A new aerial photogrammetric survey method for recording inaccessible rock art

Author’s Accepted Manuscript A New Aerial Photogrammetric Survey Method for Recording Inaccessible Rock Art Stephen Berquist, Giles Spence-Morrow, Felipe Gonzalez-Macqueen, Branden Rizzuto, Willy Yépez Álvarez, Stefanie Bautista, Justin Jennings www.elsevier.com/locate/daach

PII: DOI: Reference:

S2212-0548(17)30030-9 https://doi.org/10.1016/j.daach.2018.03.001 DAACH72

To appear in: Digital Applications in Archaeology and Cultural Heritage Received date: 23 June 2017 Revised date: 12 January 2018 Accepted date: 13 March 2018 Cite this article as: Stephen Berquist, Giles Spence-Morrow, Felipe GonzalezMacqueen, Branden Rizzuto, Willy Yépez Álvarez, Stefanie Bautista and Justin Jennings, A New Aerial Photogrammetric Survey Method for Recording Inaccessible Rock Art, Digital Applications in Archaeology and Cultural Heritage, https://doi.org/10.1016/j.daach.2018.03.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

A New Aerial Photogrammetric Survey Method for Recording Inaccessible Rock Art Stephen Berquista, Giles Spence-Morrowa, Felipe Gonzalez-Macqueenb, Branden Rizzutoa, Willy Yépez Álvarezc, Stefanie Bautistad, Justin Jenningsa a

University of Toronto

b

Western University

c

Royal Ontario Museum

d

Stanford University

Introduction Although archaeologists are increasingly deploying Unmanned Aerial Vehicles (UAV’s) in their work, the predominant usage has been to collect images for producing Digital Elevation Models (DEM’s), Digital Terrain Models (DTM’s), and georeferenced orthophotos of archaeological sites and landscapes (Wernke et al 2016; Bikoulis et al 2016). Archaeologists often discuss incorporating UAV's into archaeological survey, but workflows that do so effectively are still uncommon. This is particularly true in regards to surveys targeting smaller objects such as surface artifacts or petroglyphs that are not visible from a distance. In fact, to our knowledge, no such survey has been successfully implemented (see Field et al 2017). In the case of petroglyphs, this objective is often further complicated by surfaces that do not permit horizontal flight paths or pre-programmed flight polygons. In this paper we propose a novel survey method that will produce an analytically viable geospatial registry of petroglyphs with thematic and relational attributes. This method, relying on the use of a UAV, may be particularly useful in contexts with challenging topography or time constraints. Most efforts to digitally document petroglyphs to date focus on individual rock art panels or even the geometry of individual petroglyphs rather than on the “representational spaces” (Hubbard & Ruppel 2016). A smaller number of studies propose a workflow for digitally modelling the thematic constellations of rock art across a larger geographic expanse (Alexander et al 2015; Jennings et al 2014). These studies, however, have been conducted across relatively flat, accessible areas, with individual petroglyphs documented through pedestrian survey. Aside from being time-consuming endeavors, these research designs are often difficult to implement in contexts where rock art is difficult or dangerous to access, such as a high cliff face (Jennings et al 2017; Mark & Billo 2016) or the sides of a gorge (Fowles & Albert 2016). What is more, effective geospatial analysis of such contexts requires both precise locations of individual petroglyphs and an accurate three-dimensional terrain model. These data are difficult to simultaneously generate with previously proposed workflows. We argue that survey with an unmanned aerial vehicle (UAV) coupled with structure-from-motion photography, or

photogrammetry, allows archaeologists to quickly and accurately emplace petroglyphs within complex topographies, thus serving as a critical aid for conservation and research. This article presents our workflow to contextualize large expanses of rock art on a multi-faceted cliff face. We demonstrate our proposed workflow in reference to the site of Quilcapampa in the Majes district of southern Peru, a 70-hectare settlement situated directly above a two-kilometer escarpment characterized by a high density of incised petroglyphs (Figure 1). The petroglyphs, defined by Chandler et al (2007: 11) as: “engraved features, where rock has been scraped or pecked away in a subtractive process”, are largely contemporaneous with the fluorescence of the site from AD 800-1500 (Jennings et al. 2017). Both the site and the stability of the cliff face have been threatened in recent years by ongoing road construction. Documenting the petroglyphs has thus been of pressing importance, both as a digital heritage preservation effort and as a critical tool for archaeological interpretation. In this article, we briefly review existing methods of petroglyph documentation and digital heritage reconstruction. We then present our workflow for using a UAV to create a precisely georeferenced registry of petroglyphs on a three-dimensional model of a vertical cliff face. We close by outlining the value of this model for future geospatial analysis, site visualization, and public archaeology. Explanation of previous methods in documenting rock art As Lewis-Williams remarks (2006), archaeological interpretation of petroglyphs and their broader contexts have historically been significantly influenced by recording technologies. Documentation techniques may also impact preservation efforts. Archaeologists have traditionally employed a number of methods for documenting petroglyphs, including wax and latex rubbing, sketching, tracing, casting, and simple photography (Chandler et al 2007; Plets et al 2012). As both Bahn (2010) and Plets et al (2012) note, many of these techniques are intrusive and risk damaging the rock art. Moreover, many petroglyphs “are finely incised figures which are impossible to detect with these techniques” (Plets et al 2012: 139). Artists must pick and choose which lines are part of the artwork and which are natural cracks in the rock face (Bahn 2010). Indeed, Lewis-Williams (2006) demonstrates that even the advent of simple photographic recording techniques opened a new world of fine detail for researchers. LewisWilliams goes on to argue that the primary effect of the photographic revolution was a shift in how archaeologists organized and interpreted rock art: the necessity of attaching descriptive captions to collections of images prompted researchers to narrativize their interpretations of rock art panels. Moreover, film photography afforded a relatively limited quantity of image captures, producing a marked tendency to prioritize “good, framed picture(s)” that fit a particular interpretation (LewisWilliams 2006: 358). Ground-based digital photography has circumvented this difficulty, but still limits consideration of relationality and context. Geospatial analysis requires that petroglyphs be arduously geolocated and cross-referenced with the digital images within a landscape model that must be generated separately. Another concern is that such techniques are time-consuming, even as archaeological landscapes are increasingly at risk. Rock art in particular is threatened by both natural and anthropogenic factors. Tectonic activity can cause panels to fall from an escarpment (Jennings et al 2014), for example, and acid rain and pollution have intensified the erosion of rock art via wind, water, freeze-thaw cycles, and biotic pressure (Chandler et al 2007: 143). Climate change has been particularly damaging in the central Sahara, as increasing aridity alters the sandstone matrix (Hansen 1999). The greatest peril is development. Chandler et al (2007: 145), for example, note that petroglyph contexts in the Altai region

of Russia are increasingly threatened by pipeline projects. As mentioned previously, our site of Quilcapampa is endangered by a project paving the gravel road that sits just above the petroglyphs. Such infrastructure projects are important however, and not all archaeological heritage can be preserved. The best for which we can often hope is that stakeholders will permit researchers time to adequately document the remains. Systematic and efficient methods of data collection are thus invaluable to rock art researchers. Digital Recording of Rock Art As specialists in vulnerable material heritage that typically rely on destructive techniques, archaeologists are increasingly using technologies that reconstruct the three-dimensional form of artifacts, excavation units, and archaeological sites. Previous research has demonstrated a variety of useful methods for digitally documenting archaeological heritage (Fritz 2007; Wernke et al 2016; Sanz et al 2010; Pierdicca et al 2016). In this study, we have relied primarily on photogrammetry at varying scales to document and reconstruct both petroglyphs and their broader environmental context. Photogrammetry, more properly termed structure-from-motion photogrammetry, is a technique for extracting threedimensional point coordinates from sequences of two-dimensional images. Though photogrammetry is now most commonly conducted by software packages such as Agisoft Photoscan Pro, the basic technique can be done by hand and is nearly as old as the photograph itself. Indeed, past researchers have utilized analog photogrammetry to document petroglyphs in Britain (Atkinson 1968) and Australia (Rivett 1983). In the past decade, photogrammetric software packages have greatly improved in efficiency and decreased in price (de Reu et al 2013; McCarthy 2014), contributing to a growing ubiquity of digital photogrammetry and other forms of three-dimensional scanning in documenting artifacts, excavations, edifices, and even entire landscapes. Paleoanthropologists have reconstructed digital models of objects as small and fragile as Neanderthal teeth (Benazzi et al 2011). Such models provide a framework for analyzing morphology, biometrics, and wear patterns. Photogrammetry of artifacts such as lithics and bone tools can also generate important metrics regarding wear pattern, retouch intensity, depositional damage, morphometrics, and manufacture techniques (Bretzke and Conard 2012; Clarkson 2013; Davis et al 2015; Gingerich et al 2015; Grosman et al 2011; Morales et al 2015; Riddle and Chazan 2014). Photogrammetric recording of artifacts in the field is becoming increasingly common (Potter et al 2016), as is the documentation of excavation units at different stages of excavation (Doneus et al 2011; De Reu et al 2014) using both terrestrial and UAV generated imagery (Sauerbier and Eisenbeiss 2010). Three-dimensional capture of edifices is also proliferating. Photogrammetry and three-dimensional scanning greatly facilitate “measured drawing, reconstruction and restoration projects” (Yilmaz et al 2007). Archaeologists and heritage professionals may also employ photogrammetry to conduct architectural analysis (de Luca et al 2011; Colosi et al 2013) or track long-term structural and cosmetic damage to heritage structures (Arias et al 2005). These techniques can be expanded to an urban scale using balloons, kites (Altan et al 2004), aerial imagery (Colosi et al 2013; Wernke et al 2016), or terrestrial survey using spherical photogrammetry (de Amorim et al 2013; Fangi 2011; Fangi and Pierdicca 2012; Fangi et al 2013). Photogrammetry is also increasingly being mobilized to capture entire archaeological landscapes (Bikoulis et al 2016; Wernke et al 2016), though its implementation at this scale is still largely restricted to prospection.

Though photogrammetry is generally simpler and cheaper than other methods of three-dimensional surface capture, photogrammetry can produce very similar results to laser scanning or tacheometry, particularly for macro-scale objects (Grussenmeyer et al 2008; Yastikli 2007). Eisenbeiss and Zhang (2006), Grussenmeyer et al (2008), and Fassi et al (2013) all observe only minor deviations in surface data, wireframes, and cross sections generated by these different methods in the capture of architecture (Grussenmeyer et al 2008; Fassi et al 2013) and landscapes (Eisenbeiss and Zhang 2006). Eisenbeiss and Zhang (2006) note that the main issue regarding photogrammetric modelling of macro objects such as buildings or sites is occlusion, a scenario more easily circumvented using a UAV. Ideally, different methods of three-dimensional data collection will be combined (Guidi et al 2011; Guidi et al 2014). As Fassi et al (2013) note, all data collected through photogrammetry or other digital recording techniques should be verified to whatever extent is possible. Photogrammetry and other digital recording methods have already been employed in petroglyph research, yielding a wide variety of workflows and digital products. One of the most productive avenues thus far has been the millimetric and submillimetric reconstruction of incision patterns as a means to analyze how incisions were made and in what order (Hurst et al 2009; Fux et al 2009; Alexander 2015; Escarcena et al 2011; Chandler et al 2007; Sanz et al 2010).) There is relatively little literature on macroscale digital reconstruction of petroglyphs. Alexander et al (2015) constructed a complex apparatus for precisely imaging and georeferencing a 400 m2 rock art panel in the Italian Alps. This device was intended as a “micro-range” scanner that could precisely reproduce the geometries of the actual petroglyphs. This method is beneficial in that the structure and depth of incisions can be reconstructed and analyzed in order to intensively study individual petroglyphs. The panel can be reproduced precisely in digital or physical form. This is valuable because “interactive 3D visualization and analysis requires radiometric surface properties beyond photo-texture, because the 3D content will be viewed under varying artificial illumination conditions” (Alexander et al 2015: 182). As demonstrated by Alexander et al (2015) such high-resolution models can be integrated into macro-scale landscape models, thus serving the additional purpose of heritage preservation. Similar workflows using only commercial-grade digital cameras and a total station have also resulted in high resolution models of rock art that are geolocatable within a scalar digital environment (Escarcena et al 2011; Chandler et al 2007; Sanz et al 2010). Yet very few archaeologists have attempted to use photogrammetry to produce data for geospatial analysis on a macro scale. As vector data are necessary for such analyses, researchers looking to produce analytically viable geospatial datasets have typically employed methods for capturing highprecision point data. As an example, Jennings et al (2014) carefully delineated rock art panels with a Real-Time Kinematic Global Navigation Satellite System (RTK-GNSS) differential GPS. This instrument is accurate to within a 5 mm range and thus allows for careful mapping of rock art. Though Jennings and his colleagues did not utilize photogrammetry, they did carefully code each panel along the escarpment by physical attributes such as: “panel composition, style, number of phases, evidence of superimposition, coordinates, elevation, extent, density, orientation, condition (variation in the degree of weathering or patination), completeness (whole, fragmented or dislodged fragment of a larger panel) and its setting, which involved determining whether a panel was in a primary or an altered position” (Jennings et al 2014: 5). This coding was then used to generate maps of the different attributes. This approach offers conceptual guidance for conducting systematic geospatial analysis of the thematic ecology of the rock art panels.

As we argued in our introduction however, none of these methods are applicable to topographically challenging contexts, and most are time-consuming when implemented on a large scale. Moreover, only Jennings and colleagues (2014) produced a georeferenced registry of petroglyphs within a broad geographic expanse, with techniques that required geolocating each individual petroglyph in the field. Producing the dataset required recording multiple physical attributes in the field. This registry was generated as a vector file separate from any digital environment, necessitating acquisition of highresolution topographic data. These are inconveniences in the best of circumstances, and potentially substantive obstacles under certain conditions. We thus turn to the site of Quilcapampa in order to present a workflow that addresses these issues and combines the most effective aspects of the workflows discussed above. Quilcapampa Antigua Quilcapampa Antigua is a 70-hectare archaeological site in the Sihuas Valley of Southern Peru. Located approximately 1700 meters above sea level, the site was occupied from the ninth to the sixteenth century AD. Quilcapampa is notable in that the white sandstone cliffs on which it is situated provide a stable foundation for construction, in contrast to the loose conglomerate comprising much of the valley. As Jennings et al (2017) note, the cliffs also provide a visually striking canvas on which to carve petroglyphs (Figure 2). The rock art varies greatly in terms of theme, complexity, density, and location. Many petroglyphs are simple designs and not very deeply incised. Others are intricate, deeply carved pieces that can be seen from across the narrow valley. Based on stylistic measures, the first petroglyphs date to a few centuries before the founding of the settlement, although most examples overlap with Quilcapampa’s occupation (Chumpitaz Llerena and Rodríquez Cerrón 2014; Van Hoek 2015). Recording Quilcapampa rock art is via pedestrian survey alone is difficult since many petroglyphs are obscured, and inaccessible without risking bodily injury. Our workflow remotely captures rock art data in such a manner that it can be analyzed later in the lab. Pedestrian Survey with GPS and Camera The initial research phase comprised low-resolution pedestrian survey using a handheld GPS and camera. Pedestrian survey allowed us to definitively delineate the extent of the area covered by the petroglyphs, an approximately 2 km stretch of cliff face correlating closely with geological characteristics. Survey also aids in forming a sense of place, facilitating an understanding of how the petroglyphs interact with a human visual field as one moves through the landscape. Aside from any phenomenological benefit, this is valuable in that certain times of day may be unsuitable for photography due to lighting. If reflectance is high then it can prove difficult to capture images of the petroglyphs, especially those that are not deeply incised. Finally, pedestrian survey is a useful method for determining specific targets for later aerial survey. We thus recorded all petroglyphs that were easily accessible with the handheld GPS and camera. We assigned each petroglyph an image number and cross-referenced it with the corresponding GPS coordinates. However, the area that can be surveyed is limited due to issues of accessibility and safety. The cliffs are steep and much of the sediment is not stable. The ancient footpath has largely collapsed. Some areas cannot be surveyed at all, while others can only be seen from a distance, making it difficult to identify the thematic elements of the petroglyphs or to precisely geolocate them. More importantly, the resolution of most handheld GPS units does not lend itself to high geospatial precision. Most smaller GPS units will only be accurate to within a few meters. The sharp topographic variation exponentially

exacerbates even the slightest imprecision. Larger differential GPS units are unwieldy in such an environment. In sum, these conditions do not facilitate a comprehensive or rigorous analysis of the petroglyphs or their environmental context. Initial Aerial Survey Satellite imagery of the landform is not sufficient for analytical purposes. Aside from a lack of resolution in most commercially available digital elevation models (DEM), the file format creates a “waterfall effect” in which ledges or overhangs can form an illusory surface. A triangulated irregular network (TIN) model that permits a fully three-dimensional surface is of far more analytical value. Although a variety of techniques can produce a TIN model, the cheapest and simplest method is to collect imagery with a UAV that can then be processed into a photogrammetric model. To obtain suitable imagery for photogrammetry, the camera moves incrementally as it captures a series of images in order to yield an overlap sufficient to produce a stereoscopic image, with the Photoscan manual recommending an overlap of 60%. The overlapping images are then used to find common points that lie along the edges of Euclidean surfaces. This information is used to triangulate the camera positions at which the images were captured, solving for the position of the common points. This process alone will not properly scale or georeference the model unless geospatial metadata is attached to the image files. For the model to be scaled properly the analyst must know precise dimensions of at least some of the objects being reconstructed. Likewise, to georeference the model in a real coordinate system, the analyst must know the three-dimensional coordinates of multiple points within the model. Thankfully, many commercially available UAV’s include an onboard GPS system. While a lack of GPS precision can affect the precise geolocation of the model, UAV imagery still produces a TIN in which the internal geometry is consistent once the GPS positions of the images are aligned. More precise georeferencing can be established within the model through the selection of points with known coordinates (Ground Control Points or GCP’s). At Quilcapampa, we conducted an initial low-resolution aerial survey before moving to a higher resolution survey. This permitted us to create a preliminary three dimensional model of the cliff face at a macro scale (e.g Alexander et al 2015). This is a valuable tool in identifying the parts of the cliff face that would be covered by higher-resolution transects and determining the best locations from which to fly the drone along said transects (Figure 3). For this initial survey we used a DJI Inspire 1 with a Zenmuse X3 12 megapixel camera. We flew the UAV 50-100 meters from the cliff face, conducting 3-4 sweeps. We used Agisoft Photoscan to process the images into a three-dimensional model at the default settings. High-Resolution Aerial Survey Higher resolution survey necessitates more planning, as higher resolution photographs require increased proximity to the target. This means that a greater proportion of the battery life of the vehicle must be devoted to smaller coverage area. Higher-resolution transects are also considerably more difficult to plot, as they require close proximity to the cliff face. We thus switched to a UAV with longer flight time, the DJI Phantom 4, with a 12 megapixel camera with hyperfocal length of one meter. Although the Phantom 4 does have forward and downward facing sensors, the transects required to produce a photogrammetric model of a vertical surface require the one sensor to face towards the cliff at all times. This means that the UAV is often moving laterally, a direction in which the sensor does not

produce a reading. Previous surveys attempting to document small targets at a distance of 25 meters have been unsuccessful (Field et al 2017), necessitating flight paths as close to the cliff face as possible. Wind gusts and updrafts can potentially unbalance the UAV to a degree that it can make catastrophic contact with the cliff face. It is thus important to maintain visual contact with the UAV at all times. For this reason, we worked in two-person teams throughout the high-resolution survey. One person maintained constant visual contact, while the other controlled the flight path and camera. Given the sinuous contours of the cliff face, we decided to zonate the survey area based on breaks in the natural topography. We designated either quebradas (dry ravines) or promontories in the cliff face as the edges of each coverage zone. A typical zone measured between 50 and 100 meters. To minimize geometric distortion, before documenting a zone we distributed four targets at the base of the cliff and four at the top to serve as GCP’s. We georeferenced each GCP by assigning an x, y, z, coordinate to the center of every target with a Leica TPS1200 Total Station oriented to the Arequipa base station, thus ensuring geometric and geospatial consistency. We took care to slightly overlap photographic coverage of each zone. This was important not only to ensure full coverage, but also because the edges of photogrammetric models can be slightly distorted when images are not collected at precise intervals through automated flight paths. Not extending the coverage slightly beyond the designated zone can affect the x, y, and z coordinates later assigned to petroglyphs on the edges of a coverage area. Overlap allows for the georeferencing of different sections against each other. This was of particular importance in this instance, as it was typically very difficult to set georeferenced GCP’s on the actual cliff face. We began the high-resolution survey at the southernmost extent of the petroglyphs in an area with extensive tree cover at the bottom of the cliff face, presenting us with early logistical difficulties. We made sure to always keep the drone within our line of sight, moving slowly as we took photographs of the cliff face. We began at the top, and flew horizontal sweeps across the coverage area at 10-15 meter distance from the cliff face. We moved slightly lower with each sweep. As we encountered clusters of petroglyphs we paused our flight path to capture more extensive imagery to be processed separately. After finishing the high-resolution sweeps, we moved away from the cliff face in order to fly three lowerresolution sweeps at a distance of 30-40 meters in order to make sure that the entire cliff face could be modelled. After finishing one coverage zone in this manner, we moved to the next. The number of coverage areas that could be completed in a single day was limited however. Slower flight paths more rapidly consume the battery life of the vehicle. Moreover, lighting conditions after mid-morning were not suitable for capturing high quality images of petroglyphs in this location. Each evening after returning from the field we sorted through the UAV images in order to identify petroglyphs not captured in the initial pedestrian survey. If we had failed to capture sufficiently highresolution images of a petroglyph or a cluster of petroglyphs, then we returned to the location the next day prior to surveying further coverage areas. Processing of photogrammetric model Data processing is the key step in implementing a successful UAV survey. Previous efforts (Field et al 2017) have attempted to identify targets in the orthophoto or digital model. Both products reduce image resolution to a point where smaller targets are not identifiable. Our workflow successfully addresses this issue.

Here we take Section F2 of the escarpment as a representative sample to illustrate our workflow. As the initial step in data processing, we generated a photogrammetric model of Section F2 of the cliff, with petroglyph locations marked by annotations (Figure 6). Ideally, this step will be completed before the end of the field campaign, as processing errors may necessitate additional data collection. In order to generate the best possible model, we recommend first locating the GCM’s in the two-dimensional images and marking their coordinate location before processing by dropping a pin on the target. The standard photogrammetric workflow in Agisoft includes five critical steps: ‘Align Photos’, ‘Optimize Photo Alignment’. ‘Build Dense Cloud’, ‘Build Mesh’, and ‘Build Texture’ (in order). We have generated our best results by setting ‘Key Point Limit’ in ‘Align Photos’ to 100,000, and setting ‘Tie Point Limit’ to 10,000. We normally set ‘Custom Face Count’ in ‘Build Mesh’ to 1,000,000, and ‘Texture Size’ in ‘Build Texture’ to 16,386. We typically leave other settings at default. We stress that at this stage this model is intended as a precision tool for petroglyph registry. It does not function well as a representational tool. Photogrammetric modeling at this scale typically only captures petroglyphs as textures1 rather than three-dimensional renderings. While raster images2 are useful in thematic identification of individual petroglyphs, a systematic geospatial analysis of the thematic ecology requires each petroglyph to be converted to vector data and assigned attributes. Jennings et al (2014) accomplish this through physically locating each petroglyph in the landscape and using a DGPS to mark it as a point in Cartesian space. This method is time consuming however, and terrain features may limit access. Close-range UAV survey permits much more rapid and effective collection of data that can be examined in a laboratory or office setting. Previous attempts at aerial survey of small targets have examined the final orthoimage generated by the software. To identify petroglyphs however, it is necessary to examine individual images, which are higher resolution than a model or orthoimage of a large area. We have examined each image and placed point markers on every unique petroglyph (Figure 8). The Photoscan software automatically locates a marker on the corresponding x, y, z coordinates of the three dimensional photogrammetric model, as demonstrated in this model of Section F2 (Figure 9). Both the point coordinates and the threedimensional model are easily exportable from Agisoft Photoscan into a number of geospatial applications. Additional environmental and relational attributes can be imported directly from the terrain model. The only additional information requiring manual entry is the thematic content of the petroglyph. By pausing flight paths to focus on clusters of petroglyphs, we were able to generate sufficient imagery to produce much higher resolution models of discrete clusters of rock art, as we demonstrate for a subsection of Section F2. These higher resolution models can be hyperlinked directly to the lower resolution model or geolocated within the lower resolution model as a chunk in Agisoft Photoscan. These models are not of a resolution high enough to create depth maps of the incisions as described by Sanz et al (2010) and Zeppelzauer et al (2016). However, they do reveal even the shallowest markings as long as images of said markings were captured by the UAV camera. This can provide a valuable tool for contextualizing the rock art and examining petroglyphs from multiple perspectives without distortion (Figure 7). Panels of interest can be flagged for more precise documentation of the three-dimensional geometry. 1

A texture is an image file overlaid onto the surface of a three-dimensional model. Rasters are graphic files comprised of pixels, whereas most geospatial analyses utilize vector data (points, lines, and polygons that include attribute data). 2

Comparing the data collected through pedestrian survey with the data collected by the UAV in Section F2 of the escarpment reveals notable discrepancies in both the number of petroglyphs identified and the coordinates assigned to them (Figure 10). Figure 4 is a representative sample of the images that we captured from the best vantage points available during the pedestrian survey. We were able to identify 2 individual petroglyphs and 1 petroglyph cluster through this method. Figure 5 shows the GPS points marking each petroglyph location. As discussed above, the GPS points do not correspond precisely to their coordinate positions in real space. Petroglyph coordinates derived from UAV survey correspond precisely with the topography of the cliff face, and an additional 18 petroglyphs were noted in the UAV images. While one of the GPS points from pedestrian survey, visible only from a distance, is coded as a “complex scene”, drone imagery permitted us to delineate individual petroglyphs in a laboratory setting and examine them free of distortion. It also permits us to include appropriate elevation data that was difficult to determine from initial pedestrian survey (Figure 11). We are thus confident that this survey method is much more accurate, comprehensive, and systematic than pedestrian methods alone. Discussion and Potential Applications We foresee two primary applications for the models generated by this workflow. The first is as the basis for a virtual reality model that can be both an important visualization tool for researchers and a compelling element of a public archaeology project. The digital model of Section F2, discussed above, provides proof-of-concept. When uploaded to a digital platform such as Sketchfab and viewed through a virtual reality headset, the model is scalable and manipulable. Users are able to interact with it as they would a physical object, but with a greater degree of flexibility and control. Aside from being an engaging experience, there is a great deal of potential to easily visualize petroglyphs from multiple perspectives and distances, thus observing details that were not noted during fieldwork. The virtual reality model also provides an excellent science communication tool, as other researchers and the public can directly engage with the model. More detailed three-dimensional scans can easily be located within the model, allowing researchers to study incision patterns. The second avenue of interest is in producing a systematic geospatial analysis of the thematic ecology of the petroglyphs. Following Jennings et al (2015) we are coding the attributes of each individual petroglyph into the point file representing that petroglyph. We also intend to code experiential attributes, such as visibility and insolation, that can be generated in GIS software from the threedimensional model of the cliff face. In doing so we hope to better understand how certain themes are distributed in space and how different physical, environmental, and experiential attributes may characterize certain themes. Forthcoming publications will explore both of these applications in greater detail. Conclusion The high-resolution aerial survey detailed in this article represents the first attempt to record and digitally reconstruct a primarily vertical archaeological context. We believe that its further implementation would represent a major step forward for rock art research. Archaeologists will be able to more systematically and comprehensively map both rock art and its context in fine detail. Just as importantly, close-range UAV survey is also a much quicker and safer method for collecting data. Our workflow is particularly useful for other archaeologists conducting research on petroglyphs in other topographically difficult settings, such as gorge slopes (Fowles and Alberti 2016) or high cliff panels (Jennings et al 2017; Mark & Billo 2016). Yet we believe that our methods are also more broadly

applicable to archaeologists seeking to record and analyze macro-scale areas, whether those be sites or landscapes. The UAV captures far more data at a much quicker rate than other methods that we have described above. When situated within the workflow that we have presented, the UAV imagery also permits the careful demarcation of petroglyphs in a fully analyzable file format. While the UAV is not suitable for many modes of archaeological survey, it is of great benefit when it can be flown close enough to targets to study the petroglyphs in the resulting images. References Alexander, C.; A. Pinz; and C. Reinbacher. 2015. Multi-scale 3D rock-art recording. Digital Applications in Archaeology and Cultural Heritage 2(2015) 181–195 Altan, M.O.; T. M. Celikoyan; G. Kemper; G. Toz. 2004. Balloon photogrammetry for cultural heritage. SS 4: CIPA — Low-Cost Systems in Recording and Managing the Cultural Heritage Arias, P.; J. Herráez; H. Lorenzo; C. Ordóñez. 2005. Control of structural problems in cultural heritage monuments using close-range photogrammetry and computer methods. Computers & Structures. 83(21–22), 1754-1766 Atkinson, K.B. 1968. The Recording of Some Prehistoric Carvings at Stonehenge. The Photogrammetric Record. 6(31) 24-31 Bahn, P. 2010. Prehistoric Rock Art: Polemics and Progress. Cambridge University Press. Benazzi S.; Viola, B.; Kullmer, O.; Fiorenza, L.; Harvati, K.; Paul, T.; Gruppioni, G.; Weber, G.W.; Mallegni, F. 2011. A reassessment of the Neanderthal teeth from Taddeo cave (southern Italy). J Hum Evol. 2011 Oct;61(4):377-87 Bikoulis, P.; F. Gonzalez Macqueen; G. Spence-Morrow; W. Yépez Álvarez; S. Bautista; and J. Jennings. 2016. Combining high resolution IKONOS satellite and Unmanned Aerial Vehicle (UAV) acquired imagery to map broad ritual landscapes: preliminary survey results and workflow from the Quilcapampa Geoglyph Survey (Sihuas Valley, Peru). Article submitted to the Journal of Archaeological Science. Bretzke K.; and Conard N.J. 2012. Evaluating Morphological Variability in Lithic Assemblages Using 3D Models of Stone Artifacts. Journal of Archaeological Science 39:3741–3749 Chandler, J.; P. Bryan; and J.G. Fryer. 2007. The Development and Application of a Simple Methodology for Recording Rock Art Using Consumer-Grade Digital Cameras. The Photogrammetric Record 22(117): 10–21 Chumpitaz Llerena, D.; and M. Rodríguez Cerrón. 2014. Los Petroglifos de Chillihuay: la imagen antropomorfa (del formativo al período de integración Wari). http://www.rupestreweb.info/petroglifoschillihuay.html, accessed on March 9, 2016. Clarkson C. 2013. Measuring Core Reduction Using 3D Flake Scar Density: A Test Case of Changing Core Reduction at Klasies River Mouth, South Africa. Journal of Archaeological Science 40:4348–4357. Colosi, F.; Gabrielli, R.; Orazi, R.; Malinverni, E.S., 2013. Discovering Chan Chan: modern technologies for urban and architectural analysis. Archeol. Calcolatori 24, 187–207.

Davis L.G.; Bean D.W.; Nyers A.J.; and Brauner D.R. 2015. GLIMR: A GIS-Based Method for the Geometric Morphometric Analysis of Artifacts. Lithic Technology 40: 199–217 De Amorim, A.; Fangi, G.; Malinverni, E.; 2013. Documenting architectural heritage in Bahia, Brazil, using spherical photogrammetry. In: ISPRS International Archives XL-5W, vol.2, 219–224. De Luca, L.; Busayarat, C.; Stefani, C.; Véron, P.; Florenzano, M. 2011. Asemantic-based platform for the digital analysis of architectural heritage. Comput. Graph. 35 (2), 227–241. De Reu, J.; Plets, G.; Verhoeven, G.; DeSmedt, P.; Bats, M.; Cherretté, B.; DeMaeyer, W.; Deconynck, J.; Herremans, D.; Laloo, P.; et al., 2013. Towards a three-dimensional cost-effective registration of the archaeological heritage. J. Archaeol. Sci. 40(2), 1108–1121. Doneus, M.; G. Verhoeven; M. Fera; Ch. Briese; M. Kucera; W. Neubauer. 2011. From Deposit to Point Cloud – a Study of Low-Cost Computer Vision Approaches for the Straightforward Documentation of Archaeological Excavations. Geoinformatics FCE CTU. 6(2011) 81-88. Eisenbeiss, H.; and L. Zhang. 2006. Comparison of DSMs generated from mini UAV imagery and terrestrial laser scanner in a cultural heritage application. ISPRS Commission V Symposium 'Image Engineering and Vision Metrology Escarcena, J.; E. Mata de Castro; J. L. Perez Garcia; A. Mozas Calvache; T. Fernandez del Castillo; J. Delgado Garcia; M. Ureña Cámara; J. Carlos Castillo. 2011. Integration of photogrammetric and terrestrial laser scanning techniques for heritage documentation. Virtual Archaeology Review 2(3) 2011: 53-57 Fassi, F.; L. Fregonese; S. Ackermann; V. De Troia. 2013. Comparison between laser scanning and automated 3D modelling techniques to reconstruct complex and expensive cultural heritage areas. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XL-5/W1, 2013. 3D-ARCH 2013 - 3D Virtual Reconstruction and Visualization of Complex Architectures, 25 – 26 February 2013, Trento, Italy Fangi, G. 2011. The multi-image spherical panoramas as a tool for architectural survey. In: CIPA Heritage Documentation, Best Practices and Applications, vol. XXXVIII-5/C19,.21–28. Fangi, G.; Malinverni, E.; Tassetti, A. 2013. The metric documentation of Cham towers in Vietnam by spherical photogrammetry. ISPRS Ann. Photogramm. Remote Sens. Spat. Inf. Sci.1(1),121–126. Fangi,G.; Pierdicca, R.; 2012. Notre Dame Du Hautby spherical photogrammetry integrated by point clouds generated by multi-view software. Int. J. Herit. Digit. Era 1(3),461–480 Field, S.; M. Waite; L. Wandsnider. 2017. The utility of UAVs for archaeological surface survey: A comparative study. Journal of Archaeological Science: Reports 13 (2017) 577–582 Fowles, S.; and B. Alberti 2016. 2016. Surface revelations: epistemologies and ecologies of rock art. To appear in Anthropology of the Arts: A Reader, edited by Gretchen Bakke and Marina Peterson. Bloomsbury. Fritz, C.; and G. Tosello. 2007. The Hidden Meaning of Forms: Methods of Recording Paleolithic Parietal Art. Journal of Archaeological Method and Theory Vol. 14(1) 48-80

Fux, P. M. Sauerbier, T. Kersten, M. Lindstaedt, H. Eisenbeiss. 2009. Perspectives and Contrasts: Documentation and Interpretation of the Petroglyphs of Chichictara, Using Terrestrial Laser Scanning and Image-Based 3D Modeling in New Technologies for Archaeology. Springer. pp 359-377 Gingerich J.A.M.; Sholts S.B.; and Wärmländer S.K.T.S. 2014. Fluted Point Manufacture in Eastern North America: An Assessment of Form and Technology Using Traditional Metrics and 3D Digital Morphometrics. World Archaeology 46:101–122 Grosman L.; Smikt O.; and Smilansky U.; 2008. On the Application of 3-D Scanning Technology for the Documentation and Typology of Lithic Artifacts. Journal of Archaeological Science 35:3101–3110 Grussenmeyer, P.; T. Landes; T. Voegtle; K. Ringle. 2008. Comparison methods of terrestrial laser scanning, photogrammetry and tacheometry data for recording of cultural heritage buildings. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing Guidi, G.; Russo,M., 2011. Reality-based and reconstructive models: digital media for cultural heritage valorization. SCIRES-IT 1(2), 71–86. Guidi, G.; Russo, M.; Angheleddu, D., 2014. 3D survey and virtual reconstruction of archeological sites. Digit. Appl. Archaeol. Cult. Herit. 1(2), 55–69. Hansen, J.W. 1999. The state of rock art preservation and climatic change: an example from the Central Saharan Desert, pp. 57-59 in J.C. Dean, ed. Image Past, Image Present: The Conservation and Preservation of Rock Art. IRACX Proceedings, vol. 2. American Rock Art Research Association: Tucson, Arizona. Hubbard, T & S. Ruppell. 2016. The (Dynamic) Mind in the Cave: Representational Space of Cave Paintings and Petroglyphs. Empirical Studies of the Arts. 35(1): 67-92 Hurst S.; E. Louden; E. Johnson. 2009. Unravelling the Cowhead Mesa Petroglyphs with 3-d Laser Scanning and High-resolution Digital Photography. Rock Art Research: The Journal of the Australian Rock Art Research Association (AURA) 2009 (26)1: 43-53 Jennings, R.; A. Parton; H.S. Groucutt; L. Clark-Balzan; P. Breeze; N.A. Drake; A. Alsharekh; M.D. Petraglia. 2014. High-resolution geospatial surveying techniques provide new insights into rock-art landscapes at Shuwaymis, Saudi Arabia. Arabian Archaeology & Epigraphy. 2014(25) 1–21 Jennings, J.; S. Berquist; G. Spence-Morrow; P. Bikoulis; F. Gonzalez-Macqueen; W. Yépez Álvarez; and S. Bautista. In Press. A Moving Place: The Two-Millennia-Long Creation of Quilcapampa in In Powerful Places in the Ancient Andes, edited by J. Jennings and E. Swenson. Albuquerque: University of New Mexico Press. Lewis-Williams, J.D. 2006. The Evolution of Theory, Method and Technique in Southern African Rock Art Research. Journal of Archaeological Method and Theory. 13(4) 343-377 Mark, R.; E. Billo, 2016. Low altitude unmanned aerial photography to assist in rock art studies. In: The SAA Archaeological Record Special Issue: Drones in Archaeology 16 (2). pp. 14–16.

McCarthy, J. 2014. Multi-image photogrammetry as a practical tool for cultural heritage survey and community engagement. Journal of Archaeological Science 43 (2014) 175-185 Morales J.I.; Lorenzo C.; and Vergès J.M.; 2015. Measuring Retouch Intensity in Lithic Tools: A New Proposal Using 3D Scan Data. Journal of Archaeological Method and Theory 22:543–558 Pierdicca, R.; E. Frontoni; E.S. Malinverni; F. Colosi; and R. Orazi. 2016. Virtual reconstruction of archaeological heritage using a combination of photogrammetric techniques: Huaca Arco Iris, Chan Chan, Peru. Digital Applications in Archaeology and Cultural Heritage 3(2016) 80–90. Plets, G.; G. Verhoeven; D. Cheremisin, R. Plets; J. Bourgeois; B. Stichelbaut; W. Gheyle; and J. De Reu. 2012. The Deteriorating Preservation of the Altai Rock Art: Assessing Three-Dimensional Image-Based Modelling in Rock Art Research and Management. Rock Art Research 29(2) 139-156 Porter, S. T., Roussel, M., & Soressi, M. 2016. A Simple Photogrammetry Rig for the Reliable Creation of 3D Artifact Models in the Field Lithic Examples from the Early Upper Paleolithic Sequence of Les Cottés (France).Advances in Archaeological Practice, 4(1), 71-86. Remondino, F., 2011. Heritage recording and 3D modeling with photogrammetry and 3d scanning. Remote Sens. 3(6), 1104–1138. Riddle, A. and M. Chazan. 2014. Stone tools from the inside out: radial point distribution. World Archaeology. Volume 46(1: Digital Imaging), 123-136 Rivett, L.J. 1983. The application of photogrammetry to the recording of rock art and archaeological sites in the Kakadu National Park. In D. Gillespie (ed) The rock art sites of Kakadu National Park, pp.87-125. Australian National Parks and Wildlife Service, Special Publication 10 Sanz, J.O.; M. de la Luz; G. Docampo; S. Martinez Rodríguez; M.T. Rego Sanmartín; and G. Meijide Cameselle. 2010. A simple methodology for recording petroglyphs using low-cost digital image correlation photogrammetry and consumer-grade digital cameras. Journal of Archaeological Science 37 3158-3169 Van Hoek, M. 2015. Rare Petroglyphs of Skelton-Anthopomorphs in Caravalí, Arequipa, Peru. In Adoranten ed. 2014, pp. 88-96. Scandinavian Society for Prehistoric Art, Tanums Hällristnings Museum, Underslös. Wernke, S. G. Or C. Hern ndez A. Rodríguez A. Traslavi a and G. Marcone. 2016. Beyond the Basemap: Multiscalar Survey through Aerial Photogrammetry in the Andes. In E.W. Averett, J.M. Gordon, and D.B. Counts, pp. 251-278 Mobilizing the Past for a Digital Future: The Potential of Digital Archaeology. Grand Forks, ND: The Digital Press at the University of North Dakota. Yastikli, N. 2007. Documentation of cultural heritage using digital photogrammetry and laser scanning. Journal of Cultural Heritage. J. Cult. Herit. 8. 423-427 Yilmaz, H.M.; M. Yakar; S.A. Gulec; O.N. Dulgerler. 2007. Importance of digital close-range photogrammetry in documentation of cultural heritage. Journal of Cultural Heritage 8(4), 428-433

Zeppelzauer, M. G. Poier; M. Seidl; C. Reinbacher; S. Schulter; C. Breiteneder; H. Bischof. 2016. Interactive 3D Segmentation of Rock-Art by Enhanced Depth Maps and Gradient Preserving Regularization. ACM Journal on Computing and Cultural Heritage. 9(4) 1-30

Figure 1. Location of Quilcapampa Antigua in the Majes Area of Southern Peru

Figure 2. Cluster of petroglyphs on limestone cliffs within the F2 study area

Figure 3. Location of the F2 study area in relation to Quilcapampa Antigua (A), orthophotographs of the F2 area from above (B) and perpendicular to cliff face (C)

Figure 4. Photographs of various petroglyphs taken during pedestrian survey

Figure 5. GPS coordinates of a subset of petroglyphs recorded during pedestrian survey overlaid on low resolution DEM of F2 Study area

Figure 6. Final photogrammetric model of F2 study area in Agisoft Photoscan (top) and resulting high resolution orthophotograph of petroglyph at various scales (below)

Figure 7. Comparison of original pedestrian survey photographic recording of a cluster of Petroglyphs (left) to a photogrammetrically produced high resolution orthophoto of the same section of the F2 cliff face based on photographs taken with a UAV (right)

Figure 8. Marking petroglyph locations on images taken during UAV flight within the Agisoft Photoscan interface.

Figure 9. Three dimensional location of petroglyph locations as mapped within the resulting photogrammetric model produced from UAV imagery.

Figure 10. Comparison of results of pedestrian and UAV based survey of the same section of the F2 study area overlaid on a high resolution DEM produced photogrammterically

Figure 11. Comparison of pedestrian and UAV based GPS location data for photographs of the same set of petroglyphs, showing increased accuracy of elevation data for individual engravings when recorded at the correct altitude.