A version of this paper appears in Virtual Reality in Archaeology, 2000, edited by J. A. Barcelo, M. Forte, and D. H. Sanders, Archeopress, Oxford.It is available through http://amazon.co.uk

The reconstruction of prehistoric architecture using three-dimensional (3D) modeling software involves a series of compromises concerning what should or should not be portrayed in a 3D model. In the majority of cases, the data recovered from prehistoric architectural features are inadequate for the complete virtual reconstruction of a structure and its contents. The underlying issue is therefore focused on how "real" we should make our models, especially since the realism inherent in 3D modeling is easy to construe as the past reality when in fact a large proportion of what the viewer sees is heavily dependent on numerous inferences made by the modeler in the absence of relevant data.

This chapter discusses the many issues surrounding the process of creating virtual prehistoric architecture. In addition to the central concern of how to balance realism vs. reality, the discussion considers how decisions in the creation of virtual architecture are further constrained by the goals of the project and the intended audience, the desired product, the quality of archaeological information, and technological capabilities. The chapter ends by examining how these issues have been addressed in 3D reconstructions that I have made of prehistoric architecture from the Southwest United States.

Archaeologists have been involved in 3D reconstructions of prehistoric architectural features for as long as the discipline has existed. Sketches showing intact structures embellished with complete artifacts and surrounded by human figures have often accompanied both popular summaries of prehistoric societies and technical site reports. The intent of these sketches has always clearly been to portray how the architecture might have looked in the past when it was in active use by prehistoric groups. Such depictions have never been so realistic that they were confused with a past reality; in other words, viewers of these sketches have always been able to tell that their creators employed a considerable amount of artistic license to "bring the past alive." If, for example, roofing material was not recovered from the archaeological site, the sketch could simply show a dirt roof without having to consider details of how this roof might have been reconstructed or what the interior ceiling looked like. Alternatively, the sketch artist could have angled the reconstruction or carefully placed scenery such that the roof was not even visible. In the past decade, however, the accessibility of sophisticated 3D modeling and rendering technology has blurred the distinction between what archaeologists actually know about the architecture they are reconstructing and the realism that the model necessarily conveys. It is much more difficult to gloss over missing information when one is creating a 3D model of a prehistoric structure (Figure 1). If the archaeologist was unable to identify the roofing material, then what should the modeler do about a roof on the virtual reconstruction? Should the modelers leave a roof off, or should they put a "dirt-colored" square in its place? And if they do the latter, how thick should it be? With 3D models, unknown aspects of the prehistoric architecture cannot be as easily avoided.

Figure 1: The archaeological remains of the structure known as Chimney Rock, located in southwestern Colorado, do not provide many details on how the structure was roofed. For example, although this model includes "firewalls" that extend above the roof, their presence is not indicated archaeologically. They are included based on analogy with similar structures built by the Hopi and Zuni, the probable descendents of the prehistoric groups that occupied Chimney Rock.

The reason that this issue is important is that the ability to generate photorealistic renderings using actual materials and textures that archaeologists have recovered from actual prehistoric contexts can make it difficult for the end user to distinguish between the model’s realism and the archaeological reality. The exceptional quality of 3D architectural renderings has made it substantially more difficult for viewers of visual reconstructions to determine which features were actually identified by the archaeologists and which represent more tenuous interpretations that the modelers were forced to make as part of the modeling process. In essence, the process of 3D modeling has made virtual reconstructions appear realistic whether or not they accurately reflect how the architecture really appeared in the past.

The importance of realism vs. reality in 3D modeling depends to a large degree on the goals of the reconstruction and on the constraints that affect the modeling process. The following sections explore these issues and how they can affect how virtual reconstructions are created and presented. This discussion is intended primarily to introduce important issues and is not intended to represent a rigorous program for guiding 3D reconstructions, for the specific criteria and constraints facing any given reconstruction will be unique to every project.

Many of us who have been creating 3D reconstructions of prehistoric architecture began our modeling with no clear vision as to the function of our models. Certainly, in my case, 3D modeling began with little vision beyond surmising "wouldn’t it be cool if I could do this." Only later did I realize the potential that 3D visualization held for achieving both research and pedagogical goals. As will be described later, my first models suffered greatly by not having clearly identified goals. Now, before I begin a 3D model of a prehistoric architectural feature, I carefully consider the purpose of the reconstruction and try to design the model accordingly. Particularly important is reaching compromises between three possible goals that may place competing demands on the modeling process: research, pedagogy, and public consumption.

The virtual reconstruction of architectural features has yet to play a substantial role in archaeological research. An excellent example of the use of 3D modeling for addressing research questions is provided by the work of Philip Peterson, David Fracchia, and Brian Hayden on pithouses at the Keatley Creek site in British Columbia (1995). These researchers began with the identified floor area of an approximately 2,000-year-old pithouse and attempted to determine how the superstructure might have been constructed. Based on this reconstruction, the researchers could determine which parts of the structure would have been illuminated by sunlight at various times of the day. Data on artifact frequencies and feature location were then layered onto the floor of the structure and then compared with the information on illumination. This helped the researchers to determine usage areas in the structure. For example, they discovered that heavily retouched scrapers appeared most frequently in parts of the structure that were lit by midday sun (Peterson et al. 1995:33).

While this example shows the potential for using 3D reconstructions to address research questions, few other archaeologists have pursued this kind of study. Virtual reconstructions could be used to determine how much material would be required to construct walls of an architectural feature, or different theories of how a roof might have been built could be evaluated. A potentially promising area of investigation is the use of 3D models to evaluate architectural features that are hypothesized to have served as markers of astronomical events such as solstices or equinoxes. At the very least, the process of creating architectural models can identify inconsistencies in the actual archaeological data and rectify incorrect assumptions about the appearance of prehistoric features. For example, through their 3D reconstructions of habitations in the Central American community of Ceren, researchers determined that previous assumptions that interiors were dark and gloomy were incorrect (Novitski 1998:35). Other archaeologists have used 3D modeling in research directed towards preservation and management of archaeological resources on public land (e.g., Goodpasture 1997).

The modeling of prehistoric architecture to address research questions is obviously concerned with the faithful replication of the archaeological record. Since the goal is to address questions about the past reality, achieving a completely realistic view of the architecture is not important. It may not be important for the walls to be covered with realistic textures or for artifacts to be placed on the floor. Exactly what is modeled will be determined more by the research agenda than by the goal of creating a detailed representation of the past.

Virtual reality reconstructions are especially useful for teaching students about prehistory. For example, at the University of California at Santa Barbara, Brian Fagan and George Michaels have developed software for an introductory course to archaeology that provides multimedia learning experiences for students (Fagan and Michaels 1992). Included are fly-throughs of prehistoric architecture from Sumerian Ur and Maya Tikal. The software for this course has recently been made available to select campuses over the Internet, allowing many more undergraduates to experience virtual architecture as an integrated part of their learning experience. A somewhat different project by John Rick at Stanford University presents a 3D model of architecture from Chavín de Huántar in Peru and then uses this as a "virtual fieldsite" through which students can make observations, form hypotheses about the site, and then test them with their explorations of the site model. Similar pedagogical uses of 3D models are becoming increasingly common (e.g., Flanagan 1998).

Figure 2: The 3D model of the prehistoric structure of Kin Tl’iish, which is located in northwestern New Mexico, features rooms complete with pottery and blankets hung over poles. Because this model was created for teaching students about the prehistory of the Southwest, I took considerable liberty in populating the interior with artifacts to increase the realism of the experience. However, the structure itself has never been excavated, and therefore there is no information to suggest exactly what features were found in each room. Click on the image for a Quicktime VR view of this room.

A potential problem is that, unlike models created to address research problems, those created for pedagogical purposes are expected to be complete representations of the architecture. This in turn usually requires a number of interpretations that may not be readily supportable with the actual archaeological evidence recovered from the structure’s remains (Figure 2). For example, if no organic remains were identified during excavations, should baskets or wooden bowls be put in the 3D reconstruction? If they are not there, then students may be led to believe that such objects were never used by the prehistoric society in question. The essential problem is that if the model is a perfect recreation of the archaeological record, the result may be an austere representation that can mislead students into thinking that the structure was devoid of household goods. A similar but perhaps even more important problem is whether or not people should be included in a model, for their absence dehumanizes the representation of prehistory.

On the other hand, by providing a fully detailed image of the past, we run the risk of undermining students’ imagination and interpretive skills for bridging between the archaeological data and a vision of what the past was like. One possible solution to this dilemma is to use the actual creation of 3D models as a learning experience for more advanced students. By directly involving students in creating a representation of the past from static archaeological data, not only will they gain a better understanding of prehistory but they will also be exposed to the interpretive challenges inherent in such an undertaking. At the same time that they are learning the subject matter, they are able to draw upon a generational interest in digital technology and develop their own computer skills. This is perhaps the 21st-century equivalent of having grade-school students build architectural models out of tongue depressors and papier-mâché, but the advantage with digital models is that the students can create the models using accurate measurements and actual "materials" that were used in the original structure.

Probably the most common goal for producing virtual reconstructions of prehistoric architecture has been for consumption by the general public. And undoubtedly the most widespread products in which 3D models have appeared are in computer games. From fanciful representations such as those found in Tomb Raider to the more complete models presented in Qin: Tomb of the Middle Kingdom, most young people today are likely to first experience prehistoric architecture in the context of digital entertainment. Of course, these experiences usually provide a bizarre perspective on both prehistory and archaeology, often combining features from disparate cultures and time periods. But at the same time many people become intrigued by these representations and often pursue this interest through more traditional channels to learn about the past. Computer games therefore cannot be disregarded as completely irrelevant to the discipline and to the goal of exposing people to prehistory.

Virtual models of prehistoric architecture created for public consumption are always fully realistic representations, and, especially in the case of computer games, they may even be populated with human figures. The production of these models therefore require the greatest number of interpretive leaps between what is known archaeologically and the realism that they convey. In some cases, and particularly in computer games, the models are completely made up. In others, however, such as the game of Qin: Tomb of the Middle Kingdom, the models actually do have some basis in archaeological reality (Novitski 1998:26—27). In Qin, although the model is of an actual tomb and its surrounding structures, the tomb that is represented has not yet been excavated by archaeologists. However, in order to be as accurate as possible, the modelers turned to archaeological records from similar discoveries and used actual objects and textures recovered from other sites to complete their representation of the tomb.

Public consumption of virtual reconstructions also occurs widely on the Internet. For the most part, these representations appear to have been created to achieve either research or pedagogical goals and then placed on the Internet. In other cases, architects or digital artists with an interest in modeling prehistoric structures have created websites presenting their recreations. Either way, the wide availability of these models to people all over the world makes this an important arena for the representation of prehistoric architecture. Most people visiting these sites are formal or avocational students of archaeology and prehistory, and as such the interpretive problems associated with such models are similar to those discussed in the previous section. Care should be taken by the model’s creators both to demonstrate the purpose of the 3D model and its accuracy in representing what is actually known about the architectural features it purports to recreate. Architectural reconstructions presented on the Internet have the most potential to be both misrepresented and misinterpreted since so many people with a wide variety of backgrounds both produce the online materials and view the resulting models.

The goals of a virtual reconstruction and the intended audience are not the only factors that structure how the modeler moves between the reality of the archaeological record and the realism represented by the 3D model. Also important are the quality of the available archaeological information from which to draw interpretations as well as the technological capabilities available to the modeler.

The ability to create accurate 3D reconstructions of prehistoric architecture is clearly dependent on the degree of preservation of the prehistoric material and the quality of the archaeological research (Figure 3). Many of the largest and best-known prehistoric structures all over the world were either looted long ago or they were excavated in the 19th or early 20th centuries when modern techniques of data recovery were not available. While structural information is likely to have been preserved and identified in these early investigations, critical archaeological data on the contents of these structures were often discarded and not sufficiently recorded, and now we can only surmise what the interiors might have looked like. In other cases, such as in the pithouses digitally reconstructed by Peterson et al. (1995), much of the structure was organic and has long since decomposed, leading the modelers to rely on ethnohistoric information on how similar pithouses might have looked.

Figure 3: Kin Tl’iish, a prehistoric structure found in northwestern New Mexico, has not been excavated and the exact configuration of rooms is unknown. Although the remaining mound suggests that some portions of the original structure were multi-storied, this too cannot be ascertained without further excavation. The 3D model, therefore, provides an interpretation that may in fact be wrong. Providing a description of the interpretive leaps required to produce the model is important to mitigate against the reconstruction being interpreted as a completely factual snapshot of the past. Click on the image for a preliminary fly-through of this model.

In contrast, contemporary archaeological projects have access to modern techniques of data recovery that can provide modelers with detailed information about the architecture and its contents. In special cases, such as the University of Colorado investigations at Ceren in El Salvador, the structures were buried by ash from a volcanic eruption and the degree of preservation is therefore extremely high. Both the archaeologists and the modelers in these situations do not have to make as many interpretive leaps as do those investigators working on material from older and less well-preserved excavations. Because 3D models convey a strong sense of realism no matter how complete the data on which they are based, scholars working with and presenting these virtual reconstructions should include descriptions of the underlying archaeology and the interpretive leaps that had to be made to build the model.

Access to appropriate technology and skills can play a major role in determining the scope and quality of a 3D model of prehistoric architecture. If the archaeologist is attempting to construct the model alone, he or she might not be expected to have either the modeling skills or the access to powerful hardware and software needed to produce a superior model. Accordingly, the most successful examples of high-quality reconstructions involve collaborations between archaeologists and either architects or computer artists specializing in architectural visualization, in which case the degree of communication between those with the archaeological knowledge and those with the modeling skills becomes critical. The increasing desire to create 3D models of prehistoric architecture is also beginning to sustain a small private industry of professionals with both the tools and experience to create superior 3D reconstructions. Many of these professionals have been trained both in archaeology and in computer graphics, making them perfect for bridging between the prehistoric reality and a model’s realism.

Of course, access to professionals requires funding that has not traditionally been available to most archaeologists. Although many examples of the successful visualization of prehistoric architecture involve collaborations between academic departments or other in-house arrangements, archaeologists should consider seeking funding from extramural sources (e.g., Flanagan 1998). Money both to develop "digital labs" as well as funds earmarked for pedagogical purposes and public outreach might be obtained from large public-funded organizations such as the National Science Foundation and the National Endowment for the Humanities. For example, Education Development and Demonstration grants from the National Endowment for the Humanities are directed towards new electronic technologies that enhance students’ understanding of humanities subjects.

The various factors described in this chapter provide often conflicting influences on the creation of 3D models of prehistoric architecture. Projects will therefore greatly benefit by a full consideration of these issues and how they might determine what is or is not feasible. In this section, I consider the lessons that I have learned in the production of 3D models of prehistoric Anasazi architectural features from the northern American Southwest. This discussion will particularly focus on my reconstruction of a great kiva, a ceremonial structure that appeared in Anasazi communities between approximately A.D. 800 and 1150. The reconstruction of the great kiva was structured by two major goals. The original impetus for pursuing the visualization of one of these structures was pedagogical. Although in my courses I was able to show slides of the ruins of great kivas, these images never adequately portrayed what one of these structures would have been like when they were in use (Figure 4). Great kivas were built mostly below the ground-level and then completely roofed, and they would have been dark and smoky places that clearly conveyed a mysterious and sacred atmosphere. The current ruins, on the other hand, are roofless and bright and accordingly appear much more open and accessible than the original structures really were. Therefore, the primary goal of modeling a great kiva was to better convey to students what it really would have been like to have been inside one of these structures. My intention was to produce a detailed reconstruction complete with artifacts, a fire in the firebox, and perhaps even background music, with the product available both over the World Wide Web and on a CD-ROM that I could use in the classroom. A secondary goal for the project was to pursue research questions during the production of the model. In particular, I was interested in determining how great kivas would have been roofed, a question that has been pursued in other contexts (e.g., Lightfoot 1988; Snygg and Windes 1998), but which has never been addressed using 3D modeling. My intention was to evaluate a number of hypothesized methods for roofing such a structure based on both the known size of the structure and knowledge about available timbers in the area. Unfortunately, the production of the great kiva model ran into three interrelated problems. First, a major problem that I faced in achieving the desired detail was the relatively incomplete state of the archaeological data. A number of great kivas have been investigated, but almost all of the full excavations occurred earlier in the 20th century; most current projects only test portions of the structure. Because I needed a complete excavation in order to build a full 3D model, I chose the great kiva at a site named Chetro Ketl in Chaco Canyon in northwestern New Mexico. This structure was intermittently excavated over a period of 12 years beginning in 1920. The excavations were first conducted by the Museum of New Mexico and then continued both by the Museum and the School of American Research. My source of archaeological data came from Vivian and Reiter (1960), but this report contains very little information on artifacts recovered during excavations and few architectural details beyond descriptions of masonry features that were identified (Figure 5). For example, both because of poor preservation and incomplete reporting of the work, no information was available on the kinds of intact vessels found in the structure, the nature of the plaster covering the walls, or the roofing material that may have been used. The creation of a complete 3D model therefore involved a number of interpretive leaps that relied on my knowledge of Anasazi prehistory and on excavations of other similar structures (Figure 6). For example, the plaster and pictographs that appear in the model were derived from investigations at other, much smaller kivas; no investigation of a great kiva that I know of has ever identified intact plaster with artwork on it. Particularly challenging was determining the function of the "vaults" located in the floor of the structure. The archaeologists who excavated the kiva felt that perhaps the vaults functioned exclusively as repositories of ceremonial goods. Other archaeologists have argued that these features were used as huge firepits, but no evidence of burning was found in the vaults. My interpretation of the function of these features was derived from ethnographic examples of Hopi kivas, which exhibit similar vault-like features that were covered with planks or leather and used as drums. A variety of such interpretations had to be made and are outlined in text that accompanies the great kiva presentations. These descriptions of the inferential process have actually been very useful for illustrating to students how archaeologists derive conclusions from archaeological data. For several of the features, I actually involved advanced students in the interpretive process, assigning them small research projects to determine how best to reconstruct and model specific features within the great kiva.

Figure 4: Chetro Ketl, a ceremonial kiva found in Chaco Canyon in northwestern New Mexico, is now a completely exposed archaeological site whose features including the once-concealed niches along the interior wall, are easily visible. This, however, is not at all what this structure would have been like when it was in use by prehistoric people living in the canyon, for it would have been plastered and completely roofed. A complete interactive tour designed for pedagogical purposes can be found by clicking on the image. Figure 5: The excavation of Chetro Ketl occurred in a number of stages earlier in the 20th century. Very little information is readily available regarding the artifacts that were found within the structure. In order to create a realistic, if not real, 3D model of the interior space complete with portable artifacts, excavation records from other similar structures were examined. Clicking on the image will provide a description of the actual archaeological record from Chetro Ketl. Figure 6: No artwork was identified on the remaining plaster found in Chetro Ketl, but for pedagogical purposes, prehistoric imagery was placed on the model’s walls, supported by analogy with similar structures in which artwork was identified. Similarly, the function of the mysterious vaults seen at the base of two of the posts has been debated by archaeologists. Based on ethnographic analogy, we interpreted these features as large foot drums. The ambivalence surrounding their function, however, cannot be easily conveyed in a realistic 3D model. Figure 7: One of the research questions that guided the reconstruction of the Chetro Ketl kiva was to determine how such a large structure could have been roofed given the relative paucity of wood in Chaco Canyon. After investigating some of the ideas for roofing a kiva that have been proposed in the archaeological literature, a roof consisting of a minimum number of rare large beams and a multitude of smaller pieces of wood was designed for the 3D model. Unfortunately, the resulting 1,000-piece roof became very difficult to render.

A second problem that was encountered during the creation of the great kiva 3D model was how best to accommodate both the goal of providing sufficient realism for pedagogical purposes while also pursuing research questions. Because of my interest in determining how a great kiva would have been roofed, the majority of effort in the modeling process was focused on reconstructing the roof (Figure 7). While this research agenda was, in my opinion, sufficiently addressed, the result was a 3D model of the roof that included almost 1,000 objects and that consumed both my effort and computer power to create and to render. Although I can now calculate how much timber would have been required to roof the Chetro Ketl great kiva and how this roof was likely designed, the subsequent detail in the roof far exceeds the needs of the model for pedagogical purposes.

This brings us to the final and perhaps most difficult problem that I faced in creating the 3D model. Because I was trying to achieve realism for pedagogical purposes while also trying to make the roof as realistic as possible to answer my research questions, the resulting model became very complex and unwieldy. With well over 1,000 objects, each draped with high-quality textures, as well as a rotoscoped movie of a fire in the firebox, by the time that the model was complete it was nearly impossible to render it. Because I had no funding for this project, the software (Specular [now Metacreations] Infini-D) and the hardware (Apple Quadra 800) were not powerful enough to achieve both my pedagogical and research goals at the same time. The result was that I was actually unable to fully render the model until over a year after it was first created, at which point I had access to more powerful equipment.

My latest projects in architectural visualization have endeavored to accommodate the lessons learned during the creation of the great kiva 3D model. I am currently attempting to reconstruct prehistoric great houses, quasi-ceremonial structures that were often found in the company of one or more great kivas within many Anasazi communities. This attempt is purely for pedagogical uses, and I have accordingly simplified many aspects of the model. For example, instead of a fully detailed roof, I am using a single flat object on which is mapped a photograph of a well-preserved interior ceiling from an actual great house, drastically reducing the processing power needed to render the structure. For this model, I also have access to comparatively complete archaeological records, facilitating the inferences needed to bridge from the archaeological reality to the realism allowed by the visualization process.

The production of 3D reconstructions of nonextant prehistoric architecture requires a careful consideration of the amount of realism conveyed and the faithfulness of the model to archaeological reality. At the basic level, these two objectives will never be commensurate simply because the archaeological record is necessarily an incomplete reflection of prehistoric reality. Further complicating the modeling process are the goals of the model, the intended audience, the quality of archaeological information, and the access to sufficient technology and skills. As the examples in this chapter have demonstrated, these factors can often work against each other, substantially complicating the production of the 3D model. Reconstructions will therefore benefit by a careful consideration of all potential constraints prior to the actual modeling process. Perhaps most importantly, the presentation of the final model should be accompanied by a description of the original archaeological material and the inferences that were made to build a complete model. As 3D modeling software and hardware become more sophisticated, the amount of realism that can be created can mislead end users into believing that the model represents true prehistoric reality. While the power of these models to advance our pedagogical and research goals is undisputed, we need to be aware of the dangers of misrepresenting the past.