that have affected the whole of the study during the last hundred years and the scale of the maps, it was often difficult to find reliable GCPs. Many elements of the landscape have disappeared since the late nineteenth century AD, and most urban areas have changed dramatically, such that their layout as shown in the maps is no longer recognisable. The most useful elements for the retrieval of GCPs were crossings within the water channel network. Channels were carefully mapped during the nineteenth century AD, due to a combination of their lineal nature, the detailed surveying that went into their construction, and the core objectives of the original surveys, all of which mean that they are a source of reliable GCPs. Much of the large network of channels that is still being used has existed since the late nineteenth and early twentieth centuries, including many crossings and junctions. These offer well-distributed and reliable GCPs that can be correlated to the features still preserved today that are clearly visible and easy to identify in high-resolution aerial and satellite imagery. For similar reasons regional roads and railroads were preferred elements for the extraction of GCPs. For areas that lacked these features, elements such as crossings of local roads or the central points of villages were selected.
After the selection of GCPs and the creation of links between map features and their real coordinates in high-resolution satellite imagery, an evaluation process for ascertaining the reliability of the georeferenciation followed for each map was developed. The initial four points that were used to position in the map frame were eliminated as they introduced large errors, and the residual error values for each of the remaining GCPs’ (the difference between the location provided as a result of the transformation and the actual coordinates of the GCP) were calculated. The root mean square error or RMSE is a measure of how well GCPs conform to the real coordinates using the root mean square sum of all GCPs’ residual error values. RMSE values are thus dependent on the transformation method employed, that is, the mathematical approach adopted to deform the image to adapt it to the distribution of real coordinates. The coordinates of the selected GCPs and those of the corresponding points from the original image map file were stored in a table. This table was used to establish the links between GCPs and original points and, during the same process, to calculate residual error for each GCP and RMS value for the whole map using different transformation methods.
The transformation methods chosen for the rectification and georeferenciation process took into account the characteristics of the original 1” to 1-mile maps. Zero and first order polynomials, Helmert transformation and other similar methods were not adequate given the internal distortions in the maps. These distortions were not just due to the quality or accuracy of the maps but, more importantly, to the types of affects that the paper have suffered over time (all maps presented folding marks) and in the digitisation process (maps digitised in 2008–2009 were photographed in keeping with University Library practices of the time, and those digitised from 2010 onwards were scanned using a drum scanner).
Thankfully, the maps maintained a high degree of integrity and their high quality and precision advised against the use of higher transformation orders that can produce important deformations in the margins and in areas where no GCPs are available. Consequently, two transformation methods were tested: a second order polynomial, which is a standard method in the georeferenciation of maps, and the
Adjust transformation, which is an algorithm implemented in ArcMap that combines a polynomial transformation and triangulated irregular network interpolation approaches [
82]. In most cases, the
Adjust transformation was superior to the results of a second order polynomial and, therefore, this method was employed to generate rectified georeferenced maps.
As shown in
Table 1, the georeferenciation process produced maps with an average RMSE of 0.000430° (equivalent to around 47 m at this latitude) using a second order polynomial with values between 0.000641° and and 0.000141°. Using the
Adjust transformation the average RMSE was 0.000102357° (
c.10.3 m) with values between 0 and 0.000244° (
c.26.8 m). The rectified maps produced using the
Adjust transformation were considered to have enough spatial accuracy to extract features that could be checked during fieldwork. With a maximum RMS value of 26.8 m (which could have been larger depending on the area explored in each individual map), these provided accurate locations for the central point of mounds. Mounds, being systematically much larger than these error values, would consistently fall within the location suggested in the rectified maps.
Each individual map sheet in the Survey of India 1” to 1-mile series covers an area of 683.610 km
2, and the resolution of the images used meant that there was a pixel size of between 2.69 m × 2.68 m to 2.65 m × 2.73 m, though for some images this reached 5 m. The area covered by the 64 maps shown in
Table 1 is approximately 27,000 km
2, and spans large parts of modern Haryana and Punjab in northwest India (
Figure 7). This area is archaeological significant and was selected as it spans a variety of different climate zones and distinct ecological contexts [
83].
5. Identifying mound features on Survey of India 1” to 1-mile maps
With the maps georectified, it is possible to extract features of archaeological interest with a high degree of accuracy, and the manual detection and digitisation was carried out by several of the co-authors. Two GIS shapefiles were created to store the results of the digitisation, one for points and another for linear features. Both shapefiles had an associated table with columns destined to store information about the digitised features. These information categories included location coordinates, feature type, relief height and size, which were chosen to offer the maximum possible detail about the original features to assist future analysis (see below).
The point layer shapefile was dedicated to the digitisation of mound features depicted in the maps and recorded a single point per mound located in the geometrical centre of the feature. The table associated with this layer included information about the individual maps, edition, and year of publication of the map from which each point was retrieved, but also important data about each individual feature that was recorded. The type and colour of the line used to represent the mound features was noted as it provides an important indication of how the surveyor perceived the feature while in the field. As noted above, the methods used by the Survey of India meant that features and/or mounds that were similar in terms of size, shape and height might have been drawn using different methods, including ‘horizontal’ continuous contour lines, discontinuous ‘form-lines’, shaded relief, ‘vertical’ hatching or using a combination of these approaches (see below).
The size of the mounds was measured using the largest axis of the feature, and we opted to divide them into three categories: (1) mound features measuring up to 200 m, which were the most common and typically measure around 100 m; (2) mound features between 200 and 400 m, and (3) mound features with diameters of more than 400 m. As many of the Survey of India 1” to 1-mile maps provide information on the relative height of the mound features in feet, it was also often possible to include a measure of height, which in combination with the measure of sizes provides a useful way to characterise the volume and character of each mound feature. Spot heights on these mounds were taken during the process of surveying, and the elevation of the mounds may have made them suitable places for circuit stations or survey points.
The second shapefile layer aimed at recording linear features. These appear to consist mostly of earthworks and relict field systems, and lines were recorded following the axis of the features. Apart from the information relative to the map from which linear features were extracted, the table associated to the linear feature shapefile layer included information on the type of line and relative height of the feature. The tables for both layers also included a field in which notes about the digitisation process could be included.
Digitisation was carried out using both ArcMap and QGIS. The manual digitisation of map features followed a systematic grid to ensure no areas were left uninspected. Observation of individual sheets and sets of sheets showed that mounds could be represented using three main approaches:
(a) ‘shaded mound features’, which were delineated with graded stippling and are perhaps the most common way of representing mounds, particularly those in the small size (1) category (
Figure 8a). According to the Survey of India 1” to 1-mile map legends, these features were defined as ‘sand-hills’, and the darker shading around the edges and/or a spot height indicates that they were formally surveyed (
Figure 8a);
(b) ‘form-line mound features’, which were delineated using a discontinuous horizontal brown or black line (
Figure 8b) that was usually employed for medium size (2) and large size (3) mounds. The use of discontinuous ‘form-lines’ indicates that they represent clear areas of elevation. The use of the horizontal rather than vertical lines to depict elevation is potentially related to the shallowness of the slope. It is likely that the form-lines were used as break-lines marking the transition between the flat plain and the elevated mound. The presence of a spot height presumably marks the highest point of the mound. In the process of hill-sketching, Surveyors were advised to use 25 feet contours ([
9], p. 557), so an elevation of up to 7.5 m would have only warranted one contour. The choice to use form lines rather than contours to depict such mounds is interesting, as it suggests that they were recognised as not being natural hills, but there are no clear references to methods for depicting such features in either edition of
A Manual of Surveying for India [
9,
65];
(c) ‘hachure mound features’ were delineated using vertical lines to depict elevation, which was potentially related to the steepness of the slope. Vertical hachures were sometimes used to represent small (1) and medium size (2) mounds (
Figure 8c).
Figure 9 shows the range of mound feature types that can appear on one sheet. This includes instances where a combination of approaches was employed to represent a mound feature, and in these cases, we used a ‘combined’ category.
The use of continuous black lines to demarcate small features can be confusing, but it appears that these typically represent ponds (
Figure 10a,b). When ponds contained water they were coloured blue (
Figure 10a), while those that were dry at the time of survey were represented as a continuous black line without a colour fill (
Figure 10b). Lines formed of points also appear, and appear to represent areas under cultivation (
Figure 10c, also
Figure 10a).
Figure 11a,b show the types of relationships that existed between some linear and mound features. Mound features are visible to the top left of
Figure 11a [(b) form-line] and towards the centre of
Figure 11b [(c) hachure marked with a relative height of 8 feet]. It is evident that some the lineal mound features follow the orientation and direction of then-modern roads, suggesting that they potentially follow earlier routes.
Figure 11b also shows how water channels go around mound features, potentially because of the topography created by the mound, which needed to be avoided for the water to flow.
Figure 11c shows a combined mound feature, where a discontinuous form-line co-occurs with graves. In these instances, it is likely that the graves were relatively modern, and were excavated into an existing mound feature.
Almost 9000 mound features (
Table 2) have been identified on 40 of the 64 Survey of India 1” to 1-mile map sheets listed in
Table 1 (
Figure 7). Within the area investigated for this study, shaded, form-line and hachure features were all common, but there is considerable variability in the number and size categories of mounds attested on individual sheets. The size 1 features in each category were the most abundant, but proportionally significant numbers of size 2 and 3 features were also located.
The variability in mound feature occurrence indicates that it is not yet statistically robust to attempt to produce summative data on the average number of mound features and the frequency of specific types per sheet, or to consider extrapolating this across a larger area. In future, when sheets that cover the full range of climate and ecological zones have been assessed, it will be possible to carry out predictive modelling of the number of mound types and size categories that might be expected in different zones.
6. Testing the Archaeological Viability of the Identification of Mound Features
The viability of the categorisation outlined here has been tested in the field over several field seasons [
84,
85,
86,
87] (
Figure 12), and the detailed results of this analysis are the subject of a separate paper [
88]. To facilitate the process of ground-truthing, each feature location was included in a field survey table and assigned Historical Feature Identification Numbers (hf_id), which will be used to create a comprehensive listing of preserved and potentially lost archaeological sites. Each hf_id was accompanied by information about its feature type and size category to allow the field survey team to assess the probability that a feature identified on the historical maps remains identifiable in the contemporary landscape, and the probability that the hf_id is (or more correctly, was) an archaeological site.
A sample of mound features in the size (2) or size (3) (i.e., those greater than 200 m across on the 1” to 1-mile maps) categories were visited and assessed to determine whether they were extant archaeological sites. This resulted in the identification of in excess of 200 archaeological sites within a delimited study area that were previously unknown [
86,
87,
88]. In the early stages of this survey, it became clear that the smallest features on the historical maps were rarely preserved. As a result, in each survey unit, historical mound features in the size (1) category were visited until at least ten features tested negative. In many instances, this meant that every historical mound feature identified in a survey unit was visited, but in some, several size (1) historical mound features were not visited as their likelihood of being a preserved archaeological site was demonstrably low. It is clear that it is not feasible to simply assume that all mound features were archaeological sites, not least because there is variation in the detail and clarity of each feature, and some of these features are almost certainly geomorphological in origin. Importantly, there is significant variation in the number of mound features on each map, as some have 400+ features, while others have as few as 60. This variability by sheet no doubt reflects variation on the ground, with some areas having more features, and potentially more sites. In addition to checking mound features, it is also possible to visit locations that include topographic words within the name to ascertain whether archaeological mounds are present. It has been noted previously that a number of modern villages in this region are elevated, and are likely built on top of archaeological sites. Ground-truthing is the essential component for demonstrating the veracity of the historic map dataset. It is also important to remember that many sites have been lost as agricultural development in the region has unfolded, which is a factor that will need to be considered in future studies.
7. Discussion and Conclusions
Imperial mapping projects in South Asia were inevitably geared towards the systematic documentation of the landscape to facilitate military domination, administrative control and economic exploitation. The Survey of India and the Archaeological Survey of India began as large-scale systematic documentation projects, and there are clear instances where information about archaeological sites that had been documented made its way onto 1” to 1-mile maps (e.g.,
Figure 5). There are, however, also numerous instances where otherwise undocumented archaeological sites were recorded on these maps (e.g.,
Figure 6 and
Figure 12), but the historical significance of these mound features does not appear to have been recognised formally at the time. It appears unlikely that the surveying teams were aware of the archaeological significance of what they were recording by and large, but the appearance of village names with some appellation of mound incorporated suggests that there was some awareness amongst local populations that these mounds were something unusual. Furthermore, many of these mounds are likely to have had significance for local populations as they were areas of sacred spaces as cemeteries, areas of economic exploitation – most particularly silt, sand and brick extraction - and their elevation means that they were not well suited for irrigation supported agriculture. This last factor also means that many mounds are currently under threat due to pressure from extensive and intensive farming and the ready availability of bulldozers.
These maps provide a new resource that can now be used to take major steps towards understanding long-term trajectories of human occupation in South Asia, and they will make it possible to develop new inclusive, comprehensive, and decolonised records of these evolving social landscapes. The systematic documentation of mound features visible on historic maps will make it possible to filter and query the data set at a later stage to select specific types of features, create thematic maps, quantify aspects of those features, and perform statistical analyses. Although substantial numbers of archaeological sites have already been documented across northwest India [
75,
76,
77,
78], attempts to conduct systematic survey using historic maps as a key data source has shown that the currently ‘known’ sites represent only a fraction of the actual archaeological settlements in the region [
84,
85,
86,
87,
88]. Although a large number of the almost 9000 mound features visible in the historical maps that have been studied are likely to be natural, if even one tenth of them turn out to be archaeological sites, then the number of known sites in the region will increase dramatically. The Survey of India 1” to 1-mile maps thus have the potential to revolutionise our understanding of the archaeological landscapes of South Asia.
Beyond their use for identifying and locating hitherto unrecognised archaeological mound sites, these 1” to 1-mile maps have considerable potential for reconstructing other types of landscape features that are also often overlooked, including palaeochannel levees, relict sand-dunes, raised road-ways and relict field-systems [
89]. This historical map data set can also be used to reconstruct historical landscape dynamics [
90], and/or the development of land-use practices, hydrological schemes and irrigation, and urban growth from the late nineteenth and early twentieth centuries.
As with other remote sensing and aerial photography datasets, there are inevitably a range of limitations to the Survey of India 1” to 1-mile maps, not least the fact that some areas were mapped repeatedly, while others do not appear to have ever had maps produced and/or made publically available. There are, however, extant archives that contain maps of areas of archaeological interest, and further research will require the establishment of new collaborations involving scholars and governmental institutions in Pakistan, India and Afghanistan. The 1” to 2-mile and 1” to 4-mile maps typically cover the interstitial areas, and they also document important archaeological data, but these series are inevitably of lower resolution. Nonetheless, they are an additional data source that also needs to be considered.
It is important to acknowledge that the methods outlined here for making use of these Survey of India 1” to 1-mile maps are only likely to be useful for identifying mounded sites, and will not be suitable to aid detection of a wide range of other features of archaeological significance. Therefore, it is essential to integrate the use of these historical maps into comprehensive approaches that make use of the full suite of earth observation and remote sensing techniques, potentially integrating open-source multi-spectral data and the computational power of platforms like Google Earth Engine to identify hydrological and topographic features not easily visible on the surface [
35,
90,
91]. It is also imperative that these remote prospection approaches are co-ordinated with large-scale ground-truthing surveys, that will verify which of the mound features are archaeological sites, and establish a reliable chronology for those sites and the associated landscapes. The combined analysis of these maps and the ground-truthing of the mound feature data will also make it possible to use machine learning-based approaches to carry out site detection across very large areas.
It is particularly timely to be able to add new evidence for the distribution of old settlements. The sobering truth is that in many instances, mound features in South Asia that were recorded in the early twentieth century are no longer extant, and in areas where farming is becoming increasingly mechanised and urban growth is unabated, mounds are disappearing with increasing speed. When such circumstances are in play archaeological sites and cultural heritage are clearly at-risk, and large-scale integrated mapping and survey projects cannot be commenced soon enough.